”i. 2c -0 f Ink! I.” I... - youbmrzcnwmniru. . . - 4?".er .I .n ‘ 5 ) ‘flr. .1 . f n S . rfis§ « , . .1.. ,_ . .J .nnts. ‘ 5 , frw. )cl 1.1,” . n; | $1.1. atu$¢ ’60., .1 a. 3.1.0 .. ‘.1( v v. I Ov- V .- iumfl “3.. Irony. It! i .‘ I . .‘ . , n ‘ I , .l ‘. . 1.. ‘ x.‘ n: . Expat? : 331.1. ,r......-.....!...rr. Ln... , .. ll ll llllll llllll ll H l ll‘llllll ll 3 1293 10145 6642 This is to certify that the thesis entitled Risk and Uncertainty in Political Choice: Candidates' Policy Positions in Congressional Elections presented by Eugene Jay Alpert has been accepted towards fulfillment of the requirements for Ph.D. degree in Political Science pQW/Zd flags, Major professor Date November 17, 1977 0-7639 , W25 '¢¢t.l it“ . -' a; the requirement! M ' rm RISK AND UNCERTAINTY IN POLITICAL CHOICE: CANDIDATES' POLICY POSITIONS IN CONGRESSIONAL ELECTIONS BY Eugene J. Alpert AN ABSTRACT OF A DISSERTATION Submitted to Michigan State University in partial fulfillment of the requirements for the degree of DOCTOR OF PHILOSOPHY Department of Political Science 1977 I..‘. . a... . ‘5?- '4. "' ‘vA'. 0". ..~ 5...: ABSTRACT RISK AND UNCERTAINTY IN POLITICAL CHOICE: CANDIDATES' POLICY POSITIONS IN CONGRESSIONAL ELECTIONS BY Eugene J. Alpert This study involves the analysis of decision making by congressional candidates under conditions of risk and uncertainty. Decision making under uncertainty occurs if each action chosen leads to one of a set of possible out- comes, each occurring with a certain probability. The probability is either unknown or estimated by the decision maker. If there is some expected loss that can occur, then the decision making involves a level of risk. In a political campaign, the action often involves the choice of a public policy position that will lead to a minimization of the expected loss of votes in the election. Since there is often uncertainty about the true distribu- tion of voter opinion on an issue, a candidate can use subjective and objective information to estimate the majority position. A theoretical framework is thus developed to determine the effects of risk and uncertainty On the selection of a public policy position. The model includes concepts associated with Bayesian decision making, since the choice of a public policy .4. Q.‘ ‘ulq Ovl 0.. «~- .0.‘ '(’ (H n: I .’ I U! ‘)o 1"“ in. n q u u . ‘I ~w—————_—__fi— Eugene J. Alpert position involves the use of prior information to establish and revise estimates of district opinion as a campaign progresses. The Bayesian analysis allows us to deduce a number of hypotheses relating the independent variables of uncertainty and risk to the candidates' choice of a public policy position, the dependent variable. The 1958 Representation Study, conducted by Warren E. Miller and Donald E. Stokes at the University of Michigan, includes data used to test the hypotheses. The study con— tained interviews with 251 Congressmen from the 85th Congress and their challengers in the 1958 midterm election. The basic findings show that candidates who were opposed for election and who perceived less uncertainty about district opinion were more likely to choose a public policy position close to what they perceived to be the opinion of a majority in their district than those candidates who perceived more uncertainty. Opposed candidates who per- ceived themselves to be in a situation where uncertainty and risk made the potential for a loss of votes high were also more likely to follow what they perceived to be district opinion. Generally, the results indicate that risk and uncertainty could be used to explain the circum- stances in which candidates would be more likely to follow their perceptions of district opinion. The research provides empirical information about the linkage between public opinion and public policy, which can assist in the determination of the responsiveness of i‘ " I A Eugene J. Alpert representatives to their constituents' opinions. For example, majority party candidates from safe districts and incumbents in general were more likely to follow district opinion than those from the minority party in safe districts and those who were nonincumbents. This is interesting, because, as earlier results from these data have shown, the constituents from these districts were not likely to vote for their representatives on the basis of the candi- dates' legislative positions. The present study provides support for the use of subjective decision making techniques in studying the perceptions of political actors. As a result, additional avenues of research in this area can be identified, as well as areas where a reconceptualization according to the frame- work of Bayesian decision making can provide stronger results. This is illustrated in the thesis by the examina— tion of the "marginality hypothesis," which showed that a subjective decision making approach could help to understand the contradictory findings in the literature. In this way, the relationship between campaign decision making and legislative decision making can be more fully explored. l I... 4 'un I I (II III All (‘9 O ACKNOWLEDGMENTS I would like to thank my dissertation committee chair— man, David W. Rohde, for his continued support and assist- ance during the writing of this thesis. His comments, con— structive criticisms and careful readings of the manuscript have helped me clarify many of the ideas contained herein. The other committee members, Cleo H. Cherryholmes and Joseph A. Schlesinger, also contributed to the development of this research with their helpful comments. Their assistance was deeply appreciated. There are also a number of individuals who have inspir- ed me and directly or indirectly influenced my interests in this area of research. These include Samuel J. Bernstein, Richard F. Fenno, David B. Meltz, Alvin Rabushka, William H. Riker, and Kenneth A. Shepsle. From my undergraduate days at the University of Rochester and through my graduate study at Michigan State University, these individuals have motiva- ted me by their research and insights in political science. I would also like to thank the Departments of Political Science at Michigan State University and the University of Florida for their support in supplying computer time and access to the data provided by the ICPR. At Michigan State, Harriet Dhanak and Elizabeth Powell were particularly helpful in this area. Rachel Orsak typed the manuscript in ii -iil v s V' :~ 2. ~—» 2‘ A- U. what must have been record time, Lynn Dally drew many of the figures, and Kathy Larsen and Jackie Wishert helped prepare the bibliography. It is acknowledged that neither the ICPR nor the original collectors of the data for the 1958 Representation Study, Warren E. Miller and Donald E. Stokes, bear any responsibility for the analysis or interpretations presented herein. My sincere thanks and a debt of appreciation go to my parents, Esther and Theodore Alpert, who have helped support my education and whose encouragement was always welcomed. A very special acknowledgment is owed my wife, Heidi, who has assisted me during various stages of the research and in many valuable ways, while patiently waiting for this long task to find its way toward completion. I dedicate this dissertation to Heidi. TABLE OF CONTENTS Page List of Tables. . . . . . . . . . . . . . . . . . . . vi List of Figures . . . . . . . . . . . . . . . . . . . x Chapter I. Risk and Uncertainty in Political Decision Making . . . . . . . . . . . . . . . . . . . . l 1. Introduction . . . . . . . . . . . . . . . l 2. Representatives and Responsiveness . . . . 3 3. Constituency Influence in a Legislature: A Theoretical View . . . . . . . . . . 7 4. Political Perceptions: A Survey of the Art I n o o o n o n n I a I a o o u o 15 A. Informing the Ignorant B. Rationalizing Reality C. Overcoming Uncertainty D. Risky Choices. . . . . I O O I I I .I I O D I 9 D p..- \D 5. A Need for Theory. . . . . . . . . . . . . 33 Footnotes. . . . . . . . . . . . . . . . . . . 38 II. A Bayesian Model of Political Choice . . . . . 50 1. Introduction . . . . . . . . . . . . . . . 50 2. Bayesian Decision Theory . . . . . . . . . 52 3. The Bayesian Decision Model. . . . . . . . 56 4. The Decision Problem . . . . . . . . . . . 60 5. Conclusion 0 I I C I O O I I O I I I I O I 77 Footnotes. . . . . . . . . . . . . . . . . . . 79 III. Candidate Decision Making. . . . . . . . . . . 93 1. Choice and Constituency Influence. . . . . 93 2. The 1958 Representation Study. . . . . . . 98 3. The Hypotheses . . . . . . . . . . . . . 105 4. Summary. . . . . . . . . . . . . . . . . . 127 Footnotes. . . . . . . . . . . . . . . . . . 130 iv Chapter Page IV. The Analysis of Uncertainty and Risk . . . . . 135 1. Introduction . . . . . . . . . . . . 135 2. The Statistical Analysis . . . . . . 136 3. Uncertainty, Information and Effort. . . . 141 4. Policy Positions, Uncertainty, and Risk. . 165 5. Summary. . . . . . . . . . . . . . . . . 184 Footnotes. . . . . . . . . . . . . . . . . . . 187 V. Responsiveness, Representation, and Roll Calls 193 1. Introduction . . . . . . . . 193 2. Incumbency, Election Outcome, and Policy Positions. . . . . . . . . . . . . . 193 3. Marginality and Responsiveness . . . . . . 206 4. Incumbency, Election Outcome, and Roll Call Positions . . . . . . . . . . . . 213 Footnotes. . . . . . . . . . . . . . . . . . . 223 VI. Models, Decisions, and Representation: An Assessment . . . . . . . . . . . . . . . . 226 1. Introduction . . . . . . . . . . 226 2. A Bayesian Analysis of Political Choice. . 227 3. Toward a Representative Democracy. . . . . 237 Footnotes. . . . . . . . . . . . . . . . . . . 242 Bibliography. . . . . . . . . . . . . . . . . . . . . 245 Table 3.1 3.12 3.13 3.14 3.15 3.16 LIST OF TABLES Type of Candidate . . . . . . . . . . . . . Type of Candidate in Relation to Incumbency Type of Candidate By Competition. . . . . . Type of Candidate By Incumbency . . . . . . Type of Candidate By Election Outcome . . . Type of Candidate: Opposed Candidates by Incumbency . . . . . . . . . . . . . . . Incumbency By Election Outcome: Opposed Candidates (Unweighted). . . . . . . . . Incumbency By Election Outcome: Opposed Candidates (Weighted). . . . . . . . . . Description of Variable 0095: Knowledge of District Opinion . . . . . . . . . . . . Description of Attitude Scales: Variables 0042, 0054, & 0065 . . . . . . . . . . . Description of Variables: Perception of District Opinion By Issue Area . . . . . Description of Variable 0046. . . . . . . . Description of Variable 0094: People Interested in the Issues, and Variable 0169: People in District Know Candi— date's Stands. . . . . . . . . . . . . . Four Expected Loss Situations . . . . . . . Creation of a New Variable: “Expected Loss From Issues" . . . . . . . . . . . . . . Creation of a New Variable: "Expected Loss From Stands" . . . . . . . . . . . . . . Page 101 102 103 103 103 104 104 105 106 107 108 111 114 116 118 119 Table 3.17 Page Frequency Distribution of Variable "Expected Loss From Issues" (Opposed Candidates) . . 121 Frequency Distribution of Variable "Expected Loss From Stands" (Opposed Candidates) . . 121 Description of Variable 0159: Perception of Party Strength . . . . . . . . . . . . . . 125 Knowledge of District Opinion By Incumbency, Election Outcome, and the Number of Times a Candidate Ran For Congress . . . . . . . 143 Knowledge of District Opinion By Incumbency: Contingency Table. . . . . . . . . . . . . 143 Knowledge of District Opinion By Election Outcome: Contingency Table. . . . . . . . 144 Knowledge of District Opinion By Number of Times a Candidate Ran for Office (Congress): Contingency Table. . . . . . . . . . . . . 4 Creation of Variables: "Expected Loss From Foreign Affairs," "Expected Loss From Social Welfare," and "Expected Loss From Civil Rights". . . . . . . . . . . . . . . 147 Expected Loss By Incumbency . . . . . . . . . 149 Expected Loss By Election Outcome . . . . . . 149 Expected Loss By Number of Times Candidate Ran for Congress . . . . . . . . . . . . . 149 Description of Variables Identifying Sources of Campaign Information. . . . . . . . . . 153 Dependence on Information Sources by Incum- bency, Election Outcome and Number of Times a Candidate Ran for Office . . . . . 155 Dependence on Information Sources By Knowledge of District Opinion. . . . . . . . . . . . Dependence on Information Sources by Knowledge of District Opinion Controlled for Incumbency and Election Outcome. . . . . . 158 Table 4.13 4.26 Expected Loss By Dependence on Information Sources. . . . . . . . . . . . . . . . . . Dependence on Information Sources By Expected Loss Controlled for Incumbency and Election Outcome . . . . . . . . . . . . . Description of Variable 0206: Estimated Chance of Winning Election . . . . . . . . Description of Variable 0170: Perceived District Knowledge of Candidate as a Person . . . . . . . . . . . . . . . . . . Perceived Constituent Knowledge of a Candi— dates' Stands and Knowledge of Candidate as a Person By Estimated Chance of Winning Election . . . . . . . . . . . . . Candidates' Policy Position By Perception of District Opinion. . . . . . . . . . . . Foreign Affairs Policy Position By Perception of District Opinion. . . . . . . . . . . . Social Welfare Policy Position By Perception of District Opinion. . . . . . . . . . . . Civil Rights Policy Position By Perception of District Opinion. . . . . . . . . . . . Candidates' Policy Position By Perception of District Opinion Controlled for Knowledge of District Opinion. . . . . . . . . . . . Candidates' Policy Position By Perception of District Opinion Controlled By Perception of People's Interest in the Issues . . . . Candidates' Policy Position By Perception of District Opinion Controlled By Perception of People's Knowledge of Candidates' Stands . . . . . .'. . . . . . . . . . . . Candidates' Policy Position By Perception of District Opinion Controlled By Expected Loss . . . . . . . . . . . . . . . . . . . Candidates' Policy Position By Perception of District Opinion Controlled By Risk: Noncompetitive Districts . . . . . . . . . viii Page 160 162 166 167 167 170 170 171 171 172 174 175 177 180 Table Page ' 4.27 Candidates' Policy Position By Perception of District Opinion Controlled By Risk and Expected Loss: Noncompetitive Districts. . . . . . . . . . . . . . . . . 182 4.28 Candidates' Policy Position By Perception of District Opinion Controlled By Expected Loss From Stands: Competitive Districts. . . . . . . . . . . . . . . . . 183 5.1 Policy Position By Perception of District Opinion Controlled By Incumbency and Election Outcome . . . . . . . . . . . . . 195 5.2 Policy Position By Perception of District Opinion Controlled For People Know Candi— dates' Stands and By Incumbency and Election Outcome . . . . . . . . . . . 197 5.3 Policy Position By Perception of District Opinion By Expected Loss From Stands and From Issue Areas Controlled By Incumbency and Election Outcome . . . . . . . . . . . 200 5.4 Policy Position By Perception of District Opinion By Risk (Noncompetitive Districts) Controlled By Incumbency and Election Outcome: Opposed Candidates . . . . . . . 203 5.5 Policy Position By Perception of District Opinion Controlled By District Competi- tiveness: Opposed Candidates . . . . . . 209 5.6 Policy Position By Perception of District Opinion Controlled By Risk For Noncompeti- tive Districts and By Incumbency for Competitive Districts: Opposed Candidates . . . . . . . . . . . . . . . . 210 l 5.7 Policy Position By Perception of District Opinion Controlled By Estimated Chance of Winning Election: Opposed Incumbents . . 211 5.8 Frequencies for Roll Call Scale Items: Foreign Affairs, Social Welfare, and Civil Rights . . . . . . . . . . . . . . . 215 5.9 Roll Call Scales By Perception of District Opinion and By Public Policy Position. . . 216 ix 'LIST OF FIGURES Figure Page 2.1 The Decision Problem . . . . . . . . . . . . . 59 2.2 A Uniform Probability Distribution of a Parameter e. . . . . . . . . . . . . . . . 65 2.3 A Spiked Probability Distribution of a Parameter e. . . . . . . . . . . . . . . . 66 2.4 A Subjective Probability Distribution of a Parameter e. . . . . . . . . . . . . . . . 69 2.5 A Decision Matrix. . . . . . . . . . . . . . . 72 2.6 A Decision Matrix Specifying Losses as Outcomes . . . . . . . . . . . . . . . . . 74 3.1 Paths of Constituency Influence. . . . . . . . 95 3.2 Revised Paths of Constituency Influence. . . . 97 5.1 Causal Relationship of Perception of District Opinion.to Public Policy Position and to Roll Ca11.Position . . . . . . . . . . . . 217 5.2 Gamma Values for Relationship of Perception of District Opinion to Public Policy Position and to Roll Call Position: Foreign Policy 217 5.3 Gamma Values for Relationship of Perception of District Opinion to Public Policy Position and to Roll Call Position: Social Welfare 218 5.4 Gamma Values for Relationship of Perception of District Opinion to Public Policy Position and to Roll Call Position: Civil Rights . 218 .0 .a v. s. .a 3.. .‘s ru i. 4‘ Vs .3 Hi. 4: s: A a A v A.» n v .s . A a Ft 6 \ 2. .1 -... L .. at y. s. a. s u. L.» Li. .s a n... a: h. .u e at. 40 .lu .nt .1- nu a» I it CHAPTER ONE RISK AND UNCERTAINTY IN POLITICAL DECISION MAKING 1. Introduction Political actors are often required to make a series of decisions, that is, to choose. Choice is defined as a basic social act that transforms the essentially private thoughts of individuals into "public activity."1 The act of decision by an individual is a manifestation of his desire to achieve a particular goal in the most efficient way he knows how. A person who is confronted with a set of choices will therefore try to make an optimal or rational decision; one that is "best“ for him. This implies that a rational decision may be different for different persons, depending upon how they evaluate the possible consequences of a decision. In addition, the rational decision may depend upon the decision criteria applied, since the use of objective or subjective probabilities can indicate different choices, neither of which can be considered "wrong“ or “irrational."2 A rational decision may also depend upon the relevant information that is available. There are consequently a number of parameters the values of which must be estimated in order to make an optimal choice. Since information about 1 ”I 2 these parameters will vary across a set of individuals, they must make choices under conditions of sisk and uncertainty.3 Decision making under uncertainty occurs if each action chosen leads to one of a set of possible specific outcomes, with each outcome occurring with a certain probability, which is unknown. When the probabi- lity is known or estimated by the decision maker, it is termed decision making under iisk. kisk in another sense means that there may be some expected loss as a result of choosing an action. Since there is uncertainty regarding the occurrence of these factors, the rationality of a decision maker can therefore be interpreted as the effi— cient use of contextual information so as to produce actions consistent, s priori, with his preferences. This study is concerned with the decision making of congressional candidates under similar conditions of risk and uncertainty and the extent to which their behavior coincides with that expected under the rubrics of a rational choice model. By employing the concepts asso- ciated with subjective decision making, a set of hypotheses are developed and tested to determine how candidates' per- ceptions of their environment determine the type of public policy positions they are likely to choose in order to attain their desired goal of winning political office. In this chapter, we will examine previous attempts to study the perceptions of political actors and try to show the significance that the present study has for the tn rf 5.1 I'" r.— () (D rf fl) er, development of a theory in political science. In the first section, the problem of the responsiveness of legislators to their constituents is discussed, and then a decision— theoretic model is examined to illustrate the importance of investigating the perceptions of political actors. The latter part of the chapter continues the literature review by outlining the empirical research that has been conducted to determine the relationship between perceptions and poli- tical behavior in various political settings. In the following chapter, a subjective decision making model, employing the concepts of Bayesian decision theory, is applied to the study of a congressional campaign in order to investigate the processes through which candidates decide on public policy positions. The results of this research will provide us with some empirical evidence about the effects of risk and uncertainty on the nature of represen- tative government in the United States. 2. Representatives and Responsiveness A basic process that is important in a democratic society is that of responsiveness. Responsiveness connotes a conscious and deliberate effort by a politican to match his decisions of public policy to the opinions of most people in his constituency.6 Responsiveness exists 1) when a politician perceives his constituency's opinions correctly and 2) when he acts in accordance with his per- ceptions of constituency opinion.7 l “a . '0‘: s'ng ovr‘ ”A. ‘5». uv-0 ‘ «as. | .-:~ This topic is an important concern because it involves the activities of individuals who are often elected to re- present us. Political decision makers elected to positions in government are likely to have as their goal the attain- ment and retainment of public office. In order to achieve their goal, they often rely upon a number of resources, one of which is the responsiveness of their policies to public wants and needs, or at least the appearance of responsive- ness.8 As Hershey argues: Even if voters do not force their representa- tives to be responsive, responsiveness may still occur. If leaders feel, rightly or wrongly, that the voters are determined to have their views represented, then learning and expressing public opinion may be seen as the road to public approval and reelection. This alternative rests on the plausibility of the idea that leaders feel voters are issue-oriented, even though we know most voters are not. Thus, the perceptions of political decision makers are strongly linked with one of the basic fundamentals of our democratic principles, that is the extent to which the representatives represent their constituents. Since not all representatives are likely to be exposed to exactly the same quantity and quality of information, and will undoubt- edly exhibit some bias in interpreting the information they receive, it becomes an important question as to the extent to which these governmental representatives try to fulfill their responsibilities. Also, if one is concerned about democratic theory and popular control in a democratic _government, then only through the study of perceptions can \I l*.' y. M. IV an n be; a. a: V3... Q» flu we determine whether in fact some responsiveness to public opinion does exist. The concept of responsiveness, however, may only go as far as the coincidence of politicians' policy positions with constituency opinion guarantees political rewards. If a politician does not believe people know anything or even care about a particular issue, then does it really matter what stand he takes? If we at least assume that politicians desire election to office, and that they will choose public policy positions that they believe will help them in attaining their goal, then we must investigate why they behave in certain ways, why some seem more responsive than others, and why less responsive ones are nevertheless successful. Jones10 appears to be on the right track in investi- gating this problem, for he asks if representatives even attempt to be responsive, or what factors are operating to facilitate or even discourage responsiveness. In a study of the 1969 Texas legislature, Jones examined the causal relationship between representatives' policy attitudes and roll call behavior, and between their perceptions of consti- tuency attitudes and roll call behavior. He found that the relationship between attitudes and roll call behavior was much more important than perceptions and roll call behavior. This was also the case for representatives from marginal districts, so we find that competitiveness does not facili- tate responsiveness. Only those representatives who adopted a. w. "l the role of a delegate, rather than of a trustee or poli- tico, appeared to follow their perceptions of constituency opinion, rather than their own attitudes.ll Responsiveness, according to this study, is based primarily on the particu— lar role orientation of the representatives. If responsiveness is based on the predisposition of the representative, the question is, to what extent is there likely to be popular control over public officials and their policies? In investigating this question, Sullivan and O'Connor listed four necessary conditions for a strong linkage between constituency attitudes and public policy:12 1. voters must perceive the issue positions of the candidates, 2. voters must cast their ballots on issue grounds, 3. opposing candidates must differ attitudinally on the issues, 4. winners must vote in accordance with their pre- election attitudes. The last two conditions are of particular significance, because without a difference among the candidates on the basis of their policy positions and some consistency between their campaign positions and roll call voting, there is certainly no overwhelming reason for the voters to become informed on the issues or vote on the basis of issues.13 In order to answer some basic questions about our representative democracy, we must examine the nature of the responsiveness of politicians to constituency opinion and EX ”A! by. Y. Qua 83'. “n b“ the degree of popular control over public policy. To do this, we need to fully understand the linkages between constituency opinion and public policy. While much research has been concentrated on whether citizens vote on the basis of issue content and perceived differences in the candidates,14 we have only an incomplete picture of how politicians perceive their constituents' opinions and how they decide on a public policy position. In this study we will be primarily focusing on the respon— siveness of candidates for congressional office and the extent to which there appears to be some popular control by the citizens over their leaders' policies. In this way, we can provide additional information about the relationship of constituency influence within a campaign environment to that of the legislative arena. Significantly, the study will be organized within a theoretical framework that will allow us to derive some lawlike generalizations about the nature of representative government in the United States. 3. Constituency Influence in a Legislature: A Theoretical View The political arena in which we can most readily measure the responsiveness of a politician is a legislature. In contrast to a politician in the executive branch, a legislator must often take clear cut stands on issues by answering series of roll call votes.15 Traditionally, his responsiveness to constituency opinion has been measured by his voting record. Along with constituency influence, other k n-“- An V.. «a- Us... 2e '5 2 v- N“- Vo- _ .— be.~ up? ‘0'. n p important variables, such as party influence, have been examined to determine their influence on congressional voting.16 Until recently, many of the findings of these studies were either contradictory or apparently so unrelated that the development of a theory of constituency influence based on these studies was precluded. One example of the kind of contradictory evidence that posed more questions than could be answered was that discovered by Miller and MacRae.l7 Miller found that representatives from safe districts were in greater agreement with their constituents than those from marginal districts. MacRae, on the other hand, found just the opposite; representatives from marginal districts were more in agreement with their constituents than those from safe districts. A resolution of this contradiction, as well as a breakthrough in studying constituency influence, has come from a model developed by Morris Fiorina. The model was a formal, deductive approach that utilized a Bayesian decision making framework.18 Basically, Fiorina presented a new approach to the problem of an elected representative choosing an optimal strategy of roll call voting to assure himself of a certain minimally acceptable level of the probability of reelection. By employing a decision-theoretic approach, he deduced hypotheses incorporating the parameter of a representative's perceptions about the state of the worldlg and was able to reorganize much of this seemingly contradictory evidence (n r‘f of In .1- In In LA 'ra' L4. about representative-constituency relationships in the literature into a coherent framework. Although the connec- tion between the formal model and the testing of its hypo- theses was not entirely a direct one, Fiorina offered several convincing arguments and additional data that helped support the hypotheses. Since Fiorina's model has contributed to the develop- ment of a theory of constituency influence, and has relied upon the concept of legislators' perceptions, it would be worthwhile to examine his approach at some length. This will help lay some of the groundwork for the present study, in which the perceptions of congressional candidates are investigated. In Fiorina's model, the states of the world were repre— sented by g, the degree to which a representative believed that a group in his constituency "cared" about the repre- sentative's vote on an issue, and might be drawn into the campaign, either in support or in opposition to the legis— lator. Two kinds of goals were postulated: l) to maximize the probability of winning or 2) to maintain a minimum acceptable level of the probability of winning. A “maxi- mizer' would then choose an alternative that yielded the largest expected increase in the probability of winning. A "maintainer“ would adopt an optimal voting strategy that is a discrete probability distribution over the set of alternatives, such that the expected value of voting over [Jilllllkl .A‘w- , 10 time would equal zero. This would result in a "break even" situation in which the legislator's personal probability of election would remain at an acceptable constant level.20 From these and other assumptions, Fiorina was able to deduce hypotheses about the best strategies for represen- tatives with either maximizing or maintaining goals. Our concern is what effect the subjective estimate, 9, of drawing a group into the next election campaign, had upon these strategies. Once this is determined, one can begin to see the importance that the estimation of this parameter has upon other types of political choices, including those outside the legislative environment. To begin, let us consider some of Fiorina's hypotheses. First, he found that in a homogeneous constituency a legis- lator should not vote any differently when his estimate of g was high, than when it was low. That is, his perception of 9 made no difference in his voting decision, assuming rationality. For this case, the maximizer had a dominant strategy and the maintainer had a mixed strategy over the set of possible alternatives.21 In the heterogeneous case, with two groups in the constituency with diametrically opposed policy preferences, thenwximizerhadrundominant strategy when his perception of E for both groups was equal (91 = 92). In fact, all the strategies of the maximizer were equivalent in their leading to a decline in his subjective probability of reelection. In this situation, the maximizer would lose "a. 'U. I". in (I. ‘ov J “v ,.. .Iab 4'... a bead w-s «a. . n2. EX I a $4.. 9: (Y: ta '0 (D 50 11 votes no matter what his choice. In contrast, a maintainer was required to vote for the preferences of the stronger group to maintain a certain probability of reelection. These results lead to Fiorina's Proposition 10, which stated that in cases when 91 = 92, a representative's voting does not vary with his estimate of g 1 In the general heterogeneous case, the estimate of El 22 and 92. and 92 did make a difference. When El # 92 was assumed, the existence of an optimal voting strategy for the maintainer was postulated and it was discovered that the maximizer and the maintainer may be led to vote predominantly with the weaker group.23 The following propositions derived by Fiorina are stated below in order to summarize as well as emphasize the significance of the subjective estimate of g in a represen- tative's voting decision. We see that especially in the general case of 91 f 92 that the estimate of Q can have some nonobvious consequences: Proposition 17: Ceteris paribus, with a two-group const1tuency, as the maximizer raises his estimate that the stronger group cares, the likeli- hood that he votes with them increases. Proposition 18: Ceteris paribus, with a two-group constituency, as the maximizer raises his estimate that the weaker group cares, the likelihood that he votes with them increases. Proposition 19: Ceteris paribus, with a two-group const1tuency, as a maintainer raises his estimate, C , that the stronger group cares, he minimum probability, Q, that he must vote Ivar; .5 n3 .QIA in“ c 0. 12 with them, decreases, subject to the provision that no change in his optimal strategy occurs. Proposition 20: Ceteris paribus, with a two-group const1tuency, as a maintainer raises his estimate, C , that the weaker group cares, the minimum probability (1‘- Q), that he must vote with them decreases, subject to the proviso that no change in optimal strategy occurs. 4 The first two hypotheses may not be surprising, but Propositions l9 and 20 may seem nonobvious. When a group in a maintainer's district increases its interest in the representative's position on an issue, the propositions predict that the chances of his voting with them on the issue declines. Although this may be part of a maintainer's reelection strategy, he may nevertheless find it difficult to increase his probability of reelection or maintain it at an acceptable level unless he can accurately assess the value of E for each important group in his district. If he misperceives whether a group cares or not, he may act in a manner that is rational, but not acceptable as far as the implications of his decision are concerned. The study of a political decision maker's perceptions of his political environment is therefore an interesting question, since his actions may not be understood except within the framework of a positive theory. Such a study is also interesting in a normative sense as well, for we are interested in the responsiveness of representatives to the majority will and the extent to which they live up to the ideals of a democratic society,25 as expressed by Burke, :A' .U. 13 for example: ...it ought to be the happiness and glory of a representative to live in the strictist union, the closest correspondence, and the most unreserved communication with his consti- tuents. Their wishes ought to have great weight with him; their opinions high respect; their business unremitted attention. It is his duty to sacrifice his repose, his pleasure, his satisfactions, to theirs--and above all, ever, and in all cases, to prefer their interests to his own. Some attention will be paid to a comparison of the empirical results of this study with a normative interpretation of representation. Of particular significance is the fact that this research will extend the traditional analysis of the link- ages between citizen and representative by moving beyond the legislative forum to the campaign environment. Politi- cians may have to make a number of policy decisions that can be affected by the parameters of their environment within which they may be competing for another term of office. .Risk and uncertainty play an important part in the nature of these constraints on decision making and it is important to know how much they can affect the nature of representa- tion. Since nonincumbent challengers occasionally defeat an incumbent seeking an additional term, it would also be worthWhiletxastudy how risk and uncertainty affect them as well as the much-observed incumbent legislators. In this sway, we can begin to understand representation as a process lfather than as a product of the electoral system. To do 14 this, we must study how politicians develop their attitudes and formulate their perceptions of constituency opinion. Since the process is dynamic, there is also an opportunity for a politician to revise his perceptions and thus his policy positions. If we can identify the important inde- pendent variables which alter candidates' perceptions then we will not only be closer to explaining and predicting the kinds of public policy choices that are made, but also closer to influencing these decisions according to the normative values of our society. The approach that is well suited for this kind of analysis is Bayesian decision making, which employs subjec— tive estimates of the world,27 that can be continuously revised on the basis of new sample information to produce a "best" estimate of a parameter. Since a candidate's perceptions are actually based on prior knowledge and con- tinually revised when new information is received, there is a good likelihood that the techniques of Bayesian decision making can be applied to the study of political choice in a campaign environment. In order to proceed in the development of this analysis, the next section covers the important empirical research that has involved the study of the perceptions of political actors and the effect of their perceptions on their political behavior. .P» a. van .‘u 1AM A\v \H‘ AL 15 4. Political Perceptions: A Survey of the Art Although the Fiorina model seems adequate in formali- zing the nature of representatives' choices and decisions in a legislative environment, it falls short of explaining the behavior of an incumbent in a political campaign. In the legislature, a representative's only opponent is him- self; his own actions will help raise or lower his proba- bility of reelection. A political campaign, however, intro- duces additional factors over which the candidate may have little control, including the activities of his opponent. In addition, a bad decision or a mistake in judgment during the campaign can have a very critical effect on the outcome of the election, perhaps more than anything the representa- tive had done during his entire tenure in office. No matter how careful or cautious he may be, one critical error could possibly deflate the candidate's probability of reelection to a totally unacceptable level. During the campaign, public awareness of the candidates is likely to be relatively high and it is a time when a candidate's actions may be closely scrutinized by certain groups and individuals in the district. It is therefore essential that he have not only a good perception of which rgroups care or don't care, but also a good estimate of the probability that other parameters important to the campaign ‘will attain certain critical values necessary for winning the election. Fiorina's model may be important in explain- ing legislative voting influences, but it does not consider r1 In (D ’U L... 16 the important variables related to campaigning and how they may affect the candidates' policy positions which may be carried over in some form to the legislative arena. There are constraints in legislative decision making, but the constraints in a campaign may be even more serious, espe- cially if they affect the spatial mobility of the candidates and their ability to meet the challenges of their opponents?8 The problem then becomes a question of reconciling the difference between what is promised during a campaign and what limited policy options are available in the legislature. Occasionally, a legislator must vote for a bill that does not represent his most preferred policy position rather than face the prospect of no bill at all. It is therefore important to study the perceptions of the candidates in an election campaign in order to gain some empirical knowledge about candidate choices that can provide the basis for the development of a model of campaign deci- sion making and a more complete explanation of the process of representation. As Kingdon explained: A full account of representation, therefore, must include representatives' perceptions of the constituents as a variable intervening between the constituents and the behavior of the elected policy-maker. These perceptions may or may not be accurate, but it is necessary to take them into account in order to explain the behavior of the politician.29 In this section we will review some of the important efforts that have been made in the study of candidates‘ perceptions and evaluate them in light of their potential for contributing to the development of hypotheses that can n ..Il. . 17 be used to help explain candidates' behavior. One of the first major studies that dealt with the problem of uncertainty in political campaigning was Kingdon's survey of a sample of Wisconsin candidates in the 1964 election.30 Despite the fact that his sample was small (N = 66), which precluded the use of significance tests, it nevertheless provides us with some insightful information that can be used to make comparisons with some later studies.31 Kingdon was concerned with the differences among poli- ticians and how their images of their constituency and beliefs about voters affected their roll call voting, public policy stands, or campaign strategies. The research provides us with information regarding each of the following areas: 1) how candidates receive information, 2) how they interpret information, and 3) how they respond to this information. These topics will be covered in each of the following sections. A. Informing the Ignorant - The first category entails an investigation into how candidates receive information about the electoral environment and how they use this infor- mation to develop a good estimate of the true state of the world. Knowing the reliability of a particular source of information is very valuable to a candidate who wants to keep in touch with reality. Too often political decision makers are told only what they are thought to want to hear, for, as Lewis Dexter has found, “...a congressman hears most 18 often from those who agree with him."32 Therefore, in order to evaluate the opinion of the electorate, and to form an optimal campaign strategy, a candidate must rely upon a number of independent sources. Because of a number of con- straints on resources that are available, he has to learn which sources are available and at what cost, as well as the utility of such information. Kingdon asked the following question of his sample: "Can you tell me your sources of information about how you will do when the votes are in?" They were also asked which of these resources they relied most heavily upon. The res- ponses most often mentioned were 1) polls, 2) party people, 3) volunteers, 4) past statistics, and 5) warmth of reception.33 The reliability of the polls, which were used primarily in the upper level races (congressional.and state), was usually accepted without qualification. Party people and volunteers were frequently used as sources, but were never- theless not relied upon as accurate sources of information. Candidates used and relied most heavily on past election statistics and the warmth of reception they received from their constituents. Some candidates were necessarily cautious about placing too much reliability on any one source, especially since factors outside their control could have influenced the election, such as the effect of national and other state contests, as well as the vagaries of predicting voter turnout at the polls. my 19 The level of the office sought reflected the extent to which the candidates relied upon sophisticated campaign techniques such as polling. State legislators felt that voters were not paying as much attention to them as to the candidates for Congress. Their campaigns had less of a direct, personal appeal to the voters, but a greater appeal to interest groups. Congressional candidates, on the other hand, were more likely to take account of the media and voter reaction to their public policy positions, since they were more likely to believe their issue positions were well 34 known and would make a difference at the polls. B. Rationalizing Reality — Although candidates may have a number of sources of information in which they have varying degrees of confidence, there may be certain biases created by the ambiance of the campaign or are inherent in the psychological drama of politics. In this regard, a strong distinction can be made between the attitudes of winning and losing candidates. Since most of Kingdon's interviews were conducted after the election, the state of mind of the candidates may have been affected by the results. In addition, the fact that one person was an incumbent and another a challenger could have subjectively affected their perceptions of the campaign. Indeed, this distinction between winners and losers was confirmed by Kingdon's research. He found that losers developed rationalizations for defeat, which showed up in a number of their beliefs, whereas winners tended to pp 20 congratulate themselves and the district for having such good judgment.35 Winners ranked the importance of their personal characteristics high as well as the importance of issues in the campaign. The losers felt that the party label was too much of a factor, and that the voters were not informed on the issues.36 Further study in this area has been conducted by Kim and Racheter, who attempted to test Kingdon's "congratula- tion-rationalization“ hypothesis, which states that winners tend to develop complimentary beliefs about voters, while losers tend to rationalize their defeats by downgrading voters' competence.37 Kingdon could not successfully test this hypothesis himself, because he could not compare the attitudes of the winners amilosersbefore the election with their attitudes after the election. In contrast, Kim and Racheter conducted a pre- and post-election survey to deter- mine the attitudes of candidates in the 1970 Iowa general election.38 Using the election outcome as the independent variable, they found that winners did not develop more complimentary beliefs, nor did the losers develop more pejorative beliefs about voters. However, considering the original cognitive dissonance states as the major independent variable, among the winners, the “high dissonance“ group (the candidates who had unfavorable beliefs about the electorate at the outset of the campaign) tended to upgrade their beliefs more than the “lbw dissonance" group (the candidates who had favorable 21‘ t 7' ‘J; (n 011 21 beliefs about the electorate at the outset of the campaign). Among the losers, the "high dissonance" group tended to downgrade the voters more than those of the "low dissonance" group. The "congratulation—rationalization" effect was supported only after a refinement of the original analysis, but it points out that candidates' perceptions are influ- enced by their original perception of the election outcome?9 Another study which illustrates the importance of attitudes on perceptions was conducted by Hedlund and Friesema.40 As part of the Iowa Legislative Research Pro- ject, they interviewed members of the 1967 legislature and asked the subjects to predict their own district's majority preferences on four constitutional amendments that were on the ballot. The purpose was to determine whether those representatives who adopted various representational roles differed in perceiving and responding to constituency opinion.41 Out of 181 legislators, one-third predicted the results for all four amendments and another one-third correctly pre- dicted the results for three of the four issues. Four legislators failed to predict any of the results, and the difference in the prediction rate between the two legisla- tive chambers was not significant.42 When considering the role orientations of the legisla- tors, however, significant differences were apparent.43 It was discovered that in the Iowa legislature, delegates were least able to predict their districts' responses, which was 22 astonishing, because it was precisely this group that felt strongly about following their own districts' opinions as closely as possible. The second most accurate group was the one composed of politicos, while the most accurate represen- tatives were the trustees. Even on questions designed to assess the different perceptions of the legislators toward their constituents, the same relative orderings were found? Those who won reelection did better at predicting three out of four of the results, but a higher percentage of those defeated were correct on all four (31% vs. 27.8%). Inter- estingly, those who did not run for another term for reasons of retirement, primary defeat, or other reasons, made the best predictors (40.5% correct on all four issues).45 Given the interpretation of role orientation, these results were quite surprising, since we would normally expect the dele- gates to be most accurate in assessing district opinion. Hedlund and Friesema found that those candidates who would most likely be in close contact with their consti- tuents were not usually more correct in assessing the ballot results. This could have been because they were either following their own opinions or there was a great amount of uncertainty about the results. Unfortunately, they did not control for the margins of the winning propositions, nor did they control for other factors that could indicate the degree of uncertainty in perceiving the majority attitudes in the districts.46 Clearly, there is a need to relate this kind of phenomenon within some kind of theoretical framework 5 23 in order to explain these results. In a similar study, Erikson, Luttbeg, and Holloway tried to assess the accuracy of Florida state legislators in predicting the results of a statewide referendum.47 They improved upon the previous study by asking the legislators to predict the actual percentage of the vote for each of the three issues on the ballot. This was especially useful, since nearly all legislators correctly predicted the major- ity position on each issue, thus, a comparison of the legis- lators' average estimates could be made.48 Veteran legislators (those with two prior two-year terms) and delegates were the least accurate assessors of constituency opinion, while the junior legislators and trustees had the smallest error of prediction, following Friesema and Hedlund's findings that those who claimed to be most concerned about assessing constituency opinion were the least able to do so accurately.49 We emphasize that the immediate question is not really the accuracy of perception, but why different types of legis- lators have different perceptions and how these were developed. Role orientation and years in office appear to be important factors in influencing behavior, but the prob- lem is that we have so few well confirmed hypotheses that can assist us in building a concatenated theory to explain legislative behavior. Such a theory could help us to explain why these factors are important and how the per- ceptions of decision makers affect the public policy 24 decisions they make. Another study which provided insight into the formation of candidates' perceptions was conducted by Bullock.50 The purpose of the study was to investigate candidate evalua- tions of the effects of party identification, presidential coattails, candidate personalities, issues, and redistrict- ing on their election chances. The data were based on questionnaires sent to challengers and incumbents from 30 congressional districts. This was the first major study since the 1958 Representation Study to compare the percep- tions of candidates from the same electoral districts.51 While many of the hypotheses tested by Bullock were compar- able to those tested by Kingdon, Bullock's study had the advantage of being able to place the responses of the can— didates side by side on a district level of analysis.52 Concerning the election outcome, Bullock found that incumbents were more likely to be pessimistic, while the challengers were more likely to be optimistic about the results. In addition, more incumbents than challengers accurately predicted the outcome.53 Bullock was also concerned with the effect of presi- dential coattails on the perceptions of the candidates. The perceived partisan implications of the national election were hypothesized to have an effect on the reasons for the election or defeat of the candidate. He tested three hypo- theses about the effects of the 1972 election: 1) few freshmen or challengers will report being aided by the 25 performances of their party's presidential nominee, 2) congressmen initially elected in an off-year may have a greater sense of independence from the futures of the party and its presidential nominees than do congressmen initially elected on the ticket with a president of their own party, and 3) Democrats will tend to believe they were harmed by the presidential race and Republicans will generally deny that the presidential election had any effect.54 The findings indicated that most of the incumbents and challengers perceived that indeed there was a coattail effect and that challengers were more likely to report a coattail effect than incumbents. Also, Democrats more often than Republicans perceived that the presidential race affected their election.55 The coattail effect was seen as a possible excuse for blaming the electoral outcome on sources beyond the control of the candidate. Consistent with previous research, win- ners were more likely to attribute their success to factors within their control, while losers blamed factors beyond their control. The difference between incumbents and challengers was again illustrated by the difference in their perceptions of the effect of redistricting on their election chances. Incumbents (and in this case, winners) were less likely to believe that redistricting made a difference in the election than the challengers.56 Incumbents instead emphasized their personalities and other factors amenable to their influence, it! 26 such as issues, as contributing to their success.57 A definite inverse relationship was found between incum- bents and challengers in their ordering of election factors. Generally, the greatest agreement on the rankings came from urban areas, while in the Midwest there was strong disagree- ment, stressing that perhaps district compactness or more sophisticated techniques employed by candidates in the urban areas influenced the results.58 The distinction between winners and losers and incum- bents and challengers is significant beyond the campaign environment, because candidates who believe that they were elected on the basis of issues are likely to ascribe an unrealistic amount of political interest and knowledge to the voters.59 Winners who then become legislators are thus prompted to become delegates and pay undue attention to constituent desires, even though their constituents may know very little about them or what they are doing. The Bullock study was important because it helped to verify some of Kingdon's earlier findings and presented comparisons of incumbents and challengers on the district 60 The main focus was on the level for the first time. difference in the perceptions of the incumbents (winners) and challengers (losers) and how they saw various factors affecting the electoral outcome. The value of the study, however, was limited in that some hypotheses that were tested could only be applied to the unique electoral situa- tion of the 1972 election. Also, the analysis did not 27 follow through by controlling for important independent variables which were found to be significant in other studies of candidates' perceptions. Additional research is needed to assess what these differences mean in terms of public policy formation and how these findings can fit into an overall conceptual framework of candidate behavior under conditions of risk and uncertainty. C. Overcoming Uncertainiy - The third category of major concern is the overall effect of uncertainty on candi- dates' behavior. After they have collected and evaluated information, they may form an estimate of what would be the best course of action, given a certain degree of confidence in their evaluation of the possible states of the world. That is, although they may have acquired a certain amount of information and assessed its implications, the candidates may not always be quite certain that their knowledge about campaign parameters is accurate. As a result, candidates may try to adopt a rational strategy in order to minimize their potential losses. Kingdon, for example, suggested that one of the main factors behind many of the actions of the candidates in his Wisconsin study was uncertainty: His [the candidate's] belief about his election chances probably influences his decision to enter the race in the first place, and may influence the conduct of his campaign and even his behav1or 1n off1ce, should he w1n. Candidates may even seek to perpetuate an aura of uncertainty within their own campaign organization in order 9 u . .Aa. 6m .94 A v “Us V . HM: Lt. 28 to keep their supporters from becoming too overconfident.62 This is theoretically significant in that it can affect the formation of minimum winning coalitions. According to Riker's theory of political coalitions, the greater the degree of uncertainty about a game's outcome, the larger the coalition that must be built.63 Likewise, Kingdon hypothe- sized that in the context of his study, the greater the politician's uncertainty about his election chances, the greater would be his efforts to enlarge his coalition by 64 Therefore, the inves— attracting more groups and voters. tigation of the importance of uncertainty and the means by which politicians cope with it is important in determining the characteristics of campaign behavior, as well as legis- lative behavior. One recent study that has gone beyond Kingdon's inves- tigation of the effects of uncertainty on a political 65 Although primarily campaign was conducted by Hershey. concerned with the impact of perceptions on the choice of campaign strategies, she found some interesting results in the areas related to this present discussion of uncertainty. Unlike other studies, Hershey interviewed not only candidates for congressional and statewide office, but also the campaign managers of the candidates as well.66 Alto- gether, 57 individuals were interviewed: 28 candidates and 29 managers. Unfortunately, once again we find the sample too small to allow the use of significance tests and the lack of a random sample casts some doubt on the al ‘1." (7 ! t5 :5 ‘- L 29 representativeness of the results. Despite these drawbacks, a number of significant findings about the effect of uncertainty on campaign behavior were developed and warrant further study. These important conclusions include the following: 1. Uncertain managers, and to a lesser extent, uncertain candidates, were more likely to seek voters' opinions and information about the campaign than those who were sure.57 2. Uncertainty was likely to heighten the campaigners' attention to other groups. 3. Uncertainty about the election result stimu- lates campaigners to seek out public opinion, both from voters and contributors.69 4. Incumbents were less active than challengers in determining public opinion. 5. Challengers were more likely to use public opinion polls, while incumbents were more likely to ask party leaders about the campaign.71 These results serve to illustrate that uncertainty about the political environment is an important factor in the development of strategies and the choice of campaign activities. As Hershey's findings show, under certain cir- cumstances, candidates may try to seek additional informa- tion and be more aggressive in their campaigning. However, more research is needed in order to determine how uncer- tainty in this context affects the selection of a public policy position. D. Risky Choices - Another factor that is important in.decision making is the estimation of the possible loss 'that:could occur from adopting a particular chOice of action. (A; ()4 II 30 With uncertainty, one is not sure whether even the best decision will result in an outcome that minimizes expected loss. The perception of the degree of risk present in a decision making context may thus influence the desire for more information or the search for safer alternatives. Although the term iisk_has more than one interpretation in the formal decision making literature, in political science it has been most effectively used in the study of political ambition.72 As Schlesinger notes in his book, Ambition and Politics: Nevertheless, there is structure to the oppor- tunities for political office in the United States. Any elective system of opportunities is full of risks for the politician. But if we look at the American system from the stand- point of ambitions we can see that the risks tend to foster some ambitions and reduce others. The risks for those with progressive ambitions are not equally distributed among officeholders. Career risks are maximized in a situation in which, in order to seek a higher office, a man must give up his current office. The congress- man who reaches for the Senate and fails loses everything...73 The loss involved in such decisions would include the possible loss of the office presently held, unless a politi- cian was fortunate enough to be able to hold on to one office while running for a higher one. Black has extended this interpretation of risk by defining it in terms of the investment that would be lost if a candidate were to lose an election.74 'Black described an election as a "...risk taking venture in which candidates .are forced to wager a portion of their resources in the IPursuit of office."75 The magnitude of the risk is ‘3: (1) (r) n 0. ’t1 (7‘ 31 determined by the structural characteristics of the elec- toral system, such as the size of the unit and the compe- titiveness of its election. He hypothesized that the risk of running for office is an increasing function of the size and the degree of electoral competition in the unit. Thus, as the size of the investment (risk) increases, the less committed individuals would be the ones most likely to drop out of the race:76 Besides indicating who is likely to run for office, Black's conceptual scheme could also be applied to an Officeholder who must decide whether to run for reelection or for higher office. He hypothesized that "...the greater the cumulative investment of the individual in political office seeking, the greater will be the value placed on the offices to which the individual might aspire.“77 The invest- ments made while holding an office were presumed to be transferable to other offices. The data he collected on 435 city councilman in the San Francisco Bay area seemed to confirm his hypothesis. Also of interest was that he found that uncertainty played a role in the councilmen's aspirations. Those who were certain of winning their reelection bids were also the ones more likely to aspire to higher office. The logic ibehind this presumes that the probabilities of obtaining ‘various offices are interdependent and that winning one :race increases the probability of winning a race for another off ice . 7 8 32 It should also be noted that Black was concerned with subjective probabilities, based on the perceptions of the politicians, rather than on a series of experimental trials. They represented the state of belief of the politicians at a particular time and their perceptions could easily change on the basis of new information. This study provides addi- tional conceptual support for the study of how candidates' perceptions of risk and uncertainty can affect their poli- tical choices. Rohde has also investigated the effect of risk on 79 His use of the term risk, however, political ambition. takes on a different meaning.~ Instead of considering risk as the expected loss of an investment, as Black does, Rohde was concerned with candidates' attitudes toward risk and how they might affect their decision to run for higher office. In formal terms, a candidate's attitude toward risk depends on the shape of his utility function: one with a concave utility function is risk-averse and prefers the choice that leads to an outcome with certainty; one with a convex utility function is risk-acceptant and prefers the choice that leads to an outcome with a certain probability; one with a linear utility function is risk-neutral and is indifferent between the choice that leads to an outcome with certainty and one that leads to an outcome with a certain probability. 80 Rohde's hypothesis, based on these propensities toward rflisk, was stated as follows: 33 Hypothesis 8: If two House members are presented with similar opportunities to seek higher office, and one is a "risk- taker" and the other is not, then the "risk-taker" will have a greater probability of running for higher office than the other.81 In testing his hypothesis among candidates for congressional, senatorial, and gubernatorial offices, he found that "risk- takers are more likely to take an opportunity to run for higher office than are others, and that in most comparisons the difference is substantial."82 From these studies we know something about the impor- tance of risk as expected loss and as a characteristic describing the nature of a candidate's utility function. The dependent variable has been the decision to run for a higher office. However, the use of the concept of risk has not been widespread in the political science literature and it has not been used to explain the choice of a public policy position, except in purely theoretical models.83 There is a need to relate the impact of risk within a model that assumes that less than perfect information is available (uncertainty) to an empirical situation. In this way, we could begin to test our theoretical models and discover how candidates make their choices and estimate the true value of the parameters of the campaign environment. 5. A Need For Theory In this chapter, we have attempted to establish the ‘utility of studying the perceptions of political actors in 34 order to explain the nature of representation in the United States. we also showed that two important factors, risk and uncertainty, influenced one's perceptions of the political environment. As a result of this discussion, we can proceed to try to advance our understanding of the causal linkages involved in the process of representation by developing a political decision making model that accounts for these concepts. The development of a model in this area would be important because it would help give us an understanding of the relational structure of these theoretical concepts. It would also provide additional precision and clarity of thought which could be used to help summarize the data in terms of the parameters of the model. Consequently, the model would be valuable as a means of exploration and dis- covery. For example, if the data are reliable, and if the calculus of the model describes the real relations existing in the subject matter being described, the theorems of the calculus will yield important inferences from the data. The discrepancies that are found can then be used as a kind of base line for further inquiry.84 This approach was effectively utilized by Fiorina in his model of constituency influence. For example, he was not only able to identify contradictions in previous research, but was able to deduce a set of hypotheses that included some nonobvious explanations of political behavior. The model was also able to identify gaps in our knowledge 35 and specify further paths of research.85 Although Fiorina's framework was a useful tool for discovery and illustrated a way in which subjective decision making concepts could be applied to the study of legislative voting behavior, we argued that there was a need to examine decision making in a political campaign in addition to the legislative environment. The two situations are not alike, but similar kinds of policy decisions must be made in both arenas. Therefore, it would be desirable to examine sub- jective decision making within a campaign and formulate hypotheses that describe the relationship among the rele- vant variables. In this way, we can proceed at a later time to develop a larger framework that could adequately describe the interrelationships that may exist between campaign (electoral) and legislative decision making. The latter part of this chapter included a review of the literature that served to identify some of the important variables that could be included in a campaign decision making model. For example, the importance of information and the reliance placed on various sources was illustrated by Kingdon in his study of political candidates in a Wis- consin election. This kind of research can help explain the kinds of biases that may influence the perception of political information. Research has also established that a number of inde- pendent variables can also influence the perceptions of the candidates and thus their decisions. Winning and losing 6", 55“ Cate for § tor 36 candidates were likely to have different perceptions of their electorate as well as the incumbents and challengers. Thus the perceived chances of winning, as well as the elec- toral outcome itself, were shown to be important influences on candidates' behavior. Also, the role orientations adopted by the legislators were found to be related to their per- ceptions of their constituents' preferences, although some of the findings in this area have not been conclusive. One of the two primary independent variables of this study is uncertainty. A review of the literature, especially the work of Kingdon and Hershey, has shown the significance that uncertainty has for a campaign. Their research indi- cated that uncertainty was likely to influence the search for additional information and heighten the attention paid to groups in one's district. The other primary independent variable is risk. Since the concept of risk has been interpreted in a number of ways, its inclusion within a formal model will help clarify its meaning. To summarize, risk has been.frequently used in each of the following contexts: l) to indicate the subjective probability of an outcome, given the selection of a particu- lar course of action, 2) to measure the expected losses in making a decision, such as a loss of votes or a loss of an investment, and 3) as a description of the choice environ- ment, in which the decision makers may have utility curves that are either linear, concave, or convex. In the first ssase, risk becomes an important consideration in estimating 37 the probability of a particular state of nature. In the second, risk can be interpreted as an important component in a decision calculus, and in the third instance, it can determine the choice of an individual when faced with the choice between an outcome occurring with certainty and others occurring with a known probability. We have thus laid the basic foundation for the develop- ment of a subjective decision making model by indicating the nature of the problem and the need for additional research. In the next chapter, we will relate these concepts within a theoretical framework that is an interpretation of Bayesian decision theory. This will allow us to deduce hypotheses that can help explain and predict the dependent variable of this study, the public policy choices of Candidates in congressional elections. These hypotheses will primarily relate the degree of perceived uncertainty and risk to the selection of a public policy position and enable us to determine the extent to which responsive representation is likely to occur under these conditions. CHAPTER ONE NOTES 1. Kenneth A. Shepsle, "The Strategy of Ambiguity: Uncertainty and Electoral Competition," AmeriCan Political Science Review, 66 (1972), 55. 2. Wayne Lee, Decision Theory and Human BehaviOr (New York: John Wiley, 1971), Chapter 1. 3. The terms risk and uncertainty have a number of interpretations in the decision making literature. Some writers make a sharp distinction between risk and uncer- tainty, while others, such as Shepsle, identify uncertainty as a special case of risk. See Kenneth A. Shepsle, "Essays in the Theory of Risk-Taking" (Unpublished Ph.D. Disserta- tion, University of Rochester, 1970), Chapter 1. In addi- tion, a special use of the term risk appears in the Bayesian decision theory literature. In this case, risk refers to expected opportunity losses. See Morris Hamburg, StatiSti- cal Analysis for Decision Making (New York: Harcourt, Brace and WOrld, Inc., 1970), Chapters 14 and 15, and Alexander M. .Mood and Franklin A. Graybill, Introduction to the Theory of I§tatistics, Second Edition (New York: McGraw-Hill, 1963), :pp. 165-167. These definitions will be discussed in more detail in Chapter 2. 38 39 4. Shepsle, "The Strategy of Ambiguity," op. cit., 5. By developing a theoretical framework that does not have to depend upon the strict assumption of perfect infor- mation, as many rational choice models do, we can more realistically determine the nature of choices within a charged political environment, such as a campaign, where intuition and experience are often the bases for political action. For a discussion of the assumptions of spatial models of electoral competition and other rational choice models, see William H. Riker and Peter C. Ordeshook, Introduction to Positive Political Theory (Englewood Cliffs, New Jersey: Prentice-Hall, 1973), Chapters 11 and 12. 6. Bryan D. Jones, "Competitiveness, Role Orienta- tions, and Legislative Responsiveness," Journal of Politics, 35 (1973), 925. 7. Ibid. 8. Marjorie Randon Hershey, The Making_of Campaign Strategy (Lexington, Mass.: Lexington Books, 1974), p. 19. 9. Ibid., p. 20. 10. Jones, op. cit. 11. The concept of representational role orientations ‘was developed by Wahlke et al. in The Legislative System (New York: Wiley, 1962). In this comparative study of state legislatures, a delegate was a representative who felt a.need to consult his constituents and to usually follow 'their judgment. A trustee was a legislator who viewed his 'U 1.. 40 role as essentially a free agent and could follow his own beliefs in deciding how to vote. A politico was a type of representative who adopted a role somewhat characteristic of both types. See also Wayne Francis, "The Role Concept in Legislatures: A Probability Model and a Note on Cognitive Structure," Journal of Politics, 27 (1965), 567-585. 12. John L. Sullivan and Robert E. O'Connor, "Electoral Choice and Popular Control of Public Policy," American Political Science Review, 66 (1972), 1256-1257. 13. Ibid. 14. See Gerald M. Pomper, "From Confusion to Clarity: Issues and American Voters, 1956-1968," American Political Science Review, 66 (1972), 415-429, Benjamin Page and Richard Brody, "Policy Voting and the Electoral Process: The Vietnam war Issue," American Political Science Review, 66 (1972), 979—995, and David RePass, "Issue Salience and Party Choice," American Political Science Review, 65 (1971), 389-440. 15. Under certain circumstances, a legislator may choose not to vote at all. For a discussion of the absten- tion strategy, see Morris P. Fiorina, Representatives, Roll Calls, and Constituencies (Lexington, Mass.: Lexington Books, 1974), Chapter 4. 16. See, for example, Duncan MacRae, Jr., Dimensions of .Qpngressional Voting (Berkeley: The University of California IPress,_1958), Lewis Froman, Congressmen and Their Consti- tu’encies (Chicago: Rand McNally, 1963), Julius Turner and 41 Edward Schneier, Party and Constituency: Pressures on Congress (Baltimore: Johns HOpkins Press, rev. ed., 1970), Wayne W. Shannon, Pariy,=Csnstituency and Congressional Voting (Baton Rouge: Louisiana State University Press, 1968), Jeff Fishel, Repsesentation'andikesponsiveneSS'in Congress: The Class of Eighty-Nine, 1965-1970 (Beverly Hills: Sage Publications, 1973), Lewis Froman, "Inter-Party Con— stituency Differences and Congressional Voting Behavior," American Political Science Review, 57 (1963), 57-61, Duncan MacRae, Jr., "The Relations Between Roll-Call Votes and Constituencies in the Massachusetts House of Representatives," American Political Science Review, 46 (1952), 1046-1055, Hugh LeBlanc, "Voting in State Senates: Party and Consti- tuency Influences," Midwest Journal oikPolitical'Science, 13 (1969), 33-57, Wilder Crane, “Do Representatives RepreSent?" Journal of Politics, 22 (1960), 295-299, and Warren E. Miller and Donald E. Stokes, "Constituency Influence in Congress," American Political Science Review, 57 (1963), 45-56. 17. Warren E. Miller, "Majority Rule and the Representa- tive System of Government," in Mass Politics, ed. by E. Allardt and Stein Rokkan (New York: Free Press, 1970), pp. 284-311 and MacRae, Jr., op. cit. For a discussion of the contradictions between the findings of these two studies, see Fiorina, op. cit., Chapter 1. 18. Ibid. Fiorina found that the problem lay in a Inistaken inference by MacRae. See also Fiorina's "Electoral Margins, Constituency Influence, and Policy Moderation," “9 II“. b h‘ 42 American Politics Quarterly, 1 (1973), 479-498. 19. A Bayesian analysis involves the selection of an optimal strategy under conditions of uncertainty or risk. In this case, the individual is presumed to be uncertain about which state of a specified world is the true state which obtains. In Savage's terminology, world is defined as "the object about which the person is concerned," and a state 2: the world is "a description of the world, leaving no relevant aspect undefined." For example, the state of the world might be the true probability of reelection. See Leonard Savage, The Foundations of Statistics (New York: Dover Press, rev. ed., 1970). P. 9. 20. Fiorina, Representatives, Roll Calls, and Consti- tuencies, op. cit., Chapter 2. 21. Ibid., Chapter 3. A district is assumed to be homogeneous with respect to a particular vote on an issue. The constituency, as perceived by a legislator, may only refer to a small portion of the absolute constituency, but these will be the individuals who will be the most important to consider when deciding how to vote. 22. The propositions are not formally stated in his book but are to be found in his dissertation, "Representa- tives and Their Constituencies: A Decision-Theoretic .Analysis" (Unpublished Ph.D. Dissertation, University of lRochester, 1973). The strength of a group is defined as the net increment 111 the probability of reelection that a group has the power t0 and 1. .Le; 43 to effectuate. 23. Ibid., p. 124. 24. Ibid., pp. 126-128. 25. Sources for normative theories of political repre- sentation include the following: Hanna F. Pitkin, Tks Concept of Representation (Berkeley: University of Califor- nia Press, 1967), J. Roland Pennock, "Political Representa- tion, An Overview,“ in Representgtion: Nomos X, ed. by Pennock and John W. Chapman (New York: Atherton Press, 1968), and Alfred de Grazia, Public and Republic (New York: Alfred University Press, 1961). For a discussion of the normative aspects of representation in terms of reapportionment in the United States, see Robert G. Dixon, Jr., Democratic Repre— sentation (New York: Oxford University Press, 1968). 26. Edmund Burke, "Speech to the Electorate at Bristol," in The Writings and Spseches of Edmund Burke (Boston: Little, Brown, Co., 1901), Vol. II, pp. 95-97. 27. For an introduction to Bayesian decision theory, see Savage, op. cit. 28. See, for example, the analysis by Peter H. Aranson and Peter C. Ordeshook, ”Spatial Strategies for Sequential Elections,” in Richard G. Niemi and Herbert F. weisberg, eds., Probability_Models of Collective Decision Making (Columbus, Ohio: Charles E. Merrill, 1972), pp. 298-331. 29. John W. Kingdon, Candidates for Office: Beliefs EEDd Strategies (New York: Random House, 1968), p. 7. 30. Ibid. See also Kingdon, “Politicians' Beliefs 44 About Voters,“ American Political Science Review, 61 (1967), 137-143. 31. Kingdon, Candidates for Office, op. cit., pp. 159- 160. Another reason to reject the reliability of the study was that, as Kingdon himself admits, "The data analysis began by simply running the four independent variables against everything else to see what results would obtain," (p. 160). Given his small sample size, the sampling error might be high, and therefore any relationships discovered may have occured by chance. 32. Lewis Antony Dexter, "The Representative and His District,‘I in New Perspectives on the House of Representa- Eizss, ed. by Robert L. Peabody and Nelson W. Polsby (Chicago: Rand McNally, 1963), p. 11. 33. Kingdon, op. cit., p. 90. 34. Kingdon, "Politicians' Beliefs About Voters,“ 0p. cit., pp. 140-141. 35. Ibid., pp. 139-142. 36. Ibid., Chapter 2. 37. See Chong Lim Kim and Donald P. Racheter, "Candi- dates' Perception of Voter Competence: A Comparision of Winning and Losing Candidates," American Political Science Review, 67 (1973), 906. 38. Ibid. 39. Ibid., pp. 912—913. Kim and Racheter suggested that <>ther variables may also be important in explaining candi- dateS' beliefs. These included the incumbent status of the 45 candidates, the degree of competitiveness in their districts, their career socialization, and the level of their political ambition. 40. Ronald D. Hedlund and H. Paul Friesema, "Represen- tatives' Perceptions of Constituency Opinion," JOurnal'of Politics, 34 (1972), 730—752. 41. Ibid., p. 738. 42. Ibid., p. 741. 43. For a review of the various role orientations of legislators, see Wahlke et al., op. cit. 44. Hedlund and Friesema, op. cit., p. 743. 45. Ibid., p. 745. 46. In a subsequent article, employing the same data, Friesema and Hedlund found that overall 70% of the legis- lators' votes coincided with the referenda results in their districts, indicating that legislators' perCeptions of con- stituency opinion might have an effect on their roll call behavior. However, when they controlled for role orienta- tion, the delegates were more inconsistent with the results in their districts. These results may indicate that the concept of representational role orientation may not be a meaningful way of explaining legislative behavior. See H. Paul Friesema and Ronald D. Hedlund, "The Reality of Representational Roles,“ in Public Opinion and Public Policy, ed. by Norman Luttbeg (Homewood, Illinois: Dorsey Press, 1974), pp. 413-417. The question of the utility of the role concept is 46 further complicated by Kuklinski's study which supports the traditional interpretation of role explaining legislative behavior. Among California assemblymen, he found that delegates were more representative of constituency opinion than those who were either politicos or trustees. The rela- tionship was especially strong on issues that were considered to be highly salient to the voters. See James H. Kuklinski with Richard C. Elling, "Representational Role, Constituency Opinion, and Legislative Roll Call Behavior," AmeriCan Journal of Political Science, 21 (1977), 135-147. Further tests in this area are clearly needed. 47. Robert S. Erikson, Norman R. Luttbeg, and William V. Holloway, Knowing One's District: How Legislators Predict Referendum Voting," American Journal of Political Science, 19 (1975), 231-246. 48. Ibid., p. 235. 49. Ibid., p. 241. 50. Charles S. Bullock, III, "Candidate Perceptions of Causes of Electoral Outcome," Paper presented at the 1973 Annual Meeting of the American Political Science Association, New Orleans, September, 1973. 51. See Miller and Stokes, op. cit. Although the Miller and Stokes 1958 Representation Study contained infor- mation on candidates' perceptions, only the data for incum- bents have been analyzed in publications based on this study. 52. Since in all but one of the districts surveyed the incumbent won the election, no distinction was made between 47 incumbents and winners, and between challengers and losers. 53. Bullock, op. cit., p. 4. 54. Ibid., p. 5. 55. Ibid., p. 6. 56. Ibid., p. 10- 57. Ibid., p. 13. 58. Ibid., pp. 14-15. 59. Ibid., p. 17. 60. See note 51. 61. Kingdon, Candidates for Office, op. cit., p. 84. 62. Although incumbents may be confident of their ability to win reelection, they often find reasons to create enthusiasm within their campaign organization to prevent complacency. Incumbents may set a goal higher than 50% of the vote plus 1 vote by emphasizing their desire for a strong show of constituency support. See Marjorie Randon Hershey, "Incumbency and the Minimum Winning Coalition," American Journal of Political Science, 17 (1973), 631-637. For a theoretical view of the advantages of a larger than minimum winning coalition in American politics, see Charles Adrian and Charles Press, "Decision Costs in Coalition Formation," American Political Science Review, 62 (1968), 556-563. 63. See William H. Riker, The TheOry of PolitiCal Coalitions (New Haven: Yale University Press, 1962). 64. Kingdon, op. cit., p. 85. 65. Marjorie Randon Hershey, The Making of campaign Strategy (Lexington, Mass.: Lexington Books, 1974). 48 66. Hershey's study was conducted during the 1970 election in the state of Wisconsin. 67. Ibid., p. 21. 68. Ibid., p. 22. 69. Ibid. 70. Ibid. 71. Ibid., p. 95. 72. See Joseph A. Schlesinger, Ambition and PolitiCs: Political Careers in the United States (Chicago: Rand McNally, Inc., 1966). 73. Ibid., p. 17. 74. Gordon S. Black, "A Theory of Political Ambition: Career Choices and The Role of Structural Incentives," American Political Science Review, 66 (1972), 144-159. 75. Ibid., p. 148. 76. Ibid., p. 149. 77. Ibid., p. 156. 78. Ibid. 79. David W. Rohde, "Risk-Bearing and Progressive Ambition: The Case of Members of the United States House of Representatives,“Paper presented at the Conference on Uncertainty, Political Processes and Public Policy, San Diego, California, August, 1974. 80. For a more technical discussion, see Shepsle, "The Strategy of Ambiguity: Uncertainty and Electoral Competi- tion," op. cit. 81. Rohde, op. cit., p. 24. 49 82. Ibid. 83. See, for example, Kenneth A. Shepsle, "Parties, Voters, and the Risk Environment: A Mathematical Treatment of Electoral Competition Under Uncertainty," in Probability Models of Collective Decision Making, ed. by Richard G. Niemi and Herbert F. Weisberg (Columbus, Ohio: Charles E. Merrill, 1972), pp. 273-297. 84. For a discussion of the advantages of a formal model, see Morris P. Fiorina, "Formal Models in Political Science," American Journal of Political Science, 19 (1975), 133-159, May Brodbeck, "Models, Meaning, and Theories," in Readings in the Philosophy of the Social Sciences, ed. by May Brodbeck (New York: MacMillan, 1968), pp. 579-600, and Paul Diesing, Patterns of Discoveryiin the Social Sciences (Chicago: Aldine-Atherton, 1971). 85. This is indicated by the discussion of the func- tional relationship of the variables utilized in his study. See Fiorina, Representatives, Roll Calls, and Constituencies, op. cit., pp. 83-86. CHAPTER TWO A BAYESIAN MODEL OF POLITICAL CHOICE 1. Introduction In chapter one, we discussed the significance that information can have upon the choice behavior of political actors. Information was shown to be a crucial factor in the decision making process and unless the information acquired is accurate and complete, a politician may choose an alter- native leading to a less than optimal outcome. Coping with uncertainty is therefore an important concern for a decision maker.1 The information with which a politician comes into contact is often influenced by his perception. Perception is defined as being sensitive to and developing certain interpretations of stimuli and facts.2 During the perceptual process, the individual attempts to order the stimuli and facts and interprets them, based upon his current impressions and past experiences. He then can use these interpretations as the basis for actions directed toward the achievement of his statedgoals.3 In this study we are concerned with the explanation and prediction of the policy decisions of candidates in political campaigns. Since the nature of campaigning involves the collection and interpretation of available information for 50 51 making decisions, it is important to know the nature of the decision making process. This can be determined by the development of a model which describes in abstract terms the relationship of a set of concepts within the framework of an election campaign.4 An approach to the formulation of a model of decision making that has developed into an important model for the making of rational selections among alternative courses of action when information is incomplete or uncertain is Bayesian decision theory.5 A Bayesian analysis is relevant to the problem of making optimal policy decisions in a campaign, because, as Winkler and Hays state: The motivation for Bayesian methods is essentially the desire to base inferences and decisions on any and all available information, whether it is sample information or information of some other nature.6 We can thus incorporate the knowledge, the experience, and the intuition of candidates into the decision calculus in order to explain their policy choices.7 In this chapter, we will apply the Bayesian framework to study the decision making of candidates in a campaign under conditions of risk and uncertainty. First, we will examine the basis for the Bayesian approach, which involves the use of subjective or “personalistic” probabilities, as compared to the frequentist approach to statistical infer- ence, which uses “objective" probabilities. Then a dis- cussion of the use oleayeS"theorem is followed by an explanation of the BayeSian decision model and its 52 application to a problem in political choice. In thefollowing chapter, the hypotheses derived from the Bayesian model will be operationalized for testing, using data from the 1958 Representation Study, which contains in- formation about the perceptions of congressional candidates concerning their constituencies and their campaign activities.8 Chapter four will include an analysis of the results. 2. Bayesian Decision Theory Since WOrld War II, increased emphasis has been placed on the problem of decision making when information is incomplete or uncertain. The approach that has been devel- oped to make the best decisions under such conditions is known as Bsyesian decision theory,9 named after the English Presbyterian minister and mathematician, Thomas Bayes (1702- 1761). Although he did not originate this statistical decision theory, he is recognized for what has become known as Bayes' theorem, 10 which is the essential tool of the analysis used to handle the problem of uncertainty.11 Bayes' theorem states a procedure for the revision of prior opinion about an event in the light of new information. The opinion is expressed in terms of a probability, and it is the subjective interpretation of this probability that distinguishes the Bayesian approach from the non-Bayesian approach.12 For a Bayesian, these prior probabilities are “degrees of belief" and the result of human judgment.13 53 The subjective or "personalistic" theory of probability was first introduced by F.P. Ramsey in his book of essays, The Foundations of Mathematics and Other Logical Essays,14 but was primarily developed by de Finetti,15 Koopman,16 Good,17 and Savage.18‘ In fact, it was not until the publi- cation of The Foundations‘of;Statistics by Savage that scientists began to widely accept the use of the concept of subjective probability. The basic thesis of the personalistic theory of proba- bility19 is that the probability of an event is the "degree of belief or degree of confidence placed in the occurrence of an event by a particular individual based on the evidence 20 This definition contrasts to the more available to him? traditional view in which a probability of an event is seen in terms of a relative frequency: Non-Bayesians argue that the only legitimate types of probabilities are "objective" or relative frequency of occurrence probabilities. They find it difficult to accept the idea that subjective or personalistic probabilities should be processed together with relative frequencies, as in the Bayesian's use of Bayes;l theorem, to arrive at posterior probabilities. The occurrence of some events, however, cannot validly be assigned an objective probability. For example, the statement, "The Democrats will probably win the election tomorrow," appears to be a probability statement, but it is very difficult to see how it could describe long run relative frequencies of outcomes of repeated experiments. The problem is that this event is unique and cannot be duplicated. Infor- mation regarding past events in similar situations is not 54 available and no information in the form of observed frequencies exists as repeated trials under identical condi- tions. Instead, the statement describes one's degree of belief or subjective judgment about a situation that will occur only once. Long run objective frequencies are thus incapable of interpreting many of the kinds of events that are of concern to those studying political behavior, and "if we rigidly maintain that only objective probabilities have meaning, we prevent ourselves from handling some of the most important uncertainties involved in problems of decision making."22 Subjective probabilities have the same properties as 23 and may be chosen in any manner objective probabilities, prior to the occurrence of an event and may be based in part on objective evidence. Thereafter, the change in subjective probability as a result of experience or sampling is governed by Bayes' theorem. The simplest version of Bayes' theorem24 can be stated as follows: for two events, A and B, P (B/A) P (A) P(A/B) = p(B/A)P(A) + P(B/A)P(A) where A represents the complement of the event A (that is, "not A"). The equation consists of two basic components: a prior probability, P(A), and a likelihood, P(B/A). The prior probability is the subjective probability held at the begin- ning of the investigation or experiment. The likelihood is the probability of B conditional on the occurrence of A. 55 When the probabilities are combined in the manner specified by Bayes' theorem, they form a posterior probability, P(A/B), which summarizes the state of knowledge after taking into account the new information from observing B. The equation can be restated as follows: Posterior Probability = Likelihood x Prior Probability Likelihood x Prior Probability + (l-Likelihood) x (l—Prior) As more information becomes available, the first posterior probability may be combined with the new information to form a revised posterior probability.25 At this point we have shown how subjective probabili- ties differ from those based on relative frequencies. Sub- jective probabilities are more suitable in describing the occurrence of events in a political campaign, since these are unique events which can only be represented by one's degree of belief. Through the operation of Bayes' theorem, these probabilities are combined with all relative informa- tion, subjective or objective, to produce revised probability estimates. These new posterior probabilities can then be used to select an alternative within a Bayesian decision making model. This process will be explained in the next section. We again wish to emphasize that we do not claim that individuals actually employ Bayes' theorem for revising and estimating the probability of an event. Instead, we hope to 56 show that rational decision makers act ss_i£_they attempt to revise their probability estimates in this manner for the purpose of making the “best" decision possible, given uncertainty about their environment. A model should not be judged by the realism of its assumptions, but by the accu- racy of its predictions. If our hypotheses accurately reflect our data, then we have developed a significant new tool for the discovery of new relationships and theories. 3. The Bsyesian Decision Model The Bayesian approach to decision making involves the selection of a decision rule that minimizes expected losses under uncertainty. However, the interpretation of uncer- tainty in decision theory has.been the subject of some dis- cussion and it would be appropriate at this time to specify its meaning within the Bayesian framework and discuss its relationship to the concept of risk. Decision making under uncertainty is defined as a situa- tion in which one does not know the probability of an event. However, with the use of subjective probabilities to describe the occurrence of events, Bayesian decision making under uncertainty becomes decision making under risk.26 Since under the subjective interpretation of probability it is always possible to assess probabilities for the possible events, it may not be necessary to emphasize this distinc- tion. As Shepsle argued, uncertainty is actually a degene- rate case of risk: a known probability distribution 57 collapsed on a single point.27 The risk versus uncertainty dichotomy is therefore one of degree and is essentially artificial according to the subjective interpretation of probability. Henceforth, we shall refer to uncertainty as a situation in which the probability of an event is not equal to 1.0 (not certain). This is consistent with the modern approach to refer to this entire spectrum as one of uncertainty.28 The use of the term‘sisk_will be reserved for repre- senting the expected loss, which is the loss of making an error times the probability of making the error. In Bayesian decision theory, the weighted average or expected value of these risks, using prior probabilities of events as weights, yields the expected risk of a particular strategy.29 To summarize, uncertainty refers to decision making when subjective probabilities are used to estimate the occurrence of an event and risk refers to a description of the expected loss involved in making a decision under uncer- tainty. we can now proceed to define the elements of a Bayesian decision model. Bayesian decision theory starts with the assumption that regardless of the type of decision, there are certain basic characteristics of the decision problem that can be identified. These characteristics form the basic components of the model and provide a structure for a solution to the 30 problem. The basic unit of analysis is the decision maker, who 58 is the individual charged with the responsibility for making the decision. The individual is presumed to be uncertain about which state of the world is the true state which obtains. In Savage's terminology, the 32513.15 defined as "the object about which a person is concerned," and a 53232. 9: the world is "a description of the world, leaving no relevant aspect undescribed."31 The set of states of the world are assumed to be mutually exclusive and collectively exhaustive; one and only one description of the world does in fact obtain, or is the true state Q: the world.32 An EXEEE.iS defined as a set of states, usually containing the true state of the world.33 Let 9 denote an event containing a set of n elements, (9: 61, 92, ..., On), such that each ej is a possible state of the world. Empirically, one could think of e as an unknown parameter relevant to a decision making problem in a political campaign, such as the median position of consti- tuency opinion on an issue of concern. A candidate may be uncertain as to the actual value of the median (i.e., the true state of the world). An action A, (A: A1, A2, ..., Am), is a function which assigns a consequence to each state of the world. Set A con- tains a set of alternative courses of acts, actions, or strategies taken when a state of the world, ej, obtains and results in a consequence Ei (E: E11, ..., Emn)' which is 3'" a set of specific outcomes or payoffs. A payoff is a measure of the net benefit received by the decision maker. 59 The problem is to choose the best of the alternative strategies to achieVe the highest payoff possible. An illustration of the decision problem is shown in Figure 2.1. STATESLOF'THE'WORLD ACTIONS 61 62 93. . . an A1 E11 . . . . Eln A2 . E22 . . . A3 . . E33 . . . Am Eml ° ' ' ° Emn Figure 2.1 THE DECISION PROBLEM Decision making under certainsy will occur when the true state of the world is known: P(ej) = 1.0. The decision maker would merely have to look down the column of actions in the payoff table and select the alternative Ai that maxi- mizes the value of Eij' Under uncertainty, the individual does not know the true probability of each possible state of the world, so he assigns each a subjective probability. The selection of an appropriate action is based on the outcome of a lottery, and the act with the greatest expected value will be the most desirable choice.34 Therefore, once the probabilities of the states of the world have been speCified, usually after repeated sampling, 60 and after a utility function has established the value of each possible outcome,35 the.Bayesian decision rule is applied, by choosing the action with the highest expected value, or, alternatively, by choosing the one that minimizes expected loss.36 A Bayesian decision model is very appropriate for the study of candidates'perceptions.37 Perceptions are not based on repeated trials of experiments, but on previous knowledge, experience and intuition, as well as one's biases and prejudices. An individual's estimate of the true state of the world is constantly subject to change, and if we are going to be able to explain and predict political behavior, we need to know what factors influence the choice of an estimate, what is the effect of new information, and how the individual values the outcomes. In the next section we will apply this framework for analysis to a decision making problem in a political campaign. 4. The Decision Problem In politics, the nature of the decision making process is not unlike the Bayesian procedures presented in the last section, for politicians may often base their decisions on subjective estimates that are revised as new information is received. When the final decision is to be made, the indi- vidual must assess the likelihood of particular outcomes, as well as the confidence he has in his estimates. Thus, decisions are likely to be based not only on objeCtive 61 criteria, but on subjective beliefs as well.38 We proceed in the presentation of a campaign decision problem by assuming that candidates are rational and will select courses of action that maximize their expected 39 Since we are concerned with how candidates utility. estimate and revise their subjective probability distribu- tions of the true states of the world, the rationality of a decision maker can be interpreted as the efficient use of contextual information so as to produce actions consistent, s priori, with his preferences.40 In a campaign, one of the most crucial decisions to be made by a candidate is the selection of a public policy position on an issue of concern to his constituency. In order to make a choice to maximize the probability of winning the election, a number of factors or parameters must be known. These may include the shape of the distribution of voter opinion on an issue, the salience of the issue to the electorate, the voter turnout, the spatial mobility of the candidate, and the strategy of the opposition. This topic has been of great concern to political scientists in recent years. Attempts to model the electoral process have resulted in hypotheses that specify when elec- toral situations are likely to have equilibrium positions that the candidates will adopt or describe when candidates are likely to converge or diverge in their selection of a public policy position.41 Since most of theSe models assume perfect information, 62 candidates can act under conditions of certainty. However, considering all the parameters that must be estimated in a campaign, the models appear to be unrealistic and perhaps too simplified in their approach to the study of political behavior. Attempts to complicate and generalize these models have only succeeded in moving them further from reality and empirical testability.42 In relating the spatial analogy to the study of campaign behavior, we do not assume perfect information. Instead, we assume that candidates base their decisions on subjective probabilities, derived from sampling information (random or nonrandom) and prior experience. In this way, we can explain the behavior of a candidate under the more realistic assumption of uncertainty and we can also avoid having to make interpersonal comparisons of utility. Politi- cians can and do make mistakes from misinterpreting informa- tion which may or may not be complete, so by studying candi- date decision making within this kind of framework, we should be able to explain and predict the actions of the candidates more accurately. In this model, we will select one of the important parameters about which a candidate needs information in order to make a rational decision and maximize his proba- bility of election. The parameter is specified as the true position of the median of the voter preference distribution for a salient political issue. Information about this para- meter may be obtained over the course of a campaign and a 63 posterior distribution can be formulated. If we find that candidates do make their decisions as if they have formula- ted a posterior distribution of the true value of the para- meter, then we have shown the value of this model as a tool for discovery and can extend the analysis to encompass other important factors relevant to campaign decision making. In a campaign, the candidates will often try to adopt a policy position that will insure them of at least a tie (50% of the vote) in a two-way race. In this case, the best position to adopt would be the one representing the median of the voters' preferences. In a district where the distri- bution of the voters' preferences is symmetric and unimodal, the mean position is the same as the median.43 Under uncertainty, the true position of the median is unknown and must be estimated by the candidate. It is thus necessary for the candidate to specify a subjective probability dis- tribution over the set of possible values of the median. The distribution could be formulated by the use of Bayes' theorem.44 The formula for a conditional probability dis- tribution of the discrete population parameter s is as follows: y/5 to K: II Gj) P(G = 93'1- y/é ei) P(é 91) II ML: m '<2 II This equation shows how it is possible to revise probabili- ties concerning the unknown values of a parameter 5, when sample information, represented by §, becomes available, 64 regardless of how the information is obtained. .The result is a posterior probability for each possible value of S, which can be described as a conditional probability distri- bution of 6.45 For example, if we wish to determine the probability that a coin is fair: P( P(H) = P(T) = .5), we can flip the coin and, as a result of the new information, revise our estimate of the true state of the world. The true state of the world is the true probability that the coin is fair. The following chart illustrates a possible result: BEFORE THE TOSS AFTER THE T058 2151 ;£L. £151. 19. 0.4 . 91 = .10 0.4 91 e .05 0.5 92 = .80 0.5 92 = .75 0.6 93 = .10 0.6 93 = .15 0.7 94 = .00 0.7 e, = .05 Before the toss, the probability that the coin is fair is .80. After the toss, the probability distribution changes on the basis of the new information and the probability that the coin is fair changes to .75. Another toss of the coin would produce additional information and the probability distribution of 9 would be revised accordingly.46 The number of probability distributions of the states of the world that may exist is infinite, but we can classify them into two general types: 1) those that approximate a uniform probability distribution and 2) those that approxi- mate a “spiked distribution," or one with minimum variance. 65 These two types are illustrated in Figures 2.2 and 2.3. P(e) 1.00 .00 6 Figure'2.2 A UNIFORM PROBABILITY DISTRIBUTION OF A PARAMETER 9 Figure 2.2 shows the extreme case in which either no knowledge is available or the individual has no reason to believe that one value of the parameter s, the true value of the median of the voter preference distribution, is more likely to occur than another.' This is a situation described by Borch as satisfying LaPlace's Law si_Insufficient Reason, which states that when the probability of a series of events is unknown, the events should be treated as equally probable:47 P(Ei) = l/N for all i. In selecting a position that he believes may be the median of the voter distribution, a candidate would not be considered to be very knowledgeable about constituency opinion if his subjective probability distribution for the median was represented by a distribution similar to the one in Figure 2.2. In this case, it really would not matter to the candidate which of the values he may choose to adopt, 66 since each value has been assigned the same subjective probability of occurrence. The candidate is considered "ignorant" of the true value of the median, for he has no information to indicate that one value is more likely to be the median than another. The selection of a policy position becomes essentially a ”guess”, and with a wider range of choices, there is a greater likelihood of making the wrong decision. Given the ignorance associated with this kind of dis- tribution, a decision maker may not be very confident about his selection of the median position, which in the case of a uniform distribution is essentially a random choice. However, with a uniform distribution, the incentive would exist for the candidate to gather additional information in order to make his estimates more precise. Theoretically, this may be accomplished by sampling and using this new information to revise the distribution through the use of Bayes' theorem. It would be rational for a candidate to seek additional information until such time as the cost of the information is greater than or equal to the value of the increased precision of the subjective estimate of the true state of the world (unless of course the election comes first). (See Figure 2.3.) 1 On the other hand, a "spiked“ distribution represents just the opposite situation, in which only one value is considered likely to.occur. Figure 2.3 shows a distribution in which the probability of all values of e are equal to 67 P(G) 1.00 .00 9 Figure'2.3 A SPIKED PROBABILITY DISTRIBUTION OF A PARAMETER e 0.0, except one, which is equal to 1.0. The variance of the distribution equals 0.0. In this case, the candidate would be unlikely to choose any other position, since the proba- bility of one particular value is so much higher than the rest. As an example, in frequentist (objective) terms, if one were to toss a coin 100 times and it came up heads 100 times, then the probability of a head would be 1.0, and if asked to predict the next toss, one would of course choose heads.48 The contrast between a candidate with a uniform distri- bution and one with a spiked distribution of the possible states of the world is that the former would have a much greater incentive to seek out additional information. The candidate with a spiked distribution would not be expected to be able to add much to the precision of the information already received. This does not necessarily mean that the 49 information is correCt, but that it is at least consistent. In the selection of a value of e to adopt as a public 68 policy position, the candidate with a spiked distribution would be likely to select the value with the greateSt proba- bility, and would probably repeat that choice if given the Opportunity. He would therefore be confident about his decision, but not necessarily certain of the outcome, since nature can still play strange games. The candidate with the uniform distribution would be less confident about his deci- sion (since any choice would be as likely as another) and if given a number of opportunities to select a public policy he believed to be the median, would not have any rational reason to keep selecting the same position. Since each position would be equally likely to occur, a random selection of positions would be an acceptable strategy. The result is that compared to a less confident candidate, a confident one would be more consistent in his choices, selecting the posi- tions with thegreatest probability. The actions of the confident candidate would thus be more predictable than those of one who chose in a more randomized fashion. (See Figure 2.4.) Figure 2.4 represents another example of a spiked dis- tribution, one with small variance, but not as extreme as the one in Figure 2.3. In this case, there are a few values that show a high probability. We could still consider the candidate to be fairly confident of his choice of a public policy position if he were to choose one of the few values that cluster around a certain position. For example, it appears the 95, 96: and 97 have the greatest subjective 69 P(e) 1.00 ®_-—————-—— 0.00 91 92 93 94 7 98 99 910 9 0‘ Figure 2.4 A SUBJECTIVE PROBABILITY DISTRIBUTION OF A PARAMETER 9 probability of being the true median position. Over the long run, he would not have to always choose 96 (the value with the highest probability), but by choosing these values most often he probably would not be too far off the mark, according to his estimates. In any case, his choices would remain fairly consistent, with only an occasional and infre- quent deviation. As the distribution approaches a more uniform distribution, with the values becoming more equi- probable, his choices over the long run would be more random and thus less predictable. We have thus shown why a confident candidate in a campaign would be more likely to choose a position close to what he perceives as the most likely position of the median of the voter distribution. In contrast, the less knowledge- able and therefore less confident candidate would also be less predictable in terms of his selection of a public policy position. These conclusions can be stated in the 70 form of the following proposition: ”Proposition 1: Ceteris paribus, the smaller the variance of the subjective proba- bility distribution of a parameter 6, the more likely a candidate will consistently choose his best estimate of e as the true value of 9. The converse is also expected to be true: as the variance of the probability distribution of 9 increases, the less likely a candidate will choose his best estimate of the true value of 9, since the values will approach equiprobability. This "uncertainty" or lack of confidence in one's choice of the probability of a state of the world which leads to a less consistent chOice underlies the notion of ambiguity in terms of a probability distribution of proba- bilities, or, as Savage calls it, a "second-order" distri- bution of probabilities.50 For example, consider a situa- tion in which a subject is given two urns, A and B, with urn A containing 5 red and 5 black balls, and urn B con- taining 10 balls of unknown color (either red or black). If asked to select a ball from one of the urns and predict the color of the ball, which urn would the subject choose, urn A or urn B? Since he has no knowledge of the distribu- tion of the balls in urn B, he could apply LaPlace's Law of Insufficient Reason and assume the selection of either a red or black ball is equally likely. The selection of a red or black ball is also equally likely if a ball is selected from urn A, so there is actually no reason to select one urn over the other in order to improve the prediction of what kind of 71 ball would be selected. There is, however, reasonable evi- dence to believe that urn A will be chosen, since the subject was told by the experimenter what the true distribution of balls in urn A was. The subject was thus more "certain" or more "confident" that he knew the true distribution of balls in urn A.51 When applying this result to candidates in a political campaign, we can predict that the uncertainty regarding the true distribution of an unknown parameter can affect the kinds of choices made by a candidate, depending upon his degree of knowledge and eXperience, as well as his perceptions. We can also examine the selection of a policy position within the framework of the normal form of a decision matrix.52 For the sake of simplicity, let us assume that there are two possible states of the world: 1) the median of the voter distribution in a constituency is 91, 2) the median of the voter distribution in the same constituency is 92. A candidate has essentially two choices: 1) select Al, the perceived position of the median at 61, or 2) select A2, the perceived position of the median at 62. W, X, Y, and Z represent the set of possible payoffs when an action is chosen and one of the states of the world obtains. (The payoffs can be either positive or negative.) (See If the state of the world is such that 91 is median and Al is chosen, then the candidate could receive W votes (cell 1). Likewise, if 62 is the median and A2 is chesen, then one could expect to Figure 2.5.) the true expect to true receive Z 72 STATES OF THE WORLD ACTIONS e1 e2 1) 2) A1 w x 3) 4) A2 Y Z Figure 2.5 A DECISION MATRIX votes (cell 4). If the candidate chooses incorrectly, i.e., chooses A1 when 62 is the median (cell 2), or chooses A2 when 91 is the median (cell 3), he could receive a smaller number of votes (or lose votes), represented by X and Y, respective- ly. In order to decide whether to choose A1 or A2 under uncertainty, the expected value of A1 and A2 can be calcula- ted. The action with the greatest expected value will then be the one chosen:53 Let P = p(91), (l-P) = p(92), such that P + (l-P) = 1.0, 0 5 P s 1.0 E(A1) = PW + (l-P)X E(A2) = PY_+ (l-P)Z For example, if p(el) = .8 and p(92) = .2, then the expected value of A1 = .8W + .2X, and the expected value of A2 = .8Y + .2Z. The payoffs of W and Z would be greater than X and Y, since W and Z represent outcomes when one picks the true state of the world. The expected value of A1 would be higher than A2 in this case (assuming WeZ). If the proba- bilities were reversed, then A2 would be chosen. When 73 P = (l-P), there is no dominant choice. The question that still remains, however, is how confi- dent is the candidate that his estimates of p(el) and p(92) are true. If he is not confident, then his probability distribution of 9 will have a high variance, and as the variance decreases, his confidence will increase. According to Proposition 1, a confident candidate would be more likely to trust his judgment and go with his best estimate of the true value of the median. One who is less confident would consider a wider range of choices. This proposition can go far in explaining the selection of a public policy position of a candidate, but we also have to consider the nature of the incentives and rewards that would be needed to encourage the collection of information by candidates to learn district opinion. Clearly, the can- didate who runs unopposed need not concern himself with estimating the median of the voter distribution during the general election, but the case is quite different when a candidate perceives there is some iisk involved in the selection of a public policy position. Risk is defined as expected loss. This means that the expected loss of votes as a result of adopting a particular policy position is determined by a loss function. Formally, a loss function L(A,e) is a real-valued, non-negative func- tion which reflects the loss in taking action A when 9 is the true value of the parameter. The loss is zero when A is the best action for 6. The risk is thus the expected value 74 of the loss functiongiven an action and the values of the unknown parameter.54 The values in the matrix represent the isss from choosing an action, rather than the payoffs, as shown earlier. ’The'task is to adopt a decision rule to minimize expected loss. Although the criteria of maximizing expected payoffs and minimizing eXpected loss yield identical results, we will henceforth refer to the decision matrix in terms of loss, since this is the method most often used in the analysis of decisions when sample information is obtained, and it allows us to use the concept of risk.55 STATES‘OF'THE'WORLD ACTIONS e1 92 1) 2) A1 0 X 3) 4) A2 Y 0 Figure‘2.6 A DECISION MATRIX SPECIFYING LOSSES AS OUTCOMES Figure 2.6 is a decision matrix in which the components in the cells represent positive losses. The zeros represent no loss, but X, Y in this case represent the number of votes lost from choosing an action, given a particular state of the world. The expected loss (risk) for each action is: E( L(Al,9) ) =_po.+ (1-P)x <1-P)x PY E< L(A2.e) ) PY,+ (l-P)O A decision maker can then choose either A1 or A2, whichever minimizes the expeCted loss.' For example, if X and Y are 75 equal and (1-P) is greater than P,_then A2 will minimize the expected loss. If there is no risk in choosing a policy position that is not the median position (if the expected loss is zero for each action), then a decision may be made on the basis of some other factors that are of salience to the voters. These may include satisfying an intense minority in order to gain their support and use of their resources, or perhaps the candidate may choose to merely satisfy his own policy pre- ferences. Candidates who are unopposed will perceive no risk and may be free to adopt any policy position, except that instead of perceiving risk in the general election, it may arise during a primary contest, which, in many one-party areas, is tantamount to election. Consideration of the con- sequences of uncertainty and risk in primary contests is beyond the scope of this investigation. When there is risk perceived in the_general election, in order to prevent any loss of votes in an uncertain world, the candidates might be expected to make a greater attempt to determine voter sentiments in order to adopt a position that comes closest to minimizing expected loss. This will presumably depend upon the confidence of the candidate and the cost of such information. The greater the confidence of the candidate, the lees likely he will engage in a large scale effort to gatheradditional information, since it would be expected only to confirm what he already knew. Less con- fident candidates who had the necessary reSources would be 76 expected to try to gain additional information in order to make the best decision.56 To summarize, when there is no risk associated with choosing one alternative over another (the expected loss for each alternative is zero), there is no reason for a candidate to favor a particular alternative. However, when there is some risk associated with a set of alternatives, there is an incentive for the candidate to choose a position to minimize his expected loss. In addition, when the risks associated with each alternative are non-zero and equal, then the choice of an alternative can be chosen randomly, but as the risks for each alternative change, the best choice to make would be the one based on the candidate's best estimate of the state of the world. This leads us to the statement of Proposition 2: Proposition 2: Ceteris paribus, as the risk from chbosing an alternative increases, the more likely a candidate will choose his best estimate of 9 as the true value of e. This implies that both risk and uncertainty play an important role in the selection of a public policy position by a candidate. Risk is important because it determines the expected value of the outcome of each decision and uncer- tainty determines the degree of confidence attached to the estimates of the true value of e, which in turn determines the expected losses (risks). It also means that the rela- tionship in Proposition 1 would be strengthened by consi- deration not only of the uncertainty regarding the 77 probability of occurrence of a parameter, but also the degree of risk involved. With no risk, it would not matter which position was taken, at least within the framework of this model. If the risks were greater than zero, then considera- tion of risk and uncertainty would be important in the selec- tion of a policy position.57 Also, since these concepts are based on candidates' perceptions of the world, if we discover that candidates act as if they followed their per- ceptions, then these results can have important implications for the popular control of public policy. 5. Conclusion In this chapter, the basic framework of a Bayesian decision model has been formulated to analyze the subjective decision making of politicians seeking electoral gain under conditions of risk and uncertainty. The propositions derived from the Bayesian model explain why candidates would be more likely to follow their best estimates to make a decision and why others would be more willing to adOpt a more randomized strategy. One important consideration in the development of the Bayesian model has been the concern for empirical relevance that can assist us in the development of a political theory. As a result, the mathematical complexity of the model has been maintained at a level that simplifies the realities of a political campaign, but can be easily modified to accomo- date more complex situations. The simpler fOrm may have 78 resulted in fewer hypotheses, but their empirical implica- tions are potentially quite fruitful, as shown in the following chapters. In the next chapter, the two propositions are opera- tionalized with the consideration for the data that are available, which include the 1958 Representation Study, and given both the data's strengthsanxilimitations. Although not all of the hypotheses stated in chapter three will be for- mally and directly deduced from this model, they will gen- erally reflect the relationships of the basic concepts of Bayesian decision theory. The testing of these hypotheses will be intended to serve the purpose of providing empirical knowledge that can be considered in the construction of more complex models and in the collection of a more specialized set of data. CHAPTER TWO NOTES 1. As Downs states, "...it is the basic force affecting all human activity...it shapes the nature of [eVery signifi- cant institution in society]." See Anthony Downs, An Econo- mic Theory of Democragy_(New York: Harper and Row, 1957), p. 13. 2. E. Frank Harrison, The Managerial Decision Making Process (Boston: Houghton-Mifflin Co., 1975), p. 158. 3. Ibid., p. 160. For some recent research on the nature of the perceptual process, see John R. Bergen, "The Structure of Perception," Journal of the Association for the Study of Percsption, (1969), 1-19, and Sheldon S. Zalkind and Timothy W. Costello, "Perception: Some Recent Research and Implications for Administration," Administrative Science Quarteriy, (1962), 218-235. 4. From the calculus of the model, one can manipulate variables, while holding others constant, in order to derive a set of testable hypotheses. If the results of the model coincide closely to the data, then we can proceed to develop a body of knowledge that can lead to the development of a well-defined theory. For a discussion of the model building process, see Paul Diesing, Patterns of DiScOVery in the Social Sciences (Chicago: Aldine-Atherton, 1971), Abraham 79 80 Kaplan, The Conduct of Inquiry (San Francisco: Chandler, 1964), Richard Rudner, The Philosopky of Social Science (Englewood Cliffs, New Jersey: Prentice-Hall, 1966), and Ralph M. Stogdill, ed., The Process of Model-Building in the Behavioral Sciences (New York: Norton, 1970). 5. This may also be referred to as simply statistical decision theory, but the term "Bayesian" usually implies the additional use of subjective probabilities. See Morris Hamburg, Statistical Analysis for Decision Making (New York: Harcourt, Brace and WOrld, 1970). 6. Robert L. Winkler and William L. Hays, Statistics: Probability, Inference, and Decision (New York: Holt, Rine- hart and Winston, 1975), p. 473. 7. Bayesian decision theory is basically a prescriptive theory, rather than a descriptive one. That is, it presents the principles and methods for an individual to make an opti- mal decision under risk and uncertainty, but it does not claim to present an accurate or complete description of how actual decisions are made in the real world. Instead, it allows us to postulate goals for decision makers and deter- mine whether they act ss ii_they had these postulated goals and used a decision calculus to arrive at their choices. For a discussion of the "as if" criterion, see Milton Friedman, "The Methodology of Positive Economics," in Esssys in Positive Economics, ed. by Milton Friedman (Chicago: University of Chicago Press, 1953), pp. 3-44, and Ernest Nagel, "Assumptions in Economic Theory," 81 American Economic Review, 53 (1963), 211-220, as well as Otto Davis, "Notes on Strategy and Methodology for a Scientific Science," in Mathematical Applications in Poli- tical Science, ed. by Joseph Bernd (Charlottesville, Virginia: University Press of Virginia, 1969). Friedman's argument was essentially that a formal theory should not be judged by the realism of its assumptions, but by the accuracy of its predictions deduced from the model. This prescriptive approach has the advantage of even- tually enabling us to describe the actual way in which decisions are made. As Shepsle notes of his decision making models: It may be argued that as the criteria upon which prescriptive rules are based approach the criteria used by real decision makers, as the resources with which our theoretical decision- maker is endowed reflect the resources of a real decision-maker, and (we speculate) as the decisions increase in importance to the decision- maker, the prescriptive theory we have developed becomes increasingly suitable as a descriptive theory. See Kenneth A. Shepsle, "Essays in the Theory of Risk-Taking" (Unpublished Ph.D. Dissertation, University of Rochester, 1970), p. 63. 8. For a brief description of the 1958 Representation Study, see the Guide to Resources and Services, 1976-1977 (Ann Arbor, Michigan: Institute for Social Research, 1976), pp. 180-181. A more detailed description will be given in chapter three. 9. For a description of Bayesian decision theory, see for example, John Aitchison, Choice Against Chance:‘ An 82 Introduction to Statistical Decision Theory (Reading, Massachusetts: Addison-wesley Publishing Company, 1972), Ward Edwards and Amos Tversky, eds., Decision Making: Selected Readings (Baltimore: Penguin Books, 1967), George P. Box and George Tiao, Bayesian Inference in Statistical Analysis (Reading, Mass.: Addison-Wesley Publishing Co., 1972), Henry E. Kyburg and Howard E. Smokler, eds., Studies in SubjectiveProbability (New York: John Wiley and Sons, 1965), Dennis V. Lindley, Introduction to Probability and Statistics From a Bayesian Viewpoint, 2 vols., (Cambridge, England: Cambridge University Press, 1965), Lindley, Making Decisions (New York: John Wiley, 1971), Lawrence D. Phillips, Bayesian Statistics for Social Scientists (New York: Thomas Y. Crowell Company, 1973), John W. Pratt, Howard Raiffa and Robert Schlaifer, Introduction to Statistical Decision Theory (New York: McGraw-Hill, 1965), Howard Raiffa, Decision Analysis: Introductory Lectures on Decision Making Under Uncertainty (Reading: Mass.: Addison-Wesley Publish- ing Company, 1968), Robert Schlaifer, Analysis of Decisions Under Uncertainty (New York: McGraw-Hill, 1969), Samuel Schmitt, Measuring Uncertainty: An Elementary Introduction to Bayesian Statistics (Reading, Mass.: Addison-Wesley Publishing Company, 1969), and Robert L. Winkler, An Intro- duction to Bayesian Inference and Decision (New York: Holt, Rinehart and Winston, 1972). The reference to Winkler contains an excellent_bibliography of additional articles and texts in this area. 83 10. See Thomas Bayes,_"An Essay Towards Solving a Problem in the Doctrine of Chances," Philosophical Trans- actions of the Royal Society, 53 (1763), pp. 370-418. Reprinted in Biometrika, 45 (1958), 293-315. Bayes did not present Bayes' formula in the form presently familiar to scientists, but he apparently understood how to make the same calculation. See wayne Lee, Decision Theory and Human Behavior (New York: John Wiley, 1971), p. 47. ll. Hamburg, op. cit., p. 42. 12. When a personalistic interpretation is applied to statistics, it is called Bayesian statistics. 13. Phillips, op. cit., p. 63. 14. (London: Kegan Paul, 1926). 15. Bruno de Finetti, "Foresight: Its Logical Laws, Its Subjective Sources," translated by Kyburg and Smokler, op. cit. pp. 93-158. 16. B.0. Koopman, "The Axioms and Algebra of Intuitive Probability,“ Annals of Mathematics, Ser. 2, 41 (1940), 269- 292, "The Bases of Probability,“ Bulletin of the American Mathematical Society, 46 (1940), pp. 763-774, and "Intuitive Probabilities and SequenCes,5 Annals of Mathematics, Ser. 2, 42 (1941), 169-187. 17. I.J. Good, Probabiliiy and the weighing of Evidence (London: Charles Griffen and Co., 1950). See also Good's more recent work, The Estimation'of'Probabilities: An Essay on Modern Bayesian Methods (Cambridge, Mass.: MIT Press, 1965), which includes an extensive bibliography. 84 18. Leonard J. Savage, The Foundations of Statistics (New York: Wiley,_l954). A second revised edition has been published by Dover.Press (1972). See also "The Founda- tions of Statistics ReConsidered," in Proceedings of the Fourth Berkeley Symposium (Berkeley: University of Califor- nia Press, 1961), pp. 575-586, Reprinted in Kyburg and Smokler, op. cit. 19. See Henry E. Kyburg, Jr., Probability and the Logic of Rational Belief (Middleton, Conn.: Wesleyan University Press, 1961), Chapter 3, for a description of the personalistic theOry of probability. 20. Hamburg, op. cit.,p. 12. 21. Ibid., p. 766. For a statement on the use of objective frequencies, see R. Von Mises, “Probability: An Objectivist View,“ in Elementary Statistics for Economics and Business, ed. by Edwin Mansfield (New York: W.W. Norton and Co., 1970), pp. 59-67. As Phillips notes in regard to the use of subjective probabilities: 'Many statisticians have been reluctant to adopt Bayesian ideas because they feel that prior Opinion is vague and incapable of being quanti- fied. (Phillips, op. cit., p. 53.) However, this may be more a reflection of the complexity of events, rather than an inability to quantify judgments. At this time, there are a number of methods available for measuring subjective judgments, but the question pre- sently before theorists involves which methods should be used. One procedure that has been investigated involves the 85 use of computers to decompose complex events into simple ones. See Ward Edwards, L.D. Phillips, W.L. Hayes, and B.C. Goodman, "Probabilistic Information Processing Systems: Design and Evaluation," IEEE TransaCtions on Systems'Science and'Cybernetics, SSC-4 (1968), 248-265, for more details. For examples of empirical research on the estimation of subjective probabilities by human subjects, see Lee, op. cit., Chapter 3, and Edwards and Tversky, op. cit. The results seem to indicate that while there is likely to be a good correspondence between objeCtive and subjective probabili- ties, subjective probabilities do not consistently sum to 1.0, nor does the multiplication rule (see note 23) for independent events always hold. However, subjective proba- bilities do tend to become more consistent for adults than for a child. See Lee, op. cit., p. 65. 22. Hamburg, op. cit., p. 766. 23. The properties of objective probabilities are based on the laws of probability. These include the following: 1) probabilities cannot be less than zero nor greater than one, and the probability of a sure event is 1.0, 2) the probability of either of two mutually exclusive events occurring is equal to the sum of their individual probabilities, and 3) the probability of two joint indepen- dent events occurring is equal to the product of the proba- bilities of each event. For a complete description of the laws of probability, see Phillips, op. cit., Chapter 3, or most textbooks on probability theory. 86 24. For the derivation and proof of Bayes' theorem, see Winkler and Hays, op. cit., pp. 93-94. The general formulae for Bayes' theorem for discrete and continuous events are shown on pages 472 and 494, respectively. 25. In the continuous case, the estimation of a pos- terior distribution can be facilitated when the likelihood function and the prior distribution form natural conjugates. For example, if the prior distribution can be expressed as a beta distribution, then the posterior distribution is also a beta distribution. Likewise, a gamma and Poisson distri- bution combine to form a gamma posterior distribution, and two normal distributions form a normal posterior distribu- tion. See Ira Horowitz, An Introduction to Quantitative Business Techniques (New York: McGraw-Hill, 1972). 26. Winkler, op. cit., p. 221. 27. Shepsle, "Essays...," op. cit., p. 56. 28. Hamburg, op. cit., p. 213. 29. Ibid., p. 697. 30. Ibid., p. 615. 31. Savage, Foundations of Statistics, op. cit., p. 9. 32. Ibid. 33. Ibid., p. 10. 34. The expected value is obtained by multiplying the value of the consequences of each act by the probabilities of each of the possible states of the world, and summing the products. One can then choose the action with the highest expected value. 87 35. Different utility functions may be used to indicate different values that individuals place upon different outcomes. See Winkler, op. cit., Chapter 5. It is at this point where the interpretation of what Shepsle refers to as the "risk environment" becomes significant as a factor in the selection of an optimal strategy. For a discussion of three types of utility functions (convex, concave, and linear), see Kenneth A. Shepsle, "Parties, Voters, and the Risk Environment," in Probability Models of Collective Decision Making, ed. by Richard G. Niemi and Herbert F. Weisberg (Columbus, Ohio: Charles Merrill Co., 1972), pp. 273-297, and for an earlier source, Kenneth J. Arrow, "Alternative Approaches to the Theory of Choice in Risk- Taking Situations,“ Econometrica, 19 (1951), 404-437. 36. There are actually a number of different decision rules that could be applied to this problem besides the maximization of expected utility. These include the maximin principle, Savage's minimax regret principle, and Hurwicz's pessimism-optimism index. For a discussion of the impli- cations of these decision rules, see Shepsle, "Essays...," op. cit. and D.J. White, Decision Theory (Chicago: Aldine, 1969). 37. Although Bayesian procedures have not been widely applied to problems in political science, their use has increased in recent years. Besides Fiorina's application to the study of legislative decision making, (Fiorina, op. cit.), it has caught on predominantly in the area of 88 international politics. Examples of the application to the study of international politics include: Eugene J. Alpert, "Capabilities, Perceptions, and Risks: A Bayesian Model of International Behavior,“ International Studies Quarterly, 20 (1976), 415-440, J.D. Ben-Dak and K. Finsterbusch, "Bayesian Analysis: Applications for the Study of Foreign Behavior," in Patrick McGowan, ed., International Yearbook of Foreign Policy, 1974 (Beverly Hills: Sage Publications, 1974), pp. 269-306, and Patrick McGowan, "A Bayesian Approach to the Problem of Events Data Validity in Comparative and International Political ReSearch," in Comparative Foreign Policies, ed. by James N. Rosenau (Beverly Hills: Sage Publications, 1974). For examples of the use of Bayesian decision theory in related fields, see Gudmund Iversen, "Statistics According to Bayes," in Sociological Methodology 1970, ed. by E.F. Borgatta (San Francisco: Jossey-Bass, 1970), pp. 185-199, and V.M. Rao Tummula and Richard C. Henshaw, eds., Concepts and Applications of Modern Decision Models (East Lansing, Michigan: Graduate School of Business, Michigan State University, 1975). 38. For a collection of articles illustrating the behavioral aspects of subjective decision making, see ward Edwards and Amos Tversky, Op. cit. For a discussion of the use of information, see Avner M. Porat and John H. Haas, "Information Effects on Decision Making," Behavioral Science, 14 (1969), 98-104, and Robert Radlow, "Decision Making and the Theory of Learning," in Decision'and‘ChOice: ‘Collections 89 of Sidney Siegel, ed. by Samuel Messick and Arthur H. Brayfield (New York: McGraw-Hill, 1964), pp. 267-275. 39. For a discussion of rationality and its use in the social sciences, see Paul H. Conn, David B. Meltz, and Charles Press, I‘The Concept of Political Rationality," Polity, 6 (1973), 223-229, Arthur S. Goldberg, "Social Determinism and Rationality as Bases of Party Identifica- tion," American Political Science Review, 63 (1969), 5-25, and Herbert A. Simon, "A Behavioral Model of Rational Choice," Quarterlnyournal of Economics, (1955), 99-118. 40. Kenneth A. Shepsle, "The Strategy of Ambiguity: Uncertainty and Electoral Competition," American Political Science Review, 66 (1972), 568. 41. For a summary of the spatial modeling literature, see William H. Riker and Peter C. Ordeshook, Introduction to Positive Political Theory (Englewood Cliffs, New Jersey: Prentice-Hall, 1973) and Michael Taylor, "Review Article: Mathematical Political Theory," British Journal of Political Science, 1 (1971), 339-382. 42. For additional criticisms of spatial models, see Donald Stokes, “Spatial Models of Party Competition," American Political Science Review, 57 (1963), 368-377 and Brian Barry, Sociologists, Economists and Democracy (London: Collier-Macmillan, 1970). 43. See Duncan Black, The Theory of Committees and Elections (Cambridge: Cambridge University Press, 1958) for a discussion of the importance of the median in an electoral 90 strategy. 44. See Note 7 for a discussion of the as _i_f_ criterion. 45. Winkler, op. cit., pp. 74-76. The formula for a continuous probability distribution is shown in Winkler, pp. 143-145. 46. A more elaborate discussion of this example may be found in the article, Jack Hirshleifer, "The Bayesian Approach to Statistical Decision," in Edwin Mansfield, ed., op. cit., pp. 85-101. 47. Karl Borch, The Economics of Uncertainty (Prince- ton: Princeton University Press, 1968), p. 78. For n events, the uncertainty is greatest when all events are equiprobable. See Lee, op. cit., p. 265. While LaPlace's Law of Insuffi- cient Reason has many proponents and can be legitimately used in estimating the probabilities of events, it has had its detractors. See David W. Miller and Martin K. Starr, The Structure of Human Decisions (Englewood Cliffs, New Jersey: Prentice-Hall, 1967), pp. 123—124. Also, Shepsle points out that its use can be criticized if one does not know the finite bounds of the events. In addition, the different descriptions of what is meant by "collectively exhaustive events,“ i.e., the possible states of the world, may result in different choices. See Shepsle, "Essays in the Theory of Risk-Taking," op. cit., pp. 51-54. 48. Predicting the outcome of the toss of a coin may in fact be easier than estimating the true value of the median. At least with a coin toss, one may have one chance 91 or more, but second chances in politics are often rare. This is often the contrast between an event that can be described by objective rather than subjective probabilities. See Savage, "Probability: A Subjectivist View," in Mansfield, op. cit., pp. 102-112. 49. Box and Tiao, op. cit., p. 21. 50. Savage, op. cit., p. 58. 51. According to research in this area, subjects appear to prefer situations in which the probabilities are well- specified than ones with ambiguous probabilities, even though they should be indifferentaccording to LaPlace's Principle of Insufficient Reason. See Lee, op. cit., pp. 119-122 and Savage, 0p. cit., pp. 57-59. 52. Theoretically, all extensive game forms of a decision making problem can be expressed in normal form if the number of moves in the game is finite. See Edna E. Kramer, The Nature and Growth of Modern Mathematics (Green- wich, Conn.: Fawcett Publications, 1970), pp. 384-387. 53. In decision theory, one does not always use this criterion to select an alternative. Other methods are available, as explained in Note 36. 54. See Alexander M. Mood and Franklin A. Graybill, Introduction to the Theory of Statistics, Second Edition (New York: McGraw-Hill, 1963). pp. 165—167. 55. See Winkler, op. cit., Chapter 5. 56. In Pruitt‘s study of the information needs of decision makers, it was found that individuals try to reduce 92 risk by collecting information on various alternatives in order to assess the advantages and disadvantages of each. Pruitt also found that once a decision was made, much more information was required to make a decision maker change his mind. See Dean G. Pruitt, "Information Requirements in Making Decisions," American Journal of PSychology, 74 (1961), 433—439. 57. We have assumed that the payoffs and losses asso- ciated with each action and state of the world are known, as determined by the decision maker's utility function. This simplifies the present model, but allows for the deduction of additional hypotheses at a later time. For example, a subjective probability distribution can be attached to the parameter representing the true value of each outcome, introducing an additional consideration of risk and uncer- tainty in determining the true values of the outcomes. rn CHAPTER THREE CANDIDATE DECISION MAKING 1. Choice and Constituency Influence In a campaign in which a candidate is opposed for election and finds it necessary to take a stand on issues that he believes concerns his constituency, he will want to adopt a position that can raise his probability of election to an acceptable level. There are often three possible choices that can be made in the selection of a policy position: 1) he can adopt the policy position that he perceives to be adopted by the most number of people in his district, 2) he can adopt the policy position that best represents his own personal policy preferences and convic- tions, or 3) if the candidate is the incumbent, he can adopt the same policy position he has formulated through his voting record in the legislature. These choices of course are not mutually exclusive, and any candidate would be ineulenviable position if all three choices represented the same policy position. These three choices are components of the constituency influence model described by Warren Miller and Donald Stokes in their article, “Constituency Influence in Congress,"1 which was based on the analysis of the 1958 Representation Study. The primary purpose of the study was to: 93 94 ...provide the first direct confrontation between the policy preferences of the electorate and the policy acts of its elected representatives.... The general research objective...is the study of conditions under which policy agreement between constituent policy preferences and congressional roll call behavior is maxi— mized and minimized.2 Two paths of constituency control were 1) for the district to choose a representative who shared similar attitudes so that by following his own convictions, he also would follow his constituents' will; and 2) for the repre- sentative to follow his perceptions of district opinion in order to gain reelection. The primary paths which describe the linkage between constituency opinion and legislative behavior are shown in Figure 3.1. Of course, additional conditions must be met to ensure complete responsiveness. These include the ability, as well as the opportunity, to vote on legislative issues that reflect the policy positions that the represen- tative desires to express. Often the pressures to vote with one's party or in support of the administration may preclude the representative from following his district's preferences or his own on every vote, and vice-versa.39 (See Figure 3.1.) The linkage model may explain the paths of influence from the constituency to the roll call, but it does not take into account the activities of the representative during a campaign. There is no reason to believe that the policy position taken by the candidate in the Campaign will be reflected by the kinds of roll calls that a legislator may 95 Representative's Attitude ’P Constituency's Representative's Attitude Roll Call Behavior J I Representative's Perception of Constituency's Attitude Figure 3.1 PATHS OF CONSTITUENCY INFLUENCE4 be expected to answer. His campaign positions may in fact be his "real" positions, but his roll call positions may only reflect the idiosyncracies of the chamber's rules and membership. Also, if there is limited programmatic support for candidates based on their legislative record, then an investigation of their political behavior should not over- look the campaign environment as a separate forum for the airing of political views and public policies.5 In limiting the scope of the present study to the campaign environment, we intend to isolate some important variables that contribute to the behavior of the political candidate. As a result, the roll call behavior of the incumbent will be of only indirect concern. Instead, we will focus on the decision of a candidate, who may or may not be the incumbent, and whether he is likely to rely more 96 upon his perception of constituency opinion (his best estimate) or his own personal attitudes (or other considera— tions) in selecting a public policy position. Since the challenger of today can be the incumbent of tomorrow, it is important to explain the decisions of all the candidates, and not just the incumbents. The Miller and Stokes paradigm of constituency influence needs to be expanded to include a two phase process describing the linkage between the consti- tuency and the candidates, as well as the constituency and its representative. If there is a significant difference in these two processes, then we will want to know why candi- dates "change their tune" from the campaign to the legisla- ture. Figure 3.2 shows the revised linkage scheme. (See Figure 3.2.) The diagram shows that in order for complete respon- siveness to occur, the candidate should have a roll call record consistent with his perceptions of district opinion and comensurate with the public policy position he espouses during the campaign. Over a period of time, it may be necessary to change one‘s original stands, but a low corre— lation of one's campaign position and legislative roll call position would indicate that there was some misrepresenta- tion of the public policy preferences of the individual. As the next election approaches, and the memories of the last campaign fade, the representative can rely more heavily on either his attitudes or his perceptions in selecting a public policy position. 97 MUZWDHmZH NUZMDBHBmZOU ho mmfidm QMmH>mm N.m shaman mospfluum mosmsuwumcoo mo soHumoouom m.mumcauamo e7 cowuwmom now>mnmm HHmo Haom season oflansm mcsusun< m.m>wumusmmmnmmm m.mumofiosmo m.>osmsuwumsou 6 monunuua m.oumoflosmo 98 2. The‘l958 Representation Study In earlier chapters we have described the framework within which candidate decision making can be studied, as well as the concepts that are important in investigating this kind of problem. We now turn to a description of the data set that will be used to test the hypotheses about candidate behavior to be presented in the next section. Since the number of hypotheses that can be tested are limited by the available data, the data set is being pre— sented at this point in order to establish the context within which the hypotheses will be tested. In order to completely test the Bayesian model, we would need a survey of candidates that included questions about their perceptions, the means through which they received and evaluated information, their policy positions, and how their perceptions and policies changed over time. Since the resources to conduct such a survey for this study are not available, we shall instead utilize the data col- 1ected from the 1958 Representation Study. Although the study does not fulfill all our data needs, the study is quite comprehensive in its scope and can provide some important information that can be used as a basis for future investigations. Although the Representation Study was conducted during 1958-59, it has only recently beCome available for public examination through the services of the Inter-University Consortium for Political Research.6 As a result, only a few 99 articles based on the study have been published, mostly by the original investigators and their associates.7 Most of the articles have been concerned with the incumbents' behavior, rather than that of the nonincumbents in the sample.8 Despite the relatively few articles based on these data, and their primary concern with incumbents, the 1958 Representation Study remains one of the most significant studies of congressional candidates ever undertaken.9 The 1958 Representation Study was conducted between November 1958 and March 1959 by the Survey Research Center of the University of Michigan. It was done in conjunction with the 1958 National Election Study, which was a national sample of citizens of voting age. The Representation Study included interviews with a sample of incumbents from the 85th Congress, challengers of these incumbents, and in the case in which the incumbent did not run for reelection, the individual who was chosen to succeed the incumbent as the party's nominee (called the incumbent's successor). The sample of congressional candidates was chosen from the districts surveyed in the election study (114 districts) and in 32 other districts, for a total of 146 congressional districts.10 The purpose of the study was to "research the relation- ship between the attitudes and behavior of the electorate and the attitudes and behavior of their representatives."11 The principal investigators, Warren E. Miller and Donald E. Stokes, were especially concerned with whether their model 100 of constituency influence applied to different policy areas. They collected information about the attitudes of consti- tuents (from the election study), candidates, and legisla- tors in three main issue areas: foreign affairs, social welfare, and civil rights. The data were organized into three files: 1) a candi- date file, consisting of data on the 251 individuals inter- viewed. This group included incumbents and challengers in the 1958 congressional election, but in the districts in which the incumbents did not run for reelection, attempts were made to interview the incumbents' successors as well. This file is the one that will be used to test the hypothe- ses, since we are concerned with the candidate as the basic unit of analysis. 2) A district file, which uses the district (N = 146) as the basic level of analysis and the candidates are separated according to political party. Constituency data are also included in this file. 3) A district file, similar to the previous one, except that instead of organizing the data into Democratic and Republi- can components, they are divided according to which party controlled the congressional seat in the district. In order to establish proportionality across the districts, a weight variable was designated for each district 12 When the candidates are and another for each candidate. used as the unit of analysis (and when constituency data are not being used), a weighting factor for each candidate must be employed. Candidates from each of the districts are 101 weighted by either a value of 4.0 or 7.0 for a total of 1364 (13 l @ 4.0 + 120 @ 7.0): The candidate weights of 4 and 7 are the smallest whole integer values which maintain the original sampling probabilities given to congressional districts and provide a represent- ative sample of congressional districts in the United States.13 The individuals sampled were distributed from 39 differ- ent states and included 139 Democrats and 112 Republicans. When the weights were applied, the total number reached 1364: 781 Democrats and 583 Republicans. Table 3.1 TYPE OF CANDIDATE Variable 0005: Unweighted Weighted Code Description Frequency Frequency 0 Incumbent Opposed in General Election 94 511 1 Incumbent Opposed only in Primary 9 54 2 Incumbent Not Opposed in General Election or Primary 18 123 3 Nonincumbent who won General Election 26 137 4 Nonincumbent who lost General Election 94 493 5 Nonincumbent who had been in Congress before 2 11 6 Incumbent who sought election to other office ‘ . 4 16 7 Incumbent who retired from office’ A 3 12 8 Incumbent defeated in Primary Election 1 7 TOTAL 251 1364 102 'Tablgi3.2 TYPE OF CANDIDATE IN RELATION TO INCUMBENCY Variable 0007: Unweighted Weighted Code Description Frequency_y Frequency 00 Nonincumbent nonsuccessor who was defeated by an incumbent successor 6 30 10 Unopposed 85th Congress incum- bent who was reelected 27 174 11 Albert Thomas, Special Case of Code Ten 1 7 20 Opposed 85th Congress incumbent who was reelected 83 446 30 85th Congress incumbent who was defeated 10 61 40 85th Congress incumbent who did not run for Congress in 1958 8 35 50 Incumbent successor who defeated a nonincumbent nonsuccessor 9 42 60 Incumbent successor who was defeated by a nonincumbent nonsuccessor 5 26 70 Nonincumbent nonsuccessor who defeated an incumbent 12 69 80 Nonincumbent nonsuccessor who defeated an incumbent successor 5 26 90 Nonincumbent nonsuccessor who was defeated by an incumbent -:85 448 TOTAL 251 1364 The primary concern in this study is with those candi- dates who were opposed in the 1958 congressional elections. The categories used to select the opposed candidates are shown in Table 3.3. (See Table 3.3.) Tables 3.4 and 3.5 show the categories of candidates by incumbency and election outcome. Since we are concerned with the popular control of public policy and the 103 Table 3.3 TYPE.OF CANDIDATE BY COMPETITION Variable 0005: Unweighted Weighted Code Description Frequency Frequency 1,2,5,6,7,8 Unopposed in General Election 36 216 0,3,4 Opposed in General Election 215 1148 TOTAL 251. 1364 responsiveness of public officials, these tables provide us with some background information that will be of later use. Table 3.6 shows the distribution of opposed candidates by incumbency. Table‘3.4 TYPE OF CANDIDATE BY INCUMBENCY Variable 0005: Unweighted Weighted Code Description Frequency Frequency 0,1,2,6,7,8 Incumbent 129 723 3,4,5 Nonincumbent 122 641 TOTAL (251 1364. Table‘3.5 TYPE OF CANDIDATE BY ELECTION OUTCOME Variable 0007: Unweighted Weighted Code Description Frequency Frequency 20,50,70,80 WOn General Election 109 583 00,30,60,90 Lost General Election 106 565 10,11,40 Unopposed or Did Not Run 36 216 TOTAL 251 1364 104 Table 3.6 TYPE OF CANDIDATE: OPPOSED CANDIDATES BY INCUMBENCY Unweighted weighted Incumbents 93 507 Nonincumbents 122 . 1641 TOTAL 215 1148 One advantage of using the 1958 congressional elections as a data base is that it was a midterm election year and presidential politics played a much less important role in influencing state and local elections.14 Another factor is that since the party in control of the White House lost seats in the House of Representatives in 1958, we have a situation in which a large number of incumbents did not return to Congress.15 This fact will allow us to make comparisons between the effects of incumbency and election outcome, although the relationship between them remained strong (gamma = .928). Tables 3.7 and 3.8 show the distri- bution of the candidates based on incumbency and election outcome. Table 3.7 INCUMBENCY BY ELECTION OUTCOME: OPPOSED CANDIDATES (UNWEIGHTED) WOn Lost Incumbents 89.2% 10.8% (83) (10) Nonincumbents 21.3% 78.7% (26) (96) TOTAL (109) (106) Gamma = .936 Incumbents Nonincumbents TOTAL 105 Table'3.8 INCUMBENCY BY ELECTION OUTCOME: OPPOSED CANDIDATES TWEIGHTED) WOn Lost 88.0% 12.0% (446) (61) 21.4% 78.6% (137) _(504) (583) (565) Gamma.=.-928 The Hypotheses Now that the theoretical and empirical foundations of the study have been presented, we can now turn to the opera- tionalization of the two propositions presented in chapter two. To review, Proposition 1, which shall be referred to as the "uncertainty" proposition, is restated: deration for the limitations of the Representation Study, Proposition 1: Ceteris paribus, the smaller the variance of the subjective proba- bility distribution of a parameter 9, the more likely a candidate will consistenty choose his best estimate of 9 as the true value of e. In order to operationalize the proposition with consi- 16 we present the following interpretations of Proposition 1: Hypothesis‘l: Ceteris paribus, candidates who believe they khOw how people in their district feel about the issues are more likely to adopt a policy position that is close to what they perceive to be the majority opinion in their district. 106 The confidence, or variance of the probability dis— tribution of e, is represented by the extent to which the congressional candidate believes he knows the opinion of people in his district. His best estimate is his perception of constituency Opinion. His policy position is the posi— tion that represents the stands on the issues that are made known to the public. To test Hypothesis 1, for each of three issue areas, seven variables from the Representation Study are used. Tables 3.9, 3.10 and 3.11 show each of these variables and how they were recoded. Table 3.9 DESCRIPTION OF VARIABLE 0095: KNOWLEDGE OF DISTRICT OPINION Variable 0095: Knowledge of District Opinion Q: Do you think that you know how the rank and file voters in your district feel about issues like those we've talked about? Original Recoded all the time most of the time some of the time, sometimes, some issues seldom, not very often none of the time, never (1,2) most of the time (3) some time (4.5) seldom coo (”NI-4 Ulnb UNH FREQUENCIES: OPPOSED CANDIDATES ' Unweighted Weighted 1. most of the time 138 753 2. some time 18 87 3. seldom l9. . 103 TOTAL 175 943 107 Table 3.10 DESCRIPTION'OE ATTITUDE SCALES: VARIABLES 0042, 0054, & 0065 Variable 0042: Foreign Policy Attitude Scale Frequencies* Unweighted Weighted 34 187 0. Isolationist 40 220 l. Neoisolationist 63 339 2. Pro-Con 43 223 3. Neoactivist 35 176 4. Activist Variable 0054: Social Welfare Attitude Scale Frequencies* Unweighted Weighted 50 263 0. Conservative 49 274 1. 12 63 2. 11 56 3. 94 496 4. Liberal Variable 0065: Civil Rights Attitude Scale Frequencies* Unweighted Weighted 59 344 0. Conservative 20 104 l. 34 184 2. 102 513 3. Liberal *Opposed Candidates Only 108 'Table‘3.11 ' DESCRIPTION OF VARIABLES: PERCEPTION OF DISTRICT OPINION BY ISSUE AREA Variable 0195: Perception of District Opinion on Foreign Policy Q: How do people of your district feel about an active internationalist policy? WOuld you say that... Frequencies* Unweighted‘Weighted 50 317 1. most of them are opposed 41 224 3. they are fairly evenly divided 77 380' 5. most of them are in favor 37 175 7. not much district opinion on this (missing data) 5 20 8. don't know (missing data) Variable 0196: Perception of District Opinion on Domestic Issues Q: How do the people of your district like public power and public housing? Wbuld you say that... Frequencies* Unweighted Weighted 58 337 1. most of them are opposed 51 264 3. they are fairly evenly divided 70 358 5. most of them are in favor 23 113 7. not much district opinion on this (missing data) 4 22 8. don't know (missing data) Variable 0197: Perception of District Opinion on Civil Rights Q: How do the people of your district feel about desegre- gated schools and federal action to protect civil rights? WOuld you say that... Frequencies* Unweighted Weighted 28 187 1. most of them are opposed 21 102 3. they are fairly evenly divided 119 587 5. most of them are in favor 31 184 7. not much district opinion on this (missing data) 7 34 8. don't.know (missing data) *Frequencies for opposed candidates only. 109 The candidate attitude measures were based on multiple item scales, which were constructed according to the statis- tical procedures used to create Guttman scales. The assign- ment of the scale scores to individuals was done on the basis of the number of positive responses elicited by a set of questions in each of the three policy areas: foreign affairs, social welfare, and civil rights.17 The attitude scales represent the dependent variable, the candidates' public policy position. In their article, “Constituency Influence in Congress," Miller and Stokes used these scales to represent the congressmen's attitudes, or own personal policy preferences, separate and distinct from their roll call positions. In the campaign model we have presented here, these variables are interpreted as the candidates' public policy position and not as their own personal attitudes, although in some cases they may be the same position. Once their personal attitudes are known to the public, it is difficult to distinguish between their public and private position, or even difficult for a candi- date to admit that his own public policy position is really different from what he really believes in. If they are indeed different, then their personal policy preferences are unknown and not measured by the data. This interpretation is bolstered by the following argument. The actual wording of the questions that were used to construct the attitude scales indicates that the questions were not really asking for the candidates' 110 personal Opinions, but for their policy positions on the issues. For example, Variable 0038 reads: Now I would like to ask for a brief summary of your views on certain issues. I have a number of items here and I know that many of these questions are complicated ones. But what we are intereSted in are the basic stands that underlie your evaluation of specific bills or policies.18 (emphasis added) Variable 0045 reads: I know that a member of the House sometimes isn't able to vote on things that really reflect his own position. How well did (have) House roll calls dealing with foreign policy allow(ed) you to express your basic position on foreign affairs when you were in thefiflouse.19 (emphasis added) Variable 0046 reads: Reasons why the roll call votes have not reflected the (congressman's) basic position on foreign affairs.20 The fact that the frequency for response number 3 to Variable 0046 was zero21 (See Table 3.12) indicates that if their roll call position was different from their policy position, it was ngt_because they were following district opinion, but because they were attempting to follow their policy position, as measured by the attitude scale. The wording of the questions thus indicates that the respondents were expressing what they believed to be their open, public positions on the issues, and not necessarily their personal policy positions that may have in fact been different from the policies they advocated in the public forum. Given this interpretation, we are unable to determine the respondents' personal attitudes, but we can investigate at least whether 111 Table”3.12 DESCRIPTION'OF’VARIABLE‘0046 Variable 0046: Reason Why Roll Call Votes Have Not Reflected Congressman's Basic Position on Foreign Affairs Either-Or nature of roll call votes on bills does not allow for shades of opinion, considera- tion of particular provisions of bills. R‘s roll call Votes reflected wishes of R's desire to support President, Administration, Party Leadership in House. R's roll call votes reflected opinion in his Necessity of compromise; all legislation must Not much legislation in this area; Congress has not voted, has not been able to vote, on impor- tant questions in this area; policy has been decided by President, Administration. House rules and procedure; important decisions made in standing committees, by vote of Committee of the Whole rather than by roll call votes; floor consideration of bills brief, does not deal with basic questions; too many bills in final days of session; etc. Objective, implications of bills not always clear at time of roll call vote. R's roll call votes have reflected tactical considerations rather than basic position, e.g., R is really for a program, but has voted against it so that it will be re-evaluated. Frequency: All Candidates Weighted 178 l. 14 2. 0 3. district. 36 4. be a compromise. 48 5. 69 6. 15 7. 39 8. 7 9. Other reasons. 2322 . Inappropriate; no reasons mentioned, not congressman, etc. 112 their public policy positions were associated with their perception of constituency opinion. PrOposition 2, the risk hypothesis, predicts that the relationship between the perceived policy position of the district and the policy position of the candidate will be higher when there is some.expected loss or risk involved in the decision. For convenience, we restate Proposition 2: Proposition 2: Ceteris paribus, as the risk from choosing an alternative increases, the more likely a candidate will choose his best estimate of e as the true value of e. To operationalize the concept of risk, the Representa- tion Study provides a number of indicators that can represent a situation in which some expected loss could occur, if the candidate makes the"wrong" decision. Hypotheses 2 and 3 state how the idea of risk can affect the choice of a public policy position by a candidate: Hypothesis 2: Ceteris paribus, candidates who believe people in their district are interested in the issues are more likely to adopt a policy position close to what they perceive 'to be the majority position in their district. Hypothesis 3: Ceteris paribus, candidates who beIieve peopIe in their district are aware of their stands are more likely to adopt a policy position close to what they perceive to be the majority position in their district. These two hypotheses reflect the kind of incentives that may influence a candidate to be more responsive to whatluaperceives to be his district's opinion. Hypothesis 2 113 refers only to "issues" and Hypothesis 3 refers only to "stands,f and not specific policy areas. However, we can order these items with regard to the degree to which they refer directly to a candidate's position. Clearly, Hypo- thesis 3 is the more personal of the two, so we would therefore expect.a. stronger relationship to occur than with Hypothesis 2. Also, it may very well be difficult to distinguish district interest in the issues or stands of the candidate according to issue area, since if they are interested in one issue or stand, there may be some spill- over to another policy area. Table 3.13 shows the form of the variables, Variable 0094 and Variable 0169. (See Table 3.13.) To test Hypothesis 2, the variables measuring the candidates' perceptions of constituency opinion (see Table 3.11 for Variables 0195, 0196, and 0197) and their policy positions (see Table 3.10 for Variables 0042, 0054, and 0065) are required. The relationship between these variables for each policy area is controlled by Variable 0094, "People Interested in the Issues" (Table 3.13). To test Hypothesis 3, the relationship is controlled by Variable 0169, "People in the District Know the Candidate's Stands'I (Table 3.13). These results will indicate the extent to which candi- dates would be more likely to adopt their perception of constituency opinion because of some possible loss that may occur. Loss has been defined as the value of the outcome that results from the choice of an alternative given a 114 'Table 3.13 DESCRIPTION'OF’VARIABLE'0094:"PEOPLE'INTERESTED IN THE ISSUES, AND VARIABLEjpiGB: PEOPLE INDISTRICT KNOW CANDIDATEFS STANDS Variable 0094: People Interested in the Issues Q: How many people would you say there are in your district who are really interested and who keep up to date on issues like those we've talked about? Original Recoded 1. most; 80% or over 1. (1,2,3) most, some 2. many; 50-79% know 3. some; 20-49% 2. (4,5) none, very 4. very few; l—l9% few know 5. none; 0% Frequencies: Opposed Candidates weighted 500 1. most, some know 539 2. none, very few know Variable 0169: People in District Know Candidate's Stands Q: How much do you think the people of your district know about your stands on issues like those we've talked about? Original Recoded 1. know a great deal; stands widely 1. (1,2,3) know a known great deal and 2. know a good deal; stands fairly know some things widely known 2. (4,5) don't know 3. know some things; stands known at all somewhat 4. don't know very much; stands not very well known 5. don't know anything; stands not known at all Frequencies: Opposed Candidates Weighted 702 1. know a great deal and know some things 317 2. don't know at all 115 particular state of the world. Hypotheses 2 and 3 refer to a situation in which the losses in a decision matrix are perceived to be nonzero. We are predicting that as a candi- date perceives increasing scrutiny or knowledge of his policy positions by people in his district, the greater will be the value of the loss he perceives will occur. The way in which he can minimize that loss would be to choose his best esti- mate of the majority opinion in his district. Therefore, the association between his perception of district opinion and his policy position should increase to show a strong relationship. The actual measurement of expected loss is difficult to achieve from the data provided by the Representation Study, but we can identify the relative values attributed to four possible situations, as shown in Table 3.14. (See Table 3.14.) Table 3.14 presents decision situations involving the perceptions of the candidates with regard to their knowledge of district opinion and whether they believe people are interested in their position. 91 and 92 are two possible states of the world: the median policy position on an issue of concern to the voters in a congressional district. Al and A2 are two possible choices that a candidate can make: A1 = choose 91 as a policy position, and A2 = choose 62 as a policy position. The values, W, X, Y, and Z represent the values of the possible losses. When a loss is less than zero, the outcome is considered to be a net payoff. 116 Table 3.14 FOUR EXPECTEDgLOSS SITUATIONS SITUATION 1: EXPECTED LOSS = MODERATE TO HIGH AssumptiOns: Candidate Knows District Opinion Candidate Believes People are Interested in His Position P(el) > .5, P(ez) < .5 7 91 92 A1 w 5,0 x > 0 A2 Y > 0 72.5.0 SITUATION 2: EXPECTED LOSS = HIGH AssumptiOns: Candidate Does Not Know District Opinion Candidate Believes People are Interested in His Position P(el) = P(ez) = .5 .el . 92 A1 . w): o x > 0 A2 .Y.>.0 2.: o SITUATION 3: EXPECTED LOSS = LOW Assumptions: Candidate Knows District Opinion Candidate Does Not Believe People are Interested in His Position P(el) > .5, P(ez) < .5 91 92 A1 w‘: 0 , x 5'0 A2 Y‘: o ng o SITUATION 4: EXPECTED LOSS = MODERATE TO LOW Assumptions: Candidate Does Not Know District Opinion Candidate Does Not Believe People are Interested in His Position 9(91) = P(ez) = .5 91 92 A1 AW *0 x,: 0 Y A A2 ,:_0 z,_ IA <3 NOTE: A1 = select 91, A2 = select 92, and in all cases P(el) = 1.0 - P(92)' 117 Situation 1 in Table 3.14 shows the probability of 61 to be greater than .5. Thus, the decision maker is likely to choose Al, assuming the values of W and Z are about equal. The amount of expected loss involved in choosing Al would be perceived to be minimized since the candidate would choose his best estimate of 61, which is Al. We shall classify this situation as one of moderate'gg high expected loss. In situation 2, the average risk involved is likely to be much greater, since P(el) = .5. With no information to indicate one state of the world is more likely than another, the likelihood of making a wrong choice would be 50%. Thus the chances of losing votes would be much more likely than the previous situation, and so we would classify this as a high risk or high expected loss situation. In situations 3 and 4, the losses are all equal to or less than zero, since the candidate perceives that people are not interested in his position. From his point of view, it would not matter which position he adopted. However, if his perception about this situation proved wrong, his losses would likely be lower if his perceived probability of 61 was .greater than .5, as in situation 3. In situation 4, the probability of 61 is .5, and there is a greater chance of some loss when the choice of a policy position is based on a random choice. We therefore classify situation 3 as one of low expected loss and situation 4 as low 59 moderate expected’loss. These situations can be operationalized by the creation 118 of a new variable which takes into account the candidate's perception of district opinion and the degree to which he believes people in his district are interested in his posi- tion. The candidate's perception of district opinion is measured by Variable 0095, Knowledge of District Opinion. The perceived interest of one's constituency in policy issues is measured by a total of two indicators, Variable 0094, People Interested in the Issues, and Variable 0169, People Interested in Candidate's Stands (see Table 3.13). Tables 3.15 and 3.16 show how each of these two variables could be combined with the variable, Knowledge of District Opinion, to create a set of two new variables, "Expected Loss From Issues", and "Expected Loss From Stands." Table 3.15 CREATION OF A NEW VARIABLE: "EXPECTED LOSS FROM ISSUES" Variable 0094: People Interested in Issues Variable 0095: Egowlsdge 0E, MOST PEOPLE SOME PEOPLE NONE District Opinion MOST 17 2) 3y TIMES MODERATE MODERATE LOW , LOSS LOSS LOSS SOME 4) 57 6) TIMES MODERATE MODERATE MODERATE - LOSS LOSS LOSS 8 T 9) SELDQM HIGH MODERATE MODERATE LOSS LOSS LOSS 119 ' Table 3.16 gREATION OF A NEW VARIABLE: "EXPECTED LOSS FROM STANDS" Variable 0169: ’People Know Candidate's Stands Variable 0095: Knowledge of DiStrict Opinion MOST-PEOPLE SOME PEOPLE NONE MOST 1)MODERATE 2)MODERATE 3) Low TIMES LOSS LOSS. LOSS SOME 4)MODERATE 5)MODERATE 6)MODERATE TIMES LOSS LOSS LOSS 7) HIGH 8)MODERATE 9)MODERATE SELDOM LOSS. LOSS LOSS We shall first examine Table 3.15, which defines the variable, "Expected Loss From Issues." Cell 7 best repre- sents a case similar to situation 2 (Table 3.14), in which expected loss is high. When people are interested in the issues, but the candidate is not aware of district opinion, the risk will be at its highest. Cell 3 represents situation 3 in Table 3.14, in which the expected loss should be at its lowest. In this case, the candidate believes he knows district opinion most of the time, but since the people in his district are perceived not to be interested in the issues, the candidate's choice of a public policy position will not greatly affect his chances of winning. This is the low expected loss case. In the remaining cells, the expected loss can vary from 120 low to high. On the average, the expected loss can be con- sidered moderate, so the cells have been designated with this label. Table 3.16 illustrates the creation of the second new variable, "Expected Loss From Stands." It is formulated in the same manner as "Expected Loss From Issues." The only difference is the replacement of Variable 0094, People Interested in Issues with Variable 0169, People Know Candi- date's Stands. It has the same interpretation, except it measures the expected loss (risk) perceived from Situations in which the candidate believes people in his district know his stands on the issues. Although Variable 0094 and Variable 0169 are similar, it is expected that the use of "Expected Loss From Stands" will yield stronger results, since it measures to a greater extent the degree to which the candidate believes his constituents base their electoral decisions on the candidate's issue positions. Tables 3.17 and 3.18 show the frequency distribution of cases that fall within the High Loss, Moderate LoSs, and Low Loss categories. In an analysis, some difficulty may occur because the number of cases in the High Loss categories are relatively small. Therefore, the categories will be combined with the Moderate Loss categories. Some conceptual rigor may be lost in the recoding, but we prefer to use as much of the available data as possible to determine whether there are at least some generalized statements that can be made about candidates who are in a position to perceive low expected 121 loss as compared to those who perceive higher expected loss. Table 3.17 FREQUENCY DISTRIBUTION OF VARIABLE "EXPECTED LOSS FROM ISSUES”fl (OPPOSED CANDIDATES) Unweighted Weighted 5 29 High Loss 82 445 Moderate LOSS 58 322 Low Loss 145 796 TOTAL Table 3.18 FREQUENCY DISTRIBUTION OF VARIABLE "EXPECTED LOSS FROM STANDS" (OPPOSED CANDIDATES) Unweighted Weighted 8 47 High Loss 100 550 Moderate Loss 35 182 Low Loss 143 779 TOTAL To summarize, we have identified three types of expected loss for each of two variables, "Expected Loss From Issues" and "Expected Loss From Stands." These were recoded into two basic expected loss categories: High Loss and L !.£2§§- This classification allows us to describe the activities of candidates who have varying confidence in their perception of district opinion. For example, when candidates do not believe they know district opinion, but perceive people are interested in the issues and/or their stands, there is a 122 greater incentive for them to try to reduce the uncertainty in order to make the best possible policy decision. However, when the candidates believe they know district opinion and the people are not interested in the issues and/or their stands, their expected loss is perceived to be low. As a result, there is less incentive to collect new information and also less concern with trying to follow district opin- ion. This discussion leads us to a statement of the next hypothesis: Hypothesis 4: Ceteris paribus, candidates who percéive themselves to be in a High 'Expected Loss situation are more likely to adOpt a policy position close to what they perceive to be the majority position in their district than those who perceive themselves to be in a Low Expected Loss situation. This hypothesis can be tested using the variables "Expected Loss From Issues" and "Expected Loss From Stands" as control variables, with the candidates' policy positions in each of the three issue areas as the dependent variables, and their perception of district opinion as the independent variable. We expect that the results Obtained from the latter variable may be more decisive since it more directly concerns the citizens' perception of the candidate, rather than just the issues. This discussion has concerned the actions of a candi- date when he perceives some degree of expected loss. As defined earlier, expected loss is equivalent to risk. We have used the phrase "expected loss" interchangeably with 123 "risk", but favored the former in order to distinguish it from the following interpretation involving risk as an expected loss of an investment, rather than as an expected loss of votes. In a campaign, a candidate may be in one of the following three situations: 1) the nominee of the majority party in a safe district, 2) the nominee of the minority party in a safe district, and 3) the nominee of a party in a competitive district. The expected loss, or risk, is now defined as the expected loss of an investment. This includes all the tangible and intangible resources spent in order to attain political office,including investments made in pre- viously held offices which served as stepping stones for higher office. The investments made by candidates in each of these three situations are expected to vary. In the first case, the majority party candidate is likely to have the most investment to lose. Often one has to spend considerable time and money to win the party's nomination and a defeat in the general election would likely be a considerable loss not only to the majority candidate, but to his party as well. In the second case, the minority party candidate is not likely to invest as much in the race, since his chances of success are slim. It is indeed rare for a candidate to quit his job or go heavily into debt to run in a contest he is likely to lose. This assumes of course that the minority party candidate's goal is the nuclear office and is not 124 making the race for some other goal, such as appointment to a prestigious post by a high elected official of his party. The nature of the investment made by a majority party candidate is likely to be much higher than a minority party candidate. The expected losses of the two types of candi- dates will also differ and therefore, according to the model, their selection of a public policy position will also differ in the manner described by Hypothesis 5: Hypothesis 5: Ceteris paribus, in noncompetitive districts, the greater the risk, the more likely candidates who are opposed in the election will adopt a public policy position close to what they perceive to be the majority position in their district. Risk is defined in this case as expected loss of an investment and is operationalized by Variable 0159, Percep- tion of Party Strength. Table 3.19 describes the variable and outlines the recoding procedures. (See Table 3.19.) The risk perceived by candidates in competitive dis- tricts is classified as “competitive risk", in order to distinguish it from the ordinal risk associated with the two types of candidates from noncompetitive districts. In competitive districts, there is likely to be a mixture of candidates with different levels of investment, but these levels cannot be determined from the available data. Therefore, we are not attempting to compare the risk of candidates in competitive districts with those in noncompe- titive districts. Essentially, more factors can be con- trolled by comparing the level of risk within each of the 125 Table 3.19 DESCRIPTION OF VARIABLE 0159: 'PERCEPTION OF PARTY STRENGTH Variable 0159: Perception of Party Strength Q: How about the relative strength of the parties in the district. Over the years, has the district been a safe district, a fairly close district, or what? Original 0. safe Democratic district 1. fairly safe district; usuallygoes Democratic 2. fairly close; Democrats usually have edge 3. fairly close district; goes back and forth 4. fairly close district; Republicans usually have edge 5. fairly safe district; usually goes Republican 6. safe Republican district Recoded High Risk: A Republican who answers 5,6 A Democrat who answers 0,1 Low Risk: A Republican who answers 0,1 A Democrat who answers 5,6 Competitive A Republican who answers 2,3,4 Risk: A Democrat who answers 2,3,4 Frequencies: Opposed Candidates Unweighted ‘Weighted 61 322 High Risk 80 422 Low Risk 56 317 Competitive Risk 197 1061 TOTAL 126 two types Of districts. However, while comparisons can be made within noncompetitive districts, the data are not available to identify high and low risk candidates in com- petitive districts. If some comparison could be made between the expected loss of investment perceived by candidates in competitive districts and those in noncompetitive districts, we might expect that on the average the “competitive risk" might fall somewhere between "high risk" and "low risk." The average level of risk for candidates in competitive districts is an empirical question, but we can at least compare the associa- tion between the candidates' perceptions of district opinion and their policy positions for each level of risk (high, low, competitive) to determine if the ordinal scale has some validity. In Hypothesis 5, we controlled the basic relationship between a candidate's perception of district opinion and his public policy position by type of district (competitive or noncompetitive) and by type of risk (high risk, low risk, and competitive risk). We can go one step further by con- trolling for expected loss of votes to give us our final hypothesis: Hypothesis 6: Ceteris paribus, when controlling for expected loss of votes, the higher the risk (expected loss of investment), the more likely candidates who are opposed in the election will adopt a public policy position close to what they perceive to be the majority position in their district. 127 This is consistent with the previous hypothesis, but emphasizes the importance of risk in determining the candi- dates' policy positions. “Expected loss of votes" was ori- ginally defined as risk and the second interpretation of risk was "expected loss of investment." By controlling for both types of risks we would expect a confirmation of the predicted stronger association for the high risk rather than the low risk situation. Again we speculate that competitive risk, when controlled by expected loss of votes, will result in a level of association between the candidates' perception of district opinion and their public policy position that lies somewhere between the high risk and low risk values. The expected loss of votes in Hypothesis 6 can be operationalized by using the previously defined variable, "Expected Loss From Stands", shown in Table 3-16. The expected loss of investment has also been previously defined in Table 3.19. 4. Summary This chapter has described the operationalization of the uncertainty and risk propositions first presented in chapter two. From these two propositions, we proceeded to formulate six hypotheses that could be tested using data from the 1958 Representation Study. The 1958 Representation Study is the only major study of congressional candidates that attempted to measure candi- dates' perceptions of their district, their opposition in 128 the campaign, the effectiveness of their campaign activities, and in the case of incumbents, their congressional activi- ties. It is an ambitious study, encompassing 146 congres- sional districtszuM1251 incumbents and challengers. Since there was a long delay in making the data available to the public, researchers have only recently begun to analyze its findings. It is recognized that there are some problems inherent in the secondary analysis of survey data. Nevertheless, the advantages of using these data to test portions of the Bayesian model appear to outweigh any disadvantages. First of all, as Fiorina states, "...since these data are all that are available, we will accept them as sound..."22 At present there is no evidence to dispute the reliability of the data. Also, as long as we do not attempt to correlate constituency opinion with the candidate file data, we avoid the criticism that the small size of the sample of constituents in each congressional district (up to 17 per district) is unreliable for generalizing about constituents' true preferences. The candidate data are the primary focus of this research. Finally, since a model is an interpretation of a theory, there is more than one interpretation that can be used to operationalize a set of hypotheses. The Miller and Stokes data represent one interpretation and it is an empiri- cal question whether it is the most valid and reliable one. Nevertheless, it is recognized that additional tests will be needed, based on other data, in order to confirm the truth 129 value of the hypotheses. In order to do this, it is necessary to try to falsify the hypotheses, as Popper 23 recommends, and that is the subject of the next chapter. CHAPTER THREE NOTES 1. American Political Science Review, 57 (1963), 45- 56. See also the article by Stokes and Miller, "Party Government and the Saliency of Congress," Public Opinion Quarterly, 26 (1962), 531-546. 2. Warren E. Miller, “Majority Rule and the Represent- ative System of Government,“ in Cleavages, Ideologies, and Party_Systems, ed. by E. Allardt and Y. Luttunen (Helsinki: Transactions of the Westermarck Society, 1964), pp. 345-346. 3. See, for example, John Jackson, Constituencies and Leaders in Congress (Cambridge: Harvard University Press, 1974), as well as the earlier works of Julius Turner and Edward Schneier, Party and Constituency: Pressures on Congress, rev. ed. (Baltimore: Johns Hopkins Press, 1971) and Wayne Shannon, Party, Constituency_and Congressional Voting(BatonRouge: Louisiana State University Press, 1968). 4. Miller and Stokes, "Constituency Influence in Congress," American Political Science Review, 57 (1963), 50. 5. Stokes and Miller, op. cit., and Stanley R. Freed- man, "The Saliency of Party and Candidate in Congressional Elections: A Comparison of 1958 and 1970,“ in Public 130 131 Opinion and Public Policy, ed. by Norman R. Luttbeg, rev. ed. (Homewood, Illinois: Dorsey Press, 1974), pp. 126-131. 6. The official title of the survey is-The‘1958 American Representation Study: CongreSSmen and Consti— tuents, (SRC 433). The first printing of the codebook for the District File is August 1971 and for the Candidate File, 1970. 7. Miller and Stokes have been working on a manuscript, The Structure of Representation, but in a 1974 communication with Professor Stokes, he indicated that "Our manuscript is' not yet in a form that you could draw much substance from it." 8. The primary articles based on the 1958 Representa- tion Study include: Miller, op. cit., Miller and Stokes, op. cit., Stokes and Miller, Op. cit., Stokes, "Compound Paths in Political Analysis," in Mathematical Applications in Political Science V, ed. by James Herndon and Joseph L. Bernd (Charlottesville, Virginia: University of Virginia Press, 1970) (revised in American Journal of Political Science, 18 (1974), 191—214), Stokes, “Electoral System and Representation: United States and the United Kingdom," paper presented for delivery at the 1967 Annual Meeting of the American Political Science Association, Chicago, Illinois, September, 1967, David R. Segal and Thomas B. Smith, “Congressional Responsibility and the Organization of Constituency Attitudes,“ in Political Attitudes and‘Public ‘Opinion, ed. by Dan Nimmo and Charles Bonjean (New York: 132 David McKay, Inc., 1972), pp. 562-568, and Charles F. Cnudde and Donald J. McCrone,_"The Linkage Between Constituency Attitudes and Congressional Voting Behavior: A Causal Model," American Political Science Review, 60 (1966), 66-72. A dissertation based on the Representation Study is Helmut Norpoth, "Sources of Party Cohesion in the United States House of Representatives" (Unpublished Ph.D. Dissertation, University of Michigan, 1974). See also his article, "Explaining Party CoheSion in Congress: The Case of Shared Policy Attitudes," American Political Science Review, 70 (1976), 1156-1171. 9. Other studies of congressional candidates include: Robert Huckshorn and Robert Spencer, The Politics of Defeat (Amherst: University of Massachusetts Press, 1971), David Leuthold, Electioneering in a Democracy (New York: John Wiley, 1968), John Kingdon, Candidates for Office (New York: Random House, 1968), Jeff Fishel, Party and Opposition (New York: David McKay, Inc., 1973), and Charles S. Bullock, III, "Candidate Perceptions of Causes of Election Outcome," paper presented at the 1973 Annual Meeting of the American Political Science Association, New Orleans, 1973. Although these studies were quite ambitious, none matched the large sample of the Representation Study. 10. The original sample was 151 districts, but no can— didate or incumbent interviews were obtained for 4 districts, so they were dropped. Constituency data were available from 114 districts. 133 11. See Study Description, SRC Codebook 433. 12. For information on the sampling and weighting procedures used in the study, see Miller and Stokes, op. cit., and Stokes, "Electoral System and Representation," op. cit. 13. See Study DescriptiOn, SRC Codebook 433. 14. For the factors influencing outcomes of midterm elections, see Edward R. Tufte, "Determinants of the Out- comes of Midterm Congressional Elections," American Politi- cal Science Review, 69 (1975), pp. 812-826, and James E. Pierson, “Presidential Popularity and Midterm Voting at Different Electoral Levels," American Journal of Political Science, 19 (1975), 683-694. 15. For the 86th Congress, there was a net change of +49 Democrats and -47 Republicans, taking into account special elections and appointments between elections. Source: Congressional Quarterly weekly Report, 32 (November 9,1974), p. 3105. 16. Limitations of the data source are essentially those which are inherent in the use of most kinds of secon- dary data. These include the fact that rarely does one find that the original researchers have asked the questions in their survey according to the requirements of other analysts. Also,-secondary analysis allows for the increased chance of error in coding and interpretation. A strict test of the Bayesian model will have to wait until a future time, but meanwhile, the Miller and Stokes data provide the closest 134 means, as well as the best means available for testing the model. For a discussion of the issues involved in the secondary analysis of survey data, see Herbert H. Hyman, Secondary Analysis of Sample SurVeys (New York: John Wiley, 1972). 17. Three additional variables that can indicate an expected loss situation based on the relative importance attached to the three issue areas of foreign affairs, social welfare and civil rights will be introduced in the next chapter. They are not presented at this time because they are less directly related to constituents' interests. Instead, they deal with how important the candidates perceive these issue areas to be of importance to the voters. 18. Candidate File‘COdebOOk, 1971, p. 28. 19. Ibid., p. 31. 20. Ibid. 21. The frequencies for the equivalent question with regard to the social welfare and civil rights issue areas were also zero. 22. Morris P. Fiorina, Representatives, Roll Calls, and Constituencies (Lexington, Mass.: Lexington Books, 1974), p. 18. 23. Karl R. Popper, The LOgic of Scientific Discovery (New York: Harper and Row, 1968), Chapter 4. CHAPTER FOUR THE ANALYSIS OF UNCERTAINTY AND RISK 1. Introduction According to the Bayesian model, uncertainty and risk are important motivating factors which influence the deci- sion making by candidates in an election campaign. Two propositions relating to uncertainty and risk were opera- tionalized to permit us to state six hypotheses that could be tested using the 1958 Representation Study. These hypotheses describe the conditions under which candidates are more likely to follow their best estimate of the state of the world. The state of the world in this case is the true policy position which lies at the median of the voter's policy preference distribution. The analysis is divided into basically two parts. The first concerns an investigation of the perceived uncertainty of the candidates and the kinds of sources they were likely to depend upon for information. The second part involves the testing of the six hypotheses dealing with uncertainty and risk and their effect upon the candidates' policy positions. Throughout this chapter, we will be using two addition- al independent variables as controls: incumbency and elec- tion outcome. Research has shown that incumbents are likely 135 136 to have different perceptions and expectations from non- incumbents and these differences can be illuminated by controlling for incumbency.l Election outcome has not been a theoretical concern, because we have not included voters' perceptions in the model. Therefore, we cannot predict electoral outcomes, but we can determine whether the elec- tion results reflect the candidates' desires to win election by attempting to follow district Opinion. In a representative democracy, candidates are expected to mirror their constituents' opinions, but, as we have seen, the incentive is not always present, nor is complete represen- tation always possible to achieve. By examining the candi- dates' choice of a public policy'position according to the results of the 1958 election, we can obtain some evidence to determine whether the people are sending to Congress candidates who are at least trying to be representative of what they perceive to be their district's majority opinion. A.more detailed consideration of the responsiveness issue will be presented in the next chapter. 2. The Statistical Analysis As a prelude to the statistical analysis to follow, some discussion concerning the use of the statistical measures is in order. Most of the data in the 1958 Representation Study were 2 either on a nominal or an ordinal scale of measurement. Some researchers find this no hindrance in making the 137 necessary assumptions and hunches to permit the use of statistics normally reserved for interval scale data. In fact, Shively believes, One mark of a good researcher should be that he boldly seeks out all chances--not just the obvious ones, not just the safe ones—-to raise the level of measurement in his work.3 By the “safe ones" he means measuring at a level at which we can be relatively confident of the things we say about the results, and at the cost of saying less interest- ing things about the variables being measured.4 The positions of both social scientists and statisti- cians do not indicate clear agreement on this point. Abelson and Tukey, for example, have examined the problem of assigning metric values to an ordinal scale, but found the problem was often that "When we say we only know rank order, we actually know more than this, but don't know how to express what else it is we know."5 Blalock has discussed this problem with regard to scale construction: These examples should be sufficient to indicate that it is often not a simple matter to decide what type Of scale can legitimately be used. Ideally, one should make use of a data— gathering technique that permits the lowest levels of measurement, if these are all the data will yield, rather than using techniques which force a scale on the data.6 The use of a particular scale is important because it establishes bounds on the appropriateness of statistical operations,7 hence, we will be guided by Galtung's advice: 138 But in the measurement of correlation or agreement, the rule is invariably that the lowest level determines what coefficient to use.8 The reasoning behind such cautious behavior is well stated by Singh: ...often times it is neither reasonable nor necessary to treat ordinal variables as interval variables and for that matter the practice can be quite reasonable under most circumstances. It is not being suggested that we abandon our approach toward mathema- ticization but it should be kept in mind that a faith blended by trust in mathematical jargon rather than the logic of mathematics is no panacea for constructing causal models. It might be added that the process of mathema- ticization and for that matter use of higher levels of statistical techniques is not only commendable but a necessary first step toward our eventual goal of theory construction from axiomatic and deductive perspectives. EEE.EE must be aware of what our inputs'are in_con- strucEing such’fibdels.9__(emphasis added7——- Since the data from the Representation Study consisted mostly of nominal and ordinal data, it would therefore be advisable to use only the statistical techniques appro- priate for these levels. This would preclude the use of regression analysis, since nominal and ordinal scales10 do not permit mathematical operations on their values.11 The data from the Representation Study were analyzed by employing the Crosstab and Frequency Programs Of the Statistical Package for the Social Sciences, version 6.02.12 In the following tables, the statistic that will be reported is Goodman and Kruskal's Gamma.l3 Gamma is usually used to measure the association between two ordinal variables 139 and is symbolized by: Gamma =_E_ZIEL. P + Q Where P = the number of concordant pairs Q = the number of discordant pairs A concordant pair exists if in a contingency table a case falls below and to the right of another case, and a discor- dant pair exists if a case falls below and to the left of another case.14 In comparing various ordinal measures of association, Buchanan commented that the use of gamma has an advantage in that it can vary from -l.0 to +1.0, so as to create some uniformity in our measures, and because it is sensitive to limited, curvilinear or triangular associations, where other measures, such as tau-c, are not.15 Singh compared five measures of association to Kruskal's criteria that a measure should have, and found that gamma was able to satisfy these criteria: 1) simplicity of interpretation, 2) reasonable sensitivity to form of dis- tribution, and 3) relative simplicity of sampling theory.16 Gamma therefore seems to be an adequate and appropriate measure to be applied to the kind of data present in the Representation Study. There is, however, one note of caution that needs to be considered in the use of gamma. This is the case in which the unequal size of the marginals of the independent variable causes a distortion of the value of gamma. AS Bruner states, 140 A relationship that is large when column marginals are equal may shrink when they are not. And a relationship that looks small when marginals are grossly unequal, may be large when they are equalized.17 Bruner explained that when one is working with an implied causal hypothesis, which is concerned with the effect of group differences, especially in terms of condi- tional probabilities, comparisons of gamma between tables may be interfered with by distortions due to marginal disparity.18 Gamma is not affected by column-marginal disparity in two—column tables, but in others, it is necessary to transform the cell frequencies by equiweight- ing. Equiweighting maintains the same column percentages, but provides new column total raw frequencies. The formula for equiweighting is as follows: I — u a ij - aij (l/C) (N/nj) Where aij the cell frequency before equiweighting a'ij = the new cell frequency after equiweighting c = the number of categories in the indepen- dent variable the column total before equiweighting of the column in which a cell is located N = the total number of cases in the table.19 nj Each new cell frequency can then be used to recalcu- late gamma in order to eliminate variations of gamma as a result of column-marginal disparities, which is important when one is asking a quasi-experimental question, one about the conditional probability Of the various dependent outcomes for each category of the independent variable,20 and not when one is 141 asking about the causal impact of differences in the independent variable upgn the dependent outcome for the whole sample. This procedure of equiweighting the gammas will be performed on the tables presented herein, excluding those that are primarily illustrating association, rather than causation, and those in which no controls have been added to the tables. However, in order to present the gammas in an orderly fashion, the original gammas will be given, and in cases where the equiweighted gammas are relevant and significantly different will they be shown and presented in footnotes in the appropriate tables.‘ The gammas are calculated using the SPSS Crosstab routine and the equi- weighted gammas are produced from a desk calculator. 3. Uncertainty, Information and EffOrt In this section, we shall investigate the perceived uncertainty of the Opposed candidates in the 1958 congres- sional elections and hOw the candidates acted within their environment to deal with this uncertainty. Of particular interest will be an examination of the effort.that they may have made to decrease uncertainty about district opinion, especially in light of any expected losses from making decisions under uncertainty. The first factor to be considered is the perceived uncertainty within the campaign environment. Uncertainty has been proposed as an important intervening variable that can be used to explain candidates' policy positions. 142 Before examining these positions, we would first like to know if different kinds of candidates are more uncertain than others. For example, is uncertainty related to incumbency, election outcome, and the number of times a candidate has run for office? (See Tables 4.1, 4.2, 4.3, and 4.4.) Table 4.1 presents the_gamma associations of the relationship between Variable 0005, Knowledge of District Opinion, and variables indicating incumbency, the election outcome, and the number of times the candidate ran for Office. Tables 4.2, 4.3 and 4.4 present the frequency distributions exhibited by theSe relationships. The ques- tion asked by Variable 0005 was, "Do you think that you know how the rank and file voters in your district feel about issues like those we've talked about?" The results , show that incumbents were more likely to know district opinion "most of the time," while nonincumbents were seldom likely to know how people in their district felt about the issues. Among winners and losers, the relationship is stronger, with winners more likely to perceive knowledge of district opinion most of the time and losers more likely not to perceive knowledge of district Opinion compared to the winners. The relationship between knowledge of district Opinion and times ran for Office is the weakest of the three, but indicates that of the candidates in the “seldom" category, 65.5% had neVer run for Congress previously. 143 Table 4.1 KNOWLEDGE OF DISTRICT OPINION BY INCUMBENCY, ELECTION OUTCOME AND THE NUMBER OF TIMES A CANDIDATE RAN ’FOR'CONGRESS Variable 0095 Knowledge of District Opinion By: "Opposed Candidates Incumbency .47 (N = 943) Election Outcome .57 (943) Times Ran for Congress .37 (890) Table’4.2 ’KNOWLEDGE OF DISTRICT OPINION BY INCUMBENCY: CONTINGENCY'TABLE Variable 0095 Knowledge of District Opinion By Variable 0005 Incumbency Knowledge of District Opinion Most of the time Some times Seldom Incumbengy 'Incumbents Nonincumbents 37.5% 72.9% (N = 392) (361) 10.7% 7.9% (48) (39) 1.8% 19.2% (8) (95) 100% 100% (448) (495) 144 Table'4.3 KNOWLEDGE OF DISTRICT OPINION BYJELECTION OUTCOME: 'CONTINGENCY TABLE Variable 0095 Knowledge of District Opinion By Variable 0007 Election Outcome 'Election outcome Knowledge of District Opinion Winners Losers Most of the time 88.9% 69.1% (455) (298) Some times 8.0% 10.7% (41) (46) Seldom 3.1% 20.2% .(16) (87) 100% 100% (512) (431) Table 4.4 KNOWLEDGE OF DISTRICT OPINION BY NUMBER OF TIMES A CANDIDATE RAN FOR OFFICE (CONGRESS): CONTINGENCY TABLE Variable 0095 Knowledge of District Opinion By Variable 0254 Number of Times a Candidate Ran for Office (Congress) Knowledge Times Ran For Office of District inion 8+ 7 6 5 4 3 2 l 0 Most of 83.9% 100 100 100 100 88.6 76.2 66.4 74.7 the time (73) (40) (41) (36) (43)(93) (48) (79) (266) Some 16.1 0 O 0 0 7.6 6.3 21.8 9.8 times (14) (0) (0) (0) (0) (8) (4) (26) (35) Seldom 0 0 0 0 0 3.8 17.5 11.8 15.4 (0) (0) (0) (0) (0) (4) (11) (14) (55) 100% 100% 100% 100% 100%100% 100% 100% 100% (87) (40) (41) (36) (43)(105) (63)(ll9) (356) 145 These findings help support our thesis that candidates with more experience are likely to be more confident about knowing district opinion. In addition, it appears that the voters elected a group of candidates who perceived them- selves to be somewhat more likely to know district opinion than the ones previously elected. The next question is.concerned with whether candidates that perceive the possibility of some expected loss occurring from their decisions differ according to incum- bency, election outcome, and the number of times they ran for Congress. we can answer this by examining the relation- ship of each of these three variables with the two variables measuring expected loss, which were developed in Chapter three. These variables were "Expected Loss From Issues" and ”Expected Loss From Stands." The former was based on a combination of Variable 0095, Knowledge of District Opinion and Variable 0094, People Interested in the Issues. Variable 0094 asked the question, "How many people would you say there are in your district who are really inter— ested and who keep up with the issues like those we talked about?" Candidates were divided into three categories and later recoded into two: low expected loss and high expected loss. Similarly, Variable 0095, Knowledge of District Opinion, was combined with Variable 0169, People in District Know Candidate's Stands, which asked the question, "How much do you think people of your district 146 know about your stands on issues like those we've talked about?" to produce two additional expected loss categories of high and low expected loss. These two variables do not measure expected loss by issue area, but only to the extent to which candidates believe their constituents are "interested in the issues" or "aware of their stands."' In order to obtain some measure of expected loss according to the issue areas of foreign affairs, social welfare and civil rights, we can combine Variable 0202, Importance of Issues to Voters with Variable 0095, Knowledge of District Opinion, to create three new variables: Expected Loss From Foreign Affairs, Expected Loss From Social Welfare, and Expected Loss From Civil Rights. These variables measure the amount of loss perceived when the candidates' knowledge of district Opinion is combined with their perception that people in their district place some importance on these issues. The pro- cedure for creating these three new variables is shown in Table 4.5. (See Table 4.5.) To summarize, the concept of expected loss is now operationalized by five different variables: 1) Expected Loss From Issues represents a situation in which a candi- date may perceive to be in a position to lose votes because people in his district are interested in the issues, 2) Expected Loss From Stands represents the case in which the perceived loss may result from people knowing or being aware of the candidate's stands on the issues, 3) Expected 147 Table 4.5 CREATION OF NEW VARIABLES: "EXPECTED LOSS FROM FOREIGN AFFAIRS," :ExPECTED LOSS FROM SOCIAL WELFARE ISSUES," AND "EXPECTED LOSS FROM CIVIL RIGHTS ISSUESI-' Variable 0202: Importance of Election Issues in District: Q: WOuld you say that any of these broad categories of issues were particularly important to the voters in your district in the election? What were they? 00. No; None; or Not Opposed in Either Primary or Election 10. Yes, Foreign Affairs 20. Yes, Domestic Issues 30. Yes, Civil Rights Issues 40. Yes, Foreign Affairs and Domestic Issues 50. Yes, Foreign Affairs and Civil Rights 60. Yes, Domestic Issues and Civil Rights 70. Yes, All Three Categories: Foreign Affairs, Domestic Issues, and Civil Rights 80. Yes, Not Available Which Category Reclassification: Variable 0095 Issue Area* Knowledge of District Opinion 'Important Not Important Most of the time Moderate Loss Low Loss Some times Moderate Loss Moderate Loss Seldom High Loss Moderate Loss Frequencies: 'Expected Loss From: Foreign Affairs Social Welfare Civil Rights Unw. weighted Unw. Weighted Unw. Weighted High Exp. Loss 3 15 1 4 l 4 Mod. Exp. Loss 44 245 65 359 47 236 Low Exp. Loss 97 526 78 423 96 546 *The "Important“ and “Not Important" categories were created for each of the three issue areas according to the responses to Variable 0202. For example, for Foreign Affairs, the "Important" category included all those who responded 10,40, 50, and 70. The “Not Important" category included those who responded 00,20,30, and 60. Code 80 was a missing data category. 148 Loss From Foreign Affairs represents a possible loss situation when foreign affairs is believed to be an impor- tant issue among the voters, 4) Expected Loss From Social welfare represents a possible loss situation when social welfare issues are believed to be important to the voters, and 5) Expected Loss From Civil Rights can occur when civil rights issues are important to the voters. Tables 4.6, 4.7, and 4.8 show the relationship of these measures of expected loss to l) incumbency, 2) election outcome, and 3) the number of times a candidate ran for office. (See Tables 4.6, 4.7, and 4.8.) In Table 4.6, the highest association is Expected Loss From Civil Rights and Incumbency. .The negative asso— ciation indicates that incumbents were more likely to perceive low expected loss than nonincumbents. Nonincum— bents were more likely to perceive some expected loss. In Table 4.7, the highest association is again the one with Expected Loss From Civil Rights. The interpre- tation is that winners were more likely to perceive lower expected loss than the losers. Similarly, in Table 4.8, the strongest association is with Expected Loss From Civil Rights, in which candidates who had more experience running for office were more likely to perceive low expected loss when people in their district were perceived to consider civil rights an important issue. 149 Table 4.6 EXPECTED LOSS BY INCUMBENCY ”Opposed'Candidates Gamma TN)— Expected Loss From Issues -.04 (796) Expected Loss From Stands .21 (779) Expected Loss From Foreign Affairs -.24 (786) Expected Loss From Social Welfare .11 (786) Expected Loss From Civil Rights -.52 (786) Table 4.7 EXPECTED LOSS BY ELECTION OUTCOME Opposed Candidates Gamma (N) Expected Loss From Issues .08 (796) Expected Loss From Stands .35 (796) Expected Loss From Foreign Affairs -.10 (786) Expected Loss From Social Welfare .29 (786) Expected Loss From Civil Rights -.32 (786) Table 4.8 EXPECTED LOSS BY NUMBER OF TIMES CANDIDATE RAN FOR OFFICE Expected Expected Expected Expected Expected Qpposed Candidates Gamma (N)_ Loss From Issues -.05 (747) Loss From Stands .19 (744) Loss From Foreign Affairs -.09 (737) Loss From Social Welfare -.02 (737) Loss From Civil Rights -.40 (737) 150 It appears that although civil rights was perhaps an important and volatile issue in their districts, incumbents, winners and those with campaign experience perceived them- selves in a low expected loss Situation, compared to non- incumbents, losers, and inexperienced campaigners. Of course, if the candidates' perceptions are wrong and they presume low expected loss and don't worry about their positions on civil rights, the result could be their defeat. However, it appears that on this issue above the others, the incumbents, winners, and experienced campaigners are more likely than their counterparts to perceive they know district opinion most of the time and that people in their district don't think civil rights is an important issue. There may be a nwmber of reasons for this, one of which may be the fact that the election was decided on other issues which were kept alive by the incumbents to avoid discussing civil rights or because the incumbent had well represented his district in this area and there was low expected loss because a large majority was perceived to agree with his voting on this issue. Challengers may perceive more expected loss because of their perception that civil rights is indeed an important issue that could be potentially dangerous if they do not accurately perceive majority opinion. With the other issues, the majority position may be less Clear to the incumbent and so we find less of a dis— tinction between the experienced and less experienced 151 candidates. The next topic to be discussed after this considera- tion of the uncertainty and risk perceived by the candi- dates involves the type of sources the candidates relied upon for information. The Bayesian model is based On the assumption that more information is preferred to less infor- mation and that candidates who are less informed will be more likely to seek information until its cost becomes prohibitive or the election interrupts his search. The model does not predict which sources the candidates would rely upon, but based on what we have learned about the difference between the perceptions of incumbents and non- incumbents, winners and losers, and experienced and unex- perienced campaigners, it is possible that some empirical generalizations can be gleaned from the data. The 1958 Representation Study does not include an extensive analysis of attempts by the candidates to collect information, but there is some information that may prove useful. There are two potential drawbacks. One is the fact that the survey is of course a oneéshot case study and no comparisons can be made to determine whether some source of information changed their perceptions. The Second is that these questions were posed to candidates after the election, so it is difficult to be able to determine the causal direction of the empirical generalizations. Also, the "congratulationérationalization" effect discovered by Kingdon may operate, causing the winners and loserstx>change their perceptions based on the election outcome.22 Never- theless, the data may provide some interesting results that could be generalized to similar types of candidates and indicate some important avenues of future research. The sources of information that candidates were asked about included newspapers, public opinion polls, people in their party organization, personal contacts, (and for in- cumbents only, their mail). Table 4.9 shows the questions that were asked and the responses received. Table 4.10 shows the relationship between dependence on each of these sources with incumbency, election outcome, and the times a candidate ran for office.23' (See Tables 4.9 and 4.10.) The results in Table 4.10 show a positive relationship for newspapers and a negative relationship for people in the party organization and personal contacts. This means that incumbents, winners, and frequent campaigners were more likely to depend on newspapers than nonincumbents, losers, and people running for Congress for the first time. On the other side, these latter types were more likely to rely on peOple in their party organization and on their personal contacts than were the incumbents, winners, and frequent campaigners. There was no difference in their use of polls or dependence on their mail. These findings appear to be consistent with Kingdon's research in which the candidates seemed to be suspicious of the party organization as a source of information because party workers tended to overestimate the candidates' 153 Table'4.9 DESCRIPTION OF VARIABLES IDENTIFYING 'SOURCES OF CAMPAIGN INFORMATION Variable 0120: Dependence on Newspapers Q: How much do you depend on newspapers and editorials to tell you what opinion is in your district on issues like these? Frequency: All Candidates Weighted 71 1. Very much; Chief way of finding out what opinion is ‘ 244 2. Quite a bit 246 3. Somewhat 420 4. Not very much 290 5. Not at all 93 9. NA Variable 0121: Dependence on Public Opinion Polls Q: How much do you use opinion polls to measure district opinion? Frequency: All Candidates Weighted 62 1. Very much; chief way of finding out what opinion is ' 76 2. Quite a bit 151 3. Somewhat 212 4. Not very much 792 5. Not at all 71 9. NA Variable 0123: Dependence on People in the Party Organization Q: How much do you depend on people in the party organization to measure district opinion? Frequency: All Candidates Weighted 1. Very much; chief way of finding out what opinion is 186 2. Quite a bit 246 3. Somewhat 293 4. Not very much 403 5. Not at all 99 9. NA 154 Table'4.9 (Cont'd) Variable 0125: Dependence on Personal Contacts Q: Frequengy: Wei hted 912 1. 205 67 74 14 92 2. 3. 4. 5. 9. How much do you depend on personal contacts to measure district opinion? All Candidates Very much; chief way of finding out what Opinion is Quite a bit Somewhat Not very much Not at all NA . 155 Table 4.10 DEPENDENCE ON INFORMATION SOURCES BY INCUMBENCY, ELECTIONOUTCOME, AND NUMBER OF TIMES A "CANDIDATE RAN FOR OFFICE A. Dependence on Source By Incumbency: ‘Opposed Candidates Source: Gamma "(N) Newspapers 1 .43 (1091) Public Opinion Polls .11* (1106) People in the Party Organization -.44 (1071) Personal Contacts -.61 (1075) B. Dependence on Source By Election Outcome: Opposed Candidates Source: Gamma (N) Newspapers .38 h (1087) Public Opinion Polls .03* (1102) People in the Party Organization -.27 (1067) Personal Contacts -.49 (1071) C. Dependence on Source By Number of Times a Candidate Ran For Office:r Opposed Candidates Source: Gamma (N) Newspapers .28 (1017) Public Opinion Polls . -.03 (1032) People in the Party Organization —.36 (1000) Personal Contacts -.43 (997) *Chi-square value for this table is not significant at .05 level of significance. 156 popularity and election chances.24 The more experienced candidates are likely to know this while such contacts may help give the nonincumbents the encouragement and confidence that is welcomed in an uphill race against an incumbent. Nevertheless, the incumbent may be more skeptical of per- sonal contacts through experience. As Kingdon quotes one candidate, "People are friendly and nice, but you can't rely on it for votes."25 0n the campaign trail, many people will be polite, but often the number of friendly voters who speak to the candidates can exceed the number of votes they receive in the election. An incumbent may know this situa— tion well, as everyone after the election claims to be the one to contribute to the candidate's victory: I've never met people while I was campaigning who weren't going to vote for me, and you never meet people afterward who voted against you. If incumbents do not depend on sources that tend to be unreliable, then is their perceived knowledge of district opinion related to the sources they do depend on? Table 4.11 displays the relationship between Variable 0095, Knowledge of District Opinion, and the candidates' depen- dence on each source of information. (See Table 4.11.) Candidates who believe they are likely to know district opinion most of the time are more likely to depend on newspapers and public opinion polls. Since newspapers and especially public opinion polls are likely to be more reliable than the other sources, they may have some influence on the perceptions of the candidates, 157 Table 4.11 DEPENDENCE ON INFORMATION SOURCES BY KNOWLEDGE "OF'DISTRICT OPINION V Opposed Candidates Source: Gamma (N) Newspapers .39 (917) Public Opinion Polls .49 (920) People in the Party Organization —.13* (888) Personal Contacts —.10 (905) * Chi-square value for this table is not Significant at .05 level of significance although we cannot verify this as the true causal direction. When we control for incumbency and election outcome, the results shown in Table 4.12 were obtained. (At this point we are dropping the use of Variable 0254, Number of Times a Candidate Ran for Congress, since it has tended to mirror the results obtained from the other two control variables.) (See Table 4.12.) The largest Changes in the relationship between dependence on a source and knowledge of district opinion when controlling for incumbency occur for dependence on public opinion polls and people in the party organization. For the former, the incumbents were much more likely to depend on polls when they felt they knew district opinion. There was also an increase in a positive direction for dependence upon people in the party organization among incumbents, but the X2 value for that table was not significant. 158 Table 4.12 DEPENDENCE ON INFORMATION SOURCES BY KNOWLEDGE 9E DISTRICT OPINION CONTROLLED FOR INCUMBENCY AND ELECTION OUTCOME* A. Incumbency ’Opposed Candidates Source: ‘Incumbents 'Nonincumbents Newspapers .37 (434) .27 (483) Public Opinion Polls .60 (440) .43 (480) People in the Party Organiza- tion .32 (419)** -.18 (476) Personal Contacts .01 (429) .17 (476) B. Election Outcome Opposed Candidates Source: Winners Losers Newspapers .33 (498) .28 (419) Public Opinion Polls 1.00 (504) .39 (416) People in the Party Organiza— tion .06 (490)** -.15 (398) Personal Contacts -.21 (486)** .22 (419) * It is appropriate to use the equiweighted gammas when one is stating a causal hypothesis in terms of conditional probabilities. Although some causal inferences can be made from these data, we are not testing any causal hypotheses at this time, but merely showing the associa- tion between variables. Therefore, in Tables 4.11, 4.12, 4.13, and 4.14, we are interpreting the original gammas. **The chi-square values for these tables are not significant at the .05 level of significance. 159 When the relationship is controlled by election out- come, public opinion polls appear to be a more popular source for winners. Dependence on personal contacts shows a reversal of the direction of the relationship shown for incumbency. This may tend to give support to our earlier statements about the reliability of these sources. It seems that the more reliable the source of one's information, the more likely are his chances of winning. Winners, for example, tended to rely less on personal contact than the losers, but much more on public opinion polls, which tend to be more objective. We have considered the relationships between the dependence on sources and knowledge of district opinion, which gave us an indicator of uncertainty, and controlled them by incumbency and election outcome. The effect of risk can be observed by cross-tabulating the candidates' depen- dence on sources with the expected loss variables, which measure the risk when people are believed to be interested in the issues, when-they are believed to be aware of the candidates' stands, and when they believe any of the three policy areas involve important issues. Table 4.13 shows the gamma values for theSe relationships. (See Table 4.13.) The significance of the civil rights area shows up again. The negative gammas are interpreted to mean that candidates who perceived low expected loss were more likely to depend on newspapers and public opinion polls. Only dependence on personal contacts showed a strong positive EXPECTED LOSS BY DEPENDENCE ON INFORMATION SOURCES Dependence on 160 Table 4.13 Newspapers By: Opposed‘Candidates Expected Loss From Issues .19 (770) Expected Loss From Stands .11 (753) Expected Loss From Foreign Affairs .03 (760) Expected Loss From Social Welfare -.16 (760) Expected Loss From Civil Rights -.47 (760) Dependence on Public Opinion Polls By: Expected Loss From Issues .06 (777) Expected Loss From Stands -.08 (760) Expected Loss From Foreign Affairs -.32 (771) Expected Loss From Social welfare —.04*(77l) Expected Loss -.45 (771) From Civil Rights Dependence on People in the Party Organization By: Expected Loss From Issues .30 (741) Expected Loss From Stands .21 (731) Expected Loss From Foreign Affairs .11 (735) Expected Loss From Social Welfare .02*(735) Expected Loss From Civil Rights .04*(735) Dependence on Personal Contacts By; Expected Loss From Issues .07 (766) Expected Loss From Stands -.20 (745) Expected Loss From Foreign Affairs -.1l*(752) Expected Loss From Social welfare -.l7*(752) Expected Loss From Civil Rights .43 (752) *The Chi-square values for these tables are not significant at the .05 level of significance. 161 gamma value. Generally,_though,_the results indicate a very weak association between dependence on sources and risk. The next step is to control for incumbency and election outcome, as shOwn.in Table 4.14. (See Table 4.14.) For incumbency, the most changes were produced in the area of expected loss from civil rights issues. A negative gamma is interpreted to mean that when expected loss was low, candidates were more likely to depend on the sources indicated. Incumbents were more likely to follow this pattern than nonincumbents for dependence on newspapers, public opinion polls, and people in their party organiza- tion when expected loss from civil rights was the indepen- dent variable. Only in the use of personal contacts did nonincumbents show a positive association of dependence with high expected loss. This trend appears also in the following cases: dependence on newspapers by expected loss from stands and domestic issues; dependence on public Opinion polls and expected loss from foreign affairs and domestic issues. In general, incumbents shOw a greater association between low risk and high dependence on sources and nonincumbents are characterized by higher risk and greater dependence on sources. Since the causal sequence of these factors is difficult to determine from the data, no definite conclu- sions can be drawn, but the results of these associations at least indicate that the incumbents use what they perceive are reliable sources from experience and this 162 Ammmc am.ulaeav as.) Ammeva-.- reams oo.e- magmas He>ao some amen empowers Ammmcrao. Amaevaeo.u Ammacrea. Amemc mm.) mummams Hmeoom some mmog cmnomoxm Ammmcraa.uxmaec em.) Ammavram.n Adams as.) anemone dmemuom sons mmoq venomoxm “came ma.uxo~ev mo. lemme ca.) roams mo.) mccmum scum mmoq venomoxm Lemme Ho. Ammevama. Areal mo.) Ammmcama. mmsmmH sous mmoq empowers "hm maaom coacwmo Deanne so mosmpscmma Ammmc Na.uxmoev me.) Amway ma.) Ammmv mm.) magmas Ha>eo some mmoq cmuomoxm Ammmv «a. Emcee he.) Immevaeo. Ammmv we.) dammed: Hmeoom some mmoq empowers Ammmcaeo.unmoav we. Amuse ea. Ammmv ca. anemone cmewuom some mmoq empowers Amemc om. Aoaavamm.u Ammmc mm. lemme em.) museum sous mmoq cmuomoxm lemme ea.uxmaac me. leave ea. Adamo om. mmsmmH sous mmoq cmuomoxm mummoq mumssflz wusmnesosflooz mpcmnfidosH chouso sofluomam MOCOQEOOCH "mm mummomm3mz so moscpsumco MSOUBDO ZOHBUMAM DZfl MuzmmSDUZH mom QMAAOMBZOU mmOA Omaommxm wm mmumDOm ZOHBflSmOmZH ZO mUZmDmemQ vH.¢ OHQMB 163 .HT>OH mo. may no assuamasmfim DOC mum mosacb sundowlwso czar Ammmvame. lemmcamm. Isaac em. nemmvamo. magmas Ha>eo some mmoq cmuomoxm Ammmcrem.nlemmvrmo.u Amaevraa.a Aammcraa.u mummams Hmeoom some mmoq empowers Ammmvaem.uxemmvaaa.u Amaavro~.n Aammvaa~.n nuanced amamuom some mmoq cmuumoxm Amamcrma.nimoavrmo.u lemme no.) Lemme we.) museum sous mmoq cmuomoxm Ammmc ee.uxmoav am. Amaevemo. Amemv no. mesmmH some mmod cmuomdxm "Nm mpomucou Hms0mucm so cosmpccmoo lemmvaeo. “Hoecrmo.u leaavrmo. lemmc.~m.n magmas Ha>ao scum mmoq cmuomoxm xammvaao.uAHOecr~H. AHHevrmo. lemmcaec.u mummams Hmaoom some mmoq cmuomoxm Lemmcrao.nxaoav «a. leave om. lemme so.) nuanced cmamuom some mmoq cmuumoxm lemme mm. “meet me. lemme em. Amemcrmo.u mccmum scum mmoq cmuumoxm Ammmv mm. Amovc em. Ahead Ha. lemmvrmo. mesmmH sous mmoq cmuomoxm "Mm sowomeCcmno human can Cw mamomm Ap.unoov vH.v magma so mosmpcmmmo 164 helps them confirm their perceptions of low expected risk when their campaign.began. Nonincumbents, who would be likely to start their campaigns with a higher degree of risk, would be more likely to want to change their situation by depending more on their favorite sources. Thus, the effect of incumbency on increasing the gammas in a negative direction and decreasing the negative gammas (or making them indicate a more positive relationship) may be accounted for by the starting perceptions of risk: incumbents, perceiving low risk, seeking information, and nonincumbents, perceiving high risk, also seeking information. The explanation has some basis since, for example, the gamma for expected loss from civil rights and incumbency is —.52,_indicating low risk is perceived by incumbents and high risk for non- incumbents. The gamma for election outcome and expected loss from civil rights is -.32, so we would not expect such a pro- nounced change in the partials, and this appears to be the case as shown in Table 4.14. The most interesting results show that when winners perceived low expected loss from issues, they were less likely to depend on these sources than those who perceived high expected loss. Also, losers who perceived low expected loss from issues were more likely to depend on personal contacts. On the whole, there are no predominant trends from the data based on election outcome. However, the comparison between incumbency and election outcome helps to underline 165 the value of a theoretical framework. We are examining the decision making in a campaign from the viewpoint of the candidates and we are able to make some generalized state- ments about the class of individuals we have identified as incumbents and nonincumbents.‘ Election outcome is instead determined by the voters and is based on their perceptions. we would therefore expect to be able to explain the data for incumbency rather than for election outcome, since the theory tells us what we should look for in the data. Election outcome serves as a comparison to indicate to us what types of candidates people are electing to office. If there are indeed differences between the two sets of data, with the effects of incumbency easier to explain than those of election outcome, it is tempting to speculate that since winners become incumbents, there might be something in the nature of the legislative process that neatly orders the behavior of these candidates so that their actions become more predictable. If this is so, then the discovery of a theory which can explain their decision making processes is very much a desirable objective. 4. Policy Positions, Uncertainty, and Risk One important aspect of responsiveness is that the candidates believe that people care about certain issues and that their constituents' knowledge of the candidates' stands will affect the outcome of the election. Before we begin to test the uncertainty hypotheses, we will try to 166 determine the extent to which the candidates believed that the issues and their stands affected their chances of winning. The Representation Study included one question that asked the nonincumbents to assess their chances of election. Table 4.15 contains the exact wording of the question and the possible responses. Table 4.15 . DESCRIPTION OF VARIABLE 0206 ESTIMATED CHANCE OF VICTORY Variable 0206: Estimated Chance of Victory (Nonincumbents only) Q: When you decided to run, what did you think your chances were of winning? Freguency: weighted 63 1. Thought had excellent, very good chance of winning 119 2. Thought had good chance 148 3. Thought had some chance 140 4. Thought had little chance 130 5. Thought had no chance at all 764 9. Don't know, Inapp., NA In addition, we were able to determine the extent to which all opposed candidates felt people in their district knew them as a person. (See Table 4.16.) As a result, we can make a Comparison, at least among nonincumbents, of the association of their eStimated chances of election and 1) how well they believe people know their stands and 2) how well they believe the people know the candidate as a person. Table 4.17 shOws the results of this comparison. 167 Table 4.16 DESCRIPTION OF VARIABLE 0170 PERCEIVED DISTRICT KNOWLEDGE OF CANDIDATE AS A PERSON Variable 0170: Perceived District Knowledge of Candidate as a Person Q: How much do you think peOple in your district know about you as a person? Frequency: All Candidates Weighted 311 1. Know a great deal; very widely known 485 2. Know a good deal; fairly widely known 192 3. Know some things; known somewhat 218 4. Don't know very much; not very well known 8 5. Don't know anything; not known at all 150 9. Don't know or NA Table‘4.l7 PERCEIVED CONSTITUENT KNOWLEDGE OF CANDIDATE'S STANDS AND KNOWLEDGE OF CANDIDATE AS’A PERSON BY ESTIMATED CHANCE OF WINNING ELECTION Opposed Nonincumbent Variable 0169 People Know Candidate's Stands By Variable 0206 Estimated Chance of Winning Election: .45 (530) Variable 0170 People Know Candidate as a Person By Variable 0206 Estimated Chance of Winning Election: .22 (559) 168 The gammas of .45 and .22 indicate that the nonincum- bents associated their.chances of winning more with con- stituent knowledge of the Candidates' stands than with knowledge of the personal qualities of the candidates. This helps to indicate the relative importance that the issues and the policy positions of the candidates are perceived to have in the campaign, at least among nonincum- bents. Although comparable data for incumbents were not available, the nonincumbents are likely to be relatively less known for both their stands and their personal characteristics than the incumbent. The data indiCate that, at least at first, candidates believe that their stands are the more important influence on their election chances. With some knowledge that candidates believe that their issue positions are related to their chances Of electoral success, we shall proceed to test the three hypotheses that deal with the effeCt of uncertainty on candidates' policy positions. These hypotheses were operationalized from the propositions originally introduced in chapter two and derived from the Bayesian decision making model. The 1958 Representation Study is the source of the data used to test these hypotheses. -Also, only the data for opposed candi- dates will continue to be reported, since the model is concerned only with the majority of candidates who have the postulated goal of winning the general election by defeating a challenger. Since there is likely to be very 169 little uncertainty about the winner and practically no risk involved in the election of an unopposed candidate, the inclusion of this type of candidate in the analysis would be inappropriate. Each of these hypotheses are concerned with the policy positions adopted by the candidates in the areas of foreign affairs, social welfare (also called domestic issues), and civil rights. In most cases, the hypotheses predict when the association between the candidates' perceptions of district opinion and their public policy positions will be higher or lower. The important factors of uncertainty and risk are the conditional control variables which will deter— mine the type of change in this basic relationship, with the candidates' policy position as the dependent variable, and their perception of district opinion as the major indepen- dent variable. Table 4.18 presents the gamma values of this relationship for each of the three issue areas. Tables 4.19, 4.20, and 4.21 show the distribution of the cases in the contingency tables. (See Tables 4.18, 4.19, 4.20, and 4.21.) Hypothesis 1 is stated as follows and the results from its test appear in Table 4.22. (See Table 4.22.) Hypothesis 1: Ceteris pgribus, candidates who believe they know how people in their district feel about the issues are more likely to adopt a policy position that is close to what they perceive to be the majority opinion in their district. 170 Table 4.18 CANDIDATES' POLICY POSITION BY PERCEPTION OF DISTRICT OPINION Op osed Candidates Variable 042 Foreign Affairs Policy Position By Variable 195 Perception of District Opinion on Foreign Affairs: .43 (914) Variable 054 Social Welfare Policy Position By Variable 196 Perception of District Opinion on Social Welfare: .42 (959) Variable 065 Civil Rights Policy Position By Variable 197 Perception of District Opinion on Civil Rights: .51 (876) Table‘4.l9 FOREIGN AFFAIRS POLICY POSITION BY PERCEPTION OF DISTRICT OPINION Variable 195: PerceptiOn of District Opinion Variable 042: Most Evenly Most in Policy PosiEion Opposed Divided Favor Isolationist 92 12 27 29.0% 5.4% 7.2% Neo-Isolationist 89 54 38 28.1% 24.1% 10.2% Pro-Con 75 77 123 23.7% 34.4% 33.0% Neo-activist 32 58 103 10.1% 25.9% 27.6% Activist 29 23 82 9.1% 10.3% 22.0% TOTAL 317 224 373 100% 100% 100% 171 'Table 4.20 SOCIAL WELFARE POLICY POSITION BY PERCEPTION OF DISTRICT OPINION Variable 196: Perception of District Opinion Variable 054: Most Evenly Mostiin Policy Position Oppgsed Divided Favor Conservative 125 42 48 37.1% 15.9% 13.4% 105 48 63 31.2% 18.2% 17.6% 14 18 16 4.2% 6.8% 4.5% 7 37 12 .2.1% 14.0% 3.4% Liberal 86 119 219 25.5% 45.1% 61.2% TOTAL 337 264 358 100% 100% 100% Table 4.21 CIVIL RIGHTS POLIQY POSITION BY PERCEPTION OF DISTRICT OPINION Variable 197: PerceptiOn of DiStrict Opinion Variable 065: Most Evenly Most'in Policy Position Opposed Divided Favor Conservative 124 23 81 66.3% 22.5% 13.8% 14 7 72 7.5% 6.9% 12.3% 7 19 112 3.7% 18.6% 19.1% Liberal 42 53 322 22.5% 52.0% 54.9% TOTAL 187 102 587 100% 100% 100% 172 Table 4.22 CANDIDATES' POLICY POSITION BY PERCEPTION OF DISTRICT OPINION CONTROLLED FOR KNOWLEDGE OF DISTRICT OPINION Variable 0195 Knowledge of District Opinion Knows Most Knows Seldom of the Time Some Time Knows Foreign Affairs (.43)* .52 (608) -.13 (71) .18 (69) Social Welfare (.42)* .59 (610) .61 (79) -.37 (88) Civil Rights (.51)* .65 (587) .56 (69) -l.00 (64) *These numbers are the gamma values of the basic relation— ship between the candidates' policy position and their perception of district opinion as shown in Table 4.18. They are presented here for purposes of comparison. The hypothesis is tested by controlling the basic relationship shown in Table 4.18 by Variable 0195, Knowledge of District Opinion. The hypothesis is supported by the fact that for each issue area the gammas increased when the Opposed candidates believed they knew how people in their district felt about the issues. Also, those who believed they seldom knew district opinion were much less likely to follow their perception of district opinion. These results provide evidence for the validity of the decision matrix, in which the knowledgeable (or confident) candidates would be expected to follow district opinion and the “ignorant" candidates would assume a more randomized strategy, and that is what has occurred here. The results for the case in which the candidate believes he knows district opinion some of the time warrants some discussion. For foreign affairs, candidates were not 173 likely to follow district opinion, but were very likely to do so for social welfare and civil rights. First, the relatively small N for this category (and the "Seldom Knows" category as well) prompts some questions of reliability, but the testing of the other hypotheses may indicate whether the basic trends are correct. Second, it appears that the perceived relative importance of the issues is a potential source of influence in determining the candidates' policy position. Foreign affairs has traditionally been the issue area in which politicians have been given more discretion by their constituents in choosing a policy position. This may explain the gamma of -.13 compared to the larger values of .61 and .56 for social welfare and civil rights, respec- tively. The effect of the perceived importance of these issues on the candidates' policy positions will be examined further in regard to the risk hypotheses. We now turn to the testing of Hypothesis 2. Hypothesis 2: Ceteris paribus, candidates who believe people in their district are interested in the issues are more likely to adopt a policy position close to what they perceive to be the majority position in their district. By combining the categories of "most people interested" and "some people interested" to keep the number of cases in the cells at a high enough level for analysis, the results in Table 4.23 were obtained. (See Table 4.23.) The hypo- thesis is supported by the fact that the gamma values are higher than the original relationship shown in Table 4.18 174 Table’4.23 CANDIDATES' POLICY POSITION BY PERCEPTION OF DISTRICT OPINION CONTROLLED BY_PERCEPTION OF PEOPLE'S INTEREST IN THE ISSUES Variable 0094 People Interested in the Issues People People Interested** Not Interested Foreign Affairs (.43)* .49 (411) .41 (420) Social Welfare (.42)* .54 (394) .27 (478) Civil Rights (.51)* .66 (410)*** .40 (397) * These values are the gammas for the original relationship between candidates' policy position and their perception of district opinion, originally Shown in Table 4.18. ** This category is based on the combination of the cate- gories: "Most People Interested" and "Some People Interested." ***When equiweighted, this gamma value becomes .60. The change for the other gammas is only slight. for the case in which people are perceived to be interested in the issues. When the candidates perceive people are not intereSted in the issues, the values decline, as expected. Again, the weakest area is foreign affairs, which shows the least Significant change in gamma from the original value of .43. Before passing judgment on the application of the model to the area of foreign affairs, we should examine the results of Hypothesis 3. As stated earlier, we would expected that since the phrase "interested in the issues" did not specifically refer to the candidate, we would expected that when we control for the degree to which people are perceived by the candidate to be interested in his stands, that we would be able to capture more of the 175 concept of risk into the analysis. Hypothesis 3 is stated as follows: Hypothesis 3:‘ Ceteris paribus, candidates who Believe people in their district are aware of their stands are more likely to adopt a policy position close to what they perceive to be the majority Opinion in their district. ‘Table‘4.24 POLICY POSITION BY PERCEPTION OF DISTRICT OPINION CONTROLLED BY PERCEPTION OF P'EOELET'S KNOWLEDGE OF CANDIDATE? STANDS variable 0169 People in District Know Candidates' Stands People People ‘Know Stands ‘Don‘t Know Stands Foreign Affairs (.43)* .47 (587) .15 (222) Social Welfare (.42)* .49 (607) -.04 (245) Civil Rights (.51)* .69 (563) .13 (224) *These values are the gammas for the original relationship between candidates' policy position and their perception of district opinion, originally shown in Table 4.18. Table 4.24 shows the results when the basic relation- ship is controlled by Variable 0169, People in the District Know the Candidates' Stands. The categories of "Know a Great Deal" and “Know Some Things” are combined into one category, "People Know Stands." The gammas reflect excep- tionally well the expected relationship when the candidate perceives that people in his district know something about his issue positions. The results are especially striking because the gammas for the "People Don't Know Stands“ category Show almost no 176 relationship between the candidates' perception of district opinion and their own policy positions. Only when they perceive that the people are aware of their stands do they adopt a policy position consistent with their perception of district opinion. This appears to be especially true regarding Civil rights issues, which were of particular concern during the 1958 congressional elections. These results signify that when candidates perceive that there is some potential loss of votes as a result of some knowledge acquired by their constituents, they will adopt a policy position that coincides more closely with their perception of district opinion. Hypothesis 4 opera- tionalizes the concept of expected loss, which considers not only the constituents' perceived knowledge, but also the candidate's confidence in their knowledge of district opinion. Hypothesis 4: Ceteris paribus, candidates who perceive themselves to be in a High 'Ex ected Loss Situation are more IiEely to adopt a policy position close to what they perceive to be the majority position in their district than those who perceive themselves to be in a Low Expected Loss situation. Table 4.25 shows the original relationship between the candidates' perceptions of district opinion and their public policy position on each of the three issue areas, controlled by each of the five variables that were created to measure expected loss. (See Table 4.25.) For each of the expected loss variables, the categories of High and CANDIDATES' 177 Table'4.25 OPINION CONTROLLED BY EXPECTED LOSS A. Expccted Loss From 'Expected“Loss High** 'Expected‘LOSS Issues: Foreign Affairs (.43)* Social Welfare (.42)* Civil Rights (.51)*a B. Expected Loss From Stands: Foreign Affairs Social Welfare Civil Rightsb C. Expected Loss From .50 .43 .64 .48 .52 .70 (373) (365) (363) (486) (506) (462) 'EaChTPolicyfArea: Foreign Affairs Social Welfare Civil RightsC .53 .50 (222) (326) .54 (201) POLICY POSITION BY PERCEPTION OF DISTRICT LOW EXpected'Loss .56 .47 .56 .33 .30 .45 .51 .42 .69 (262) (284) (249) (130) (126) (123) (400) (338) (402) * These values are the in Table 4.18. gamma values for the original rela- tionship between the candidates' policy position and their perception of district Opinion, originally shown **The High Expected Loss category combines the responses to the High Expected Loss and Moderate Expected Loss categories. a,b,c: a. .57, .55 b. .61, .51 C. .40, .77 When the gammas for these categories are equi- weighted, the following gammas result: 178 Moderate Expected Loss were combined because of the small number of cases in the High Expected Loss category. The results show that there was not much change in the original relationship when controlling for Expected Loss From Issues. For each issue area, the relatiOnship is slightly enhanced, indicating that to some extent the control is a conditional causal factor. When the control variable Expected Loss From Stands is introduced, the relationship predicted by the hypothesis occurs.) The values for the.High Expected Loss category are all higher than the original gammas and the Low Expected Loss gammas are all lower (except when the equiweighted gammas for civil rights are considered). The last part of the table shows the controls of Expected Loss From Foreign Affairs, Social Welfare, and Civil Rights. These results do not appear to present con- Clusive evidence in support of Hypothesis 4. It would certainly have been preferable to have a variable that could directly measure expected loss (risk), but it is noteworthy that among the variables in the survey that are available, the one that does appear to best measure the extent to which candidates perceive themselves to be in a potentially risky situation is the one that provides the best reSultS. This speaks well for the utility of theory development, especially Since formal models can direct us to appropriate kinds of data that are necessary to test our hypotheses and tell us what the reSults are 179 likely to mean. The problem we face here is the limited availability of the data that measure precisely that which we would like to measure. NeVertheless, as long as we can recognize the limitations and try to work within them, the preliminary results of this study can have important impli- cations in terms of the direction of future research. The empirical research itself is important, because different measures of the same kinds of variables can extend the original interpretation of the model and provide addi- tional richness to the research and the implications of its findings. For example, we have interpreted expected loss to be an indicator of risk, but risk, in a more general sense, can represent the possible loss of an investment rather than just a possible loss Of votes. Hypothesis 5 enables us to broaden the interpretation of risk and provide some justifi- cation to the pursuit of the various effects that risk can have in other kinds of political decisions, such as running for higher office, running for reelection, or retiring from office. Hypothesis 5: Ceteris paribus, in noncompetitive districts, the greater the risk, the more likely candidates who are opposed in the eleCtion will adopt a public policy position close to what they perceive to be the majority position in their district. We classified candidates who ran for office repre— senting the majority party in a noncompetitive district as being in a High Risk situation, since they had the most to lose (althOugh they were most likely to win). Minority 180 party candidates were in a Low Risk situation because they were likely to have made less of an investment in running for an office they did not have much chance of winning. Table 4.26 shows the gammas for these situations. Table'4.26 CANDIDATES' POLICY POSITION BY PERCEPTION OF DISTRICT OPINION CONTROLLED BY RISK:V NON- ‘COMPETITIVE‘DISTRICTS Risk ‘High Risk Low Risk Foreign Affairs (.43)* .48 (278) .23 (282) Social Welfare (.42)* .68 (283) .09 (334) Civil Rights (.51)* .81 (258) .41 (293) *These values are the gammas for the original relationship between candidates' policy position and perception of district opinion, as shown in Table 4.18. For all three issue areas, the results Show an increase in the basic relationship between the candidates' policy positions and their perception of district opinion for majority party candidates in the High Risk category. In the Low Risk category, the gammas decline sharply. This means that majority party candidates are very likely to follow district opinion while minority candidates' poliCy positions Show very little relationship (except for civil rights) to their perception of majority opinion in their district.27 Thus far, we have controlled the basic relationship by expected loss of votes (Hypothesis 4) and by expected loss 181 of investment (Hypothesis 5). The sixth hypothesis controls for both types of risks and these results are Shown in Tables 4.27 and 4.28. (See Tables 4.27 and 4.28.) Hypothesis 6: Ceteris pdribus, when controlling for 'expected loss of votes, the higher the risk (expected loss of investment), the more likely candidates who are Opposed in the election will adopt a public policy position close to what they perceive to be the majority position in their district. Tables 4.27 and 4.28 show the relationship between the candidates' policy position on an issue and their perception of district opinion, controlled by risk (expected loss of investment) and expected loss of votes (which was an alter- native definition of risk). To measure expected loss of votes, we are using "Expected Loss From Stands", since this was found to be a good measure of risk. The first table contains cases in the noncompetitive districts and the second one contains those in competitive districts. Unfortunately, the number of cases in the Low Expected Loss category in Table 4.27 has dropped to the point where it is difficult to make any reliable judgments about the results in that part of the table. Observing the values in the High Expected Loss section, we see the hypothesis is confirmed for both social welfare and civil rights. In both of these cases, the basic relationship increased for the High Risk category and decreased for the Low Risk category. Compared to the results in Table 4.26, which tested Hypothesis 5, the gamma for social welfare increased from .68 to .72 (High Risk) and from .81 to .85 for civil 182 .COASHQO uowuumwp mo coaummoucm Remap pom GOHpflmom MOHHOQ .mcumpwpsco on» smoSucn QHQMCOADMHTH pwaaouusooss Hecamfluo on» How mcEEmm may cum mo5am> cassette .muumo aufinoswfi ecu mo mmosu mopsaosfl anomwpco Roam 30A map can .musmm mnemonme may Eoum mcumpfipsco mo upmamsoo muomwumo xMHm swam one «R .mcDEMH ms» co mpsmum .mcumpwpsco Ono msfi3ocx uoauumflp Hams» cw mamoom Eoum mmOH pcuoooxc mcusmmcfi cancfiuc> mags R Ammv mv. Amav oo.H Ammv mv. Amway mm. «RRAHm.V nunmfim Ha>flo Achy hm. Amav oo.H| onav mo. Aaomv mu. taramv.v canvass HcHoom Amhv ma. Amav we. Aboav mm. Ammav ow. rttfimq.v msficmmd smflmuom .2.me 30a .3032 SE «a? .4me seem «mmoq pcuoomxm 304 «mmoq pwuommxm swam mBUHMBmHQ m>HBHBmszUZOZ «mmOA Omaummxm QZd MmHm Nm DMAAOMBZOU ZOHZHmO BUHmbHQ ho ZOHBmmummm wm ZOHBHmom NUHAOm .mMBflnHQZflo Amway hw. Awmav Hm. Ammav om. Amway an. mummamz HMfloom AHmHV mm. Aomav Hm. imomo ms. Ammo mm. muflmuma cmflmuom mmoq mmoq mmoq mmoq cmuommxm cmuommxm cmuommxm Umuommxm 30A swam 3oq swam madmnEdocwsoz mucmnasocH 200 Ammv mv. Abomv mo. Ahmv hm. Ammmv mu. munmwm HH>HU Ambv om. Ahmmv om. Amvv mm.| Ammmv mm. mummamS HMHoom Amhv ha. Amamv mm. Aamv hm. Ammmv mm. mufimmmd cmwmuom mmoq mmon moon mmoq Umuommxm pmuommxm pmuommxm cwuommxm 30A can: 304 sons mucwnadocficoz mucmnfidocH mwumwwmcmo cmmommo (MNocwnEsocH hm Umaaouucoo mwcmum scum mmoq pmuommxmem GOHGHQO DUAHumflo mo nowumvoummxwm COADHmom moflaom mzobmbo oneomqm ozm wozmmzsozH um mmuqommzoo .mmmmm mmmmH 20mm aza mnzaem 20mm mmoq omeommxm mm onszo aonemHo mo onemmommm wm onaHmom onqom m.m OHQMB 201 AHhHV Hm. AHOHV mm. AHMNV on. Aooav on. munmwm HH>HU Amway ma. Ammav Ho. Amway mm. Roomy up. mummawz Hafioom Amado ms. Ammo mm. lemme Hm. AFNHV mm. unammma cmfimuom mmoa mmoq 11mmmm1I. mmoqux omuowmxm omuowmxm pwuommxm cwuommxm 304 Beam sou roam mummoa muwccaz mwymmwccmo cmmommo .umaoouso cowuomam mmpwaaouucou .mu£mwm Hfi>fiu.msm .mnmmamz Hmfloom .muH8mm¢ cmflmuom Scum mmoq omuowmxm mm.coflcwmo uoflhumwo mo coaummonmm Mm coauflmomnwoaaom Ammo mv. Amway me. Ammv oo.H Aommv mm. munmwm HH>HU Aamv oe. Ammav mm. Amvv mo. Avamv no. . mummamz Hmwoom Ammo mm. Amway Hm. Amwv me. Ahamv mv. muwmmum cowmuom mmoq .mmoq mwoq . mmoa, Umuommxm wmuommxm cmuommxm cmuommxm 30a swam son swam mummoq mnmccHB mmumwflocmo.cmmommo "meoopso commomam mm Umaaoupcoo m©GMDm scum mmoqwomuowmxm mm soacw O HUflHDMfia.Mo coflAmmoumm mm,s0fiuwmom Nmflaom Ac.ucoov m.m OHQMB 202 are likely to elect representatives who do try to follow district opinion, especially in regard to an issue that is likely to have a high degree of salience among the citizens. Foreign affairs during this era was not likely as salient an issue as civil rights, and the data in Table 5.3 seem to mirror this assessment. Table 5.4 shows the results from controlling for the other risk indicator, the variable measuring the expected loss of investment, and comparing incumbency to election outcome. The High Risk category is composed of majority party candidates in a safe district, and the Low Risk category contains the minority party candidates in safe districts. The candidates from the competitive districts are included in an additional column, but are not con- trolled for risk. (see Table 5.4.) The analysis is limited by the increased number of cells with too few cases, but there does not appear to be much change between the gammas for High Risk incumbents and High Risk winners, which have sufficient cases to be significant. In the competitive districts, the gammas for social welfare decreased slightly and increased for civil rights from .40 to .58. The nonincumbents from competitive districts had low associations between candidates' policy positions and their perceptions of district opinion, and the gammas for the losers show that the citizens rejected candidates from competitive districts who were not too likely to follow district opinion. ASOHC mo. Amway mm..,. “Haas Hm. Amado ow. magmas Hfl>flu Aomav «N. Amway no. Aoaav mm.l Amway hm. mummamz Hmfioom AmHHV mm. Amway ¢m. AMHHV hm. Acnav om. muwmwmd :mflmuom mummoq mumccfl3 mpcmnasocwcoz muswnfisocH umzoaaom mm mum xmflm an wmaaouucoo no: .mmpmvflpcmo unflnumflo m>fluwummaoo How mmEEmm one "muoz immmv Hm. Amflv u- loss mm.u _lov~v mm. musmflm Hfl>flo Assay Ho.u Ammo co. isms so. 1mmmo mm. mumuams HmHoom AHNNV ma. Ammo so. last me. Ammmv ma. whammmm cmhmuom xmflm 30a xmflm nmwm xmflm sou xmflm smflm mummoq muwccfl3 mEoouso coauomam ‘Nm cmaaouucoo xmwm mm coflcflmo uowuumfla mo coaumwwumm mm coHuflmom mowaom .203 Amsmo as. Ammo om. Amav In- Amowv mm. musmnm Hfl>flo Amomv mo. Ammo ms. Ammo oo.H loans on. mummamz Hmfloom wismmv an. Ammo ow. Ammo oo.H imamv Hm. muammma cmamuom xmflm 30g xmnm‘mmflm xmflm 309 swam swam mpcmnesocflcoz muchEDOGH NucwnfidocH mm omaaouucou xmam mm GOHQHQO uofiuumfla mo coflummonmm NW coHuHmom Nmflaom mnemoHozmo ammommo "mzooaoo oneomqm oza wozmmzoozH um omqqomezoo AmmpHmemHo m>Hewwmmzoozozv smHm wm onszo aonemHo mo onsmmommm mm oneHmom mquom v.m wanna 204 As far as responsiveness is concerned, we did not determine whether the winners actually represented their constituency bettethh n the set of incumbents before them, but we did determine whether or not they were likely to adopt policy positions close to what they perceived to be the majority position in their districts. Incumbents and winners were more likely to follow district opinion than the nonincumbents and losers, indicating that they may have had rational reasons for doing so. Indeed, even incumbents in safe districts were likely to follow district opinion. .The model explains that this is because of.the high expected loss of investment that would occur if they did not reflect district opinion. When there are rational reasons not to follow district opinion, such as in low expected loss (of votes or of investment) situations, candidates expectedly do not follow district opinion. Instead, they may feel free to either adopt their own personal attitudes (if they differ from the majority opinion), or second, they may be obligated to follow the dictates of their party organization and its activist supporters, who may not represent the mainstream of opinion in the district. Third, as the model suggests, they may be uncertain about district opinion and may not be very confident about their own estimates of district opinion. Therefore, they would rely much less on their perceptions and select some other possible position. Although it cannot be confirmed.from the data we have, there may be an important underlying explanation for some of 205 the negative values for the low loss and low risk cate- gories when controlling for incumbency and election outcome. For the hypotheses tested in chapter four, the prediction for these categories was no relation between policy position and perception of district opinion. The results generally coincided with our prediction and rarely were there any negative values at all. However, when we control for incum- bency and election outcome, a number of negative associa- tions do arise. As stated previously, the theory does not predict any relationship for incumbency or election outcome, but the results may be worth investigating. These negative values (in Tables 5.2, 5.3, and 5.4) may represent cases in which candidates consciously chose to adopt policy positions other than those perceived to be majority opinion in order to satisfy other groups or individuals in the district that have a special influence upon the candidate. Another reason may be that the negative correlations represent the positions of candidates who were unable to moderate their positions adopted during a primary campaign. This point is related to the problem that a candidate may have in his campaign: to maximize votes or maximize resources from supporters who have preferences that conflict with the majority of the voters. These factors may indicate the nature or significance of the negative gamma values and thus be an important area for future research, since these questions go beyond the scope of this study. 3% 206 3. Marginality and Responsiveness One of the more interesting implications of the results in chapter four concerns the effect of the perceived compe- titiveness of the district on the relationship between can- didates' perceptions of district opinion and their policy positions. The study of the effect of electoral margins on political behavior has a long history, going back at least as far as MacRae's study of the Massachusetts House of Representatives,2 but, as first mentioned in chapter one, the results of some of these studies have often been contra- dictory and confusing.3 The Bayesian decision model, how- ever, presents a perspective for examining the decisions of candidates from marginal (competitive) districts and safe (noncompetitive) districts. In this section we intend to show how the results of our study can help ameliorate the seemingly contradictory findings of previous efforts. A recent reappraisal of the literature on the effect of marginality on constituency influence was published by Fiorina,4 who applied his decision-theoretic model to the study of this question. 'In his article, Fiorina discussed the following generalization that is prominent in the constituency influence literature: The marginality hypothesis: the less confident the representative is about his chances to be re-elected, the more he votes in accordance with the interests of his constituency.5 Fiorina concluded that many of the researchers who found the marginality hypothesis to be true were "not 207 justified by the data they analyzed."6 This was primarily because the questions asked by the researchers were not answered by the data. In fact, Fiorina concluded, "only Miller's research7 bears directly on the question of constituency influence on safe versus marginal represen- tatives."8 As a result, the contradiction between Miller's findings, which work against the marginality hypothesis, and those of previous researchers, is reconciled. However, Miller's findings, which were based on the 1958 Representation Study, appear to be counterintuitive to the marginality hypothesis, which predicts that legislators from safe districts are less likely to follow district opinion because their seats are safe and they "can concern themselves with a broader state interest."9 It also pre- sumes that legislators from competitive districts would feel more pressure to pay heed to constituents' interests, since their seats would be vulnerable to challengers. Instead, Miller found that congressmen from safe districts presented a more balanced voting record between the influences of constituency and party interests. They were also found to repreSent their majority party constituents better than themarginal district congressmen (for social welfare and civil rights), which tends to disprove the marginality hypothesis.10 Although the marginality hypothesis could not explain these results, they nevertheleSs fit well within Fiorina's model,11 for his hypotheses describe the voting decisions 208 of legislators according to the size and number of groups in their district to which the congressmen are obliged to respond. Although the marginality hypothesis may be “intuitively reasonable," its explanatory power is quite limited when compared to data that could be used to test it. Actually, the marginality hypothesis is not all wrong, because it does rely upon the condition that legislators want to represent their constituents most accurately when there is the greatest risk. The problem therefore lies in the conceptualization of what is meant by risk. Proponents of the marginality hypothesis have presumed that candidates in marginal districts perceive the greatest risk, and therefore are more likely to follow district opinion than legislators from noncompetitive districts. Taken from the viewpoint that each legislator is the basic unit of analysis, this is not necessarily true, since the perceptions of each legislator are likely to be different. Taken a step further, we cannot strictly compare legislators from safe and com- petitive districts and say one group is going to be more responsive than another. Instead, we have to examine their perceptions of risk and uncertainty and divide’them into high and low risk groups and then make comparisons. As we saw in chapter four, there is a difference in responsiveness within noncompetitive districts, and we can only presume that this may very well be the case within competitive districts as well. 209 If we examine the difference between candidates from marginal and safe districts according to the degree to which they are likely to select policy positions which coincide with their perceptions of district opinion, we find the results shown in Table 5.5. Tablé’5.5 POLICY POSITION BY PERCEPTION OF DISTRICT OPINION CONTROLLED BY DISTRICT COMPETITIVENESS: 'OPPOSED CANDIDATES All Opposed 'Candidates 'NoncompetitiVe ‘Competitive Foreign Policy .43 (914) .33 (560) .55 (283) Social Welfare .42 (959) .43 (617) .41 (273) Civil Rights .51 (876) .60 (551) .36 (253) Candidates from noncompetitive districts are more likely to follow district opinion primarily on civil rights issues, while candidates from competitive districts are more likely to follow district opinion on foreign affairs issues. When we controlled for expected loss of investment, as in Table 4.26, we found that the candidates in the non- competitive districts showed a remarkable difference. The high risk candidates increased their support of district Opinion and low risk candidates decreased their support of district opinion. As an imperfect indication of high risk candidates in competitive districts, we can look at the gammas for incumbents from competitive districts, origi- nally shown in Table 5.4 and displayed again in Table 5.6. 210 Table‘5.6 POLICY POSITION BY PERCEPTION OF DISTRICT OPINION CONTROLLED BY RISK FOR NONCOMPETITIVE DISTRICTS AND BY INCUMBENCY FOR COMPETITIVE;DI§TRICTS: OPPOSEDrCANDIDATES NoncompetitiVe 'Competitive Districts ’DIStricts High RiSk Low Risk 'Incumbents Nonincum. Foreign Affairs .48 (278) .23 (282) .60 (170) .37 (113) Social Welfare .68 (283) .09 (334) .77 (163) -.29 (110) Civi1 Rights .81 (258) .41 (293) .40 (142) .31 (111) The comparison between incumbents and nonincumbents in competitive districts suggests that (if we assume for the moment that the incumbents perceive high expected loss of investment and nonincumbents perceive low expected loss of investment), it is the risk they perceive rather than the marginality of the district itself that causes this differ- ence in the gammas between the two groups. Merely to examine the gammas between the competitive and noncompeti- tive districts tell us very little, since the competitive district candidates showed a higher association for foreign affairs, noncompetitive district candidates showed a higher association for civil rights, and the gammas were about the same for social welfare (Table 5.5). However, when we break down the candidates into categories measuring the perception of risk, the striking differences appear. The conclusion we can draw from this is that it is not so much the marginality of the district, but the candidates' perception of how much risk and uncertainty is present. 211 We can examine the effect that the perceived chance of election has upon candidates by controlling the relationship between candidates"perceptions of district Opinion with their policy positions by their estimated chance of election, measured by Variable 206. This question, "When you decided to run, what did you think your chances were of winning?“ was asked only of nonincumbents in the survey, but the results in Table 5.7 are quite instructive. Table'5.7 POLICY POSITION BY PERCEPTION OF DISTRICT OPINION CONTROLLED BY ESTIMATED CHANCE OF WINNING ELECTION: OPPOSED NONINCUMBENTS Policy Position By Perception of District Opinion Controlled By Variable 206, Estimated Chance of Winning Election: Estimated Chance of Winning Good Chance Some Chance Little Chance Foreign Policy (.32)* .54 (144) .28 (133) .05 (156) Social Welfare (.13)* .60 (141) -.32 (126) .09 (212) Civil Rights (.44)* .47 (145) .28 (116) .56 (184) *These gammas are the ones representing the relationship between policy position and perception of district opinion for opposed nonincumbents. We see that when the candidates believed their chances of election were good, they were very likely to match their perceptions of district opinion with their public policy positions. The relationship appears to be linear for foreign policy and curvilinear for Civil rights and to some extent for social welfare. The point is that we cannot make valid generalizations about constituency influence by 212 relying solely on the competitiveness of the district. As we have seen here, at least for the nonincumbents, the perception of a good chance of winning the election increased the gammas as Compared to nonincumbents as a whole. The validity of the marginality hypothesis therefore lies not in its statement of the effect of marginality as an objective measure, but as a subjective one, which cuts across the conceptual boundaries between safe and marginal districts. It is the candidates and the legislators that perceive the marginality and merely observing the margin of the vote is not enough to determine the safeness of a district. We are arguing therefore that the marginality Of the district is not the overriding factor that it may appear to be regarding constituency influence, but that other factors that indicate the degree of risk and uncer- tainty, subjectively perceived by the candidate or legis- lator, should be examined. The significance of the Bayesian model is well illus- trated by this example, because it allows us to perceive the nature of candidate (as well as legislator) decision making within an entirely new perspective. The marginality hypothesis appears intuitively reasonable: legislators who are in more danger of losing their seats are more likely to follow district Opinion. However, it has not been properly measured, because researchers have been concerned with identifying marginal and safe districts by objective 213 criteria and not subjectively. The marginality hypothesis has thus suffered from incomplete conceptual development, especially because it is not accompanied by a conceptual framework. The Bayesian model instead points the way toward the type of data that are.needed to test the model and what the results are likely to mean within a larger context. This interpretation of the marginality hypothesis helps to emphasize the utility of the subjective decision making approach to the study of political behavior. It recognizes that there is a need to reconceptualize our ideas about poli- tical behavior in many areas and to be more concerned with studying the perceptions of political actors and learning more about how they collect and interpret the information they use to formulate their perceptions and beliefs about the states of the world. Politicians may not actually think in terms of a decision matrix and expected losses, but the results from this study seem to indicate there is strong reason to believe that they act‘ag if they have made similar calculations. 4. Incumbency, Election Outcome, and Roll Call Positions To complete the examination of the linkage process involving the perceptions of the candidates and their choice of a public policy position, we turn to the final step in that process, the casting of roll call votes by the successful candidates.' If the candidates who win the election attempt to be truly responsive, then there should be a strong positive association between their perception 214 of district opinion and their roll call behavior. Also, if candidates do not misrepresent themselves in an election campaign, then their public policy positions should be strongly and positively associated with their roll call behavior. The data from the Representation Study included a set of roll call scales created from votes cast by the incum- bents during the 85th Congress.12 Table 5.8 shows the frequencies of the responses to the roll call scales for the areas of foreign affairs, social welfare, and civil rights. (See Table 5.8.) Scale scores for the 86th Congress would have been preferable, since the linkage process involves the winning candidate casting his votes in the new Congress. Neverthe- less, we do not expect that the roll call scales for the 85th Congress would cause any great variations in the results. The data for the gamma associations between the roll call scales and perception of district opinion and between the public policy positions of the candidates who were the incumbents and of those incumbents who won the 13 Figure 5.1 shows the election are shown in Table 5.9. paths of the relationships, and for easier reference, Figures 5.2, 5.3, and 5.4 are presented to illustrate the causal relationships that are expected to hold, with the accompanying gammas. FREQUENCIES FOR ROLL CALL SCALE ITEMS: 215 Table 5.8 FOREIGN AFFAIRS, SOCIAL WELFARE, AND CIVIL RIGHTS Variable 0364: Weighted Frequency 126 134 7 0 25 7 11 35 86 67 161 Code 00. 01. 02. 03. 04. 05. 06. 07. 08. O9. 10. Foreign Policy Roll Call Scale No Positive Responses Conservative One Positive Response Two Positive Responses Three Positive Responses Four Positive Responses Five Positive Responses Six Positive Responses Seven Positive Responses Eight Positive Responses Nine Positive Responses Ten Positive Responses Liberal Variable 0362: Social Welfare Roll Call Scale Weighted Frequency Code 150 0. No Positive Responses Conservative 40 1. One Positive Response' 69 2. Two Positive Responses 30 3. Three Positive Responses 28 4. Four Positive Responses 47 5. Five Positive Responses 58 6. Six Positive Responses 111 7. Seven Positive Responses 122 8. Eight Positive Responses Liberal Variable 0363: Civil Rights Roll Call Scale Weighted Frequency Code 197 0. No Positive Responses Conservative 7 1. One Positive Response 21 2. Two Positive Responses 7 3. Three Positive Responses 32 4. Four Positive Responses 54 5. Five Positive Responses 378 6. Six Positive Responses Liberal 216 AHmwv on. Ammvv mm. Amamv mm. GOHummonom mm sofluwmom mOfiHom lqmev om. Ammmv as. muamflm Hfl>fio laces mm. Amame mm. manuamz anacom lasso om. immmv we. whammma amfimuom cofiuflmomxwmfiaom GORPMOOHOA mm Hamo Haom mm Hamo Haom mOPwumucmmmummm mmmumcoo numm Omuomamm Ahmmv Hm. Ammvv mm. Avvov hm. coawmooumm.mm GOADHmom Nowaom Amway mm. lmsmo HS. muamflm HH>NO inset mm. loose om. mHMMHms Hmfloom Ammee mm. Amovv om. muflmmm< cmflmuom soflpflmomDNOAHom (coflummoumm mm Hamo Haom mm Hamo Haom «mnoccwz mmoumcou zumm ZOHBHmom NUHAOQ UHQmDm Mm DZ< 7)) ZOHZHAO BUHMBmHO m0 ZOHBmmommm Mm mMQ¢Um Afldu AAOM m.mumHDMB "mucmnasocH 217 PUBLIC POLICY_ POSITION PERCEPTION OF DISTRICT OPINION IIIII‘I‘7“r-——-_l_________5ROLL CALL C BEHAVIOR Figure~5.l CAUSAL RELATIONSHIP OF PERCEPTION OF DISTRICT OPINION TO PUBLIC POLICY POSITION AND ROLL CALL POSITION PUBLIC POLICY POSITION .57 (.53) .55 (.50) PERCEPTION OF DISTRICT OPINION» IIII“““‘-~—~____~__~__‘~>ROLL CALL .30 BEHAVIOR (.76) Figure‘5.2 GAMMA VALUES FOR RELATIONSHIP OF PERCEPTION OF DISTRICT OPINION TO PUBLIC POLICY POSITION AND ROLL CALL POSITION: FOREIGN AFFAIRS* , *These values are for opposed incumbents of the 85th Congress. The numbers in parentheses are for those incumbents who won the election in 1958 to the 86th CongreSs. 218 PUBLIC POLICY. POSITION .68 (.65) .59 (.63) PERCEPTION OF DISTRICT OPINION I_IIII“““‘--—~—__.________9ROLL CALL POSITION .60 (.56) ’Figurei5.3 GAMMA VALUES FOR RELATIONSHIP OF PERCEPTION OF DISTRICT OPINION TO PUBLIC POLICY POSITION AND_ ROLL AL 'P O : I ' LF * PUBLIC POLICYA POSITION .61 (.70) .83 (.90) PERCEPTION OF DISTRICT OPINION IIIIIIIII7:;I“‘7“--~ll__~_~9§OLL CALL (.77) OSITION Figure 5.4 GAMMA VALUES FOR RELATIONSHIP OF PERCEPTION OF DISTRICT OPINION TO PUBLIC POLICY POSITION AND ‘ROLL CALL POSITION: CIVIL'RIGHTS* *These values are for opposed incumbents of the 85th Congress. The numbers in parentheses are for those incumbents who won the election in 1958 to the 86th Congress. 219 If the incumbent candidates attempted to be responsive, the gammas should be high for at least paths A and C, as Shown in Figure 5.1. That is, the representatives are adopting their policy positions in their campaign or in the legislature so as to closely coincide with their perceptions of district opinion. Path B represents the linkage between their public policy position as publicly stated and their roll call position. Since it is not always possible for a legislator to vote his policy position, these data will indicate to what extent the citizens have elected represen- tatives who vote according to their public policy positions. In foreign affairs, the relationship is highest for path B, the linkage between perception of district opinion and roll call position. 'It appears that the incumbents are not as likely to vote their public policy position as well as their perception of district opinion. In the end, the incumbents seem to be representing their constituents, to the degree the representatives correctly perceive district opinion. On social welfare issues, the incumbents best repre- sent their constituents through the adoption of public policy positions that are close to their perception of district opinion, but this association declines when carried on to the incumbents' roll call voting. An explanation for this may be that since there are so many complex issues involved in social welfare, it may not be possible for representatives to be able to vote according 220 to their perceptions of district opinion. That is, the alternatives that are placed before them may not exactly coincide with what they belieVe their constituency wants and they are therefore unable to vote their perceptions of district opinion. For Civil rights, the strongest association is between roll call voting and policy position, indicating that incum- bents vote according to their announced stands. The second strongest association is between their perception of district Opinion and their roll call position. All the gammas in this area are high, but the gamma of .83 especially shows that the linkage between the incumbents' stated policy position and their roll call voting is very strong. If we examine the difference in the gammas for the incumbents who ran for reelection and thOse who won the election, we see that the gammas declined slightly in the area of foreign affairs, but probably not enough to be significant. In the area of social welfare, they stayed basically stable, but increased the most for civil rights issues. Again, the saliency of the issue may have had some effect on these gammas, since it was likely that civil rights issues were the most salient, and foreign affairs the least. This discussion helps to show that there may be some important differences between the campaign environment and the legislative arena. There are obvious differences with regard to the fact that incumbents must campaign against a 221 designated opponent and employ a variety of multi—media techniques to get their message across to the voters. In the legislature, the incumbents' opponents may very well be only themselves, since their record will establish the ammunition which could be used by opponents in the next campaign. There are also differences in the kinds of choices that can be made in the legislature and in the campaign. A comparison of Fiorina's subjective decision making model14 with the one presented in this study serves to illustrate these differences. In fact, Fiorina seems to agree that legislative and campaign decision making should be studied within different contexts: We argue that there is no necessary correlation between a representative's relative positions in the legislative arena and in the constituency arena.15 This study can therefore provide a means of linking the two arenas in order to achieve a more complete understanding of the dimensions of legislative decision making. That is, instead of formulating models that include only legislative behavior, the models should include some provision for accounting for campaign policy positions, since incumbents must face the voters when their public policy positions do not coincide with their roll call positions. Responsiveness is thus not merely the linkage between perception and roll call, but also between perception, policy position, and roll call. And, according to results shown here, we can state the generalization that the more salient the issue, the more 222 likely the candidate's policy position in the campaign will be the most important indicator of that candidate's roll call vote on that issue. In this regard, civil rights was the most salient and foreign affairs was the least salient issue. Responsiveness appears to occur in either case, for the incumbent's roll call positions were strongly associated with either their perception of district opinion or their public policy position. Incumbents are likely to be more responsive, because, according to our model, they have the most to lose, and they are more likely to believe they know district opinion, so they vote their perceptions. The con- sequence is that as long as incumbents are likely to believe they are well informed and have a good chance to win re-' election, they are likely to continue to try to be responsive to what they perceive to be the majority opinion in their district. This concludes the examination of the data. Chapter six will present an assessment of the significance of these results, as well as some discussion of the strengths and weaknesses of the study. CHAPTER FIVE NOTES 1. Martin Tolchin, "Byrd Persuasive as Senate Chief," The New York Times, 27 March 1977, p. 44. 2. Duncan MacRae, Jr., "The Relation Between Roll Call Votes and Constituencies in the Massachusetts House of Representatives," ‘American'Political Science Review, 46 (1952), 1046-1055. 3. Morris P. Fiorina,‘Representatives, Roll Calls and Constituencies (Lexington, Mass.: Lexington Books, 1974), Chapter 1. See also his "Electoral Margins, Constituency Influence, and Policy Moderation: A Critical Assessment," American Politics Quarterly, 1 (1973), 479—498. 4. Ibid. 5. Ibid., pp. 494-495. 6. Ibid., p.'494. These studies include the following: MacRae, Op. cit., Thomas Dye, ''A Comparison of Constituency Influences in the Upper and Lower Chambers of a State Legislature,“ Western Political Quarterly, 14 (1961), 473-481, Samuel Patterson, "The Role of the Deviant in the State Legislative System: The Wisconsin Assembly," Western Political Quarterly, 14 (1961), 460-473, and P. Personen, "Close and Safe State Elections in Massachu- setts," Midwest JOurnal of Political Science, 7 (1963), 223 224. 54-70. 7. warren E. Miller, “Majority Rule and the Repre- sentative Systems of Government," in Mass PolitiCs, ed. by E. Allardt and S. Rokkan (New York: Free Press, 1970), pp. 284-311. 8. Fiorina, op. cit., p. 486. 9. W. Crane and M. Watts, State Legislative systems (Englewood Cliffs, New Jersey: Prentice—Hall, 1968), p. 87. 10. Fiorina, op. cit., p. 485. 11. Miller, op. cit., pp. 356-376. 12. For a detailed description of the scale items and the creation of the scales, see Footnote 6 in the Candidate File, 1958 Representation Study Codebook, SRC No. 433. 13. The statistics reported in Table 5.9 are different from those reported by Miller and Stokes in their article, "Constituency Influence in Congress," American Political Science Review, 57 (1963), 45-56. This difference is that here we are considering only the opposed incumbents and not the entire sample of incumbents from the study. Also, we are using gamma as a measure of association, while Miller and Stokes chose to make the assumption that the ordinal data could be interpreted as interval data, which is an assumption that attracted some criticism (see chapter four of this work). The statistics presented here can actually help to broaden their analyses since they did not publish the correlations for all paths of the linkage process for foreign affairs and social welfare issues. 225 14. Fiorina, Representatives, Roll Calls and Consti- tuencies, op. cit. 15. Fiorina, "Electoral Margins, Constituency Influence...", op. cit., p. 495. CHAPTER SIX MODELS, DECISIONS, AND REPRESENTATION: AN ASSESSMENT l. IntroductiOn In chapter one, we stated that this study was concerned with determining the extent to which a rational choice model could explain and predict the behavior of political actors in order to develop a set of lawlike generalizations. These generalizations could then contribute to the development of a formal deductive theory of political behavior. In addi- tion, by addressing the question of rationality versus the desire for a representative government, we could make some statements, based on these results, about the responsiveness of public officials to citizen preferences. This final chapter offers an assessment of this research in terms of its contribution toward theory development in political science and its implications for electoral control of public policy. We will begin by briefly reviewing the major aspects of the Bayesian decision model and the results from the hypo- thesis tests. Next, the results will be examined in light Of previous research, followed by a discussion of the significance of this research for a representative democracy. 226 227 2. A Bayesian Analysis of Political Choice The Bayesian approach to decision making has in the last twenty years become widely accepted in its application to business management problems as a means for making optimal or "best" estimates, based on subjective as well as objective information. Only recently has this approach been considered by social scientists as a tool of discovery for developing lawlike generalizations about social and political behavior.1 The model has been shOwn here to be an appropriate means of using all available information in order to make decisions under conditions of risk and uncertainty. In political science, rational choice models have often made the assump- tion of perfect information to develop determinative models or have relied upon objective probability models for developing hypotheses that predict the likelihood of events.2 The Bayesian model can be considered more appropriate in many circumstances because it relies upon not only objective probabilities, but also on subjective probabilities based on the perceptions of political actors. The individual is still the unit of analysis, but instead of presuming that all individuals are exposed to the same quantity and quality of information, the Bayesian model allows for analyses which compensate for variegated interpretations of information by different types of individuals. These people may perceive high or low uncertainty and high or low risk, which will affect their political deCisions. The Bayesian model is 228 therefore a powerful alternative decision making model, because of its ability to include all available information and revise individuals' subjective estimates based on new information. The reconceptualization of political problems that have been studied by political scientists have proved to be quite fruitful. For example, Fiorina's3 study of legislative voting behavior not only opened new avenues of research, but was able to advance the study of constituency influence by indicating how apparent contradictions in the literature could be discovered and explained through this approach and by generating a set of nonobvious hypotheses. Fiorina's model, hOweVer, was concerned only with legislative decision making and no clues were given as to how legislators formulated their estimates of the states of the world, or how various levels of risk and uncertainty were likely to affect their decisions. In this study, we have applied the Bayesian framework to the study of campaign decision making and the kinds of policy decisions candidates are likely to make in a situation involving risk and uncertainty. We contended that in order to describe the complete process of constituency influence, it was necessary to develop a two-stage model, with the first stage linking the constituents with the candidates in a campaign, and the second, as investigated by Miller and Stokes,4 linking the constituents with their representatives, the winners of the campaign. Since candidates are subject to different kinds 229 Of constraints and pressures in a campaign in comparison to the legislative arena, what the candidates promise their constituents is not always what they are able to (or maybe even want to) deliver through legislative voting behavior. By studying campaign policy making in a separate model, we can work toward identifying the important factors that influence public policy, which can in turn provide a better description of the overall process of representation. Then the relationship between the policy positions of the politi- cian gpa candidate and politician gga legislator could be more thoroughly investigated, and a determination made as to why winning candidates' policy positions do not coincide with their legislative voting behavior. We developed a subjective decision making model similar to Fiorina's in many respects, and developed concepts relevant to campaign behavior. However, we went beyond his basic framework to consider hypotheses related to uncer- tainty as a process in which candidates would try to seek information about the world in order to increase their confidence in their estimates of district opinion. The desire for a model to include the concept of uncertainty was expressed by Shepsle: Not included [in his paper], for example, are any manifestations of uncertainty in candidate decision making except as they arise in the game context of strategy selection. Thus, it was supposed that there is no uncertainty in candidate information about voter preferences or strategy constraints. It would be of great interest to incorporate questions of this sort. 230 Moreover, it should be emphasized that the results in this paper depend upon highly restrictive assumptions. Increased mathematical generality, then is a first order concern.5 The model we developed was presented in a simplified form, but increased mathematical generality could easily be achieved by considering continuous, rather than discrete distributions. Also, our assumptions regarding candidate information was not overly restrictive, since perfect information was not assumed, only that some candidates were likely to be more confident about their knowledge than others. Thus, what we have done has been to present a new approach to the study of political behavior, which, when applied to decision making situations, can develop new kinds of testable hypotheses that can encompass a broader area of concern than present research currently affords. It has the advantage Of a formal deductive system, since it is based on a well developed set of mathematical struc- tures. Therefore, our main task is to interpret these structures in political science terms and test the hypotheses to determine whether individuals act as if they follow the decision calculus of Bayesian decision theory when making political decisions. Chapter three presented an interpretation of the Bayesian model by describing how candidates in a political campaign are likely to make a decision regarding an optimal policy position that would minimize their expected loss under uncertainty. The goal was to test the propositions 231 derived from the Bayesian model which described the effect of risk and uncertainty on candidates' policy positions. To operationalize and test these propositions, we relied upon the data supplied.by.the 1958 Representation Study, conducted by Miller and Stokes. The analysis was restricted by the fact that the survey did not contain questions that specifically asked for the exact kinds of information that were desired for a critical test of the propositions. Consequently, we had to use the questions that best approx- imated the concepts described in the model and thus formu- lated a set of hypotheses based on the information available from the study. Among the drawbacks in using the data were the usual problems of secondary analysis of survey data. Also, the time frame in which the survey was conducted did not allow control for all internal validity factors and the complete establishment of causal relationships, as in the situation of dependence on sources for information. Despite these caveats, the data set remains an outstanding source of information about candidate perceptions, including nonin- cumbent challengers as well as incumbents. To date, no major survey has attempted to duplicate the thoroughness, size, or scope of this survey. It is an important source of information that had gone largely untapped because its public release was delayed until about 1971. The first set of publications authOred by the principle investigators are now considered classics and are frequently cited in the 232 literature.6 Therefore, it is justifiable to first consider the data that are available to determine the goodness of fit of the model before additional resources are expended.7 In this way, the development of a theory can be based on a set of accumulated knowledge, acquired through various interpretations of the model. That is, if the model holds up well against a less than perfect interpretation, then this portends even-more satisfactory results when additional tests are performed. In considering the appropriateness of a secondary analysis of a data set, one should consider the results of the hypothesis tests before making a definite judgment. That is, one should not be critical of the assumptions of a model (which include the interpretation of the validity of the questions posed in the survey) until the predictions of the model can be compared to reality. In this respect, the 1958 Representation Study provides some very encouraging results that point toward the value of the Bayesian decision making model as applied to the study of campaign decision making. When the hypotheses were tested, the following results, briefly stated, were obtained: 1) Opposed candidates who perceive to know how peOple in their district felt about the issues were more likely to follow their perceptions of district opinion than those who seldom knew district opinion. 233 2) Opposed candidates who perceived that people in their district were interested in the issues were more likely to follow their perceptions of district opinion than those who believed people were not interested. 3) Opposed candidates who perceived people in their district were more likely to know the candidates' stands on the issues were more likely to follow their perceptions of district opinion than those who believed people did not know the candidates' stands. 4) Opposed candidates who perceived high expected loss of votes from perceiving people in their district were aware of their stands were more likely to follow district opinion than those who perceived low expected loss. 5) Opposed candidates in noncompetitive districts who were in a high expected loss of investment category (majority party candidates) were more likely to follow district opinion than thOse in a low expected loss category (minority party candidates). These findings were based on the uncertainty and risk propositions that were derived from the Bayesian model. The uncertainty proposition predicted that greater uncertainty would influence candidates to be less likely to adopt their perceptions of constituency opinion since they would be less confident that their perceptions would be correct in identifying the true value of a campaign parameter. The 234 risk proposition stated that candidates who perceived greater risk (expected loss) would have a greater incentive to minimize their expected loss by more closely following district opinion as they perceived it. Whenever the best available indicators of risk and uncertainty were tested, the results confirmed that the variables representing risk and uncertainty were important conditional control variables that frequently enhanced the difference between candidates in the high and low categories. The results varied somewhat according to the particular issue area, but the overall pattern was that when the variables of risk and uncertainty were considered, the difference between the categories of candidates increased. These results help to establish the validity of this approach to studying political decision making. Fiorina was only able to test his model indirectly, and although we do not consider this the beSt possible test of the Bayesian model, we have been able to go two steps further than Fiorina: l) we have provided a relatively more direct test ofta subjective decision making model with good results and 2) we have described not only an optimal or preferred choice, but have also described the process of making the choice by considering that candidates act as if they followed Bayes' theorem to revise their estimates of the true state of the world. This model could also be applied to other kinds of campaign decisions involving unknown parameters. 235 With regard to the second point, Fiorina described C, the probability that a group in a legislator's district cared about the way a legislator voted on a particular issue. In Chapter 4 of his book, he described the process by which a legislator might estimate the value of 9, instead of considering g_as agiven,_which was Fiorina's assumption throughout most of his discussions.8 He stated that Q_was a function of group organization, group cohesion, intensity of preference, and a representative's past voting record.9 Just as we were able to show how candidates could estimate the probability of e, Fiorina's legislators could be Shown to estimate g_in the same way, by employing Bayes' theorem. If the data were available to operationalize Bayes' theorem, one could estimate the value of g, which could then be revised after each vote cast by the legislator. As a result, the difficulty that Fiorina discussed in regard to measuring C_may not be as burdensome a task as he proposed: Enough has been written in this section to indicate the dimensions of the problem we face, and why we began by taking estimates of §_and g_as givens. The questions raised will be topics of future research for a rather lengthy future. Isolating the major variables affecting group strength and concern should not be terribly difficult. But theorizing about the relationships among them and carrying out the measurements necessary for empirical estimation pose no easy task. Yet if the §j and Cf are important variables in a representative 3 voting decision problem, then eventually we must face up to these problems.10 Our point is that it may be possible to estimate‘gj (group strength) and gj, not by considering all the factors 236 that Fiorina discussed, but instead by reconceptualizing the problem in terms of subjective probabilities and the use of Bayes' theorem. Estimates of group strength or the degree to which a group is concerned about legislative behavior could be revised according to new information, either subjective or objeCtive. The research presented in this study thus builds upon the analysis of Fiorina and, by testing a set of related hypotheses, provides additional confirmation of the utility of studying subjective decision making processes in a political environment. In addition, it has helped to explain Miller's findings, which were contradictory to the results of other studies dealing with constituency influence,11 and therefore contributes to a body of empirical literature as well. It also provides a conceptual linkage between poli- tical behavior in a campaign environment and legislative behavior. Both have their roots in constituency influence, yet few studies have attempted to consider both as being part of a larger process of decision making. Our study, along with Fiorina's, helps substantiate the feasibility of studying both processes within one kind of theoretical framework, instead of considering each environment as entirely separate and deserving independent analyses. As a result, greater explanatory hypotheses about political behavior could be developed. With the exhibition of the.utility of the Bayesian decision making model in the context of both the legislative 237 and campaign environments, we would expect that more research will be devoted to this area in the future. The present model can be expanded to consider additional hypotheses based on new assumptions, either more or less restrictive, but the collection of appropriate interview data must remain a high priority as well. Our purpose was to show how a subjective decision making model could be used to concep- tualize and explain candidate decision making under condi- tions of risk and uncertainty. In the process, we have provided a more realistic model, while hopefully maintaining the flexibility and precision to qualify as an important tool of discovery. 3. Toward a Representative Democracy One of the four conditions for a strong linkage between constituency attitudes and public policy was that winners must vote in accordance with their pre-election attitudes. To help achieve responsiveness, these attitudes must at least be in accordance with their perceptions of constitu- ency opinion. In investigating these necessary requirements for a representative democracy, we found that representatives do attempt to follow district opinion, despite whatever ignorance or disinterest is attributed to the voting public. we found that of the opposed candidates, the incumbents were more likely to follow what they perceived to be the opinion of a majority in their constituency than nonincum— bents, and winners were more likely to do the same as 238 compared to the losers. Also, incumbents and winners who perceived people in their district were aware of the candi- dates' stands on the issues were more likely to follow district opinions than nonincumbents and losers, respec- tively. These results indicate that among the candidates studied, representation is likely to occur, not only in the sense that candidates who are incumbents follow their perception of district opinion, but also because citizens continue to elect candidates who are more likely to follow district opinion. Incumbents are therefore likely to be winners and both groups shOwed a high association between their campaign policy positions and their perceptions of district opinion. There is, of course, no absolute level that determines what is an "acceptable" level of representation, but what is important is that incumbents and winners compared to the nonincumbents and losers, are relatively more concerned with following district opinion. The results also showed that the level of association is likely to be higher when the issue is perceived to be more important than the others, as shown in the area of civil rights. Although these representatives appear to be following their perceptions, their actual motives may not be entirely altruistic. This is apparent because in cases where there was shown to be low expected loss, candidates were not very likely to follow their perceptions of district opinion. Only when the stakes are high or when they perceive some 239 electoral loss as a result of being unaware of constituency opinion, do they eXhibit a higher association of their policy positions with their perceptions of district opinion. It appears that as long as representatives and candidates in a campaign believe that the people are concerned about the stands of their representatives, politicians will more likely be responsive. These results coincide well with Fiorina's conclusions that: Instead, in a world of uncertainty, representa- tives may find it rational to act as if consti- ' tuents were watching, mass constituency ignorance to the contrary notwithstanding.12 In sum, responsiveneSs occurs because candidates and representatives perceive, rightly or wrongly, that the citizens or groups in their districts care. The question, though, is why? Why do they believe people care, and do our results imply that nonincumbents and losers don't care? The theory we have presented cannot account for the perceptions that candidates or representatives formulate, it only des- cribes how those perceptions are formed. However, we can speculate that the reason is because of a self selection process that occurs during the electoral process. Kingdon and others13 have found a difference between the perceptions of winners and losers and how the outcome of an election is likely to influence their perception of the voters and of themselves. In Kingdon's study, the winners in his sample developed complimentary beliefs about the voters and the 240 losers developed rationalizations for their losses. Kim and Racheter14 investigated this "congratulation-rationali- zation" effect and found that other factors besides election outcome were possibly more important in determining candi- dates' attitudes. These included competitiveness, career socialization, and political ambition. Fishel, Huckshorn and Spencer, and Leuthold15 have also investigated the losers in congressional elections, but no one clear explanation emerges to shed light on the reason why nonincumbents would be less receptive to following district opinion. However, the problem may not be serious as far as the prospects for a representative democ- racy are concerned. Since nonincumbents occasionally win office and are then likely to follow district opinion, it may actually be the winning that evokes a sense of responsi- bility that nonincumbents do not perceive as strongly until after they are in office. Once in office, the risks of the game increase, there is suddenly more to lose and there is the possibility, however remote, that the citizens may defeat them in the neXt election. Incumbents have a greater chance of reelection, but they are not unbeatable. In order to exhibit some control over the course of events, incum- bents may perceive that they should at least try to follow district opinion in order to reduce the uncertainty of their electoral chances. Nonincumbents may not perceive the policy positions to be as crucial a factor as organizational or financial resources are in order to reduce their uncertainty 241 about their chances of election. .As Downs16 indicated, when the party in power must declare its policies first, under uncertainty, the party out of power need only wait until their Opponents make a mistake and take the minority position on an issue. Then the party out of power can take the majority position and win the election on that one issue. It is therefore the incumbent who may perceive he is being watched, while the nonincumbent is relatively freer to take a position that is not the majority position, except on the most important ones, such as civil rights. The implications for a representative democracy are basically good. We see that representatives are likely to follow district opinion because they perceive people in their district do care about the positions they take. The people's responsibility is therefore to inform their repre- sentatives as to the true position of the majority, whomever that may comprise. CHAPTER SIX NOTES 1. See Gudmund Iversen, "Statistics According to Bayes," in Sociological Methodology 1970, ed. by E.F. Borgatta (San Francisco: Jossey-Bass Publishers, 1970), pp. 185-199. 2. See William H. Riker and Peter C. Ordeshook, Introduction to Positive Political Theory (Englewood Cliffs, N.J.: Prentice-Hall, Inc., 1973) and Richard G. Niemi and Herbert F. Weisberg, eds., Probabilipy Models of Collective Decision Making (Columbus, Ohio: Charles E. Merrill Co., 1972) for a discussion of these models. 3. Morris P. Fiorina, Representatives, Roll Calls, and Constituencies (Lexington, Mass.: Lexington Books, 1974). 4. Warren E. Miller and Donald E. Stokes, "Constituency Influence in Congress," American Political Science Review, 57 (1963), 45-56 and Stokes and Miller, "Party Government and the Saliency of Congress," Public Opinion Quarterly, 26 (1962), 531-546. 5. Kenneth A. Shepsle, "The Strategy of Ambiguity," American Political Science'Review, 66 (1972), 555. 6. Miller and Stokes, op. cit. and Stokes and Miller, op. cit. 242 243 7. It should be noted that the major criticism of the data has been that the correlations between constituency opinion and legislators' perceptions, attitudes and voting behavior were subject to too much sampling error. This charge is based on the small number of cases in each district, which ranged up to about 17 persons. The present study avoids the problem completely, because we are con- cerned only with the perceptions and position of the candi- dates and do not attempt to correlate them with constituency opinion. 8. Fiorina, Op. cit., p. 83. 9. Ibid., p. 84. 10. Ibid., p. 86. 11. See warren E. Miller, "Majority Rule and the Repre- sentative System of Government," in Cleavages, Ideologies, and Party Systems, ed. by E. Allardt and Y. Littunen (Hel- sinki: Transactions of the Westermarck Society, 1964), pp. 343-376. 12. Fiorina, op. cit., p. 124. 13. John Kingdon, Candidates for Office (New York: Random House, 1966), Robert Huckshorn and Robert C. Spencer, The Politics of Defeat (Amherst: University of Massachu- setts Press, 1971), David Leuthold, Electioneering in a Democracy (New York: John Wiley, 1968), Jeff Fishel, Party and Opposition: Congressional Challengers in American Politics (New York: David McKay, Inc., 1973). 244 14. Chong Lim Kim and Donald P. Racheter, "Candidates' Perception of Voter Competence: A Comparison of Winning and Losing Candidates,“ American Political Science Review, 67 (1973), 906-913. 15. Fishel, op. cit., Huckshorn and Spencer, op. cit, and Leuthold, op. cit. 16. Anthony Downs, An Economic Theory of Democracy (New York: Harper & Row, 1957). BIBLIOGRAPHY, BIBLIOGRAPHY Abelson, Robert P. and John W. Tukey. "Efficient Conversion of Non-Metric Information into Metric Information." Quantitative Analysis Of Social Problems. Ed. Edward R. Tufte. Reading Mass.: Addison-Wesley, 1970, 407-417. Adrian, Charles and Charles Press. -"Decision Costs in Coalition Formation." 'American Political ScienCe Review, 62 (1968), 556-563. Aitchison, John. Choice Against Chance: ‘An Introduotion to Statistical Decision Theory. Reading, Mass.: Addison- Wesley Publishing Company, 1972. Alpert, Eugene J. "Capabilities, Perceptions, and Risks: A Bayesian Model of International Behavior." Inter- natiOnal Studies Quarterly, 20 (1976), 415-440. Aranson, Peter H. and Peter C. Ordeshook. "Spatial Strate- gies for Sequential Elections." »Ed. Richard G. Niemi and Herbert F. Weisberg.~ Probability Models of Collec- 'tive Decision Making. Columbus, 0516: Charles E. Merrill, 1972, 298-331. Arrow, Kenneth J. "Alternative Approaches to the Theory of Choice in Risk-Taking Situations."‘ Econometrica, 19 (1951), 404-437. Barry, Brian. Sociologists, Economics and DemOcracy. London: Collier-MacmiIlan, 1970. Bayes, Thomas. "An Essay Towards Solving a Problem in the Doctrine of Chances." Philosgphical‘TranSactions of the ngal Society, 53 (1763), 370-418. Reprinted in Biometrika, 45 (1958), 293-315. Ben-Dak, J.D. and K. Finsterbusch. "Bayesian Analysis: Applications for the Study of Foreign Behavior." International Yearbook Of Foreign Poligy. Ed. Patrick McGowan. Beverly Hills: Sage Publications, 1974, 269-306. Bergen, John R. ."The Structure of Perception." ‘Journal'of ‘ the Association for the Study of Perception, (1969), 1-190 ‘ 245 246 Black, Duncan. The Theory_of COMmittees and Elections. Cambridge: Cambridge University Press, 1958. Black, Gordon S. "A Theory of Political Ambition: Career Choices and the Role of Structural Incentives." American Political Science Review, 66 (1972), 144-159. Blalock, Hubert M., Jr. ‘Social Statistics. New York: McGraw-Hill, 1972. Borch, Karl. The Economics of Uncertainty. Princeton: Princeton University PreSs, 1968. ' Statistical Analyses. Reading, Mass.: Addison-Wesley Publishing Co., 1972. Boyle, Richard P. “Path Analysis and Ordinal Data." 'American Journal of Sociology, 75 (1970), 461-480. Brodbeck, May. "Models, Meaning, and Theories." Readings in the PhilOSOphy'of‘the Social Studies. Ed. May Brodbeck. New York: Macmillan, 1968, 579-600. Bruner, Jere. "What's the Question to That Answer?: Measures and Marginals in Crosstabulation." ‘American Journal of Political Science, 20 (1976), 781-804. Buchanan, William. "Nominal and Ordinal Bivariate Statis- tics: The Practitioner's View." ‘American Journal Of Political Science, 18 (1974), 625-646. Bullock, Charles S., III. "Candidate Perceptions of Causes of Electoral Outcome." New Orleans, September, 1973. (Paper presented at the 1973 Annual Meeting of the American Political Science Association.) Burke, Cletus J. "Measurement Scales and Statistical Models." Contemporary Problems in Statistics. Ed. Bernhardt Lieberman. New York: Oxford University Press, 1971, 92-101. Burke, Edmund. "Speech to the Electorate at Bristol," The writings and Speeches of Edmund Burke. Boston: Little, Brown, COO, 1901' II, 95.-97o Cnudde,_Charles F. and Donald J. McCrone. "The Linkage Between Constituency Attitudes and Congressional Voting Behavior: A Causal Model." American Political Science Review, 60 (1966), 66-72. congressional Quarterly Weeklnyeport, 32 (November 9, 1974). 247 Conn, Paul H., David B. Meltz, and Charles Press. “The Concept of Political Rationality.“ Polity, 6 (1973) 223-229. ‘ Coombs, Clyde H. "Theory and MethOds of Social Measurement." Research Methods in the BehaviOral Sciences. Ed. Leon Festinger and Daniel Katz. New York: Holt, 1953, Chapter 11. Costner, Herbert L. "Criteria for Measures of Association." American Sociological Review, 30 (1965), 341-353. Crane, Wilder. “Do Representatives Represent?“ 'Journal of Politics, 22 (1960), 295-299. ------------ and M. watts.‘ State Legislative Systems. Englewood Cliffs,_New Jersey: Prentice-Hall, 1968. Davis, Otto. “Notes on Strategy and Methodology for a Scientific Science." Mathematical'Applications in Political Science. Ed. Joseph Bernd. Charlottesville, Virginia: University Press of Virginia, 1969. de Finetti, Bruno. "Foresight: It's Logical Laws, Its Subjective Sources.""StudieS‘inygubjective Probability. New York: John Wiley and Sons, 1965, 93-158. de Grazia, Alfred. 'Public and'Rgpublican. New York: Alfred University Press, 1961. Dexter, Lewis Anthony. "The Representative and His District." New Perspectives on the House of Representatives. Ed. RoBert L. PeaBOdy and Nelson H. Polsby. Chicago: Rand McNally, 1963, 3-29. Diesing, Paul. Patterns of Discovery in the Social Sciences. Chicago: AldIhe-Atherton, 1971. Dixon, Robert G., Jr. Democratic Representation. New York: Oxford University Press, 1968. Downs, Anthony. An Economic Theory of Democracy. New York: Harper and Row, 1957. Dye, Thomas. "A Comparison of Constituency Influences in the Upper and Lower Chambers of a State Legislature." Western Political Quarterly, 14 (1961), 473-481. Edwards, ward and Amos Tversky, eds. Decision Making: Selected Readings. Baltimore: Penguin Books, 1967. ------------ , L.D. Phillips, W.L. Hayes, and B.C. Goodman. "Probabilistic Information Processing Systems: DeSIgn 248 and Evaluation.“ ‘IEEE'Transactions on Systems Science and Cybernetics, SSC—4 (1968), 248-265. Erikson, Robert 8., Norman R. Luttbeg, and William V. Holloway. "Knowing.One's District: How Legislators Predict Referendum Voting." ‘American Journal of Political Science, 19 (1975), 231-246. Fiorina, Morris P. "Representatives and Their Constituen- cies: A Decision-Theoretic Analysis." Unpublished Ph.D. Dissertation, university of Rochester, 1973. ------------ . "Electoral Margins, Constituencies, Influence, and Policy Moderation:‘ A Critical Assessment." American Politics Quarterly, 1 (1973), 479-498. ------------ . RepresentatiVes, Roll Calls, and Constituen- cies. Lexington, Mass.: Lexington Books, 1974. ------------ . "Formal Models in Political Science." American Journal of Political Science, 19 (1975), 133-159. , * Fishburn, Peter. Decision and Value Theory. New York: John Wiley and Sons, 1964. Fishel, Jeff. Party_and Opposition. New York: David McKay, Inc., 1973. ------------ . Representation and Responsiveness in Congress: The Class of Eighty-Nine, 1965-1970. Beverly Hills: Sage Publications, 1973. Forbes, Hugh D. and Edward R. Tufte. "A Note on Causation in Causal Modelling." American Political Science Review, 62 (1968), 1264-1269. ------------ . "Communications to the Editor." American Political Science Review, 62 (1968), 1270-1271. Francis, Wayne. "The Role Concept in Legislatures: A Probability Model and a Note on Cognitive Structure." Journal of Politics, 27 (1965), 567-585. Freedman, Stanley R. "The Saliency of Party and Candidate and Congressional Elections: A Comparison of 1958 and 1970." Public OpiniOn and Public Policy, rev. ed., Ed. Norman R. Luttbeg. Homewood, Illinois: Dorsey Press, 1974, 126-131. Friedman, Milton. "The Methodology of Positive Economics." Essays inPosiEive Economics. Ed. Friedman. Chicago: University of Chicago Press, 1953, 3-44. 249 Friesema, H. Paul and Ronald D. Hedlund. "The Reality of Representational Roles." Public Opinion and Public Policy. Ed. Norman Luttbeg. Homewood, Illinois: Dorsey Press,_l974, 413—417. Froman, Lewis. "Inter-Party Constituency Differences and Congressional Voting Behavior." American Political Science Review, 57 (1963), 57-61. ------------ . Congpessmen and Their Constituencies. Chicago: Rand McNaily, 1963. ------------ . "A Realistic Approach to Campaign Strategy and Tactics." The Electoral Process. Ed. M. Kent Jennings and L. Harmon Zeigler. Englewood Cliffs, N.J.: Prentice-Hall, 1965, 1-20. GaltUng, Johan. Theory and Methods of Social Research. New York: Columbia University Press, 1967. Goldberg, Arthur S. "Social Determinism and Rationality as Bases of Party Identification." American Political Science Review, 63 (1969), 5-25. Good, I.J. Probability and the Weighing of Evidence. London: Charles Griffen and Co., 1950. ------------ . The Estimation of Probabilities: An Essay_ on Modern Bayesian Methods. Cambridge, Mass.: MIT Press, 1965. Goodman, Leo A. and William H. Kruskal. "Measures of Association for Cross Classifications." Journal of the American Statistical Association, 49 (1954), 732-763. ------------ . "Measures of Association for Cross-Classifi- cations, II: Further Discussion and References." Journal of the American Statistical Association, 54 (I959), 123-163. Hamburg, Morris. Statistical Analysis for Decision Making. New York: Harcourt, Brace and WOrld, Inc., 1970. Harrison, E. Frank. The Managerial Ogcision Making Process. Boston: Houghton-Miffin Co., 1975. Hays, William L. Statistics for the social Sciences. New York: Holt, Rinehart and Winston, 1973. Hedlund, Ronald D. and H. Paul Friesema. "Representatives' Perceptions of Constituency Opinion." ‘JOUrnal'of Politics, 34 (1972), 730—752. 250 Hershey, Marjorie Randon. "Incumbency and the Minimum Winning Coalition." American Journal of Political Science, 17 (1973), 631-637; ------------ . The Makingycf Campaign Strategy. Lexington, Mass.: Lexington Books, 1974. Hirshleifer, Jack. "The Bayesian Approach to Statistical Decision." Elementary Statistics for Economics and Business. Ed. Edwin Mansfield} New York: W.W. Norton, 1970, 85-101. Horowitz, Ira. An Introduction to Quantitative Business Technigues. New York: McGraw-Hill, 1972. Huckshorn, Robert and Robert Spencer. The Politics of Defeat. Amherst: University of Massachusetts Press, 1971. Hyman, Herbert H. “Secondary'AnalysiS'of‘Sample SurVeys. New York: John Wiley, 1972. Iversen, Gudmund. "Statistics According to Bayes." Socio- ‘logical Methodology 1970. Ed. E.F. Borgatta. San Francisco: Jossey-Bass, 1970, 185-199. Jackson, John. Constituencies and Leaders in Congress. Cambridge: Harvard University Press, 1974. Jennings, M. Kent and L. Harmon Zeigler, eds. 'The Electoral Process. Englewood Cliffs, N.J.: Prentice-Hall, 1966. Jones, Bryan D. "Competitiveness, Role Orientations, and Legislative Responsiveness." Journal of Politics, 35 (1973), 924—947. Kaplan, Abraham. The Conduct of Inquiry. San Francisco: Chandler, 1964. Kelley, Stanley, Jr. Political Campaigning. washington: The Brookings Institute, 1960. Kim, Chong Lim and Donald P. Racheter. "Candidates' Percep- tions of Voter Competence: A Comparison of Winning and Losing Candidates." American Political Science Review, 67 (1973), 906—923. Kingdon, John w. "Politicians' Beliefs About Voters." American Political Science Review, 61 (1967), 137-143. """"""" . ' Candidates' for ’O'f'f'i‘ce'f ' Beliefs and Strate— jgigg, New York? Random House, 1968. 251 Koopman, B.O. "The Axioms and Algebra of Intuitive Probability." :Annals of Mathematics, Ser. 2, 41 (1940), 269-292. ------------ . "The Bases of Probability." 'Bulletin of the American Mathematical Sociepy, 46 (1940), 763-774. ------------ . "Intuitive Probabilities and Sequences."' Annals of MathematiCS, Ser. 2, 42 (1941), 169-187. Kramer, Edna E.‘ The NatUre'and Growth of Modern Mathematics. Greenwich, Conn.: Fawcett Publications, 1970. Kuklinski, James H. and Richard C. Elling. "Representational Role, Constituency Opinion, and Legislative Roll Call Behavior." American Journal'of PolitiCal ScienCe, 21 (1977), 135-147: Kyburg, Henry E., Jr.‘ Probability and the Logic of Rational Belief. Middleton, Conn.: Wesleyan University Press, ------------ and Howard E. Smokler, eds. 'Studies in Subjec- tive Probability. New York: John Wiley and Sons, 1965. Labovitz, Sanford. "The Assignment of Numbers to Rank Order Categories." American Sociological Review, 35 (1970), 515-525. Lamb, Karl A. and Paul A. Smith. Campaign Decision-Making: The Presidential Election of 1964. Belmont, California: Wadsworth Publishing Co., Inc., 1968. LeBlanc, Hugh. “Voting in State Senates: Party and Consti- tuency Influences." Midwest Journal of Political Science, 13 (1969), 33-57. Lee, wayne. Decision Theory and Human Behavior. New York: John Wiley, 1971. Leege, David C. and wayne L. Francis. Political Research. New York: Basic Books, 1974. Leuthold, David. Electioneering in a Democracy. New York: John Wiley, 1968. Levin, Murray B. The New MethodolOgy: ‘A Study of Political Strategy and Tactics. Washington, D.C.: American Institute for Political Communication, 1967. Lindley, Dennis V. ‘Introductionito Probability and Statis- ‘tics From a Bayesian'vieWpoint. 2 vols. Cambridge, England: Cambridge University Press, 1965. 252 ............ . Making Decisions. ‘New York: John Wiley, Luce, Duncan and Howard Raiffa. ‘Games and Decisions. New York: John Wiley and Sons, 1957. MacRae, Duncan, Jr. "The Relations Between Roll-Call Votes and Constituencies in the Massachusetts House of Representatives." American Political Science Review, 46 (1952), 1046-1055. ------------ . Dimensions of congressiOnal Voting. Berkeley: The University of California Press, 1958. Mansfield, Edwin, ed. Elementarnytatisti s‘f r Economics and Business. New York: W.W. Norton and Co., 1970. McGowan, Patrick. "A Bayesian Approach to the Problem of Events Data Validity in Comparative and International Political Research." Comparative Foreign Poligy. Ed. James N. Rosenau. Beverly HiIls: Sage Pfiblica- tions, 1974. McGrath, J.E. and M.F. McGrath. "Effect of Partisonship on Perception of Political Figures." 'Public Opinion Quarterly, 26 (1962), 236-248. Miller, David W. and Martin K. Starr. 'The Structure of Human Decisions. Englewood Cliffs, New Jersey: Prentice-Hall, 1967. Miller, warren E. and Donald_E. Stokes. "Party GOvernment and the Saliency of Congress." Public Opinion Quarterly, 26 (1962), 531-546. ------------ . "Constituency Influence in Congress." American Political Science Review, 57 (1963), 45-56. ------------ . "Majority Rule and the Representative System of Government." Mass Politics. Ed. E. Allardt and Stein Rokkan. New York: Free Press, 1970, 284-311. Mood, Alexander M. and Franklin A. Graybill. Introduction to the Theory of Statistics. Second Edition. New York: McGraw—Hill, 1963. Morris, Raymond N. "Multiple Correlation and Ordinally Scaled Data." Social Forces, 48 (1970), 299-311. Nagel, Ernst. "Assumptions in Economic Theory." 'American Economic Review, 53 (1963), 211-220. 253 Nexon, David. "Asymmetry in the Political System: Occa- sional Activists in the Democratic and Republican Parties.“‘ Amenican Political'ScienCe‘Review (1971), 707-716. Nie, Norman H., C. Hadlai Hull, Jean G. Jenkins, Karin Steinbrunner, and Dale H. Brent. Statistical Package for the Social Sciences. New York: McGraWbHill, 1975. . Niemi, Richard G. and Herbert F. Weisberg: eds. Probability Models of Collective DeCision Making. Columbus, Ohio: Charles E. Merrill Co., 1972. Norpoth, Helmut. "Sources of Party Cohesion in the United States House Of Representatives." Unpublished Ph.D. Dissertation, University of Michigan, 1974. ------------ . "Explaining Party Cohesion in Congress: The Case of Shared Policy Attitudes." American Political Science Review, 70 (1976),_1156-ll7l. Page, Benjamin and Richard Brody. "Policy Voting and the Electoral Process: The Vietnam War Issue." American Political Science Review, 66 (1972), 979-995. Paletz, David. "The Neglected Context of Congressional Campaigns." Polity (1971), 195-218. Patterson, Samuel. "The Role of the Deviant in the State Legislative System: The Wisconsin Assembly.“ Western Political Quarterly, 14 (1961), 460-473. Pennock, J. Roland. “Political Representation, An Overview." Representation: Nomos X. Ed. Pennock and John W. Chapman. New York: Atherton Press, 1968. Personen, P. “Close and Safe State Elections in Massachu- setts." ‘Midwest Journal of Political Science, 7 (1963), 54-70. Phillips, Lawrence D. 'Bayesian Statistics for Social Scientists. New York: Thomas Y. Crowell Company, 1973. Pierson, James E. "Presidential Popularity and Midterm Voting at Different Electoral Levels." American Journal of Political Science, 19 (1975), 683-694. Pitkin, Hanna F. The Concept of Representation. Berkeley: University of California Press, 1967. 254 Pomper, Gerald M. "From Confusion to Clarity: Issues and American Voters, 1956-1968." 'American Political Science Review, 66 (1972), 415-429. Pool, Ithiel de Sola, Robert P. Abelson, and samuel Popkin. Candidates, Issues and Strategies. Cambridge, Mass.: MIT Press, 1965. Popper, Karl R. The Logic of Scientific Discovery. New York: Harper afid Row, I968. Porat, Avner M. and John H. Haas. "Information Effects on Decision Making." Behavioral Science, 14 (1969), 98-104. Pratt, John W., Howard Raiffa, and Robert Schlaifer. Introduction to Statistical Decision Theory. New York: McGraw—Hill, 1965. Pruitt, Dean G. "Information Requirements in Making Decisions." American Journal of Psychology, 74 (1961), 433-439. Radlow, Robert. "Decision Making and the Theory of Learning." Decision and Choice: Collections of Sidney Siegal. Ed. Samuel Messick and Arthur H. Brayfield. New York: McGraw-Hill, 1964, 267-275. Rai, Kul B. and John C. Blydenburgh. PoliticalScience Statistics. Boston: Holbrook Press, 1973. Raiffa, Howard. Decision Analysis: Introductory Lectures of Decision Making Under Uncertain_y. Reading, Mass.: Addison-Wesley Publishing Company, 1968. Ramsay, F.P. The Foundations of Mathematics and Other Logical Essays. London: Kegan Paul, 1926. Ranney, Austin and Willmoore Kendall. Democracy and the American Party System. New York: Harcourt, Brace, and werld, 1956. RePass, David. "Issue Salience and Party Choice." American Political Science Review, 65 (1971), 389-400. Riker, William H. The Theory of Political Coalitions. New Haven: Yale University Press, 1962. ............ and peter C. Ordeshook,1ntrOductiOn to Positive ’ Political TheOry. Englewood, Cliffs, New Jersey: Prentice-Hall, 1973. 255 Ritchie, B.J. "An Exploratory Analysis of the Nature and Extent of Individual Differences in Perception." Journal of Marketing Research, 11 (1974), 41-49. Rohde, David W. “Risk—Bearing and Progressive Ambition: The Case of Members of the United States House of Representatives." San Diego, California, 1974. (Paper presented at the Conference on Uncertainty, Political Processes and Public Policy.) Rudner, Richard. The Philosophy of SOcial Science. Englewood Cliffs, New Jersey: Prentice-Hall, 1966. Savage, Leonard J. ‘The'Foundations of Statistics. New York: Wiley, 1954. ------------ . "The Foundations of Statistics Reconsidered." Proceedingyof the Fourth Berkele‘ S sium. Berkeley: University of California Press, 1961, 575-586. ------------ . The Foundations of Statistics. Rev. Ed., New York: Dover Press, 1970. Schlaifer, Robert. Analysis of Decisions Under Uncertainty. New York: McGraw-Hall, 1969. Schlesinger, Joseph A. Ambition and Politics: Political Qareers in the United States. ChiCago: Rand McNally, Inc., 1966. Schmitt, Samuel. Measuring Uncertainty: An Elementary Introduction to Bayesian StatistiCs. Reading, Mass.: Addison-Wesley PubliShing Company, 1969. Segal, David R. and Thomas S. Smith. "Congressional Respon- sibility and the Organization of Constituency Attitudes." Political Attitudes and Public Opinion. Dan Nimmo and Charles Bonjean. New York: David McKay, Inc., 1972, 562-568. Shannon, wayne W. Party, Constituency and Congressional Voting. Baton Rouge: Louisiana State University Press, 1968. Shepard, Roger N. "Metric Structures in Ordinal Data." Journal of Mathematical Psychology] 3 (1966), 287-315. Shepsle, Kenneth A. “Essays in the Theory of Risk-Taking." Unpublished Ph.D. Dissertation, University of Rochester, 1970. ------------ . "Parties, Voters, and the Risk EnVirOnment: A Mathematical TreaUment of Electoral Competition Under 256 Uncertainty." Probability Models of Collective Decision Making,. Ed. Richard G. Niemi and Herbert F. Weisberg. Columbus,.Ohio: Charles E. Merrill, 1972, 273-297. ------------ . -"The Strategy of Ambiguity: Uncertainty and Electoral Competition." American Political Science Review, 66 (1972), 555-568. Shively, w. Phillips. The Craft of Political Research. Englewood Cliffs: Prentice-HaIl, 1974. Simon, Herbert A. "A Behavioral Model of Rational Choice." Quarterly Journal of Economics, (1955), 99-118. Singh, B. Krishna. "Sampling Theory and Ordinal Measures of Association: A Sociological Perspective and a Critique." Proceedings of the American Statistical Association, Social Statistics Section, 1973, 409-411. Stevens, S.S. "Scales of Measurement." Handbook of Experi- mental Psychology. Ed. S.S. Stevens. New York: John Wiley, 1951, 23-30. ------------ . "Measurement, Statistics, and the Schemapiric View." Science, 161 (1968), 849-856. Stogdill, Ralph M. ed. The Process of Model-Building in the Behavioral Sciences. New York: Norton, 1970. Stokes, Donald. "Spatial Models of Party Competition." American Political Science Review, 57 (1963), 368-377. ------------ . "Electoral System and Representation: United States and the United Kingdom." Chicago, Illinois, September, 1967. (Paper presented at the 1967 Annual American Political Science Association.) ———————————— , "Compound Paths in Political AnalYSiso" Mathematical Applications in Political Science V. Ed. James Herndon and Joseph L. Bernd. Charlottesville, Virginia: University of Virginia Press, 1970. ------------ . "Compound Paths in Political Analysis." American Journal of Political Science, 18 (1974), 191- 214. Suits, Daniel B. "Use of Dummy Variables in Regression Equations." Journal of the American Statistical 'Association, 52471957), 548-551. 257 Sullivan, John L. And Robert E. O'Connor. “Electoral Choice and Popular Control of Public Policy." American Political Science Review, 66 (1972), 1256- 1268. Taylor, Michael. "Review Article:» Mathematical Political Theory." British Journal of Political ScienCe, l (1971), 339-382. Tolchin, Martin. "Byrd Persuasive as Senate Chief." ‘New York Times, 27 March 1977, 1, 44. Tufte, Edward R. "Improving Data Analysis in Political Science." WOrld Politics, 21 (1969), 641-654. ------------ . "Determinants of the Outcomes of Midterm Congressional Elections." American Political Science Review, 69 (1975), 812-826. Tummula, V.M. Rao and Richard C. Henshaw, eds. 'Concepts and Applications of Modern Decision Models. East Lansing, Michigan: Graduate SOhool of Busifiess, Michigan State University, 1975. Turner, Julius and Edward Schneier. ‘Party and Constituency: Pressures of Con ress. Rev. ed., Baltimore: Johns Hopkins Press, 1 . Von Mises, R. "Probability:~ An Objectivist View." Elemen- tary Statistics for Economics and Business. Ed. Edwin Mansfield. New York: W.W. Norton and Co., 1970, 59-67. Wahlke, John C., Heinz Eulau, William Buchanan, and Leroy Ferguson. The Legislative System. New York: Wiley, 1962. Weisberg, Herbert F. "Models of Statistical Relationship." American Political Science Review, 68 (1974), 1638- 1655. White, D.J. Decision Theory. Chicago: Aldine, 1969. Wilson, F.P. "Critique of Ordinal Variables." Causal Models in the Social Sciences. Ed. Hubert M. Blalock, Jr., Chicago: Aldine-Atherton, 1971, 415-431. Winkler, Robert L. An Introduction to Bayesian Inference and Decision. New York: Holt, Rinehart and Winston, 1972. 258 ———————————— and William L. Hays. 'Statistics: 'Probability Inference, and Decision. New York: Holt, Rinehart and Winston, 1975. Zalkind, Sheldon S. and Timothy W. Costello. "Perception: Some Recent Research and Implications for Adminis- tration." Administrative Science Quarterly, (1962), 218-235. "IilliflllllfllflMilli