decomposing intergenerational income elasticity
DESCRIPTION
Decomposing Intergenerational Income Elasticity. The gender-differentiated contribution of capital transmission in rural Philippines. Leah Bevis & Christopher B. Barrett, Cornell University Calvin College Summer Seminar on the Economics of Global Poverty August 2013. Background. - PowerPoint PPT PresentationTRANSCRIPT
Decomposing Intergenerational Income Elasticity
The gender-differentiated contribution ofcapital transmission in rural Philippines
Leah Bevis & Christopher B. Barrett, Cornell UniversityCalvin College Summer Seminar
on the Economics of Global PovertyAugust 2013
Background• Equality of socio-economic opportunity
o Often proxied by intergenerational income elasticity (IGE) estimates:
• Multiple possible pathways behind IGE and the pathways matter to policy design:o Intergenerational transmission of educationo Intergenerational transmission of healtho Intergenerational land transfers o Assortative marriageo Migrationo Productivity due to correlated unobservables
• Yet there has been very limited exploration of multiple possible pathways, esp. in developing countries
ijjoij eybby 1
2/17
• How high is IGE/socio-economic mobility in rural Philippines?
• Which pathways account for estimated IGE?
• Do these pathways vary by child or parent gender?
• Does migration affect capital transmission or income transmission?
Core questions
3/17
Conceptual ModelPer Becker-Tomes model, child adult income results from parental investment in child capital stocks and productivity.
Parental capital stocks may affect child capital stocks in 3 ways:
1) PC directly transmits to child capital (e.g., health, land)
2) PC affects parental income, which constrains investment in child capital (e.g., education) given borrowing constraints
3) PC affects parental preferences and expectations, which affect investment in child capital (e.g., marriage, education)
Also may be intergenerational productivity correlation due to genetics, natural/social environment, unobserved skills, etc.
Parent Land
Parent Education
Parent Productivity
Parent Health
Parent Income
Child Land Child /Spouse Education
Child Productivity
Child Health
Child Income
Conceptual ModelNaïve regression vs. pathway decomposition
Capital transmission pathways
Income transmission pathways
1) Estimate naïve IGE, w/ and w/o correction for measurement error in and transitory shocks to parent income (instrument w/initial period parent expenditure). Strategy resolves downward bias due to using short-term income.
2) Estimate intergenerational capital transmission
where E=education, H=health, L=land, S=spouse education, X=other covariates, ij is child, j is parent. Use OLS-IV.
ijL
ijL
jLLj
LHj
LEj
Lij eXLHEyL
ijS
ijS
jSLj
SHj
SEj
Sij eXLHEyS
5/17
ijH
ijH
jHLj
HHj
HEj
Hij eXLHEyH
ijE
ijE
jELj
EHj
EEj
Eij eXLHEyE
Estimation Strategy
ijjoij eybby 1
3) Estimate IGE via OLS-IV using different specifications to isolate pathways of transmission: i)Here π captures productivity transmission independent of child capital accumulation. λi estimates are returns to child capital. ii)where μ captures both direct productivity and indirect liquidity effects of parental income. λ are returns to parent capital.iii)allows testing of the exclusionary restriction (λp2 = 0) that parental capital has no direct effect beyond that on child capital accumulation and productivity transmission.
ijy
ijy
ijLiij
Hiij
Eijij eXLHEyy 1111
ijy
ijy
jL
jH
jE
jij eXLHEyy 2111
ijy
ijy
ijLiij
Hiij
Eij
Lj
Hj
Ejij eXLHELHEyy 3222222
5/17
Estimation Strategy
• Bukidnon: a rural, landlocked province of southern Philippines
• Gathered over two decades:o 1984: 448 families relying primarily on
agricultural income, largely sugar, corn or rice
o 2003/2004: revisited original families, tracked children to new homes in local, peri-urban & urban locations
• “Split” vs “migrant” children o As children, not significantly different
except by gender and birth order.o By adulthood migrants better
educated, wealthier
Data
7/17
Data
8/17
Mean Values Daughters
Mean Values Sons
Mean Values Migrants
Mean Values Non-Migrants
Child Age (years) ‘84 9.6 10.4 9.5 10.3
Father Age (years) ‘84 40 40 40 40
Mother’s Education (years) ‘84 5.9 5.6 5.9 5.6
Father’s Height (years) ’84 161 161 161 161
Parent Landholdings (hectares) ‘84 2.3 2.6 2.6 2.3
Parent Weekly Income (Philippine Peso) ‘84 270 301 288 279
Child Age (years) ‘03 29 30 29 29
Child Household Size (persons) ‘03 ` 7.2 4.6 5.3
Child Education (years) ‘03 9.7 8.6 9.8 8.8
Spouse Education (years) ‘03 9.3 10.1 10.2 9.1
Child Height (cm) ‘03 150 163 155 156
Child Landholdings (hectares) ‘03 0.1 0.3 0.2 0.2
Child Weekly Income (Philippine Peso) ‘03 1830 1805 2439 1326
Naïve IGE Estimates
9/17
Once control for permanent income and life cycle effects, IGE much higher than w/o controls and higher than OECD countries. No stat. sig. difference b/n daughters/sons, migrants/non-migrants
Table A4: IGE across Various Estimation Approaches Estimator:
(3) OLS
(4) IV
Dependent Variable: Log Child Inc Log Child Inc Independent Variable: Avg Parent Inc Parent Inc ‘84 Instrument: NA Parent Expen ‘84 IGE All Children: 0.262*** 0.500*** (0.0538) (0.120) IGE Daughters: 0.248*** 0.537*** (0.0633) (0.163) IGE Sons: 0.269*** 0.434*** (0.0877) (0.146) IGE Splits: 0.280*** 0.474*** (0.0571) (0.103) IGE Migrants: 0.238*** 0.574*** (0.0739) (0.197)
Robust standard errors in parentheses Father’s age and child’s age are controlled for quadratically in all regressions
*** p<0.01, ** p<0.05, * p<0.1
Capital Transmission Pathways
9/17
Key results:- Liquidity effects limited. Parental income exerts a significant positive effect only on daughters’ (and sons-in-law) education.
- Direct intergenerational human capital transmission is key. Esp. mothers’ human capital on both sons and daughters: children’s education and height, and their spouse’s education.
- Parent land affects no child capital stocks, perhaps due to 1988 land reform and mass exits from agriculture 1984-2004.
- Mother’s education negatively related to daughters’ height, seemingly through effect on maternal labor supply and thus on child feeding practices.
9/17
Capital Transmission PathwaysTable 2: Intergenerational Capital Transmissions for Daughters (OLS-IV)
(1) (2) (3) (4) Daughter
Education Daughter Height
Daughter Landholdings
Daughter Spouse Education
Log Parent Income ‘84 1.642** 0.763 0.231 1.949** (0.702) (2.882) (0.169) (0.909) Parent Land ‘84 0.0727 -0.121 -0.00330 -0.0398 (0.0604) (0.408) (0.0192) (0.0995) Mother’s Education ‘84 0.348*** -0.578* -0.00206 0.159* (0.0807) (0.351) (0.0131) (0.0833) Father’s Education ‘84 0.149* 0.559 0.00871 0.258*** (0.0846) (0.359) (0.0143) (0.0860) Mother’s Height ‘84 -0.0837** 0.438*** -0.0102 -0.0158 (0.0352) (0.107) (0.00652) (0.0410) Father’s Height ‘84 0.0353 0.0460 -0.00635 0.00308 (0.0346) (0.195) (0.00626) (0.0421)
Son
Education Son
Height Son
Landholdings Son
Spouse Education Log Parent Income ‘84 1.311 -0.330 -0.132 -1.162 (1.211) (1.850) (0.389) (1.325) Parent Land ‘84 0.194 0.236 0.0752 0.224 (0.148) (0.241) (0.0674) (0.174) Mother’s Education ‘84 0.268** -0.190 0.0650** 0.321*** (0.120) (0.234) (0.0326) (0.121) Father’s Education ‘84 0.147 0.264 -0.0436 0.0505 (0.115) (0.206) (0.0462) (0.115) Mother’s Height ‘84 -0.00627 0.362*** 0.00531 0.0202 (0.0509) (0.0963) (0.0127) (0.0447) Father’s Height ‘84 -0.00461 0.296*** -0.0224 -0.00566 (0.0469) (0.0746) (0.0139) (0.0436)
Robust standard errors in parentheses Controls include household size, gender-specific birth order dummies,
location (barrio) & ethnic groups dummies
10/17
Income Transmission Pathways
- Naïve IGE estimates statistically insignificantly different between daughters and sons, migrants and splits (0.43-0.57).
- But once we control for capital transmission, very different pathways appear by child gender.
- In both cases, landholdings and spouse education strongly affect children’s adult income – and is each strongly associated with parent land and maternal education, respectively. But the estimated marginal effects of landholdings (spouse education) are far higher for daughters (sons).
- For daughters, IGE runs primarily through intergenerational productivity correlation, while for sons no such effect exists.
10/17
Daughters’ Income Transmission Pathways
Very strong intergenerational productivity/unobserved capital transmission effect.
Daughters’ landholdings and (own and spouse) education also play a role.
Some residual effect of maternal human capital and parental landholdings.
Table 5: Decomposing Intergenerational Income Elasticity for Daughters (OLS-IV) (1) (5) Income Income Parent Income ‘84 0.537*** 0.779** (0.163) (0.344) Parent Land ‘84 -0.0558** (0.0267) Mother’s Education 0.0614* (0.0357) Father’s Education -0.0325 (0.0275) Mother’s Height 0.0359*** (0.0114) Father’s Height -0.00619 (0.0117) Own Education 0.0632* (0.0346) Spouse Education 0.0356* (0.0193) Own Height -0.00605 (0.00730) Landholdings 0.307*** (0.112) Age Controls: Yes Yes Additional Controls: No Yes Observations 240 220 R-squared 0.114 0.433
Robust standard errors in parentheses Age controls include quadratic terms for child and father age
Additional controls include parent household size, gender-specific birth order dummies, location (barrio) & ethnic groups dummies
*** p<0.01, ** p<0.05, * p<0.1
10/17
Sons’ Income Transmission Pathways
By contrast, no intergenerational productivity transmission effect.
Sons’ landholdings and (own and spouse) education key income determinants.
Some residual (negative?) effect of parental human capital.
(1) (5) Income Income Parent Income ‘84 0.434*** -0.00884 (0.146) (0.385) Parent Land ‘84 -0.0285 (0.0459) Mother’s Education 0.0511 (0.0382) Father’s Education -0.00327 (0.0312) Mother’s Height -0.0598*** (0.0186) Father’s Height -0.0245* (0.0132) Own Education 0.0367* (0.0223) Spouse Education 0.131*** (0.0229) Own Height 0.0217 (0.0140) Landholdings 0.161*** (0.0605) Age Controls: Yes Yes Additional Controls: No Yes Observations 182 157 R-squared 0.050 0.594
Robust standard errors in parentheses Age controls include quadratic terms for child and father age
Additional controls include parent household size, gender-specific birth order dummies, location (barrio) & ethnic groups dummies
*** p<0.01, ** p<0.05, * p<0.1
10/17
Which Pathways Explain Most IGE?
Use change in r2 from adding variable to naïve specification as upper bound estimate and from dropping variable from full specification as lower bound in establishing relative importance of different paternal capital/income pathways in explaining adult child income. Punch line: maternal education biggest factor, but much variation.
Table 9: Pathway Explanatory Power Model Sub-Sample Proportion of 𝑅2Explained by Each Parent Capital Level
Productivity Parent Land Maternal Education
Paternal Education
Maternal Height
Paternal Height
Naïve IGE Regression
Daughters 0.011 0.448 0.407 0.128 0.006 Sons 0.101 0.366 0.383 0.147 0.003
Migrants 0.104 0.631 0.265 0.000 0.000 Splits 0.001 0.438 0.535 0.007 0.020
IGE Decomposition
Daughters 0.780 0.001 0.149 0.001 0.067 0.002 Sons 0.117 0.179 0.222 0.064 0.309 0.101
Migrants 0.028 0.008 0.679 0.007 0.035 0.244 Splits 0.508 0.069 0.127 0.060 0.015 0.221
1. IGE is high (~0.5) in rural Philippines2. But there are sharp gender differences in the pathways
behind these IGE estimates.3. For sons, IGE operates through land and spouse education
capital transmission, with some residual (negative!?) association with parental human capital.
4. For daughters, IGE also operates through human capital transmission, but with some residual association with maternal human capital and very strong intergenerational productivity transmission.
5. Parent income does not affect most child capital levels, which are driven primarily by direct transmission of parent human capital stocks. But education is a normal good.
Conclusions
11/17
6. Parental landholdings play a surprisingly negligible role in intergenerational income transmission, perhaps due to major land reform episode.
7. Marriage markets are crucial institutions limiting equality of opportunity.
8. Patterns also differ for migrants and non-migrants (not shown due to time).
9. Mothers’ human capital is important and transmits relatively equally to sons and daughters. Fathers’ human capital much less important and rarely equal to both sons and daughters.
Conclusions
11/17
Thank you for your time,attention and comments