Позитивные изменения. Том 1, №1 (2021). Positive changes. Volume 1, Issue 1 (2021)

If there are repeated samples over two time periods, the model tested using this method is written as follows: y = B0 + e1dB + 0d2 + 1d2dB + u, where y is the outcome of interest, d2 is a dummy variable for the second period, dB is a dummy variable for the experimental group.

Thus, the sample in our study will also represent two groups, which will be expressed in the two groups of companies, one of which was a violator of the law (experimental) in one of the two periods, and the other did not carry out such actions in any of the periods (control). Thus, the variable d2 in our equation characterizes the period corresponding to a particular observation, and the variable dB will represent the difference between the experimental and control groups. As a result, this method gives us the opportunity to identify the net impact of the program on any individual social or economic component.

The second major work of David Card was the study of the impact of the minimum wage on employment. The hypothesis was tested that the introduction of the minimum wage would affect employment. The study was based on the fact that on April 1, 1992, New Jersey was supposed to raise the minimum wage from $4.25 to $5.05 per hour. The researchers collected data on the employment rate at fast food restaurants in New Jersey and Pennsylvania before and after the increase in New Jersey. The result showed that the increase in the minimum wage led to an increase in employment, since it actually increased in New Jersey.

Another fundamental task, which is generally expressed as the search for the causal relationship between decisions (individual or public) and economic results, was investigated by Joshua Angrist. “Cause-and-effect” issues required a special language. Angrist in his works relied on the model of “potential outcomes” invented in the seventies by Donald Rubin.

The initial assumptions of the model were the following: each individual has a set of potential outcomes that can happen to him, depending on what his decision will be. For example, if a person has a toothache, then he can either take an analgesic pill, or go to the dentist, or do nothing. Each of these solutions will lead to some potential result. The actual data show only one of these outcomes, we do not know the outcome if the same individual would have made a different decision. This fundamental problem of causal analysis has no solution, we will never be able to measure the impact effect for a particular individual. At the same time, under certain conditions, we can measure some average effect.

In their 1994 paper, Imbens and Angrist show how to apply this methodology to instrumental variables. Instead of coming up with potential prerequisites to evaluate the effect for everyone, the researchers turned the task around and asked for whom we can evaluate the effect with reasonable prerequisites. The answer turned out to be simple and intuitive: the average effect can be calculated for those individuals who changed their decision under the influence of the tool — the so-called Local Average Treatment Effect.

Based on the new methodology, Angrist and Kruger studied the impact of education on wages. Their task was to exclude other influencing factors — a person's abilities or his family background. Scientists decided to use data on the year of birth of students to predict how many students will study at school. It was assumed that the year of birth is not related to the origin and innate abilities, respectively, it does not affect the success and salary level of a person in the future. They calculated for a large sample that in fact the impact of training on earnings turned out to be greater than previously estimated using traditional methods. This is how the standard for this kind of analysis was established.

Instead of coming up with potential prerequisites to evaluate the effect for everyone, the researchers turned the task around and asked for whom we can evaluate the effect with reasonable prerequisites. The answer turned out to be simple and intuitive: the average effect can be calculated for those individuals who changed their decision under the influence of the tool.