Quantitative research can be purely descriptive techniques or causal impact analysis and can be historical or prospective. Most quantitative research is descriptive and historical, such as describing the earnings of male and female workers over the past 20 years, or the proportion of children of different races who are poor. But descriptive analyses tend to invite causal inference because when we see a gap in earnings or different trends in poverty, we naturally want to explain why such patterns exist and to project future conditions.
There are many barriers to this kind of extrapolation. For example, historical data represent one possible outcome of a partly random process, and we cannot easily see the appropriate counterfactual condition—what might have been under alternative scenarios. One serious threat to interpreting any pattern is that traits and participation in programs are typically not randomly assigned, which can introduce selection bias.
There are a variety of experimental techniques to sidestep this problem, and there are quasi-experimental methods that seek to replicate the results of an experiment that was not or could not have been run. Quasi-experimental methods include techniques such as regression adjustment, propensity score methods, panel regression, instrumental variables, and regression discontinuity methods.
There are related problems that these methods can also address, under additional assumptions. One is that historical data may not represent the population of interest, either because the data were not selected to be representative (as with sites that volunteer to participate in an experiment) or because the target population is different from the available sample (for example, from another geographic area, or existing at a future date).