=> yujiri.xyz | Argument

Statistics are not the trump card you think

last edited 2024-11-05

Most times I see someone use a statistical argument online, I shake my head and scroll down to the next post. Here's why.

Real-world situations usually have too many variables for statistics to be useful

Obviously, to use statistical data to establish that A and B are causally related you have to isolate A and B in your comparison. It won't do to sample a bunch of red squares and blue spheres and conclude that blue objects roll better than red objects. But this is exactly what many statistical argments are.

I once argued with a Jordan Peterson acolyte about whether there are intrinsic psychological differences between men and women. The acolyte claimed there was some study showing that in cultures that try to treat men and women as if they don't have intrinsic psychological differences, the differences become more extreme. I pointed out that even if his claim about the study was accurate, it's still a bad argument because the way a culture treats gender is never the only difference between it and another culture. A real society has thousands too many variables for this to be a reasonable argument. But he got mad at me for refusing to treat his "study" as instant proof of his claim and wouldn't continue the debate.

Statistics often measure the wrong thing

I've seen people argue for gun confiscation saying that America has more gun homicides per year than a bunch of other (smaller) countries who have stricter anti-gun laws... only to find out that they were measuring the gun death rate as a flat number rather than a proportion of each country's population. Of course America will have more gun homicides if you measure it that way because America has more people than those other countries.

(Not to mention that argument is atrocious anyway for, again, ignoring the thousands of other variables involved.)

Similar common fallacies to that conclusion include counting suicides and defensive killings of criminals in one country but not in the other, etc.

Who made the judgements?

This doesn't apply to all statistics, but for ones about things that are open to interpretation, there's another problem. For example, statistics about sexual harassment are often quoted as facts about how many people have experienced it, but who decided which cases count as harassment?

Statistics can have ambiguous implications.

Even if you prove there's a non-coincidental correlation between A and B, that doesn't prove that A is the cause of B. You still have to consider the possibility that B is the cause of A, or that a certain C that wasn't considered in the study causes both B and A, or that a combination of A and C is required to cause B.