Social Class and Changing Families in an Unequal America

Marcia J. Carlson, Paula England, Marcia J. Carlson, Paula England

• Publisher: Stanford University Press
• Year: 2011
• ISBN:
  9780804779081
• Edition: 1
• Topic: Ciencias sociales
• Subtopic: Matrimonio y sociología familiar

American families are far more diverse and complex today than they were 50 years ago. As ideas about marriage, divorce, and remarriage have changed, so too have our understandings about cohabitation, childbearing, parenting, and the transition to adulthood. Americans of all socioeconomic backgrounds have witnessed changes in the nature of family life, but as this book reveals, these changes play out in very different ways for the wealthy or well off than they do for the poor.

Social Class and Changing Families in an Unequal America offers an up-to-the-moment assessment of the condition of the family in an era of growing inequality. Highlighting unique aspects of family behavior, it reveals the degree to which families' varying experiences are shaped by social class. This book offers a much needed assessment of contemporary family life amid the turbulent economic changes in the United States.

CHAPTER ONE

Birth Control Use and Early, Unintended Births

Evidence for a Class Gradient

Paula England, Elizabeth Aura McClintock, and Emily Fitzgibbons Shafer

Family patterns have always differed by social class in America. Generations ago, lower- or working-class individuals married earlier than the middle class, and all classes typically started childbearing shortly after marriage. In the past, less-privileged young adults typically had a first birth at about 18–20 years of age; they do the same today except they are often not married (Ellwood and Jencks 2004; Rindfuss et al. 1996). The shotgun marriage—a marriage that occurs in response to a premarital pregnancy—has gone out of fashion (Akerlof et al. 1996), yet an unintended pregnancy often galvanizes couples to solidify a several-month-old romantic relationship and even move in together (Reed 2006). About 80 percent of nonmarital births are to romantically involved couples (England and Edin 2007, Chapter 1), but they typically break up within a few years. Each partner may go on to have another unintended birth with another partner, leading to multiple-partner fertility and complex households. As Sara McLanahan illustrates in this volume, from a child’s point of view, this produces many transitions and instability in household composition. The pattern is most pronounced among poor blacks, but non-college-educated whites and Latinos are increasingly following these patterns as well.

Middle-class childbearing remains largely in marriage, and is increasingly delayed until well after marriage. As a result, class differences in the age of mothers at the birth of a first child have grown appreciably (Ellwood and Jencks 2004; Rindfuss et al. 1996). Class differences in the instability of children’s circumstances have long been present because of class differences in divorce rates, but today such differences are magnified by the less durable unions into which the children of disadvantaged parents are born.

Recent academic discussions of class differences in family patterns focus on the “retreat from marriage,” and many policy discussions suggest the need to encourage marriage as a way of reducing instability for less-privileged children. Cultural changes have raised the economic and relational standards for marriage (Cherlin 2004). At the same time, in recent decades, the earnings of men without a college degree have fallen, creating a situation where the poor are unlikely to ever meet their own economic standards for marriage (England and Edin 2007). The retreat from marriage has undoubtedly increased family instability and complexity, as well as the class gradient in such instability. However, we believe that the retreat from marriage has been overemphasized.

In this chapter we focus on a more causally “upstream” set of causes for class differences in family patterns. As we will show, more-advantaged youths begin engaging in intercourse slightly later and, as young adults, use birth control (contraception and abortion) more consistently. As a result, they are much less likely to become parents early, or to have unintended births at any age. While early births are not always unintended, and not all unintended pregnancies are early, the two phenomena are empirically linked: a national survey asking women about their childbearing between 1997 and 2002 found that 78 percent of births to women under age 20 resulted from unintended pregnancies, compared to 45 percent among women 20–24, and 24 percent among women 25–44 (Kissin et al. 2008). This is probably because few see the teen years as appropriate for childbearing, and because anyone who has a high propensity for unplanned pregnancies because of inconsistent contraceptive use will probably have an unplanned pregnancy shortly after the initiation of sexual activity. As Bongaarts (1978) has pointed out, sex and birth control are the proximate determinants of fertility. It follows that class differences affecting early fertility must operate through these proximate determinants. Once premarital sex is ubiquitous, unintended fertility is particularly likely to flow from lack of consistent use of birth control. Class differences in unintended fertility, then, are likely a result of class differences in birth control use.

Our focus here is on class differences, not race differences. Racial differences in family patterns are well documented in past research. To maintain the focus on class, in our analysis we combine all racial groups, statistically adjusting out of any class differences the amount that reflects the different racial composition of those in various class locations.

Our concept of class is gradational. We are interested in how the plethora of correlated measures of hierarchical position that go into what we typically mean by class or socioeconomic status—income, occupation, education, and so forth—affect behaviors that lead to early, unplanned births. Given this broad view, we use the terms “class” and “socioeconomic status” (hereafter “SES”) interchangeably. We recognize that for some outcomes, predictive power is increased if discrete classes are posited (Weeden and Grusky 2005). Here, however, we utilize theories that suggest gradational processes. As we consider behavior that spans late childhood and early adulthood, both the class of one’s family of origin and one’s own emergent social class are relevant, although our analyses emphasize the former. Most of our own empirical analyses here use an individual’s mother’s education as the most readily available measure tapping class background, but we augment analyses with some other measures, and we review research using other measures as well.

Our goal is to document and suggest explanations for class differences in early and unplanned births. We combine a review of past research with a presentation of our own data analysis. The evidence from past analyses and our own analysis shows that, although their fertility goals early in life are the same as those of their more-privileged counterparts, individuals from lower-class backgrounds start having intercourse at a slightly earlier age, use contraception substantially less in their teens and early adulthood, have more teen births, abort less when they have a pregnancy, and are more likely to have unintended pregnancies. After presenting the evidence, we review three theoretical perspectives on what it is about class that leads to the behaviors proximate to early, unplanned pregnancies. The perspectives focus on (1) opportunity costs, (2) social roles, and (3) efficacy and self-regulation. We note that current evidence gives us little ability to adjudicate between these perspectives and recommend that future research focus on such adjudication.

DATA AND METHODS

In addition to reviewing others’ evidence, we provide some illustrative analyses of class differences in the behaviors leading up to early and unintended births. Drawing on three data sets, we present the analyses in a common format. We present class differences in various dependent variables by showing predicted scores for individuals whose mothers have one of three education levels: less than high school, high school, or a four-year college degree. (In one case, when reporting premarital conceptions from the Current Population Survey (CPS), a large national probability sample, we use the respondents’ own education because the data do not include parental education.) In each case, the predicted scores for each education level come from a regression that uses education along with controls for race (and sometimes other variables), with controls set at their mean value.¹

Interactions by race are common in the literature. Although we do not report the results below, we have determined that, for each of the predicted values we present, our basic conclusions about an educational gradient hold for both blacks and whites (although the magnitude of the association may differ somewhat). Thus the reader can be confident that, at least in direction, any class difference we talk about can be found in both the black and the white population (the size of the sample for other groups would not sustain separate analysis). Often, though, there are race differences in levels within class, even when there are no differences in direction in the effects of class by race. These race differences in levels are beyond our topical scope here.

Add Health Data

We use data from the first, second, and third waves of the National Survey of Adolescent Health (Add Health), a nationally representative longitudinal survey of adolescents in grades seven to twelve at the time of the initial interview (Bearman, Jones, and Udry 2004; Chantala 2006).² The sample is school-based: 134 public, private, and parochial schools were selected and a sample of 27,000 adolescents was selected from these schools for extensive in-home interviews. Approximately 21,000 of these students completed the in-home interviews. Their parents were also interviewed at home. To collect sensitive information more accurately, students used an audio-computer-assisted self-interview device (audio-CASI) for several sections of the interview. The initial (Wave I) interviews were collected during the 1994–95 academic year and the second interviews (Wave II, the first follow-up) were collected about a year later. The third wave of interviews (Wave III, the second follow-up) was conducted in 2001–02.

In this analysis, all explanatory variables except the respondent’s age and college status are measured at the initial interview (Wave I, average age about 14.5). Age at first sexual intercourse is measured in whichever interview the respondent first reports having had sexual intercourse (if ever—about 12 percent of respondents were still virgins in Wave III). Sexual activity and contraception use are measured in both the Wave II and Wave III interviews (at average ages of approximately 16 at Wave II and 21.5 at Wave III), and all other outcomes are measured at Wave III, when the respondents provide a full history of their lifetime sexual and romantic relationships, pregnancies, and births. Although we analyze age of first sex for both males and females (using sex-specific models), we restrict analyses of contraception, pregnancy, and abortion to women because men may not always know when their partners are using contraception (particularly hormonal contraception) and may not know about pregnancies with casual partners and/or pregnancies that ended in abortion.

We describe the measures below:³

Mother’s education: This is our key independent variable. We have collapsed the more detailed categories of the original variable to classify mothers of respondents as having not completed high school, having completed high school or some college, or being college graduates. Their education is taken from the parent home interview (usually with the mother) at Wave I, and if not available there, it is taken from the respondent’s report at Wave I.

Enrolled in or graduated from college: An alternative independent variable indicating respondents’ own prospective class is whether, at Wave III, when respondents averaged about 21.5 years of age, they were either still enrolled in college or had graduated, as opposed to having never enrolled or quit. This was measured based on questions in Wave III about current enrollment and educational attainment. These results are presented in Appendix Table 1.A.2.

High school grade-point average: Another alternative independent variable indicating prospective class is the youth’s self-reported grade-point average at Wave I (age about 14.5). These results are presented in Appendix Table 1.A.1.

We have five key dependent variables regarding sex, contraception, and early childbearing:

Age at first intercourse: Because individuals are likely to be most accurate in recalling the timing of recent events, we measure age at first intercourse at whichever interview the respondent first reports having had sexual intercourse. By Wave III, at age averaging about 21.5, about 88 percent of respondents have had sexual intercourse.

Sexual activity and contraceptive consistency: In Wave II and Wave III, we classify respondents as: sexually inactive for the past twelve months (sexually inactive includes virgins); sexually active and “always” using contraception; or sexually active and using contraception less consistently (or never). We use this three-category classification as a dependent variable in our regression models. We use regression-predicted means for these categories to calculate the proportion that is sexually active, and the proportion of the sexually active not always contracepting. We restrict this analysis to women.

Abortion: For women who have been pregnant, this indicates whether the respondent has ever had an abortion (rather than having the pregnancy end in birth or miscarriage). In analysis not shown, we also estimate the probability that a pregnancy ended in abortion (rather than in birth or miscarriage). We conduct this analysis for all pregnancies and, in results not shown, for unintended pregnancies. Both measures include any pregnancies before Wave III. We restrict this analysis to women.

Teen birth: This indicates whether the respondent gave birth to at least one child before age 20. It is taken from reports at all waves. We restrict this analysis to women.

Additional variables entered as controls:

Race: All models control for race. Racial groups are non-Hispanic white, non-Hispanic black, Hispanic, and other. Race is reported by the respondent and is also evaluated independently by the interviewer (relying on physical indications). We use primarily the respondent’s self-report of her or his race, but when it is missing, we use the interviewer’s report.

Age: Age is measured in years and is the respondent’s age on the day of the Wave III interview.

Mother’s age at respondent’s birth: We control for the mother’s age when she gave birth to the respondent. Mother’s age at her first birth (i.e., if the respondent has older siblings) is not known.

Intact family: This measure indicates whether the respondent was living with both biological parents, or with original adoptive parents, at the first interview (at average age about 14.5). Parents are considered original adoptive parents if the respondent was adopted by a two-parent family by age 1.

Individual-level models are estimated using weighted data and adjusting for the stratification and clustered sampling design; not making these adjustments could bias the standard errors (Chantala 2006). Whether a pregnancy ended in abortion, which is measured at the pregnancy level, is clustered by respondent (correcting for the nonindependence of observations). Models estimated with unweighted data and without clustering are also consistent with the models that use weighted data and clusters.

To improve our ability to assess causal effects, all time-varying independent variables (except age) are lagged to the previous wave to ensure that predictive variables are measured temporally prior to outcome variables. However, when the outcome variable is a measure o...