EC1P1.

Economics
London School of Economics
Assignment 3
Michaelmas term 2022
Professor Ricardo Reis

If at any point in this assignment you feel that anything is unclear, please make
additional assumptions that you feel are necessary and state them clearly.

1. A researcher compares two groups of people that are identical in all the characteristics that she
can measure, but perhaps are different in characteristics that the researcher does not observe. She
tries to answer one research question by comparing an average outcome in one group versus the
other. Discuss in each case below whether there is a selection bias, and whether you think that the
actual answer to the question is likely lower or higher than what the researcher calculated.

(a) What is the effect of education on wages? Compared the average wage of a group of people
with a university degree with the average wage of group with only secondary education.

(b) What is the effect of being bilingual on getting into the best university? Compared the share
of bilingual children who were admitted at LSE with the share of mono-lingual children who
did not make it.

(c) What is the effect of a learn-to-code program on ability to find a job in the future? Compared
share of those who found a job among those who took the program (and were selected to be
in it by the entrepreneur who created the course) with share among a group that applied but
were not selected in.

(d) What is the effect of social skills on professional success? For a group of couples that were
due for promotion in their jobs within the next year, gave one of them an intensive course on
social skills and the other not, and compared the shares who got the promotion.

(e) What is the effect of public infrastructure on economic performance? Within the same region,
picked a group of villages at random and invested in sanitation, roads, schooling, while for
another group of villages did not, and compared average wages between the two sets of villages
a few years later.

(f) What is the effect of giving recent graduates seed money to start a tech company on it
becoming a unicorn (worth more than a billion)? Pick the 200 graduates from the LSE
department of economics this year, and give half of them a million pounds to start a fintech,
measuring 5 years later how many in each group have a company worth a billion.

1
(a) What we observe are the wages of the group with higher education and the wages of the group
with only secondary education. Note that we do not observe what the wage of the higher
education group would have been if they only had secondary education.

Comparing the stated averages will likely overstate the effect of education on wages. Those
who choose to enroll in higher education likely had some unobserved features that would
make them more successful so the effect is overstated. What kind of features do you think
these might be?

(b) Bilingual children often come from immigrant families, or from families that value language
proficiency. This could imply something about how much time and effort the family invests
into their children’s’ education, which definitely matters for getting into university.

So there is a potential selection bias here, but it could go either way.

(c) The entrepreneur wanted to show the success of the program, so she likely picked the better
applicants who were more likely to find a job anyway. Hence, the effect is overstated as
the difference between the two groups accounts for the effect of the coding program and the
inherent higher ’quality’ of the applicants chosen for the program.

(d) If Bob and Manuel are in a couple, and Bob learns social skills, surely he will teach some
of them to Manuel. Hence, both of them are gaining some social skills and the effect is
understated as there is some spill-over of the treatment itself to the non-treated group (which
is different from spill-over effects from treatment).

It is not explicitly stated how it was decided which person in the couple was chosen for the
social skills course. If it was not random, then there might be a selection bias introduced as
well.

Note that we are looking across many groups of couples here rather than just at an individual
couple - why is this important?

(e) The villages are randomly chosen for treatment, so there should be minimal selection bias so
long as the treatment group and non-treatment group consist of a large number of villages
each.

However, as the villages are in the same region they will inevitably interact through trade.
People may move from the unprivileged villages to the benefiting villages, possibly as economic
migrants and sending remittances back to their home villages. The benefits of treatment will
spill over to the untreated villages which will lead to the effect being understated.

2
 (f) Graduates are randomly selected for treatment, so there should be no selection bias (although
all the graduates are from the LSE Economics department - do you think this causes any
issues?).

The average treatment effect is likely to be very close to zero for almost all students, since
becoming a unicorn is so unlikely. That is not to say that all the graduates do not start a
tech unicorn, but the vast majority will not.

Thus the average treatment effect disguises massive inequality: for one lucky graduate, this
treatment may have been the difference between starting a unicorn or not, while for most it
had no effect.

2. Consider an omniscient social planner ruling over a society of 20 agents where 10 of the agents
have A-preferences:
√
ua (ca ) = 2 ca

and 10 have B-preferences:

√
ub (cb ) = cb + 100

Our planner has 100 units of the c good to allocate to the two groups, so:

10ca + 10cb = 100

The social planner aims to allocate c between the two groups in order to maximise the sum of
total utility in society. However, they might not always care about both groups equally and so
they weigh the utility of group A by λ and group B by 1 − λ when summing up the total utility in
society.

(a) What is the objective function that the planner will want to maximize, taking into account
the relative size of each group, their preferences and how much the social planner cares about
each group?

The social planner cares about the sum of total utility in society, but weighs the utility of
group A by λ and the utility of group B by 1 − λ.

Therefore, the social planner wants to maximise:

√ √
10λ2 ca + 10(1 − λ)( cb + 100)

But note that maximising this will be no different to maximising:

√ √
λ2 ca + (1 − λ)( cb + 100)

3
(b) Maximize that objective function subject to the constraint that total consumption is 100, by
replacing out ca in the objective function and taking derivatives with respect to cb . Show
that the optimal choice of cb /ca is:
2
cb
  
1 1−λ
=
ca 4 λ

Once we substitute the budget constraint to replace ca the problem becomes:

p √
max λ2 10 − cb + (1 − λ) cb
cb

The first order condition is:

 −1/2
λ2(1/2)(−1) (ca )−1/2 + (1 − λ)(1/2) cb =0

Rearranging this gives the required result.

(c) Say that the social planner cares about both groups equally so λ = 1/2. At the optimal
allocation, why is cb < ca intuitively?

As the social planner values the utility from each group equally, we can just focus on the
marginal utilities for group A and group B. Intuitively, as the the marginal utility of A agents
is higher than B agents, maximising the simple sum of utility in society will require giving
more consumption to A agents than B agents.

We can look at this a bit more formally as well. The marginal utilities for the two groups are:

∂ua
MUa = = (ca )−0.5
∂ca

∂ub
MUb = = 0.5(cb )−0.5
∂cb

The marginal utilities tell us how much the utility of an agent increases if they receive more
consumption. Here, both types of agents have diminishing marginal utility of consumption
as the marginal utility of extra consumption falls as the amount of consumption an agent
already has increases.

What should be clear from the above is that if ca = cb , then the marginal utility of A
agents is higher than the marginal utility of B agents. This means that A agents value extra
consumption more than B agents if they have the same amount of consumption. If this is the

4
 case, then taking away some consumption from B agents (thus losing M U b ) and giving it to
A agents (thus gaining M U a ) will increase the sum of total utility.

Hence, if we are thinking about allocating consumption in order to maximise the sum of
utility, it must be the case that ca > cb . The optimal allocation of consumption in this case
where λ = 1/2 will be where M U a = M U b , which is exactly what the FOC is stating.

cb
(d) Why does the ratio ca fall when λ increases? Explain intuitively.

cb
From the result in part (b), it is clear that an increase in λ will lower ca - but what is the
basic explanation for this?

An increase in lambda means the social planner cares relatively more about group A and
relatively less about group B. This will mean that the optimal allocation of consumption will
cb
involve more ca and less cb which will lower the ratio ca .

More formally, we can look at the marginal gain in social utility when increasing ca or cb
which we can find by taking the derivative of the social planner’s objective function with
respect to ca and cb . This is effectively what we did when taking the FOC and is how we
derived the result in (b).

(e) Prove the following claim: If 0 < λ < 1, the choices of the social planner are always Pareto
optimal.

If we have 0 < λ < 1, then the social planner puts some positive weight on every agent. This
means that a higher utility for any agent whilst keeping all other utilities unchanged will
always increase social welfare and be a good thing in the eyes of the social planner.

Hence, a situation where it’s possible to increase one agent’s utility with no cost to any other
agent would not be maximizing social welfare. If a Pareto improvement is possible, then the
social planner cannot be making the optimal choice.