fr-Z .1 (v " LAM (Iflm _ .'?..'h{ 1""1‘. {ivy-Ft $9.13 P 13.5} Nit-2?. 4 ’1. ,/ ‘7 "11:sz /, '1'. u. . .. . . flipéinfa; "45;" 'fi “I . 4"}?! I if , f ”:5; '"z 2.2 "i?! 5 ['11.] _ ../l' f... , Inn-r IRIS"— ,,_ 'Ir “,1 r '. . "3‘...” ‘11; “rifpné‘c'rh v:- If LEA}? . L_-V ‘1'- ‘1‘!“ , ' 1* .. («71.3— a. . ' ‘33,:3‘. r -41, I;' 9.1.). g, I :1. 13' q 1 33/51“. "1 ”$1. .ng Ifi- {$3.03 KW?» 1:35;}‘33Rf. , g: (‘9 "l:\‘.:' x‘&%13f¥,{h 1353, 11:55:1‘» 1‘ . ¢ .‘.5~‘.{\~.'} ”wk-3&4; L12» \‘- ‘6‘“5‘2.‘ \: I! r‘. ‘\ ‘2‘, f." 4' J ‘ ‘ Vfl§ g; 3 >é§kh(‘:i - ‘ ”fix 1 iv an“ UV. W} .‘J 1'7, ‘ 3; . 3:11:53} . Km; “y? .1, \ 5i“ -.' _.".‘ 3:“ .r,\‘.., «.‘r V 1w. ’f’m’ ., k‘ ‘1 x‘ )' 0.}! l f ' 1'35: 1 -. ‘0‘va ”,1 ‘r @xafiaw Y ‘11 gii§§f .1 ( :3“! J‘ '- ¢ :‘fififih‘g‘ 13;;‘5'1. 23:1,... 0:5 $3}? £37”; ‘1-..'n”.", "#13 % ‘45 . _ _ . . \1. ‘ .‘§\ ., -“ ‘ EA‘ ':\; §:\\1‘z_" $335{\ V‘1‘-:v“§; \z‘.‘ ‘3‘} .\}.§1.'-“\‘1 gym; 5 - ‘11:"V?:AH) 3%}: ‘: 1‘ 9. t p, ' 1.1%...- "h .‘-' ti.‘ *3 1‘!“ A . . 1 ,2} :33“- v- n ‘K ‘ ._ , ' ‘1 v 33:51; 955: (h; 33,3”, )5? 3‘34} 7 . {HY/in. .‘ "r- "I . v . arr/.3? "$1?" 1-,“, {xi-z"? 3 I ' J. .1 l " WEE. ” .31, 249:1.“14’éy ”{- 3/14“" 3" Jlr"'a{m MSW I,“ ‘F" ,, f ) +7511“??? :IJJ ., ‘ " ”I; 31.; 15/” 3'" .1ff’l‘”, I - I 9. "£41,": I I . , :4?‘,)' .12? .2435 $75.1”. "- [L ‘I:I:‘V/” 5(l.‘(,1_'1-,~ '5; V" ”:33". 5.4:: 5%,.“ «I. ‘91“? Jig)!” :1; I: ‘ If? ”JG ,f 5'}; . : " If” , , ".5. .4233? flag :I 1. (Z/ . 7/7 I 'v'.“»».5%" £{:‘% ”"9 {g} “II... %;J £3 .V . I.:/ 42,}?! .f :4},;:.::(' J "f; 1” ,";,j// ‘42,» .41.,in ,1: a LA Lyinqullr/f. rl'I' ‘3 "/ llft’l’l' 5775, .77.. {x v ,I ,3.- -/ in. p“..’. 9.7.5, ;: (Vt-(V flag? I‘- f." I fin’z; [II-’1' ’1‘, 0:1,!!! :3”, - A,” 'i‘Ifi" "#1 6'5“? .j/{Xr'f r 1;": -‘ . 1‘15",sz "J; '1'» jé; £5513. .g'i, 1,1,! _J-- .. '7‘, 7-.- J 5v 1’? , “15:41.!” I" J"~ 1'”): "pg: . ”179/ [47” :I'fr: #4371! 1/55? 7 67:91?" '3. hg‘I/t‘ 1:16 52...: h ‘ '. a? 37",” I“ ,’ ‘ . 7‘? v“. ‘ $3.156": JI.’ 5:}: {n ' ‘7 ”I" .1 ,. 7.1. I a} ,‘w H... . 6151M? v’f; xfl¢2£%?z§.wfiqgfi’ “fizz? Eda”? .h‘g'l, 5:./ - '-J""'I.'Jl"’ ‘. r: n 'f ’f {31;} '€;f/;.’.; .1." mm: ’11,}? .’.”/.ur_ I; ffl'I'ur.‘ . ,3, ,4 "'1' (13%:ch Jfifi’fi 57-46%: ail:’:r’”f" I- 3???? 505:”: “I “I ’ ... ‘?:;l:d ,JV’ ,‘fég‘zxéjv/ g.:_.::' 1;. . - ' 3?] ’- . ‘ {991,1é1',‘ :’ It"?! 1-}? :1. ”"ng, "fly. I:- 3",}: f”?! L. "- I i ""1 ” ’1: {If “WW7" 7" ~'-'.’. I“ H "fih ~ . .J 6;; W ”a"? . ,,.. W f I 2% "v .1: ' , " fig .. 1;?" ‘ I ’ fix. (Rum; “7... . .. "’ 4””. "NE :31 1:.— 70". r I '- £1.25}: - ‘17..."{172'54-17/10' MICHIGAN STATE UNIVER RRRRRRRRRRRRR ._.. lllllfllllllfllJIIINIHHHIIHlllllllllllWlllHll 31293 00534 4472 LIERARY Michigan State University This is to certify that the thesis entitled The Eficd’ op DISLNP‘M (.3) DIHCo p. Mun/Ian dn/ Exams; v “ of Pr or K Win/1c 0,0 #1:: Sent me on amlwle CMAJ( presented by 5q\c\\ Abdullah Dado“ has been accepted towards fulfillment of the requirements for Méi'h": degreeinjm '0 c7 '1 jaw Major pof‘so Date S/lf/QX/ 0-7639 MS U is an Affirmative Action/Equal Opportunity Institution MSU RETURNING MATERIALS: Place in book drop to LJBRARJES remove this checkout from —:-— your record. FINES will be charged if book is returned after the date stamped below. THE EFFECT OF DISCREPANCY, DISCONFIRMATION, AND EXTENSIVENESS OF PRIOR KNOWLEDGE OF THE SOURCE ON ATTITUDE CHANGE BY Saleh Abdullah Dabil A THESIS Submitted to Michigan State University in partial fulfillment of the requirements for the degree of MASTER OF ARTS Department of Sociology 1988 Q” 5/ 7 U. ABSTRACT THE EFFECT OF DISCREPANCY, DISCONFIRMATION. AND EXTENSIVENESS OF PRIOR KNOWLEDGE OF THE SOURCE ON ATTITUDE CHANGE BY Saleh Abdullah Dabil An experiment was conducted to investigate the effect of message discrepancy and message disconfirmation on attitude change. The amount of knowledge about the source’s previous behaviors is the third variable. gThe more prior data points, the more confident should be the subjects in their expected position of the source. 'The more confident the subjects about their expectation the more the M disconfirmation should affect the attitude change; Our experiment of (£3300) found a significant effect of discrepancy on attitude change. The relationship between discrepancy and attitude change seemed linear as predicted. We also found a significant effect of disconfirmation on attitude change. We found that attitude change is a monotonically decreasing function of disconfirmation. For the third variable there was no significant effect of the amount of knowledge about the source's previous behaviors on attitude change. We found that the large amount of knowledge about the source's previous behaviors did not enhance the effect of disconfirmation on attitude change. 1i Dedicated to my parents Abdullah & Norah iii ACKNOWLEDGMENTS I wish to express my appreciation to my advisor, Dr. Stan Kaplowitz for his guidance, kindness and understanding through the course of my thesis. I also, thank Dr. Clifford Broman and Dr. S. F. Camilleri for serving on the thesis committee and for their thoughtful suggestions. To my wife Mashail and my children Abdullah and Nora, I give my thanks and my pledge to compensate them for the intermittent periods in which they went without husband and father. iv TABLE OF CONTENTS List of Tables ..................................... vi Theory and Literature Review ...................... 1 Method ............................................ 19 Overview ............. . ......................... 19 Features Common to All Experimental Conditions.. 20 Manipulation of Independent Variables ........... 22 Major Dependent Variable ..... . .................. 26 Other Dependent Variables ....................... 26 Data Collection ................................. 28 Result ............... . ............................. 30 Elimination of Subjects. ........................ 30 Data Transformations ...... ..... ................. 30 Manipulation Checks................... .......... 35 Experimental Effect on the Subjects' Position... 39 Experimental Effect on Possible Mediating Variables.......... ......................... 44 Discussion and Conclusion .......................... 49 APPENDIX A ......................................... 56 APPENDIX B ......................................... 70 APPENDIX C... ...................................... 71 List of References ....... . ......................... 76 10- LIST OF TABLES Experimental design of Discrepancy and Disconfirmation ............................. Transformation Procedures ...................... Normality for Dependent Measures ............... Homoscedasticity for Dependent Measures ........ The Geometric Means for FUTRATIO ............... Experimental Effect on Attitude Change ......... Geometric Mean(958 confidence interval) of Sentence for Defendant x Recommended by Subjects: by Discrepancy and Disconfirmation level..................... .................. Geometric Mean of Sentence for Defendant x Recommended by Subjects: by Disconfirmation and Amount of Knowledge .......... . .......... The Geometric Means of BADPREV by Df and Knowl. The Means of DEPXBAD by Df and Rnowl ........... vi 25 32 33 34 39 41 42 44 45 48 - THEORY AND LITERATURE REVIEW Among the variables which have been found to influence attitude change are message discrepancy, and message disconfirmation. Discrepancy is the difference between the position advocated by the mes;:;e and the initial attitude of the recipient. Disconfirmation is the degree to which the position advocated and the position expected from the source differ. Our first task is to review the discrepancy literature. Next, we shall review the disconfirmation literature. Finally we shall introduce our current study. Aronson, Turner, and Carlsmith (1963) took the theory of cognitive dissonance as an approach to study the relationship between message discrepancy and opinion change. The theory suggests that when there is a discrepancy between the position advocated by the message and the initial opinion of the receiver, the receiver will be in the state of dissonance. Whenever dissonance occurs, it needs to be reduced. Someone who receives the message and can not communicate with the source, is likely to reduce the dissonance by one of two ways: either changing his/her own opinion or derogating the communicator. According to Aronson et a1. (1963), if the source is ighly crediblelhthere will be opinion change. If u ,g...w-‘-‘ however, the source is less credible, derogation will 1 2 be the easiest avenue to reduce the dissonance. Aronson et a1. (1963) tested the following hypotheses. 1) For a highly credible source, the greater the discrepancy between the communicator opinion and the opinion of the recipient, the greater the opinion change. 2) If the communicator is only a mildly credible one, increases in discrepancy will increase opinion change up to a point. As the discrepancy becomes more extreme, the recipient will resort to disparagement, rather than opinion change, and the opinion change will decrease. Aronson et al. (1963) found a significant effect of the discrepancy on opinion change. The study also discussed why previous studies tended to find linearity or curvilinearity relationships between discrepancy and opinion change. Linearity results when the communication is viewed as highlyflcredible. Curvilinearity results when the communicator is viewed as only moderately credible. Bochner and Insko (1966) hypothesized that opinion change would be curvilinearly related to discrepancy for both high and low credible source. They hypothesized that the hump on the curve would occur further out on the discrepancy dimension for the high credible source than for the low credible source. The study measured three dependent variables: opinions regarding how many hours sleep per night a person should receive, communicator disparagement, and 3 communication disparagement. This study, unlike Aronson et a1. (1963), treated communication and communicator disparagement separately. There were three independent variables manipulated: discrepancy, source credibility, and order of measuring dependent variables (opinion-disparagement or disparagement-opinion). The result of this study showed that there is a curvilinear relationship between discrepancy and opinion change for medium credibility sources (which confirms the hypothesis). For the high credible source, the curve dropped markedly at that end, but the entire curve was nonetheless significantly linear and not significantly curvilinear (which did not support the hypothesis). Hence, this result supported Aronson et al.'s (1963) findings. Jaccard (1981) studied message discrepancy and attitude change. In this study, three independent variables were manipulated:_di§g;gpangy, the_confidence- anmindividual haswinmhigmgr her own position, andmthe €01,115idenceeenjndixidual..,.h§.§s...i§l£h§.-.§§urce (credibility of the source), and each variable has three levels (lngwmodergte, and high). This study resulted in a significant main and interaction effect of all the three independent variables on attitude change. But the effect varied according to the type of the confidence. When the subject is more confident in the source than in his/her own position, the relationship 4 between discrepancy and attitude change seemed linear. When the subject is more confident in his/her own position than in the source, or if his/her confidence in the source and in the self were equal there was little attitude change and there was not any notable arelationship between discrepancy and attitude change. Some aspects should be added to the discussion of discrepancy and its effect on attitude change. Lange and Fishbein (1983) studied the difference between ‘peripheral dimensionsjandifocal dimensions. The focal \ dimension is the positional differences, regardless of the categories to which these positions belong. The peripheral dimension is the different categories of the positions. For example, when a student gets a grade of 50 or over, he or she will pass the test but less than 50 is not a passing grade. So, 49 or less are in a different category from 51 or over. The difference between 50, and 70 is 20, which is the difference on the focal dimension. The difference between failure and success is a difference on the peripheral dimension, because they belong to two different categories. Lange and Pishbein (1983) found an increase of belief change, as the discrepancy increases within the same category of peripheral dimension. And the category differences between the recipient's position and source's position on a peripheral dimension decreased the receiver's acceptance of a message. This 5 resulted in less attitude change. They found that belief change is significantly greater if the position of source and position of receiver were in the same category of peripheral dimension, than if they were in the different category of peripheral dimensions. Hence, a discrepancy within one category has more effect on attitude change than a discrepancy in a different category in opinion change. Therefore, the position of the source and the position of the recipient should be in the same category of peripheral dimension for discrepancy to cause a great deal of attitude change. The previous studies showed some conditions affect the relationship between discrepancy and attitude change. The credibility of the source is an important variable.( If the source is highly credible, the relationship tends to be linear. If the source is moderately credible, the relationship is curvilinear. Jaccard (1981) found linear relationship between discrepancy and attitude change when the subject was more confident in the source than in his/her own position. For the discrepancy to cause the greater amount of attitude change, the positions advocated should be belong to the same category of peripheral dimension as the subject's initial position. Aside from studies investigating the relationship between discrepancy and attitude change, there are also studies devoted to the effect of the disconfirmation on 6 attitude change. The disconfirmation is the degree to which the position advocated by the source is different from the position expected from the source. Eagly, Wood, and Chaiken (1978) studied the effectiveness of message disconfirmation in attitude change. The study manipulated both confirmation and disconfirmation as result of three types of biases: knowledge bias, reporting bias, or both. Eagly et a1. (1978,P:424) defined these two biases as follows: Knowledge bias refers to a recipient's belief that a communicator's knowledge about external reality is nonveridical, and reporting bias refers to the belief that a communicator's willingness to convey an accurate version external reality is compromised. The study created the knowledge bias by presenting the communicator himself as having a strong commitment to some values (pro-environment or pro- business). Reporting bias was created by portraying the communicator's audience as having a strong commitment to some values (pro-environment or pro- business). In all conditions, the communicator advocated the pro-environment position. So, the disconfirmation occurred when the source's background (and/or audience) was pro-business and confirmation occured when the source's background (and/or audience) was pro- environment. 7 Eagly et al.(1978) found that regardless of the kind of source bias, disconfirmation of expectancy led to more persuasion than confirmed expectancy; and the recipients rated the communication moresunbiased-in the disconfirmed expectancy condition, than in the confirmed ones. Wood and Eagly's (1981) study also studied the effects of disconfirmation on attitude change. An attributional analysis was made to explore the stages by which the persuasive messages led to opinion change. This study found that if the message is disconfirming, the recipients attribute the disconfirmation to factual evidence, and then view the communicator as unbiased. This enhances opinion change. In the case of expectancy confirmation, the subjects attribute the communicator's position to the communicator's background and view the communicator as biased. WOod and Eagly's (1981) study found that while perception of the communicator as unbiased increases opinion change, perception of the communicator as biased increases message comprehension. As the message comprehension increases, the opinion change increases. Although the perception of the communicator as biased led to more comprehension which, in turn, leads to more opinion change, the disconfirmed message was still more persuasive than the confirmed message. Wood and Eagly (1981) explained the reasons for this effect by concluding that perceiving the communicator as 8 unbiased, leads to attitude change by only one causal link (unbias ---> opinion change). But perceiving the communicator as biased needs two causal links (bias --) message comprehension ---> opinion change). So, the two causal links weakened the ability of a communicator seen as biased to cause opinion change. The question of the effect of the confirmation (confirmed vs. disconfirmed expectancy) on the amount of attribution processing was investigated by Pyszczynki and Greenberg (1981). They examined the attributional analysis and found that people may engage in more thorough attributional processing, after observing unexpected events, than when they observe expected ones. This finding indicates that the disconfirmation enhances attributional processing. This finding helps us to know the process by which the confirmation and disconfirmation effect might occur. Hunt and Rernan (1984) tested the Eagly et a1. (1978) paradigm in a commercial advertising context. It was hypothesized that disconfirmed bias-related expectancy makes recipients:(l) accept the veridicality of the message, (2) view the message as unbiased, (3) judge the communicator as credible, (4) and engage in more message processing. They declared that the study supports the above hypotheses, except for the third one (disconfirmation causes the communicator to be viewed as more credible). Subjects in disconfirmation group did not rate the I \ 9 communicator more honest or sincere than did the subjects in confirmation group. So, there was no effect of the disconfirmation variable on whether the communicator was rated as credible. Most of the previous disconfirmation studies found an effect of disconfirmation on attitude change. Both Eagly et al. (1978), and Wood and Eagly (1981). found that the communicator was perceived as unbiased when the expectancy was disconfirmed. Both Pyszczynki and Greenberg's (1981) and Hunt and Rernan (1984) found that the disconfirmation caused more attributional processing. The previous studies all investigated discrepancy and disconfirmation separately. There are also studies which have studied both the two variables in one study. Studying both of discrepancy (Dp) and disconfirmation (Df), in one study, has its advantages since the seeming result of any one of these two variables might be a spurious result of the other variable. By means of the orthogonal design of these two variables, we can disentangle their effects. Most of the existing studies in disconfirmation and discrepancy, had colinear designs, where the correlations between disconfirmation and discrepancy \\were close to l or -1. It is a difficult in the / colinear designs to disentanglefThe effect of these two l variables. We can quantify discrepancy (Dp) as follows: 10 Dp=§A'£o. and disconfirmation (Df) may be quantified analogously: DfBEA‘EE. Where 29 is the initial attitude of the subject, and g; is the position expected from the source. In most of the previous studies, disconfirmation and discrepancy designs were colinear as clarified by Raplowitz, Fink, Nemecek, Mulcrone, and Atkin (1988,P:3) as following: By treating discrepancy and disconfirmation as characteristics of an experimental condition rather than of the individual, the typical study treats 2p and g; as if they were constant across subjects receiving the same source description. Given this assumption, we see from the following equation: $322+!!!) '23:)" that disconfirmation and discrepancy differ by a constant (2, - 2;) and are, therefore, perfectly colinear. Hence, it is possible that the observed relationship between discrepancy and attitude change is spurious, with disconfirmation being the true causal variable. Two experiments by Raplowitz et al. (1988) have tried to disentangle the effect of the discrepancy (Dp) and the disconfirmation (Df) in attitude change by examining them in the same study. 11 The first study was conducted in a small church- oriented college. The study sought students' opinions about the number of chapel services per week to be requested. The discrepancy was manipulated by subtracting the initial opinion of the subjects about the chapels, from the position of the source. The disconfirmation was manipulated by varying both the source and (hence, the expected position) and the actual position of the source. This study found a significant linear effect of the discrepancy on attitude change, but no effect of disconfirmation. The second study investigated the effect of discrepancy, disconfirmation, and focus of attention on attitude change. This study had many levels of disconfirmation, whereas most other studies dichotomized this variable (disconfirmation vs. confirmation). The topic of criminal sentencing was chosen for the second study because it is possible to express the message and the responses on a numerical scale with ratio level measurement. Armed robbery from a bank was chosen as the crime based on a pilot study. The crime was chosen from 18 different crimes because the recommended years imprisonment for this crime had the lowest ratio of variance to mean. For all conditions, subjects read that 10 years is the guideline for armed robbery. The guideline of 12 armed robbery of ten years was based on a pilot study. In the pilot study, subjects were asked about the appropriate sentence for armed robbery, and the median sentences found to be 10 years. In the experiment, variance in initial attitude was reduced by making the subjects believe that the sentencing guideline of ten years was based in a consensus of legal expert and was publicly accepted. Judge Walters was the source in all conditions. In all conditions, subjects read a statement about Judge Walters' experience in the judicial system, including his sentences for three previous armed robbers. Subjects, also, read a statement from Judge Walters as he sentenced a certain new defendant (Defendant X) for the same crime (armed robbery). This study varied the positions advocated by the source (Judge Walters). The positions of the source are the different sentences for Defendant X. By subtracting the initihl attitude from the position of the source, the difference is the degree of discrepancy. Disconfirmation was manipulated by creating various expectations of the source, before presenting the Judge Walters' sentence for Defendant X. The disconfirmation is the difference between the sentence Judge Walters was expected to pass, and the ”adtual” sentence he passed for Defendant x. Subjects were given the Judge Walters' three previous sentences for 13 armed robbery, and then they were asked about their expected sentences from the source as a manipulation check for their expectation. The design of the second study was perfectly orthogonaln/which means that the k,._. correlation between discrepancy and disconfirmation \ is O.’ This enables us to separate their effects. J The subject's focus of attention was directed to either source's bias or to the message.{ A source focus was created by asking subjects to keep gpgge Walterg; point of view and his degree of strictness in mind as they read his sentence for Defendant X. A message focus was created by asking subjects to think about the reasons Judge Walters gave for his sentencte They measured attitude change by asking the subjects' position on the appropriate sentence for defendant X, after receiving the message from the source (Judge Walters). The result of the second study found a highly significant linear relationship between discrepancy and attitude change, a marginaly significant quadratic effect of disconfirmation, no significant effect of focus, and no significant interaction effects. The authors proposed one possible reason for the weak effect of disconfirmation on attitude change. The reason is that subjects might not be very confident about their expectation from the source. With large number of previous sentences, the previous defendants will be regarded as typical 14 robbers. If subjects perceived the previous armed robbers as typical, they will be more confident in the expected position from the source. When the subjects get more confident about their expectation, they will be shocked if their expectation is disconfirmed, and will change their initial attitude about the appropriate sentence for armed robbery. If the subject change his/her initial attitude)he/she is likely to give Defendant X a different sentence from the orginal initial attitude. As the confidence about the expected position from the source increases, the effect of disconfirmation on attitude change may increase3l In Xaplowitz et al. (1988) the expectation was formed by only three previous sentences and this small) sample may not have been enough to make subjects confident about their expectation as Raplowitz et al. (1988,P:42) stated: The subjects, however, may have not been very confident about this expectation. This may explain why, when their expectations were disconfirmed, they did not strongly conclude that Defendant X must be atypical (worse than the typical robber). Instead, they may have concluded that the three previous robbers were atypical (less bad than the typical robbers). For this reason, subject should have a good deal of confidence in the prior expectations of the source to have an effect of disconfirmation. 15 This study suggested a further research which our study will take in consideration as Raplowitz et al. (1988,P:43) noted: This discussion has a clear implication for further research. A study should be done which varies the confidence of the expectation, by varying the number of prior data points the expectation is based on. We hypothesize that many prior data points should lead to higher confidence. This, in turn, should lead to more of an effect of disconfirmation than should fewer data points (less confident). Our study has taken the Kaplowitz et al. (1988) study's suggestion for further researchti Discrepancy 23a disconfirmation are again varied. The number of prior datapoints on which the expectation is based, is also varied, to manipulate the third variable. The prior behaviors are either three (for small number of data points) or one hundred (for large number of data points). By manipulating number of data points, we are also manipulating the amount of knowledge about the previous behaviors of the source. The following hypotheses have been formulated for this study: 1) There is a positive linear relationship between discrepancy and attitude change. This hypothesis was based on the finding of Kaplowitz et al. 16 (1988). 2) There is a quadratic relationship between disconfirmation and attitude change. We also predict that the moderate disconfirmation has the greatest attitude change. This hypothesis is based on Kaplowitz et al.'s (1988) finding. \, 2 /“*fi 3) There should be interaction effect of J‘ / disconfirmation and the amount of knowledge about the affl: source's previous behaviors on attitude change. When there is a small amount of knowledge about the source's previous behaviors, there should be little effect, if any, of disconfirmation on attitude change. When there is a large amount of knowledge about the source's previous behaviors, there should be a greater effect of disconfirmation on attitude change. We next have some hypotheses dealing with the specific situation dealt with in Kaplowitz et al. (1988) which we are replicating) 4) With the increase of disconfirmation, the judgements of badness of previous defendants decrease. The judgements of badness of previous defendants decrease because with greater disconfirmation Defendant X's sentence (the new sentence) is increasingly greater than the previous defendants' sentences. This makes subjects conclude that Defendant X must be worse than the previous defendants, therefore, the judgement of badness of previous defendants will decrease. 5) The larger the amount of knowledge about the 17 source's previous behaviors, the lggg the effect of disconfirmation on judgements of badness of previous defendants. .7 With large amount of knowledge about the source's previous behaviors, increasing new sentence will not have much effect on the subjects' confidence about the badness of previous defendants because previous sentences are averages of large number of cases of armed robbery (100 sentences). The previous defendants should, therefore, be assured typical. 6) With the increase of disconfirmation, the judgements of badness of Defendant X increase. The judgements of badness of Defendant X increase because with greater disconfirmation Defendant X's sentence is increasingly greater than the previous defendants' sentences. This makes subjects conclude that Defendant X must be worse than the previous defendants and then the judgements of badness of Defendant X will increase. 7) The larger the amount of knowledge about the source's previous behaviors, the 39;; the effect of “ disconfirmation on judgements of badness of Defendant X. With large amount of knowledge about the source's previous behaviors, increasing the new sentence should have little effect on the subjects' confidence about the badness of previous defendants because previous sentences are averages of large number of cases of l8 armed robbery (100 sentences), but with Defendant X (one case only) subjects will conclude that Defendant x must be worse than the average armed robbery. METHOD Overview. Subjects were told that we are doing research regarding criminal justice system. Questionnaires were distributed to subjects, containing the pre-treatment measures, the experimental treatment, and the post- treatment measures. The questionnaire provided ten years in prison as the guideline for armed robbery, then some information about a judge and some previous sentences he had passed for armed robbery. Then the questionnaire provided a judge's recent sentence for the same crime. Subjects were then asked to indicate their view of the recent case as their final attitude measures..FOur independent variables are discrepancy, disconfirmation, and amount of knowledge about the source's previous behaviors; This study based mainly on Kaplowitz et al.'s study (1988) with some differences. Kaplowitz et al.'s study has manipulated two different foci of the subject's attention: source focus, and message focushg whereas our study manipulated only the source focus in all conditions. Another difference between these studies is that, we have manipulated amount of knowledge about the source's previous behaviors as the third independent variable. This variable was not manipulated in Raplowitz et al.'s study, but was suggested by Kaplowitz et al.'s study (1988). 19 20 Featgres Common to All Experimental Conditions The topic of criminal justice system was taken from Raplowitz et al. (1988) study. The subjects in all conditions read that the Michigan sentencing guideline for the crime of armed robbery is ten years. They were further told that this guideline is a consensus of legal experts, and supported by a large majority of the public. (See P.59 of Appendix A). After that,fwe asked them about the appropriate \u. sentence for armed robbery} /This serves as the measure /‘ of the subjects initial attitude. Subjects in all conditions read a description about certain judge, called Judge Walters (JW). (See P.61 of Appendix A). They read that (JW) is a judge in a large metropolitan area. He is in his fifties, has gray hair, is married, and has.grown children. This descriptions was based on a pilot study by Xaplowitz et al. (1988). In this pilot study, subjects were asked to describe what they think the typical judge is. The questionnaire provided information about Judge Walters experience in the judicial system, including his prior sentencing behaviors. It states the number of prior sentences he has passed for armed robbery, and states the average, the smallest, and the greatest of those sentences. The smallest sentence was always 208 below the average, and was created via multiplying the average by 0.8. The greatest sentence 21 was always 258 above the average, and it was created via multiplying the average by 1.25. (See P.61 of Appendix A). For all conditions, subjects read the same message from the source (Judge Walters). The message explains the sentence (Judge Walters) passed on Defendant X. Defendant X's sentence was varied in each condition for the purpose of varying the discrepancy and the disconfirmation. (See P.63 of Appendix A). In all conditions the following speech was given by Judge Walters, before pronouncing Defendant X's sentence: By threat of force and violence, you gained access to money which was not rightfully yours. You brandished a lethal weapon and made quite clear that you would not hesitate to use it if your crime were in any way resisted. Since there was no resistance, you did not fire your weapon, but the terror you instilled in all of those present will be with them for a very long time. Clearly, you played a major role in the planning and execution of this crime. Finally, your record shows that this is not the first time that you have violated the laws which create a civilized society. . . . Therefore, I sentence you to [number of years] in the penitentiary. Prior to measuring the final attitude, in all 22 booklets, subjects read the following statement to prevent them from feeling an obligation to keep their initial attitude: In this study, you expressed an initial opinion, but since then, you have received additional information and have had additional time to think about this issue. Therefore, please feel free to change your views. After this statement, subjects were asked to state what they think that should have been Defendant \ X's sentence. This is the final attitude-. (See 9.65 id of Appendix A). In addition to measuring the final . attitude, subjects were asked their views of Judge Walters, the criminals, the crime of robbery, some memory questions about this specific crime, and about their thoughts while deciding the appropriate sentence for Defendant X. These were possible intervening variables. 2 \ Manipulation of Independent Variables Message discrepancy (Dp) is the difference between the message advocated by the source (2;) and the initial attitude of the subject (29). Thengp) is the individual initial attitude which, for most subjects, should be the guideline. Since, in our pilot study, and Kaplowitz et al.'s study (1988), the median initial attitude of the subjects was ten years, which is the guideline, we regarded ten years as the initial attitude (29) of the typical subject. 23 The position advocated by the source (2;) was varied to different levels (10, 22.5, and 50 years sentences) to create low, medium, and high discrepancy. (See P.63 of AppendixA). The reason of choosing 10 is to have a condition with zero message discrepancy. We chose 50 because we needed to have an extreme message in order to get an extreme message discrepancy. A sentence of 22.5 years was chosen to make the ratio of successive steps constant. We expressed the discrepancy and disconfirmation not by the raw numbers but_bythe logarithms because ”if pairs of stimuli have the same gppig, the logarithms of the responses to them will be equally far apart.” (Raplowitz et al. 1988). For the discrepancy we needed three levels with constant ratio between adjacent sentences. In order to get these three levels, we went through three steps; 1) We want lowest sentence to be 10 and highest to be 50. 2) Therefore, our sentences are 10, 10x, and 10x”. Since 10x3 = 50, x = 2.24. 3) Therefore, if D1) =- 1092.24 (EA/go) = Ln(_P;AJ;_P:oS/)/Ln(2.24). our three levels of discrepancy became O, l, and 2. Message disconfirmation (Df) is the difference between the message advocated by the source (2;) and the position expected from the source (2;). We formed the expectancy by varying the average of previous sentences(that Judge Walters gave for armed robbery. There were five different averages: in the different conditions (2, 4.5, 10, 22.5, and 50 years). (See P.61 24 of Appendix A). These five numbers were chosen as the F" source's position (new sentences). According to the ‘s—.—.-__g ..——~ procedure for getting the discrepancy the base of the logarithm is 2.24, so, we took this base also to produce three levels of disconfirmation. In order to get the disconfirmation levels we used the following formula: Df-I 1092.24i2a/Erl = 1n(§A/EF) / ln(2.24). As the result of this procedure, we got three levels of disconfirmation (0, 1, and 2). (01"75. , Due to the three different positions advocated by ( 1,) the source (Ea), and the five expected positions from the source (2;) we could have 15 different combinations (3 X 5) as our experimental design, but we confined it to only 9 combinations (3 x 3). We confined the design A - to the nine combinations, for two reasons. First, we do not want to have a negative discrepancy or a negative disconfirmation.z Second, having these particular combinations makes our design perfectly J orthogonal. The correlation between discrepancy and) disconfirmation in our design is O, which means that 1 2” our design is perfectly orthogonal. Supposedly, if we' included all possible combinations, the correlation would be 0.5, which is neither perfectly orthogonal nor perfectly colinear. For the experimental design of discrepancy and disconfirmation see table 1. 25 Table 1 Experimental design of discrepancy and disconfirmation The Previous New Sentence For Average Defendant X Sentence for NEWSENT (2;) Judge Walters 7 PREVAVE (2;) 10 22.5 50 50 * * 0/2 22.5 * 0/1 1/2 10 0/0 1/1 2/2 4.5 1/0 2/1 * 2 2/0 * * Note. Dp is the number of the right of the slash. Df is the number of the left of the slash. An asterisk (*) means that the combination did not appear in the design. The third independent variable is amount of knowlege of the source's previous behayiprs. In our study, we manipulated this by varying the number of previous sentences given by Judge Walters for armed robbery. The smaller number is three previous sentences. The larger number is one hundred previous sentences. Subjects read either that Judge Walters has 26 passed three, or one hundred, previous sentences for armed robbery. (See P.61 of Appendix A). This ; variable was created to test our hypothesiSJthat \ \ disconfirmation affects the attitude, more when there is large amount of knowledge, than when there is small amount of knowledge of previous behaviors. So, we have 18 different conditions (3 x 3 x 2), 3 Df, 3 Dp, and 2 different amount of knowledge. Major Dependent Variaplg ii {fThe final sentence recommended for Defendant Xris our major dependent variable. (See P.64 of Appendix A). We tried to make most subjects to choose ten years as their initial attitude. The effect of the independent variables on attitude change measured by looking to how the final attitude was affected according to the different levels of the independent variables. Other Dependent Variables As a manipulation check for the subjects' expectation, we asked about the average sentence they expected from Judge Walters in the future. (See P.62 of Appendix A). We expected the answer to this question to be predicted by the average of the previous sentences. The question was stated as following: If, in the future, Judge Walters passed 100 additional sentences for armed robbery, what is your best ggess as to what the average of the additional sentences will be?. 27 Subjects were then asked how sure they were about their expected average. Then, subjects were asked about their expectation of smallest and largest possible values of the average to be passed by Judge Walters in the future. These questions were provided to check the manipulation of amount of knowledge about the source's previous behaviors and to see how .--'——- _.__.__-' confident the subjects about their expectation in large ,r—s— A—r‘ ah... amount of knowledge vs. small amount of knowledge. (See P.62 of Appendix A). We have some magnitude/scaling questions. The, questions measure the subjects' view of the criminals and the crime. Some questions were about Defendant X's sentence, such as how surprising and howfiunexpected ' this sentence was. Some questions were asked about how bad the previous armed robbers were and how bad Defendant x was. (See P.66 and 67 of Appendix A). These questions allow subjects to respond to unbounded scale starting from zero up to any number, The scale was unbounded at the upper end. One hundred units was set as the moderate response for twoe of magnitude scales (SURPRISE and UNEXPECT). For example, if the subjects were asked about how surprising Defendant X's sentence was, they were told to use 100 as moderately surprising. In two of the magnitude scaling variables, namely, how bad were the previous armed robbers and how had was Defendant X, one hundred unit was set as the average response. For example, 28 before subjects were asked about how had previous armed robbers, they read the following statement (See P.67 of Appendix A): We are now going to ask you how gag you think various criminals are. To measure badness of criminals, let us now use a new "yardstick". As your new "yardstick”, imagine that the average armed robber is 100 unit bad. You will then be comparing the badness of various criminals with the badness of the average armed robber. The magnitude scaling questions also measure some cognitive processes implicated in any attitude change as intervening variables. Subjects were asked about the guideline and Defendant X's sentence to make sure that the subjects remembered the main features and serve as a manipulation check for discrepancy. 2192 Callectm A sample of three hundred undergraduate students in a communication class at Michigan State University I J- participated voluntarily. :The students were offered some class credit for participating. Before distributing the questionnaires, we read the oral instruction, which indicated the cover story of our study, that is study of public attitude towards the criminal justice system. Also we explained to the subjects how to answer the magnitude scaling questions 29 and some relevant information. After That subjects received consent forms. The consent forms included some information about their right of withdrawing from the participation any time and other information. KThe various experimental forms were ordered and distributed randomly among the subjects. By interspersing the conditions, we gave all the conditions the chance to be distributed equally. The questionnaire took approximately 45 minutes to be answerd. After all the questinnaires were collected, the subjects were told about the actual purpose of this study. RESULTS Elimination ofpgpbjects We wanted subjects to think that the purpose of our study is knowingrthe attitude toward criminal justice systemJ We found whether they believed this by our question as to the purpose of the study. Some subjects showed that they did not believe in the cover story by saying the purpose of this study is to see how student affected by the Judge Walters' view, saying this is a psychological experiment, or saying the purpose is to see how students keep their view unchanged. So, those subjects who did not believe in the cover story or said they heard or participated in this study before, were excluded from the analysisi‘ We excluded those subjects who heard or participated because Raplowitz et al.'s study (1988) was done prior to this study and the data were collected in 1985) Both of mentioned study and ours used the same topic and similar questions. Due to this elimination we end up with 283 subjects from the original of 300 subjects. Data Transformations {Normality and homoscedasticity;are important assumptions in order to perform analysis of covariance and numbers of statistical test. Examining of \dependent measures indicated that our raw data did not \ ’ “ -if meet these assumptions. 3O 31 To meet the assumptions, we did two different ways. First, we reduced the effect offoutliers by recoding the most extreme responses to be less extreme. Initially, any response greater than 10,000 was recoded to 10,000. For two magnitude scaling questions, the highest numerical response, which was greater than 3,000 was recoded to 3,000. These two variables were SURPRISE and UNEXPECT. The basis for these recoding, was the ratio of the maximum value to the next to the highest value. If the ratio was more than three, the maximum value was recoded to 3,000. Second, we transformed dependent measures by the logarithm since, the logarithm is an appropriate function to use with positive skewed data. Since, we have a zero value in most of variables, and the logarithm can not be used with zero value, we added iconstant values for any variableywhich has zero value Ibefore taking the logarithm; For each variable which had zero values, we tried different constants, choosing the constant