Development Economics
Lecture 1: Introduction and RCTs
1: Introduction
1.1: Why development economics (skip)
1.2: Development economics and growth theory (skip)
1.3: Micro-foundations of development
Move away from macro-perspectives => microfinance (savings, credit, insurance) and investments in
human capital (health, education, entrepreneurship training)
Reasons why these factors are constrained in developing countries:
Market failures: credit, insurance, land
Govt failures
Institutional failures
To understand these factors, we need to immerse ourselves into the lives of the poor
Measuring poverty
Deaton and Dupriez: calculate the cost of a basket typically consumed by the poor how
much it costs to buy the basket => below the cost = poverty household will have different
preferences use price index with weights based on the consumption of the poor from the
household surveys instead of aggregate consumption from national accounts => housing,
clothing, educational fees, food, etc in the basket instead of cars, electronics etc national
accounts (such as Sala-i-Martin has been using)
Worldbank estimate: 1.3 billion out of 5.2 billion world population were in this situation
Poverty = relative concept poverty lines are much higher for richer countries (so many
more ppl can be considered poor)
Recent advances in machine learning allow to construct alternative measures of regional
poverty:
o Initial work explored nightlights data (satellite photos taken at night), strongly
correlated with growth: more light = richer but they are less effective at
differentiating between regions at the bottom of the income distribution
o More recently, high-resolution satellite imagery is used to study daytime pictures
look at types of roofs in certain areas or look at the pavement of roads metal roofs
and paved roads = better separate poor from ultra-poor regions
o Regional usage of mobile phones correlate with regional distribution of wealth: usage
of phones: some phones cost only 10-15 euros coverage rate is high, a lot of ppl
have this the amount of times the phone is used differs => bcs ppl still need to buy
data etc
Evolution of poverty (using the 2005 poverty line): absolute poverty by region
In 1981: absolute poverty was severe in the most populous regions of LIC
By 2005, the poverty gap had fallen to 10% or less everywhere but in SSA
1
,Extreme poverty: $1.90/day (in 2011) = threshold
Economic lives of the poor:
How does someone live on $1 per person per day? (now +- $2)
What do they spend their money on?
How do they earn their living?
What kind of infrastructure do they have access to?
What kind of markets do they have access to?
Lives of the poor: data HH level survey data from 13 countries important heterogeneity in living
standards both within and across 13 countries
Lives of the poor: fertility housing = fixed cost more ppl under same roof share this fixed
cost that's why ppl live with their parents, large # of children, multiple women living within same
families => aspect where we see very large disparities slightly less poor: more split in different
housing
Lives of the poor: expenditures
Food is a major expenditure for the poor (56%-78% of consumption)
Are ppl eating enough? studies show that this is not the case
You can have much more calories by changing your diet there is clearly preference for
certain type of food (rice & wheat) despite that you could get more calorie intake by changing
your diet (other grains)
Non-trivial sums on alcohol & tobacco (higher than on education for a majority of countries)
also large expenditures in festivals suggests that the poor do have some margin of
choice and choose not to exercise it in the form of buying more food why?:
o Higher discount rates in general (low life expectancy)
o Self-control problems: everyone has this, the rich ppl too there is always a nicer
alternative, also same with type of expenditures you make on food you might feel
more for a certain type of food consequences of making a mistake is very different
than for richer ppl
o Social norms, preferences, institutional constraints
In most places: urban areas come with more substantial costs than rural areas
Lives of the poor: access to infrastructure road, electricity, water, and sanitation have a direct effect
on welfare, but might not be reflected in income measures goes back to HDI => researchers try to
measure more not only variation across countries but also within countries enormous between-
and withing country variation in access to services and infrastructure
Poverty is more visible in cities but if you measure the basic access (not good, just basic access) it's
still must easier to get access in urban area than in rural area not as visible, bcs ppl have space in
rural area but they have less
Health and education:
Most LIC made some attempt to ensure poor HH have access to primary schools and basic
health centers the quality of the facilities tends to be low, even when they are available, and
it is not clear how much they actually deliver almost everyone is enrolled in school, but the
quality? are they actually going? => in terms of enrollment a lot of had been obtained, the
same with access to hospitals the big problem is quality of school or health care how to
motivate teachers? Etc
Very common to be sick and need to visit doctors: Leads to high share of expenditures on
health, which increase steeply with income
2
, Huge mortality rates (before year 1) per 100 living births.
Expenditures on education are low (bad quality public schools), but increase steeply with
income as well
Occupational choice?:
Self-employed (“entrepreneurs”): distribution of occupation in low vs high income countries
high income: laborers, there is a contract and employer and you know the # of hours you
will work
Very few are only farmers often only small plot of land and they sell if they have left over
after own consumption => many ppl combine in low income countries, often with a business
instead of wage labor
Small enterprises: no employees, very low level of working capital those who own land
own tiny amounts of it, not irrigated this reflects lack of employment opportunities as well
as lack of access to financial markets
An important policy question is whether to subsidies or encourage this form of “petty”
entrepreneurship (e.g. through microfinance) or whether to expand formal sector employment
opportunities you want entrepreneurs to grow => they move towards medium or large size
# employees they can offer stable jobs to more ppl and job opportunities for other ppl =>
on the other side a lot of petty entrepreneurs: earning a very small amount of money, not
paying taxes, ...
What should happen?: try to create more opportunities for petty entrepreneurs of attract
foreign investors?
Financial access:
Loans:
o The poor have a hard time accessing loans from formal sources, even in urban areas
where banks are within reach.
o Most of the loans they receive are from informal sources (relatives, shopkeepers,
moneylenders).
o Average interest rates are extremely high (e.g., it is almost 4% per month in the
Udaipur survey).
Savings
o Saving at home is subject to negative real rates of return, pressure by the kith and kin,
theft, and temptation to spend.
o Access to formal savings accounts is growing.
o A growing literature looks at whether reducing the cost of financial intermediation or
providing commitment devices is effective.
o Very few are insured, another topic to discuss given the considerable risks faced.
2: Randomized control trials
2.1: Introduction
RCTs = Randomized control trials = how to run field experiments very popular: many experiments
in the US and LIC
More and more of theories in devt economics have been tested using randomized control trials e.g.
studies to access on savings
Looking at causality => our cities are doing well, but it's actually not easy to implement
Rubin Causal Model:
Two ingredients:
3
, o Treatment: 1 variable di ∈ {0,1}
o Output variable of interest yi
Key element, the notion of potential outcomes:
o y 1i : output when treated
0
o y i : output when not treated
y i: the observed outcome in other words: y i= y 0i ( 1−d i )+ y 1i d i
Causal effect of treatment: for an individual i the treatment d i has a causal effect on the outcome yi if
1 0
the event di = 1 instead of di = 0 implies that y i = y i instead of y i= y i in this case the causal effect
1 0
of di on yi is c = y i − y i => the identification and the measurement of this effect is logically
impossible
The fundamental problem of causal inference: we cannot observe both treatment and
control group for the same individual always 1 observed and 1 missing given
that we can't know for the same person what happens if person is treated or what if
person is untreated => we need to construct the counterfactual = a person who is as similar as possible
to the treated person but who is not treated through randomization
Selection bias:
0 0
Can we replace the missing E( yi ∨d i=1) by the observed E( yi ∨d i=0) and consider
E ( y i |d i=1) −E ( y i ∨d i=0) ? => RDD = regression discontinuity design
1 0
[
E ( y 1i |d i=1) −E ( y 0i|d i=0 ) = E ( y 1i|d i=1 ) −E ( y 0i |d i =1 ) ]
[
+ E ( y i |d i=1 ) −E ( y i |di =0 )
0 0
]
= program effect + sample selection bias
First line gives us program effect (think about the treated outcome with if they were untreated)
and the 2nd line is the sample selection bias (= what would have been the outcome if you
would not treat ppl in the control group vs the treatment group) SSB should be zero
problematic if it is not zero
Example: we're interested in a scheme for unemployed ppl give certain training to help ppl
be better at finding new job => reduce time to find job or in other words increase probability
that ppl find job after 1 month = outcome variable that I'm interested in you make it
available for 100 ppl (control) but only 80 show up 20 don't show up bcs they don't think
they need it and they are good at it and they disappear but they are the best ppl bcs they would
have found a job easier likelihood that you find job now with new sample of 80 is going to
be smaller so negative sample selection bias => underestimation of real effect of the program
New example: same but 20 ppl who drop out bcs they are lazy they are the ppl who would
otherwise have a harder time finding a job => the probability of finding a job is going to be
bigger bcs the lazy ones disappear => positive sample selection bias
SSB: usually large compared to program effect (= problematic) highly depends on choice
of control groups (sign might be reversed) you can see this in the previous example (either
lazy or the best ppl who drop out)
How to eliminate SSB?: randomly assign ppl to a program (like tossing a coin: random
assignment) LLN: random assignment makes groups on average comparable ensures
treated and nontreated are equal in all aspects but treatment status
2.2: How to randomize in practice?
Three types of randomization: lotteries, phase in, encouragement design
4