OK, let’s talk about statistics. When you have one of the best statistics professor in the world, you will feel more and more pains when you do OSL ( ordinary least square) , because he always reminds you of the assumptions you need to have in order to use OLS in order to get the minimal least squares error between the observed the Y and predicted Y.
We all know that using OSL regression we have to meet its assumptions. However, few people really care about what what is the philosophy behind OSL regression. First of all, if we want to use OSL to approval a theory, the foundation of this way to appraise a theory is based on Karl Popper’s philosophy regarding counterfactual. A theory should be falsifiable, but even we reject H0, it does not mean that H1 is right. So what actually we are doing is to just to just reject H0, and we have not build a theory yet. When will we able to truly build a theory, OSL actually cannot accomplish this task at all.
Secondly, OSL assumption is built on the design of RCT. Sr. Fisher was the first one to propose that using RCT to “randomize for confounders ” ( this actually where one of the assumptions randomizing residuals comes from). However, in real life, we can never achieve the requirements of RCT and in real life what RCT can do is very limited. For example, within a sample less than 800, we will not be able to find the interaction terms. This implies that the by-product of certain drug will not be able to be found in the RCT trail.
Thirdly, before we build the model, in the social science filed, we can never achieve the assumptions of exogenous of variable. We cannot measure a lot of concepts and constructs in our field, but those concepts and constructs are no way like the concepts in the physics such as mass and volume. ( btw. those concepts never be observed in real world, and can never actually measure in real world, they are just abstract concept, if you go back and read about concepts in physics and other natural science).
The majority of social science in the past used OSL model. We have and are still misusing OSL and believe that OSL can build theories. Misusing OSL models in the social science indicate the a false confident in positivist paradigms and a lack of confident of our own disciplines. Social science always tries to imitate natural science. It, for me is a wrong direction.
I am not sure what is the best tool to answer research question in social science, maybe we just have to bear with the misuse of OSL for a while. Although there are sophistical statistical tools developed these days to take care of the flaws of OSL and other linear models, we do need to reflect on the limitation of models and foundations of our field. BUT most importantly, we should never forget that Social Work as a discipline, our goal is to improve human conditions. We should not become a “ist”, not matter frequentest, bayesianist, linear thinkers or system thinkers…. We should not categorize ourselves according to methods. Methods are just tools to help us improve human conditions.