~o,_.--o. -'_ V>~b.-u -_ .‘- “0‘- - c, A, '4 \ '3‘. ‘ 1-“ ‘. A u ‘9 ‘ V. It ' ‘3 v. "N k- ‘_ :‘Lru v ‘1‘ K o- " “fl .‘l I c.- h"~ ‘\ ~".‘. , a ‘V; R:- ‘ 4‘ , d. ‘ ‘ , \ N’. v?" Fr’ ‘« “A 5 cu: ‘. ‘« ¢‘~ a “ ABSTRACT CHARACTERISTICS OF LOW INCOME RURAL FAMILIES RELATED TO EXPENDITURE AND CONSUMPTION PATTERNS: AN ANALYSIS OF RURAL POVERTY FOR PUBLIC PROGRAM PURPOSES by Pius Weisgerber Despite current interest in measures which can gauge family level of living status, little empirical work has been done on either a criterion or possible predictors for low-income living levels. This study examines both family expenditures and the value of family consumption as possible level of living criteria; it then examines a number of family characteristics for their capacity to predict the consumption level of low-income rural families. A multivariate regression model is the chief technique employed to show the effect of several social, economic, and demographic character- istics in determining low income spending and consumption. As dependent variables, "total expenditure" and "total consumption" were specified along with six subclassifications of these same two measures. As independent variables, each of the following factors were considered: money income, housing, family size, chief source of income, age and education of the homemaker, race, farm and tenure status, and the proportion of food produced at home by the family. Data were from three geographic areas in which low income was considered a problem. Analysis of the data was conducted in four phases: Phase I -- Money income, housing index, and family size were tested as predictors of the living level. Initial assumptions were that the effects of the three variables were continuous and could be fitted as linear variables. Phase II -- Additional variables were introduced and.more freedom allowed in the way the independent variables were permitted to Operate, 0“». g. 5. a.. .~ ’ ‘A ‘VHw-C ' “VA. VJ nus. - ‘1. .‘_‘ _ '1 ..u .- ;‘~-~¢0¢_‘ - . . “ +':‘ a - . v..- A~ in hair." I V. “ ~v..b-.._ ._ . . -- ‘55” s .— ‘ih-A '1‘, V-. . ..'.-~ ‘ r , fl 4" ‘ _, -._~ -~ .4... h- ~ 7": -: A..r,.~__ ‘ ‘ ~~ l fi‘ v~"rx, :.'~.§.. *‘~;‘ .A a...“ v‘ 'V - h. ’ \ - , n.‘~ A“ .1 L‘v".‘ ‘. in, A. .‘ID ’- "‘~} 6-: ,. - ) ,;p-§."'a J ~. ‘ I v~v . I... “O—- A . I‘r u‘J“ ‘ \n: ’p‘: 5 r’ V 5 .' .. LC. — \r‘b --‘ R I a. ‘1‘.) A. r“ “ 4c-_( . F u ' ‘ v5.4 42“.»...- ~ f‘ H. ‘hy‘-~ .fsu‘. "F C a» -t' u~vh . ‘ c-~ a- ‘L_. 1. I o“: . ' h r. ~ A". f " ¢.. ..‘ 5.: ,3» ‘Lbd ‘ -‘n. | “w \ 7 ‘ O A '3 v '1 3 .F ‘\ K'” a 'J‘-‘ “A ‘4 e.g., transformations were introduced so that specifications could conform to curvature; categorical variables were used to reflect discrete effects when continuous variables did not seem to fit. Dummy variables were specified for all characteristics other than housing index and family size. Phase III -- Additional exploratory regressions were run allowing for the possibility that an "additivity" specification over the whole range of a continuous variable might not be warranted. This method, which was used to test the hypothesis of null interaction effects, used interaction terms in an extension of the dummy variable regression technique. Phase IV -- A final model incorporating insights gained from preliminary analyses was tested. Joint variables were specified which presumed to represent a valid approximation to the underlying theoretical structure of consumption behavior among low-income rural families. The race, age, and major source of income factors were most salient for identifying and segregating the "best" set of family subgroups. "Propor- tion of home produced food" categories were also important in predicting more basic family consumption. Housing index and log of family size were also important predictors in the final model but the slopes of the consumption functions for these two variables were known to vary for different categories of other family characteristics. The conclusion arrived at was that the true effect of the housing index and family size could best be determined by running separate computations for certain population groups such as (a) owners and renters, (b) whites and nonwhites, and (c) families earning most of their incomes from wages, salaries, professions or business, and all other families as a group. «H... u -u._.. CHARACTERISTICS OF LOW INCOME RURAL FAMILIES RELATED TO EXPENDITURE AND CONSUMPTION PATTERNS: AN ANALYSIS OF RURAL POVERTY FOR PUBLIC PROGRAM PURPOSES By Pius Weisgerber A THESIS Submitted to Michigan State University in partial fulfillment of the requirements for the degree of DOCTOR OF PHILOSOPHY Department of Agricultural Economics 1966 ~— _—— ~— .— u.~ ' P-l..‘“ I I -"‘-a‘ ' I “-‘l A. s o. “~.4 'r - " § . out. i .,nr. \ 'vd.“, ‘1~~ ' .' ‘4_“ ."-—‘\ A.-._”_ 1“ r - v'-.._' x.- J. _ -“ ‘ i 'm ' - _-‘ . .’ “.‘ ~ .g\’ ‘- -A‘ , “J- l‘. . . .K c. «‘0 ‘ V‘ ‘ V- ‘N. _..- n r- ‘C. ‘» "& AC KNOW LED GLEN TS Acknowledgment is made of the assistance and encouragement received throughout this study from Dr. James T. Bonnen, Michigan State University. Also, acknowledgment is made to Robert B. Reese, who by relieving the author of many other duties as an employee of the U. S. Department of Agriculture, afforded the time that was needed to do this study. Special appreciation is extended to Faith Clark, Jean Pennock, and Mary Jane Ellis of the Consumer and Food Economics Research Division, ARS, USDA, who supplied the carefully collected data without which this study could not have been undertaken. My thanks also to the secretarial, clerical, and computer programming staff of the Marketing Economics Division, ERS, USDA, for their vital assistance in bringing this study and manuscript into being. Special thanks are extended to Mrs. Rita B. Witten who did nearly all the desk calculator work which went into this study. V vur-vp- ‘. -O'-‘ ,, r“_‘ 6. -- I ~ ‘0... | v y‘\ .- -~ «9 “. ~ «--‘ U-_.. "Q - l ‘ V. -" -.-- I ’ ‘ ~ ‘1 A V , , - "a. . r_-_ ‘ '. '— “‘ ‘ I l “ _ ~-\ O M. -\ 4" ‘-. I a - .r~ . _ J-'v“~ . ~~\;«. K \ ‘ N .7 ‘ _ ‘. .4 .- b ' I... ~. It. - &. ‘ p ,- ‘t x, y H ’ ‘ . “'v \‘ \l. \‘ ; ’~ ~ "\‘ A‘ ‘\ u‘ d.__.-‘ .3 ‘~ \‘ 33 I .’~ w: ’ 1 \“ y w. ‘ ’ .‘:‘.I ‘n. y... ‘ ‘Q \- p._ \ 'h ~'. F'" ‘ , ') ‘._‘. I 0 - .\ V‘. ~, .. . \ y’v “ ‘ a .‘III‘D \ 2‘).~ bus IVE‘A.‘ ‘\ \_‘ \ ~ TABLE OF CONTEIES (HMPTER FAG” I . INTRODUCTION Qualifications for Public Assistance . . . . . . . . . . . . . . . 2 Objectives of the Study . . . . . . . . . . . . . . . . . . . . . A Previous Work Done . . . . . . . . . . . . . . . . . . . . . . . . 5 me Primm‘y Data: 0 O O O O O O O O O O O O O O O O O O O O O O O 0 10 Design and Methodology . . . . . . . . . . . . . . . . . . . . . . 13 II. SPECIFYING THE MODEL The Role of Proxy Variables . . . . . . . . . . . . . . . . . . . 2O Specifying the Dependent Variables . . . . . . . . . . . . . . . . 22 Specifying the Independent Variables . . . . . . . . . . . . . . . 2o Analytic Procedure . . . . . . . . . . . . . . . . . . . . . . . . 33 III. EAST TEXAS A General Description of the Area . . . . . . . . . . . . . . . . Al Description of the Survey Sample . . . . . . . . . . . . . . . . . A2 The Analysis Pklase I O O O O O O O O O O O O O O O O O O O O O O O O O O 0 1+3 Pilase II 0 O O O O O O I O O O O O O O O O O O O O O O O O O O 55 Phase III 0 O O 0 O O O O O O O O O O O O O 0 0 O O O O O O 0 ()6 Phase IV C O O O O O O O O O O O O O O O O O O O O O O O O O O 83 IV. BLACKLAND TEXAS A General Description of the Area . . . . . . . . . . . . . . . .106 Description of the Survey Sample . . . . . . . . . . . . . . . . .l07 The Analysis Phase I O O 0 O O O O O I O O I O O 0 O O 0 O O O O O O O O 01.03 Phase II 0 O O O O O O O O O O O O O O O O O O I O O O O O O .111 Phase III 0 O O O O O O O O O I O O O O O O O O O O O O O O 0 111+ Phase IV 0 O O O O O O O O O I O O O O O O O O O O O O O O O .121 V. SOUTH CENTRAL KENTUCKY A General DescriptiOn of the Area . . . . . . . . . . . . . . . .135 Description of the Survey Sample . . . . . . . . . . . . . . . . .136 Q‘. CHAPTER VI. VII. VIII. IX. The Analysis Phase I O O O O C O O O O O O O O O I O O O O O I C O O 0 Phase II 0 O O O O O O O 0 O O I O O O O O O O I O O O O 0 Phase III 0 O O O O O O O O O O O O O O O O O O O O O O 0 Phase IV C O O O O O O O O O I O O O O O O O O O O O O O 0 THE STUDIES COI-IIPARED Population Comparisons . . . . . . . . . . . . . . . . . . . . Sample Comparisons . . . . . . . . . . . . . . . . . . . . . . Analytic Comparisons Phase I O O O O O O O O O O O O O O O I O O O O O O O O O Pllase II 0 O O O O O 0 O O O O O O O O O O O O O O O O O 0 Phase III C O O O O O O O O O O O O I I 0 Phase IV 0 O O O O 0 O O O O O O O O I O O O O I O O C O O COECLUSIOHS Empirical and Theoretical Conclusions -- Dependent Variables . Empirical and Theoretical Conclusions -- Independent Variables Methodological Conclusions . . . . . . . . . . . . . . . . . . APPENDIX BIBLIOGRAPHY PAGE ~13? .1110 .lh2 .lA9 .163 .16A .166 . 169 .175 .177 .183 186 19k s-‘_ .r v‘ U“: .. . . ‘ . ‘u ‘ -‘ ~‘éh ___ C 5'- O“- -""n.,. :" -. ”v. "I” u... v. ., .. '- B ‘N ‘ .‘4- "' “u ‘b. ‘1 h | u ‘H V ‘\ ¥~ __ . CHARACTERISTICS OF LON IRCOME RURAL FAMILIES RELATED TO EXPENDITURE AND CONSUMPTION PATTERNS: AN ANALYSIS OF RURAL POVERTY FOR PUBLIC PROGRAM PURPOSES CHAPTER I INTRODUCTION The Economic Opportunity Acts of l96u and 1965 draw increasing and renewed focus on a special segment of American society -- the poor and the impoverished. The Congress has enacted several public programs to help those in very low income households achieve a better level of living. Most antipoverty programs attempt to help people rise above the poverty level by providing them the means to improve their employability, or by giving them better opportunities for greater self-help. Some other programs are designed to alleviate the effects of poverty, but do not move the recipient poor out of the ranks of the impoverished on any self- sustaining basis. The aged, the ill, and the incapacitated are essentially isolated from the productive world around them, and they can be aided only by direct income transfers or specialized programs. A direct attack on their poverty status can be made with such assistance as improved low-cost housing or supplemental food. The purpose is to upgrade the living con- ditions of those who are in poverty. It is with this latter type of program that the present thesis is chiefly concerned. The ultimate objectives of the President's "war on poverty" are clear but there are and will continue to be many unanswered questions. There are uncertainties about many things, including the way prOposed programs should be designed and implemented and about the degree of success which alternative approaches will have in operation. There is a need also for periodic assessment of a program's achievement. Our state of knowledge concerning the rural poor is so inadequate that we are not even very sure how best to approach research related to the problem. This thesis is somewhat of a pilot study designed to provide some basic information about the levels and patterns of consumption among low income rural families. It examines samples of households drawn from among "open country" dwellers in three low income areas of Kentucky and Texas. Knowledge gained from the study is intended to help public program administrators to better understand consumer behavior of low income rural people and provide for a more realistic adaptation of antipoverty programs. Qualifications for Public Assistance The process of classifying individuals or households as to eligibility for particular benefits presents many problems. A concise, simple defini- tion of need will not detect important distinctions in actual well-being. More precise or more realistic definitions, on the other hand, may defy statistical analysis with data currently available. Nevertheless, a boundary about the poor must somehow be described, even if it only grossly defines a level below which family resources must not fall if a minimum living standard is to be maintained. This boundary or minimum standard is not established simply on the basis of characteristics of the low income population, but by an ultimately arbitrary determination made in the political decision making process and involving a wide range of matters including administrative and political feasibility. The process by which needy families are certified as eligible to participate in a particular program is typically a large scale undertaking. The qualifying test for participation has to be applied to many thousands of families; therefore, it_is not feasible to apply a very complex defini- tion of need to each individual family. Instead, the measures used to establish need should be objective and readily obtainable, requiring a minimum expenditure of time by State and local certifying agencies. Thus, . . .. ,. _. \‘"“:~ «2".31 ~vu§ ~:\'~--- . . - .‘ ~— . ._, it "5 "V “ ' "~V'-‘ .v “‘5 ‘s-u _ h . ;‘~. “2"“:- ‘ Vn~-;-s.‘ . “T :- gv‘. yl «A Va, “u‘ ~. A-.. I‘ iuth“ nu , ‘4 vw. ' Au.” H “ v-.. ‘ y. W v.‘-‘-‘ H -.-__ “A . ~§\.K,-_ ‘vygl _._ h-O‘ .. _._‘ r‘ .~‘- — .. Q .. , a. 'N r‘ n A.._' . “‘7. I v. ’,~ "‘“v :.‘ « -- - -J‘.. Q.- 71“» ‘ - 1-. “"~ “‘t. - K V ' x A . I» "ck II- :- «V‘, :- ‘fi" .‘. - ~.‘ 7“ u; e s" _ \‘ I. ‘u: ~ \ P' r '5‘ - “‘ ‘y 2m, . ‘g‘; n M‘ ‘5. . N“~s -‘Q s ‘ u . 2m. . \.‘ ,- htflr.‘ . ‘ r- -"\I #— ‘ P v n u \ fl . __. ‘5‘7‘n‘ V‘u "".‘ ‘ ‘u -Q .. u ‘2‘). ‘ ‘. N" g >. ‘t. t": hm \ \n~ 3. a limited number of characteristics associated with low income families and some specific set of needs (e.g., housing, food, medical care, etc.) are selected for use in identifying the population to be served by a specific program. Perhaps the most common measuring rod used for gauging the living level, and thus the need for assistance, is the current money income. However, money income is not a good.measure of family living status because nonmonetary income often comprises a large portion of rural family resources. Furthermore,.money incomes fluctuate a great deal among low income families and thus the current income rate does not necessarily reflect average family well-being. The asset level of the family is another factor that is commonly taken into account to help determine the family's need for public aid. Where there are sufficient liquid assets, the consumption level may be reduced very little following a sharp reduction in family income. On the other hand, a family with few or no assets to draw upon may have to reduce its spending severely when there is a large drop in income. Family size is still another variable that is frequently associated with classifying families when their need for assistance is being deter- mined. It seems reasonable to use higher income cutoffs for larger families and lower cutoffs for smaller families. This approach presents a more realistic approximation to the minimum needs of a family than would any single across-the-board income cutoff for all families. Differences in family size are also important in characterizing the low income population. The incidence of low incomes is particularly high among the aged and in rural and nonwhite families. l/ The aged are generally associated with small families, while rural and nonwhite families tend to be larger than the national average. l/ Source: Economic Report of the President, transmitted to Congress January l96h, p. 81, Tables ll, 12. :- F‘- * 5“— -.- . - ‘wfir"-rvé- \. “-t‘v. uC.“ . . . ‘ "w'fir -~,.' ... I. -.“ b.“~ .- Q . ‘ CH‘ 0.: "~ 5v“ —. “Q V- V..“‘ \ | 9". .r;'—. v-‘ ‘- “,2 a..- N «.v ‘ ‘-'C. - 5‘ A‘ V‘ A‘- . - v‘\_ ._ a,“ u- 4., kl.‘ . - ~r- p sea T‘,I\ “. p. u‘.d\ \ . Q‘. n \ _. H A. \- “~¥ . ‘ . ‘. \‘i §~ 5‘ \ \. \C“— ‘- \ 5. \J « Objectives of the Study As noted in the previous section, when families are certified for program participation, among the measures most commonly obtained are their monetary income, family size, and the level of their liquid assets. An important objective of this study is to measure the degree of association which exists between characteristics such as these and the rural family's level of living. For the purpose of this study, the level of living or well-being of the rural family will be measured alternatively by their annual expenditures on family living and by the value of their total family consumption for one year. These two measures will be described more fully in Chapter II. It is recognized that in the family experience, the actual living level encompasses more than either the level of expenditure or the level of consumption. As Davis points out, the level of living is made up of a complex combination of consumption, working conditions, possessions, freedoms, and "atmosphere" and the balance or harmony among them, in relation to needs and felt wants. l/ In this study, however, consumption or expenditure are the only components which are measured and available, and the term "level of living" is used in the thesis when referring to those measures in a generic sense. The terms "level of consumption" and "level of expenditure" are reserved for use when referring specifically to one or the other of the two measures. The poor in this study will also be classified by various additional characteristics to see which family traits have a degree of association with expenditure and consumption. The measures and traits that show a relation with the living level can then be combined in a multivariate 1/ Joseph S. Davis, "Standards and Content of Living," American Economic Review, Vol. xxxv (March 1915), p. 7. . “.2 “us?" fiv' - y- .»u~.g., v . u "" ““r- r". .“" x .- “.»-u u.... . "F’:"P.-‘- \“ ”M u‘. i..- -. "" ‘r »._‘A. ‘. 1-. VJ b... ~_U- O 5. regression model. The results should provide information of use to policy makers and administrative decision makers in identifying the salient characteristics of the rural poor. This kind of information is necessary to the process of identifying the poor, both within an area and in specifying the geographic distribution of the poor between political sub- divisions -- and thus in designing a system for allocating program funds. It also is necessary in the process of designing or redesigning programs for the rural poor. A further objective of the study is to gain a better insight into methods for studying low-income, low socioeconomic status families. It was considered likely that a different set of influences would be strategic in determining the behavior of low-income families, as compared with groups belonging to higher levels of income. Previous Work Done Certain evidence does exist that lower income households behave in a manner which is distinct from that of the main body of households. For example, in a study which examined factors associated with the variance in food expenditures among a national sample of households, Herrmann found an atypical expenditure behavior on the part of households from the lowest socioeconomic categories of that study. l/ On the basis of his study, he concludes that neither social class nor education of the homemaker seem particularly significant except insofar as they separate out the "deviant" extreme low-income group. He suggests separate study of the deviant group. Another study that appears to attribute a distinct mode of behavior to low-income families was reported by Rockwell. In his study based on the same survey as that examined by Herrmann —- namely, the 1955 Household Food l] Robert O. Herrmann, Household Socioeconomic and Demographic Characteristics as Determinants of Food Expenditure Behavior. Unpublished Ph. D. Thesis. Michigan State University, 196+, p. 82. J a flaw —v.—.... .A . A. .__. W... .45.... -.n-O...‘..__ ._ _‘ _ - . van-Is - _ . p 'I' cv- - -~u| um... . v . --,.- "__- . \ I _ .V‘flv "u-V . .- . ‘v u- . C? ’r; o.“ c... Hk" V -—.._ . . . - r ._ , .__ A \I “‘ ‘v ~ .4 m.-.“ A.— .. . “‘“u 1;- - .. _ ‘M ‘|"..~- ,. H. ’u- _ r - " y «- ‘ a. . ‘H- 5 xi... "v,-. — --... . ‘V-r’>._ sv , v ‘¢.~‘-v'.‘ _ '5‘. .‘ _“k ‘ ‘N. .. v._ _, -.. _‘ ~ A ,' ' . _. y .. h'r. , ’2“ .4 . ‘ “~.‘ , . ". . s...- ",.._ “--.J .- .. - 7.. E.“ c.‘ v» A if A . ‘~ a: ‘..- ‘ "‘ . r -“ ~ ’.v~ ~‘ _ "n. ‘ ‘ a. .‘J ‘ UK“ ;_ y,‘ x V“ ' ‘ ~“- g», ‘~~.' ‘8. s wg‘ L- ‘ e ) L, - , Consumption Survey -- Mr. Rockwell reports as follows: "The income elasticity of demand for fluid whole milk was found to be only moderately high, and then only in low income households." 1/ In a more general statement he says: "Low-income nonfarm families whose incomes increase significantly tend to expand their food purchases more, percentagewise, han do high-income families receiving proportionate income increases -- as one would expect on the basis of angel's law. Low-income farm families, being less dependent on purchased food, tend to increase their food con- sumption very little in response to significant income increases." In still another study where Martin David examined family composition and consumption, he describes his results that related to automobile con- sumption. 2/ David found significantly different regressions for high and low income groups -- the marginal prOpensity to consume out of the incremental income was found to be greater among the lower income brackets. Nearly all research relating to low-income families is reported in a manner similar to that of the three examples cited above. In other words, low-income family behavior is usually examined within a study of broader scope, and the results pertaining to low-income families are thus reported as an incidental part of a larger study. The results are in general what one would expect from economic theory. Besides the analyses that have been based on individual cross-sectional studies, there has also been considerable work done to compare the findings of studies of a similar nature conducted in two or more periods of time. The results of the 1955 Household Food Consumption Survey have been compared, for example, with a similar U. S. national food consumption survey which was 1/ George R. Rockwell, Jr., Income and Household Size--Their Effects on Food Consumption. U. S. Department of Agriculture, MRR-3AO, June 1959. Summary. ' g/ M. H. David, Family Composition and Consumption, "Contributions to Economic Analysis" No. XXV; Amsterdam: North Holland Publishing Co., 1962, pp. 86-89. - x .'.H ~£._ ”H . Humv 5-5» a _ '5‘- \ ue,__nfip - \ v I"“Uu.‘ . \ 9,... _ .,._ " . A“... “a“. """*t-—-, " ‘ _ ,‘ yv4~ ‘- a. ( ' 'a. C“v».‘ ' no ~n'—«—--_:”v » J . .5, . - '1‘: . .‘n‘b... \:H~ ‘ , “.2. .l r a ‘.‘.‘-"\A ~ ‘ -.-. w._ e H V .“ h A 7-“-.r* ' ’ .- V ‘dv. I 4- .‘ . . 3"f " bd‘ut‘ ‘.G .v ‘ q J": a,“ Uu~-' - v a' V- I ‘ ‘s “b‘: v ‘- .“ v.~“ .5 ‘v ‘~~) ‘A . ' c. w ,R‘ U “wig, ‘._ 9“: "w. “h- . *s_:>\ ‘ .‘b. p A .X‘ 7. made in 19A2. Most of the comparisons between the two points in time are made between similar groups of families such as the same income class and within the same urbanization class. Marguerite Burk reports an interesting comparison between the l9h2 and 1955 studies that concerned low-income farm households. She sought to explain a rise in the market value of food consumed at home which had occurred among low-income farm households between 19u2 and 1955. ;/ After considerable research, Burk found the significant element to be the different sample composition among low-income farmers that existed in the two survey periods. In 1955 the ratio of cotton farmers (especially sharecroppers) to tobacco and general farmers, had declined substantially from what it had been in 19h2. A study of separate survey reports for Southern farming areas provided the information needed to attribute a large part of the food consumption increase to the decrease which had occurred in the proportion of cotton farmers. Besides the large scale studies that have been made relative to food consumption only, there have also been a number of national surveys made on consumer expenditures for all goods and services. The Bureau of Labor Statistics, U. S. Department of Labor, conducts such surveys periodically so that the weighting factors used for the consumer price index may be kept current. Perhaps the most comprehensive BLS study was made in 1950 and 1951, covering 1,500 items of budget information collected from each of 12,500 families and individuals in all income and occupational classes. Many volumes of statistical data based on the 1950 and 1951 surveys have been published, g/ and there has also been a series of research studies of the economic and social aspects of consumption and savings. 1/ Marguerite C. Burk, Influences of Economic and Social Factors on U. S. Food Consumption, Minneapolis: Burgess, 1901, p. 93. 2/ The data tabulations and cross—tabulations are available in Studies of Consumer Expenditures Incomes and Savings, Philadelphia: University of Pennsylvania, 1956, 18 volumes. V..~ ._ ‘ \ o. .I ‘- an r a... _ u ‘ I ‘~~.A, 1 -cc '~'-u. . l a. ,_. _... l v... “,3 ._ K A. .-". D -7 V V. V5‘_V- . - l ‘l . ‘ ....,J . I. '.‘.. .4.\... V-.,_ — LS-xDH,‘ ._,_ ho..- ‘_... - o a S “ P05,“ -. vv ‘u-._. v~~ --‘P~ ». l .i - O . ~V. ‘ V . L ) LIN“ N- hu._’* C. o 5 a- ‘ ‘1‘; ‘4 . . ~ h a a _ ‘A ~..," “A q. ‘ u. . \ ‘- 7 IAN-hr ‘ Mu.) ~f‘ A...» ~~ ‘ . \— c 1 ~ .‘ . ' ‘ I .J‘w, J I a \.-:g A,. J u '» .‘- ‘A 2 v3; ,. ‘-l‘3";_- “-Q n ' . ‘v._‘ ' . "e “M 3 ‘ ~ . ‘PI U“; .\_ ‘ -“‘u P. r . P. f- ‘. * , ‘5 ('3 ‘ . “A ‘ in“ ' b‘~-A“ ' ‘ ‘1- ‘ \K v Q ‘_. v ‘.' ‘ 't '--:, \‘ nun; ‘ ‘ s a. I w I I u 1 L ‘ o C o o 8. One of the important research papers based on the 1950 BLS cross- section data was one in which Crockett and Friend fitted multivariate regressions to ungrouped data. 1/ One of their chief aims was to measure the net effect of income on consumption, while holding constant a number of other family characteristics which were correlated with income and which therefore would bias the income effect if not taken explicitly into account. The main part of the two author's regression analysis was devoted to after tax incomes in the $1,000 to $9,999 range. It was considered likely that such incomes would be reported more accurately and be less affected by transitory or abnormal elements than the incomes at either extreme. The two researchers did, however, also investigate the consumption behavior of the extreme low-income families, as well as those who reported incomes over $10,000. Their appraisal of the overall consumption is as follows: "A complete demand function may thus be presented by 3 straight lines, one corresponding to extremely low incomes, the second corresponding to incomes in the $1-10 thousand range, and the third corresponding to the upper income extreme. In general, both the extreme slopes are flatter than the central slope giving an S-curve effect." 2/ The authors could see the likelihood of certain confounding factors prevailing at the lower end of the income scale. They surmised that when they were dealing with the poorer families, assets and borrowing capacity would become much more important relative to income in the determination of consumption. They also hypothesized that the very low-income families would have larger values of "transitory" income and that the observational errors would be much larger. lf—Jean Crockett and Irwin Friend, "A Complete Set of Consumer Demand Relationships," Proceedings of the Conference on Consumption and Saving, Vol. I, Irwin Friend and Robert Jones (Eds.) Philadelphia: University of Pennsylvania, 1960, pp. 1-92. g/ Ibid, pp. 10-11. P___________,_.. ’_——- . '— M ,——\.‘~ _—_\_ F—~ ._.—— 21—“ J .. “ '. -... .. mung“ -V"’-‘ .-. I .. "‘ v~ 9“,", ”"‘: :MV 0-,. . ‘ "“ w-- \ . 9...... g‘- , ‘ a 1 . "v- cu— 5.. v.._ . " ‘:.“\r,‘ _ “o‘_ u _' __ ‘- V“ 0‘- r,‘ - n "‘ 1 .. "nq " \_ § § .14 _- c . "V'. 'TN , ~.. .. .1 '4. a H 9.,- ” , a: ~ _ K... , ., ‘ ‘. ._q M _ v - K ' . . ~ . a)” nun‘ . ~., ~ . 1“. “x F ‘J ,4 . $4 ( "ra' . .- “ Vi“. ~ '1'..~ ‘\- 51‘ ,— I “‘ ‘ \p‘. "y L’."‘ . *‘s -~ 3 s A, ~« “ uv‘ ‘1 '- ‘A‘ ”L. .N K'- "L' ‘p‘: "4;, ,-,. \g'-< ~. - ‘ l y ‘ N x I a. h. . I .‘4 40 U. ”-AA '“‘.., \ k.“ *Ner‘: ~ ‘- \"-"'_ V‘ r‘ "« ~ ~‘l 9. The 1950 and 1951 BLS data were restricted to U. S. urban areas only. However, in 1955 the U. S. Department of Agriculture conducted a nation- wide survey of 3,8h5 farm-operator families. 1/ The major objective of this survey was to obtain expenditure data for the development of price index weights; however, it has also served very readily as a source of information for studying rural family living expenditures. 2/ In 1960 and 1961 a new national survey of consumer expenditure was made, this time as a joint project between the Bureau of labor Statistics and the Department of Agriculture. To date, the BLS has published a con- siderable number of general purpose tables in its "BLS Report Number 237 and 238" series. It is anticipated there will be sufficient numbers of low-income households within each urbanization class 2/ so that considerable analytical work with low-income families will be feasible for any single urbanization class. The sample is not adequate, however, to do a separate study of a geographic subdivision say as small as a standard metropolitan statistical area. In the Detroit SMSA, for example, a report based on data collected from one-half the full sample had only 19 families with incomes under $3,000. 5/ In recent years a number of surveys have been made which were conducted exclusively in economically depressed areas. They had as their goal the study of consumption behavior of needy families as a group unto themselves. Three such surveys were made in the period 1957-59 in depressed rural areas of Kentucky and Texas where low farm income was considered a major problem. l/IFor further information about this study see: U. S. Bureau of Census. U. S. Census of Agriculture, 195%. Vol. III, Special Reports Part II, Farmer's Expenditures in 1955 (Washington: USGPO, 1956). 2/ See, for instance, Laura Mae Webb's Report to the 38th Annual Outlook Conference, U. S. Department of Agriculture, Washington, D. C., Nevember 16, 1960. g/ The specified urbanization classes are: urban, rural nonfarm, total nonfarm, and rural farm. E/ See Survey of Consumer Expenditures 1960-61, U. S. Department of Labor, BLS Report Number 237-1, November 1962, (Advance Report). I . _ [- .fi. 1|” . u I .- _' "-.v a V! r‘“ w: . ' A-v. . v. . n 5 . z . AQ. i‘am, __-.. . «HQ, . U “‘- I ‘.. 'h‘ . ' '. _ . — . “h. ' . .,' s; L . o .4.“_ «-. -'-\~ -| “ v.2 -\ - H' v ~ I ~ ‘-.; .7 v— 5. 1. . ~..:. \ 'A ‘N Nr‘u . . . \ a ‘Q \ V‘ v r .‘ C. - ‘nfi. L.‘ h. ( 4'~’.§ E 'c N '7. § LA ‘~- 'ci *Q P 4:. .. a3 . 1‘ ‘9.- .‘ ~‘~ '*;~'~‘ . I. 4“ Q I 5-: 1 . \~A 151‘ . A - v JV’V ., ‘l .5 g- ,— U.- ' I '¥ ‘h‘ - a I :N‘I‘». .‘ ->\\ ‘1'. A5. 10. It was from these surveys that the primary data for the present study are drawn. These data are discussed in the next section. Since 1961, a number of research agencies of the U. S. Department of Agriculture have done several studies centering around the Pilot Food Stamp Program. Research related to this program included surveys of sales in retail food stores and of attitudes and opinions of low-income families, members of the food industry, welfare workers, and others. There has also been a series of special studies designed to measure the economic and dietary effects of the program. 1/ The present study aims to contribute additional findings which can help public administrators and others who are concerned with programs devoted to the welfare of needy families. The Primary Data The surveys which form the primary basis for this study were not conducted specifically for the purpose to which they are employed here. Nevertheless, there is much to recommend these data for a preliminary study of low-income rural families. Our systematic knowledge of low-income rural dwellers is very meager at this stage, and we can profitably look at the information at hand as a pilot study before embarking on a more ambitious project. The data employed for the present analysis were collected in surveys of household living expenditures and consumption conducted by the Consumer and Food Economics Research Division, Agricultural Research Service, U. S. Department of Agriculture. The surveys provide information for studies of family living in low income rural areas of South Central Kentucky, East Texas, and the Northern Blackland area of Texas. They are based on samples l/Three reports evaluating the Pilot Food Stamp Program have been published by the Department of Agriculture: The Food Stamp Program, AMS-472, April 1962; Effect of the Pilot Food Stamp PrOgram on Retail Food Store Sales, AER-8, April 1962; and Food Consumption and Dietary Levels Under the Pilot Food StamprrOgram, AER-9, June 1962. hav~a~ .. . .WJM; u~ up... . r rnbvdv ‘y .1. '|. ,WA < can. s.cv..¢v. -, £7‘;:~__ ev' - '“‘ V- v... .9 .c‘, l a "‘ vu- b. “.3“ ‘ ‘ '1 A ,__ ._-_ I-'.. a - O‘J.‘ LA“; A. U. 'H . - . an ’ " 'I A,_ '.;»w'r‘s‘ - In“- I'V' J C\. ”" u‘: _ .. v A“..‘:, - ’ . \. U .A-u - 9 .“-~. I». “' y“ a.“ -\ -.. \ -. ‘fifio‘ \ .'uy 3‘ §’ ~ | , .._ v . >3, . ‘ M-“'» 4‘; fl ‘\ ‘Q- ‘ ~ J. y. ‘ " C N ”a ' _ r‘n. .‘ J.~‘>\- V‘— ‘C‘.-'-». s“ ~AA V"¢ n’n _ , K a" , \v I ‘V‘. V . . ‘V ~'\ V 1 v4- .. . .-,. | K‘. N -. a ‘1“: ’- n a. .- ’v ,V, § J- Ilrv‘ u" r “\H‘ 0, v. n- .. J :1 :~.'\ \ \ ., , . ‘U ~ " \' .— \, ‘ . A e. 03 ;.\'._r,,‘ \‘Ve..h .. _ . , ‘J ’./ '\ I‘LL“- «. , 5 nix, »‘ ., 1": V ’ ‘ ., 11. randomly chosen from a five county area in each of the three regions. About 350 open country inhabitants were interviewed in each region. The areas sampled represent appropriate populations for our purpose, since they have a high incidence of poverty within their boundaries. Furthermore, they may be considered as somewhat typical of two depressed areas of the United States. South Central Kentucky may be fairly similar to other rural areas in the poverty stricken Appalachian region. East Texas and the Texas Blackland, on the other hand, are closely associated with the old cotton producing areas of the South and may fairly represent this type of depressed region. There was a further assurance of high poverty concentration since the observations were drawn from the rural or "open country" portions of the counties that were surveyed. Rural areas contain a disproportionately large number of the Nation's poor -- 32 percent of the rural and 16 per- cent of the urban population had money incomes under $3,000 in 1959. The incidence of poverty is particularly high among those who are farmers or are connected with farming. In 1932, forty-five percent of farmers and fann managers earned money incomes of less than $3,000. Poverty was even greater among farm laborers and foremen, with 56 percent earning below $3,000. 1/ The study sample included only those families headed by both husband and wife. A further qualification was that the family had to have been in existence for at least one year at the time the interview was conducted. Data from the three sample areas are not strictly comparable. The family living data collected in Kentucky were for the 12 month period, September 1956 through August 1957. However, income data for the same study were for calendar 1956 and thus predate the rest of the data somewhat. The income figures for Kentucky are given as net before taxes. In contrast, l/Source: Economic Report of the President, Transmitted to Congress January 196A, p. 81, Tables 11, 12. ‘p—‘H __ -..-..1.- -- 3%-. V. ‘ -M...” » '\ H" ‘ v 0 has __. ‘ .._ -. “ *-..‘ - ‘9 n._ .-. :r“" 9‘ - ‘l ‘Mvu..-- - fir .- ‘9 ‘ V‘“. v. _ a “’ --_ m V. ~_‘_ -‘ v _-.;‘_*~“. ‘s.._‘.'~- "V. ‘_.'- ~_ K A ? c.‘ ... .. ‘ v-. “ HMJ“ ~ ~_ .— Q-‘ " ‘7‘ ~ .H: ._,‘v “‘ . \‘_ >. _ n ‘H. .— “-s\‘ ‘V a“ “.I. 4s. _ .. ‘ v v ‘F . “- '2 ‘r U .“_ s '2.“ ‘. I ‘0‘ r ._ U“ - l2. income figures for the two Texas studies are given as net after taxes. In further contrast to the Kentucky study, the income for the two Texas studies was for the same period as all other data in the same sample -- calendar 1958 for East Texas and calendar 1959 for the Blacklands. Kentucky income figures differ from those of the Texas studies in yet another respect. The farm component in the Kentucky income figures consists only of net cash farm receipts. The farm component of the income figures of the Texas households, on the other hand, reflects changes in farm inventory as well as cash receipts. The questionnaires used in the three surveys were designed to obtain all expenditures for family living for one year. An additional feature permitted the determination of a more comprehensive, more stable measure of the level of livi N. This measure, termed the value of consumption, includes, besides money expenditures on nondurables, the valuations placed on home produced food, fuel and clothing, as well as a year's depreciation on the family home, its furnishings and equipment, its stocks of clothing, and the family share of automobiles and trucks. These valuations were determined by professional home economists of the Family Economics Branch, Consumer and Food Economics Research Division, U. S. Department of Agriculture. The relative merit of the annual consumption value over annual expenditures, as a measure of the level of living, should show up especially in the present study of low income families. 1/ Within rural life the consumption of home produced (rather than purchased) goods and services, is most frequently a major component of family consumption in the rural low income sector. 1/ Data from the Kentucky and East Texas studies have been used in comparing the two measures of the level of living in an article by Mary Jane Ellis, "Value of Consumption: An Improved Measure of the Level of living." Family Economics Review, U. S. Department of Agriculture, June 19U1, pp. 3-8. __ n5 ._. .x ..v N ‘ .A. a. . . . . .2 3.. .. . _ ... .... : t .u p; .L 1; .Av . A“ 1» Av Lv L. .. Cu .h. — ‘ .4 .. .. a. .: B s. a... .... ... '1‘ l LII! l“! [11. .e i.“ o vk~ .\ . Q A] A: H. 5.; ‘II‘ L 1 ... . a Z .1 ..‘ a; ;. {u x I L. \,.~ n. x u .. A; a... h. .C r; 13. Design and Methodology The technique most commonly used to isolate the effects of given independent variables on a dependent variable, involves cross-classifying two or more independent variables and averaging the dependent variable for each cell defined. Ezekial and Fox describe this process as follows: ”Analysis by averages where there are two independent variables, involves classifying the records first by one variable, then breaking each of the two resulting groups into several smaller groups according to the values of the second variable. If a third independent variable were to be considered, these groups would be broken up into still smaller groups, according to the values of the third variable. Then the values of the dependent variable as well as each of the independent variables would be averaged for each sub— group. This process is known as subclassification or cross-classification."l/ The two authors regard cross-classifying and averaging as an intuitively obvious method to employ for analysis when large numbers of observations are available. 2/ Regarding this same technique of analysis, Daniel Suits says: "In general, the function treated as a table is the easiest to understand, and, where adequate data are available, is the most complete and sophisti- cated representation possible." 3/ The method of cross-classification and averaging has been used extensively in analyzing the three surveys which contain the primary data for the present study. The Family Economics Branch, Consumer and Food Economics Research Division, USDA, has used this technique to cross-classify certain basic variables with other family characteristics to determine the effect of these factors on family expenditures and consumption. Since a l/Mordecai Ezekial and Karl A. Fox, Methods of Correlation and Regression Analysis--Third Edition. New York, Wiley and Sons, Inc. pp. 389-390. g/ Ibid, p. 388. fi/ Daniel B. Suits, tatistics: An Introduction of Quantitative Economic Research, Chicago: Rand McNally and Co., 1963, p. 87. r- . ‘4 _.—..-‘ -v- ‘\ a. . i 1.. _H. a“ .n. .. n« k: r a... is a n .1 _ . Z . . . . . — . _ \ s u .‘L a e. r. .C . x . q . t; . . x x; f: L. . fie .rtu .. .n A.\\ s ‘1. 2\ 11+. fairly detailed breakdown of items of expenditure and consumption was employed, the data could be presented by only two or three cross-classifying variables. 1/ The procedure of holding, say, two variables constant while the effects of a third variable are being examined, is not an adequate analytical tool when it would be desirable to control the effects of 3 or more variables. With respect to consumer behavior there are usually several important factors that are closely correlated with the factor whose effects are being studied. Thus, if the studied factor is to be properly isolated, a method must be found that will hold the effects of the correlated factors constant. As we shall see in subsequent chapters, failure to control the effects of correlated characteristics may affect the estimates obtained to a marked extent. The main analytical technique employed in the present study is a single equation multiple regression. It is obvious that economic phenomena are affected by many factors and the fluctuations in a dependent variable may be due to the interaction of many forces. Food expenditures, for example, may vary with numbers and ages of family members, the childhood or ethnic background of parents, the price of food in stores, the kinds and values of home produced foods, and many more factors. It is essential, therefore, that a multivariate function be employed so that a number of different independent variables can be explicitly specified as influencing the dependent variable. But we can never list and measure all of the factors which influence the dependent variable. In addition to the factors mentioned above, there are literally hundreds of other variables which have a greater or lesser effect on food purchases. Of these, only a few can be specified in any l7IIn connection with these studies the Agricultural Research Service has published Home Economics Research Report No. 26: Rural Family Spending and Consumption in a Low-Income Area of Kentucky, by Jean L. Pennock. The same author is preparing a report based on the two Texas studies as well. A“ 3x....- -\ h" ~*--'- .-,. A. y..u_ y , ~ —,iA:-\. . \ _. -A . ~. ~_" . . «-‘-_ I ‘F' ”0., ---\, Iv ' _ - _ 5. . v _‘ 3 c ‘\ ~ ‘ \ ~‘._“ .-V' ,‘ ~-fl .‘»‘. ‘L' . .- ..._‘.\ n” ‘ 4‘ ., ‘ v H ‘A n. 'v " . \— u. ~ . .. , -. vL ‘ t ‘._ I ~ V v ‘. s. L ‘ - \.. ‘ u I. H, . .‘_ -‘ . . u ,‘ - ‘x , \T_. 3. ..-|, p - _~ .\_ “ “ ‘v .\ ‘\ \ \‘ " v . - ~- ‘5 fl \- ‘ a- 'v a \ P‘ . ‘. ..._ c~ \— t - . b \‘ U .Q .- a. _ ‘- ‘ ‘— *Ja‘. ‘ -. \_ ‘ ~ \ — ,a .N ‘ _ - ‘s ‘. V»“_‘ .3 ‘ ‘\ “. y —‘ ‘ i ‘0' . ., h \V ‘x ‘M 15. one function being considered. The dependent variable is, therefore, not an exact function of the designated independent variables, and is for that reason called a statistical or stochastic function. The specified variables comprise the systematic component of such a function and the non- specified variables make up the stochastic component. The statistical function employed takes the general form: Y = f (x1, X2, ... Xk, xk + 1, ... Xn) + u where Y is family expenditure or consumption, Kl to Xk consist of continuous independent variables, Xk + l to Xn represent one or more sets of discrete independent variables, and u represents the myriad of nonspecified, non- measured variables. The continuous variables, X1 to Xk, are also referred to as numerical, interval, or sealed variables, whereas the discrete sets Xk + l to Xn, are variously referred to as dummy or categorical variables. The first problem in specifying the function lies in the decision of which dependent variables one should specify. The answer to this problem depends largely on what it is that we really want to explain or study, e.g., are we interested in only food expenditures, or do we want to look at the reader total of family consumption? The next problem is in specifying the independent variables which on theoretical grounds or from previous empirical evidence are believed to affect the dependent variable. This process often ends up in a compromise between the measures that are thought to be pertinent and those which are actually measurable and available to be related to the dependent variable. A third important specification problem revolves around the designation of the prOper form that should be attributed to each variable, i.e., should data be presented in their original form or with a log, square root, or some other transformation; should the function be fitted linearly, linearly in logs, or curvilinearly; etc. The objective is to obtain an appropriate fit for each variable. -~ ~.-- \‘ In“- . ‘ ‘NAF,:_, _. chuwaA- ... . — . . ‘ H. ‘n‘ _‘ . . I a. ...» 'lA-A V v~ u . r"... 0-: ‘ r‘ - u "‘.-m~__. r‘f ., t I- ’ 3 x... . tau-d Vu-~ - .... VD . ' - r- . " 5.” V V' ‘ ' 5' - . - y 'f . II I Q - . H I O 0 3' ~ « " .":‘-\ ‘b 'N‘- I . ~ . ‘ a- n . _- ,_ ,. -‘ “..b A . “ru. , “ F¢- .‘4 N “'49., v . ~4.. ' . ‘F {1“ 7" .H ~~t .9 ‘ A ‘v . ‘ fih~ A . k I c: ,c : ‘ug ‘ . \‘ ~‘ ‘ . '~£.."“: ‘L ~“" g L“. -~ .A .- . . h"’\ .4 . 2's y“ m “"0 ”C . ‘.._ w - “-n h C‘n‘ . v~ . ‘ u‘ n n “. ' .:d¢“. . ,‘F “ I!" a ' V \ “n V‘ The regression analysis employed in this study is a stepwise procedure. _l/ Besides the complete multiple regression equation obtained at the end of a computer run, a number of intermediate equations are also available. These equations are obtained by adding one variable at a time and this results in the following intermediate equations: Y be + bixl Y b'0+ b'le+ 13'ng Y = bno + 1)"le + b"2X2 + b"3X3 The variable added is that one which makes the greatest improvement in the "goodness of fit." The coefficients represent the best values when the equation is fitted by the specific variables included in the equation. An important property of the stepwise procedure involves the facts that (a) a variable may be indicated to be significant in an early stage and thus enters the equation and (b) after several other variables are added to the I'e£7:.'3."ession equation, the initial variable may be indicated to be insignifi- cant - The insignificant variable will be removed from the equation before adding an additional variable. Therefore, only significant variables are 1“cluded in the final regression. Certain variables such as income and family size are singled out for Special consideration and study. These variables can be forced into the equa-tion and held there even if their coefficients are insignificant initially, or if they should lose their significance as other variables are added or removed. Therefore, the final multiple-regression equation :17 For a discussion of this procedure see M. A. Efroymson, "Multiple R r; eoresSion Analysis" in Mathematical Methods for Digital Computers, Anthony Ra o f 20§°t0n and Herbert S. Wilf, (EdsJ John Wiley and Sons, Inc. 1900; PPo 191- ("I 17. will contain all forced variables regardless of their significance, and any others of the specified variables which came into the equation and remained significant at a designated F test level. For those variables not forced into the regression but brought in as the stepwise technique selects them, the criterion or decision rule for selection is as follows: If the variance reduction obtained by adding a variable to the regression is significant at the specified F level, this variable is entered into the regression. I" .. I U / 'l , ‘w ‘- ‘¢ 4' A. \‘ ‘ h 9Q \ v CHAPTER II SPECIFYING THE MODEL The investigator of consumption behavior does not have a fully systematic or adequate body of knowledge to which he can turn for guidance. However, he may obtain guideposts for his own purposes if he examines the various theories and postualtes found not only in economics but also in the other behavioral sciences. Economists tend to take as their starting point the enabling conditions of income, assets, and credit which are available to the consumer house— hold. Their point of view received its beginnings in the economic conditions that existed over a century ago when nearly all income went into subsistence. The masses of people were so poor that demand for a good was thought to be limited only by the consumer‘s ability to buy, or otherwise gain command of the good. More recently, economistst have given recognition to the fact that social and psychological variables play an important role in economic matters. Demand is now considered a function not only of the ability to buy but also of the willingness by the consumer to make the purchase. In other words, it is now granted that consumers have considerable latitude in their actions, and many of their demands are influenced by their per- ceptions, attitudes, and expectations. Questions concerning consumer behavior are now being explored by an ever widening group of social scientists on the basis that consumer actions are motivated by many factors rather than one factor of overwhelming importance. They are attempting to determine the extent to which the consumer's actions are associated with rationality, and to what degree they may be responding merely as creatures of habit and inertia. They are also interested in the extent to which the consumer may be a product of his l9. sociocultural setting and of the group or groups to which he belongs. Various hypotheses are postulated concerning the role that should be assigned to such attributes as the physiological drive, the social mores, and to such facts that the consumer grew up in an affluent society. l/ The validity of many hypotheses can tentatively be tested by the employment of mathematical models of behavior. A behavior model, however, becomes researchable only when each of the explanatory variables can be represented by some objective measure. Thus, variables such as income, which ideally come in quantified continuous form, may plug into this model in an essentially unmodified form. Some other variables may be given numerical content by various scaling procedures. Although the resultant scales are usually quite arbitrary, they often do represent a useful further step in the direction of quantification. Variables can often be quantified and become part of an empirical model if the characteristic considered is presented as a set of mutually exclu- sive subclassifications. These subclassifications, which are termed categorical or dummy variables, are supplied with numerical values and can then be placed in a regression equation. The procedure for using the dummy variable technique is briefly but well outlined by Suits. g/ There is a tendency to restrict the explanatory variables to readily measured demographic and financial variables. Statistical groupings based on characteristics like age and income are not always the most pertinent classifications we could select, but in many cases at present they are the only variables readily available for statistical study of consumption, l/ Some notable cross disciplinary investigations have been made through work carried on at the Survey Research Center, University of Michigan. One product based partly on this work was a book by George Katona: The Powerful Consumer (New York: McGrew-Hill, 1960). g/ Daniel B. Suits, "Use of Dummy Variables in Regression Equation." Journal of the American Statistical Association. Vol. 52, pp. 548—551. December 1957. in..-'—~ rw-. LA‘- behavior. Many other family traits or characteristics that economic or social theory consider to be relevant to consumer behavior are so complex and often so subjective that they are difficult to observe or to measure. Even though considered important, some phenomena may not lend themselves to any adequate description. The Role of Proxy Variables Hopefully, though, the classifications we do use in our models can capture a number of subjective intervening variables that relate to con- sumption or expenditures. This is accomplished when certain of the specified variables act as a proxy for forces that are not directly specified and in many cases are not even observed. The proxy variable is expected to explain much the same portion of variance in consumption that would be explained by a direct measure of the unobserved variable. In no case would the proxy variable be able to capture exactly the theoretical construct in which we are interested. It is possible though, that a given variable can be proxy, in an imprecise way, to several differ- ent constructs. An example of such a proxy variable is illustrated by age classifications of the family homemaker. When homemakers are categorized into age groups, the main consumption effect we hope to capture usually relates to stages of the family life cycle. These same age classifications, however, may reflect consumption behavior that is not directly attributable to the life cycle into which they fall. Following are three examples of behavior that may be traced to other theoretical constructs, and yet they may be captured by life cycle classifications: (l) The age group that grew to maturity in the 1930's could have attitudes toward family expenditures which were uniformly different from the group which reached maturity in the post World War II period. These “...-... .. 5‘V-o«_~‘- . 1.” . —‘ -v I - i. _ -. ...- V‘-‘ \v ‘c. .- _. . - , h Q. N - ‘A ‘. Vv . “o I ~-. - 7‘- v -y “~H‘ "\ ‘~‘ s . v_~ a v~‘ ~.‘_ _‘ - .h -“- ‘v ‘ "‘v 'A‘ ..,. in ‘7 . ‘1 \ . ’2‘ ~ \ ‘ . .“ . \ . “§._' ... \. Q Ce ‘ u’ ‘ g‘ .. f .“ ‘- ‘- ‘ ‘\ .—fl b‘_ ‘v p»-\ x ’, - '- ‘— - \e \.: ~ \\ ‘.. \. ,‘. a ‘. r \' \ ’ V r- “ ’\ A 3'} \ ~ ! \ . \ \‘. ,1 s “- s \‘ ‘y ‘— ‘ _ \ - ‘<,-.‘. x ‘ V, 21. attitudinal differences are suggested by the fact that vastly different economic conditions prevailed in the two periods. (2) An older group could reflect many effects and a younger group very few of various influences that relate to the ethnic background of the older group -- even though they were of the same ethnic origin. (3) Different age groups may exhibit different degrees of adapability to a subsistence type of living, and this fact might be expected to reflect in their family spending behavior, e.g., an older generation may be able to live on less cash income, by virtue of the fact that they grew up in a situation that demanded a great degree of subsistence living and, there- fore, they learned to get along with less cash. Along a similar vein, if we classifv the family decision makers by their home ownership or their racial origin, we may also be delineating groups with similar attitudes and aspirations. Renters and sharecroppers who have practically no prospects for ownership, could be expected to regard the future somewhat differently from landowners or homeowners. Furthermore, there is apt to be a functional relationship between an individual's achievement motivation and his racial status. Kurt Lewin and his co-workers suggest that levels of aspiration are reality oriented in that they are only slightly higher or slightly lower than actual levels of accomplishment. l/ We know that, on an average, the accomplishment levels in our present society are quite different for whites and nonwhitus. They also vary with tenure status, especially when we compare sharecroppers with owners. l7Kurt Iewin; Tamara Dembo; Leon Festinger; and P. S. Sears, "Level of Aspiration." In J. McV. Hunt (Ed.), Personality and Behavior Disorders. (New York: The Ronald Press Company, l9hh). _— ~—'—‘-.’—-—-— ’— .— . A» .“._~ ‘ _ a» ._V_ ‘- . ‘ -,'_V ‘ ‘ _.~".- _. u.‘ .. - . "“'~-0 - ‘ ‘3‘"- -..-J‘__~‘ ‘ . _- O \‘ v -4. -. I-.." s“ ‘- ‘- . .fi . r.‘_" A ‘>-~_,.‘_- .N. [U D; O the Dependent Variable As pointed our in Chapter I, an important objective of this study is to examine the relationship between the family living level and the charac- teristics thought to be associated with the living level. To achieve this objective, it is necessary first of all to specify the dependent variable which is to represent the level of living. The current income is sometimes used as a measure of the family living level. When it relates to medium or high income people with a steady income, money income may be fairly adequate as a level of living index. But it can prove quite unsatisfactory when used to measure the well-being of low-income families, or persons who earn their livelihood mainly from farming or other endeavors where year to year income can vary greatly. Money incomes fluctuate a great deal if the household wage earners are employed only intermittently, or if they frequently work at '0 rather than at a regular line of worl. The amount of ass'stance that poor families receive from public or private sources also varies with fluctua- tions in the family's eligibility to receive help. Farm incomes exhibit considerable instability since they vary with the weather, prices received and paid, the type of farm program currently in effect, as well as other matters. Low-income farmers are also subject to various degrees of underemployment, varying with the amount of capital the family can bring to bear on their farming operations, and with the opportunities for off-farm employment. A better measure of the level of living is found in the amount of goods and services consumed over a given period of time, such as one year. The measure most commonly available for this purpose is the family's total outlay for the goods and services which it has purchased with private resources at its command. Public services received, such as education and 1. HI .Ia, ’3’.) (.Jo police protection, are difficult to evaluate and are generally not included in any level of living measure. There are of course other factors, such as proximity to good highways or schools, which are not easy to evaluate and therefore are not usually included in a scaled index of living. The three surveys from which we draw our primary data have provided for two different numerical measures of the living level: (I) The money expenditures made during the survey year for goods and services that are normally associated with family living, regardless of the amounts of such goods and services actually consumed by the family in the same year. (2) A.measure of the value of goods and services actually consumed in the survey year, regardless of the family outlay for such goods and services during the year. i The first of these measures will be referred to as "family expenditures;' the second as "value of family consumption." Family expenditures include the total outlay during the survey year for food that was consumed both at home and away from home. Included also, are expenditures on clothing, clothing services, costs of operating the household, repairs and replacements to the home, and home furnishings and equipment. Rent paid on the home was also included. Improvements and additions to the home were regarded as a capital investment, and none of it was charged to current living expenses. Other items included medical care, recreation, reading, education, personal care, and tobacco. The family share of automobile and truck operating costs was also included along with other transportation expenses. If an automobile had been purchased during the year, ten percent of its purchase price was added to family expenditures. The value of family consumption includes the valuations attached to consumption of durables as well as nondurable goods and services. Home produced food, and food received as pay or gift are assigned values then added to purchased food to arrive at a value of total food consumed. Home produced fuel is also given a value and this amount is added to cash out- lays in order to determine the total cost of household operations. The value of durables consumed was determined by assigning an annual "use-value" to each of the durable goods used by the family. To the extent that it appeared feasible, house furnishings and equipment were inventoried and a use-value assigned on the basis of original cost and age. The use- value of clothing was based on an estimate of the length of time garments are kept in inventory, and the average value of the garments. se-value of the family dwelling was made up of two components: (1) depreciation on the current value of the dwelling and (2) an allowance for maintenance, insurance, taxes, nd return on investment. Similarly, a depreciation rate was applied to cars and trucks and a proportion was determined as the family share of use-value of automobiles and trucks. A detailed description of the derivation of various consumption values can be found in U. S. Department of Agricultureh Home Economics Research Report No. 26. l/ Subclasses of the Dependent Variable It seemed appropriate to subdivide the total expenditure and total consumption into additional dependent variables for examination and study of additional hypotheses. For instance, the amount of food that is l/ Pennock, Jean L., Rural Family Spending and Consumption in a Low Income Area of Kentucky, Home Economics Research Report No. 2o, lQCh. Appendix B. ,— O T 2 purchased or consumed by the family appears to be a pertinent dependent variable to consider. Food is the single most important item in the family budget, and it tends to become a larger proportion of the total budget as the total purchasing power of the family becomes smaller. A combination of the basic items, food, clothing, and shelter are another appropriate subclassification. Since we are concerned primarily with poor people, we should investigate the extent to which our sample of low-income families appears to approach a subsistence type of living. We regard a subsistence economy as one in which most people devote all their income to food, clothing and shelter and where consurer discretion is therefore largely absent. The designation "more basic" xpenditures or consumption is used to identify this group of subsistence items. It might be -‘gucd that the "more basic" designation ougat to include some other items such as medical care. But we know that many poor people avail themselves of proper medical care only in the most dire of circum- stances. In fact, it is readily apparent that many rural people with limited means strive harder for such modern day "necessities" as refrigerators, automobiles, and TV sets, than they do for adequate medical services. l/ A final subclassification was designated "less basic" expenditures or consumption, since it consisted of the residual after the so-called.more basic items of food, clothing, and shelter had been subtracted out of the total. The three most important items included in this less basic designa- tion were transportation, medical care, and home furnishings and equipment. l7 The author is not able to cite any research in support of this thesis, but a number of observations from his experience would tend to support the proposition. A rationalization for the priority given these "necessity" items is not hard to find: peOple who have lived in rural areas can attest to the importance of both car and refrigerator to the rural dweller. And as Katona points out, op cit, p. 26, watching TV was one of the least expensive of leisure time activities in the post World War II period. [D 0\ Thus, there are eight separate dependent variables which will be examined in eight separate regression equations for every phase of the analysis that is undertaken. The variables are as follows: Y1 - Total family expenditure Y2 - Total family consumption Ya - Total food expenditure Yu - Total food consumption Y5 - More basic family »xpenditure Y6 - More basic family consumption Y7 - Less basic family expenditure I8 - Less basic family consumption The "more basic" designation includes all food, all clothing, and the costs of household use, operations and maintenance. The "less basic" designation makes up the remainder of all family expenditure or consumption for the survey year. §pecifying the Independent Variables We have varying reasons for specifying the particular independent variables that are selected to go into a statistical model. A first requirement is that the variable be numerical, scaled, categorical, or that it be convertible into one of these quantified forms. t is equally apparent that the variable selected must have relevance in theory or an explanatory value in the judgment of the researcher. Of course, in the case where secondary data are used the nature and limitations of the data as they were gathered often modify the form in which an independent variable is finally specified for the model. One has to contend with the limited number of observations that are available, bofli in total and within given subclassifications. Errors of observation fi—_—A—._— n~; ku..- "a... R2 0 must be considered along with sampling variability when selecting the factors best equipped to reduce the residual error. The way in which the scaled and dummy variables are presented, is also considerably dependent on the nature of the survey data. Vari bles for Preliminary Examination Family income usually ranks among the most important of explanatory variables in the study of consumption behavior. It is immediately apparent though, that total income is comprised of a number of components. Money or cash income comes from the earnings of land, labor, and capital and the remainder derives from various transfer payments. Nonmoney income usually comes in several forms of "income in kind." Information on total cash income was collected in the field surveys. The figure used for East Texas and Blackland Texas was money income after deduction of personal taxes. In the Kentucky study, information on taxes was not available and therefore the income figure used was "cash income before taxes." Among our designated "poor" people, however, the income before taxes figure should not run a great deal higher than income after taxes. For the purpose of preliminary examination, money income was studied as a continuous variable and fitted as a linear function in the regression equation. Among economically deprived people, especially those living in rural areas, cash income is apt to fluctuate a great deal. Crops and business returns can have severe ups and downs. The availability of jobs to family wage earners may be very intermittent. Outside aid coming from private and public sources may arrive only in sporadic fashion. Two items that can add stability to the income picture of rural low- income families are nonmoney income and the assets which are held by a 9 family. In a rural consumption study, the more important components of nonmoney income are usually home produced food and fuel plus imputed rents received from the family dwelling. A lesser item, home sewing, can cut back substantially the amount of cash outlay needed for clothing. Among the families interviewed in all three surveys, home produced food was the most significant item of nonmoney income. Furthermore, a large majority of families produced at least some of the food they con- sumed. Hence, the ratio "home produced food as a percentage of total food consumption" was utilized to indicate the effect of nonmoney income. Although home fuel production and home sewing were significant, only a small proportion of families engaged in these activities. The home produced food variable was subdivided into A classes and these were run as dummy variables in the regression equation. It was hypothesized that a linear specification might have produced a biased regression coefficient due to an expected curvature in the relationship-- hence the specification of dummy v “iables which could conform with any curvature that was present. Assets can also help stabilize the welfare status of families who are subject to large fluctuations in the incomes of the family earners. Earnings from interest, dividends, royalties, and real estate rentals all represent a fairly steady income flow. Assets which do not earn direct returns can, furthermore, serve as the basis for obtaining credit when liquid assets do not suffice. Unfortunately, it was not possible to include an asset variable in the regression model since the surveys did not include much information on family assets. The quality of homes in which 1 w-income families resided, was considered a measure that might be closely related with the living level in general. Since quite detailed housing characteristic data had been 30. collected, the author undertook to build an index intended to reflect the adequacy of the housing in which each family lived. A scaling procedure was used which rated 17 different aspects of housing, placing each on the plus or minus side of an arbitrarily selected average. The net index calculated for each home was independent of the tenure status of the home's occupant. A complete account of the technique used to determine the so- called "housing index" is found in Appendix B. It was expected that the housing index would fit into the regression model as a continuous linear function. Source of income was chosen as another variable that might relate importantly to the living level. Families were classified according to the source from which they received the major portion of their money incomes. The classifications used were: (1) salaries and wages, (2) pro- fessions and business, (3) interest, rent, and transfer payments, and (h) income from farming operations. Age of the homemaker was the characteristic selected to represent variations caused by stages in the family life cycle. Age could readily have been run as a continuous variable, but this was not considered the correct form in which to place this variable into the equation. Different bundles of goods and services apply to different stages of the life cycle because of events that conform to various age spans. Therefore, discrete classifications to conform with life cycle stages seemed more appropriate than a continuous mathematical function. The families were classified so as to represent three groups -- headed in turn, by a younger, a middle- aged, and an older homemaker. The age ranges for each of the categories were selected in part on the basis of the frequency distribution of the homemakers' ages in the surveys. 5.. u“, mg... ., -.~ ~‘ .... —. ‘.._ " . I I ‘p ‘ , --.. ’ . ‘ i “, I” 5“ v.‘ ufl.‘ ‘. , l ‘~ _- “A ~v _~ 4 , ~ , \_‘ -. ‘\ x. "i s A u y . a \ \ U x ‘ . ‘ l. U.) The life cycle stages were thought to be more closely associated with the wife's age than with the age of her husband. She marries at a con- siderably younger age than does her husband, especially in rural areas. And the standard deviation from the average age is smaller for the women than it is for men. In the Kentucky sample, which was the most rural of our surveys, the average age of the homemaker at her first marriage was 20.h years. The standard deviation around this average was 5.h years. The average husband was married at age 25.5 with a standard deviation of 7.9 years. Since the age at which the wife marries is more nearly predictable than the husband's, it is also easier to delineate the period of house- hold formation on the basis of the wife's age. Likewise, her age becomes a better guide to predicting the arrival and departure of children from the home. The shorter age range in which women can beget children, helps further to predict reliably the time when children are in the home, and their probable ages. The education of the homemaker is also o’ten considered a factor that has a bearing on the level at which a family spends or consumes. Increased education presumably provides the homemaker with better skills in managing the home and in making purchases for the family. The extent of her dependence on habit and inertia in management and buying, should also under- go certain modifications with greater education. Her tastes would change along with her attitudes toward such things as advertising and the intro- duction of new products. A better knowledge of nutrition could be expected to bring about some changes in food purchases and in home production of food. There are at least two reasons for considering the education of the wife in preference to that of her husband. With regard to family purchases 2. LU and consumption, the wife is assumed to be connected more closely with the various decision making activities. Furthermore, in the samples of families interviewed, the wives were not so highly concentrated in the lower education levels as were their husbands. Thus, they could provide more observations to test the effects of the middle and upper levels of education. Tenure status may influence the living level in a number of ways. Home owners can usually have more adequate housing, since the decision to make repairs or improvements to the home is their own and does not need to have the concurrence of a landlord. Owners of entire farms also have greater flexibility in the way they use their land or other farm resources. For instance, they may devote considerable resources to producing income in kind such as food or fuel. They do not have to answer to a landlord, should production of these products for family consumption cut signifi- cantly into commercial farm production. For similar reasons, farm owners are more free than renters to accept off-farm employment. Eage_was regarded as an important characteristic for a model on spending or consumption behavior. It is recognized that this variable has con- siderable intercorrelation with a number of other classifications that were specified. Nonwhites, for example, tend to be concentrated among the lower income, lower education groups. They are typically engaged in non- farm pursuits or in very small, uneconomic size farming efforts. The quality of their home is vastly lower than that of whites, on an average. But a number of differences between whites and nonwhites are, no doubt, attributable to different motives, attitudes, and expectations. The segregated conditions which have existed between the two races have probably produced a number of traits in each race that can be expected for the present to generate different consumption patterns. :a'w fill-8. :- u -. ' I . .av u, I— -.~,‘- \ . m. - 33- Analytic Procedure In studies that are dependent upon previously collected survey data, the variables specified are of necessity confined to factors which are measurable with the data collected in the survey. The variables we can afford, therefore, are often only partially effective in capturing the theoretical constructs we might wish to examine. The nature and extent of the data set limits on the analysis, but, within these limits, the analyst must seek out every opportunity to approach the pertinent classifications and the proper functional forms. There is a further need to direct the analysis in such a way as to locate the important interactions among the variables. Then a new set of classi- fications, including combination variables, can be chosen to represent the constructs which seem to underlie consumption or spending behavior. Steps in Building the Model The steps used in building the model consist of four successive phases: Phases I, II, and III are exploratory, aimed at discovering the more decisive factors which determine the family level and pattern of living; Phase IV analyzes and discusses a final regression model. Phase I considers three independent variables which, it was believed, should be analyzed separately from several other explanatory factors. Money income, family size, and housing adequacy were intended to serve in place of the three factors, income, family size, and assets, which are normally considered by case workers in establishing eligibility for family assistance. In the surveys used for this study, measures of income and family size were collected but no measure was obtained which could properly account for net assets controlled by the family. Instead of studying the effect of assets, the influence of the housing index (which was discussed 31+. earlier in this chapter) will be examined. It was hypothesized that housing adequacy may serve partially as a proxy for assets. But even if it should not capture the same effects as a direct measure of assets, the housing variable appeared to be an appropriate variable to consider in any case. Due to its importance in low income budge s, the adequacy of a family's housing may of itself, to a significant degree, reflect their need to participate in anti-poverty programs. In the initial analysis, the factors of money income, housing, and family size are run as independent variables against the eight previously described dependent variables. Since the investigator can find no contrary evidence to begin with, we start with the simplifying assumption that the effects of the three variables are linear and that they can be fitted as continuous variables. The initial analytic process can, therefore, give an indication of the effectiveness of money income, housing index, and family size as predictors of the living level. In addition, we are afforded the opportunity to start out with a model that is simple and manageable. After examining the eight regression equations of the first analysis, more freedom is permitted in the way the independent variables may Operate. If there are indications, for example, that the effects of a continuous variable arenot linear, transformations are permitted to reflect possible 'curvilinear effects. Alternatively, the use of dummy variables can be employed to conform with any curvature that is present. Tb accomplish the latter, the conventionally measured variable is partitioned into intervals, a set of dummy variables is defined in terms of the intervals, and the dummy variables then serve in place of the continuous variable. l/ Treating a numerical variable as a categorical variable is particularly desirable l/IThis method is suggested by Suits in his journal article; op cit, p. 551. 0,..- in... i p.” ...- ‘n '4 .. ‘V 35. if the researcher has doubts concerning the validity of the numerical scale that goes with the measured variable. In Phase II several new explanatory factors were added to the three already specified. This step recognized that additional variables should be taken out of the unspecified category and placed in the equation. The new variables were all specified as categorical variables because there was no a priori notion about the particular mathematical form which the true function would take. The method of regression analysis employed was the stepwise procedure discussed in Chapter I. After forcing the three Phase I variables to come into the regression, the categorical variables were then permitted to enter only if they could contribute to explained variance at a pre- designated F test level. Thus, for instance, if the category of families with older homemakers entered the regression it would be an indication that the families of this cateogry consumed at a level significantly different from the combined group of families headed by middle aged and younger homemakers. If the regression coefficient associated with the older homemaker group is negative, it indicates the older families consumed less than the average of the remaining families; if the coefficient is positive, it indicates the opposite effect. The combined younger and middle aged group is the "base" group with which the older group compares. Suppose now that after the older homemaker group entered the regression equation, there was still a significant difference (at the specified F level) in the consumption levels of the groups headed by middle aged and younger homemakers. Then in a subsequent sweep of the stepwise process we could expect either one of these two categories to enter into the regression. Let us say the middle aged group came in with a positive i‘v‘ , -, ha. .— . " _' ‘ -\ 5.- . -'r. .... . M‘s ‘-. ~ n ,. ‘a 4 .0. § A V, v q ‘- .-' s‘.“ .~_ '1‘ _A‘ ‘, ( coefficient; and let us say that the coefficient for the older group, although it would necessarily adjust, continued significant at the specified F level. The younger family group would now become the "base" category with which the other two categories compared. The negative coefficient associated with the older group indicates they consume at a lower level than the younger group; the positive coefficient with the middle aged group indicates they consume at a higher level than the younger group. The entry of a single category from a set of dummy variables indicates that this particular variable significantly increases the sum of squares explained. We know this is so by definition of the way the stepwise regression technique operates. However, lack of significance in any single class of a set of dummy variables reveals nothing about the possible signi- ficance that the set of classes would exhibit if taken jointly or as a group. The procedure for testing the null hypothesis that a subset of the coefficients in a regression are all (jointly) equal to zero is straight- forward, and.may be found in many standard references. In the present investigation, time and other resources did not permit the many computer runs that would have been required to test the significance of all pertinent subsets of coefficients. The objectives of the Phase II analysis are only exploratory, searching for the more decisive characteristics which influence consumption behavior. To accomplish this purpose, the stepwise technique was regarded as adequate to give an indication of the factors that should be considered in the final regression model to be tested in Phase IV of the analysis. The categorical variables that enter the Phase II regressions may be regarded as representing the more important effects which determine fluctuations in family living. 'v- bu.- A‘- ~..v *,A~ \‘vu ...“ d.-- . a w » u ~.. ‘~\. 37. The F level specified in Phase II is the 5 percent level throughout, and a number of subclassifications from several dummy variable sets will enter the regression at this level. When all the classifications have entered which can significantly contribute to variance reduction, a final (complete) multiple regression equation is printed out. In each of the three analysis chapters the complete equation for each of the 8 dependent variables is reported in a table. There are numerous intercorrelations among the independent variables specified. Hence, any test to see if a given factor contributes signifi- cantly to explained variance, is contingent upon what other factors have been considered, i.e., the apparent influence of any one independent variable depends on what other variables have been allowed for. Thus, for instance, the effect of money income on family spending is modified in the regression equation every time we add or subtract a variable that is correlated with money income. We may, for example, introduce a second independent variable "income in kind" and allow for its effects. This will cause the regression coefficient and standard error of the coefficient for money income to be altered, depending on the degree of correlation between the two types of income. The addition of a third correlated variable, say nonwhite consumers, would further mediate the influence of money income on spending. This process could be continued until all hidden correlated variables had been taken out of the unspecified category. In Phase III the analysis becomes even more complex when we begin to search for hidden interactions. l/ Tb expose interactions, we must not only specify the important interacting variables, but also discover how l/ Hidden interactions and correlations are discussed by Daniel B. Suits in his book -- Statistics: An Introduction to Quantitative Economic Research. Rand McNally & Company, 1963, pp. lll-ll5. A" bs ‘h . ‘ A ‘q ‘a 38. the impact of one variable depends on the values of the others. Students of consumer behavior already recognize a number of interaction effects, e.g., (l) substantial reductions in money income receipts are accompanied by large or small declines in consumption, depending on the amount of assets and income in kind at the family's disposal; (2) consumption increases with the size of the family, more or less depending on whether the family is well to do or has severe income limitations. Other examples could be cited. The stepwise regression technique is employed again in Phase III, and this time it selects from among several sets of dummy variables made up of interaction terms. They are parallel sets to those employed in Phase II because the interaction terms are combinations of the categorical and con- tinuous variables which were specified in Phase II. For example, if the continuous variable family size were designated X and three mutually exclusive homemaker age groups were Al, A3, and A3, then the interaction terms XAl, XAg, and XA3, would be one set of dummy variables. Suppose XA3 were selected under the decision rules to be brought into the regression equation. It would mean that the consumption function for the A3 group of families s10ped significantly differently from the families in A1 and A2. Note that when dealing with interaction terms we are looking for differences in the Elgpg of the consumption function 1 , whereas in Phase II we were looking for differences in the level of consumption. Hence, whereas in Phase II the purpose of the selection process was to identify family groups which consumed at different levels, in Phase III the purpose is to identify groups with a different pattern of consumption with respect to some (continuous) variable. l/ See Suits journal article, 0p cit, p. 550. 39- A more complete explanation of the way the stepwise selection technique seeks out interaction effects is presented in the Phase III discussion of Chapter III. The investigator needs to direct his analysis so as to discover hidden interactions and thereby achieve the greater insight that the interactions afford. We are assured by noted workers in this field that we can expect distorted results from analyses that assume away all interaction effects. l/ Morgan and Sonquist observe that: "Those of us who have looked closely at the nature of survey data, have been increasingly impressed with the importance of interaction effects and the useful way in which allowing for interactions between measured factors gets us closer to the effects of more basic theoretical constructs.“ g/ In Phase IV the final regression analysis attempts to incorporate the insights gained from the preliminary analyses. Evidence of existing interaction warns us of the need for something other than a simple additive regression equation. This problem is partially handled by the creation of complex variables which make it possible to build some of the interaction effects right into the new classifications. It is hoped that the new classes specified, will also represent a valid approximation to the under- lying theoretical structure of consumption behavior among low-income rural families. Chapter III follows the general procedure outlined in this section in analyzing the survey that was conducted in East Texas. Chapters IV and V are devoted to a similar analysis of the Texas Blackland and South Central Kentucky respectively. Chapter VI will make comparisons among the three areas and will attempt to arrive at those generalizations that appear likely on the basis of the analysis of these surveys. Chapter VII will draw conclusions. IZSee Suits book, op cit, p. 110. g/ James N. Morgan and John A. Sonquist, "Problems in the Analysis of Survey Data and a Proposal." Journal of the American Statistical Assn., D- .. r-n III- HA/f“ ‘.f\fl T. no. Most of the discussions and analyses that follow deal with families reporting less than $2,500 money income. These families were selected to represent the so-called "low-income" portion of the total sample in each survey. Whereever the discussion deals with the total sample rather than the subsample of families reporting under $2,500 income, it will be so indicated. CHAPTER III EAST TEXAS A General Description of the Area The five counties surveyed for this study were Anderson, Cherokee, Nacogdoches, Rusk, and Smith. They are a set of contiguous counties located approximately in the East Central portion of the State of Texas. (See Chart I, Appendix A) The U. S. Census of POpulation reveals that these five counties had a total population of 2l2,000 in 1960. Of those with native parentage, non- whites made up 28 percent. Males 25 years and older had a median education of 9.2 years l/; females 25 and older averaged 10.3 years. Nonwhites 25 years and older (male and feamle combined) averaged 7.7 years of schooling. Manufacturing is the chief industry group, employing about 18 percent of all employed persons in the five counties. Retail trade employs about 16 percent and agriculture ranks third with 10 percent. But among non- whites, the classification "personal services" was the largest industry group employing about 28 percent of all employed nonwhites, manufacturing employed 17 percent; agriculture, forestry, and fisheries employed 16 per- cent, and wholesale and retail trade employed 12 percent. The median annual income for the five counties was $3,886, but 27 percent of the families made less than $2,000. 2/ Among nonwhites, nearly 52 percent made less than $2,000 per year. About half the population was classified "rural" and 30 percent of the rural peOple were nonwhite. The median rural income was a little over $3,000 and 36 percent of all rural families had incomes under $2,000. l/The median for each county was multipliedtw'the number of males 25 and older in that county, and the 5 products were added to make one total. This total was then divided by total males 25 and older in all 5 counties to yield a weighted median. 2/ These multicounty income statistics are weighted medians constructed in the same manner as the education data above. l\\ M2. Families classified "rural farm" made up less than one-fifth of the total rural population. The median rural farm income was $2,570 and M2 percent had incomes below $2,000. Description of the Survey Sample The total sample consisted of 350 randomly selected families who lived in the open country portion of the five counties at the time the survey was conducted in the spring of 1959. Their net incomes after taxes, in 1958, ranged from a negative $h,300 to nearly $20,000. One hundred eighty-nine of these families had incomes below $2,500, and for the purpose of this study they are labeled "low-income" or "poor" families. It should be recognized that an across-the-board cutoff point will overemphasize the number of small families and underestimate the number of large families that are classified as impoverished. But in order to avoid a complicated classification process, a concept of absolute poverty was adoptec.nere. The cutoff level of $2,500 annual money income is the same level used by Robert Lampman for a family of four in his report to the Joint Economic Committee, Congress of the United States in 1959. _1/ When all families in the sample were classified according to the chief source of their income, wages and salaries accounted for 5h percent; professions and business 8 percent; interest, rent, and transfer payments made up 23 percent; returns from farming 15 percent. Roughly 22 percent of the families did not produce any of the food they consumed, and an additional 16 percent produced less than one-tenth of their food supply. Nineteen percent of the families produced more than hadi‘the total food they consumed. \ 17 Study Paper No. 12 -- The Low Income Population and Economic Growth. U3 S. Government Printing Office, Washington, D. C., 1959. \ a; A» ., I ...... :3s 1+3. Nonwhites comprised 22 percent of the sample families, and they all produced at least some of the food consumed in their households. The bulk of nonwhite households fell in the class "10 to 29.9 percent of home pro- duced food." In the subsample of low-income families, wages, salaries, professions, and business provided the chief income for about #0 percent. Another #0 percent of the families received their main income from transfer payments, interest, and rent. The remaining 20 percent obtained most of their income from farming. Fourteen percent of the low-income families produced no food of their own, and an additional 11 percent produced less than one-tenth of the total amount of food they consumed. Twenty-six percent produced.more than half their food. The proportion of nonwhites rose to nearly one-third of the 189 low-income families. The Analysis Phase I -- Money Income, Housing Index, and Family Size as Explanatory Factors The first set of regressions were intended to test the effectiveness Of money income, housing index, and family size as predictors of the family living level. We have a Special interest in these three household traits Since they could very well serve as qualifying characteristics when (Nartifying needy families for public aid programs. The object is to find OLFt how closely the three factors correlate with the spending or consump- tiJDIl levels of the low-income families in this rural sample. Table 3.1 gives results of regressions for eight dependent variables on‘ lilflcome, housing, and family size. Each independent variable was run as a"1¥ifitnear continuous function and there was found to be little intercorrelation &MC>r1{3. the three variables. The simple correlation coefficient between money an. income and housing index was -0.11; between money income and family size, 0.06; between housing index and family size, -0.28. It is notable that the regression coefficient for money income was insignificant, at the 5 percent level, in all cases except equation 3 where food expenditure was the dependent variable. Normally one would expect the money income variable to show high significance in all equations, because income is such an important enabling condition for a family's purchases or consumption. Failure of the income variable to Show significance in the Table 3.1 regression, can be interpreted as an indication that money income has little effect on a low-income rural family's living level. Another possibility is that the money income function had so much curvature that it was not amenable to study by linear approximation. Table 3.1-~Computed t values in regressions of the living level on three household characteristics, low-income rural families, East Texas, 1958 Independent variables Dependent variables 3 ggzgge : Hihdigg : Fgmiiy R2 (1) Total expenditure ...........: 1.672 : 6.653 : 5.u01 : .235 (2) Total consumption ...........: -0.5iu : 10.u33 : 7.356 : .uie (3) Food expenditure ............: 2.363 : b.287 : 7.005 2 .237 (h) Food consumption ............: -0.567 : 5.559 : 7.289 : .258 (5) More basic expenditure ......; 1.555 ; 8.222 ; 6.9h0 ; .266 (6) More basic consumption ......2 -0.9h5 ; 11.2u2 ; 9.750 ; .h36 (7) Less basic expenditure ......; 1.252 ; h.9h9 ; 2.151 ; .109 (8) Less basic consumption ......; 0.081 ; 0.086 2 2.522 ; .159 Critical t value at 5 percent level ==l.973 Critical t value at 1 percent level ==2.600 In contrast to the insignificant effects of money income, the Coefficients for the housing adequacy index were highly significant in L— .14 every equation. The computed "t" values (t 2 b/sb), pertaining to the housing index, ranged from a low of 4.29 in equation 3 to 11.2h in equation 6. The critical t value at the 5 percent level was 1.973; at the 1 percent level it was 2.600. The family size variable was also significant in all eight equations if the 5 percent test level is employed. But at the 1 percent level it failed to show significance in the last two equations. Apparently the effects of family size are not so significant when they relate to less basic expenditures or consumption. The effects of the three independent variables were able to explain only about 12 percent of the variation in less basic expenditure. But in equation 6, the coefficient of multiple determination, R2 = 0.h9h, indicates that nearly 50 percent of the total variation in more basic consumption depends upon the effects of the three variables. The R2 is largest in equation 6; both housing index and family size have their greatest influence in this equation. 1/ An important reason for the housing index coming in so strongly in equation 6 is the high correlation which exists between the index on one hand and the shelter component of more basic consumption on the other. The explanation for the high correlation is found in the way the two factors l/ Throughout this study, the EB (adjusted coefficient of multiple determination) is presented in the tables rather than the unadjusted R2. Thus, when the amount of variance explained by one regression is compared with the amount explained by another, the R2 compensates for any differences in degrees of freedom, i.e., the adjustment in R2 compensates for the bias resulting from unequal degrees of freedom in the equations compared. There is an exact functional relationship between R2 and R2 since N - 1 .~ R2.i- [(i—R2)WTI:I were N = number of observations and P 3 number of independent variables. were calculated -- essentially the same aspects were considered in determining both the value of the shelter component and also the housing index. It is important that we not attach too much significance to the high correlation between housing index and shelter cost. The value of the shelter consumption, after all, comprises only about one-eighth the value of total family consumption. The value of shelter consumption, further- more, is included in only two of the eight dependent variables -- equations 2 and 6. Further Investigation of the Money Income Variable In order to learn more about the money income variable, an analysis parallel to that of Table 3.1 was run on the total sample of 350 families. In this run, the money income had a highly significant effect (significant at the 1 percent level) on all the dependent variables except food consump- tion. Computed t values were as high as 9.h7 in equation 1. The comparative significance of money income in the total sample and its lack of significance in the subsample of poor families, as well as the signs on the coefficients, suggested that a best fitting income function might be roughly U-shaped rather than linear. This estimate of the situation was largely confirmed when still another set of regressions was run on the 123 families which had reported less than $1,500 income. The coefficients for money income were coming in negative in all the equations, and were in fact significant at the 1 percent level in the four equations predicting the consumption levels -- equations 2, h, 6, and 8. It was hard to find a logical explanation that could account for a negative slope of consumption on income. Therefore, further investigation was undertaken to discover the peculiarities which pertained to this par- ticular situation. MY. A theory has been propounded for some time now that consumption is more nearly a function of long-term average income than of the current income. 1/ The author believes that this theory applies particularly well to the case of farmers. Many of them are plagued by large fluctuations in their year to year incomes due to vagaries of the weather and other causes. A household by household investigation of the East Texas survey showed that many of the very low incomes could be traced to farming operations which had turned out poorly for the year that was reported, 1958. There was additional evidence that the larger of the negative incomes tended to be associated with some rather sizeable farming operations. It appeared, in fact, that a number of these farm families did not properly belong with a classification of "poor" families at all. Adverse conditions had put them in this class in 1958, but their long time average could, and probably would, place them with a different group entirely. Here then, is a possible and plausible explanation for the comparatively high consumption levels reported by families with large deficits in their 1958 farming returns. Presumably, they were consuming according to a long- term income potential, and in several cases it was considerably above a so-called low-income level. Iearly all families with incomes approximately in the range of $1,000 to $1,500 were mainly older couples and nonwhite families whose long-term income potential was not very different from the incomes they did report in 1958. Their consumption levels, therefore, reflected more nearly an income potential that one would associate with the genuinely poor. 1/ Milton Friedman's permanent income hypothesis is among the better known of hypotheses along this vein. In his book, A Theory of the Consump- tion Function, Princeton University Press, 1957, pp. 21-22, he describes a permanent component of income as analogous to the "expected" value of a probability distribution. He also describes a transitory component to be interpreted as all "other" factors which the affected unit would treat some- what as "accidental" or "chance" occurrence. #8. In order to get an overall picture of the way money income was related to consumption in East Texas, a scatter diagram (Figure 3.1) was drawn. Each dot in the chart depicts the total family consumption associated with a particular income level. The household which had nearly $20,000 income was omitted entirely, leaving an income range from -fh,300 to $9,2h0 among the remaining 3h9 observations. In an effort to delineate groups with somewhat similar characteristics, four classifications of income groups were defined as follows: (1) incomes below $u00, (2) suoo - $1,u99, (3) $1,500 - $2,u99, and (4) $2,500 and above. A simple least squares regression line was fitted to each group. An inspection of Figure 3.1 reveals a sharply negative slope in the regression line for the lowest income category, a slightly positive slope in the next two classes, and a fairly significant upward slope in the class $2,500 and above. It is apparent from Figure 3.1 that if money income were to be treated as a continuous variable, a quadratic rather than a linear function could better capture the curvature that is present. A parabolic function, for instance, would represent a relatively good fit, but this is not the type of function we would expect to fit the more typical case. In a study dealing with rural families that are truly poor people, there may be little or no empirical support for a parabola type function. Certainly, a sample of strictly poor families would not contain any observations like the larger farmers that contributed so greatly to the negatively sloping portion of the regression in Figure 3.1. Since we do not have a good logical basis for selecting a particular quadratic function, the use of dummy variables represents a very flexible alternative approach for analyzing our income-consumption relationship. This method tends to avoid the biases which result when an incorrect Total ‘ Fun 1: Co as umption (D 0115.:- 5) 80. m 08. 4 coo. w 08. a A.3H “V g 0 H 0 : h—Ha‘ 0! h coo. w 08. w 08. a 80. m- 80. :. __ — ‘1 # COO _ _ : ho In 0 h Gave :59 a In ondkdnlo “in O .IIO owv on cash H: «II I. a M 90 no : HI» s C i“ «no «I‘M 0:0 05-5 0 haullh . .Oldd 0% H.n Oksla ~ 50. mathematical function is fitted to a set of statistical data. _1/ The regression coefficient associated with a dummy variable merely reflects the level at which the respective income class is consuming; it does not have reference to the regression slope of consumption on income for that particular class. The level of consumption can therefore be compared among the several income categories, without regard to the mathematical function that would fit best if continuous income, instead of the larger intervals, were used. It is necessary when dealing with a set of dummy variables to omit one of the subclassifications from the regression equation. The emission of one Class prevents the system from becoming overdetermined, and thereby ensures the proper conditions for a determinate estimate of each parameter. The Omitted class becomes the base with which other "specified" classes in the Same set are compared. Each estimated coefficient represents the difference in the living level between the omitted class, and the class with which the Coefficient is associated. The results from a new set of regressions are given in Table 3.2. In this analysis, money income was run with the employment of dummy variables in the manner described above. The variable selected for omission was the Class with incomes below $h00. The housing index was again run in linear Ciontinuous form as had been done previously. The family size, this time, was entered with a log transformation rather than as actual family size. The log transformation produced a curvilinear regression which has consumption increasing at a decreasing rate as the household size goes up. Theoretically, this type of function should provide a reasonably good fit in an analysis of low-income families. When there are inadequate resources to Clevote to family living, as is often the case with poor families, the di lrSee Suits' journal article, 0p cit, p. 557, for a more complete 8(mission in the use of dummy variables to capture curvilinear effects. 51. Ukable 3.2--Computed t values in regressions of the level of living on certain household characteristics, low-income rural families, East Texas, 1958 Independent variables Continuous : Ihurmr 1 variables : variables : Dependent : . : Log : Income class a? : fig variable :Hous1ng. U . $300 to 83$1,500 to . a °family ° . index , ‘L . .4. . (7., ° ° size ' Wl)h99 ' W3)u99 (1) Total expenditure........: 7.11M : 6.69M : 0.072 : 2.251 2 .332 (2) Tbtal consumption .......:11.23 i 8.369 : -l.h33 : 0.748 : .502 (3) Food expenditure ........: 4.5U3 : 7.7%7 : l.h&9 : 2.911 : .298 (4) Food consumption ........: 5.76% i 7.870 : -l.622 : -0.0M0 : .329 (5) More basic expenditure ..: 6.583 i 7.883 : 0.009 : 1.927 : .3u9 (6) More basic consumption ..:l2.093 :10.929 2 -2.178 : -0.h53 : .503 (7) Less basic expenditure ..: h.382 : 3.267 ; 0.135 : 1.870 E .220 (8) less basic consumption ..; 6.329 2 2.9uu ; -0.222 ; 1.580 2 .207 g/ The omitted class was "families with incomes below $h00." t.05 I 1.973 t.Ol a 2.600 IXer capita consumption is forced to decrease with an increase in the size (If household. Furthermore, it is reasonable to expect that some economies CI? scale would occur in household operations as new family members are arided. The costs of shelter and household utilities ought to rise by less tfllan a constant proportion with increased family size. Some scale economies STMDuld also appear in the process of preparing food for larger families. CVNE researcher found the per person cost for food to be about 20 percent 11338 in a 6 person family than in a 2 person family with a given standard or (iietary adequacy, and where the effects of income and family composition haéi been allowed for. ;/ h 1] Janet Murray, "Per Person Food Cost Differentials in Large and onmllml Families," U. S. Department of Agriculture, Family Economics Review, Sept. 1960), pp. 3-11. [- r.‘ 2.2;. Thus, it appears reasonable to try fitting a cruve whose slope declines with increased size of household. The general form for such a function can 'bc represented by a semilogarithmic equation Y s a + blogX, where Y ,reprcsents the living level and X represents family size. The results indicated that log family size did give a better fit than ‘was obtained with the original data. The new form contributed substantially 'to the larger coefficients of multiple determination found in Table 3.2, o i . . . compared with respectlvc Rr's 01 Table 3.1. The hou51ng index also per- formed considerably better when run in conjunction with log family size and Classifications of income (Table 3.2), than when it was combined with linear :forms of income and family size as reported in Table 3.1. The computed t values associated with the income classes showed that tile lowest income families (omitted class) were relatively high consumers orl an average. The reason for this perverse relationship has already been Staggested: a number of the negative incomes were reported by fairly large sxzade farmers. The t values in several equations indicated that a number Of' the dummy variable consumption levels were not significantly different (5ft the 5 percent level) from the level of the omitted class. But in the PITESent case where we appear to have a U-shapcd step function, a test of 3r€x1ter interest may be the one based on the null hypothesis: there is no .L Sié§rrificant difference in the level of consumption between the $h00 - $l,h99 and £551,500 - $2,199 classes. y If this null hypothesis can be stated as Em. _. 133 _ 0, then the statistic t: b)!__b SCJ C33 - 20311 chh U.) ha 8 the t distribution with N-5 degrees of freedom; where b1“ b3 are e813- nrnneites of Bu, B ; C33, Chh) and C3p are corresponding elements of the ’3 J 3L7 This test was suggested by Dr. R. L. Gustafson, Department of . Age: _ C=‘LJ.l_tura1 Economics, hichigan State University. ~ 53- inve *se moment matrix (properly scaled, of course, so that so Cii is the standard error of the bi); and Se is the regression "standard error of estimate." Notice that with bit.) 0 and b3 < O, bh - b3 is larger than either of the individual coefficients. then, for instance, this significance test was applied in the ease of total consumption (equation 2), the computed t value for the null hypothesis Eh - B3 = O was 2.0915 and this is significant at the 5 percent level. Ordinary t tests of the separate null hypotheses Bl, . O and B3 = 0 produced t values of 0.748 and -l.lt33 respectively, indicatiw that neither Bh nor B3 were significantly different from zero. This demonstrates that it is quite possible for the difference between two income classes to test out significantly even though the individual coef- ficients do not test out significantly different from zero. Although the dummy variable specification for money income tests out somewhat better than the linear continuous form, the tentative conclusion is that little correlation exists between the money income variable and the living level for low income rural families. Only among families reporting more than $2,500 income per year, does money income demonstrate considerable Predictive power. The error of observation in reported money incomes also contributes to low Predictive power. It is the author‘s impression from his own experience that family incomes are among the least accurately reported of the usual kind-S of data that are collected in a typical sample survey. While there will be a few overstatements of income, the understatements are certainly more - ‘ Ctommon. And, whereas some of the understatements are due to modesty or fa'LIILty memory, the majority can be attributed to conscious or inadvertent C011 cealment on the part of the respondent. Another difficulty arises when 511*. the homemaker rather than the family head is interviewed. The homemaker often does not know the correct information relative to family income. _l_/ It seems a prudent course, therefore, to exercise caution in evaluating results of analyses relating to family incomes where these incomes have been reported through sample surveys. A parallel analysis to that of Table 3.2 was also run on the total sample. In this run, the log family size variable did not show an improved fit over that which actual family size had demonstrated. It should be recalled that in the subsample with incomes under $2,500, the transformed variable had shown a substantial improvement over original data. The reason for this difference can probably be attributed to different levels Of resources available for family living. The large families of the top income class, usually had sufficient means to maintain a more or less stable per capita living level. But the large families among the poor, were often forced by their limited buying power to follow a declining Per capita living level. larger families in the top income class were especially better fed and better housed than those from the three lower income classes. In the Class with incomes above $2,500 and with 6 or more members per household, tne average value of food consumed was $1,880; the average value placed on their shelter consumption was $19.2. The 27 families in this class averaged 7 ° 2 members . \ 1 In making these observations concerning the accuracy of incomes :jported in sample surveys, the author is influenced primarily by his own Hoser‘iences as an interviewer and also as a supervisor of interviewers. ta; e‘Jer, other authors have referred to this problem especially as it per- Stagtcg to the underreporting of incomes. See, for instance: Bureau of Labor 10971 stics, Family Income, Ikpenditures, and Savings in 1950, Bulletin No. Eva; (Revised) pp. 7-8; Office of Statistical Standards, Statistical Ntion Report No. 5, December 1961+, pp. 7-lO; Department of Agriculture, % to the 38th Annual Outlook Conference given by Laura Mae Webb, pp. l—2. g In the class with incones below $2,500 and with a or more members in the household, the average value of food consumed was $1,5H2, and the average value placed on their shelter consumption was only $177. The 18 families in this group averaged 8.l members. Phase II -- Additional Explanatory Factors Specified in the Regressions This phase of the analysis permits several additional independent variables to be included in the multivariate regressions. Taking extra 'variables into explicit account removes them from the unspecified category, thereby reducing the residual variance and increasing the variance explained by the regression effects. This process also reduces the biases xfhich are always present so long as any variables correlated with both the dependent variable and at least one of the independent variables, remain \Jithin the unspecified group. In addition to the three independent variables considered in the IPhase I analysis, the following characteristics and their subclassifications ‘Were specified for a new set of regressions: 1. Chief source of income (a) Wages and salaries (b) Professions or business (c) Transfer payments, rent, and interest (d) Farm sources 7 2. Age of homemaker (a) Under #0 (b) no to 5A (c) 55 and older 3. Education of the homemaker (a) less than 8th grade (b) Grades 8 through ll (c) Grade 12 and higher 1. 52111.- . _ 53. h. Tenure of home by farm status (a) Owner farm (b) Renter farm (c) Owner nonfarm (d) Renter nonfarm 5. Race (a) White (b) Nonwhite The reason for selecting these particular characteristics and these Ewirticular subclasses were discussed in Chapter II. Dummy variables were defined on each subclass named under the above fuousehold characteristics, and then placed in a regression along with HKDney income, housing index, and family size. The latter three variables there again entered in linear continuous form for the first set of analyses Hinder Phase II. Table 3.3 gives computed t values f r all the variables that were £3ZPEI<2ified and brought into the equations. Since the three continuous Vtxrtiables were forced and held in the equation, there is a t reading on GEUEEI of them regar ess of the influence they had on the dependent ‘VaJ?fiLable. The categorical variables, however, were brought in (one at a tllnfi3) by a stepwise regression procedure, and only those variables entered t O o o o o o o 1 116’ <3quation which represented a Significant improvement in the gooancss O - o f‘ i?lft of the regress10:. IIf a classification variable enters the regression, it means that COno . .. . . . .. . . . .. onJiuptlon of families in this particular class1fication ls Signilicantly dif‘ I?€31?ent from families in other classifications (of the same dummy Val?‘ J‘Q-‘ble set) which did not enter the regression equation. " '. '1‘. . Total ciO ;1. em“ Less . can . .o L Table 3.3--Computed 1; values in regressions of family levels of living on household characteristics, low-income rural families, East Texas, 1958 : Independent variables : Continuous variables : Dummy variables : : z : : Age of : Education of : Tenure of home by : : ‘ ChiEf source 0f income : homemaker : homemaker : farm-nonfam status : Dependent 21401183? : Housing : Family :- ages : Profes- :Transfer : : . : 21,955 =Grades : : . : : : : . variables :income: index : size ; or Isions ipayments,=F :Under: 1+0 :55 :than : 8 :Grade :Owner :Renter: Owner :Renter : : : z ‘ ‘ ‘sala-= or =rent, or 3 arm: “0 = t° =and =8th =thru =12 and :farm : farm : non- :non- :White: Nonwhite: ‘ ' 1 ries =business=interest : : : 51+ :over:gmde: 11 :above : : : farm :farm ; 3 : We 5 f i i E = . f s ' ' : 3 3 = f 3 3 3 : expenditure I-o.i76 1 6.701 1 n.289 1 n.27l; : : : : I : : h.l6i : : I : : : .3u8 Total = : : ‘ : : : : : : : : : : z ; : consumption 2-0363 2 10.253 2 6.097 I I : -4-016 I I I I I I I 3'97“ 2 I I I Z I I A93 Food ' : : : : : : : : . : : : expenditure ; 0.707 1 3.297 : 7.154 : 3.3a; I : 2 I : I 2 £198: ‘. 1 I 1 2-366 : : “8-82“ I 321+ Food 2 : : : : : : : : : : : : g consumption 2 0.379 I 3.606 2 7.213 : : -2366 : 1 : I I : 2 2 I I —3J+80: I I -2.613 : 3%. ”fiber : f i i f f s 5 3 i i 3 3 _: 3 5 3 3 3 3 exPenditure 1 1.753 1 5.721 1 5.735 1 1 2 ~3-345 I I I I ' I I “.050 I I I I I I I -355 More basic 3 f 3 3 f f 3 f f 3 f f 3 3 5 f f f f ‘. consumption ;-0.l39 : 10.907 2 8.686 I : : -3.]_85 : I : I : 2.798 _3.059: : I .5’48 Less basic : : z : exrenditure 1-0.lou 1 5.010 1 1.09% 1 3.9501 1 1 1 1 1 1 1' 1 2-670 1 I I . I I I -195 Less basic 1 consumption ;-1.1L1+2 I 6.121 1.292 I #258: z 1 : : 2 I : I 3,900 - I : I I I : .275 t105 = 1.973 t.01 = 2.600 58 . The classifications for which there is a t reading given in Table 3.3 were therefore the only categorical variables which were a part of the final multivariate regression reported in the table. The 5 percent signi- ficance level was the critical level designated both for entering variables, and also for removing any variables which had dropped below significance as other variables were entered. The entry of these particular categorical variables caused a downward adjustment in the t values for the three continuous variables (Table 3.3 cf. Table 3.1). The adjustments in the t values occurred because the inter- correlations among the continuous and new categorical variables caused adjustments in the regression coefficients and also in the standard errors of these coefficients. The new parameter estimates retained fewer biases which were there because of the (previously hidden) correlations that were brought out by the more inclusively specified regressions. When all the adjustments had been reflected in the final equations, the money income variable showed less than a significant effect on any of the 8 dependent variables. The influence of housing index and family size continued highly significant, however. Only in its effects on the less basic items of living, was family size relatively insignificant as an explanatory factor. In each of the 8 equations at least one of the classifications delineating the chief sources of family income entered the multivariate regression. Families whose chief income was from wages and salaries were identified with a higher living level than the remainder of low-income families. This was true in four of the equations. As a counterpart in the remaining four equations, the recipients of transfer payments, rent, and interest were identified with a lower level of living than was enjoyed by the rest of the families. I“ [at ‘.._‘.,‘. V‘-‘ . ~ -..- v a... 14‘0-4‘3‘al. A {Al 59. None of the "age of homemaker" categories entered any of the equations at the 5 percent significance level. But there was a relatively high correlation between families with an older homemaker, and families whose chief income was from transfer payments, rent, and interest. This "source of income" category, therefore, contributed to an attenuation of the effect of the "age of homemaker" category and vice versa. The result is that only one of the two classifications comes in as a significant variable. When the same Phase II variables were run using the total sample, an age of homemaker class entered 5 of the 8 equations. In each of the 5 instances, a chief source of income classification had also entered and both remained as significant variables in the final equation. It is PrObable, therefore, that if chief source of income were omitted entirely as an explanatory factor, the category "homemaker 55 and older" would come into all the equations as a significant variable.. The homemaker‘s education was another factor that contributed SubStantially to explaining variability in the level of living. Families Whose homemaker had a grade l2 or higher education spent and consumed at a Substantially higher level than those headed by lesser educated home- makers. Only when predicting total food consumption did none of the education categories enter at the 5 percent probability level. Grades 8 through ll homemakers appear to have spent significantly less on food than their better and lesser educated counterparts. The author was not able to find a suitable explanation for this. Neither was there a satisfactory explanation for the "owners nonfarm" to be consuming food and other more basic items at a lower level than families in the other Cla SSifications. Chance may have been the main factor in producing these be“ Qults, but the difficulty in explaining these particular outcomes 814%- . . . E:ested the need for specifying an additional variable-«one that would ¥ A ~'~ VV . k- ‘_..i u‘v 60. help account for more of the variations that relate to food purchases and consmnption. The proposed new variable, a measure of home produced food, is discussed more fully in Phase III of this chapte where it is first specified in the regressions. The tenure and farm status classification did not demonstrate an identifiable pattern of behavior when only Table 3.3 results were considered. But with the help of two other separate analyses (one with the total sample; another with families whose incomes were {MOO to EMA-99) a pattern did seem to emerge. The pattern that emerged was that farmers spend less on the more basic living items, and also less on total expenditures, but they tend to consume more food than do nonfarmers. These results lend support for the practice employed above, which Combines home tenure with farm-nonfarm status, when both these traits are Specified in the regression. If instead of the joint variable we were to Specify owners vs. renters only, we could expect distortions produced when an interaction effect caused the regression outcome to be dependent on the farm-nonfarm makeup of the sample. An increased proportion of farmers Would lower the per family amount spent on several items of living and raise the amount of food consumed. Increasing the proportion of nonfarmers WCuld have the Opposite effect. The difference in the living levels between the owner class and the renter class could therefore be great or Small, dependim on the proportion of farmers to nonfarmers that comprise each class. Hidden interactions are discussed and analyzed more inlly in the next phase of this chapter. In that part of the analysis another interaction Was . . . . . cllscovered involVing owners and renters, but related this time to the housing index. Race is the last variable that was specified in the regressions re ported in Table 3.3. Nonwhites differ significantly from whites only / 01. with respect to the food variable, i.e., nonwhites purchased less food and they also consumed less food. But in a similar analysis of the total sample, nonwhites were shown to purchase and consume not only less food, but less of the more basic items generally. A more detailed investigation indicated, however, that it was only the food and shelter items (rather than clothing) of which the white families used substantially more. Tne clothing purchases by nonwhites, in fact, actually exceeded those of the whites . _The $1+OO - $1,169 Income Class Analyzed When the 96 families with incomes $400 to $l,h99 were analyzed as a separate group, even the food variable no longer showed a significant difference between the races. The reason appeared to be that the whites of this small subsample were mostly older couples who like the nonwhites had a relatively low living level. These two classifications actually make up the bulk of the sample -- thirty-three of the 96 families were nonwhite, and 55 out of 96 had homemakers in the top age class. The results of the eight final regressions on families with incomes $2+OO to $1,199 are given in Table 3.1+. One category of each of the characteristics, "chief source of income" and "education of homemaker” again erHiered the regression of this smaller subsample of low income families. Tenure of home by farm status continued to make little contribution to eXPlained variance. The race variable showed significance only with resDect to less basic expenditures; but this time, interestingly, the non- whitesv spent at a higher level than the whites. A detailed investigation showed that nonwhite families of this small Substample had outspent the white families in several less basic categories lncliliding house furnishings and equipment, transportation, personal care, ’3’ East Texas, 1958 Table 3ah-~Computed t values in regressions of expenditure or consumption on household characteristics, low-income rural families with income $400 to $1,500, : Independent variables : 2 Continuous variables Dummy variables : 1 : : . , _ e of Education of = Tenure of home by 3 : Sepengent : Money : Housing: Family: Chlef source Of income homemaker : homemaker. f farmjnonfarm status. 3 haee E E? arla .88 zincome: ' 3 ‘ : 3 _: : : . 3 - ~ - . o . ~ . = : Index = Slze : wages: PIOfeS = ::::= ‘ ‘ “0 ‘55 ‘i::: ‘ Grades: Grade : Own rzRenter: owner : Renter2White: Nonwhite: : ' 2 : or : along 2 Paym :Farmtunder: to :and :8th : 8 thru: 12 and: e : 3 non— : non- : ; : ’ ' 1 ' sala-. or ' rent, or. ' MO ' 5h iover' 3 ll 2 above 3 farm : farm 2 farm : farm ' : : n,‘ ‘ I Z . : ries :business: interest: 1 2 . :grade. . . : . : : : . Total : 2 2 2 : : : : : : : : : : z z z : : 3 expenditure 1 2.93u f n.996 : 2.u56 f f'-u.392 f f f f f f ; 3.017 ; 3 ; I 1 1 ; .h3o Total ‘ 2 : : : : : : : : : : : : : : : : : “nmpmn 12.185 3 8-573 2 3-657 3 E E ”-981 i i E f f i 3 M12 3 : : : : : ; .615 F001 : : : : : : z : : : : : : : : : : z : : expenditure 22.775 i 11.356 E ham 3 E E -’+-203 f f f E 3 $3361L E : : : : : : ; .397 Food : : : : : : : : : : : : : : : : : : : E ”Esme“ H.923 3 3-101 3 3-978 3 f 3 4.135 f E f i i f 3 32-9963 : : : : ; .367 ”we basic . = = ? = 2 i 3 = i i s 3 s 3 i i f 2 expenditure ; 3.179 : 5.648 5 M16 : E : -3-900 : : i : : 2-2-883: : : : : : 3 z #33 More basic ' 1 z : : : : : : : : : : : : z : : i 3 WWW ;2.1+os 3 9.2.15 3 5.6683 3 3 4-695 3 3 3 i E 3 32.559: : i-3-ou83 : z ; .636 Less basic . f = 5 3 § : : : : : : E f i E E : f : exPenditure ; 1.837 ; h.oou 3-0.u80 f f f -3.389 f 3 f f f f I 2.75% 3 ; 3 : :2.17h; : .268 I... has. i a s 2 z E 2 E 3 a 5 s i i 3 3 E °°nsumption ; 1.022 ; n.903 ; 0.197 :3.197 ; 3 ; ; ; ; ; 1 ; h-328 ; ; g 2 1 g : .375 t . '05 = l~973 - t.01 = 2.600 2M . . __ < ,V “‘“WMV M-~uww| A ' «......4..." .. .x ,. , ‘ -. C\ LC 0 recreation, and education. It is understandable, however, that whites should have spent less on these items because the white family component consisted mostly of families with older homemakers. These older families were small spenders on both the basic and nonbasic items. The analysis of the families in the $hOO to $l,h99 income class affords a good illustration of some of the added insights which a step- wise regression can provide. The final equation of this procedure is reported in Table 3.h for the present sample. But the stepwise method also points out three or four intermediate equations in addition to the final equation results. One can learn certain information from the inter- mediate equations which is not apparent from a study of the final equation alone. The results found in Table 3.h would indicate that money income contributes substantially to explaining variance in the dependent vari- ables. But this same money income variable is shown to be less significant as an explanatory factor in the first intermediate equation where only housing index and family size have been controlled. It is the entry of the transfer payments, rent, and interest category which gives a sizeable boost to the money income coefficient in all eight equations. In the stepwise analysis that has been employed for studying the present problem, the "transfer payment, rent, and interest” class was the first of the categorical variables to enter after the continuous variables had been forced to enter the equation. The second intermediate equation, therefore, gives parameters that were estimated in a regression where money income, housing index, family size, and a source of income classifi- cation were the independent variables. 6h. Table 3.5 gives computed t values pertaining to money income, befor and after the source of income" variable was part of the regression. The table also gives Ra's for the first and second intermediate equations. Table_3.5--Computed t values pertaining to the money income variable and fiz's for explained regression before and after the inclusion of a source of income variable, rural East Texas families with incomes $MOO to $1,u99 in 1958 is: .L \ Computed t value for monev income and 1 J First intermediate : Second intermfyiate o Dependent variable equation 2] : equation 8 t value 3 ha i t value I F2 Total expenditure .........: 1.72 I .286 - 2.727 1 .380 Total consumption .........: 0.85h : .h60 ; 1.87% 3 .536 Food eXpenditure ..........: 1,5ho : ,gul : 2,u66 : .329 Food consumption ..........: o,¢93 : .211 : 1,64h : .311 More basic expenditure ....: 2,033 : .316 : 2,917 : .388 More basic consumption ....: 0.738 : .h98 : 1,828 : .578 a/ The specified independent variables were money income, housing index, and family size. b/ The specified independent variables were money income, housing index, family size, and the classification of families whose chief source of income was from transfer payments, rent, and interest. The evidence has been that families with chief income from transfer payments, rent, and interest live on a considerably lower plane than the average of other classes. Furthermore, the families with this particular source of income are found more frequently near the $l,h99 level of income than down near the $MOO level. Thus, after this classification of families has been allowed for and controlled, there is an increase in the slope of the money income function. A directly opposite effect was obtained when the entire East Texas sample was analyzed, using the same variables. The entry of the same C\ \fl "source of income classification" into the regression, caused the money income coefficient to be adjusted downward instead of upward as in the jprevious case. The explanation is found once again in the frequency dis- ‘tribution of the various source of income categories. Families with incomes chiefly from transfer payments, rent, and interest were now concentrated more heaving near the lower end of the overall income distri- 'bution. Consequently, when their classification entered the regression, the Inoncy income coefficient was adjusted downward. "source The opposite adjustments to the money income coefficient, when (If income" entered, would probably not have been discovered if the print (Jut of the intermediate equations had not been available for each analysis. (Ehe availability of the intermediate analysis also makes it possible to (determine the amount that each entering variable contributes toward total explained variance . The analytic value of the stepwise regression technique is further iLLlustrated in Phase III where it proves very helpful in locating the lridden interactions. The employment of stepwise regression, however, is not without certain