This article was downloaded by: [Memorial University of Newfoundland] On: 30 January 2015, At: 03:24 Publisher: Routledge Informa Ltd Registered in England and Wales Registered Number: 1072954 Registered office: Mortimer House, 37-41 Mortimer Street, London W1T 3JH, UK

Biodemography and Social Biology Publication details, including instructions for authors and subscription information: http://www.tandfonline.com/loi/hsbi20

Predictors of family‐size preferences, 1945–1977: A multivariate analysis a

Judith Blake & Jorge H. Del Pinal

b

a

School of Public Health and Department of Sociology , University of California , Los Angeles, California b

School of Public Health , University of California , Los Angeles, California Published online: 23 Aug 2010.

To cite this article: Judith Blake & Jorge H. Del Pinal (1979) Predictors of family‐size preferences, 1945–1977: A multivariate analysis, Biodemography and Social Biology, 26:4, 302-313, DOI: 10.1080/19485565.1979.9988388 To link to this article: http://dx.doi.org/10.1080/19485565.1979.9988388

PLEASE SCROLL DOWN FOR ARTICLE Taylor & Francis makes every effort to ensure the accuracy of all the information (the “Content”) contained in the publications on our platform. However, Taylor & Francis, our agents, and our licensors make no representations or warranties whatsoever as to the accuracy, completeness, or suitability for any purpose of the Content. Any opinions and views expressed in this publication are the opinions and views of the authors, and are not the views of or endorsed by Taylor & Francis. The accuracy of the Content should not be relied upon and should be independently verified with primary sources of information. Taylor and Francis shall not be liable for any losses, actions, claims, proceedings, demands, costs, expenses, damages, and other liabilities whatsoever or howsoever caused arising directly or indirectly in connection with, in relation to or arising out of the use of the Content. This article may be used for research, teaching, and private study purposes. Any substantial or systematic reproduction, redistribution, reselling, loan, sub-licensing, systematic supply, or distribution in any form to anyone is expressly forbidden.

Downloaded by [Memorial University of Newfoundland] at 03:24 30 January 2015

Terms & Conditions of access and use can be found at http://www.tandfonline.com/ page/terms-and-conditions

Predictors of Family-size Preferences, 1945-1977: A Multivariate Analysis

Judith Blake and Jorge H. Del Pinal

Downloaded by [Memorial University of Newfoundland] at 03:24 30 January 2015

School of Public Health and Department of Sociology, University of California, Los Angeles, California: and School of Public Health, University of California, Los Angeles, California ABSTRACT: This paper focuses on a set of substantive and methodological problems in the study of family-size preferences. Substantively, we ask whether traditional social-demographic variables explain proportionately more or less variance in family-size preferences during the mid-1970's than during the period following World War II. Additionally, we inquire whether individual variables, such as race, religion, education, and community size, perform differently in recent years than they did in the postwar period. Methodologically, we question whether analogous trends emerge from including diverse indicators of family-size preferences. Using Multiple Classification Analysis, we find no diminution over the time period in the proportion of variance explained by the socialdemographic predictors considered. In fact, these variables appear to increase in predictive power over time. However, some variables have clearly changed in marginal importance since World War II. Religious affiliation and community size have declined, and race has increased. Educational level has maintained a major position throughout. Finally, the differentials in reproductive preferences are very congruent regardless of the dependent variable used—family-size ideals, preferences, intentions, or ideals for the average American family.

Over the past thirty-five years demographers have tried to show that social and demographic variables are powerful predictors of family-size preferences. On the whole, this effort has been disappointing. Although religious, racial, educational, occupational, and residential divisions have produced some differences in familysize desires, the amount of variance explained has typically been small and is believed to be diminishing. As long age as 1962, Ronald Freedman remarked on the declining role of the "traditional" differentials in family-size preferences (Freedman, 1962). It is now commonplace to assume that social and demographic predictors of reproductive variation have continued to lose importance (Westoff and Westoff, 1971). Indeed, one of the most frequent justifications for new approaches to studying fertility is that the traditional variables have lost most of

their explanatory power (e.g., Turchi, 1975). For example, stimulated by microeconomic theory concerning fertility, demographers have increasingly evoked as predictors various measures of income and other economic indicators (Freedman, 1963; Mincer, 1963; Easterlin, 1969; Cain and Weininger, 1973). However, no systematic trend analysis of predictors of family-size desires, using modern multivariate techniques, exists. Hence, it is unclear whether the explanatory power of social and demographic variables actually has diminished over time. Do traditional predictors in fact explain less variance in reproductive preferences today than they did, say, during the post— World War II period? Which predictors are the most promising? Do their effects on family-size preferences remain constant over time? The present paper is designed to address these questions. It is

302

Downloaded by [Memorial University of Newfoundland] at 03:24 30 January 2015

Vol. 26, No. 4

Family-size Preferences

stimulated by the conviction that we need to reassess, quantitatively, which types of fertility predictors are important today as against the past—in this case, the period following World War II. It is our suspicion that the study of population is accumulating a great deal of "theory" about some predictors that may have little explanatory power, while at the same time it is neglecting others whose track record may be more impressive. In this report we have confined ourselves to a consideration of trends in the importance of selected predictors over time. An analysis of trends in the nature of these relationships will be forthcoming. We have utilized a series of Gallup surveys on ideal family size between 1945 and 1977—17 surveys in all—to analyze possible changes in the socioeconomic predictors of family-size preferences. This series on family-size preferences is the longest available for national samples of the U.S. population of both sexes and voting age. Additionally, we wished to validate our findings by analyzing other measures of family-size preferences from five national fertility surveys conducted in the United States between 1955 and 1973. These surveys are described in the following section. All of the data sets have been analyzed using AID (Automatic Interaction Detector) and MCA (Multiple Classification Analysis). MATERIALS AND METHODS The national Gallup surveys used here all contained the same question on family size preferences: "What do you consider the ideal size of a family—a husband, wife, and how many children?" Most of the studies also included comparable data on selected predictor variables—race, religion, education, occupation, family income, and community size in addition to age and sex. Hence, taking account of

303

some possibly confounding demographic factors such as age and sex, we can study major social and demographic differentials in family-size preferences over time. We can also examine the importance of income in view of the theoretical interest in this variable. Occupation of the chief wage earner as a predictor was dropped from the regressions because a high proportion of older respondents were retired (and no occupational data were available for them), and the marginal importance of occupation was very slight for all of the surveys. The marginal importance of marital status was also so slight that it was not worth including in the regressions. Because some of the surveys, dating from the 1940's and 1950's, did not include one or more predictor variables, we have also presented a time series based on those data that were available from the earlier period through 1977. The five fertility surveys were stratified national samples of women in the reproductive ages (15^9). The 1955 and 1960 Growth of American Families studies and the 1965 National Fertility Study were based on surveys of currently married women. The 1970 National Fertility Study included ever-married women and the 1973 National Survey of Family Growth was based on single and ever-married women. In order to insure comparability among the fertility surveys, we analyzed only currently married women. The fertility surveys contained a variety of questions concerning the dependent variable, family-size preference. These questions will be discussed in detail in a later section. The predictor variables were similar to those in the Gallup surveys, except that the fertility studies allow us to use husband's income rather than family income. The latter is believed by some to confound income effects (Mincer, 1963; Easterlin, 1969; Cain and Weininger, 1973).

Downloaded by [Memorial University of Newfoundland] at 03:24 30 January 2015

304

Blake and Del Piñal

The research questions we have posed are answered most efficiently through the use of multiple-regression techniques. We have employed Multiple Classifcation Analysis (MCA), an elaboration of "dummy variable" multiple-regression analysis (Andrews et al., 1973), which allows for categorical variables as well as numerical values. The method makes no assumptions about linearity and can demonstrate possible nonlinearities as a normal feature of its operation. However, since MCA assumes a high degree of additivity, it is not appropriate to analyze data in which there are major interactions unless it is possible to construct interaction variables. In order to detect interactions and other anomalies, we first analyzed all of the surveys with AID (Automatic Interaction Detector) (Sonquist et al., 1973; Sonquist, 1970). The AID analyses, plus MCA tabulations, showed some of the Gallup studies to be interactive and some clearly additive. For each of the interactive surveys, the anomalies were taken into account through interaction terms. In all, ten Gallup studies emerged that were either additive or nearly so. In the presentations that follow, the R2 (adjusted for degrees of freedom) is given for all of the studies, including those where interaction terms have been used. The remaining analyses assess the importance of particular predictors. Unfortunately, the importance is difficult to evaluate when a variable in question is part of an interaction term. Indeed, if a true interaction exists, then the variables interacting do not have separate identities, and their effects should not be assessed independently. Consequently, for assessing unique effects, we used the ten Gallup polls where the additivity assumption was justified and, hence, no complex interaction terms were required. The same procedure was used to analyze the fertility studies. Only one was deemed interactive.

Social Biology

We are concerned here with the proportion of the variance in family-size preferences (R1) we can "explain" by socialdemographic variables and how this changes over time. However, a change in R2 may occur for a number of reasons. For example, there may be a shift in the variance of the dependent variable. A great deal of dispersion in the dependent variable will, other things being equal, increase the probability that a given set of predictors will be able to explain a high proportion of this variance, whereas a diminution in the dispersion of this variable may require a very fine explanatory net in order to generate much power. Conversely, a change in the variability of the predictors can alter their ability to explain variance. By the same token, a change in the intercorrelation among predictors may also change the proportionate level of explanation. Finally, an overall change in R2 can come about because the importance of individual predictors (the zero-order correlation with the dependent variable) shifts over time. In the course of the discussion that follows, we shall see that R2 has differed because particular variables have changed in unique importance. RESULTS THE GALLUP SURVEYS

Compared with the early post-WorldWar-II period, what proportion of the variance in family-size preferences can be explained thirty years later by traditional social-demographic variables—age, sex, race, religion, education, and community size? What is the independent contribution of each of these predictors to explained variance? Has their marginal importance changed over time? And, finally, what is gained in explanatory power by adding an economic variable such as family income? Table 1 presents the amount of ex-

Vol. 26, No. 4

Family-size Preferences

305

TABLE 1 VARIANCE IN FAMILY-SIZE PREFERENCES EXPLAINED BY SELECTED SOCIAL-DEMOGRAPHIC CHARACTERISTICS: 17 NATIONAL SURVEYS IN THE UNITED STATES, 1945-1977 (R2 ADJUSTED FOR DEGREES OF FREEDOM)

Downloaded by [Memorial University of Newfoundland] at 03:24 30 January 2015

Model No.'

1 2

Aug. Jan. Mar. May Feb. June May Mar. Dec. May June Sept. Feb. Oct. Sept. April Feb. 1945 1947 1952 1953 1957 1959 1963 1965 1967 1968 1972 1972 1973 1973 1974 1975 1977

3

t t 0.06 t 0.07§ 0 06§ 0 09 0 08 0 08 0 07 0 10 0 09 0 09 0 10 0 V 0 11 0 12 f 0 05 •f0 04 0 07 0 12 0 10 0 05 0 10 0 10 0 10 0.04 OOS 0 04 0 04 nos 0 04 0 08 0 04

N_.... X.... SD.... V ....

2,8802,7981,9701,5001,3641,3561,3701,2941,4201,5661,4921,4791,4641,4841,5011,5471,486 3.53 3.26 3.30 3.30 3.40 3.54 3.50 3.39 3.32 3.37 2.94 2.92 3.08 2.97 2.82 2.85 2.62 1.40 1.28 1.29 1.30 1.19 1.22 1.23 1.12 1.18 1.31 1.41 1.30 1.49 1.32 1.31 1.32 1.20 1.96 1.64 1.66 1.69 1.42 1.49 1.51 1.26 1.39 1.72 1.99 1.69 2.22 1.74 1.72 1.74 1.44

'Model 1: Age, sex. race, religion, education, community size, family income; Model 2; Age. sex. race, religion, education; Nlodel3: Age, sex. race, education, community size; ¿V. number of respondents in regression: A\ sample mean; su. standard deviation; V. unbiased sample variance. Ellipsis dots indicate that the survey was interactive. tNo data on religion. tNo data on religion and income. §No data on income.

plained variance for 17 Gallup surveys according to three models. The first model includes all of the basic background variables. Models 2 and 3 are shown for additive surveys only, and where the designated predictors were available. In Model 2 the number of predictors was cut to the five having temporal priority in the life cycle—age, sex, race, current religious affiliation, and education. Model 3 includes those variables for which we have predictors as far back in time as 1943. Turning first to the changing importance of traditional background factors as predictors of family-size preferences, in Model 1 we see that proportionately more variance was explained during the 1970's than in 1952 or during the 1960's. At no date do these variables explain much of the variance. There is a high of 12 per cent in 1974 and 1977, but the recent record is better than in the early years of this series when they accounted for no more than 6 to 8 per cent. Approximately the same proportion of variance can be explained over the years by reducing the number of predictors to five—age, sex, race, religious affiliation, and education (Model 2).

Again, with this model, more variance is explained by these predictors during the 1970's than previously. It is possible to carry Model 3 back to 1945, and in this series we can clearly see the trend toward greater explanatory power over time— from 4 per cent in 1945 to 10 per cent in 1975. Although we have not computed Fratios because they are inappropriate for national stratified samples (see Glenn, 1977, p. 42), it is unlikely that this regular progression of increasing R2 is due to chance factors, particularly since a number of early studies have low R2's and a number of later ones have higher R2's. The increasing ability over time of background variables to explain familysize preferences is unlikely to be due to a change in variance in the dependent variable, family-size preferences. As may be seen from Table 1, the variance in ideal family size did not follow a steadily increasing course. It generally decreased from the 1940's to the early 1960's, then increased until the early 1970's, and has declined since then. As for changing variance in the predictors, MCA predictors are categories (dummy variables) and

Social Biology

Blake and Del Final

306

TABLE 2 MARGINAL IMPORTANCE OF SELECTED SOCIAL-DEMOGRAPHIC CHARACTERISTICS IN EXPLAINING FAMILY-SIZE PREFERENCES: SEVEN NATIONAL GALLUP SURVEYS, 1952-1975 AVERAGE PERCENTAGE CHANGE IN EXPLAINED VARIANCE (R2)

FOUR POLLS PRIOR TO 1970

Downloaded by [Memorial University of Newfoundland] at 03:24 30 January 2015

VARIABLES IN MODELS

Age and sex Age, sex, and race Age, sex, race, relig Age, sex, race, relig., educ. Age, sex, race, relig., educ. commun, size Age, sex, race, relig., educ, commun, size, fam. income*

THREE POLLS AFTER 1970

Percentage Increasct

Averaee R2"

Percentage Incrcaset

0.019 0.020 0.042 0.053

8' 104 29

0.019 0.057 0.076 0.108

206 35 42

0.068

27

0.110

2

0.070

2

0.112

2

Average R2

"This comparison excludes the 1957 poll which had no information on income. tCalculated from unrounded average R

therefore have no variance in the statistical sense. We have also made the categories for each variable as comparable as possible throughout. Hence, changes in predictor variation are not, in this type of analysis, responsible for changes in R2. The problem of intercorrelation among predictors is obviated, to a great extent, by the fact that we use the marginal or unique importance of independent variables throughout. Finally, changes in firstorder relations between predictors and family-size preferences will be discussed when the results for individual predictors are presented. Our results so far contradict the impression that the social-demographic variables explain less variance now than in the past. We believe that this impression may stem from the changing importance over time of particular variables, a supposition that can be tested by examining the marginal effects of the predictors. We assess the unique or marginal effect of a predictor by running two regressions, one run with and one run without the vari-

able in question. The percentage change in R2 (adjusted for degrees of freedom) shows the additional explanatory effect achieved by adding a particular predictor. We have added the predictors into the regression according to their temporal priority in the life cycle. We start with ascribed characteristics such as age, sex, and race, then add characteristics acquired early in life, such as religious affiliation and educational level, and finally current characteristics such as community size and family income. A notable finding in this analysis is that the effects of certain predictors have changed substantially over time (Table 2). Race. Among the ascribed characteristics, race (essentially a white/black dichotomy) has most radically changed its marginal importance over the time period we are considering. Prior to 1970, adding race to age and sex in the regression increased explained variance (R2) by an average of 8 per cent. Since 1970, race adds over 200 per cent to explained variance (compare the first and second rows of

Downloaded by [Memorial University of Newfoundland] at 03:24 30 January 2015

Vol. 26, No. 4

Family-size Preferences

Table 2). The marked increase in the importance of race (net of other variables) as a predictor is not simply due to the fact that the Gallup samples have become racially more heterogeneous over time (Glenn, 1977). Rather, it results from a real change in the zero-order relationship between race and family-size preferences. Moreover, this relation does not "wash out" when other predictors are added to the analysis. In sum, family-size preferences of blacks and whites now differ more than they did previously, both before and after we have controlled for the effects of other predictors. Looking at these data alone, we would have to conclude that race is a relatively promising predictor of family-size preferences. To understand how it operates on such preferences is of course another matter, but to attempt such understanding appears to be increasingly worth the effort. Religious Affiliation. Throughout the regressions, respondents were classified as Catholics, Protestants, Jews, without a religious affiliation, or "other." The "other" category is relatively small, rising from 1 per cent in 1963 to only 5 per cent in 1977. A comparison of the second and third rows in Table 2 demonstrates the marginal effect of religious affiliation before and after 1970. Clearly, the unique effect of religion has declined markedly over time. Prior to 1970, adding religion to age, sex, and race increased the variance explained by 104 per cent. Since 1970, the marginal effect has declined to 35 per cent. This trend is also present in the first-order relation of religion and family-size preferences. Educational Level. The marginal importance of education as a predictor has generally increased throughout the period (Table 2). The addition of education to age, sex, race, and religion typically raises the variance explained after 1970 by 42 per cent, whereas the comparable figure prior

307

to 1970 is 29 per cent (compare the third and fourth rows of Table 2). This result reflects a marked rise in the zero-order relationship between schooling and familysize preferences, a relationship that is not vitiated by the association of other predictors with education. Education, like race, is thus a basic background predictor that in recent years has attained uniquely important status in explaining family-size preferences. Community Size. The fairly large effect community size once had on reproductive preferences had been dissipated by the early 1970's. Table 2 shows that this variable increased R2 by 27 per cent prior to 1970 (compare the fourth and fifth rows). Since then, this increase has diminished to 2 per cent. This closing of the rural-urban differential (our variable distinguishes rurality from urban communities of various sizes) in family-size preferences reflected in the first-order relationship, together with the diminished religious differentials, may explain the widespread perception of an overall drop in the predictive power of sociodemographic variables. Family Income. The rationale for adding income last is that, to a large extent, income is determined by some of the other predictors already entered in the regression—particularly race and education. The marginal effect of this variable is typically very low regardless of the timeperiod considered, although the zeroorder correlation rivals that of religion (compare the fifth and sixth rows, Table 2). In other words, when temporally prior variables such as education are taken into account, the marginal importance of family income as a predictor of family-size preferences is very small. We find little support for the notion that family income is. an important predictor in this type of analysis. To sum up our results so far, it seems

308

Social Biology

Blake and Del Pinal TABLE 3

VARIANCE IN FAMILY-SIZE PREFERENCES EXPLAINED BY SELECTED SOCIAL-ECONOMIC AND DEMOGRAPHIC CHARACTERISTICS: FIVE NATIONAL FERTILITY STUDIES IN THE UNITED STATES, 1955-1973 (R2 ADJUSTED FOR DEGREES OF FREEDOM)

FECUND MARRIED WOMEN

Expected Family Size

Au. MARRIED WOMEN

Ideal Average American Family

Ideal Averace American Family

Downloaded by [Memorial University of Newfoundland] at 03:24 30 January 2015



GAF No.*

3. .... ....

SD . . . .

V

....

GAF

NSFG

NFS

NSFG

GAF

NFS

NSFG

1965t

1970$§ 1973$

1955+

1961)

1965$

1973$

1955+

1960

1965$

1973$

0.14 0.12 0.14 1,782 2,372 3.35 3.42

0.14 0.12 0.13 3,124 3.41

0.16 0.14 0.15 4,668 2.99

0.16 0.15 0.15 4,516 2.69

0.08 0.05 0 08 1,775 3.40

0.10 0.07 0 10 2,331 3.52

0.10 0.08 0 10 3,160 3.22

0.10 0.09 0 10 4,273 2.42

0.07 0.04 0 07 2,684 3.41

0.09 0.07 0.09 3,195 3.55

0.12 0.10 0.12 5,544 3.33

0.10 0.09 0.10 6,976 2.44

1.56 2.43

1.73 2.99

1.50 2.25

1.35 1.82

0.96 0.92

1.17 1.37

1.06 1.12

0.80 0.64

0.97 0.94

1.17 1.37

1.11 1.23

0.83 0.69

1960

0.16

1 2

N X

1955t

NFS

1.74

3.03

•Model 1: Age. race, religion, education, community size, husband's occupation, husband's income; Model 2: Age. race, religion,education: Model 3: Age. race, religion, education, community size, husband's occupation: ,V. number of respondents in regression; A\ sample mean; SD. standard deviation: V. unbiased sample variance. tGAF, 1955, includes only white respondents. For example, in Model 2. the explained variance is only for age, religion, and education. The ellipsis dots indicate that, with respect to family size expectations, it is also interactive. {Appropriate weights were used in the regressions to take account of the oversampling of blacks. §Number of children intended.

clear that the entire group of social and demographic predictors of family-size preferences has not lost explanatory power over time. Actually, the group has gained power since our first observations after World War II, a fact that cannot be explained by the changing variance in the dependent variable. Thus, considering major social and cultural cleavages in the society, people's family-size preferences have become more different over time, on the average, rather than more similar. However, now people differ along different dimensions from those of the past. We find that the marginal effects of individual predictors have changed over the years. Family-size preferences appear to have become more similar across religious and rural-urban groups while becoming more different across educational and racial groups. Family-size preferences among income groups appear to be mainly a function of other variables, since inclu-

sion of this predictor adds very little explanatory power. These results suggest strongly that more attention is required to understand the fertility differentials of the present, rather than assuming them away because the major cleavages of the past—religion and rural-urban differences—are no longer so sharp. For example, is the racial split simply another expression of educational variability? Or are there cultural and social components to black fertility that may transcend, or even short-circuit, educational attainment? Given the poor showing of income as a predictor, must we conclude that economic theories that relate resources to family-size desires are poorly specified? Is it possible that economic considerations come into play only after the effects of other incentives have been operative? Is it also possible that such incentives may even cancel out the effects of economic rationality?

Vol. 26, No. 4

Family-size Preferences

INDEPENDENT VERIFICATION:

THE 1955,1960,1965,1970, AND 1973

Downloaded by [Memorial University of Newfoundland] at 03:24 30 January 2015

FERTILITY SURVEYS

Would findings similar to those from the Gallup series result from using different indicators of reproductive preference (family-size expectations or intentions instead of ideals) and differently composed samples (women in the reproductive years instead of adults of all ages and both sexes)? As a test, we have performed a congruent analysis on the 1955 and 1960 Growth of American Families Studies (GAF, 1955 and I960); the 1965 and 1970 National Fertility Studies (NFS, 1965 and 1970); and the 1973 National Survey of Family Growth (NSFG, 1973). In all of the surveys, women were asked about their family-size expectations or intentions. And, except in 1970, women were asked a standard question on their views of the ideal family size for the average American family. It will be remembered that the Gallup question did not refer to the average American family. The questions on expectations/ intentions have been tabulated only for women without known fecundity impairments. Obviously, the response to questions on expectations or intentions may be biased by the knowledge of subfecundity or sterility. Since there were many more voluntary sterilizations in 1970 and 1973 than in 1955, we did not wish to bias any possible real differences among the survey responses because of these changes. The question on ideal family size for the average American family was tabulated for both fecund women and all women. Except for income and occupation, most predictor variables were similar to those in the Gallup surveys. In the fertility surveys we were able to use husband's income rather than family income as suggested by economists (Freedman, 1963; Mincer, 1963; Easterlin, 1969). Additionally, we

309

have data on the occupations of the husbands in the fertility studies. As with the Gallup series, we generally find an increase rather than a decrease in the proportional variance explained by the background predictors at the later dates (Table 3). This occurs despite a decline in the variance of the dependent variable since 1960. Thus, neither data set substantiates a recent contraction in the proportion of variance explicable by social-demographic predictors. Congruent with the Gallup series, virtually all of the explained variance is due to four basic predictors—race, age, religion, and education. As between Models 1 and 2, it is evident that, in recent years especially, very little was added by husband's income, community size, and husband's occupation (Table 4). This is true for the fertility studies regardless of whether expectations or ideals are considered. We may now assess the contribution of individual predictors. Because of the sampling of the black population in the earlier surveys, strict comparability among the surveys cannot be assured. For this reason, we do not consider the marginal importance of race. Religious Affiliation. In conformity with a temporal decline in the zero-order importance of religion to family-size preferences, when religious affiliation is added to age and race (Table 4), its declining marginal importance in later years as against 1960 is evident. With regard to expectations or intentions, in the 1960 data, the amount of explained variance more than triples when religious affiliation is added to age and race, whereas a similar addition in 1973 produces an increase of 37 per cent. The tabulations on ideal family size for the average American family in the fertility surveys show a similar trend. Thus, both the Gallup data and the fertility studies agree on the generally declining marginal importance of religious

Downloaded by [Memorial University of Newfoundland] at 03:24 30 January 2015

TABLE 4 MARGINAL IMPORTANCE OF SELECTED SOCIAL-DEMOGRAPHIC CHARACTERISTICS IN EXPLAINING FAMILY-SIZE PREFERENCES: FOUR NATIONAL FERTILITY SURVEYS, 1960-1973

EXPECTED* VARIABLES

1960

1965

1970i

iDEALt

IDEAL*

1973

1973

1960

1965

197011

1973

... ... ... ... ...

0.041 0.070 0.090 0.096 0.100

0.013 0.053 0.069 0.077 0.094

0.029 0.054 0.096 0.102 0.118

... ... ... ... ...

0.049 0.065 0.088 0.095 0.101

S"

0.098 0.098 . . .

0.103

0.095 0.118 . . .

0.104

2.

1960§

1965

197011

Adjusted R

Variables in Model Age, race Age, race, relig Age, race, relig., educ Age, race, relig., educ, husband's occup Age, race, relig., educ, husband's occup., comm. size Age, race, relig., educ, husband's occup., comm. size, husband's income

0.018 0.080 0.120 0.132 0.143

0.033 0.062 0.086 0.073 0.106 0.118 0.1210.140 0.151 0.124 0.147 0.154 0.135 0.1510.155

0.142 0.136 0.159 0.159

0.012 0.050 0.068 0.084 0.098

0.013 0.042 0.076 0.082 0.099

341 49 10 8

121 66 2 9

70 32 5 3

37 28 2 1

317 37 23 17

218 79 8 20

n 8S 3 D.

3 5'

Percentage increase in/?

Marginal Effect of Adding Religion to age, race Education to age, race, relig Husband's occup. to age, race, relig., educ Comm. size to age, race, relig., husband's occup. educ. .. Husband's income to age, race, relig., educ., comm. size, husband's occup

ÇS

71 29 6 4

298 30 12 23

87 79 6 16

34 34 9 6

0

"Fecund currently married women. tCurrently married women. îNumber of children intended. §The 1955 survey included only white women and hence docs not appear in this table. IIThe 1970 survey did not ask a question on family-size ideals. #Computcd from unrounded R2.

s.E. C/3

ES

S] 5"

2?

Downloaded by [Memorial University of Newfoundland] at 03:24 30 January 2015

Vol. 26, No. 4

Family-size Preferences

affiliation, and this finding holds in the fertility studies for both expectations and ideals. Education. The unique effect of education does not increase over time in the fertility surveys (Table 4); in fact, it decreases, reflecting a slight decline in the zero-order relation. However, regardless of the dependent variable used, the unique role of education relative to the unique role of other variables rises appreciably. For example, regarding family-size expectations for 1960, the addition of religion caused a 341 per cent increase in R2, whereas adding education resulted in an additional 49 per cent increase. For 1973, the respective figures were 37 per cent and 28 per cent. It may be that the trend results for education in these data, as compared with the Gallup surveys, are biased by the fact that earlier fertility studies were weighted in favor of more highly-educated women. Community Size. Congruent with the Gallup surveys, the single-order and the marginal effect of community size in the fertility surveys has declined greatly. For example, the percentage increase by virtue of adding community size to age, race, religion, education, and husband's occupation in 1960 was 8 per cent, whereas in 1973 it was 1 per cent. For ideals, in the same period of time, the percentages decreased from 17 to 4 and from 23 to 6, depending on whether fecund currently married women or just currently married women were considered. Husband's Occupation. Correlatively, the role of husband's occupation in the fertility surveys has also declined over time. Along with community size, husband's occupation has thus become a marginally unimportant variable during recent years. Husband's Income. Finally, turning to husband's income, the results are congruent with those for family income on the

311

Gallup surveys. The marginal effect of husband's income in the fertility surveys is slight or nonexistent (Table 4), despite a fairly constant zero-order relation with family-size preferences. It does appear that there is a slight increase in importance in recent years, but the 1973 data still only show a 2 per cent increase in R2 from adding husband's income to the other variables. We do not find this result surprising since it appears that regardless of the economic-status measure used, very little explanation is achieved (see Curtin, 1976; Beaujotetal.,1978). SUMMARY AND CONCLUSION This paper focuses on a set of substantive and methodological problems in the study of family-size preferences. Substantively, we ask whether traditional socialdemographic variables predict more or less variability in family-size preferences during the mid-1970's than in the period following World War II. Additionally, we inquire whether individual variables such as race,, religion, education, and community size perform differently in recent years than in the post-war period. We also question whether analogous trends emerge from using diverse indicators of family-size preferences—personal familysize ideals, expectations, and intentions, and ideals for the average American family. Employing Multiple Classification Analysis, we use two types of data sets. Initially, we concentrate on 17 Gallup surveys spanning the period 1945 to 1977. Each contained a question on family-size ideals, and all included virtually identical data on predictor variables—age, sex, race, religious affiliation, educational level, family income, and community size. Then, we turn to the 1955 and 1960 Growth of American Families studies, the 1965 and 1970 National Fertility studies,

Downloaded by [Memorial University of Newfoundland] at 03:24 30 January 2015

312

Blake and Del Piñal

and the 1973 National Survey on Family Growth—all covering married women in the reproductive ages. These studies, in addition to including information on predictor variables that is analogous to the Gallup data, contained questions on reproductive preferences that differed from the one in Gallup on ideal family size. The questions in the fertility surveys related to personal expectations and intentions, plus a question on the ideal family size for the average American family. Substantively, our analysis suggests that there is no diminution over the time period in the proportion of variance to be explained by the social-demographic predictors considered. In fact, these variables appear to increase in predictive power over time. However, we must also emphasize that at no time do the variables explain more than a relatively small percentage of the variance in family-size preferences—a high of 12 per cent in the Gallup series in 1974 and 1977, and 16 per cent (for family-size intentions) in the 1973 National Survey of Family Growth. Clearly, if one wishes to think in terms of explained variance, one needs to search for additional predictors. In fact, when we turn to the marginal importance of our

Social Biology

predictors, we see that most of the variance explained is due to age, race, religious affiliation, and education. Income— husband's or family—makes little or no contribution once causally prior predictors are taken into account. Some predictors have clearly changed in importance over time. Religious affiliation and community size have declined in importance, and race (at least in the Gallup surveys) has increased. Educational level has maintained a major position throughout. Finally, a methodological finding: The differentials in reproductive preferences studied are highly congruent between the two kinds of data sets—the Gallup surveys and the national fertility studies. In effect, the Gallup series on ideal family size produces a picture concerning socialdemographic differentials that is very similar to the one depicted by expected/ intended (as well as ideal) family size in the fertility surveys. Moreover, a comparison of the computations for ideal and expected/intended family size in the fertility studies also indicates internal congruence. We believe it is significant that diverse questions on reproductive preferences from disparate studies will produce findings that are so similar.

REFERENCES ANDREWS, F., J. MORGAN, J. SONQUIST, and

L.

KLEM. 1973. Multiple classification analysis. 2nd ed. Univ. of Michigan Press, Ann Arbor. BEAUJOT, R., K. KROTKI, and P. KRISHNAN. 1978.

Socio-cultural variations in the applicability of the economic model of fertility. Pop. Stud. 32: 319-325. CAIN, G., and A. WEININGER. 1973. Economic

determinants of fertility: Results from crosssectional aggregate data. Demography 10: 205-223. CURTIN, R. 1976. Patterns of income and fertility among American households, 1967-1973. In G. Duncan and J. Morgan (eds.), 5000 American families. Univ. of Michigan Press, Ann Arbor.

EASTERLIN, R. 1969. Toward a socio-economic theory of fertility: A survey of recent research on economic factors in American fertility, p. 127-156. In R. Behrman and L. Corsa, Jr. (eds.), Fertility and family planning: A world view. Univ. of Michigan Press, Ann Arbor. FREEDMAN, D. 1963. The relation of economic status to fertility. Amer. Econ. Rev. 53: 414-426. FREEDMAN, R. 1962. American studies of family planning and fertility: A review of major trends and issues. In C. Kiser (ed.); Research in family planning. Princeton Univ. Press, Princeton. GLENN, N. 1977. Cohort analysis. Sage Publications, Beverly Hills.

Vol. 26, No. 4

Family-size Preferences

MINCER, J. 1963. Market prices, opportunity costs and income effects, p. 67-82. In C. Christ et al.(eds.), Measurement in economics. Stanford Univ. Press, Palo Alto. SONQUIST, J. 1970. Multivariate model building. Univ. of Michigan Press, Ann Arbor.

Downloaded by [Memorial University of Newfoundland] at 03:24 30 January 2015

SONQUIST, J., L. BAKER, and J. MORGAN.

1973.

313

Searching for structure. Univ. of Michigan Press, Ann Arbor. TURCHI, B. 1975. Microeconomic theories of fertility: A critique. Soc. Forc. 54: 107-125. WESTOFF, L. A. and C. F. WESTOFF. 1971. From

now to zero—Fertility, contraception, and abortion in America. Little, Brown, Boston.

Predictors of family-size preferences, 1945-1977: a multivariate analysis.

This article was downloaded by: [Memorial University of Newfoundland] On: 30 January 2015, At: 03:24 Publisher: Routledge Informa Ltd Registered in En...
815KB Sizes 0 Downloads 0 Views