Estimate Effect Sizes

post by Vaniver · 2014-03-27T16:56:35.113Z · LW · GW · Legacy · 5 comments

I think the following recommendation might reduce a particular negative reaction to blog posts and claims by 10%-50%, and generally improve the precision of discussion by a slight amount.

There are many phenomena that we observe that we aren't certain how to explain, and discussing potential explanations is a good way to converge on better explanations. Many phenomena are complicated, and so simple explanations are unlikely to explain everything- but a combination of simple explanations might explain most (or even all) of the phenomena.

One common way to approach this issue is to throw out a list of possibilities that could explain the phenomenon. This can run into trouble several ways: a list that is meant to explain only a fraction of the total effect could be read as exclusive, the list might come across as mutually exclusive instead of cooperative, or the relative importance of the various proposed possibilities might not be clear. It can also be bothersome to have to list the entirety of the possible alternatives or co-factors whenever you want to focus on a particular explanation.

So consider estimating the effect size as a way to clarify and limit your claims. Numerical estimates are likely to be more communicative than verbal estimates (what does it mean to be a "minor" factor, or a "significant" factor?), but harder to generate. Leading with your estimate, rather than adding it as a caveat at the end, is likely to be more communicative. This often degrades the flow of an argument, but as the point is often to prevent people from disconnecting from the argument, flow afterwards is not going to be meaningful for those people.

Counterargument: Robin's post Against Disclaimers seems relevant. This post was inspired by the negative reaction to this article by Yvain.

5 comments

Comments sorted by top scores.