2.4 Sampling in Practice

The sampling methods above deal with situations in which the population is hypothetically obtainable, but it is not feasible due to time or resource constraints. For example, a company could run a concrete election poll by calling up every single registered voter (the population), but that would cost too much and take too long. What happens in the situation where the population is unobtainable, meaning that at any point in time there will be some obtainable portion of the population because it hasn’t occurred yet. For example, if I wanted to analyze US unemployment rates, I couldn’t possibly consider future rates that haven’t been observed yet. In situation like these, one must take time to consider exactly what population you want to draw inferences from and draw their sample accordingly.

A quick example of data sampling in my own research is as follows. Some of my research deals with how bank lending responds to changes in the stance of monetary policy.1 Since bank lending data is coming in daily, it is clear that the entire population is unobtainable. However, selecting a sample is not simply collect as many observations as possible because we must be clear about what population we want to actually talk about. In my example, I want to talk about how bank lending responds to monetary policy shocks in normal times. This means that observations in the sample cannot be impacted by episodes where monetary policy differed from what is currently considered normal. This restricts my sample to be after World War 2 and before episodes of unconventional monetary policy (i.e., anything post-2007).

What happens if characteristics of the population potentially changes? That’s easy - you repeat the analysis with an updated sample and acknowledge that you are drawing inferences on a potentially different population. That is what I am currently researching. In particular, I am determining how bank lending responds to monetary policy under unconventional monetary policy practices of paying interest on excess reserves. This requires a data sample of observations appearing after 2007.

  1. Dave, Chetan, Scott J. Dressler, and Lei Zhang, (2013). The bank lending channel: a FAVAR analysis. Journal of Money, Credit, and Banking 45(8). 1705-1720.↩︎