The Endogeniety Problem

Inside, outside, leave me alone
Inside, outside, nowhere is home
--The Who

Lots of empirical research is tainted with an endogeniety problem. Suppose that you are a medical researcher, and you suspect a relationship between consumption of soy and the onset of cancer.

So you commission a survey study, asking people whether they have ever had cancer and also asking if they consume products with soy in them. The test seems straightforward. If soy consumers have a higher occurence of cancer than non-soy consumers, then the results seemingly show a relationship between soy and cancer.

However, there are other factors that might influence the onset of cancer besides consumption of soy products. Age, genetics, exercise, other dietary factors, etc. If people who are prone to contract cancer by these other factors are also likely to consume soy, then your study is plagued by a selection effect. Soy consumers are not a cross section of the population and may be more susceptable to cancer.

Consequently, your survey will take on two effects. It will reflect the possibility of a positive relationship between soy consumption and cancer as hypothesized. But it will also reflect the possiblity of contracting cancer due to other peculiarities of the sample. Thus, your survey is likely to overstate the true relationship between soy and cancer.

Because the subjects were allowed to choose the treatment, the treatment choice is 'inside' or part of the study. This is the endogeniety problem.

One way to avoid the endogeniety problem in survey research is to divide subjects into control and trial groups. This may be done by flipping a coin. Another way is to divide subjects according to whether they were born on odd or even days of the month.

The priniciple is to not permit subjects to self-select into the treament and control groups. Instead, it is best to have nature choose for them. In survey research, the key to avoiding the endogeniety problem is to conduct a 'natural' experiment.