Identity, correlation and causation

From Race and Gender Shape Views of Science in Surprising Ways by Mark Strauss.

In part this article is reporting the already well known fact that opinions on different subjects vary significantly by categories of identity such as race, class, gender, orientation, political affiliation, religion, level of education attainment, regionalism, nationality, age, profession, industry, familial circumstance, etc. Women believe X more than men. Mainline Protestants believe Y more than Catholics. Young people believe Z more than older people. And so on. This is the stuff of textbooks.

What I found interesting was the subtle framing of the narrative of the article. I do not know whether this arises from unfamiliarity with maths and statistics or whether it has some other basis. For example, take the third paragraph.

A survey published Wednesday by the Pew Research Center reveals that Americans’ views on many science-related policies are more likely to be shaped by demographics—age, race, gender, religion, and education—than by political beliefs, often with surprising results.

[snip]

The extent to which those factors—either alone or combined with others—influence people’s views varies by topic (see chart above).

The subtle inference is that Identity X causes you to hold Opinion Y. What the data is of course saying is that if you have Identity X, you are more likely to hold Opinion Y. The data says nothing about causation. If you are interested in Opinion Y, you want to know the causative reason people hold that opinion, not whether there is a correlation.

For example, there may be a correlation between the level of education attainment at the Master's degree level and the opinion that the rate of income tax ought to be lower. It is not correct to say that having a Master's degree shapes or causes you to have the opinion that income tax rates ought to be lower. The reality is that we don't know why there is that correlation because it is simply a correlation. With investigation, we might find that education attainment covaries with income because those with Master's degrees actually have higher incomes (think of all those doctors, lawyers, MBAs, etc.). So the causative relationship is actually between Those With High Incomes and Opinion That Income Tax Rates Ought to Be Lower. The correlation with education is an indirect covariance. Or possibly it is an entirely spurious correlation (see Spurious Correlations website for fun examples).

Strauss is framing his discussion as causation when it is actually correlation. This is the central problem I have with those who couch everything in terms of Identity (whether by race, gender or any of the innumerable other forms of Identity). First it too often obfuscates the distinction between causation and correlation, second, in confusing the two issues it obscures real causation and third, it discounts the interplay between multiple identities. Being Male may statistically correlate with (but does not cause) a belief that GMO is safe but perhaps that correlation is trumped by the fact that I live in a commune or that I am a member of Earth First or some other attribute. All correlations are a form of averaging and all averages disguise the degree of variance in a given population.

Anybody who trades in monolithic identities and purported uniform beliefs held by those monolithic identities is committing statistical fraud and trying to exercise ideological hegemony. Look at individuals as individuals and consider individuals in terms of the content of their character as manifested through their actions. Forget this racialist, classist misanthropic tendency to lump people into generic categories and deal with averages rather than real people in all their variety.

As long as journalists fudge the distinction between correlation and causation, we delay the day when we deal with the real world of flesh-and-blood individuals instead of an ideological world of abstractions.