david lam department of economics and population studies center university of michigan
DESCRIPTION
Impacts of teen fertility on outcomes of teen mothers and their children in South Africa: Evidence from the Cape Area Panel Study. David Lam Department of Economics and Population Studies Center University of Michigan World Bank Workshop on - PowerPoint PPT PresentationTRANSCRIPT
1
David LamDepartment of Economics
and Population Studies CenterUniversity of Michigan
World Bank Workshop on "Tackling Adolescent Reproductive Health: Impacts and
Interventions to Address Them"December 1, 2009
Impacts of teen fertility on outcomes of teen mothers and their children in South Africa: Evidence from the Cape Area Panel Study
Support for this research was provided by the U.S. National Institute of Child Health and Human Development and the William and Flora Hewlett
Foundation.
2
Cally ArdingtonUniversity of Cape Town
Nicola BransonUniversity of Cape Town
David LamUniversity of Michigan
Murray LeibbrandtUniversity of Cape Town
Letícia MarteletoUniversity of Michigan and University of Texas
Vimal RanchhodUniversity of Michigan and University of Cape Town
This presentation draws on a number of papers produced by various combinations of the following project team:
This work was produced as part of the “Global Teams of Research Excellence in Population, Reproductive Health, and Economic Development” sponsored by the William
and Flora Hewlett Foundation and the Population Reference Bureau
3
Background of the Cape Area Panel Study• Study began in 2002 with 4,752 14-22 year-olds
– Collaboration of University of Cape Town and University of Michigan– All areas and all population groups in Cape Town are represented– Integrated survey of education, employment, sexual behavior, health
• Wave 2, 2003 and 2004• Wave 3, 2005
– Successfully reinterviewed about 85% of original young adult sample
• Wave 4, 2006- UCT, Michigan, Princeton collaboration– Tracked all young adults, plus all members of original CAPS
households who were age 50+ in 2006, plus all children of female CAPS young adults
• Wave 5, 2009 (young adult sample, includes HIV testing)• Public access data
– Integrated Wave 1-2-3-4 data available at www.caps.uct.ac.za
4
Teen childbearing in South Africa
• Total Fertility Rate of 2.9 is lowest in Sub-Saharan Africa
• Relatively high rates of teen childbearing– 24% had a birth by age 18; 50% by 20
• Significant fractions of teenage mothers return to school after having their child – Over 50% of 15-17 year-olds with a child were in
school
• Most teen childbearing is non-marital– Only 18% of 20 year old mothers had ever been
married
Source: 2001 South African Census Data
5
Using CAPS to study the impact of teen fertility – three approaches
1. Compare CAPS young adult respondents who were born to teen mothers with those born to non-teen mothers
2. Look at the outcomes of the children of CAPS YA respondents, comparing those with teen versus non-teen mothers
3. Look at the educational outcomes of YA respondents with and without teen births
6
Mother was teen when YA was born
CAPS young adults
Treatment
CAPS young adults
(age 14-22 in 2002)
Mothers of CAPS
young adults
Children of
female CAPS young adults
Analysis 1
Mother was 20+ when YA was born
CAPS young adults
Control
Child of female
CAPS YA
Treatment
Child of female
CAPS YA
Control
CAPS YA had birth as teen
CAPS YA had birth at 20+
Analysis 2
CAPS YA had
teen birth
Treatment
CAPS YA did
not have teen birth
Control
Analysis 3Generation
7
1. Using CAPS young adults (YAs) as children of teen mothers
• CAPS has the mother’s age at YA’s birth for both resident and non-resident mothers
• Information on schooling and other characteristics at each age from birth based on retrospective histories
• Information on household characteristics such as income, parent’s education, and employment status of household members
• Information on up to three YAs in same household – allows sibling fixed effects
8
Sample and Methods• CAPS 2002-2006, young adult sample
• Young Adults as children of teen mothers
• OLS, with and without controls, plus sibling/cousin fixed effects
Total # of Young Adults 3,662
% born to teen mother 14.58%
# of groups (pairs/triplets) 1,045
% (#) with variation on teen mother 21% (221)
Includes all African and Coloured YA’s with mother’s age at their birth
Siblings and cousins in the same household
9
Estimated impact of being born to a teen mother
Outcome Mean OLS
OLS with
controls
Sibling fixed
effects Mean OLS
OLS with
controls
Sibling fixed
effects(1) (2) (3) (4) (5) (6) (7) (8)
[0.372] -0.202*** -0.14** 0.00 [-0.267] 0.003 -0.018* 0.0170.003 0.064 0.107 0.795 0.01 0.018
[0.737] -0.040*** -0.036*** -0.015 [0.756] -0.095 -0.237*** -0.0980.005 0.013 0.021 0.114 0.061 0.08
[0.5] -0.122** -0.097** -0.025 [0.318] 0.019 -0.063 -0.052
0.011 0.043 0.095 0.656 0.043 0.084[0.266] 0.103*** 0.097*** 0.068 [0.105] -0.031* 0 0.008
0.003 0.032 0.058 0.093 0.017 0.035[0.214] 0.062* 0.056* 0.113** [0.192] -0.041 -0.026 -0.064
0.063 0.033 0.047 0.133 0.029 0.058[0.059] 0.034* 0.033* 0.092*** [0.024] -0.036 -0.017 -0.086**
0.058 0.018 0.034 0.182 0.028 0.036
Note: Robust standard errors in italics; significance levels: *=.10, **=.05, ***=.01
Indicator: completed high school by age 20
Dropped out of school by age 16Lived with an alcoholic when growing up
Fear of physical abuse when growing up
Coloured Sample African Sample
Age standardized mathematics score
Rate of grade progression
For coloured sample, those with teen mother have 0.2 standard deviations lower math score than those with non-teen mother.
Controlling for parents’ education, childhood poverty status, and mother’s fertility reduces coefficient by 30%
Comparing siblings with and without teen mothers reduces coefficent to zero.
For African sample there is no unadjusted difference in test scores.
With controls for parents’ education, childhood poverty status, and mother’s fertility there is a difference of -.02 standard deviations.
Comparing siblings with and without teen mothers the coefficient is similar but has larger standard error.
10
CAPS respondents born to teen mothers have younger mothers – as a result their mothers have higher education. Mean education of teen mothers is 1.4 grades higher than for non-teen mothers. This creates a bias in the opposite direction of most studies of teen childbearing.
11
Estimated impact of being born to a teen mother
Outcome Mean OLS
OLS with
controls
Sibling fixed
effects Mean OLS
OLS with
controls
Sibling fixed
effects(1) (2) (3) (4) (5) (6) (7) (8)
[0.372] -0.202*** -0.14** 0.00 [-0.267] 0.003 -0.018* 0.0170.003 0.064 0.107 0.795 0.01 0.018
[0.737] -0.040*** -0.036*** -0.015 [0.756] -0.095 -0.237*** -0.0980.005 0.013 0.021 0.114 0.061 0.08
[0.500] -0.122** -0.097** -0.025 [0.318] 0.019 -0.063 -0.052
0.011 0.043 0.095 0.656 0.043 0.084[0.266] 0.103*** 0.097*** 0.068 [0.105] -0.031* 0.000 0.008
0.003 0.032 0.058 0.093 0.017 0.035
Note: Robust standard errors in italics; significance levels: *=.10, **=.05, ***=.01
Coloured Sample African Sample
Age standardized mathematics score
Rate of grade progressionIndicator: completed high school by age 20
Dropped out of school by age 16
Adjusted point estimates are larger in magnitude than unadjusted estimates for three outcomes for Africans
12
Sensitivity Checks
• Birth order effect versus teen mother effect
• Does teen childbearing affect all the teen mother’s children or only the one born to her as a teen?
12
13
Teen mother versus birth order• High correlation between being born to teen
mother and being the older sibling/cousin• Older siblings may fare better on certain
outcomes due to birth order effects• We restrict sample to YAs not born to teen
mothers and look for birth order effects• In African sample, older siblings/cousins progress
through school faster and are less likely to drop out by age 16
• Older sibling advantage might be masking a negative effect of being born to teen mother
13
14
Are all children born to teens equally affected by the initial teen birth?
• This would explain the small estimated impacts in the Fixed Effects analysis
• To test this we restrict sample to YAs born to older mothers
• We compare YAs who have older siblings/cousins born to a teen mother to YAs who do not– Mostly negative but insignificant results found for the African
sample
– Evidence of lower math scores in coloured sample
• Some support for “systematic difference” hypothesis, implying that FE estimates may not be informative
14
15
Conclusions from Analysis 1• Negative effects of having a teen mother
found for Coloured young adults• Effects decline when we include controls and
disappear when we compare siblings/cousins– Suggests that unadjusted differences result from
adverse pre-birth factors• Effects for Africans become larger when we
include controls– Teen mothers have more education because they
are younger• Effects disappear comparing siblings/cousins
– Might be a result of the fact that first-born children do better on certain outcomes
15
16
2. Health outcomes of children of CAPS respondents
• We compare children born to teen mothers with children born to mothers age 20+
• Using propensity score matching, we estimate weighted regressions with “born to teen mother” as key variable:– Step 1: Estimate the probability of being a teen mother given
pre-childbirth characteristics
– Step 2: Predict the propensity scores
– Step 3: Calculate a set of weights based on these scores to construct a counterfactual from the children born to older mothers group
– Step 4: Estimate the effect of being born to a teen mother using regressions weighted by the constructed weight
17
CAPS data – Timeline & Sample
Wave 1 (2002)4752 young adults (age 14-22)
Wave 2A (2003)1360 young adults (age 15-23) Wave 2B (2004)
2489 young adults (age 16-24)
Wave 3 (2005)all young adults (age 17-25)
Wave 4 (2006-07)All young adults (age 18-26)
pluschildren of female young adults
607 children – African and coloured first born children only
Sample selective of women who begin childbearing early
Majority of teen mothers in their late teens - average age = 17.6
Majority of older mothers in their early 20s - average age = 21.6
18
Proportion teen mother
VariableTeen
motherOlder
mother Diff.Teen
motherOlder
mother Diff.
Childhood household poor or very poor 0.08 0.03 0.05* 0.24 0.31 -0.07 Neighborhood household income (log mean) 10.79 10.83 -0.04 9.94 9.91 0.03 Wave 1 household owns 5 or more books 0.78 0.94 -0.16*** 0.66 0.56 0.10* Wave 1 household per capita income 670.0 747.6 -77.6 332.7 356.7 -24.0
Mother's education 7.54 7.80 -0.26 7.39 7.34 0.05 Father's education 8.11 8.51 -0.40 6.53 6.77 -0.24 Proportion of life lived with mother 0.87 0.87 0.00 0.78 0.77 0.01 Proportion of life lived with father 0.58 0.59 -0.01 0.48 0.50 -0.02 Prop. of life lived with maternal grandparent 0.17 0.18 -0.01 0.22 0.23 -0.01
Drug addict in childhood household 0.19 0.1 0.09* 0.03 0.05 -0.02 Alcoholic in childhood household 0.26 0.18 0.08 0.23 0.19 0.04
Highest grade at age 12 5.62 5.63 -0.01 4.79 4.60 0.19 Failed a grade by age 12 0.30 0.29 0.01 0.26 0.21 0.05 Standardized numeracy and literacy score 0.02 0.00 0.02 -0.14 0.09 -0.23** Number of students in class 39.02 35.4 3.62*** 45.18 43.69 1.49 Age at menarche 13.06 12.92 0.14 13.99 14.58 -0.59***
Sample size 169 119 140 179
Table 3. Mean characteristics of teen mothers and older mothers
0.44African
0.57Coloured
Teen mothers are generally worse off in coloured sample, but teen mothers are better off in African sample on a number of characteristics (marked in red).
19
Sample
sizeLimited controls
Full controls
Propensity score
weightedBirthweight (Z-score) 422 -0.109 -0.141 -0.255* [0.13] [0.15] [0.15]Underweight at birth 412 0.072** 0.071** 0.098*** [0.03] [0.03] [0.03]Current height for age 481 -0.272 -0.561* -1.093** [0.26] [0.32] [0.55]Stunted 481 0.01 0.022 0.157** [0.06] [0.08] [0.07]
Note: Marginal effects based on regressions. Bootstrapped standard errors in brackets; significance levels: *=.10, **=.05, ***=.01
Effect of being born to a teen mother on child health outcomes
Unlike most analyses of teen mothers, we estimate larger effects when we control for characteristics than when we don’t. For some characteristics teen mothers are better off than older mothers.
20
Conclusions of Analysis 2
• We find some evidence that children born to teen mothers are at risk of worse health – More likely to be born underweight– Have lower height for age z-scores and more likely to be stunted
• Unlike previous studies, our results do not suggest that teen mothers are inherently socioeconomically disadvantaged– P-score weighted differences are often larger than unadjusted
differences
• Differences between Africans and coloureds are large– Children born to coloured teen mothers have double the health
disadvantage seen for Africans
21
3. Analyzing the impact of teen fertility on educational outcomes of teen mothers
• This analysis uses only the CAPS young adults, comparing teen mothers to other women
• As in analysis 2, we estimate propensity score weighted regressions.– Step 1: Estimate probits using covariates, with
`treatment’ variable, predict the propensity scores. – Step 2: Match untreated observations to treated
observations, based on the pscores. – Step 3: This generates a set of weights for the
untreated group. (treated obs have weight= 1.)– Step 4: Estimate regressions by OLS weighted by the
(sampling weights x propensity score weights).
22
Table 4: Regression results, coefficients on “teen birth” variableDependent Variable
Matric by age 20
Matric by age 22
Grades by age 18
Grades by age 20
Grades by age 22 Dropout
Specification 1: -0.303*** -0.302*** -0.931*** -1.331*** -1.130*** 0.190***No sample restriction, sampling [0.026] [0.035] [0.10] [0.12] [0.14] [0.020]weights only, no covariates N 1735 1129 2224 1735 1129 2295
R2 0.07 0.07 0.05 0.09 0.07 0.03
Specification 2: -0.208*** -0.227*** -0.620*** -0.921*** -0.801*** 0.147***No sample restriction, sampling [0.025] [0.033] [0.081] [0.11] [0.12] [0.019]weights only, with limited N 1718 1118 2193 1718 1118 2258covariates R2 0.31 0.28 0.45 0.39 0.37 0.16
Specification 3: -0.125*** -0.112*** -0.382*** -0.568*** -0.358*** 0.102***Had sex by age 19, sampling [0.029] [0.037] [0.096] [0.12] [0.13] [0.024]weights only, with all N 1221 810 1467 1221 810 1486covariates R2 0.27 0.26 0.37 0.37 0.4 0.14
Specification 4: -0.100*** -0.0754* -0.348*** -0.510*** -0.274* 0.0825***Had sex by age 19, sampling [0.032] [0.039] [0.11] [0.14] [0.16] [0.026]and propensity score matching N 1218 807 1464 1218 807 1483weights, with covariates and R2 0.24 0.27 0.36 0.34 0.37 0.14common support restriction
Standard errors in brackets *** p<0.01, ** p<0.05, * p<0.1
“Naïve estimate” – no covariates or sample restrictions. Teen mothers are 30 percentage points less likely to have completed grade 12 (matric) by age 20.
Coefficient falls by 31% with controls for household and socio-economic characteristics
Coefficient falls by 58% when we look at sample that had sex by age 19 and add controls for measures of sexual behavior
Coefficient falls by 67%, but is still negative and statistically signifcicant, when we use propensity score matching weights based on the full set of covariates including measures of sexual behavior
23
Table 4: Regression results, coefficients on “teen birth” variableDependent Variable
Matric by age 20
Matric by age 22
Grades by age 18
Grades by age 20
Grades by age 22 Dropout
Specification 1: -0.303*** -0.302*** -0.931*** -1.331*** -1.130*** 0.190***No sample restriction, sampling [0.026] [0.035] [0.10] [0.12] [0.14] [0.020]weights only, no covariates N 1735 1129 2224 1735 1129 2295
R2 0.07 0.07 0.05 0.09 0.07 0.03
Specification 2: -0.208*** -0.227*** -0.620*** -0.921*** -0.801*** 0.147***No sample restriction, sampling [0.025] [0.033] [0.081] [0.11] [0.12] [0.019]weights only, with limited N 1718 1118 2193 1718 1118 2258
covariates R2 0.31 0.28 0.45 0.39 0.37 0.16
Specification 3: -0.125*** -0.112*** -0.382*** -0.568*** -0.358*** 0.102***Had sex by age 19, sampling [0.029] [0.037] [0.096] [0.12] [0.13] [0.024]weights only, with all N 1221 810 1467 1221 810 1486covariates R2 0.27 0.26 0.37 0.37 0.4 0.14
Specification 4: -0.100*** -0.0754* -0.348*** -0.510*** -0.274* 0.0825***Had sex by age 19, sampling [0.032] [0.039] [0.11] [0.14] [0.16] [0.026]and propensity score matching N 1218 807 1464 1218 807 1483weights, with covariates and R2 0.24 0.27 0.36 0.34 0.37 0.14common support restriction
Standard errors in brackets *** p<0.01, ** p<0.05, * p<0.1
Similar declines in coefficients using other outcome measures
24
Enrollment by age, women with different ages at first birth
African females
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
12 13 14 15 16 17 18 19 20
Age
Pro
po
rtio
n e
nro
lled
First pregnancy age 23+First pregnancy at age 18First pregnancy at age 17
First pregnancy at age 16First pregnancy at age 15
50% of those with a birth at age 15 are enrolled in school 1 and 2 years later
Those with teen births already had lower enrollment rates at age 14
25
Enrollment by age, women with different ages at first pregnancy
African females
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
12 13 14 15 16 17 18 19 20
Age
Pro
po
rtio
n e
nro
lled
First pregnancy age 23+First pregnancy at age 18First pregnancy at age 17
First pregnancy at age 16First pregnancy at age 15
Dropout rate after pregnancy is much higher for coloured teens than for African teens
This may partly reflect the high rates of grade repetition in African schools, which reduce stigma of going back to school
Coloured females
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
12 13 14 15 16 17 18 19 20
Age
Pro
po
rtio
n e
nro
lled
First pregnancy age 23+First pregnancy at age 18First pregnancy at age 17First pregnancy at age 16First pregnancy at age 15
26
Grades completed by age, women with different ages at first birth
African females
4
5
6
7
8
9
10
11
12
12 13 14 15 16 17 18 19 20
Age
Gra
des
co
mp
lete
d
First pregnancy age 23+
First pregnancy at age 18
First pregnancy at age 17
First pregnancy at age 16
First pregnancy at age 15
Similar schooling at age 20 for those with birth at age 15, 16, or 17
27
Grades completed by age, women with different ages at first pregnancy
Coloured teens gain very little schooling after pregnancy, while Africans continue to pass grades after pregnancy.
Coloured females
4
5
6
7
8
9
10
11
12
12 13 14 15 16 17 18 19 20
Age
Gra
des
co
mp
lete
d
First pregnancy age 23+First pregnancy at age 18First pregnancy at age 17First pregnancy at age 16First pregnancy at age 15
African females
4
5
6
7
8
9
10
11
12
12 13 14 15 16 17 18 19 20
Age
Gra
des
co
mp
lete
d
First pregnancy age 23+
First pregnancy at age 18
First pregnancy at age 17
First pregnancy at age 16
First pregnancy at age 15
28
Conclusions from Analysis 3• A large proportion of the mean difference in
schooling disadvantage of teen mothers is accounted for by pre-existing covariates.
• There remain negative and statistically significant effects of teen births on educational attainment after controlling for observeable characteristics.
• Some evidence that there may be heterogeneity depending on actual age at first birth. Schooling often does continue after teen birth.