Birth order and risky adolescent behavior.

Argys, Laura M. ; Rees, Daniel I. ; Averett, Susan L. 等

I. INTRODUCTION

There is a widespread belief that birth order is an important determinant of personality, intelligence, and economic success. This belief is supported by a number of recently published popular books, each with its own approach to the topic. Leman (2001, 16-18), for instance, argues that firstborns "are more highly motivated to achieve than later borns" and as a consequence "often fill positions of high authority or achievement." In contrast, Wallace (1999, 18), worries that firstborns "go through life feeling like they cannot measure up to the high standards their parents expected ... lacking in confidence they might drop out and refuse to compete altogether." (1)

Although there is an intuitive appeal to some of these theories, the weight of evidence suggests that birth-order effects of this nature either do not exist or are difficult to measure using the standard approaches of social scientists. A large number of empirical studies have examined the effect of birth order on test scores, education, and earnings--see Olneck and Bills (1979), Blake (1981), Hauser and Sewell (1983), Behrman and Taubman (1986), Kessler (1991), and Hanushek (1992). Taken together their results suggest that, after properly controlling for family size, birth order explains little variation in these conventional measures of success.

Another, more recent strand of research in this area examines whether the sex composition of an individual's siblings affects academic achievement as measured by years of education. In a well-known article, Butcher and Case (1994) present evidence that females raised with brothers go on to receive more education, on average, than that of females raised with at least one sister. However, even this relatively modest, albeit unexpected finding has been called into question by subsequent studies; note Kaestner (1997) and Hauser and Kuo (1998).

Using data on adolescents aged 12 through 17, we investigate the relationship between birth order and a number of behaviors that have been, for the most part, ignored in this literature. These behaviors include the probability that an individual smoked cigarettes, used substances such as marijuana or alcohol, engaged in sexual activity, or committed any one of a series of crimes. In contrast to what most empirical studies have found, we find strong evidence that birth order is related to many of the outcomes under study. We also show that at least one birth-order difference can last into early adulthood, if not beyond. Although it is difficult to isolate the precise routes through which birth order affects the behavior of adolescents, our results are consistent with the idea that peer influences play an important role.

II. WHY BIRTH-ORDER EFFECTS MIGHT EXIST

Researchers working in this area have long speculated that birth order might be related to child outcomes through parental investments in their offspring. Becker and Tomes (1976) posit a model in which parents devote more financial resources to children with lower innate ability. (2) If, as suggested by Behrman and Taubman (1986) and Kessler (1991), genetic endowments favor earlier-born children, this model would predict offsetting financial investments in later-borns. Under the assumption of "imperfect capital markets," Birdsall (1991) predicts greater expenditures on later-born children as family income rises and older siblings become financially independent, but the author also notes that firstborns do not initially have to compete for their parents' time and attention. In a frequently cited piece in the psychology literature, Zajonc (1976) likewise emphasizes the early period in a firstborn's cognitive development when parental attention is undivided. (3) In fact, most researchers working on birth-order issues have assumed that children are particularly sensitive to parental time investments and the home environment at young ages. (4) However, certain nonmonetary parental inputs, such as monitoring and supervision, may become increasingly important as a child matures, especially in the determination of risky or delinquent behaviors.

Sibling interactions represent another possible route through which birth order might be related to child behavior and subsequent achievement. For instance, older siblings could act as positive role models, their achievements adopted as goals and their failures serving as cautionary examples. (5) Older siblings might also act as caregivers or authority figures, especially when one of the parents is absent from the family unit. Alternatively, older siblings may serve as negative role models or even purposely introduce their younger brothers and sisters to certain behaviors earlier than would otherwise be the case. (6) In addition, having an older sibling may provide more opportunities to interact with, and perhaps copy the behavior of, a different set of friends. This could be especially important when the sex of the sibling is different or when there is a substantial age gap between siblings and, as a consequence, the behavior of the older sibling's peers is markedly different from the behavior of the younger sibling's peers; on this point, see Rodgers et al. (1992). (7)

These possibilities suggest that researchers searching for birth-order differences might fruitfully focus attention on adolescent risky behavior as opposed to earnings, education, or test scores. If birth order is indeed related to such behavior, it could have important implications for the health and well-being of adults as well as teens. For instance, if having an older sibling increases the likelihood that teens become sexually active, then birth order could be associated with marriage match quality. Or, if older siblings expose younger family members to the use of alcohol, marijuana, or tobacco, then birth order could lead to long-term health and work-related problems, especially if the substance use persists into adulthood.

III. PAST EMPIRICAL STUDIES

Early empirical studies in this area often confused the effects of family size and birth order. (8) These effects are interrelated because, to use an example offered by Kessler (1991), children from larger families are more likely to be middle borns as opposed to firstborns or last borns. One of the first studies to explicitly address this confusion was that of Olneck and Bills (1979). They examined the effect of birth order and family size on tests scores, education, occupation, and earnings. They found that, after controlling for family size, birth order was no longer a statistically significant predictor of these measures of achievement. However, Olneck and Bills (1979) were unable to include siblingless children in their analysis and entered family size linearly, a functional form assumption that has been criticized as being overly restrictive--see Kessler (1991).

Subsequent researchers corrected these flaws but, by and large, produced similar results. For example, neither Blake (1981), Hauser and Sewell (1983), nor Hanushek (1992) found evidence that birth order was related to educational achievement. Behrman and Taubman (1986) found that, controlling for family size, birth order was unrelated to earnings, although their results did suggest a negative relationship between birth order and years of education for women (but not for men). Kessler (1991) found no effect of birth order on either the level or growth rate of earnings.

More recently, researchers have turned their attention to examining the effects of sibling sex composition on measures of achievement. Perhaps the best-known study in this vein is by Butcher and Case (1994). Using a variety of data sources, these authors show that women raised with only brothers went on to receive significantly more education than women raised with at least one sister. This finding, however, is not easily explained. In fact, some economic models suggest that females raised with brothers should receive less education than those raised with sisters. Moreover, the reliability of Butcher and Case's empirical work has been called into question by subsequent authors. Kaestner (1997), replicated their analysis using data from a more recent cohort but could only find a similar effect for black women. Hauser and Kuo (1998) argued that Butcher and Case's findings were "no more than suggestive" and found little evidence that sex composition affects educational outcomes in an analysis of three nationally representative data sets.

A handful of studies by psychologists have examined the effect of birth order on outcomes similar to those used in this research. Using data from the National Longitudinal Survey of Youth--1979 (NLSY79), Rodgers et al. (1992) examined the effect of birth order and sibling sex composition on age at first intercourse. They found that younger siblings tended to have first intercourse earlier than older siblings, but the authors excluded siblingless children from their calculations as well as individuals who became sexually active after the age of 18. In addition, although Rodgers et al. controlled for family size, they did not include other family background measures in their analysis.

Using data on male college students, Brook et al. (1991) found that the characteristics and actions of older brothers were correlated with the probability that their younger brothers experimented with drugs. This result, however, is subject to at least two interpretations. It is possible that the behavior of older brothers directly influences their younger brothers' drug use, but it is also possible that this correlation can be attributed to difficult-to-measure environmental factors that exert similar influences on the behavior of both siblings. Such factors could include parental attitudes toward substance use or the general level of parental supervision within the family. (9)

IV. DATA AND EMPIRICAL MODEL

Our primary data source is the National Longitudinal Survey of Youth-1997 Cohort (NLSY97). The NLSY97 was launched to enable researchers to define "the transition that today's youths make from school to the labor market and into adulthood." In addition to including schooling and labor market information, the NLSY97 contains detailed questions on family background, personal characteristics, and a variety of behaviors that can be considered risky or delinquent.

Respondents to the NLSY97 have been surveyed on an annual basis beginning in 1997. The initial wave of the NLSY97 consisted of a representative sample of the U.S. population aged 12-16 years on December 31, 1996 (n = 6,748), coupled with a supplemental oversample of black and Hispanic adolescents (n = 2,236). The current study uses data from respondents between the ages of 12 and 17 who were interviewed at least once during the first three waves of the NSLY97. (10) Respondents could contribute up to three observations to the analysis. (11)

Table 1 presents unweighted means for our outcome variables by gender and the presence of an older sibling. Information on tobacco, marijuana, and alcohol consumption; sexual activity; birth control; and criminal/delinquent behaviors was collected using ACASI (Audio Computer Assisted Self-interview) technology. This technology allowed participants to read sensitive questions on a laptop screen or listen to them on headphones and then indicate their responses electronically. It was adopted by the NLSY97 with the goal of minimizing the potential influence of family members and interviewers.

To examine the relationship between birth order and the outcomes listed, we estimate the following baseline model, which focuses on the role of older siblings:

(1) [R.sup.*.sub.i] = [alpha][S.sub.i] [gamma]'[F.sub.i] + [beta]'[X.sub.i]+ [[epsilon].sub.i],

where [R.sup.*.sub.i] is an adolescent's propensity to engage in a particular risky activity, for example, substance use; [S.sub.i] is a dichotomous variable equal to 1 if the adolescent has an older sibling; [F.sub.i] is a vector of controls for family size; and [X.sub.i] represents controls such as age, race, and measures of socioeconomic status. The variable [R.sup.*.sub.i] is latent, but when [R.sup.*.sub.i] > 0, an indicator variable, [R.sub.i], can be seen to equal 1 so that Prob ([R.sub.i] = 1) = Prob([alpha][S.sub.i] + [gamma]'[F.sub.i] + [[beta]'[X.sub.i] + [[epsilon].sub.i] > 0). If the error term of Equation 1 is normally distributed, then the result is a standard univariate probit model.

Our principal interest is in the coefficient of [S.sub.i]. An estimate of [alpha] greater than zero would indicate that children with older siblings--in other words, middle borns and last borns--are more likely to engage in the risky behavior under study than are firstborns, holding family size, age, and the other factors in [X.sub.i] constant. An estimated coefficient less than zero would suggest that middle borns and last borns are less likely to engage in this risky behavior than firstborns.

The presence of an older sibling was ascertained using the NLSY97 household and non-household family rosters. These rosters also provide information on the age and gender of a respondent's siblings. Although the baseline model does not take advantage of this additional information, we incorporate more detailed sibling characteristics when we extend the baseline model below.

The vector [F.sub.i] consists of four dichotomous variables, [F.sub.1] through [F.sub.4], where [F.sub.1] is equal to 1 for siblingless children, [F.sub.2] is equal to 1 for two-child families, and so forth. The omitted category is families with five or more children. Notice that in this specification firstborn and siblingless children are treated as separate categories. Notice too that there are no interactions between birth order and family size in this specification, although we relax this restriction later in our analyses. (12)

The vector [X.sub.i] contains a variety of individual characteristics, such as the respondent's age, race, and ethnicity, as well as socioeconomic indicators (the education level of the youth's most educated parent and a classification of family income relative to the poverty level). The unweighted means of these variables and a continuous measure of family size are reported by gender in Table 2 for the full sample, the subsample consisting of observations contributed by respondents with an older sibling, and the subsample consisting of observation contributed by firstborns and siblingless children.

As expected, we observe clear differences in family characteristics when we compare the subsamples. For example, family size tends to be larger in the older-sibling sampies, whereas parental education tends to be smaller. Because these and other factors may affect the probability that an adolescent engages in risky or delinquent behaviors, it is important that they be included in the vector [X.sub.i] if one is interested in obtaining accurate estimates of the effects of birth order.

V. RESULTS

Determinants of Substance Use

Table 3 reports estimated marginal probabilities and robust standard errors for the baseline probit model in which three outcomes are considered: the probability of adolescents' ever having smoked cigarettes, the probability of their ever having consumed alcohol, and the probability of their ever having used marijuana. (13) Results for males and females are presented separately to allow for gender differences in the relationship between birth order and substance use.

In keeping with published studies by the National Center on Addiction and Substance Abuse (2002) and Pacula et al. (2001), we find strong evidence of race and ethnicity effects. Being black is associated with a lower probability of substance use across all six estimations presented in Table 3, and the marginal effect of the Hispanic indicator is always negative but fails to reach conventional levels of statistical significance in two cases. In addition, we find that adolescents who lived with both of their parents are much less likely to have used tobacco, alcohol, or marijuana, whereas household income seems to have an effect on female, but not male, substance use. Growing up in an urban area is associated with an increase in the probability of having smoked marijuana and, for females only, an increase in the probability of having drunk alcohol.

As noted, the baseline model includes only a single measure of birth order: the dichotomous variable [S.sub.i]. However, this simple specification produces striking results. We find that children with older siblings are, on average, more likely to have used tobacco, alcohol, and marijuana than their firstborn counterparts controlling for family size, age, and the other factors in the model. For instance, females with older siblings are 8 percentage points more likely to have smoked cigarettes than are their firstborn counterparts. Males with older siblings are 6.2 percentage points more likely to have drunk alcohol and are 5.1 percentage points more likely to have tried marijuana than their firstborn counterparts.

A positive relationship between older siblings and substance use is not consistent with the argument that experienced parents provide better guidance for their younger children nor with the notion that older siblings serve as positive role models. This result is more in keeping with the argument that adolescents with older siblings are "prematurely" exposed to behaviors initiated by their older siblings at later ages, although it could reflect differences in parental monitoring and supervision or even a beneficial effect of having younger siblings. (14)

In general, family size, as measured by the number of siblings in the household, seems to have little influence on the outcomes examined. The estimated family-size effects in Table 3 are never significant for males, nor is family size a good predictor of smoking or marijuana use among females. (15) However, there is evidence that family size is negatively related to the probability of female alcohol use. (16)

Table 4 presents estimates of Equation 1 in which an alternative set of substance use measures are employed. Specifically, it examines the relationship between birth order and substance use in the past 30 days. Although not shown, the full set of regressors are included. Again, the results strongly suggest that children with older siblings are more likely than their firstborn counterparts to have used tobacco, alcohol, and marijuana.

Determinants of Sexual Behavior

Table 5 presents estimates of the relationship between birth order and adolescent sexual behavior. Two dependent variables are considered: the first is an indicator of adolescents ever having had intercourse, whereas the second is an indicator of whether they used contraception at first intercourse and it applies only to the subsample of respondents who indicated that they had had sexual intercourse.

The results presented in Table 5 suggest that black males were more likely to have had sex than their white or Hispanic counterparts the opposite is true for Hispanic females. Parental characteristics and socioeconomic status are also important determinants of sexual behavior. Having an older mother is associated with a decrease in the probability of having had intercourse, and living with both parents (as opposed to living in a nontraditional family) sharply reduces this same probability for both sexes. Higher levels of parental education are associated with a reduction in the probability of having had intercourse: an additional year of education is associated with a 0.9 to 1.7 percentage point reduction in this probability.

For both males and females, we see substantial (7.4 and 4.0 percentage points, respectively) increases associated with the presence of an older sibling in the probability of having had intercourse. This result might indicate that middle borns and last borns are at higher risk of experiencing unwanted pregnancies and contracting sexually transmitted diseases. However, there exists the possibility that older siblings could mitigate these negative consequences by introducing their younger siblings to responsible methods of birth control. Using a subsample of respondents who had had intercourse, the second column in Table 5 presents estimates of the effect of birth order on the probability that contraception was used at first intercourse. They provide little evidence that birth order is related to this probability.

In Table 6 we assess the impact of birth order on current sexual behavior, using two alternative outcome measures: whether the respondent was sexually active during the past year and, for sexually active respondents, the probability of using birth control during the past year. Although not shown, the full set of regressors are employed. The results of this exercise are similar to those presented in Table 5 in that they suggest that younger brothers and sisters are more likely to be sexually active than their firstborn counterparts. The results also indicate that having an older sibling decreases the probability that sexually active females used birth control in the past year. There is no evidence of a corresponding effect for male adolescents.

Determinants of Criminal and Delinquent Activities

In Table 7 we examine the relationship between older siblings and the probability that an adolescent ever participated in a variety of illegal or delinquent activities, including carrying a handgun, being a member of a gang, stealing, destroying property, and assault.

Holding other factors constant, blacks and Hispanics are often less likely to report having engaged in these activities than whites, although blacks are more likely to have committed assault than their white counterparts, and black males are more likely to have been a member of a gang. Being raised by two parents is associated with a lower probability of engaging in a number of criminal/delinquent activities, whereas living in an urban area is associated with an increase in the probability of committing assault and increases in the probability that males belonged to a gang, destroyed property, and stole. (17)

Table 7 contains evidence that birth order is related to delinquent activities, but the estimated marginal probabilities are not always significant at conventional levels. We find that males with older siblings are more likely to have stolen and carried a handgun as compared to their firstborn counterparts. Females with older siblings are also more likely to have carried a handgun but are actually less likely to have stolen.

Table 8 examines the determinants of criminal and delinquent behavior in the past year. The results are similar to those reported in Table 7: once again we find that males with older siblings are more likely to have stolen and carried a handgun as compared to their firstborn counterparts and that females with older siblings are more likely to have carried a handgun. The negative relationship between stealing and having an older sibling for females disappears in Table 8, replaced with a positive relationship between destroying property and having an older sibling.

Extensions of the Model

The baseline empirical model only captures differences in behavior between firstborns and younger siblings and does not allow for the possibility, as some theories suggest, that sibling gender composition and age may matter. In this section we address these issues and others by extending the baseline model in a number of ways.

The first extension, labeled Model 1 in Tables 9 and 10, examines the role of older sibling gender. Specifically, we allow the oldersibling effect to differ according to whether the respondent had an older sister or brother by estimating,

(2) [R.sup.*.sub.i] = [[alpha].sub.1][OB.sub.i] + [[alpha].sub.2][0S.sub.i] + [[gamma]'[F.sub.i] + [[beta]'[X.sub.i] + [[epsilon].sub.i],

where [OB.sub.i] is a dichotomous variable equal to 1 if respondent i had an older brother and [OS.sub.i] is a dichotomous variable equal to 1 if respondent i had an older sister. (18) The estimated effect of having an older sister is often larger than that of having an older brother, but the marginal probabilities associated with [OB.sub.i] and [OS.sub.i] are statistically different at the .10 level in only 3 out of 18 estimations, suggesting that sibling sex composition is less important than simple birth order. (19)

The second extension, labeled Model 2 in Tables 9 and 10, tests if the relationship between older siblings and risky/delinquent behavior depends on birth spacing. Specifically, we estimate,

(3) [R.sup.*.sub.i] = [[alpha].sub.1][S3.sub.i] + [[alpha].sub.2][S4.sub.i] + [[gamma]'[F.sub.i] + [[beta]'[X.sub.i] + [[epsilon].sub.i],

where [S3.sub.i] is a dichotomous variable equal to 1 if respondent i had a sibling zero to three years older and [S4.sub.i] is a dichotomous variable equal to 1 if respondent i had a sibling four or more years older. (20) The results indicate that an age gap of four or more years can lead to an increased risk of substance use and sexual activity. In fact, per Table 9 the estimated effect associated with an age gap of four or more years is always significantly greater than that associated with a gap of zero to three years. Some evidence suggests that birth spacing may be important in the determination of female delinquent/criminal behavior, but the marginal probabilities are not significantly different for most of the outcomes examined in Table 10. (21)

In Model 3 we expand our search for birth-order effects by adopting a more flexible specification in which the older-sibling variable is replaced by a set of dichotomous variables indicating if the respondent was second-, third-, fourth-, or fifth-born. (22) The results of this exercise reveal that later-born children are especially prone to engaging in risky and delinquent behaviors. For instance, having been born second is associated with a .09 increase in the probability that males ever smoked, but having been born fifth is associated with a .180 increase in this probability. To take another example, having been born second is associated with a statistically insignificant increase of .025 in the probability that females had sexual intercourse, but having been born fifth is associated with a .133 increase in this probability. (23) It is tempting to view these results as evidence that additional older siblings increase an adolescent's exposure to risky and delinquent behaviors. However, it is possible that early-borns benefit from extra parental supervision when there is a last-born child in the household. (24)

In Model 4 we investigate if the effect of birth order varies systematically according to family size. Specifically, we show the results of running separate regressions by family size, using the birth-order measures from Model 3. The estimated effects from these regressions, although less precise, indicate that middle-born and later-born children are especially prone to engaging in risky and delinquent behaviors across the range of family sizes examined. However, the data provide only limited evidence that birth order and family size interact in a significant fashion. We cannot reject the hypothesis that the effect of having been born second is equal across two-, three-, and four-child families in 14 out of 18 cases. Nor can we reject the hypothesis that the effect of having been born third is equal across three- and four-child families in 12 out of 18 cases. (25)

VI. BIRTH-ORDER DIFFERENCES IN AN EARLIER COHORT: EVIDENCE FROM THE NLSY79

In this section we investigate the relationship between birth order and risky behavior using an alternative data set, the NLSY79. (26) Examining an earlier cohort serves two purposes: first, we can confirm if birth-order differences such as those documented in Table 3 existed for adolescents growing up in the 1970s; second, and perhaps more important, we can explore if these effects typically lasted past adolescence. Evidence of effects on adult behavior would suggest that birth order can have long-run health consequences and point to heretofore overlooked economic consequences.

The upper panel of Table 11 presents estimates of the effects of birth order on substance use and sexual activity for individuals between the ages of 14 and 21 in 1979. Although the results are not as strong those that we found using the NLSY97, it is nevertheless clear that birth order was an important correlate of risky behavior for this earlier cohort. According to the probit estimates presented in Table 11, males with older siblings were 5.6 percentage points more likely to smoke than their firstborn counterparts, and females with older siblings were 4.5 percentage points more likely to smoke. Likewise, the estimated coefficient of the older-sibling variable is significant and positive in the male marijuana and sexual-activity equations as well as in the female alcohol equation. (27)

The bottom panel of Table 11 presents estimates of the relationship between birth order and tobacco, marijuana, and alcohol use in 1992, when the NLSY79 respondents were between the ages of 27 and 34. The data suggest that birth order is not related to the drinking behavior of young adults. However, there is strong evidence that having an older sibling is positively related to the probability that young adults smoke, and some evidence to suggest that males with older sibling were more likely to use marijuana as young adults, although this result is not quite significant at conventional levels.

It is not surprising that birth order seems to have a longer-lasting influence on smoking as compared to that on drinking or marijuana use. Cigarettes are highly addictive, and, as Gruber and Zinman (2001) have shown, most adult smokers begin their habit as adolescents. The relationship between having an older sibling and smoking as an adult suggests that birth order should be related to easily measurable long-run health outcomes such as longevity.

VII. CONCLUSION

A general belief holds that birth order is a key determinant of an individual's personality and attainments. However, academic researchers have typically been unable to link birth order to quantifiable measures of success such as educational attainment and earnings. Here we attempt to measure the relationship between birth order and substance use, sexual behavior, and criminal/delinquent activities during adolescence. Our results provide the strongest empirical evidence to date that birth order is related to behavior.

Our primary data source is the first round of the NLSY97, which provides detailed information on a large sample of adolescents between the ages of 12 and 17. Estimating standard univariate probit models, and controlling for family size and other factors such as parental education and income, we conclude that adolescents with older siblings are more likely to have used tobacco, alcohol, and marijuana, and are more likely to have had sexual intercourse than their firstborn counterparts. In addition, the evidence suggests that male adolescents with older siblings are more likely to steal as compared to firstborns, and female adolescents with older siblings are more likely to destroy property. Having an older sibling is also associated with carrying a gun for both males and females.

A number of plausible explanations exist for these results. It may be that older siblings purposely introduce their younger brothers and sisters to behaviors that they otherwise would not have initiated or that younger siblings are simply mimicking the behavior of their older siblings. Our findings are also consistent with the idea that parents invest less time and energy into supervising their younger offspring, perhaps because of diminishing returns to parenting or because they have less energy to spend on their younger offspring. (28) However, our results are not consistent with the hypothesis that older siblings serve as positive role models or that experience leads to better parenting skills.

To further explore processes through which having an older sibling might be related to behavior, we extend our model to take into account more detailed sibling characteristics. Although we find only limited evidence of sibling sex composition effects, the evidence suggests that birth spacing is important. Having a sibling who was four or more years older is associated with a larger impact on risky behavior than having a sibling who was zero to three years older.

In addition, we find that later-born children are especially prone to engaging in risky and delinquent behaviors as compared to second-born children. For instance, we find that being born second is associated with a .09 increase in the probability that males ever smoked, but being born fifth or higher is associated with a .180 increase.

Finally, we examine the relationship between birth order and risky behavior using an alternative data set, the NLSY79. We find that for individuals who reached adolescence in the 1970s, having an older sibling is associated with an increase in the probability of engaging in risky behavior. Moreover, we find that individuals with older siblings are more likely to smoke as young adults, a result that suggests that birth order can have important healthrelated consequences in the long run.

ABBREVIATIONS

NLSY79: National Longitudinal Survey of Youth-1979

NLSY97: National Longitudinal Survey of Youth-1997

REFERENCES

Argys, L. M., and D. I. Rees. "Searching for Peer Group Effects: A Test of the Contagion Hypothesis." Paper presented at the annual meetings of the Population Association of America, Philadelphia, March 2005.

Becker, G. S., and H. G. Lewis. "On the Interaction between the Quantity and Quality of Children." Journal of Political Economy, 81(2), 1973, S279-88.

Becker, G. S., and N. Tomes. "Child Endowments and Quantity and Quality of Children." Journal of Political Economy, 84(4), 1976, S143-62.

Behrman, J. R., and P. Taubman. "Birth Order, Schooling, and Earnings." Journal of Labor Economics, 4(3), 1986, S121-50.

Birdsall, N. "Birth Order Effects and Time Allocation." Research in Population Economics, 7, 1991, 191-213.

Blake, J. "Family Size and the Quality of the Children." Demography, 18(4), 1981, 421-42.

Brook, J. S., M. Whiteman, D. W. Brook, and A. S. Gordon. "Sibling Influences on Adolescent Drug Use: Older Brothers on Younger Brothers." Journal of the American Academy of Child and Adolescent Psychology, 30(6), 1991, 958-66.

Butcher, K. F., and A. Case. "The Effect of Sibling Sex Composition on Women's Education and Earnings." Quarterly Journal of Economics, 109(3), 1994, 531-63.

Datcher, L. "Effects of Community and Family Background on Achievement." Review of Economics and Statistics, 64(1), 1982, 32-41.

Gruber, J., and J. Zinman. "Youth Smoking in the United States: Evidence and Implications," in Risky Behavior among Youths: An Economic Analysis, edited by J. Gruber. Chicago: University of Chicago Press, 2001.

Hanushek, E. A. "The Trade-Off between Child Quantity and Quality." Journal of Political Economy, 100(1), 1992, 84-117.

Haurin, R. J., and F. L. Mott. "Adolescent Sexual Activity in the Family Context: The Impact of Older Siblings." Demography, 27(4), 1990, 537-57.

Hauser, R., and H-H. D. Kuo. "Does the Gender Composition of Sibships Affect Women's Educational Attainment?" Journal of Human Resources, 33(3), 1998, 644-57.

Hauser, R. M., and W. H. Sewell. "Birth Order and Education Attainments in Full Sibships." Working Paper No. 83-31. Madison: University of Wisconsin, Center for Demography and Ecology, 1983.

Haveman, R., and B. Wolfe. "The Determinants of Children's Attainments: A Review of Methods and Findings." Journal of Economic Literature, 33(4), 1995, 1829-78.

Hill, M., and G. J. Duncan. "Parental Family Income and the Socioeconomic Attainment of Children." Social Science Research, 16(1), 1987, 39-73.

Isaacson, C., and K. Radish. The Birth Order Effect: How to Better Understand Yourself and Others. Avon, MA: Adams Media, 2002.

Kaestner, R. "Are Brothers Really Better? Sibling Sex Composition and Educational Achievement Revisited." Journal of Human Resources, 32(2),1997, 250-84.

Kessler, D. "Birth Order, Family Size, and Achievement: Family Structure, and Wage Determination." Journal of Labor Economics, 9(4), 1991, 413-26.

Leman, K. The Birth Order Book: Why You Are the Way You Are. Grand Rapids, MI: Baker Book House, 2001.

Mocan, N. H., B. Scafidi, and E. Tekin. "Catholic Schools and Bad Behavior." National Bureau of Economic Research Working Paper No. 9172, Cambridge MA, 2002.

National Center on Addiction and Substance Abuse. "Teen Tippers: America's Underage Drinking Epidemic." National Center on Addiction and Substance Abuse, Columbia University, 2002.

Oettinger, G. S. "Sibling Similarity in High School Graduation Outcomes: Causal Interdependency or Unobserved Heterogeneity?" Southern Economic Journal 66(3), 2000, 631-48.

Olneck, M. R., and D. B. Bills. "Family Configuration and Achievement: Effects of Birth Order and Family Size in a Sample of Brothers." Social Psychology Quarterly, 42(2), 1979, 135-48.

Ouyang, L. "Sibling Effects on Teen Risky Behaviors." Working Paper, Duke University, Department of Economics, 2004.

Pacula, R. L., M. Grossman, F. J. Chaloupka, P. M. O'Malley, L. D. Johnston, and M. C. Farrelly. "Marijuana and Youth," in Risky Behavior among Youths: An Economic Analysis, edited by Jonathan Gruber. Chicago: University of Chicago Press, 2001.

Rodgers, J. L., D. C. Rowe, and D. F. Harris. "Sibling Differences in Adolescent Sexual Behavior: Inferring Process Models from Family Composition Patterns." Journal of the Marriage and the Family, 54, 1992, 142-52.

Slomkowski, C., R. Rende, S. Novak, and E. Lloyd-Richardson. "Sibling Effects on Smoking in Adolescence: Evidence for Social Influence from Genetically Informative Design." Addiction, 100(4), 2005, 430-38.

Sulloway, F. J. Born to Rebel. Birth Order, Family Dynamics and Creative Live. New York: Vintage, 1997.

Tyas, S. L., and L. L. Pederson. "Psychosocial Factors Related to Adolescent Smoking: A Critical Review of the Literature." Tobacco Control 7, 1998, 409-20.

Wallace, M. Birth Order Blues: How Parents Can Help Their Children Meet the Challenges of Birth Order. New York: Henry Holt, 1999.

Wang, M. Q., E. C. Fitzhugh, R. C. Westerfield, and J. M. Eddy. "Family and Peer Influences on Smoking Behavior among American Adolescents: An Age Trend." Journal of Adolescent Health, 16(3), 1995, 200-203.

Widmer, E. D. "Influence of Older Siblings on Initiation of Sexual Intercourse." Journal of Marriage and the Family, 59(4), 1997, 928-38.

Zajonc, R. B. "Family Configuration and Intelligence." Science, 16, 1976, 227-36.

Zax, J., and D. I. Rees. "IQ, Academic Performance, Environment and Earnings." Review of Economics and Statistics, 84(4), 2002, 600-616.

Zick, C. D., and W. K. Bryant. "A New Look at Parents' Time Spent in Child Care: Primary and Secondary Time Use." Social Science Research, 25(3), 1996, 260-80.

(1.) Other popular books in this area include those by Isaacson and Radish (2002) and Sulloway (1997). Sulloway takes a more historical approach and offers evidence that birth order affects political power and scientific innovation, but the author does not investigate the effect of birth order on wages or education.

(2.) The Becker and Tomes (1976) model was an extension of the well-known child quantity-quality tradeoff model proposed by Becker and Lewis (1973), who assumed an equal level of financial investment in each child within a family.

(3.) Zajonc (1976) proposed thinking of the "intellectual environment" of the family, in which a child is raised as the average of the "absolute intellectual levels of [all of] its members" (227). When a new child is born, this average falls (the absolute intellectual level of a newborn can be thought of as zero), and the new child's intellectual environment is therefore less stimulating than that enjoyed by the older siblings. According to this argument, older siblings will usually be more intelligent and score higher on standard tests of achievement such as the SAT than will their younger brothers and sisters.

(4.) Behrman and Taubman (1986) write that "the oldest child has some periods, particularly during presumably critical early years, when he or she has less competition for mother's time" (S124, italics added). When discussing the impact of the home environment, Zajonc (1976) explicitly notes that "the later these influences occur in an individual's life, the smaller is their effect" (228).

(5.) Sociologists and developmental psychologists such as Haveman and Wolfe (1995) and Rodgers et al. (1992) have emphasized the importance of role models, including older siblings, in the determination of aspirations and behavioral norms of children and adolescents.

(6.) Data from the NLSY97 show a strong positive link between risky behaviors and age among adolescents living in the United States in the late 1990s. For instance, the probability of marijuana use increases, on average, by .034 for every year of adolescence. To take another example, the probability that an adolescent is sexually active increases, on average, by. 123 per year, according to Argys and Rees (2005). This pattern of results suggests that older siblings may "prematurely" expose their younger brothers and sisters to certain risky behaviors.

(7.) If a substantial age gap exists, older siblings may be more likely to inherit monitoring responsibilities from their parents. This could serve to magnify the effect of having an older sibling, whether positive or negative.

(8.) Haveman and Wolfe (1995) list a number of articles that find a negative relationship between educational outcomes and family size. The evidence with regard to earnings is less strong. See, for instance, Datcher (1982), Hill and Duncan (1987), and Zax and Rees (2002).

(9.) A number of studies have in fact examined if the behavior or achievements of an older sibling influence the younger sibling's behavior--see Haurin and Mott (1990), Wang et al. (1995), Widmer (1997), and Slomkowski et al. (2005)--but, to our knowledge, only Oettinger (2000) attempted to address what he terms the "unobserved heterogeneity" problem: "Of course, since the behavior of the family unit simultaneously influences the achievement of all of the children in the family and since many family factors are unobservable [to the researcher], a sibling's achievement should be treated as an endogenous explanatory variable" (632). Oettinger found that, even controlling for unobserved heterogeneity, having an older sibling who graduated from high school increased the probability that the younger sibling would go on to graduate. Oettinger argues that successful older siblings could act as positive role models "by revealing information about a youth's own potential ... or through other mechanisms" (632). He also suggests that "the achievement of siblings might have spillover effects if learning is to some extent a public good within the family unit" (632).

(10.) The NLSY97 questions pertaining to sexual activity and contraceptive use were asked only of respondents ages 14 and older.

(11.) The initial three waves of the NSLY97 took place in 1997, 1998, and 1999. By the fourth wave, almost 60% of the NSLY97 respondents were at least 18 years old. Because we restrict our sample to participants between the ages of 12 and 17, and as a result of natural attrition from the NLSY97, many respondents contributed fewer than three observations to the analysis. For instance, a total of 4,317 respondents contributed an average of 2.28 observations to the male marijuana estimation presented in Table 3. About half of these (50.4%) were observed in each of the three waves of the NLSY97 analyzed. A total of 4,082 respondents contributed an average of 2.28 observations to the female marijuana estimation presented in Table 3. Of these, 50.6% were observed in each of the three waves.

(12.) Kessler (1991), proposes an empirical model that allowed for interactions between family size and birth order.

(13.) These and all other results in the article are based on unweighted data.

(14.) It is also possible that this pattern of results reflects differences in the reporting of behavior as opposed to differences in actual behavior. However, as noted, the NLSY97 collected sensitive information using ACASI (Audio Computer Assisted Self-interview) technology in order to minimize measurement error.

(15.) A number of other studies confirm that family size is not a good predictor of adolescent smoking or substance use--see, for instance, Tyas and Pederson (1998), Mocan et al. (2002), and Ouyang (2004).

(16.) For instance, being raised as an only child (as compared to being raised in a family of five or more children) is associated with a 15.1 percentage-point increase in the probability that a female respondent reported ever drinking, whereas being raised in a four-child family is associated with a 7.6 percentage-point increase in this same probability.

(17.) Living in an urban area is also associated with a decrease in the probability that male respondents carried a handgun and in the probability that female respondents destroyed property or stole.

(18.) Of middle-born and last-born respondents, 25% had both an older brother and an older sister; 36% had an older sister but no older brother; 39% had an older brother but no older sister.

(19.) Older sisters have a significantly larger impact than that of older brothers on the probability that male adolescents ever smoked, drank alcohol, or belonged to a gang.

(20.) Of middle-born and last-born respondents, 20% had a sibling in both of these age categories; 42% had a sibling in the first category but not in the second; 38% had a sibling in the second but not in the first.

(21.) In Table 10, we find evidence that birth spacing matters in the female assault and gang equations. A number of plausible explanations are available for why the sibling age gap might be related to risk-taking behavior; however, it is difficult to imagine why it might be related to a willingness to report such behavior. Thus, we view the results from Model 2 as evidence against the hypothesis that the birth-order effects documented in Tables 3 to 6 simply reflect differences in reporting.

(22.) This last category includes respondents who had five or more older siblings.

(23.) In Table 9, the effect of having been fourth or fifth is significantly greater than the effect of having been born second in 6 out of 8 cases. In Table 10, the effect of having been born fourth or fifth is significantly greater than the effect of having been born second in only 4 out of 10 cases.

(24.) Studies of two-parent, two-child households have shown that the age of the youngest child is an important determinant of the time that parents devote to child care in general--see, for instance, Zick and Bryant (1996).

(25.) Although we did not show them here, we also estimated models in which the dependent variable was transformed in order to reflect whether the respondent engaged in a particular behavior by a certain age. We found that having an older sibling was associated with large increases in the probability of smoking, drinking, using marijuana, and engagin in sexual activity by the age of 15. However, when we examined the relationship between birth order and engaging in risky behavior by the age of 17, the results were less strong. Having an older sibling was associated with an increase in the probability that males used marijuana and were sexual active by the age of 17, and was associated with an increase in the probability of smoking for both males and females. The other estimated marginal effects were not significant at conventional levels (although they were always positive).

(26.) The NLSY79 is a precursor to the primary data source used in this investigation. It provides information on more than 12,000 individuals who were aged 14 to 21 in 1979, including an oversampling of blacks, Hispanics, and economically disadvantaged whites.

(27.) We also investigated the effect of birth order on criminal activities such as assault and robbery. Although not shown, our results suggest that there was no effect of birth order on criminal activity for this cohort.

(28.) Another possibility is that our findings reflect an interaction between sibling and parental influences. For example, parents could determine the optimal level of supervision for their children based on the assumption that sibling behavior is contagious. If this assumption is incorrect, then parents run the risk of setting overly restrictive rules (e.g., an 8:00 PM curfew) for their earlier-born offspring in a misguided attempt to shield their later-born offspring. On the other hand, if sibling behavior is in fact contagious, then it would make sense for parents to concentrate their supervision efforts on earlier-born offspring in order to take advantage of positive behavioral spillovers.

LAURA M. ARGYS, DANIEL I. REES, SUSAN L. AVERETT, and BENJAMA WITOONCHART, We would like to acknowledge support from the National Institute of Child Health and Human Development, contract number HD047661. The views expressed in this article are those of the authors and do not necessarily reflect the views of the National Institute of Child Health and Human Development.

Argys: Associate Professor, University of Colorado at Denver, Department of Economics, Campus Box 181, Denver, CO 80217-3364. Phone 1-303-556-3949, Fax 1-303-556-3547, E-mail Laura.Argys@cudenver.edu

Rees: Associate Professor, University of Colorado at Denver, Department of Economics, Campus Box 181, Denver, CO 80217-3364. Phone 1-303-556-3348, Fax 1-303-556-3547, E-mail drees@carbon.cudenver.edu

Averett: Professor, Lafayette College, Department of Economics and Business, Easton, PA 18042-1776. Phone 1-617-330-5307, Fax 1-610-330-5715, E-mail averetts@lafvax.lafayette.edu

Witoonchart: PhD Candidate, University of Colorado at Boulder, Department of Economics, 256 UCB, Boulder, Colorado 80309-0256. E-mail benjama. witoonchart@colorado.edu

TABLE 1
Unweighted Means of Dependent Variables by Gender and Presence of an
Older Sibling

 Males

 Has an Has no
 Full Older Older
 Sample Sibling Sibling

Substance Use
Ever smoked cigarettes 0.447 0.471 0.415
Smoked cigarettes in past 30 days 0.242 0.251 0.231
Ever drank alcohol 0.530 0.539 0.519
Drank alcohol in past 30 days 0.293 0.300 0.283
Ever smoked marijuana 0.267 0.271 0.261
Smoked marijuana in past 30 days 0.129 0.133 0.125
Sexual Activity
Ever had sexual intercourse 0.400 0.415 0.381
Sexually active in the past year 0.328 0.341 0.313
Used contraception at first intercourse 0.808 0.801 0.818
(if ever had intercourse)
Used contraception within the past year 0.868 0.867 0.869
(if sexually active)
Delinquent Activities
Ever carried a handgun 0.195 0.198 0.190
Carried a handgun in the past year 0.092 0.096 0.086
Ever assaulted anyone 0.281 0.281 0.281
Assaulted someone in the past year 0.075 0.073 0.078
Ever belonged to a gang 0.091 0.092 0.089
Was a gang member in the past year 0.036 0.036 0.036
Ever destroyed property 0.398 0.392 0.405
Destroyed property in the past year 0.198 0.194 0.203
Ever stole anything 0.419 0.422 0.416
Stole items worth more than $50 in 0.086 0.093 0.078
the past year

 Females

 Has an Has no
 Full Older Older
 Sample Sibling Sibling

Substance Use
Ever smoked cigarettes 0.435 0.443 0.425
Smoked cigarettes in past 30 days 0.227 0.234 0.218
Ever drank alcohol 0.516 0.510 0.524
Drank alcohol in past 30 days 0.280 0.282 0.277
Ever smoked marijuana 0.230 0.229 0.232
Smoked marijuana in past 30 days 0.098 0.098 0.100
Sexual Activity
Ever had sexual intercourse 0.336 0.326 0.349
Sexually active in the past year 0.281 0.273 0.291
Used contraception at first intercourse 0.786 0.796 0.774
(if ever had intercourse)
Used contraception within the past year 0.888 0.869 0.910
(if sexually active)
Delinquent Activities
Ever carried a handgun 0.043 0.048 0.036
Carried a handgun in the past year 0.018 0.020 0.015
Ever assaulted anyone 0.162 0.165 0.159
Assaulted someone in the past year 0.045 0.045 0.045
Ever belonged to a gang 0.044 0.046 0.041
Was a gang member in the past year 0.013 0.012 0.013
Ever destroyed property 0.220 0.228 0.209
Destroyed property in the past year 0.095 0.099 0.090
Ever stole anything 0.312 0.310 0.315
Stole items worth more than $50 in 0.035 0.035 0.035
the past year

TABLE 2
Unweighted Means of the Explanatory Variables by Gender and
Presence of an Older Sibling

 Males

 Full Has an Has no
 Sample Older Sibling Older Sibling

Has an Older Sibling 0.559 1 0
Black 0.260 0.260 0.260
Other race 0.137 0.149 0.122
Hispanic 0.199 0.219 0.174
Age 15.10 15.04 15.18
 (1.43) (1.44) (1.41)
Lives with both parents 0.542 0.585 0.488
Mother's age at respondent's 25.62 26.97 23.91
birth (5.24) (5.07) (4.95)
Highest parent's education 13.20 12.95 13.52
 (3.01) (3.16) (2.77)
Income below 125% of the 0.106 0.104 0.109
poverty level
Income above 400% of the 0.076 0.069 0.084
poverty level
Number of siblings 2.211 2.695 1.597
 (1.60) (1.64) (1.31)
Number of older siblings 0.97 1.74 0
 (1.20) (1.12)
Lives in urban area 0.717 0.716 0.718
Sample size 9859 5514 4345

 Females

 Full Has an Has no
 Sample Older Sibling Older Sibling

Has an Older Sibling 0.563 1 0
Black 0.266 0.273 0.257
Other race 0.140 0.157 0.117
Hispanic 0.205 0.228 0.175
Age 15.11 15.04 15.19
 (1.43) (1.44) (1.42)
Lives with both parents 0.517 0.569 0.451
Mother's age at respondent's 25.49 26.89 23.69
birth (5.41) (5.32) (4.99)
Highest parent's education 13.12 12.88 13.43
 (3.02) (3.11) (2.87)
Income below 125% of the 0.117 0.115 0.120
poverty level
Income above 400% of the 0.077 0.068 0.088
poverty level
Number of siblings 2.235 2.758 1.563
 (1.73) (1.83) (1.31)
Number of older siblings 0.97 1.72 0
 (1.23) (1.17)
Lives in urban area 0.729 0.733 0.724
Sample size 9328 5247 4081

Note: Standard deviations of continuous variables in parentheses.

TABLE 3
Determinants of Substance Use

 Males

 Ever Smoked Ever Drank Ever Smoked
 Cigarettes Alcohol Marijuana

Black -0.172 ** -0.202 ** -0.066 **
 (0.019) (0.019) (0.016)
Other race 0.002 -0.026 0.038
 (0.028) (0.028) (0.024)
Hispanic -0.081 ** -0.040 -0.035
 (0.025) (0.026) (0.020)
Age 12 -0.057 * -0.092 ** -0.078 **
 (0.028) (0.029) (0.026)
Age 14 0.126 ** 0.116 ** 0.163 **
 (0.021) (0.021) (0.025)
Age 15 0.210 ** 0.228 ** 0.254 **
 (0.019) (0.018) (0.023)
Age 16 0.287 ** 0.321 ** 0.350 **
 (0.021) (0.019) (0.025)
Age 17 0.360 ** 0.384 ** 0.426 **
 (0.024) (0.020) (0.028)
Lives with both parents -0.127 ** -0.094 ** -0.101 **
 (0.017) (0.017) (0.014)
Mother's age at birth of -0.003 -0.002 -0.002
 respondent (0.002) (0.002) (0.001)
Parents' highest education -0.009 ** 0.001 -0.0004
 level (0.003) (0.003) (0.003)
Income < 125% of poverty 0.014 -0.005 0.013
 (0.022) (0.021) (0.019)
Income < 400% of poverty 0.022 0.035 -0.001
 (0.024) (0.023) (0.021)
Lives in an urban area -0.022 0.027 0.033 *
 (0.019) (0.018) (0.015)
Only child 0.042 0.003 0.032
 (0.035) (0.034) (0.031)
Two-child family -0.001 0.038 0.017
 (0.025) (0.024) (0.021)
Three-child family -0.019 0.018 -0.00001
 (0.024) (0.023) (0.020)
Four-child family -0.005 -0.006 -0.002
 (0.026) (0.025) (0.021)
Has an older sibling 0.105 ** 0.062 ** 0.051 **
 (0.017) (0.017) (0.014)
Sample size 9859 9852 9843

 Females

 Ever Smoked Ever Drank Ever Smoked
 Cigarettes Alcohol Marijuana

Black -0.247 ** -0.193 ** -0.145 **
 (0.019) (0.020) (0.013)
Other race -0.019 -0.026 -0.016
 (0.028) (0.029) (0.022)
Hispanic -0.119 ** -0.039 -0.038
 (0.025) (0.026) (0.019)
Age 12 -0.110 ** -0.137 ** 0.107 **
 (0.029) (0.032) (0.042)
Age 14 0.119 ** 0.164 ** 0.259 **
 (0.022) (0.021) (0.038)
Age 15 0.199 ** 0.297 ** 0.372 **
 (0.020) (0.019) (0.038)
Age 16 0.264 ** 0.375 ** 0.427 **
 (0.022) (0.019) (0.040)
Age 17 0.306 ** 0.428 ** 0.498 **
 (0.026) (0.019) (0.041)
Lives with both parents -0.150 ** -0.130 ** -0.126 **
 (0.018) (0.017) (0.013)
Mother's age at birth of -0.002 0.001 -0.001
 respondent (0.002) (0.002) (0.001)
Parents' highest education -0.0003 0.0053 0.0017
 level (0.003) (0.003) (0.002)
Income < 125% of poverty -0.010 -0.050 * 0.001
 (0.022) (0.023) (0.019)
Income < 400% of poverty -0.085 ** -0.059 * -0.041 *
 (0.023) (0.024) (0.018)
Lives in an urban area 0.003 0.043 * 0.052 **
 (0.020) (0.019) (0.014)
Only child 0.063 0.151 ** 0.035
 (0.037) (0.033) (0.029)
Two-child family 0.040 0.125 ** 0.029
 (0.027) (0.026) (0.021)
Three-child family 0.026 0.101 ** 0.031
 (0.026) (0.025) (0.021)
Four-child family 0.015 0.076 ** -0.006
 (0.028) (0.027) (0.021)
Has an older sibling 0.080 ** 0.055 ** 0.041 **
 (0.018) (0.017) (0.013)
Sample size 9328 9323 9319

Note: Marginal probabilities from a univariate probit model are
reported. Standard errors corrected for clustering at the family level
are in parentheses. Specifications include controls for survey year and
missing race/ethnicity, parental education, interview year, and family
income.

** p < .01; * p < .05; (^) p < .10.

TABLE 4
The Effect of an Older Sibling on Substance Use within the Past 30 Days

 Males

 Smoked Drank Smoked
 Cigarettes in Alcohol in Marijuana in
 Past 30 Days Past 30 Days Past 30 Days

Has an Older Sibling 0.055 ** 0.056 ** 0.029 **
 (0.013) (0.013) (0.009)
Sample size 9031 9051 9340

 Females

 Smoked Smoked
 Cigarettes in Drank Alcohol Marijuana in
 Past 30 Days in Past 30 Days Past 30 Days

Has an Older Sibling 0.063 ** 0.033 * 0.018 *
 (0.013) (0.013) (0.008)
Sample size 8595 8728 8913

Note: Marginal probabilities from a univariate probit model are
reported. Standard errors corrected for clustering at the family level
are in parentheses. In addition to the independent variables used in
Table 3, specifications include controls for survey year and missing
race/ethnicity, parental education, interview year, and family income.

** p < .01; * p < .05.

TABLE 5
Determinants of Sexual Behavior

 Males

 Ever had Sexual Used Birth Control
 Intercourse at First Intercourse

Black 0.241 ** (0.020) 0.041 (^) (0.022)
Other race 0.042 (0.028) 0.010 (0.035)
Hispanic 0.005 (0.027) 0.017 (0.031)
Age 15 0.104 ** (0.027) -0.024 (0.040)
Age 16 0.269 ** (0.024) -0.034 (0.036)
Age 17 0.405 ** (0.025) -0.045 (0.039)
Lives with both parents -0.136 ** (0.017) -0.004 (0.021)
Mother's age at birth of
 respondent -0.005 ** (0.002) 0.003 (0.002)
Parents' highest education -0.017 ** (0.003) 0.007 (0.004)
Income < 125% of poverty 0.044 (^) (0.027) -0.016 (0.030)
Income < 400% of poverty -0.014 (0.030) -0.005 (0.040)
Lives in an urban area 0.044 * (0.018) 0.007 (0.024)
Only child 0.025 (0.037) -0.005 (0.046)
Two-child family 0.001 (0.025) 0.014 (0.028)
Three-child family -0.011 (0.024) -0.01 (0.028)
Four-child family -0.026 (0.025) 0.027 (0.028)
Has an older sibling 0.074 ** (0.018) -0.026 (0.022)
Sample size 7291 2899

 Females

 Ever had Sexual Used Birth Control
 Intercourse at First Intercourse

Black 0.034 (^) (0.020) 0.098 ** (0.025)
Other race 0.010 (0.027) -0.063 (0.047)
Hispanic -0.070 ** (0.024) -0.028 (0.040)
Age 15 0.130 ** (0.027) -0.010 (0.020)
Age 16 0.277 ** (0.025) 0.001 (0.021)
Age 17 0.400 ** (0.027) 0.00001 (0.026)
Lives with both parents -0.148 ** (0.016) -0.002 (0.002)
Mother's age at birth of
 respondent -0.006 ** (0.002) -0.204 ** (0.067)
Parents' highest education -0.009 ** (0.003) -0.042 (0.035)
Income < 125% of poverty 0.040 (0.026) 0.047 (0.041)
Income < 400% of poverty -0.057 * (0.027) 0.034 (0.029)
Lives in an urban area -0.010 (0.018) -0.033 (0.059)
Only child 0.088 * (0.036) 0.053 (0.033)
Two-child family 0.038 (0.025) 0.039 (0.033)
Three-child family 0.022 (0.024) -0.008 (0.036)
Four-child family 0.015 (0.026) 0.043 (0.026)
Has an older sibling 0.040 * (0.017) 0.011 (0.023)
Sample size 6920 2313

Note: Marginal probabilities from a univariate probit model are
reported. Standard errors corrected for clustering at the family level
are in parentheses. Specifications include controls for survey year and
missing race/ethnicity, parental education, interview year, and family
income.

** p < .01; * p < .05; (^) p < .10.

TABLE 6
The Effect of an Older Sibling on Sexual Behavior in the Past Year

 Males

 Used Birth
 Had Sexual Control Within
 Intercourse the Past Year
 in Past Year (if Sexually Active)

Has an older sibling 0.071 ** 0.005
 (0.017) (0.018)
Sample size 6566 2193

 Females

 Used Birth
 Had Sexual Control Within
 Intercourse the Past Year
 in Past Year (if Sexually Active)

Has an older sibling 0.042 ** -0.031 (^)
 (0.016) (0.017)
Sample size 6423 1857

Note: Marginal probabilities from a univariate probit model are
reported. Standard errors corrected for clustering at the family level
are in parentheses. In addition to the independent variables used in
Table 3, specifications include controls for survey year and missing
race/ethnicity, parental education, interview year, and family income.

** p < .01; (^) p < .10.

TABLE 7
The Determinants of Criminal and Delinquent Activities

 Males

 Ever Ever Ever
 Engaged Belonged Carried a
 in Assault to a Gang Handgun

Black 0.033 (^) 0.028 ** -0.071 **
 (0.018) (0.011) (0.014)
Other Race -0.013 0.012 -0.032
 (0.024) (0.014) (0.019)
Hispanic -0.041 (^) 0.020 -0.028
 (0.021) (0.014) (0.018)
Age 12 0.034 -0.020 0.012
 (0.027) (0.014) (0.023)
Age 14 0.070 ** 0.021 0.040 *
 (0.022) (0.014) (0.019)
Age 15 0.086 ** 0.041 ** 0.069 **
 (0.024) (0.014) (0.017)
Age 16 0.093 ** 0.064 ** 0.084 **
 (0.028) (0.016) (0.020)
Age 17 0.103 ** 0.062 ** 0.123 **
 (0.031) (0.020) (0.026)
Lives with both parents -0.092 ** -0.041 ** -0.028 *
 (0.015) (0.009) (0.013)
Mother's age at birth -0.002 -0.001 -0.002
of respondent (0.002) (0.001) (0.001)
Parents' highest -0.010 ** -0.004 ** -0.004
education level (0.003) (0.001) (0.002)
Income < 125% of poverty -0.004 0.013 0.033 *
 (0.018) (0.011) (0.017)
Income < 400% of poverty -0.018 0.005 -0.019
 (0.020) (0.013) (0.017)
Lives in an urban area 0.031 * 0.019 * -0.045 **
 (0.016) (0.009) (0.015)
Only child -0.032 -0.032 * -0.002
 (0.029) (0.012) (0.027)
Two-child family -0.014 -0.028 * -0.008
 (0.021) (0.011) (0.019)
Three-child family -0.014 -0.006 -0.0108
 (0.021) (0.011) (0.018)
Four-child family -0.031 -0.029 * -0.026
 (0.021) (0.010) (0.018)
Has an older sibling 0.012 0.001 0.023
 (0.016) (0.009) (0.013)
Sample size 9876 9850 9849

 Males

 Ever Ever
 Destroyed Stolen
 Property Anything

Black -0.101 ** -0.105 **
 (0.019) (0.020)
Other Race 0.008 -0.003
 (0.027) (0.028)
Hispanic -0.080 ** -0.049 (^)
 (0.024) (0.025)
Age 12 0.032 -0.030
 (0.027) (0.026)
Age 14 0.063 ** 0.079 **
 (0.022) (0.021)
Age 15 0.081 ** 0.100 **
 (0.024) (0.019)
Age 16 0.103 ** 0.145 **
 (0.028) (0.022)
Age 17 0.107 ** 0.165 **
 (0.031) (0.027)
Lives with both parents -0.066 ** -0.102 **
 (0.017) (0.017)
Mother's age at birth -0.005 ** -0.001
of respondent (0.002) (0.002)
Parents' highest 0.003 0.001
education level (0.003) (0.003)
Income < 125% of poverty -0.013 -0.011
 (0.021) (0.021)
Income < 400% of poverty 0.001 0.010
 (0.022) (0.023)
Lives in an urban area 0.050 ** 0.068 **
 (0.018) (0.018)
Only child -0.024 -0.022
 (0.034) (0.034)
Two-child family -0.038 -0.031
 (0.025) (0.025)
Three-child family -0.022 -0.019
 (0.024) (0.024)
Four-child family -0.030 -0.039
 (0.025) (0.025)
Has an older sibling 0.011 0.030
 (0.017) (0.017)
Sample size 9877 9878

 Females

 Ever Ever Ever
 Engaged Belonged Carried a
 in Assault to a Gang Handgun

Black 0.041 ** -0.007 -0.013 *
 (0.016) (0.007) (0.006)
Other Race 0.040 (^) 0.020 * -0.003
 (0.022) (0.012) (0.008)
Hispanic -0.028 0.006 -0.002
 (0.017) (0.009) (0.008)
Age 12 0.012 0.020 0.021
 (0.024) (0.018) (0.020)
Age 14 0.034 (^) 0.029 * 0.050 **
 (0.019) (0.016) (0.021)
Age 15 0.067 ** 0.032 * 0.057 **
 (0.022) (0.016) (0.021)
Age 16 0.060 * 0.030 (^) 0.059 **
 (0.025) (0.018) (0.023)
Age 17 0.062 * 0.031 (^) 0.061 **
 (0.027) (0.020) (0.026)
Lives with both parents -0.079 ** -0.034 ** -0.025 **
 (0.012) (0.007) (0.006)
Mother's age at birth -0.002 (^) -0.001 (^) 0.0005
of respondent (0.001) (0.001) (0.001)
Parents' highest -0.004 * -0.002 * -0.001
education level (0.002) (0.001) (0.001)
Income < 125% of poverty 0.013 -0.001 0.002
 (0.016) (0.008) (0.007)
Income < 400% of poverty -0.020 -0.004 -0.016
 (0.017) (0.009) (0.007)
Lives in an urban area 0.026 * -0.003 0.007
 (0.013) (0.007) (0.006)
Only child -0.034 -0.011 0.023
 (0.021) (0.010) (0.016)
Two-child family -0.025 0.001 0.010
 (0.017) (0.009) (0.009)
Three-child family -0.0007 0.012 0.003
 (0.016) (0.010) (0.008)
Four-child family -0.021 0.0003 0.014
 (0.017) (0.009) (0.010)
Has an older sibling 0.013 0.007 0.016 *
 (0.012) (0.006) (0.006)
Sample size 9337 9325 9328

 Females

 Ever Ever
 Destroyed Stolen
 Property Anything

Black -0.057 ** -0.069 **
 (0.016) (0.019)
Other Race 0.010 0.064 *
 (0.024) (0.028)
Hispanic -0.051 * -0.038
 (0.020) (0.024)
Age 12 0.443 0.347
 (0.284) (0.281)
Age 14 0.054 * -0.016
 (0.026) (0.027)
Age 15 0.033 (^) 0.058 **
 (0.020) (0.021)
Age 16 0.042 (^) 0.092 **
 (0.023) (0.019)
Age 17 0.024 0.118 **
 (0.026) (0.022)
Lives with both parents 0.012 0.129 **
 (0.028) (0.028)
Mother's age at birth -0.088 ** -0.092 **
of respondent (0.015) (0.016)
Parents' highest 0.086 * 0.012
education level (0.039) (0.041)
Income < 125% of poverty 0.004 0.008 **
 (0.003) (0.003)
Income < 400% of poverty -0.018 -0.043 *
 (0.017) (0.019)
Lives in an urban area -0.028 -0.042 *
 (0.015) (0.017)
Only child 0.024 0.005
 (0.035) (0.039)
Two-child family -0.021 -0.006
 (0.028) (0.033)
Three-child family -0.010 -0.001
 (0.021) (0.024)
Four-child family -0.003 -0.017
 (0.021) (0.023)
Has an older sibling -0.029 -0.051 *
 (0.021) (0.024)
Sample size 9341 9342

Note: Marginal probabilities from a univariate probit model are
reported. Standard errors corrected for clustering at the family level
are in parentheses. Specifications include controls for survey year
and missing race/ethnicity, parental education, interview year, and
family income.

** p < .01; * p < .05; (^) p <.10.

TABLE 8
The Effect of an Older Sibling on Criminal and Delinquent
Activities in the Past Year

 Males

 Ever Ever Ever Ever Ever
 Engaged Belonged Carried a Destroyed Stolen
 in Assault to a Gang Handgun Property Anything

Has an -0.004 -0.001 0.017 * 0.013 0.016 (^)
older
 sibling (0.007) (0.004) (0.007) (0.012) (0.008)
Sample
 Size 8085 9843 9845 8356 6484

 Females

 Ever Ever Ever Ever Ever
 Engaged Belonged Carried a Destroyed Stolen
 in Assault to a Gang Handgun Property Anything

Has an 0.005 -0.001 0.005 (^) 0.017 (^) 0.006
older
 sibling (0.005) (0.002) (0.003) (0.009) (0.005)
Sample
 Size 8404 9322 9326 8455 6766

Note: Marginal probabilities from a univariate probit model are
reported. Standard errors corrected for clustering at the family level
are in parentheses. In addition to the independent variables used in
Table 3, specifications include controls for survey year and missing
race/ethnicity, parental education, interview year, and family income.

* p < .05; (^) p < .10.

TABLE 9
Extensions to the Baseline Model: Substance Use and Sexual Behavior

 Males

 Ever Ever Ever had
 Smoked Drank Ever used Sexual
 Cigarettes Alcohol Marijuana Intercourse

Model 1

Has an older brother 0.052 ** 0.018 0.039 ** 0.041 *
 (0.017) (0.017) (0.014) (0.018)
Has an older sister 0.097 ** 0.064 ** 0.054 ** 0.064 **
 (0.018) (0.018) (0.015) (0.019)
Sample size 9859 9852 9843 7291

Model 2

Has an older sibling 0.062 ** 0.008 0.021 0.027
within 3 years (0.016) (0.016) (0.013) (0.017)
Has an older sibling 4 0.111 ** 0.096 ** 0.054 ** 0.091 **
or more years older (0.018) (0.018) (0.016) (0.019)
Sample size 9859 9852 9843 7291

Model 3

Second-born 0.090 ** 0.049 ** 0.038 * 0.062 **
 (0.018) (0.018) (0.015) (0.019)
Third-born 0.148 ** 0.095 ** 0.096 ** 0.099 **
 (0.026) (0.024) (0.024) (0.027)
Fourth-born 0.154 ** 0.096 ** 0.080 * 0.125 **
 (0.038) (0.036) (0.036) (0.041)
Fifth-born (or higher) 0.180 ** 0.103 * 0.134 ** 0.223 **
 (0.048) (0.047) (0.051) (0.052)
Sample size 9859 9852 9843 7291

Model 4

Two-child families
Second-born 0.068 * 0.048 0.035 0.072 *
 (0.028) (0.028) (0.024) (0.029)
Sample size 2902 2901 2902 2169
Three-child families
Second-born 0.083 ** 0.051 0.023 0.029
 (0.032) (0.032) (0.026) (0.033)
Third-born 0.216 ** 0.176 ** 0.126 ** 0.154 **
 (0.040) (0.036) (0.036) (0.042)
Sample size 2826 2827 2826 2057
Four-child families
Second-born 0.091 * 0.026 0.050 0.142 **
 (0.046) (0.044) (0.038) (0.048)
Third-born 0.156 ** 0.034 0.113 * 0.102
 (0.053) (0.051) (0.047) (0.054)
Fourth-born 0.194 ** 0.134 * 0.154 ** 0.198 **
 (0.062) (0.060) (0.059) (0.069)
Sample size 1737 1734 1730 1271

 Females

 Ever Ever Ever had
 Smoked Drank Ever used Sexual
 Cigarettes Alcohol Marijuana Intercourse

Model 1

Has an older brother 0.059 ** 0.034 (^) 0.037 * 0.056 **
 (0.019) (0.018) (0.015) (0.018)
Has an older sister 0.084 ** 0.053 ** 0.039 ** 0.040 *
 (0.018) (0.018) (0.014) (0.017)
Sample size 9328 9323 9319 6920

Model 2

Has an older sibling 0.029 -0.0005 0.018 -0.006
within 3 years (0.017) (0.016) (0.013) (0.016)
Has an older sibling 4 0.110 ** 0.075 ** 0.071 ** 0.090 **
or more years older (0.019) (0.019) (0.016) (0.018)
Sample size 9328 9323 9319 6920

Model 3

Second-born 0.075 ** 0.051 ** 0.036 * 0.025
 (0.019) (0.018) (0.015) (0.018)
Third-born 0.093 ** 0.063 * 0.058 ** 0.070 **
 (0.028) (0.027) (0.023) (0.027)
Fourth-born 0.101 * 0.081 * 0.088 * 0.162 **
 (0.042) (0.040) (0.038) (0.044)
Fifth-born (or higher) 0.192 ** 0.077 0.117 ** 0.133 **
 (0.050) (0.048) (0.049) (0.052)
Sample size 9328 9323 9319 6920

Model 4

Two-child families
Second-born 0.071 * 0.081 ** 0.007 0.023
 (0.029) (0.029) (0.022) (0.028)
Sample size 2845 2842 2841 2098
Three-child families
Second-born 0.069 * 0.050 0.071 ** 0.030
 (0.033) (0.031) (0.028) (0.031)
Third-born 0.104 * 0.098 * 0.080 * 0.097 *
 (0.041) (0.039) (0.036) (0.040)
Sample size 2656 2656 2653 1973
Four-child families
Second-born 0.098 * -0.007 0.017 -0.036
 (0.048) (0.045) (0.034) (0.047)
Third-born 0.077 -0.045 0.013 0.013
 (0.060) (0.056) (0.040) (0.059)
Fourth-born 0.092 -0.009 0.044 0.153 *
 (0.074) (0.069) (0.054) (0.075)
Sample size 1406 1406 1404 1047

Note: Marginal probabilities from a univariate probit model are
reported. Standard errors corrected for clustering at the family level
are in parentheses. Sample size in brackets. In addition to the
independent variables used in Table 3, specifications include controls
for survey year and missing race/ethnicity, parental education, and
family income.

** p < .01; * p <. 05; (^) p < .10.

TABLE 10
Extensions to the Baseline Model: Criminal and Delinquent Activities

 Males

 Ever Ever Ever
 Engaged Belonged Carries
 in Assault to a Gang a Handgun

Model 1
Has an older brother 0.001 -0.004 0.019
 (0.015) (0.008) (0.013)
Has an older sister 0.033 * 0.019 * 0.009
 (0.016) (0.009) (0.014)
Sample size 9876 9850 9849
Model 2
Has an older sibling 0.011 -0.002 0.017
within 3 years (0.015) (0.008) (0.013)
Has an older sibling 4 0.024 0.006 0.011
or more years older (0.017) (0.009) (0.014)
Sample size 9876 9850 9849
Model 3
Second-born 0.009 -0.002 0.030 *
 (0.017) (0.009) (0.015)
Third-born 0.016 0.003 0.001
 (0.024) (0.013) (0.020)
Fourth-born 0.015 0.018 0.007
 (0.034) (0.021) (0.029)
Fifth-born (or higher) 0.117 * 0.025 0.065 (^)
 (0.050) (0.028) (0.042)
Sample size 9876 9850 9849
Model 4
Two-child families
Second-born -0.013 -0.002 -0.018
 (0.026) (0.012) (0.022)
 2910 2901 2905
Three-child families
Second-born 0.004 -0.004 0.047 (^)
 (0.029) (0.017) (0.026)
Third-born 0.030 0.013 0.035
 (0.037) (0.022) (0.032)
Sample size 2835 2823 2820
Four-child families
Second-born 0.042 -0.007 0.048
 (0.041) (0.020) (0.034)
Third-born 0.016 0.011 -0.021
 (0.047) (0.026) (0.038)
Fourth-born 0.040 0.036 0.063
 (0.056) (0.035) (0.050)
Sample size 1736 1734 1735

 Males

 Ever Ever
 Destroyed Stolen
 Property Anything

Model 1
Has an older brother 0.009 0.033 (^)
 (0.017) (0.017)
Has an older sister 0.002 0.024
 (0.018) (0.018)
Sample size 9877 9878
Model 2
Has an older sibling 0.019 0.028 (^)
within 3 years (0.016) (0.016)
Has an older sibling 4 0.012 0.049 **
or more years older (0.018) (0.019)
Sample size 9877 9878
Model 3
Second-born 0.003 0.009
 (0.018) (0.018)
Third-born 0.037 0.079 **
 (0.026) (0.026)
Fourth-born 0.012 0.106 **
 (0.038) (0.038)
Fifth-born (or higher) 0.023 0.125 *
 (0.051) (0.051)
Sample size 9877 9878
Model 4
Two-child families
Second-born 0.024 0.003
 (0.029) (0.029)
 2911 2911
Three-child families
Second-born -0.026 -0.026
 (0.032) (0.032)
Third-born 0.090 * 0.124 **
 (0.040) (0.039)
Sample size 2833 2835
Four-child families
Second-born -0.022 0.077
 (0.044) (0.044)
Third-born -0.038 0.043
 (0.050) (0.052)
Fourth-born -0.071 0.067
 (0.059) (0.063)
Sample size 1737 1737

 Females

 Ever Ever Ever
 Engaged Belonged Carries
 in Assault to a Gang a Handgun

Model 1
Has an older brother -0.0002 0.015 * 0.018 **
 (0.013) (0.007) (0.007)
Has an older sister 0.021 (^) 0.010 (^) 0.010
 (0.013) (0.006) (0.007)
Sample size 9337 9325 9328
Model 2
Has an older sibling 0.0004 0.003 0.008
within 3 years (0.012) (0.006) (0.006)
Has an older sibling 4 0.046 ** 0.022 ** 0.013 *
or more years older (0.014) (0.008) (0.007)
Sample size 9337 9325 9328
Model 3
Second-born 0.011 0.006 0.014 *
 (0.013) (0.006) (0.007)
Third-born 0.009 0.011 0.033 **
 (0.019) (0.011) (0.014)
Fourth-born 0.057 (^) 0.022 0.020
 (0.034) (0.019) (0.021)
Fifth-born (or higher) 0.071 * 0.057 * 0.047 *
 (0.039) (0.032) (0.030)
Sample size 9337 9325 9328
Model 4
Two-child families
Second-born 0.020 -0.0004 0.011
 (0.018) (0.007) (0.009)
 2847 2843 2845
Three-child families
Second-born -0.006 0.010 0.019 (^)
 (0.024) (0.013) (0.012)
Third-born 0.005 0.013 0.033 *
 (0.030) (0.017) (0.018)
Sample size 2661 2658 2521
Four-child families
Second-born 0.010 0.018 0.029
 (0.031) (0.015) (0.022)
Third-born -0.0001 0.015 0.045
 (0.040) (0.020) (0.030)
Fourth-born 0.075 0.027 0.049
 (0.058) (0.029) (0.045)
Sample size 1408 1312 1406

 Females

 Ever Ever
 Destroyed Stolen
 Property Anything

Model 1
Has an older brother 0.032 * 0.023
 (0.015) (0.017)
Has an older sister 0.026 (^) -0.0003
 (0.016) (0.017)
Sample size 236 276
Model 2
Has an older sibling 0.033 * 0.018
within 3 years (0.015) (0.016)
Has an older sibling 4 0.039 * 0.027
or more years older (0.016) (0.018)
Sample size 9341 9342
Model 3
Second-born 0.020 0.021
 (0.016) (0.017)
Third-born 0.072 ** 0.024
 (0.026) (0.026)
Fourth-born 0.064 (^) 0.020
 (0.039) (0.039)
Fifth-born (or higher) 0.046 -0.039
 (0.046) (0.043)
Sample size 9341 9342
Model 4
Two-child families
Second-born -0.0003 0.010
 (0.024) (0.028)
 2848 2847
Three-child families
Second-born 0.041 0.035
 (0.029) (0.032)
Third-born 0.064 0.032
 (0.037) (0.039)
Sample size 2665 2664
Four-child families
Second-born -0.0002 0.036
 (0.039) (0.042)
Third-born 0.106 * 0.028
 (0.054) (0.052)
Fourth-born 0.033 -0.039
 (0.063) (0.059)
Sample size 1407 1408

Note: Marginal probabilities from a univariate probit model are
reported. Standard errors corrected for clustering at the family level
are in parentheses. In addition to the independent variables used in
Table 3, specifications include controls for survey year and missing
race/ethnicity, parental education, and family income.

** p < .01; * p < .05; (^) p < .10.

TABLE 11
Estimated Effects of Having an Older Sibling:
National Longitudinal Survey of
Youth--1979 Cohort

 Males Females

1979 interview ages 14-21
Ever smoked cigarettes 0.056 ** 0.045 **
 (0.015) (0.016)
Ever drank alcohol 0.009 0.032 *
 (0.013) (0.014)
Ever smoked marijuana 0.040 * 0.024
 (0.018) (0.016)
Ever had sex 0.027 * -0.011
 (0.013) (0.015)
1992 interview ages 27-34
Smoked cigarettes daily 0.034 (^) 0.054 **
 (0.020) (0.019)
Drank alcohol daily -0.016 -0.013
 (0.015) (0.017)
Drank >5 drinks daily 0.006 0.003
 (0.004) (0.003)
Smoked marijuana in 0.022 0.007
the past year (0.015) (0.011)
Smoke marijuana in 0.006 0.001
the past month (0.012) (0.008)

Note: Marginal probabilities from a univariate probit
model are reported. Standard errors corrected for clustering
at the family level are in parentheses. In addition to the
independent variables used in Table 3, specifications include
controls for survey year and missing race/ethnicity,
parental education, and family income.

** p < .01; * p < .05; (^) p < .10.