Disparate impact and confounding variables

A very interesting article that brings together issues of disparate impact, disadvantage, causal density, correlation versus causation, confounding variables etc., A lefty’s lament by Alvin Powell.

The issue is that left handed people are disadvantaged for a variety of reasons, compared to the right-handed population. In terms of life outcomes such as morbidity, mortality, income, education attainment, etc, a nominal reading of the data indicates that lefties suffer measurable discrimination.

Goodman looked at five large studies of individuals in both the United States and the United Kingdom. Along with data on education and employment, the research included questions about handedness. Some aspects of the studies were in line with earlier investigations, such as lefties making up 11 to 13 percent of the sample and the children of lefties being more likely to be left-handed themselves.

A child’s health early in life played a role in handedness, with lower-birth-weight babies and those experiencing complications at birth likelier to be left-handed. U.S. babies that remained in the hospital for more than a week, for example, were 50 percent more likely to be lefties.

Lefties had slightly lower cognitive skills than righties, more behavioral and speech problems, and were more likely to have learning disabilities. There was no evidence to support the commonly held belief that lefties are over-represented among the very smart.

The negative patterns identified in the research have effects that play out first at school and then at work. Goodman found that American lefties are 2.4 percentage points more likely to end their education after high school; those who do advance are 2.9 percentage points less likely to complete college. When they hit the job market, lefties are more likely to work in less cognitively demanding jobs, more likely to do manual labor, and less likely to hold professional or managerial positions.

The effect on annual earnings is significant, with lefties earning an average of 6 percent less than righties. In the United States, the wage gap translates to about $1,300 per year.

Across most of the life outcomes, lefties are disadvantaged to the tune of 2-3%. Not huge, but persistent.

But it forces three questions. 1) Is this really an outcome based on discrimination or are there other factors at play?, 2) What can be done about it? and 3) Should we do anything about it?

What is interesting is that the research by Goodman reveals that the negative outcomes are not associated with left-handedness per se.

The earnings divide disappeared, however, when Goodman compared righties with lefties who had been healthy as infants, an indication that it may result from the effects of poor infant health on development and education, he said.

So it is not discrimination that causes the different life-outcomes, or even the inconveniences of being a lefty in a right-handed world. The problem is that there is a confounding variable to do with childhood health which is the actual driver for the differences.

By knowing the real root causes it allows us to focus on the real causes of the problem, childhood health circumstances. We don't need to do anything about discrimination against lefties because it is not a cause of the differences in outcomes. Instead we can look at the real root causes but we then still need to figure out whether there is anything that can be done about childhood health related to left-handedness and whether it is actually worth intervening given risks, unintended consequences, and the small effect size.

I find this interesting because it is a clear, non-emotive, rebuttal of the approach we too often take in public discourse which follows something like these steps - 1) Identification of a differential in outcomes, 2) Assumption that the difference is due to discrimination in some fashion, 3) Failure to quantify the effect size, 4) Failure to look for alternate sources of cause, 5) Failure to deal with confounding variables, 6) Reliance on superficial correlations as cause, and 7) Development of ineffective policies to address the false root cause.