Thomas Sowell on a question many answer with confidence but may have never examined:

Virtually no one has seriously denied that discrimination and bias have resulted in various inequalities.

It is the converse proposition—that discrimination or bias can be inferred from statistical inequalities—which is the reigning non sequitur of our times, both intellectually and politically.

To prove statistically that the observed patterns of representation or reward are not due to random chance is considered to be virtual proof that they are due to discrimination—not to performance differences.

The implicit assumption is that a more or less even or random representation or reward for performance could be expected, in the absence of institutional or societal policies and practices which disadvantage one group compared to others.

Yet there has never been an even or random world, even in matters not controlled by the biases of others. Not only performance differences but also differences in luck and in many other factors wholly disrupt the simple picture of an even, regular, or balanced world. . . .

What is wholly unsubstantiated is the prevailing assumption that the world would be random or even, in the absence of discrimination or bias by individuals, institutions, or “society.”

—Thomas Sowell, The Quest for Cosmic Justice (New York: Touchstone, 1999), 62-63.

View Comments


2 thoughts on “An Unexamined Assumption: Can Discrimination or Bias Be Inferred from Statistical Inequalities?”

  1. Marcus says:

    So, I’m a statistician, and this argument may or may not be terrible and completely misunderstanding how statistics works. It’s hard to know apart from the broader context. His argument is not terrible if he’s talking about population level estimates (e.g., whites as a whole make more money on average than African-Americans). However, the argument that “you don’t know that it’s discrimination and not something else, it could be luck” seems to me to be taking the easy way out. It’s not as if arguments that there is discrimination, or that the effects of past discrimination are causing the inequality at the population level are not implausible. He would need to come up with a better explanation to account for the data.

    If he’s talking about individual studies that rely on samples (e.g., studies done that test if men are more likely to be promoted than women on the basis of managerial review of fake resumes), then what he’s saying is very unhelpful, particularly if they’re done more than once and you have confirmation of the results. Repetition reduces the odds of results being “luck.” Yes you can never measure everything that may have an impact and it’s possible that some explanatory variable is being overlooked, but you never have perfect measurement and data anywhere in life. I’m sure he doesn’t hold himself to this same standard in his own work.

    In the end, this strikes me as the type of unfalsifiable skepticism that I find to be a waste of time. Since you can’t measure perfectly, and don’t know every possible explanation you can never prove him wrong. That’s a totally irrational stance.

  2. Matt says:


    I won’t begin to catalogue them; but Sowell gives repeated examples of exactly the sort of thing he is talking about throughout many of his works. I’d encourage you to check them out. I don’t think he (Sowell) is partaking in skepticism, much less the sort you imply.

Comments are closed.

Justin Taylor photo

Justin Taylor, PhD

Justin Taylor is executive vice president of book publishing and book publisher for Crossway and blogs at Between Two Worlds. You can follow him on Twitter.

Justin Taylor's Books