The remainder of this paper is organized as follows

The remainder of this paper is organized as follows. Section 2 describes the econometric model. Section 3 explains the sources and samples. Section 4 reports and comments on the estimation results. Section 5 discusses the lifetime impact of teenage motherhood. Section 6 concludes.
Empirical model Assume that for each teenager i the impact of teenage motherhood (TM) on her education and labor market outcomes (Y) is obtained from the linear equation:where Y is the outcome for a teenage woman i who was born in state s and belongs to a particular age cohort c; TM is an indicator for early childbearing (1 if she Monastrol had a child before the age of 17; 0 otherwise); X is a set of controls including past socioeconomic status, labor market conditions (wage of unskilled women, child labor), and development indicators (infant mortality, poverty and inequality); I denotes public investment in education and health; E(ɛ∣TM, X, I)=μ+μ+μ+E(ξ TM, X, I) is the mean error term assumed as an additive function of state of birth, cohort and individual specific factors with a possibly not zero component E(ξ TM, X, I); and α is the parameter of interest. We allow for fixed effects of cohorts and state of birth to affect the outcomes through μ and μ. The indicator TM, however, can still be correlated with the woman\'s unobserved heterogeneity μ. To account for this, we construct a pseudo panel with observed means by state of birth and cohort using the PNAD household survey. The PNAD is a time series of independent cross-sections; most control variables obtained from external data which can be found at the state level but not at the individual level. We use the PNAD to build a pseudo panel of states of birth and cohorts. This methodology closely follows the approach used by Meghir and Whitehouse (1996). The following equation was estimated considering cohort and state of birth: We estimate Eq. (2) using weights determined by the number N of observations (individuals) per cell. We use White\'s consistent standard errors in order to deal with the problem of heteroskedasticity in grouped data.
Data sources and sample The main source of data is the Brazilian Household Survey, the 1981–2004 PNADs. The data are a series of independent cross-sections of individuals. Each year, the national sample comprises around 300,000 individuals who are interviewed across all of the country\'s 26 states, covering both urban and rural areas. This survey encompasses demographic and current socioeconomic information including data on education and labor market status for individuals aged 10 or more. From 1992 onwards, the PNAD has included which state the individual was born in. We match these data as follows. For women aged 23–30 who were born in state X and year 1992, we use the information on control variables from the PNAD in state X and year 1981, that is 11 years earlier, when these women were in their teenage years 12–19. Similarly, for women aged 23–30 and born in state X and year 1993, we use the PNAD controls in state X at a year later, 1982. We follow this process until year 2004, which is matched with past information using year 1992. Data on GDP per capita and public spending were gathered from the 1985–1997 period because before 1985 data was unavailable for some states. For example, we match outcomes and the main explanatory variables by state of birth X and year 1992 to data on GDP per capita and public spending in state X and year 1985, that is, 7 years before, when these women were aged 16–24. This was the best we could do with the available data on control variables. Finally, considering individual data in 1992–2004 (268,670 observations) and control variables by state and year already merged with it as explained, we group the data by state of birth and cohort (survey year minus age). The grouped data is a panel of 567 observations classed into 21 cohorts (1961–1981) and 27 states of birth. Table A.1 in the appendix provides the mean, standard deviation, source, and sample period of each variable.