Econometrics is the study of quantifying causal relationships in a social science setting.
What do a mean by a social science setting? I mean a setting where it'd be inappropriate to run a controlled experiment.
An example of an unethical experiment would be running a experiment that determines the effect of parental income on the grades of a high schooler. There are many factors that we might think influence student outcomes apart from household earnings including the quality of the teachers, the age of the students, riggor of the course, etc... If any of these variables are correlated with grades as well as parental income, then we need to be careful about how our analysis is conducted.
A possible solution from this problem is controling for these factors in our model. But what would we do to control for these variables and how would we know if our approach is correct? In order to be entirely sure of our analysis, we would have to randomize parental income for high school students. How can parental income be randomized? We could put slips of paper in a hat that determines how much money each family receives monthly and then randomly assign these incomes to households. If the household makes more money, we take it away, and if the household make less money, we give them more.
Clearly, this approach has ethical problems and even if running this experiment was considered ethical we can think of other aspects that we may. (e.g. We may think that household income before high school also influences grades.)
Econometrics provides us with tools to combat situations where we don't have as much control over our experiment as would be ideal for the purpose of answering our question.
Other examples of questions where econometrics can help include:
For me, some of the most interesting questions are causal, economic questions such as these.
I'm going to share a story that made me realize that economists are trained to think in a way unique to statisticians.
I was taking a regressions course in the Statistics department at my university. We had a course lecture on linear regression. To illustrate linear regression, the professor decided to run a linear regression of global temperature on atmospheric CO2 and methane. We were told to write the following in our notebook:
CO2 has an effect of 0.740 (C/ppm) on global temp holding methane constant. Holding CO2 constant, methane has an effect of 0.097 (C/ppm).
I was pretty shocked at the lack of understanding. This would only be true if we assumed that the disturbance term in our regression was uncorrelated with our outcome, global temperature. While a bit out of the scope of a non-technical post, this requirement for causality is surely violated.
I was taking that course with a colleague from the econ department. Though I didn't feel the need to check with the professor to make sure he understood the dubiousness of the claim, my colleague tried talking to the professor after class about possible sources of error that would invalidate the claim such as omitted variable bias but the professor waved off the objections and remained adamant that the these objections were irrelevant for this conclusion.
Learning to rigorously understand the necessary circumstances to infer causality are impressed in the minds of econometricians in every lecture. This did not appear to be the case in statistics and students who had not yet earned degrees in economics had a better understanding of this concept than a tenured, professor with a reputable PhD.
A lack of understanding of causality is not unique to statisticians. I've had experiences similar to these with professors from the engineering department who largely approach their studies the same way a physicists. In the hard sciences, it's easy to fall into the trap of not thinking about causality rigorously because the typical environment for their research is controlled and in a lab. These are not the circumstances in the social sciences.
Disclaimer: This article is not claiming that climate change is not occurring, it simply critiques the less-than-rigorous analysis conducted in an introductory regressions course in the statistics department.