Econ 980o: Health, Education and Development Lecture 1 September 18, 2008

Econ 980o: Health, Education and Development

Lecture 1September 18, 2008

Basics

Class: Thursday 2:00-4:00

Instructor: Erica FieldOffice: Littauer M30Office Hours: Friday 1:30-3:30

TF: Vanya PashevaOffice: TBAOffice Hours: TBA

Section: TBA

What this course is about

• Micro-foundations of poverty and economic development

• What health and education have in common:

– Human Capital

• Stock of skills and attributes embodied in people that raises labor productivity

Why do development economists care so much about human capital?

(1) In practice, big difference between developed & developing countries: School and literacy low, and burden of disease disproportionately high in developing countries

Much of this disease burden preventable (in theory)- In 1990, 41% of deaths due to infectious diseases

(diarrhoeal disease, pneumonia, TB)

→ Room for policy intervention

“Macro” motivation - growth accounting:

(2) In theory, human capital accumulation one of the important sources of long-run economic growth

– Endogenous growth theories rely on understanding labor productivity and innovation, and hence human capital

Empirical evidence:

Recent periods of sustained growth in total factor productivity (TFP) and reduced poverty in low-income countries closely associated with improvements in a country’s:

– Child nutrition– Adult health – Schooling

“Micro” motivation:

(3) Education and health correlated with ...

• Migration• Faster adoption of new (profitable) technologies• Child wellbeing• Marriage choices• Fertility• …..nearly everything you can think of

So important for understanding poverty dynamics

Key questions we’ll address in class: In practice, what are best ways to increase human capital formation in developing countries (and thereby encourage growth)?

– What are relative returns to various types of human capital inputs?

– What are institutional or political barriers to human capital formation?

– Are there important psychological or cultural barriers to human capital formation?

– How do we design policies to motivate investment in health and education?

Brief overview of course topics:

(1) What are the returns to human capital?

Motivation:

How much should we prioritize these over other poverty alleviation strategies?

Topics covered:• Labor market returns to health and schooling• Effect of health on schooling• Effect of education on mortality

Brief overview of course topics:(2) What are determinants of health and education production?

Motivation: What interventions will be most effective in improving health

and education outcomes in developing countries?

Topics covered• Is health a normal good (how much will health improve with

income)?• Intra-household inequality and child investments• Gender differences and “missing women”• Impact of information on health behavior• Education production• Teachers and teacher incentives• Political barriers to human capital investment

Primary objectives of class: • Get you started on careful empirical research

– Not only what are the key questions in health, education and development, but how do we figure out the answers

– Think critically about measurement/data issues

– Start thinking critically about how to estimate causal relationships instead of simple correlations

– Expose you to range of empirical methods

– Methods we’ll discuss: Regression methods, experimental approaches, natural experiments, instrumental variables, “difference in difference” estimation

Who should take this class: • Juniors thinking about senior thesis topics

• Anyone interested in research methods (e.g. graduate school)

• Anyone interested in exploring development policy questions

• Anyone else interested in the topic!

• Caveat: Need some past exposure to economics and econometrics

Assignments

• Readings

– Read ~2 journal article for each class

– Come prepared!

– Use reading guide to prepare for discussion

Reading guide questions:

Estimation strategy:

• How do authors establish causality (what is identification strategy)?

• Is the approach valid? If not, why not?

• What is most significant measurement challenge (i.e. selection bias, reporting bias) and how is it addressed?

Reading guide questions:

Criticism:

• How do findings contribute to our understanding of health, education or human behavior?

• What is most important question left unanswered?

• What is most important policy implication of the findings?

• What is leading alternative explanation for their findings?

• Are the data and measurements reliable?– Bigger problem than you would think!

Evaluation:

– 10 weekly assignments (40%)

– Research paper (50%)• First draft (25%)• Final draft (25%)

– Class participation (10%)

Weekly assignments

(Not as bad as it sounds!)

– Many just detailed review of class reading

• Next week:– Read Field, Robles and Torero article– Answer questions like the ones we just saw

– 3 problem sets designed around practice using Stata

– Second half of semester, assignments will ask for parts of your research paper

Research paper

• ~20 pages

• Empirical study of topic related to class content

• Should be used to explore potential thesis areas of interest

• Look for ideas in readings

• Discuss ideas with instructor and Vanya

• Note: No way to get behind – draft due early!

Examples of appropriate paper topics:

– Look for narrow question with a YES or NO answer

– Think about testable predictions from simple economic models

• A lot of important work begins with a search for the obvious, since data frequently show something different

Appropriate paper topics:

– Changes over time or differences across areas in school or health policies

– Differences in disease climate or nutrition driven by geography

– Cross-country or within-country differences in social norms or institutions

– Impact of climate shocks (rainfall), other unanticipated events

Specific examples:

– Do returns to primary school vary with opportunities for farming technology adoption?

– Do changes in life expectancy influence marriage age?

– Is disease environment correlated with fertility preferences?

– How does nutrition influence labor market earnings?

Prime example: health and income

The challenge of causal estimation:

(why the last question is so hard to answers)

Income–health gradient:

• Striking consistency in the association between poverty and poor health across diverse array of existing studies

• Generally robust to variation in– Measurement of poverty, health– Geographical focus– Time– Within- or across countries

The changing relationship between life expectancy and income (Preston, 1976)

Strauss and Thomas 1999

2.252.25

22

1.751.75

1.51.5

1.251.25

1.251.25

11

.75.75

.5.5

.25.25

160160 170170 180180 190190

United States

Brazil

HeightCMS

ln (wages) in Brazil ln (wages) in USA

Within-country cross-sectional variation: Relationship between Adult Height and Earnings

3 possible explanations for basic association between health and wealth:

(1) Health leads to improvements in income and economic growth

HEALTH WEALTH

“Improvement in nutrition and health may account for as much as 30 percent of the growth in conventionally measured per capita income between 1790 and 1980 in Western Europe.” R.W. Fogel 1990

What are potential pathways?Direct productivity outcomes:

– Good health increases labor market productivity of adults

– Child health leads to better cognitive skills, higher productivity of schooling

– Public health externalities

• Disease spreads! So there are important spillovers of one person’s health to another’s

Health as human capital investment

(i.e. Health to wealth via capital accumulation)

Lower life expectancy leads to:

– Lower investment in education

– Lower savings rates

– Change in population age structure

3 possible explanations for association:

(2)Income leads to improvements in health

WEALTH HEALTH

What are potential pathways?

Society level: (government expenditures on public health):

– Disease control – Sanitation

Individual level:

– Health investment (vaccinations)– Better nutrition– Health care – Indirect effects:

• Education (leads to better health information)• Housing• Lifestyle choices

3 possible explanations for association:

(3) Third factor explains both

Like what?

• Political climate (e.g. stability, civil war)

• Culture (e.g. technology adoption)

• Geography (e.g. climate, isolation)

Big question:

• How do we disentangle these three pathways?

We’ll spend the semester thinking about this and similar questions…

Aside: change over time in focus of discussion

“The influence of economic conditions on mortality has been recognized at least since biblical times.” Preston, 1976

In contrast, health to wealth relatively modern issue:

“I disregard here the few works which deal with the relatively minor effect of mortality on economic processes.” Preston, 1976

Today’s Class:

How do we isolate causal effect of health on income?

Once again, multiple potential pathways:

Direct productivity outcomes:– Good health increases labor market productivity– Child health increases productivity of schooling – Ill health has large externalities

Lower life expectancy :– Reduces investment in education– Reduces savings rates– Increases population growth

To test theory, need to isolate particular channel

This class: Does adult nutrition raise labor productivity?

What’s thought to matter:Energy (caloric content)

Specific nutrients:Iron (thought to be most important)

Deficiency leads to:

• Cognitive defects in children• Maternal deaths due to severe anemia• Decreased day-to-day productivity

Vitamin A

Deficiency associated with:

• Decreased resistance to infection

Iodine (next class)

Why is effect of nutrition on productivity relevant for scientific theory?

• Many scientific theories of bio-nutrition (specific nutrients protect health and increase physical functioning), not very good human evidence

• Evidence from lab experiments on rats, but not so straightforward to scale up in humans

• Bottom line: Lots of potentially beneficial policy interventions, hard to say what matters most, particularly with combinations of nutrients, age-specific importance

Why is bionutrition relevant for economic theory?

Nutrition and labor productivity:

• One leading hypothesis: Efficiency wage model (Ray, Chapters 8,13)

• If nutrition important for productivity, could explain why so much surplus labor in developing countries

– Surplus labor: Unemployment at the same time as MPL>0 (positive, rigid wages)

• Potentially important source of poverty traps (why we don’t see GDP convergence as predicted by growth theory)

Poverty traps:

• In a poverty trap, you’re poor because you’re sick and you’re sick because you’re poor …

• Example of a health poverty trap: – Minimum calories needed for employment

• What that means for a model of economic growth: – Multiple equilibria! Rather than converging to high

GDP, you get stuck in the bad equilibrium

A simple theory of nutrition and productivity:

• X-axis: w = calories (in simplest model, wages go only to food)

• Y-axis: e = work capacity (total number of tasks you are capable of completing in a day)

• e(w) = capacity curve (index of labor productivity)

Capacity curve relates income and work capacity

Higher income → better nutrition

The Capacity Curve

ework

capacity

w (calories)

e(w)

Key assumption:

• e(w) convex at low levels of wi, but then eventually concave

Interpretation:

• Better nutrition: Calories first used by body for basic metabolism, only after a certain level do they translate into higher capacity

• When capacity curve is steep (slope > 1), small decrease in wage lowers output by a lot

The Capacity Curve

ework

capacity

w (calories)

v1

v*

h(c)

Now add wages: v = slope is “piece rate”

Involuntary unemployment:

• Because there is a discontinuity, employers can’t adjust wage to meet labor supply

Supply exceeds demand, so jobs are rationed

• Idea: Many want to work at this wage, but they can’t bid down the wage

Why not?

• Lower wage would decrease worker’s capacity to the point where it’s no longer worth it to hire him

Implications for labor markets:

(1) Leads to involuntary unemployment: (people willing to work for lower wage but wages don’t adjust)

(2) Means that the poorest are more likely to be malnourished, and more likely to be unemployed because they are malnourished

– a poverty trap

Policy interventions can shift individuals’ capacity curve, so that involuntary labor is reduced for the neediest:

Figure 4: Effect of Supplemental Nutrition on the Capacity Curve

Employment Income

Caveat:

• Depends on extent to which capacity curve really S-shaped

• To find out, need to conduct an empirical study

A nutrition experiment (Thomas et al.):Indonesian Work and Iron Status Evaluation

• Thomas et al. studies iron deficiency in Indonesia

Hypothesis tested:

Iron deficiency →lower aerobic capacity, lower endurance, fatigue → lower labor productivity → lower earnings

Method: Randomized trial

What makes a good empirical study?

1. External validity:

• The validity of inferences about whether the cause-and-effect relationship holds in different settings

(How easily can we extrapolate?)

2. Internal validity:

• The validity of inferences about whether observed associations between program participation or a policy (X) and the target outcome (Y) reflects a causal relationship from the program/policy to the outcome

Threats to external validity

• Characteristics of the population– Example: Nutritional needs of males may not hold for females

(calcium)

• Setting or context – Example: Impact of nutrition intervention depends on the

relative nutrient deficiencies, which depend on local diet

• Exact nature of the intervention– Example: Nutrient administered as supplement may not be

absorbed by body (but nutrient still important!)

• Outcome examined – Example: Maybe nutrient matters for cognitive ability but not

energy level

Threats to internal validity:

• Reverse causation: Y caused X

• Omitted variables:

– Selection: People who participate in the program are systematically different

– History: Events occurring concurrently with treatment could cause the observed effect

• Mean reversion (before/after): Naturally occurring changes over time confused with a treatment effect

Fundamental problem of impact assessment:

• Counterfactual outcome cannot be observed because individuals cannot simultaneously participant (T=1) and not participate in a program (T=0)

→ Same individual is only observed in one of two treatment states – received treatment (i.e, the policy or program) or not – so only see one set of outcomes (YA

T=1 or YAT=0)

Problem:

Want to compare E[YAT=1] to E[YA

T=0],

… but forced to compare E[YAT=1] to E[YB

T=0]

Estimator: E[Net Effect of T] = E[YAT=1 – YB

T=0 ]

So how do we get unbiased estimate of effect of particular intervention?

Key: Develop believable counterfactual (Group B)

– Best estimate of outcome person would have had had they not participated in program or received treatment

Empirical approaches to constructing counterfactual:

1. Select the best-looking comparison group ex-post

Example: Identify group that looks just like treated group but wasn’t treated

2. Construct an appropriate comparison group ex-ante

Best example: Randomized control trials (RCT):

• Treatment and control groups randomly selected from a potential population of participants

Randomization guarantees necessary conditions for unbiased estimator:

If randomization valid, on average, two groups are identical in every way prior to being treated, so:

1. E[YBT=1] = E[YA

T=1]

(both groups have the same response to treatment),

and

2. E[YAT=0] = E[YB

T=0]

(both groups have same outcome if not treated)

Experimental versus observational study:

Random treatment assignment addresses omitted variables (third factor):

You estimate:

Real model: (Z is omitted factor)

Purpose of randomization: Choose people to be treated so that treatment, X, not correlated with any other factor, Z, i.e.,

0 eZXY zx

0 vXZ x

0x

XY x ˆˆˆ0

)(ˆzXxx

Major advantages of randomized designs

• If implemented correctly, they eliminate concerns about internal validity

• They are easy to explain and convincing to lay audience, including policy makers

Calculating the experimental effect:

Before program

After program

Change

Experimental group YE1 YE

2 YE2-YE

1

Control group YC1 YC

2 YC2-YC

1

So what is the effect equal to?

[YE2-YE

1]-[YC2-YC

1]

Do we need two periods to get unbiased estimate?

Disadvantages of randomized designs• Most experiments necessarily conducted in laboratory,

whereas policy experiments would be much better for external validity

• Field experiments costly (data collection, manipulation)

– Hard to keep well-controlled– Hard to collect large enough sample for EV

• All randomized experiments can be politically difficult (even laboratory experiments)

– No one wants the placebo!

• Some potential measurement problems:

Potential problems with experiments:

• Confounding variables – Not a problem if randomize and large enough sample

• ContaminationSolution: Measure and check

• Attrition, complianceSolution: Intent-to-treat (what?)follow everyone regardless of what they do

• Placebo, Hawthorne, experimenter effects Solution: Make treatment blind, use placebo control group

• Generalizability Solution: Select random sample, stratify sample (requires increasingly large sample)

Motivation for Thomas et al study• Iron deficiency thought to be most common bio-nutritional

deficiency in world

• In Indonesia, half of female and one-third of male adults iron deficient (likely due to diet, malaria, worms)

• Animal studies indicate causal relationship between iron deficiency and reduced maximum aerobic capacity (endurance = capacity curve)

• Strong correlation between income and iron deficiency

• What about existing studies?

Methods• Iron deficient households over-sampled. WHY?

Why is a randomized intervention necessary?

• Strong correlation between hemoglobin levels and earnings

• What might this reflect?– Omitted variables?– Reverse causality?

Measurement issues in Thomas experiment:• How did they deal with compliance?

– Pill packet monitoring (noisy): achieved 92% compliance– Showed up directly in blood levels (very good measure)– Same rate of compliance for treatment and control (Why imptnt?)

• Is attrition a problem? – Only 3% attrition at 6 months– Mainly due to mobility (not opting out)– Attrition same for treatment and control

• How did they reduce contamination?

• Randomized at household level:– reduce sharing– maintain blindness– (also increased compliance of illiterate participants)

Estimator: Difference-in-difference-in-difference

What is this?

Calculating the DIDID effect:

Before program

After program

Change

Experimental group

YE1 YE

2 YE2-YE

1

Control group YC1 YC

2 YC2-YC

1

[YE2-YE1]-[YC2-YC1]

So what is the effect equal to?

{[YE2-YE

1]-[YC2-YC

1]}-{[YE3-YE

4]-[YC3-YC

4]}

Why do they do this?

Before program

After program

Change

Experimental group

YE3 YE

4 YE4-YE

3

Control group

YC3 YC

4 YC4-YC

3

[YE4-YE3]-[YC4-YC3]

-

High Hb: Low Hb:

Results: Among low-Hb treatments• Income 20% higher

• 3.6% more likely to be in labor force (only 5% OLF at baseline)

• Hours of work unaffected

What is interpretation of income increase?– Labor productivity

Is this likely in such a short time (wages are sticky …)?– Concentrated among the self-employed (50% of sample), so

plausible

Other possible reasons?

Effects for self-employed?• Improvements mainly for self-employed

• 40% increase in hourly earnings (huge!)

• Work more or less ?– No real change

• Argument: this is short run

Results (continued):• 0.8 fewer lost days of work

• 20 minutes less sleep

• More of treated group able to carry a heavy load

• No difference in reported fatigue! (maybe because they adjust sleep?)

• Some evidence of improved psycho-social health

• No measurable effect on women

Which of these results is most believable? Most surprising?What is most likely reason for gender difference?

Lessons:

• How did they do on external validity? • Is there a lesson for economic theory?(Does it look like there’s an S-shaped capacity curve?)

• What about scientific theory?Lots of pathways, hard to clarify exactly happens inside bodyCan you think of a way to alter the experimental design that could

shed light on these?

• What are policy implications?Advantage: CHEAP intervention (increases policy relevance)Remaining issue: Opportunity cost of alternative interventions

To really inform policy, need similar results for many types of nutrients. Where would you start?