Quote Originally Posted by Tom OC View Post
Most of this theory-testing only produces R-squares of .20 or .30 at best, which means a large percentage of known factors remain unknown. That's what makes social science a soft science, I suppose. We aren't dealing with close tolerances or things with 999.99% certainty like chemistry.
Tom,

If memory serves, R-square values are a numeric expression of the probability that a given model will predict future behavior. Perhaps the problem is social scientists are trying to predict behavior. This seems somewhat difficult to accomplish in a social system where the agents have a choice, emotions, subjective rationalities, cultural forces, etc...

By way of contrast, in the "hard" sciences atoms (above the quantum level), molecules, etc... obey predictable laws. Thus, it would seem models which predict the behavior of agents (that themselves must follow predictable laws) would result in very high R-square values.

I don't mean to assert that there isn't a significant amount of stupidity in the social sciences (there is in every discipline). Rather, I would suggest less that reliable predictive models in social systems says more about the system in question and the approach to understanding it than it does about the scientists.

Often, the answer you get depends on the question you ask. Perhaps we are asking the wrong questions? I would argue the failings in social science are related to our attempt to study it as if it were a hard science; that is to say reductionist, analytical, linear thinking.

For example, if you are doing any type of research you must state your independent and dependent variables. However, social systems are not composed of independent and dependent variables, and applying such a construct is doomed to fail. The construct asks the wrong question, i.e. "What are the cause and effect relationships?" There are few cause and effect relationships in social systems because people have choices.

Social systems are composed of interdependent variables. Therefore, we cannot study one or two in isolation, but we must study the system as a whole to understand the interdependency of the variables and the emergent properties of the system.

Additionally, classical sciences attempts to remove context from the equation in order to isolate the cause and effect relationships between variables. However, context is everything in a social system. To study a social system without context is to invite failure. Results of context-free experimentation will not be useful in the "real world" because context exerts a heavy influence on behavior.

In short, social systems can't be studied like physical or chemical systems, yet this is what we are doing. As long as we continue to do so, we are unlikely to have much success.