Preliminary & incomplete, please do not cite or circulate

Biology, Stress and the Intergenerational Transmission of Economic Status Anna Aizer Brown University and NBER Laura Stroud Brown Medical School Stephen Buka Brown University December, 2007

We explore the role of stress as a mechanism behind the intergenerational transmission of economic status. We find that poor mothers have elevated levels of the stress hormone cortisol and that cortisol negatively affects the IQ and educational attainment of their offspring. This is consistent with experimental evidence based on animals linking stress hormones in utero with declines in offspring cognition and health. Moreover, we find that conditional on poverty, black mothers have higher cortisol than white mothers. Our findings suggest an important role for stress in explaining why the children of poor blacks are less likely to escape poverty than their white counterparts and underscore the importance of incorporating biological processes in explanations of economic phenomenon.

I.

Introduction

Intergenerational transmission of economics status in the US is as high as or higher than most other industrialized nations (Corak, 2004). Attempts to understand the mechanisms of transmission have sought to separately identify the forces of nature (genetic heritance) and nurture (environmental influences). These studies have often relied on samples of twins or adoptees to do so. In keeping with advances in our understanding of the importance of very early childhood experiences including the in-utero environment in shaping outcomes, recent work in this area has examined within twin differences in prenatal nutrition (as manifest in birthweight) on short and long term outcomes. We focus on the role of a different environmental factor – stress – in perpetuating intergenerational inequality. We argue that stress is more relevant than nutrition in the context of the US, where malnutrition is relatively rare. We focus on in utero exposure to stress as an important mechanism behind intergenerational transmission of economic status for two reasons: first, poverty is associated with greater levels of stress and second, recent evidence based largely on animal experiments suggests that exogenous exposure to stress in utero negatively affects offspring cognitive functioning. Given the importance of cognitive ability in determining adult education and income, this suggests that greater in utero exposure to stress among the poor has the potential to explain the intergenerational persistence of poverty observed in the US. In this paper we estimate the impact of in utero exposure to stress on offspring IQ and educational attainment using a unique dataset with detailed information on parental characteristics including prenatal stress hormones (cortisol). To establish a causal relationship between exposure to stress hormones in utero and offspring outcomes we use

2

instrumental variable techniques to mitigate against measurement error and omitted variable bias. We find that exposure to elevated maternal cortisol in utero has a negative and significant impact on offspring IQ at ages 4 and 7 as well as the probability of high school completion. However, our results suggest that increases in the level of stress matters at the top of the distribution – small changes at relatively low levels do not appear to have any significant effect on offspring outcomes. This work makes three important contributions to the literature on the intergenerational transmission of economic status. First, we show that exposure to stress in utero may be an important mechanism behind the intergenerational transmission of economic status. This factor has not previously been considered. Second, while much of the existing literature has assumed that ability is mostly a function of genetic heritance, we show that it can be heavily influenced by environmental forces. And third, our results have the potential to explain the lower rates of economic mobility observed among blacks. Conditional on poverty, we find that black mothers have much higher cortisol than white mothers. This difference can potentially explain an 11 point difference in IQ measured between black and white children at age 4. It can also explain a six percentage point difference in high school completion rates across races. This represents nearly three quarters of the measured difference in IQ across races at this age, and half of the difference in high school completion rates. The rest of the paper is organized as follows. In section II we summarize existing theoretical and empirical research on the intergenerational transmission of economic status. In section III, we provide background information on SES, stress, and cortisol. In

3

section IV, we describe the data. In section V we present our empirical results. Section VI concludes.

II.

Intergenerational Transmission of Economic Status: Previous Work Recent work on the intergenerational transmission of economic status in the US

has produced estimates of intergenerational income elasticities on the order of 0.5 with the highest estimate approaching 0.65 (Solon, 1999; Mazumder, 2005).1 Yet these numbers mask important differences by race. Hertz (2005) estimates intergenerational correlations in earnings for black and whites separately using the PSID. He finds that blacks in the bottom of the income distribution experience especially low rates of upward mobility: 17 percent of whites in the bottom decile of family income remain their as adults compared with 42 percent of black children. Hertz explores whether lower levels of parental education among black families can explain these differences and finds that they cannot, concluding that other mechanisms must be responsible.2 Bhattacharya and Mazumder (2007) develop an alternative measure of intergenerational economic mobility: the probability that an adult child’s relative position exceeds that of the parents. This alternative measure still yields racial differences in intergenerational mobility, though the gap is smaller.

1

Initial studies found generally low correlations between the earnings of fathers and sons; however, these studies were based on single year estimates of income and earnings of parents and children. Estimates based on only a single year suffer considerable measurement error which tends to bias down estimates of intergenerational earnings correlations. 2 Loury (1981) extends the Becker, Tomes (1979) model of intergenerational transmission of economic status to include credit constraints. In his model, if and when parents are credit constrained, they may not be able to invest the efficient amount in their children’s human capital. This model could explain blackwhite differences in intergenerational mobility if blacks are more credit constrained than whites as evidence suggests (Perraudin and Sorensen, 1992).

4

Most of the existing theoretical and empirical work that has sought to identify the mechanisms of intergenerational transmission of economic status has categorized the mechanisms as either nature (genetic inheritance) or nurture (the environment). In this work it is often assumed that ability is a function of nature and skills a function of nurture and that the two influences affect child outcomes in an additive manner. Data on twins and adoptees have been used to empirically identify the roles of nature and nurture in determining economic status. Taubman (1989), Behrman and Taubman (1989) and Behrman, Rosenzweig and Taubman (1994) analyze data on identical and fraternal twins in a structural framework based on established behavioralgenetic models. In twin studies, the assumption is that stronger within identical twin correlations are attributable to genetic similarities. These studies generally conclude that both environment and genetic heritance matters in the transmission of such outcomes as educational attainment, income and obesity. However, Goldberger (1978) discusses limitations to the behavioral-genetic model which include, for example, the fact that parents may treat identical twins more similarly than they do fraternal twins so that not only are their genes the same, but so is their environment. More recent work based on twins has focused on the role of birthweight as a measure of prenatal nutrition (nurture) to explain offspring outcomes (Almond, Chay and Lee (2006), Royer (2006), Black Devereaux and Salvanes (2007) and Behrman and Rosensweig (2004)). This work has generally found small short term impacts but larger long term impact of prenatal nutrition on offspring outcomes.3

3

Currie and Moretti find that mother’s born low birth weight (lbw) are 50 percent more likely to give birth to lbw babies and that the intergenerational transmission of lbw is stronger for poor mothers than others,

5

Research based on adoptees generally finds that characteristics of both adoptive parents and biological parents contribute to the education and earnings of their children (Sacerdote, forthcoming; Bjorklund, Lindahl and Plug, 2006). By including the characteristics of both adoptive parents and biological parents, Bjorklund, Lindahl and Plug (2006) estimate what portion of the transmission is due to nature (birth parents) versus nurture (adoptive parents). They also find interactive effects between the two (nature and nurture). But recent advances in the field of epigenetics suggest that these distinctions between nature (genes) and nurture (environment) are false (see Cunha and Heckman, 2007 for a discussion).4 Our work provides further evidence that this distinction may not be warranted. We show that ability, which is often assumed to be a function of nature in this literature, is strongly influenced by the prenatal environment. Our work also differs from existing work in that the prenatal environmental factor that we consider is not nutrition but the level of stress. We argue that this environmental factor may be more relevant in the US as malnutrition is relatively rare in the US (Nord, Andrews and Carlson, 2006).

III.

Background on Economic Status, Prenatal Programming and Stress

suggesting important interactive effects between nature and environment. In an attempt to control for genetic differences, Currie and Moretti analyze between sister differences and find that the effects remain. 4

Geneticists have established that one’s environment plays a critical role in which genes are expressed (made into a functional gene product).

6

SES and Stress The hormone cortisol is considered a marker for stress (Wust et. al. (2000); Van Eck et al (1996)). Given previous findings that low income individuals report greater stressful events in their lives (Dohrenwend, 1973), it is not surprising that more recent work has established higher cortisol levels among those of low economic and social status (Cohen et. al., 2006; Steptoe et al, 2003; Kunkz-Ebrecht, Kirschbaum and Steptoe, 2003). Even controlling for income or poverty, blacks appear to face more stress than their white counterparts (Geronimus et al 2006; Cohen et al, 2006). In recent work exploring racial differences in adolescent cortisol levels, DeSantis, et al (2007) found that a host of socioenvironmental factors examined failed to explain much of the observed racial/ethnic differences in diurnal cortisol rhythms. The authors concluded that other factors, including but not limited to genetic differences, exposure to prenatal stress, and racism/discrimination, may be responsible. There is experimental evidence that perceived racism is a significant stressor faced by blacks. Research exploring the role of stress in explaining black-white differences in cardiovascular disease (CVD) has found that acute exposure to racism increases cardiovascular activation among blacks and also that past exposure to racism can sensitize individuals to future stressors, increasing cardio-vascular response to both race-related and non race-related stressors (see Brondolo, et al 2003 for a review).5 These studies suggest that perceived racism may represent an acute but poorly measured stressor faced by blacks that causes a significant negative physiological response. This 5

Laboratory studies expose participants to racist stimuli in a lab (racist videos, harassment from a white experimenter) and measure cardio-vascular response; observational studies monitor cardiovascular activity throughout the day and ask the participant to keep a time diary of daily events.

7

can potentially explain the higher levels of cortisol observed among blacks, even conditional on income.

Cortisol, Prenatal Programming and Offspring Outcomes Prenatal programming refers to “the action of a factor during a sensitive period or window of fetal development that exerts organizational effects that persist throughout life” (Seckl, 1998). Almond (2006) presents empirical evidence of the importance of inutero conditions in explaining adult mortality, education and disability in the context of the 1918 flu pandemic, consistent with a prenatal programming hypothesis. Cortisol is considered a key agent in prenatal programming. Work based largely on animal studies has established a strong link between exogenous in utero exposure to stress/cortisol and poor offspring outcomes. For example, Welberg, Seckl and Holmes, (2001) administered glucocorticoids (cortisol in humans) to pregnant rats and found that the offspring of exposed rats exhibited behavioral inhibition, and impaired coping and learning in aversive situations.6 Other studies on primates have successfully mimicked the negative impact of a mild stressor during pregnancy on motor and mental development by exogenously exposing mothers to stress hormones (Schneider, 1992). Among humans, non-experimental research has shown that exposure to excessive amounts of cortisol in utero can affect the developing brain and spinal cord (Yu, Lee, Lee and Son, 2004). Huizink et al (2003) find that stress and elevated cortisol in late pregnancy were negatively related to both mental and motor development of offspring at three and eight months.

6

They concluded that prenatal programming of the HPA axis was responsible for the outcome based on studies of the areas of the brain affected.

8

But non experimental studies based on humans suffer from problems endogeneity and measurement error that experimental studies based on animals do not. To address these problems, we identify the impact of cortisol on offspring outcomes using instrumental variable methods. And by focusing on in-utero exposure to stress, our paper differs from Almond (2006) in that the in utero conditions we examine are not random (as the 1918 flu pandemic was) but are correlated with low socio economic status. In this context, prenatal programming can potentially explain not only reduced economic status later in life (as in Almond, 2006) but also the intergenerational transmission of economic status.

IV.

Data The data are a subset of the National Collaborative Perinatal Project (NCPP). The

NCPP was a collaborative multi-site study in the late 1950s and early 1960s that sought to identify the prenatal and early childhood determinants of subsequent child health and well-being. Factors investigated fell into three categories: abnormal conditions of the pregnancy, environmental factors (social and economic conditions) and biological factors in parents (Gordon et al 1972). The study comprised a prospective survey of 55,908 pregnancies between 1959 and 1965 across 12 cities. Women were enrolled primarily though public clinics where they sought prenatal care.7 Extensive data were collected at each prenatal visit, during labor and delivery and during five follow-up periods: 4

7

Unwed women who planned to put their children up for adoption and women who arrived at the hospital for delivery without any prenatal care were excluded from the study.

9

months, 8 months, 1 year, 4 years and 7 years. During follow-up periods, children were examined by physicians and psychologists who evaluated them along a wide range of neurological, development, behavioral and cognitive measures. In this study we focus on a subset of 1000 children born to mothers enrolled in NCPP through either the Providence or Boston sites for whom follow-up information as adults is available.8 Trained interviewers collected information on adult SES (education, income, employment), disease and other characteristics (Buka, Shenassa and Niaura, 2003). Data on personal and household income are heavily top-coded: more than half the sample reported household income in excess of $150,000. This distribution of income in these data exceeds that of a comparable sample of respondents in the Current Population Survey residing in Massachusetts and Rhode Island in 2002. In contrast, information on educational attainment in the NCPP and CPS are more comparable. For this reason and because educational attainment is arguably a better measure of permanent income, we rely on educational attainment as our measure of adult economic status. Maternal blood/serum collected during the third trimester of pregnancy (between 31 and 36 weeks of pregnancy) was analyzed for cortisol, testosterone, cortisol binging globulins (CBG) and sex hormone binding globulin (SHBG). Values obtained were compared to published studies of fresh samples to assess validity after 40 years of storage (Stroud et al, 2007). The results support the overall validity of these cortisol and testosterone values. However, they are measured with error. Cortisol naturally varies

8

The original cohort from the Providence NCPP site included 4,140 pregnancies, of which 3,138 subjects were assessed at age 7. The cohort from the Boston NCPP site included 13,737 pregnancies of which 8,931 were assessed at age 7. The subset of those followed up in 2002-2004 was selected from among those who had completed the 7 year assessment for a study investigating the consequences of maternal smoking. The subset of the NCPP for which adult follow-up data are available comprise a subset of those for whom maternal smoking data are available.

10

over the course of the day. Current studies of cortisol levels follow relatively standard methods of collection – taking measurements immediately upon waking and at specific intervals throughout the day. The cortisol measures in this study are spot measures and it is not known when during the day they were obtained, introducing random measurement error. To overcome the attenuation bias associated with measurement error, we use instrumental variables. We also approximate the extent of measurement error in these data to gauge the potential attenuation bias in OLS estimates of the impact of cortisol on offspring outcomes. Table 1 presents sample means for the entire NCPP cohort (all 12 cities), the NCPP Boston/Providence cohort, and the analysis sub-sample. The subset of individuals selected from the Providence and Boston NCPP for adult follow-up is slightly less educated and of slightly lower SES than those in the general Boston/Providence cohort, but the differences, while statistically significant, are small. Birth outcomes (weight and gestation) for the analysis subsample are considerably greater which we attribute to the following: 1) the analysis sample does not contain very premature or low birth weight babies that subsequently died within the first seven years of life (as the greater NCPP sample does) and 2) because cortisol measures were based on blood collected in the third trimester, all births in the analysis subsample reached at least 31 weeks gestation and usually longer, thereby excluding the most premature births. In the last column of the table are descriptive statistics from the 1960 census of women with children less than five years old residing in Massachusetts and Rhode Island.9 As is evident from the table, women in our sample are on average more likely to

9

Because of limitations of the census data, we were unable to calculate averages for Providence and Boston only.

11

be black, younger, less educated, single and poorer. This is not surprising given that the NCPP sample is urban and that recruitment for subjects in the NCPP was conducted through public clinics. The demographic composition of the NCPP aids our investigation of the intergenerational transmission of income (poverty) and differences in transmission by race.

V.

Results Estimation of the role of stress in the intergenerational transmission of economic

status proceeds in three stages. First we establish intergenerational correlations in economic status in these data and document higher rates of persistent poverty for blacks. Next we explore the role of maternal stress in mediating intergenerational correlations in economic status. To do so, we first document higher levels of the stress hormone cortisol among women of lower economic status. Second, given the medical evidence linking exposure to cortisol in-utero with future cognitive ability, we explore IQ as a mechanism by which maternal cortisol may “transmit” economic status from mother to child. For this we estimate the impact of prenatal cortisol levels on child IQ using OLS and IV techniques. Finally, we link prenatal maternal cortisol with adult economic status by providing OLS and IV estimates of the impact of maternal prenatal cortisol on the probability of completing high school. Our results show that prenatal cortisol is elevated in poor mothers and has a negative causal impact on cognitive ability and educational attainment of their offspring. However, these negative effects are observed only among those with very high levels of cortisol – in the top quarter of the distribution. We also find that conditional on poverty,

12

black mothers appear to face greater stress than their white counterparts as indicated by significantly higher cortisol levels. Racial differences in cortisol levels can potentially explain the greater persistence of poverty observed among black families relative to their white counterparts.

A. Persistent Poverty Across Generations Intergenerational correlations in education and income are present in this sample. The raw correlation between maternal education (highest grade completed) and that of her adult offspring is 0.285 and the correlation between family income during the prenatal period and the personal income of adult offspring is 0.156.10 In Table 2 we present evidence of greater persistence of poverty among blacks in these data, consistent with previous findings based on other data (Hertz, 2005; Bhattacharya and Mazumder, 2007). In panel A of the table are the share of children born below poverty who grow up to be high school drop outs (column 1) and the share with adult personal income in the bottom 15 percent of the income distribution (column 2). 11 While 14 percent of poor white children do not complete high school, 32 percent of black children don’t and while 10 percent of poor white children remain in the bottom 15 percent of the income distribution as adults, 18 percent of black children do. In panel B are similar numbers for those children born to mothers without a high school degree and intergenerational persistence of poverty is still greater among black families. 10

The correlation of 0.156 is on the lower end of the spectrum but reasonable given that the income measures are based on single years and the fact that roughly 25% of the adult personal income reports are top coded, introducing considerable measurement error, a point to which we return. 11 We do not utilize the full spectrum of adult income due to top-coding in the data: 25 percent of the responses to the personal income question are top-coded and 53 percent of the responses to the household income question are top-coded. Instead, we construct an indicator for whether the responded reported income in the bottom 25 percent of the income distribution in these data.

13

B. Maternal SES and Cortisol In these data, low social and economic status is associated with elevated cortisol levels, consistent with existing literature (Cohen et al, 2006; Cohen, Doyle and Baum, 20006). Figure 1A presents mean maternal cortisol levels by race (white/black), marital status (single/married), SES index (low/high), poverty (below/above).12 As is evident from the figure, Blacks, single mothers, poor mothers and those of low SES have higher levels of cortisol. This is consistent with evidence that individuals of lower SES report greater exposure to stressful live events (Dohrenwend, 1973). In Table 3 we explore the source of differences in cortisol measures between white and black mothers. The nine point average difference in cortisol between blacks and whites is driven entirely by those below poverty. In fact, for the non-poor, there are no racial differences in average cortisol levels. Among the poor, however, blacks have significantly higher cortisol (25 points, or 30% of a standard deviation) relative to whites. The second panel of the table suggests that this racial difference is driven by large differences at the top of the distribution of cortisol: among those in the top quartile (cortisol above 328), blacks have significantly higher cortisol levels (414 versus 391). At low and average levels of cortisol, there is no significant difference between blacks and whites. In the remainder of this section we explore IQ as a mechanism by which maternal stress may influence offspring economic status. We focus on prenatal programming of

12

The SES index is a composite index adapted from the US Census Bureau that averages centiles derived from the education and occupation of the head of the household and family income. It ranges from 0 to 9.9 with higher scores representing higher SES.

14

child cognitive ability as a mechanism given 1) animals experiments linking elevated cortisol to offspring cognition, 2) evidence that the transmission of cognitive ability and education explains three fifths of the intergenerational transmission of economic status (Bowles and Gintis, 2002) and 3) work by Neal and Johnson (1996) and Bhattacharya and Mazumder (2007) showing that pre-market differences in cognitive ability explain much of the black-white earnings difference. We begin by documenting the correlation between family background and child IQ and between child IQ and future adult economic status in this sample. This is followed by OLS and IV estimation of the impact of prenatal maternal cortisol on child IQ and high school completion, our measure of offspring economic status.

C. Family Background, Childhood IQ, and Adult Economic Status Family background is highly correlated with offspring IQ at age 4 in this sample (Figure 2). IQ at age 4 is higher for those children born to white mothers, married mothers, mothers with at least a high school degree, non-poor mothers, non-teen mothers and high SES mothers. These differences range from 7 points to 16 points (one half to one standard deviation) and all differences are statistically significant. The only family background measure with which IQ does not appear to be correlated is whether the mother was on welfare. Child IQ is also correlated with later economic status. High school drop outs have an average IQ of 94, relative to 108 for high school graduates. And among males, those with personal income below $9000 annually have average IQs at age 4 of 92, relative to 105 for those with personal income that exceeds $9000 annually. This is consistent with a large body of work suggesting that cognitive ability is

15

an important predictor of educational and economic success (Heckman, 1995; Murname, Willet and Levy, 1995) Child IQ also appears to explain some of the black-white differences in adult economic status, consistent with work by Neal and Johnson (1996) and Bhattacharya and Mazumder (2007). In Table 4 we explore the extent to which childhood characteristics can explain black-white differences in adult economic status. In the first panel of the table we present coefficients of a regression of adult economic status (probability of adult personal income below $9000 and whether a High School Drop out) on an indicator for black, including controls for gender, adult age and city of residence. In the first column the coefficient on black is 0.075 indicating that blacks are 7.5 percentage points more likely to have less than a high school degree. In the second column, we control for birthweight, in the third for 4 year IQ, in the fourth and fifth for 7 year IQ, and in the sixth and seventh columns we control for chronic health conditions in childhood. What is evident from the table is that cognitive ability as measured by IQ at ages 4 and 7 explains more of the black-white differences in high school completion and adult poverty than the other child hood characteristics. In the second panel of the table, we condition on being born below poverty and the same pattern emerges.

D. Impact of Maternal Cortisol on Offspring IQ – OLS Estimates In this section we provide empirical evidence that maternal prenatal cortisol negatively affects offspring cognitive ability, but the effect is limited to those with the highest levels of cortisol. In Table 5 we present results from the following linear

16

regression of four year IQ on maternal prenatal cortisol and a vector of family and child characteristics: IQ = α + β1Cortisol + β2X+ β3Birthweight + ε

(1)

X, the vector of family and child characteristics at birth, includes gender, maternal race, age and education, parity (number of living siblings at age 7), birth order, family income at birth, marital status, maternal score on the SRA (a test of cognitive ability), maternal religion, housing density, whether there was any pregnancy complication and whether the family resided in Providence or Boston at the time of the birth. We also control for birthweight as elevated cortisol in utero has been associated with reduced birthweight (Mulder et al, 2002). We find that cortisol has a very small and only borderline significant impact on child IQ: a standard deviation increase in cortisol leads to less than a one point drop in IQ (or five percent of a standard deviation). However, this estimate is likely biased downwards due to both measurement error in cortisol and the possible misspecification of a linear regression, as we discuss below.

Measurement Error in Cortisol Cortisol naturally varies over the period of gestation and over the course of the day. With respect to variation over the period of gestation, Harville et al (2007) record average cortisol levels of 0.28 ug/dL for pregnant women in the 14th week of gestation increasing to 0.56 for those in the 31st week of gestation. With respect to variation in cortisol levels over the course of the day, Steptoe, et al (2003) found that cortisol measures in adult men decline roughly 50 percent (in a linear fashion) over the course of

17

the 9 hour workday: from 15 nmol/l at 8 am to slightly more than 6 nmol/l by 6pm. Other research based on pregnant women recorded a mean level of cortisol of 0.56 (ug/dL) at waking which declined to 0.16 by bedtime (Jones et al, 2006).13 In our data, we do have information on the week of gestation the blood was drawn, but not on the time of day.14 If there is no relationship between time of day the blood was drawn and maternal characteristics, then this would introduce classical measurement error and lead to attenuation bias for which an instrumental variable strategy may correct. In order to assess the extent of measurement error introduced because the time of the blood draw is unknown, we simulate what the distribution of cortisol would be if cortisol were measured at the same time for all women (assumed mid day) and if it were measured at different times throughout the day, as it is in our sample. The underlying assumption is that the latter represents the distribution of a noisy measure of cortisol and the former represents the distribution of the true measure of cortisol, without noise. With these two distributions we can calculate a reliability ratio which characterizes the extent of bias in OLS estimates when cortisol is measured with error. To generate the two distributions, we proceed in three steps. First, we assume that cortisol measured without error is distributed normally around a mean of 280 (ng/ml) with a standard deviation of 20. We selected the mean of 280 because it corresponds to the mean in our sample and it also corresponds to the average cortisol at 11am measured in 100 pregnant women (three times each) by Harville et al (2007). Second, we add random noise to each woman’s cortisol measure based on a randomly assigned time of 13

Harville, et al (2007) report average cortisol among pregnant women at wake of 0.48 (ug/dL), rising to 0.56 30 minutes later and then declining to 0.28 by 11 am and 0.19 by 5pm and 0.16 by 9pm. 14 Even if we have week of gestation, gestation is measured with error in pregnancy

18

blood draw. We assume that the time of day of sample collection is distributed randomly in a uniform fashion over the course of a 9 hour day and that cortisol declines by 290 ng/ml in a linear fashion over this period (based on ranges from wakening to 5pm of 480 to 190 as documented by Harville et al, 2007). The noise is added in such a way as to preserve the mean of 280 in the noisy distribution (representing a mean preserving spread). Finally, we compare the standard deviations of the distribution of cortisol with and without noise and calculate a reliability ratio which corresponds to the following: Reliability Ratio = (σ2c + σ2v)/σ2c Where σ2c represents the variance of the true measure of cortisol (set at 400 in this exercise) and σ2v represents the variance of the measurement error or noise. The reliability ratio is equal to the ratio of the instrumental variable estimate relative to the OLS estimate and provides us with an estimate of the attenuation bias present in an OLS estimate. When we simulate the two distributions (for a sample of 1000) based on a standard deviation in the true measure of cortisol (σc) of 20, we find that the standard deviation of the noise is (σv) is 84. Based on these numbers, we calculate a reliability ratio of 19, indicating that OLS estimates of the impact of cortisol on offspring outcomes will be considerably biased downwards. In addition to the attenuation bias that results from classical measurement error, the regression may also be misspecified. Existing medical research on the impact of prenatal cortisol on offspring cognition does not yield the prediction that the relationship should be entirely linear. Results based on animal studies or persons with Cushing disease suggest only that exposure to high levels of cortisol result in diminished cognitive

19

functioning. No studies yield any predictions with respect to marginal increases in cortisol at low levels. In addition, in human studies, it has been found that those who have undergone significant trauma and suffer from post-traumatic stress disorder (PTSD) are likely to have lower cortisol levels than average, suggesting a non-linear effect. In the next section we present instrumental variable estimates of the impact of cortisol on child IQ. The instruments are designed to overcome the attenuation bias induced by classical measurement error in the cortisol variable. In addition, we explore alternative functional forms and provide evidence that a simple linear regression of the impact of cortisol on outcomes may be miss-specified.

E. IV Estimates We include three instruments for maternal cortisol. The first instrument is the week of gestation the blood was drawn. Because cortisol varies over the course of the period of gestation, when the blood sample was drawn should be correlated with the cortisol measure but should not independently affect child outcomes.15 The second instrument is an indicator for whether the woman received the drugs DES, estrogen, progesterone during pregnancy that are (now) known to increase cortisol levels. Women with relatively minor pregnancy complications such as morning sickness received these drugs, but not all women with these pregnancy complications did. Though we control for pregnancy complications in the analysis, it is still possible that these drugs may affect child outcomes independent of their impact on cortisol by, for example,

15

This instrument could raise concerns if women with blood drawn later were less likely to deliver prematurely which could independently affect offspring outcomes. Including birthweight in the regression, however, should address this concern.

20

affecting the health of the child as it develops in utero (ie – fail the exclusion restriction). To limit this possibility, we control for birthweight in these regressions – the most common measure of health at birth. In so doing, we are likely underestimating the impact of cortisol on child outcomes if part of the impact is operating through reduced health of the fetus. In Table 6 we present evidence that the administration of these drugs is uncorrelated with SES or other observed characteristics that may independently affect offspring outcomes. In the first column are results of a regression of an indicator for whether a mother received any of these drugs on maternal education, marital status, income, race, parity, child gender and pregnancy complication. The only coefficients that are statistically significant are pregnancy complication and birthweight. All other coefficient estimates are small in magnitude and imprecisely estimated and an F statistic on all family characteristics is 1.00. The third instrument is maternal testosterone which plays a role in regulation of the HPA axis and is positively correlated with cortisol levels but does not increase with exposure to stress as cortisol does. Maternal testosterone has been found to increase during pregnancy, though the cause is unknown.16 What is known is that testosterone levels decrease with maternal age and are higher in black women (Troisi et al 2003), but little else has been found to be predictive of maternal testosterone levels during pregnancy. If the increase in maternal testosterone during pregnancy directly affects

16

The exact cause for increasing testosterone during pregnancy is not known, though evidence suggests that the increase is due largely to a decrease in the rate at which testosterone is metabolized and less an increase in the production of testosterone (Bammann, Coulam and Jiang, 1980; Kerlan, Nahoul, Le Martelot and Bercovici, 1994).

21

cognitive functioning or economic status of the offspring it fails the exclusion restriction and is not a valid instrument. We argue that maternal testosterone is a valid instrument for three reasons. First, most of the existing work linking prenatal testosterone to offspring outcomes is based on rodent studies or studies of humans exposed to atypical levels of testosterone.17 Second, evidence relating prenatal testosterone to offspring outcomes is mixed at best. Putz et al (2005) conducted a review of the existing literature linking fetal testosterone levels (as measured by 2D:4D) with offspring outcomes. They concluded that “out of 57 correlations, 2D:4D correlated significantly in the predicted direction only with sexual orientation (for both sexes) and only for left hand 2D:4D…. These facts suggest a cautious approach to claims of a relationship between 2D:4D and particular aspects of the phenotype.”18 And third, most of the existing work relates fetal testosterone to offspring outcomes and our instrument is maternal testosterone. While some work has found a positive correlation between maternal and fetal testosterone, others have not (Sarkar et al 2007; Rodeck et al 2005). Consistent with the latter, we find no correlation between maternal testosterone and fetal sex in these data. To further strengthen our claim that maternal testosterone may be a valid instrument for maternal cortisol, in Figure 3 we provide evidence that maternal testosterone is uncorrelated with maternal education and cognitive ability, and family income. In column 2 of Table 6 we present results of a regression of maternal 17

Prenatal exposure to testosterone in rodents has been linked to greater “rough and tumble play” (Birke and Sadler, 1988). Girls exposed to unusually high levels of hormones via congenital adrenal hyperplasia are characterized by lower verbal IQ, enhanced spatial ability and more aggressive behavior during adolescence (Nass and Baker, 1991; Hines et al). 18 However, Carlsen, SM, G Jacobsen and P Romundstad (2006) find that increasing maternal testosterone levels from the 25th percentile to the 75th percentile reduces birthweight by 160 grams (one third of a pound). We control for birthweight in our analyses.

22

testosterone levels on maternal education, marital status, age, income at birth, religion, race, maternal SRA score, birthweight, and pregnancy complication. Only maternal age and race appear correlated with maternal testosterone levels (consistent with existing medical evidence), all other maternal and family characteristics are uncorrelated with testosterone levels during pregnancy (F statistic of 1.2). These results provide additional evidence that maternal testosterone is not correlated with observed characteristics of the mother. While this does not rule out the possibility that is correlated with unobserved characteristics that may be correlated with offspring outcomes, it does provide some evidence in support of our claim that maternal testosterone has no independent effect on offspring cognition or educational attainment and thus serves as a valid instrument in this context. The first stage is: Cortisol = α + τ1Drugs + τ1Week + τ3Testosterone + τ4X+ τ5Birthweight + υ (2)

In Table 7 are results from the first stage in which we estimate prenatal cortisol levels as a function of the three instruments and the exogenous characteristics of the family and child. Administration of any one of these drugs during pregnancy increases the level of maternal cortisol by 30 percent of a standard deviation. Testosterone levels also have a positive and significant effect on cortisol levels – a one standard deviation increase in testosterone is associated with an increase in cortisol of 10 percent of a standard deviation. A Sargan test of overidentifying restrictions yields a chi-squared statistic of .325 with a p-value of .569 (appendix table 1).

23

In the second column of Table 5 are IV estimates of the impact of maternal prenatal cortisol on offspring IQ at age 4 and column 3 presents limited information maximum likelihood (LIML) estimates. The IV and LIML estimates are similar and are both larger than the OLS estimate, but are statistically insignificant. However, as noted previously, the existing literature does not contain a clear prediction regarding the functional form of the relationship between cortisol and offspring cognition. To explore whether the impact of a marginal increase in cortisol is the same across the distribution of cortisol, in Figure 4 we plot IQ predicted as a function of the exogenous characteristics and instruments (the reduced form) against cortisol predicted as a function of the exogenous characteristics and instruments (the first stage). The slope of this relationship represents the Wald estimator. We do this separately for those with cortisol values in the bottom quartile (below 223), second quartile (between 223 and 270), third quartile (between 270 and 328) and top quartile (above 328). Cortisol only appears negatively related with offspring IQ among those in the top quartile. Based on this, we regress offspring IQ on a spline of cortisol with the knot at 328 (representing the top quartile of the distribution of cortisol). OLS, IV and LIML results are presented in Table 8. The spline is constructed such that the regression coefficients for these variables represent the slopes for the intervals (they do not represent the change in the slope).19 We find that increasing maternal prenatal cortisol when cortisol is below 328 appears to have no impact on offspring IQ. In contrast, a marginal increase in cortisol within the upper quarter of the distribution of cortisol appears to have a small, negative and significant impact on offspring cognition: a 23 unit increase in cortisol in

19

Spline cortisol <328 = cortisol*I(cortisol<328) and spline cortisol >328 = (cortisol-328)*I(cortisol>328)

24

this range (which represents the standard deviation of this range) results in less than a one point drop in IQ at age 4. However, as discussed previously we believe the OLS estimate to be biased downward due to measurement error in cortisol. The results of the simulation exercise suggest that IV estimates may be as much as 20 times larger than the OLS estimates. In columns 2 and 3 of Table 8 are the IV and LIML estimates. The IV estimates are 23 times the OLS estimates which is only slightly more than what our simulations suggested. The results suggest that among those with already high levels of cortisol, increasing maternal cortisol further still by 20 would lead to a 10 point decline in offspring IQ at age 4. Because of concerns regarding the validity of testosterone as an instrument, we also develop an alternative instrument – testosterone predicted as a function of maternal age and other exogenous characteristics.

As indicated in Figure 3 Testosterone

increases linearly with maternal age but does not appear correlated with other maternal characteristics. Thus we can predict testosterone as a function of maternal age and other exogenous characteristics: Testosterone = ρ1X2+ ρ2Maternal Age + ν

(3)

Where X2 includes the vector of maternal and child characteristics as before but excludes the continuous variable maternal age and includes instead two indicator variables for whether the mother was under 20 or over 35. We include these two variables out of concern that having either a very young or very old mother might affect offspring outcomes directly. By including controls for these three age groups (teen, age 20-34, and 35 and over), we are identifying the impact of maternal age on testosterone off of small

25

linear increases in maternal age within the three age categories. Testosterone predicted in this manner is then substituted for testosterone in equation (2) and the second stage regression includes the vector X2 rather than X. The results of IV regressions based on this alternative instrument are presented in Table 9 and are similar to previous IV results based on (unpredicted) testosterone as an instrument. Finally, we present estimates of the impact of maternal cortisol on child IQ at age 7 (Table 10). Three IQ scores were computed at age 7: verbal, performance (non-verbal) and full. Work based on individuals with Cushings disease which is characterized by high cortisol levels suggests that verbal ability would be most affected (Starkman et al, 2001). Consistent with this, we find that elevated cortisol has a large negative and significant effect on verbal IQ and (in results not presented here) a smaller and insignificant effect on non-verbal IQ. These result have important implications for our understanding of measured black-white differences in cognitive ability. Not only are blacks more likely to have cortisol levels in the upper quartile of the distribution, but conditional on cortisol in the top quartile, their cortisol levels are on average 23 points higher than those of their white counterparts. A 23 point difference in prenatal cortisol levels in this range translates to an11 point difference in offspring IQ, or roughly three quarters of the observed difference between white and black children in this sample. Given the finding of Neal and Johnson (1996) that black-white differences in pre-market factors (namely cognitive ability as measured by the AFQT score) explain most of the black-white difference in earnings, these results suggest that racial differences in environmental stress may explain the greater persistence of poverty among blacks. We explore this in the next section.

26

F. Prenatal Cortisol and the Transmission of Adult Economic Status In this section we estimate the impact of maternal prenatal cortisol on the transmission of adult economic status. As noted previously, we focus on intergenerational transmission of education in part because it is arguably a better measure of permanent income than a single year of earnings and in part because of difficulty externally validating adult income measures in these data. In restults not presented here, we find small and statistically insignificant effects of cortisol on years of education. However, if we focus on high school completion instead of years of education, we find that increases in maternal prenatal cortisol do significantly reduce the probability of high school completion. In Table 11 we provide probit and IV probit estimates of the impact of maternal cortisol on the probability of being a high school drop-out. The IV probit estimate of 0.018 is equivalent to a marginal effect of 0.0027. In our sample, 17 percent of the offspring of black mothers did not complete high school relative to five percent of the offspring of white mothers.20 The 23 point difference in cortisol between black and white mothers in this sample (conditional on cortisol in the top quartile) can explain a six percentage point difference in high school drop out rates, or half the observed black-white difference in high school drop out rates. The fact that we find positive and significant results in the lower tail of the education distribution is not surprising given that those characterized by high maternal cortisol levels are most likely to be at this margin.

20

If we condition on those with cortisol in the top quartile, black HS drop out rates are 22 percent, relative to eight for whites.

27

VI. Conclusions In this paper we explore whether stress may be responsible in part for the intergenerational transmission of economic status. We find that stress hormones are higher in low socio-economic status mothers and that exposure to elevated prenatal cortisol in utero negatively affects offspring cognition and high school completion, consistent with a prenatal programming hypothesis. These results provide further evidence that “ability” is significantly and largely shaped by one’s environment (nurture). However, these results hold only for those exposed to the highest levels of prenatal cortisol (the top quartile of the distribution). Interestingly, we find that black mothers are both more likely to have cortisol levels in the top quartile and conditional on being in the top quartile have higher cortisol levels on average than their white counterparts. These results can help explain, in part, why poor black children are less likely to escape poverty than their white counterparts and underscore the importance of in-utero conditions in explaining the intergenerational transmission of economic status. They also underscore the ways in which discrimination may affect the children of black mothers not only directly through job market discrimination but also indirectly by elevating the level of stress in their lives, thereby negatively affecting their human capital accumulation and future economic status.

28

References Almond, Douglas (2006) “Is the 1918 Influenza Pandemic Over? Long-term Effects of In Utero Influenza Exposure in the Post-1940 U.S. Population” Journal of Political Economy, 114: 672-712. Bartels, Meike, ge Geus, Eco, Kirschbaum, Clemens, Sluyter, Frans and Dorret Boomsma (2003) “Heritability of Daytime Cortisol Levels in Children” Behavior Genetics, Vol 33 No 4: 421-433. Becker, Gary and Nigel Tomes (1979) “An Equilibrium Theory of the Distribution of Income and Intergenerational Mobility", Journal of Political Economy, Vol. 87, No. 6: 1153-1189. Becker, Gary and Nigel Tomes (1986) “Human Capital and the Rise and Fall of Families” Journal of Labor Economics Vol. 4 no 3: S1-39. Behrman, Jere and Paul Taubman (1989) “Is Schooling Mostly in the Genes? NatureNurture Decomposition Using Data on Relatives” Journal of Political Economy Vol 97, No 6: 1425-1446. Behrman, Jere, Rosenzweig, Mark and Paul Taubman (1994) "Endowments and the Allocation of Schooling in the Family and in the Marriage Market: The Twins Experiment," Journal of Political Economy Vol. 102, No. 6 :1131-1174 Behrman, Jere and Mark Rosenzweig (2004) "Returns to Birthweight," The Review of Economics and Statistics, Vol. 86, No. 2: 586-601. Birke, LI and D Sadler (1988) “Effects of Modulating Neonatal Progestins and Androgens on the Devlopement of Play and Other Social Behavior in the Rat” Hormones and Behavior, 22:160-171. Bjorklund, Anders, Lindahl, Mikael and Erik Plug (2006) “The Origins of International Associations: Evidence from Swedish Adoption Data” Quarterly Journal of Economics : 999-1028. Bammann, BL, Coulam CB, and NS Jiang (1980) “Total and Free Testosterone During Pregnancy” American Journal of Obstetrics and Gyneocology. Vol. 137 No. 3: 293-298. Bhattacharya, Depopam and Bhashkar Mazumder (2007) “Nonparametric Analysis of Intergenerational Income Mobility with Application to the United States” mimeo Black, Sandra, Devereux, Paul and Kjell Salvanes (2007) “From the Cradle to the Labor Market? The Effect of Birth Weight on Adult Outcomes” Quarterly Journal of Economics,

29

Bowles, Samuel and Herbert Gintis (2002) “The Inheritance of Inequality” Journal of Economic Perspectives Vol. 16 No. 3: 3-30. Brondolo, Elizabeth, Rieppi, Ricardo, Kelly, Kim and William Gerin (2003) “Perceived Racism and Blood Pressure: A Review of the Literature and Conceptual and Methodological Critique” Annals of Behavioral Medicine Vol 25, No. 1: 55-65. Buka, Stephen, Shenassa Edmund and Raymond Niaura (2003) “Elevated risk of tobacco dependence among offspring of mothers who smoked during pregnancy: a 30-year prospective study.” American Journal of Psychiatry. Vol. 160 No. 11:1978-84. Carlsen, SM, G Jacobsen and P Romundstad (2006) “Maternal testosterone levels during pregnancy are associated with offspring size at birth” European Journal of Endocrinology, Vol 155, Issue 2, 365-370. Cohen, Sheldon, Schwartz, Joseph, Epel, Elissa, Kirschbaum, Clemens, Sidney, Steve and Teresa Seeman (2006) “Socioeconomic Status, Race and Diurnal Cortisol Decline in the Coronary Artery Risk Development in Young Adults (CARDIA) Study” Psychosomatic Medicine Vol 68: 41-50. Corak, Miles.(2004). “Do Poor Children Become Poor Adults? Lessons for Public Policy from a Cross-Country Comparison of Earnings Mobility.” UNICEF Innocenti Research Center, Florence, Italy. Cunha, Flavio and James Heckman (2007) “The Technology of Skill Formation” American Economic Review 97 (2): 31-47. Currie, Janet and Enrico Moretti (forthcoming) "Biology as Destiny? Short and LongRun Determinants of Intergenerational Transmission of Birth Weight" Journal of Labor Economics DeSantis AS, Adam EK, Doane LD, Mineka S, Zinbarg RE, Craske MG.(2007) “Racial/ethnic differences in cortisol diurnal rhythms in a community sample of adolescents.” J Adolescent Health. 2007 Jul;41(1):3-13 Fiocco, Alexandra, Wan, Nathalie, Weekes, Nicole, Pim, Heather, and Sonia Lupien (2006) “Diurnal cycle of salivary cortisol in older adult men and women with subjective complaints of memory deficits and/or depressive symptoms: Relation to cognitive functioning” Stress Vol. 9 No. 3: 143-152. Geronimus, Arline, Hicken, Margaret, Keene, Danya and John Bound (2006) “ “Weathering” and Age Patterns of Allostatic Load Scores Among Blacks and Whites in the United States” American Journal of Public Health, Vol 96 No.5: 826-833.

30

Goldberger, Arthur (1989) “Economic and Mechanical Models of Intergenerational Transmission” American Economic Review, Vol. 79 No. 3:504-513. Goldberger, Arthur (1978) "The Genetic Determination of Income: Comment." American Economic Review Vol. 68:960-69. Gutteling, Barbara, deWeerth, Carolina, and Jan Buitelaar (2005) “Prenatal Stress and Children’s Cortisol Reaction to the First Day of School” Psychoneuroendocrinology. Vol. 30: 541-549.

Heckman, James (1995) “Lessons from the Bell Curve” Journal of Political Economy 103(5): 1091-1120. Hertz, Thomas (2005) “Rages, Riches and Race: the Intergenerational Economic Mobility of Black and White Families in the United States,” in Unequal Chances: Family Background and Economic Success, edited by Samuel Bowles, Hervery Gintis and Melissa Osborne Groves (New York: Russesll Sage and Princeton University Press): 165-191. Huizink, Anja, Robles de Medina, Pascale, Mulder, Eduard, Visser, Gerard and Jan Buitelaar (2003) “Stress During Pregnancy is Associated with Developmental Outcome in Infancy” Journal of Child Psychology and Psychiatry Vol 44 No. 6: 810-818. Kerlan, V, Nahoul, K LeMartelot, MT and JP Bercovici (1994) “Longitudinal Study of Plasma Bioavailable Testosterone and Androstanediol Glucuronide Levels During Pregnancy” Clinical Endocrinology. Vol. 40 No.2:263-267. Kunz-Ebrecht, Sabine, Kirschbaum, Clemens and Andrew Steptoe (2004) “Work Stress, Socioeconomic Status and Neuroendocrine Activation Over the Working Day” Social Science and Medicine Vol 58: 1523-1530. Loury, Glenn (1981) “Intergenerational Transfers and the Distribution of Earnings,” Econometrica, Vol. 49, No. 4: 843-867. McEwen, Bruce (2000) “Allostatis and Allostatic Load: Implications for Neuropsychopharmacology” Neuropsychopharmacology, Vol. 22, No. 2: 108-124. McEwen, Bruce (1998) “Protective and Damaging Effects of Stress Mediators” New England Journal of Medicine vol 338: 171-179. Murname, Richard, Willett, John and Frank Levy (1995) “The Growing Importance of Cognitive Skills in Wage Determination” Review of Economics and Statistics 77(2):25166.

31

Neal, Derek and William Johnson (1996) “The Role of Premarket Factors in BlackWhite Wage Differences” Journal of Political Economy, Vol. 104, No. 5: 869-95 Nord, Mark, Andrews, Margaret and Steven Carlson (2006) “Household Food Security in the US, 2005” Economic Research Report No. (ERR-29) 68 pp, November 2006 O’Connor, Thomas, Ben-Schlomo, Yoav, Heron, Jon, Golding, Jean, Adams, Diana and Vivette Glover (2005) “Prenatal Anxiety Predicts Individual Differences in Cortisol in Pre-Adolescent Children” Biological Psychology Vol 58: 211-217. Olga Ousova, Véronique Guyonnet-Duperat, Nathalie Iannuccelli, Jean-Pierre Bidanel, Denis Milan, Carine Genêt, Bastien Llamas, Martine Yerle, Joël Gellin, Patrick Chardon, Agnès Emptoz-Bonneton, Michel Pugeat, Pierre Mormède and Marie-Pierre Moisan (2004) “Corticosteroid Binding Globulin: A New Target for Cortisol-Driven Obesity” Molecular Endocrinology vol 18 no.7: 1687-1696 Perraudin, William and Bent E. Sørensen (1992) “The Credit-Constrained Consumer: An Empirical Study of Demand and Supply in the Loan Market” Journal of Business & Economic Statistics, Vol. 10, No. 2: 179-192. Royer, Heather (2006) “Separated at Girth: US Twin Etimates of the Effects of Birthweight” mimeo Case Western Reserve University Sacerdote, Bruce (forthcoming) “How Large ar the Effects from Changes in Family Environment? A Study of Korean American Adoptees” Quarterly Journal of Economics Sapolsky, Robert, Uno, Hideo, Rebert, Charles and Caleb Finch (1990) “Hippocampal Damage Associated with Prolonged Glucocorticoid Exposure in Primates” The Journal of Neuroscience Vol 10, No. 9: 2897-2902 Schneider, ML (1992) “Prenatal stress exposure alters postnatal behavioral expression under conditions of novelty challenge in rhesus monkey infants” Developmental Psychobiology Vol 25 No 7: 529-540 Sekl, Jonathan (1998) “Physiologic Programming of the Fetus” Clinics in Perinatalogy Vol 25: 939-964. Solon, Gary (1999) “Intergenerational Mobility in the Labor Market” Handbook of Labor Economics, Vol 3, Eds Orley Ashenfelter and David Card. Starkman, Monica, Giordani, Bruno, Berent, Stanley, Schork, Anthony and David Schteingart (2001) “Elevated Cortisol Levels in Cushing’s Disease Are Associated With Cognitive Decrements” Psychosomatic Medicine 63:985–993.

32

Steptoe, Andrew, Kunz-Ebrecht, Sabine, Owen, Natalie, Feldman, Pamela, Willemsen, Gonneke, Kirschbaum, Clemens, and Michael Marmot (2003) “Socioeconomic Status and Stress-Related Biological Responses Over the Working Day” Psychosomatic Medicine vol 65: 461-470. Steptoe, Andrew and Michael Marmot (2002) “The Role of Psychobiological Pathways in Socio-Economic Inequalities in Cardiovascular Disease” European Heart Journal Vol 23: 13-25. Stroud, Laura, Solomon, Catherine, Shenassa, Edmund, Papandonatos, George, Niaura, Raymond, Lipsett, Lewis, LeWinn, Kaja and Stephen Buka (2007) “Long-term stability of maternal prenatal steroid hormones from the National Collaborative Perinatal Project: still valid after all these years.” Psychoneuroendocrinology Vol. 32 No. 2:140-50. Tsigos, Constantine and George Chrousos (2002) “Hypothalamic-Pituitary-Adrenal Axis, Neuroendocrine Factors and Stress” Journal of Psychosomatic Research vol 53: 865-871. Yu, In Tag, Lee, Sang-Hun, Lee Yong-Sung, and Hyeon Son (2004) “Differential effects of corticosterone and dexamethasone on hippocampal neurogenesis in vitro.” Biochemical and biophysical research communications. Vol 317 No. 2:484-90. Van Eck, Marleen, Berkhof, Hans, Nicolson, Nancy and Jose Sulon (1996) “The Effect of Perceived Stress, Traits, Mood States and Stressful Daily Events on Salivary Cortisol” Psychosomatic Medicine Vol 85. No. 5: 447-458. Walker, Brian, Phillips, David, Noon, Josephs, Panarelli, Maurizio, Andrew, Ruth, Edwards, Hugh, Holton, David, Seckl, Jonathan, Webb, David, and Graham Watt (1998) “Increased Glucocorticoid Activity in Men With Cardiovascular Risk Factors” Hypertension 31:891-895. Welberg, LAM, Sekl, Jonathan and Meghan Homes (2001) “Prenatal Glucocorticoid Programming of Brian Corticosteroid Receptors and Corticotrophin-Releasing Hormone: Possible Implications for Behaviour” Neuroscience Vol 104: 71-79. Wust, Stehan, Federenko, Ilona, Hellhammer, Dirk and C Kirschbaum (2000) “Genetic Factors, Perceived Chronic Stress and the Free Cortisol Response to Awakening” Psychoneuroendocrinology Vol 25: 707-720. Hines, M., Fane, B.A., Pasterski, V.L., Mathews, G.A., Conway, G.S. & Brook, C. (2003) “Spatial abilities following prenatal androgen abnormality: targeting and mental rotations performance in individuals with congenital adrenal hyperplasia.” Psychoneuroendocrinology 28, 1010–1026. Nass, R. & Baker, S. (1991) “Learning disabilities in children with congenital adrenal hyperplasia.” Journal of Child Neurology,6, 306–312.

33

Berenbaum, S.A. & Resnick, S.M. (1997) “Early androgen effects on aggression in children and adults with congenital adrenal hyperplasia” Psychoneuroendocrinology, 22, 505–515.

34

Table 1: Characteristics of NCPP and Analysis Sample

Maternal Characteristics Share White Share Black Share Hispanic Maternal Age Single Maternal Education Share
Entire NCPP

Bos/Pvd NCPP

Analysis Sample

0.47 0.46 0.06 24.3 0.23 10.62 0.55 0.36 0.04 47 4030 0.11 0.15 0.07

0.86 0.12 0.002 25.2 0.11 11.40 0.39 0.45 0.08 58 4903 0.10 0.19 0.23

0.87 0.12 0.00 25.0 0.09 11.12 0.43 0.50 0.05 56 4784 0.06 0.17 0.51

1.48 0.63 1.51 1.24 2.02 3.65 2.69 3.47 3.41 4 1.89 4.2 2.01 22.08

0.48 3096

0.47 3185 3239 3109

0.41 3307 3384 3258

4.46 6.43 4.46 5.62

59392

17921

1103

* difference is difference between Analysis Sample and Bos/Pvd NCPP

t-statistic difference*

Table 2: Offsrping Economic Status

Mothers below Poverty: white black

Share
Mothers without a HS degree: white black

Share
Offspring Outcomes Share <15% 0.10 0.18

Share <15% 0.11 0.19

Avg Years of Education 13.43 12.92

Avg Years of Education 13.85 13.43

Table 3: Average Coritsol Levels by Poverty and Race

All White Black Black-White Difference

White Black Black-White Difference

All 280 278 287

< Poverty 285 279 304

> Poverty 277 278 276

9

25

-2

Corisol<328 243 240

Cortisol>328 391 414

-3

23

Table 4: Childhood Characteristics and Racial Differences in Adult Economic Status

black male adult age Providence

0.075 [0.023] 0.022 [0.015] 0.003 [0.004] 0.075 [0.016]

birth weight kg

0.079 [0.023] 0.019 [0.015] 0.002 [0.004] 0.076 [0.016] 0.024 [0.015]

4 year IQ

0.056 [0.023] 0.014 [0.015] 0.002 [0.004] 0.059 [0.016]

HS Drop Out 0.036 [0.023] 0.026 [0.014] 0.002 [0.004] 0.049 [0.016]

0.045 [0.023] 0.03 [0.014] 0.003 [0.004] 0.053 [0.016]

0.067 [0.023] 0.022 [0.015] 0.003 [0.004] 0.075 [0.016]

Providence

-0.003 [0.001]

978 0.05

978 0.05

978 0.06

958 0.08

958 0.07

961 0.04

0.144 [0.064] -0.056 [0.055] 0.009 [0.016] 0.158 [0.082]

0.146 [0.065] -0.056 [0.055] 0.009 [0.016] 0.158 [0.082] 0.012 [0.053]

0.091 [0.064] -0.084 [0.053] 0.015 [0.016] 0.103 [0.081]

HS Drop Out 0.032 [0.063] -0.075 [0.052] 0.007 [0.015] 0.061 [0.080]

0.058 [0.062] -0.048 [0.052] 0.003 [0.015] 0.068 [0.080]

0.111 [0.065] -0.04 [0.055] 0.009 [0.016] 0.144 [0.082]

0.128 [0.065] -0.051 [0.055] 0.013 [0.017] 0.164 [0.082]

981 0.03

981 0.03

0.041 [0.057] -0.068 [0.049] 0.013 [0.014] 0.082 [0.074]

0.041 [0.058] -0.068 [0.049] 0.014 [0.015] 0.083 [0.074] -0.004 [0.047]

-0.007 [0.002]

7 year full IQ

981 0.03

961 0.04

961 0.04

Adult Income < $9000 0.014 0.03 0.028 [0.058] [0.059] [0.059] -0.082 -0.063 -0.068 [0.049] [0.049] [0.050] 0.016 0.006 0.01 [0.014] [0.015] [0.015] 0.054 0.044 0.058 [0.075] [0.076] [0.077]

-0.01 [0.002]

any definite chronic conditions age 7 # definite chronic conditions,age 7 179 0.13

0.046 [0.059] -0.068 [0.050] 0.013 [0.015] 0.084 [0.075]

-0.002 [0.002] -0.082 [0.059]

179 0.07

0.041 [0.059] -0.065 [0.051] 0.012 [0.015] 0.079 [0.076]

-0.003 [0.002] -0.009 [0.002]

179 0.07

964 0.03

0.012 [0.013] 964 0.03

-0.003 [0.002]

7 year verbal IQ

Observations R-squared Standard errors in brackets

0.009 [0.018] 0.008 [0.010] 961 0.04

4 year IQ

0.062 [0.028] -0.086 [0.018] 0.005 [0.005] 0.024 [0.019]

-0.002 [0.001] -0.003 [0.015]

birth weight kg

0.062 [0.028] -0.086 [0.018] 0.005 [0.005] 0.024 [0.019]

-0.002 [0.001]

# definite chronic conditions,age 7

adult age

Adult Income < $9000 0.051 0.038 0.039 [0.028] [0.028] [0.028] -0.086 -0.084 -0.083 [0.018] [0.018] [0.018] 0.005 0.003 0.005 [0.005] [0.005] [0.005] 0.017 0.004 0.008 [0.020] [0.020] [0.020]

-0.003 [0.001]

any definite chronic conditions age 7

male

0.056 [0.028] -0.08 [0.018] 0.006 [0.005] 0.024 [0.019] -0.024 [0.018]

-0.001 [0.001]

7 year verbal IQ

Conditional on Childhood Poverty black

0.06 [0.028] -0.082 [0.018] 0.006 [0.005] 0.025 [0.019]

-0.002 [0.000]

7 year full IQ

Observations R-squared Standard errors in brackets

0.067 [0.023] 0.021 [0.015] 0.003 [0.004] 0.075 [0.016]

176 0.18

177 0.16

176 0.07

-0.019 [0.054] 0.013 [0.034] 176 0.06

180 0.03

180 0.03

180 0.05

177 0.03

178 0.04

177 0.03

0.006 [0.031] 177 0.03

Table 5: Linear Impact of Maternal Cortisol on 4 Year IQ

Cortisol (ng/ml) black other race maternal education
OLS -0.008 [0.005] -5.253 [1.550] -3.444 [5.194] -10.305 [1.546] -6.64 [1.424] 0.09 [0.041] 4.589 [2.386] -3.133 [1.337] 6.448 [2.035] -0.348 [0.940] -3.751 [1.845] -3.787 [0.899] -3.012 [0.937] -4.376 [1.062] 1.322 [0.892] -0.672 [4.878] 0.233 [0.103] 0.131 [0.377] -0.502 [0.326] -1.735 [1.103] -1.157 [1.045] 953 0.25

IV -0.014 [0.035] -5.295 [1.566] -3.605 [5.268] -10.306 [1.547] -6.626 [1.427] 0.085 [0.048] 4.585 [2.388] -3.14 [1.339] 6.503 [2.058] -0.361 [0.943] -3.724 [1.852] -3.777 [0.901] -2.957 [0.983] -4.327 [1.095] 1.283 [0.918] -0.474 [4.993] 0.218 [0.131] 0.105 [0.401] -0.499 [0.327] -1.697 [1.122] -1.198 [1.069] 953 0.25

LIML -0.018 [0.044] -5.319 [1.559] -3.701 [5.254] -10.307 [1.531] -6.618 [1.413] 0.083 [0.052] 4.583 [2.363] -3.145 [1.325] 6.536 [2.048] -0.369 [0.935] -3.708 [1.836] -3.771 [0.893] -2.924 [0.998] -4.297 [1.102] 1.259 [0.922] -0.356 [5.005] 0.209 [0.144] 0.09 [0.410] -0.496 [0.324] -1.675 [1.121] -1.223 [1.071] 953

Table 6: Determinants of DES/Estrogen/Progesterone/Testosterone

black other race maternal education
Estrogen/Progesterone -0.027 [0.019] -0.022 [0.063] -0.005 [0.019] -0.018 [0.017] 0 [0.000] -0.035 [0.029] -0.007 [0.016] -0.021 [0.025] 0.004 [0.011] -0.012 [0.022] -0.01 [0.011] 0.034 [0.011] -0.012 [0.013] -0.035 [0.011] 0.816 [0.059] 0 [0.001] 0.005 [0.005] -0.009 [0.004] 0.008 [0.013] -0.011 [0.035] -0.01 [0.035] 962 0.21

Testosteron (ng/ml) 0.241 [0.059] -0.099 [0.197] 0.068 [0.059] 0.036 [0.054] 0 [0.002] 0.071 [0.091] -0.04 [0.051] -0.036 [0.077] 0.041 [0.036] 0.06 [0.070] 0.03 [0.034] 0.016 [0.036] 0.047 [0.040] 0.019 [0.034] 0.317 [0.185] -0.015 [0.004] -0.024 [0.014] -0.01 [0.012] 0.047 [0.042] 0.071 [0.108] 0.165 [0.111] 953 0.11

F(15,940)=1.00 0.45

F(13,931)=1.20 0.27

Table 7: First Stage

Estrogen/progesterone/DES gestational age(wks) at draw_date Testosteron (ng/ml)

Linear Cortisol (ng/ml) 29.458 [16.261] 1.165 [1.801] 21.6 [5.219]

Piece-Wise Linear cortisol<328 cortisol>328 28.393 1.065 [10.898] [8.432] 1.389 -0.224 [1.207] [0.934] 13.089 8.512 [3.498] [2.706]

Maternal Age black other race maternal education
-10.723 [9.452] -21.051 [31.348] -1.537 [9.330] 1.737 [8.595] -0.743 [0.245] -1.156 [14.410] -0.321 [8.080] 9.508 [12.279] -3.545 [5.729] 3.275 [11.136] 1.165 [5.429] 7.124 [5.677] 6.515 [6.423] -5.378 [5.412] 0.575 [32.284] -2.027 [0.624] -3.441 [2.279] 1.037 [1.971] 4.48 [6.658] -4.282 [6.321]

-12.333 [6.335] -15.471 [21.009] 4.016 [6.253] 4.084 [5.760] -0.487 [0.164] -0.784 [9.657] 1.264 [5.415] 5.618 [8.229] -2.903 [3.840] 2.42 [7.463] 3.568 [3.638] 2.089 [3.805] 3.424 [4.305] -3.844 [3.627] -1.819 [21.636] -1.28 [0.418] -2.623 [1.528] 1.264 [1.321] 1.501 [4.462] -5.499 [4.236]

1.61 [4.901] -5.58 [16.255] -5.554 [4.838] -2.347 [4.457] -0.256 [0.127] -0.371 [7.472] -1.585 [4.190] 3.89 [6.367] -0.643 [2.971] 0.855 [5.774] -2.402 [2.815] 5.035 [2.944] 3.091 [3.331] -1.534 [2.807] 2.394 [16.741] -0.748 [0.324] -0.818 [1.182] -0.228 [1.022] 2.979 [3.452] 1.217 [3.278]

953 0.1

953 0.1

953 0.05

Teen Mother Maternal Age >=35 Observations R-squared Standard errors in brackets

Linear Cortisol (ng/ml) 33.413 [16.373] 1.671 [1.813]

Piece-Wise Linear cortisol<328 cortisol>328 29.957 1.872 [10.911] [8.438] 1.792 0.036 [1.207] [0.933]

-2.371 [0.624] -5.196 [9.438] -22.576 [31.616] -0.045 [9.403] 2.552 [8.667] -0.736 [0.247] 0.508 [14.529] -1.316 [8.146] 8.927 [12.384] -2.879 [5.777] 4.616 [11.227] 1.886 [5.473] 7.439 [5.726] 7.471 [6.475] -4.748 [5.457] 4.972 [32.545]

-2.2 [0.530] -8.969 [6.284] -14.099 [21.057] 2.769 [6.299] 3.23 [5.787] -0.466 [0.165] -1.855 [9.735] 0.597 [5.426] 4.024 [8.252] -2.067 [3.847] 2.999 [7.491] 3.865 [3.643] 2.301 [3.811] 2.132 [4.346] -3.56 [3.636] 3.607 [21.683]

-1.46 [0.410] 3.858 [4.859] -4.541 [16.284] -6.224 [4.871] -3.002 [4.475] -0.238 [0.127] -0.651 [7.528] -2.108 [4.196] 2.788 [6.381] -0.089 [2.975] 1.003 [5.793] -2.19 [2.817] 5.183 [2.947] 2.188 [3.361] -1.42 [2.812] 6.14 [16.767]

-3.932 [2.296] 0.842 [1.987] 5.521 [6.710] -5.901 [6.363] 6.73 [10.440] 49.54 [14.20] 953 0.08

-2.446 [1.536] 0.871 [1.325] 1.589 [4.472] -6.926 [4.240] 5.897 [6.995] 28.805 [9.513] 953 0.1

-0.728 [1.188] -0.473 [1.025] 3.101 [3.458] 0.226 [3.279] 0.83 [5.409] 20.735 [7.357] 953 0.05

Table 8: Piece-Wise Linear Impact of Maternal Cortisol on 4 Year IQ

cortisol*I(cortisol<328) (cortisol-328)*I(cortisol>328) black other race maternal education
OLS 0 [0.009] -0.02 [0.011] -5.13 [1.553] -3.37 [5.194] -10.402 [1.548] -6.698 [1.424] 0.091 [0.041] 4.597 [2.386] -3.163 [1.337] 6.452 [2.035] -0.34 [0.940] -3.757 [1.845] -3.842 [0.900] -2.976 [0.938] -4.369 [1.062] 1.344 [0.892] -0.788 [4.878] 0.234 [0.103] 0.142 [0.377] -0.514 [0.326] -1.713 [1.103] -1.095 [1.047] 953 0.25

IV 0.206 [0.150] -0.48 [0.290] -1.368 [3.594] -2.486 [8.835] -13.688 [3.344] -8.515 [2.663] 0.075 [0.081] 4.833 [3.995] -4.255 [2.345] 7.14 [3.463] -0.184 [1.581] -3.689 [3.096] -5.59 [1.888] -1.204 [1.977] -3.631 [1.881] 1.667 [1.553] -2.75 [8.467] 0.125 [0.227] 0.277 [0.679] -0.855 [0.590] -0.611 [1.995] 0.576 [2.105] 953

LIML 0.225 [0.163] -0.522 [0.320] -1.025 [3.840] -2.413 [9.255] -13.99 [3.565] -8.681 [2.813] 0.073 [0.085] 4.854 [4.184] -4.356 [2.466] 7.205 [3.629] -0.17 [1.656] -3.681 [3.242] -5.75 [2.008] -1.039 [2.103] -3.561 [1.977] 1.695 [1.628] -2.921 [8.878] 0.115 [0.240] 0.288 [0.713] -0.886 [0.622] -0.508 [2.102] 0.727 [2.230] 953

Table 9: Impact of Maternal Cortisol on 4 Year IQ - Alternative IV

cortisol*I(cortisol<328) (cortisol-328)*I(cortisol>328) teenage mother maternal age 35 + black other race maternal education
IV 0.205 [0.157] -0.527 [0.295] -3.51 [3.205] 0.392 [3.247] -1.139 [3.774] -2.77 [9.355] -13.645 [3.401] -8.628 [2.779] 0.065 [0.088] 5.411 [4.271] -4.445 [2.455] 7.313 [3.698] -0.229 [1.674] -3.842 [3.278] -5.654 [1.987] -0.957 [2.003] -3.422 [1.983] 1.537 [1.645] -2.48 [8.995] 0.176 [0.762] -0.842 [0.616] -0.347 [2.121] 0.56 [2.218] 953

LIML 0.222 [0.168] -0.564 [0.317] -3.481 [3.331] 0.471 [3.375] -0.823 [3.982] -2.667 [9.715] -13.896 [3.575] -8.767 [2.901] 0.063 [0.091] 5.416 [4.433] -4.526 [2.555] 7.387 [3.843] -0.213 [1.738] -3.869 [3.403] -5.803 [2.089] -0.82 [2.101] -3.364 [2.062] 1.563 [1.709] -2.632 [9.347] 0.167 [0.793] -0.867 [0.642] -0.25 [2.213] 0.702 [2.326] 953

Table 10: Piece-Wise Linear Impact of Maternal Cortisol on 7 Year Verbal IQ

cortisol*I(cortisol<328) (cortisol-328)*I(cortisol>328) black other race maternal education
OLS 0.003 [0.008] -0.014 [0.010] -4.829 [1.363] -14.426 [4.504] -9.448 [1.370] -5.497 [1.257] 0.102 [0.036] 0.785 [2.132] -2.44 [1.164] 6.309 [1.775] 1.278 [0.826] -0.865 [1.628] 2.229 [0.789] -0.889 [0.820] -3.813 [0.928] -0.727 [0.782] -7.221 [4.231] 0.242 [0.091] 0.069 [0.337] -0.52 [0.301] -1.748 [0.962] 0.989 [0.920]

LIML 0.206 [0.126] -0.328 [0.255] -1.674 [2.959] -12.778 [6.887] -12.328 [2.938] -7.326 [2.285] 0.114 [0.064] 1.637 [3.272] -3.386 [1.899] 6.751 [2.737] 1.603 [1.261] -1.019 [2.462] 0.919 [1.483] 0.141 [1.617] -3.484 [1.509] -0.021 [1.258] -10.308 [6.657] 0.262 [0.182] 0.4 [0.548] -0.919 [0.534] -1.125 [1.586] 2.354 [1.631]

933 0.27

933

teenage mother maternal age 35+ Observations R-squared Standard errors in brackets

LIML2 0.209 [0.147] -0.485 [0.276] -0.941 [3.500] -13.248 [8.431] -13.552 [3.354] -8.024 [2.701] 0.077 [0.078] 1.688 [4.017] -3.812 [2.272] 7.122 [3.370] 1.711 [1.548] -0.905 [3.009] 0.607 [1.779] 0.977 [1.848] -3.213 [1.812] -0.235 [1.538] -9.186 [8.164]

0.351 [0.682] -1.086 [0.630] -0.651 [1.930] 2.372 [1.965] 0.354 [2.906] 5.064 [3.295] 933

Table 11: Probit and IV Probit Estimates of Impact of Cortisol on Probability HS Drop Out

Cortisol

Probit 0.007 [1.75]

Marginal Effect 0.00007

cortisol*I(cortisol<328) (cortisol-328)*I(cortisol>328) Black maternal education
0.195 [0.92] -0.078 [0.3] -0.412 [1.5] -0.011 [1.19] -0.683 [1.5] 0.358 [1.86] -0.777 [2.08] -0.012 [0.07] 0.322 [1.89] 0.151 [0.94] 0.252 [1.08] -0.016 [0.64] 0.112 [1.91] 0.000 [0.01] 0.023 [0.13] 0.054 [0.32] 0.263 [1.86] 873

0.0144 -0.0050 -0.0258 -0.0010 -0.0244 0.0334 -0.0512 -0.0009 0.0252 0.0136 0.0222 -0.0020 0.0063 0.0002 0.0059 0.0020 0.0167

Probit

Marginal Effect

-0.001 [0.34] 0.003 [1.8] 0.188 [0.81] -0.052 [0.17] -0.423 [1.39] -0.017 [2.23] -0.800 [1.66] 0.441 [2.32] -0.845 [2.19] -0.018 [0.1] 0.405 [2.47] 0.196 [1.2] 0.379 [1.69] -0.035 [1.75] 0.102 [1.57] 0.009 [0.18] 0.099 [0.55] 0.033 [0.17] 0.274 [1.82] 873

-0.00003 0.0002 0.0127 -0.0030 -0.0246 -0.0010 -0.0242 0.0354 -0.0498 -0.0011 0.0259 0.0122 0.0223 -0.0021 0.0060 0.0005 0.0058 0.0019 0.0162

IVPROBITMarginal Effect

-0.00017 [0.02] 0.018 [2.45] 0.065 [0.3] 0.037 [0.15] -0.289 [1.09] -0.008 [0.71] -0.578 [1.22] 0.355 [1.75] -0.647 [1.57] -0.010 [0.07] 0.333 [1.66] 0.054 [0.31] 0.190 [0.75] -0.010 [0.34] 0.084 [1.43] 0.019 [0.44] -0.005 [0.03] 0.023 [0.15] 0.200 [1.38] 873

-0.0000253 0.0027693 0.0100376 0.0055 -0.0429 -0.0012 -0.0595 0.0632 -0.097 -0.0015 0.0522 0.008 0.0285 -0.0015 0.0127 0.00285 -0.00074 0.0034 0.0301

Appendix Table 1: Sargan Test of Overidentifying Restrictions

Estrogen/Progesterone prescribed gestational age(wks) at draw_date Testosterone black other race maternal education
Residuals 1.195 [4.474] -0.245 [0.496] -0.049 [1.438] 0.028 [2.610] -0.138 [8.629] 0.016 [2.567] 0.045 [2.365] 0 [0.068] 0.065 [3.965] 0.064 [2.224] 0.041 [3.384] 0.105 [1.577] -0.006 [3.065] 0.017 [1.494] -0.069 [1.564] 0.081 [1.768] 0.028 [1.491] -1.3 [8.884] 0.003 [0.172] -0.023 [0.628] 0.017 [0.542] -0.022 [1.832] -0.047 [4.724] -0.006 [4.848] 953 0

Figure 1A: Average Cortisol

330 320 310 300 290 280 270 260 250 white/black

married/single

high/low SES

above/below poverty

non teen/teen

white/blue collar

Figure 1b: Cortisol Levels by Race and Poverty Status

310 305 300 295 290 285 280 275 270 265 260

White Black

All

< Poverty

> Poverty

Figure 2: Parental SES & Offspring IQ (age 4)

115 110 105 100 95 90 85 white/black

married/single

high/low SES

above/below poverty

non teen/teen

white/blue collar

0

10

.2

20

.4

30

.6

40

.8

1

50

Figure 3: Relationship between Testosterone and Maternal Characteristics

0

2 4 Testosteron (ng/ml)

0

2 4 Testosteron (ng/ml) maternal score SRA/100

6 Fitted values

0

0

5

20

10

40

15

60

Fitted values

20

age of mother, years

6

0

2 4 Testosteron (ng/ml) maternal education

6 Fitted values

0

2 4 Testosteron (ng/ml) income at birth in 1000s

6 Fitted values

80 90 100 110 120 130

80 90 100 110 120 130

Figure 4: Predicted IQ and Predicted Cortisol by Quartile

160

180 200 220 predicted cortisol - quartile 1

230

240 250 260 predicted cortisol - quartile 2 predicted 4 year IQ

270

Fitted values

80

100

120

80 90 100 110 120 130

Fitted values

140

predicted 4 year IQ

240

280

285 290 295 300 predicted cortisol - quartile 3

predicted 4 year IQ

305

Fitted values

300

350 400 predicted cortisol - quartile 4 predicted 4 year IQ

450

Fitted values

0

0

.001

Density .002 .003

.004

Density .005 .01 .015 .02 .025

Results of Simulation Exercise

100

0

Density .002 .004

.006

200 250 300 350 True_Cortisol

0

100

200 300 Actual Cortisol

400

500

200 300 Noisy_Cortisol

400

500