A User’s Guide to Statistical Inference and Regression | Hacker News Reader

A User’s Guide to Statistical Inference and Regression

mattblackwell.github.io

148 pointssebg7 comments

Man I love this first paragraph. Love it.

“Quantitative analysis of social data has an alluring exactness to it. It allows us to estimate the average number of minutes of YouTube videos watched to the millisecond, and in doing so it gives us the aura of true scientists. But the advantage of quantitative analyses lies not in the ability to derive precise three-decimal point estimates; rather, quantitative methods shine because they allow us to communicate methodological goals, assumptions, and results in a (hopefully) common, compact, and precise mathematical language. It is this language that helps clarify exactly what researchers are doing with their data and why.”

A lot of industry “data science” is moronic, which is to say not clear about assumptions or about what is and isn’t possible with the estimates. Leading case: Models are treated as causal when they are not. Antidotes to that kind of thinking are great. Hopefully this can be one of them. I’ll keep looking at it.

If you want to take this a step further, quantitative methods are about efficient data reduction. As part of this data reduction, the model’s assumptions and mathematical form take center stage in describing how you got to your “number”.

This is different from qualitative analysis because the data reduction is done “by hand” by the researcher.

The difference between the “automatic”, model-based data reduction in quantitative research and the “subjective” reduction in qualitative research is then amplified when people say that quant is more objective than qual analysis. The discussion, instead, should be about the quality of the work and whether the final conclusions are warranted by the methods instead of the method itself.

Yep. There’s unfortunately a large contingent of people, usually the people that haven’t conducted quantitative research themselves, but have maybe read some, that are just impressed by numbers. It’s like the next level up from people that say “science says that …”.

It’s the “data doesn’t lie” people.

They believe themselves to be objectively data-driven but don’t see the forest because they’re so busy looking at the trees they are shown.

It's far worse than that. People using numbers as supportive arguments often generated said numbers themselves, from scratch (meaning, they also collected the data). I'm currently redoing a study from an old prof. The old study was of a more trusted design, and found a massively positive effect. The new study (including old data) has an effect firmly grounded on zero. Those people aren't even always dishonest. They're just incompetent.

And yet… the plug and play nature of many ML methods and their frequently positive impact when applied so has probably played a large part in the growth of the field.

good one

Reply on news.ycombinator.com