ECON 980o: Health, Education and Development Lecture 4 October 9, 2008

ECON 980o: Health, Education and Development

Lecture 4October 9, 2008

Today’s Class:

Part II: Health production

Does health improve with income?

Today:Do health improvements depend on preferences within

household?

Do health improvements depend on gender preferences of household?

First, back to South Africa …

Two types of household decision-making models

1. Unitary Model

– A household behaves as a collection of individuals with agreed goals of welfare

– Assumes a benevolent decision-maker or equal decision-making power of all members (observationally equivalent)

– Means that expenditure allocations independent of who gets money (household income pooled) resources always get allocated to maximize marginal utility of “neediest” household member

2. Collective model:

– Focus on individuality of household members

– Model explicitly includes individual preferences leading to collective choice (preferences allowed to differ)

– Considers how members reconcile differing preferences (bargaining process)

– Considers variation in decision-making power of individual members, possibly determined by how resources accumulated

Unitary model: Advantages and disadvantages

• Advantages

– Impact of policy on behavior easily analyzed

– Applicable when consensus among HH members prevails

• Disadvantages

– Assumes common preference of members (agreement)

– Assumes household resources pooled (no control issues)

– Basically, just less realistic

Policy implications:Why do we care?

(1) How to measure poverty – individual or households?

If resources are not equally distributed, poverty will be underestimated

(2) How to give public transfers - to individual household members or households?

(should we worry about targeting transfers?)

How do we determine correct model in given setting?

Implications of the unitary versus collective model:

• Income pooling restriction:

If unitary model applies, income is pooled, so outcomes should be independent of who earns or controls money

• What kind of outcomes?

Easier to get predictions if have a prior about differences in male and female preferences

For instance: Male versus female goods (shoes versus cigarettes), children

How can we use South African pension experiment?

• Assume women care more about child health

• Compare weight of children living with female pensioner to weight of children living with male pensioner

• First tool: look at mean differences between implicit comparison groups

• “Descriptive statistics” key component of any analysis

• What are two key comparison groups?

What sets of variables do we want to include on table?

1. Household characteristics (all potential controls)

• Better if no difference in raw means

2. Dependent variable of interest

• NEED difference in raw means!

3. All outcome variables in analysis

• Better if difference in raw means alone

What regression do we run?

• As before, condition on presence of man or woman 55-60

Equation:

Assumption: Households with female pensioner (FP) not so different from households with male pensioner (MP), conditional on having an elderly man or woman in the house

How is this different than the Case assumption?

ijjjjjij MEFEMPFPW 4321

Reading tesults:

What do columns 1-3 tell us?

What is diff between columns 4 & 5?

Why do estimates of λ1 and λ2 increase from column 4 to column 5? (what does this tell us about direction of bias?)

What is purpose of column 7?

What else do they investigate in terms of characteristics of grandmothers and grandfathers (table 4)?

What do they find? What is the interpretation?

Results:• Impact of pension on child weight only when pension

received by woman

• Only find effects for girls

Girls who live with eligible woman experience weight gain of 1.16 SDs relative to girls who do not

Interpretation: Grandmothers protect health status of grandchildren, and either:

(a) care more about girls than boys

(b) equalize health outcomes of boys and girls

Is it possible that there are still unobservable differences between households?

Why?

How could we eliminate this concern?

• Compare within subset of households that have an eligible man or woman

How?

• Look at younger versus older children

Why is this better?

Can we use weight-for-height (WFH) for this?

Why or why not?

Weight for height is flow measure of nutrition, so will respond fast

What is solution?

Solution: Height-for-age (HFA)

Height for age is stock measure of nutrition, so captures long-run effect of inputs such as nutrition

Why might HFA but not WFA be different between older and younger kids?

Idea: Older kids have been exposed to better nutrition a smaller fraction of lives than younger children, so weight-for-age may have caught up even if height-for-age hasn’t had a chance to

Idea:

Ht Ht

}x

}x

Age Age

This DID: [0 – (-x)] = x This DID: [x – (0)] = x

Doesn’t matter if kids start out in same position (before pension), regression estimates will pick up difference in difference (which is same since slope is the same)

What would weight–for-age look like?

Estimating equation:

Difference in difference estimate, older children serve as control group

MP= male pensioner

FP= female pensioner

Y=young kid

If women feed kids more than men when they get the income, what would you expect to see?

ijijjjjjij YYMPYFPMPFPW )()*()*( 54321

Results:

Within-household HFA results same as across-household WFH results – good confirmation

Remaining concern:

1. Changes within the household in response to the pension

What kind of behavior response do we worry about and why?

Endogenous family re-composition

What exactly is the concern here?

Send kids to grandparents, grandparents flee!

What evidence does she present that it is not a problem?

Control for having living eligible grandparentNo difference in WFH using within-household estimate

What do these prove?

Remaining concern:

2. Differences in expected years of pension income for grandmothers

Why might this be a problem?

What evidence does she present that this isn’t a problem?

Conclusions:

(1) Households don’t behave as unitary model would predict

(2) Evidence of gender bias in allocation of health resources in India

.. Related to policy issue of “missing women”

What do we mean by “missing women”?Ratio of men to women in western world is close to unity

Sex ratios at birth favor boys (1.05 boys to 1 girl), but higher male mortality rates → sex ratios of 1 by young adulthood

Meanwhile, population sex ratios in number of Asian countries much higher:

• China: 1.07• India: 1.08• Pakistan: 1.11

Accompanied by:• Substantial excess female mortality• Neglect of female children

Amartya Sen estimates 60–100 million “missing women”

What does it reflect?Many believe severe sex imbalance found in parts of Asia driven

by neglect toward girls exacerbated by extreme poverty

This implies sex ratio will be sensitive to economic conditions

As poverty falls, treatment of girls will improve

Alternatively, sex imbalance could reflect household differences in preferences for girls over boys

What is a testable prediction in that case?

If that’s true, treatment of girls will only improve if female relative to male bargaining power improves

How can we proxy for changes in female to male “bargaining power”?

Qian’s paper:Examine whether treatment of girls sensitive to economic

conditions, and whether bargaining power of husbands relative to wives matters for treatment of girls

Looks at data from China

Puzzle: Sex imbalance in China increasing despite rapid economic growth (does not support poverty only hypothesis)

Important consideration: Sex imbalance in China exacerbated by one-child policy

How does one child policy interact with son preference?

How does she identify a causal effect of economic conditions on sex ratios in China?

Key Insight: In China• Women cultivate tea (comparative advantage)

• Men cultivate orchards (comparative advantage)

Why?

That means:

1. A rise in tea price → ?

→ increase relative income of women

2. A rise in orchard income → ?

→ increase relative income of men

So who does she compare?

Data:

• 1997 Chinese Agricultural Census• 1990 Chinese population census

What is outcome (Y)?

Difference 1: Compare counties in which tea is cultivated (treatment group) to counties in which no tea is cultivated (control group)

Estimator:

Why might this be biased? How does she correct for bias?

Are these counties really comparable?How can we investigate?

1. Visual inspection (are they geographically interspersed?)2. Look at county characteristics, test for mean differences in

tea and non-tea counties

Did she find any? What can she do about that?

Remember: It’s a “quasi-experiment”, not a real experiment, so there are bound to be some differences!

What kind of differences are okay?

Map 1 – Tea Planting Counties in ChinaDarker shades correspond to more tea planted per

household.

Map 2A – Garden and Tea Producing Counties

Tea counties are colored black

Map 2C – Fish and Tea Producing CountiesTea counties are colored black.

Map 2D – Agricultural Density and Tea Producing CountiesTea producing counties are outlined.Shaded counties indicate where the average land per household exceeds 4 mu.

Map 2B – Orchard and Tea Producing CountiesTea counties are colored black.

Use difference-in-differences framework:Difference 1: Counties with and without sex-specific crops that

experienced a value increase due to reforms

Difference 2: Cohorts born before and after 1979

Why does she choose 1979?

Natural experiment

Use exogenous variation in agricultural income caused by two post-Mao reforms (1978-1980)

What did these reforms do?

Caused “quasi-random” price changes in sex-specific agricultural products

First step: Look in data

Is there an apparent effect of reforms on tea income?

On sex ratios?

Key Insight

At what age should they diverge?

They diverge around age 11 (post reform)

Calculating the experimental effect with DID:Before

1979 (B)After 1979

(A)Change

Tea counties (T)

Non-tea counties (C)

N

i

BTiYN 1

1

M

j

BCjYM 1

1

So what is the effect equal to?

DID Estimator:

Do we need two periods to get unbiased estimate (why not 1 dif)?

N

i

ATiYN 1

1

M

j

ACjYM 1

1

)(1

11

N

i

BTi

N

i

ATi YY

N

)(1

)(1

1111

M

j

BCj

M

j

ACj

N

i

BTi

N

i

ATi YY

MYY

N

)(1

11

M

j

BCj

M

j

ACj YY

M

Natural experiment

So what does the second difference accomplish in words?

Eliminates time-invariant differences in preference for girls across counties

Why is natural experiment (price reform) important?

Creates counterfactual: Counties in 1965 have same characteristics/cultures but no differences in relative prices (everything controlled)

How do we turn that into a regression?(keep ignoring different years)

First, create interaction term = A*B

Y= sex ratio in county k in year t

How does she get this with only 1997 data?

Regression equation:

ktktktktkt uATTAY 321 How do we interpret each of these coefficients?

What about time trends?China during 1962 – 1990

• Migration strictly controlled

• Little technological change in tea or orchards

• Sex-revealing technologies unavailable

What assumption is necessary?

Time trends related to sex preference same in tea- and non-tea counties

Income effect versus relative income

How does Qian distinguish an income effect on preferences for girls from a change in female relative to male income?

Looks at three kinds of crops:(1) Category 1 = grain(2) Category 2 = tea(3) Category 3 = orchards

How does she use these?

Classifies counties according to whether they plant 1 or 3 using data from Agricultural census, all others plant 2

Uses same difference-in-differences framework for three types of counties:

(1) To estimate effect of adult female income on sex ratios: estimate effect of increase in relative tea value on sex ratios

(2) To estimate effect of adult male income on sex ratios: estimate effect of increase in relative value of orchards on sex ratios

(3) To estimate effect of household income on sex ratios: estimate effect of increase in relative value of sex-neutral cash crops in sex ratios

What does income in non-tea counties look like?

What are potential problems with the estimate?

Does it matter that she can’t see tea production at household level?