Household Income and Investments in Child Health and Education in Ivory Coast

1
Household Income and Investments in Child Health
and Education in Ivory Coast

• Denis Cogneau (DIAL, Paris School of Economics)
• Remi Jedwab (Paris School of Economics)

2
Introduction

• Education and child-related health demand role
  of household income ? endogenous.
• S a b x Income e
• The blindly estimated b will differ from the
  true coefficient b. Solution 1 controlling
  variables (parental education, supply, etc.).
  Solution 2 IV adequate as long as the set of
  instruments is uncorrelated with the error term
  in the main equation (or weak instruments).

3
Past Results

• Education (i) developed countries small
  positive or nil effect (Blau 1999, Acemoglu and
  Pischke 2001), (ii) developing countries high
  positive effect (Behrman and Knowles 1997,
  Cogneau and Maurin 2001)
• Health (i) developed countries health gradient
  (Case, Lubotsky and Paxson 2001), (ii) developing
  countries same (Duflo 2003)
• Related literature direct impact of income
  shocks as a result of insurance missing markets
  (Jacoby and Skoufias 1997, Thomas and al. 2003),
  impact of income volatility (Fitzsimons 2003,
  Kazianga 2005)

4
Our Natural Experiment

• Ivory Coast first cocoa-producing country in the
  world. Internally, fixed producer price by the
  state till 1998.
• From the 1980s, fall in international cocoa
  prices. 1990 first diminution of the national
  cocoa producer price in 25 years. 1994 and 1997
  increase of the same price.
• Conclusion exogenous strong decrease in the
  income of cocoa households vs. non-cocoa
  households ? differential impact on investments
  in human capital.
• Outcomes school attendance 5-17 yo, health
  status 0-18 yo and anthropometric status 0-5 yo.

5
Our Natural Experiment

• Data cross-section 1985-1988 (CILSS), 1993 and
  1998 (EP)

National fixed producer price
survey wave
1993
1998
1988
year
1985 1990 1994
1997
6
DiD-IV Strategy

• Cocoa households more affected than non-cocoa
  households ? relatively less income in 1993 vs.
  1988 ? relatively less education and health for
  their children (conversely for 1993-1998)
• Econometric specification instrument income with
  belonging to a cocoa-producing household in
  1993/1998 vs. 1988 (difference-in-difference
  with a treatment group and a control group).
  Other examples of DiD-IV Duflo 2000, Chen 2004.

7
Econometric model
For child i in household h in village v at time
t
8
Western and Eastern Forests main cocoa regions
9
Limitation 1 non-parallel trends?

• Since we control for belonging to
  cocoa-producing household, a bias is allowed
  between cocoa and non-cocoa households.
• However, better if rather similar. Guarantees
  that non-cocoa households would have equally
  modified their human capital investments
  decisions if they had faced the same income shock
  that cocoa households. Control group non-cocoa
  farmers.
• Many tests available.

10

• 1) T-tests on observable general characteristics
  1988

11

• 2) T-tests on observable specific characteristics
  1988

Once we regress active/abandon/number of inactive
days due to illness on belonging to a cocoa
household village-time fixed effects, cocoa is
non-significant.
12
Same height-for-age Z-score kernel distribution
in both groups 1988
13
Limitation 2 change in supply?

• Supply (quality of schools and health centers) is
  likely to change more in cocoa-dominant villages,
  since aggregate income is relatively lower there.
  
• Always inclusion of village-time fixed effects.

14
Limitation 3 geographical migration?

• E.g. problematical if differential migration
  (between both groups) from forest regions to
  other regions due to the negative shock
  1990-1993.

15
Limitation 4 professional migration?

• E.g. problematical if some cocoa households have
  switched to non-cocoa production due to the
  negative shock 1990-1993. Irreversibility story
  cocoa trees only mature after 5-10 years.

16
Econometric model
For child i in household h in village v at time
t
17
(No Transcript)
18
Limitation 5 other consequential changes than
income?

• E.g. problematical if the fall of cocoa prices
  alters cocoa-specific anticipated returns to
  education or borrowing constraints?
• This is very unlikely. Returns to education are
  presumably unrelated to opportunities in
  agriculture. Borrowing constraints do not depend
  upon producer prices or income, but assets and
  social capital (unchanged in the short term).
• Any change in schooling is thus only attributable
  to changes in income!

19
Econometric model
For child i in household h in village v at time
t
20
Econometric Results

1. School Attendance
2. Health Status
3. Anthropometric Status

21
(No Transcript)
22
School Attendance Linear probability model.
Controls dummies equal 1 if the household chef
is a woman, literate, has achieved primary
schooling, has spent at least one year in
secondary schooling, has obtained the BEPC (at
the end of secondary schooling), has obtained the
BAC or more (at the end or after high school),
dummies equal 1 if the household owns livestock,
a business, has a household member being a civil
servant, the ratio of extended family over
household size and the share of women in the
household. Obs. 5-17, All 18992. Obs. 5-11,
All 11782. Obs. 5-17, Forest 9111. Obs. 5-17,
Forest 5802.
23
Children 0-18
24
Health Status Linear probability model.
Controls dummies equal 1 if the household chef
is literate, has achieved primary schooling, has
spent at least one year in secondary schooling,
has obtained the BEPC (at the end of secondary
schooling), has obtained the BAC or more (at the
end or after high school), dummies equal 1 if the
household owns livestock, a business, has a
household member being a civil servant, and the
share of women in the household. Obs. All
28002. Obs. Forest 13681.
25
(No Transcript)
26
Anthropometric Status Controls dummies equal
1 if the household chef is literate, has achieved
primary schooling, has spent at least one year in
secondary schooling, has obtained the BEPC (at
the end of secondary schooling), has obtained the
BAC or more (at the end or after high school),
dummies equal 1 if the household owns livestock,
a business, has a household member being a civil
servant, or belongs to a certain ethnical group
(Akan, Krou, Mande, Voltaque, Other, Foreigner),
and the share of women in the household. Obs.
0-5, Forest 2824. Obs. 3-5, Forest 1434.
27
Conclusion

• A 10 increase in parental income (a) a 2.3 - 3
  point increase in the likelihood to attend school
  for 5-17 year-old children, (b) a 6 point
  decrease in the likelihood that a 3-5 year-old
  child is malnourished, and (c) a 1.8 - 2 point
  reduction in the probability that a 0-18 year-old
  child is sick.
• Our contribution
• (1) OLS are downward biased in LDC, robust to
  within and not due to weak instruments. Why
  (given that theoretical models emphasize an
  upward ability bias)? (i) Simultaneity (ii)
  Measurement error (iii) Omission bias.
• (2) Imperfect credit and insurance markets.
• (3) One of the first papers to empirically
  address child-related health demand.