Pirog, Jung, and Lee’s article “The Changing Face of Teenage Parenthood in the United States: Evidence from the NLSY79 and NLSY97” focuses on an important and timely topic: social changes in teenage family forming behaviors over the last two decades of the twentieth century (2017). The paper builds on extensive research on changes in teenage fertility (e.g., Vinovskis 2003; Santelli and Melnikas 2010) and tracks teen parents’ educational and labor market outcomes until age 29 to determine whether the predictors or outcomes of teen parenthood changed over the 20 year study period. This commentary builds from Pirog and colleagues’ work by highlighting some of the manuscript’s contributions to family science, describing limitations, and proposing alternate interpretations of key findings. Four issues in particular are explored: the appropriateness of the data and methods, the overlooked declining risk of teenage motherhood, the possible demographic explanations for increased risk of teenage fatherhood, and the value of identifying changes in teenage family formation patterns.

Many of the contributions of this article relate to the technical strength of Pirog et al.’s approach for assessing changes in teen parenthood (2017). In particular, the choice of data sets (NLSY79 and NLSY97) was well suited to identify changes in teenage childbearing, particularly among men. As noted by Joyner et al. (2012), there is a long history of men’s childbearing being underreported or undercounted in surveys, which has led to significant concerns about studying paternity with national samples. To this end, Joyner et al. (2012) evaluated the quality of men’s fertility data in the two NLSY cohorts by comparing men’s and women’s reported childbearing with results from Vital Statistics and the U.S. Census Bureau. The authors identified several features of the NLSY surveys that have bearing on the quality of the results presented here. First, the NLSY cohorts have relatively complete reports of childbearing, with about 97–99% of women’s early births and about 90% of men’s early births reported. The NSLY men’s data is strong relative to other national samples; compared to the NSFG, for example, the NLSY data captured roughly 10% more fathering behaviors. Past research has found that between one-third to one half of men’s nonmarital births and births within previous marriages were left out of male fertility estimates when retrospective reports were used (Rendall et al. 1999). In the NLSY, respondents were interviewed as teenagers and prospective reports of teen births were heavily relied upon for the estimates used in this paper. When retrospective reports were used because births occurred before the baseline survey, the reports were within a relatively short time from the actual event. Given this background information, we can expect that the results provided here will be a relatively accurate reflection of national rates of teenage childbearing over this time. Second, the NLSY is particularly strong in its assessment of male fertility relative to other national surveys due to its sample design that captures a greater number of non-civilian, non-institutionalized youth. Because men enter the study as teenagers and are followed over time regardless of later institutionalization or military service, the outcomes at age 29 are likely to provide a more complete picture of the consequences of teen childbearing for American men than other national samples, like the NSFG (see Joyner et al. 2012). Third, the two NLSY studies are about equal in their ability to identify early male fertility, with the caveat being that a slightly greater proportion of nonmarital births were reported among the 1997 cohort. Given the overall finding that fertility rates are reliably assessed between cohorts, the difference-in-difference approach to modeling has face validity because readers can be relatively confident that what is being measured in 1979 is the same as what is being measured in 1997.

Like childbearing, retrospective reports of cohabitation are problematic for data accuracy (Hayford and Morgan 2008), and the authors utilize the strengths of the NLSY survey design to minimize error and maximize comparability of relationship measures in both cohorts. At first blush, finding comparable measures for each cohort is difficult, given the first cohabitation questions weren’t asked of NLSY79 participants until 1990, almost a decade after the respondents were teen parents. Conversely, in the later born cohort, NLSY97 participants were asked prospective marriage and cohabitation questions in each survey wave. To create comparable measures for each cohort, the authors overlooked direct assessments and instead constructed cohabitation histories from household rosters assessed in each year for each cohort. This approach allowed the authors to overcome a major limitation of retrospective reporting with cohabitation and provided a more effective and comparable prospective measure for the two groups. Similarly, the authors identified five common family of origin indicators in both cohorts, assessed as family status at ages 14 (1979 cohort) and 12 (1997 cohort). Given prior research indicating family of origin structure is a marker of access to resources and structural supports (Amato et al. 2008), and that daughters with a history of instability are more likely to face instability in their own adult relationships (Amato and Kane 2011) and have children earlier in life (Manning and Cohen 2015), developing a comparable measure of family of origin is important when studying cross-cohort differences in childbearing and family formation.

Finally, the article is situated in a U.S. policy framework that makes the research better suited for translation, a critical feature for work that is so closely tied to government programming. The authors catalog state policies regarding abortion and contraception use which afforded the more recently born cohort a wider range of options for preventing or terminating unwanted pregnancies, potentially impacting teen birth rates. They also describe how youth from the more recent cohort were subject to the 1996 Personal Responsibility and Work Opportunity Reconciliation Act (PRWORA) that discouraged single motherhood, marriage, and cohabitation by requiring teenage mothers to live with their parents or an approved adult to receive benefits, potentially impacting residential partnering. Although the authors did not test the effect of these policies on teen childbearing and cohabitation directly, nor did they spend serious time revisiting these issues in the conclusion, nonetheless they offered a clear and convincing rationale early on for why family forming behaviors might differ based on the historical policy contexts. This approach had several payoffs: it provided a mechanism for explaining why teen partnering and childbearing might be changing between cohorts, and it was a clear example of how researchers might better communicate the value of their work to practitioners and policy makers by situating hypotheses within the policy context. Even within America’s highly polarized political system, there is growing and broad bipartisan support to promote evidence based policy making (e.g., Report of the Commission on Evidence Based Policy Making 2017) and reduce the gap between academic publications and the incorporation of results into programming and policies. Describing how the political context shapes one’s approach to research is a useful first step in breaking down barriers between the academy and policy makers (Freise and Bogenschnieder 2009; Oliver et al. 2012).

Along with the many strengths of the paper were several important limitations that have the potential to impact comparability between samples, limit the scope, or obscure the study’s findings. First, the authors’ measure of total family income at baseline is error prone. In the NLSY97, a parent of the respondent completed an interview at baseline in which they provided details on the gross total family income. For the older cohort, the authors utilized the 1979 baseline measure of family income, which from 1979 to 1986 reflected an amalgam of household income questions, including data on a parent’s report of income for the family of origin if the respondent was living in the parent’s household (survey version A) or the youth’s report of income for themselves and potentially their partner if they were at a temporary address or living on their own (survey version B or C, respectively). When the 1979 baseline report is used, the comparability of early family economic status across the two cohorts is muddled because household income may or may not reflect parental income, depending on who answered the survey item, and this is particularly important in a study considering cohabiting teen parents who would be answering this question for themselves and their partner. Although a survey was not conducted with all of the parents of 1979 respondents, a household screener in 1978 was completed by a household head and assessed total family income in order to determine whether the respondent would be part of the poverty oversample. Utilizing the 1978 household screener with the 1997 parent report would provide the authors a more optimal comparison of total family-of-origin income and would reduce bias in the model. I agree with the authors that it is valuable for family scholars to consider these measures in their models of teen fertility, as indicated by the burgeoning literature regarding selection into parenting and childbearing experiences, and how exposure to early life (dis)advantage may be linked to long term family inequalities (e.g., McLanahan 2004; McLanahan and Jacobsen 2015). Furthermore, the developing work on cohabitation among teen girls finds that, unlike older women, disadvantaged teens are less likely to cohabit as adolescents, perhaps because of the PRWORA restrictions on living with a parent or guardian to receive benefits (Manning and Cohen 2015). This suggests that the family of origin income question is not only important as a reference to structural resources available to the teenager which has been tied to higher rates of teen childbearing and early family formation (Amato et al. 2008) but may also be confounded with the likelihood of cohabiting or marrying the father (Manning and Cohen 2015). Thus, family of origin may be proxying for distinct, and potentially age-specific, family forming choices and so its measurement should be as error free as possible in order to better understand these key associations.

A second limitation regards the analytic weights used in this study. According to the NLSY97 online documentation (2017), authors are encouraged to use weights when creating descriptive statistics such as means, medians and standard deviations, however “if you are running a more complex analysis such as doing a regression, we suggest that you do not weight…(as this) may lead to incorrect estimates” (User Note #1, para. 1). When doing descriptive analyses with weights, the ‘gold standard’ of weighting is to replicate what the authors did here, and create a customized longitudinal weight, following the recommendations of the Bureau of Labor Statistics. It is also reasonable to use an existing off-the-shelf weight by selecting the weight for the terminal year in the longitudinal measure (NSLY 2017). Unfortunately, the authors chose to weight their difference-in-difference models with the same customized longitudinal weight, which goes against the recommendations of the data providers. The authors are not alone in this approach, and given other scholars have used NLS weights with multivariate models, I reached out to Jay Zagorsky, an NSL weighting expert, to discuss recommended protocol. He noted “The NLS has been consistent in its advice not to run weighted complex analysis”, but if authors choose to disregard this advice, he suggested running the difference-in-difference model three times: first on a cross-sectional sample without weighting to get an idea of the correct sign and magnitude of effects, then again with the full sample without weighting to improve the precision of standard errors and coefficients, and a third time with the full model and full sample weighted to test whether the results are in-line with expectations (personal communication, November 7, 2017). Another common way data users have addressed survey design without weighting complex analyses is to include indicators related to the oversampling of the surveys in their models; the weights correct for other things as well, but the primary reason to use weights is that NLS oversampled particular groups of people. In a cross-cohort comparison of the 1979 and 1997 surveys, for example, including indicators of race/ethnicity and the poverty screener in multivariate models help control for the oversample characteristics that may bias results to disproportionately reflect the experience of these groups.

The third, and most significant, concern with the manuscript is the key findings on women’s teen birth rates were misstated in the abstract, results, and discussion, wherein the authors suggest that teen paternity is increasing while teenage maternity is held constant. Given the goal of the paper to understand how gendered fertility rates are changing over time, the mischaracterization of results is important to correct. As noted in the Table 3 output and written results for the table, compared to the 1979 cohort, the more recently born 1997 cohort has 7.9% higher rates of teen fatherhood and 12.1% lower rates of teen motherhood, net of family structure, racial and ethnic background, parental education, and family income at baseline. Although equally significant (p < .05) and taken from the same table as the fatherhood results, the authors were virtually silent in regards to the declines in teen maternity and repeatedly mischaracterized the findings as demonstrating stability over time. Based on the table output, however, the rate of change was greater for women and in the opposite direction of the findings for men. Further, the basic thrust of the findings in the Pirog et al. (2017) manuscript are reinforced in prior research by Joyner et al. (2012) on this sample and with Vital Statistics and Census Data for the same years (note that the Pirog et al. article uses a subsample of the NLSY cohort who was 15–17 at baseline while the Joyner piece considers all teen childbearing). Even with the variability in approach due to distinctions in the sample selection, weighting, and analytic strategies, both manuscripts find a decline in women’s teenage fertility over the 18-year window of about 12–13% and a corresponding increase of between 5 and 8% in teenage paternity. In support of the Pirog et al. (2017) approach, the similarity of results with prior published work on the NLSY, Vital Statistics, and Census data indicates robust findings, even in light of the previously mentioned limitations. On the other hand, the gendered divergence necessitates exploring reasons behind both men’s and women’s fertility experiences rather than focusing exclusively on men’s changing sexual patterns (e.g., the cougar hypothesis) or changes in measurement.

Perhaps the simplest (and overlooked) explanation for the divergent gendered findings can be seen in an analysis of the long-term demographic trends (see Fig. 1) in men and women’s teen birth rates. It is generally assumed in our regression models that the changes between two time points reflect a linear change in behavior. This is a roughly accurate understanding of teen birth rates over the last 75-years, which witnessed a 59% decline in teenage childbearing in the United States from 1940 to 2015, and a 77% decline since the high point of teen fertility in the baby boom of 1957 (author’s calculations). This is not the case, however, for teen fertility over the study period. From the baby boom until the first survey, there was an average decline of 3% per year in teen fertility (Ventura et al. 2001). In contrast, during the time between the first and second NLSY cohort surveys, teen birth rates saw their first sustained increase in over 30 years, leading to a sharp 24% uptick in teen childbearing from 1986 to 1991 and then a renewed and sustained decline of about 2.7% per year from 1991 until 2000 (Ventura et al. 2001), when the youngest of the NLSY97 cohort were at risk for becoming teen parents. Given the rapid increase and then decline in birth rates between the studies, it is possible that both men and women were undergoing a similar, rather than divergent, process of change following an inverted U-shape pattern, but they may have been at different places on the downward curve because women transitioned more quickly than men to lower fertility. Figure 1 depicts the rate of teen childbearing in the United States from 1940 until 2015 (descending line with hollow circles representing each data point) and the percent of nonmarital teenage childbearing over this same period (ascending line with tick marks indicating each data point). Dashed vertical lines reflect the baseline years of the NLSY79 and NLSY97 surveys. The rates displayed follow the protocol set by The National Center for Health Statistics, which tracks the U.S. teenage childbearing as the birth rate of the population “at risk” of giving birth, that is, female teenagers aged 15–19 years old (Ventura et al. 2001, 2014).

Fig. 1
figure 1

Source: Birth rate: CDC/NCHS. National Vital Statistics System, U.S. birth rates by age of mother, 1940–2015. Retrieved from https://blogs.cdc.gov/nchs-data-visualization/us-natality-trends/. Percent unmarried: Ventura et al. (2014)

Teenage birth rate and percent of births to unmarried teens: 1940–2015.

Full size image

A second unexplored, but relatively simple explanation for why men in the newer cohort may be more likely to be teen fathers, is that women are choosing to partner with similarly aged peers, rather than the Pirog et al. (2017) suggestion that men are choosing to partner with older women (e.g., the cougar hypothesis). The age of teen partners is not assessed in this study, but we know from prior work that the trends in men’s and women’s fertility follow a similar pattern over time, although the rate of teen paternity is much lower overall. For example, 15–19 year old women in both cohorts are between 2 and 3 times more likely to be teen parents compared to similarly aged men, as depicted in Pirog et al.’s Fig. 1 (2017). In the absence of additional evidence, it seems equally, or more plausible to interpret the Pirog et al. (2017) findings as suggestive that though female teenagers are having fewer children overall, when they do, they are more likely to be having their children with similar aged peers compared to the past. There is nothing in the data presented here or in the literature more generally that leads me to agree that the cougar hypothesis is the most likely or reasonable explanation for the divergent gender findings. Further, if the hypothesis were upheld that women were partnering with similarly-aged peers, the second explanation for increases in male fertility described by Pirog et al. (2017) related to measurement error may be an overstatement. However, the alternate hypothesis of more teens having children together would be consistent with the Joyner et al. (2012) finding that men in both cohorts report roughly equivalent quality data regarding being a teen father, compared to Census and Vital Statistics data for the same years. Future work that carefully considers the age of teen childbearing partners is needed to determine which of these explanations is best supported by the data.

Who are the unmarried teenagers having children? Pirog et al. (2017) noted that over the course of the two studies, the makeup of the teen childbearing population became more concentrated around a smaller proportion of women who were raised by teen mothers and chose cohabitation over marriage. This finding is in line with what is known about family formation among teenagers more generally. Over the study period, there was a consistent and linear increase in the percentage of teen births to unwed women, from 46.1 to 77.8% (see Fig. 1). The change reflects a move away from unwed childbearing being a less common experience—less than half of teenage childbearing was nonmarital in 1979—to the normative experience—over three-quarters of teenage childbearing was nonmarital in 1997. Furthermore, this increase was part of a long-term and consistent trend toward a greater share of teen births being to unwed women, from 13.6% in 1940 to 88.7% in 2013. It is important to note, however, that though the percent of births to unmarried teens increased substantially, the birth rate among unmarried teens was quite low relative to married teens during this period (see Fig. 2). Around the time of the second cohort study, for example, the birth rate for married teenagers was roughly eight times greater than the birth rate for unmarried teenagers. Although there has been a significant convergence in birth rates by marital status from 1940 to 2013, the experience of married versus unmarried teens remained distinct over this period, reflecting the fact that most unmarried teenagers do not have children and a greater number of married teenagers do. This is likely due to the family forming decisions teenagers make after having a fertility experience: teenage who get pregnant have much greater odds of marrying (986%) or cohabiting (446%) than teenagers who do not have a conception event (Manning and Cohen 2015). And though the vast majority of teenage mothers remain single before the birth of their first child (with 22% cohabiting and 5% marrying), by the child’s third birthday, two-thirds of women form a union (59% cohabiting, 9% marrying, and 5% doing both), most often with someone other than the birth father (Manning and Cohen 2015). And among all men and women who marry as teens, 44% cohabit first, making cohabitation a common route to marriage, even in adolescence (Manning and Cohen 2015).

Fig. 2
figure 2

Source: Ventura et al. (2014)

Teenage birth rate by marital status.

Full size image

The Pirog et al.’s (2017) findings regarding changes in cohabitation over the two periods demonstrate a significant advance in what is currently known about teen parent cohabitation and later relationship trajectories. The author’s report in their Fig. 2, that the more recently born cohort is less likely to be married or cohabiting at the time of their child’s birth and they are more likely to be married or cohabiting 5 years after their fertility experience compared to the older born cohort. This delayed onset of partnering is important to study in greater detail, as prior work suggests that partnering at the time of birth tends to be with the biological parent of the child, but by year three it is more likely to be someone new (Manning and Cohen 2015). Building from these findings, future work is needed to verify whether more recently born children of teen parents in the NLSY cohorts are less likely to experience living in a home with both biological parents, as this is tied to long-term social and economic outcomes (McLanahan 2004). Furthermore, recent scholarship has begun to consider how teen childbearing and cohabitation influence the parent’s later family forming behaviors, such as Eshbaugh’s (2008) study of disadvantaged parents in Head Start and Manning and Cohen’s (2015) research on a nationally representative sample of teenagers in the NSFG. These studies have provided important first steps at understanding how teenage fertility is associated with relationship choices, especially marriage and cohabitation, but they have not provided longitudinal evidence of how these patterns might be changing over time, or how these patterns might differ for men. Because the likelihood of marrying or cohabiting increases with conception, and cohabitation is becoming the most common context for teenagers raising a young child (Manning and Cohen 2015), exploring how these events have changed provides important understandings into what it means to be a teen parent today and have implications for policy efforts aimed at reducing teenage fertility. Additionally, the family forming behaviors of adolescents set teens on a pathway for later relationship and childbearing activities (Manning et al. 2008; Meier and Allen 2008), and it is important to quantify what these relationships are and how they are changing over time in order to better contextualize research on adult family formation patterns, family inequities, and overall parent and child wellbeing.