Question

 

Assume that you are interested in estimating the effects of schooling on earnings and that you have individual data on schooling and earnings.

a) Set up an empirical model that captures the correlation between schooling and earnings and describe this. Why are you likely to overestimate the true effects of schooling on earnings with this model?

b) How would you go about to estimate the causal effects of schooling on earnings if you as a researcher had the authority to do whatever you like? You do not need to consider any ethical or financial restrictions. Describe how you would go about and why it would work using your own words.

Answer 1

(a) The model is as follows: Earnings =constant term + Beta1*education + error term

Now, the coefficient of schooling is overestimated here because of the omitted variable bias. there are several other factors that are positively correlated with both the education of a person and their wages, For example, the education level of parents. This leads to a positive bias in the estimated coefficient of education in the model.

(b) To estimate the causal impact I would run a Randomized control trial (RCT). Now, first I will randomly select a specific area, say a city. In that city, I will find out the young children who are not currently educated and have similar characteristics like family income and family education. Now, I will randomly select some students from these and send them to schools (no financial restrictions and no ethical restrictions on who should be selected). The selection has to be completely random. Thus, we will have two groups, treated who are sent to schools and untreated who are not sent to schools. After a few years, we track the earnings of these two groups, and any difference in the earnings can be attributed to the treatment i.e. sending to school. We have a causal impact as we have exploited external variation and the randomness of slection ensures that there are non confounding effects