3 ms·
I think this is a weird view of how bias works. statistical models are imperfect models and are always going to have lots of bias or variance. We typically want
by jtmcmc 8y ago
I think this is a weird view of how bias works. statistical models are imperfect models and are always going to have lots of bias or variance. We typically want to bias models using domain expertise or other external factors.
This is like being mad at a company for using L1 regularization
- humanrebar 8y agoThe difference is how much practical diversity we'll have in models. With enough effective choice, the severity of biased models goes down, at least somewhat. Models biased in different directions can provide a variety of perspectives.
- _dps 8y agoThese are different uses of the word "bias". In the statistical sense there is an objectively true right answer and "bias" refers to the difference between a model's average answer and the average true answer. It has little to do with the lay use of the word to mean the application of an inappropriate subjective value judgement in a situation with no one true answer (unless you take an extremely anthropomorphic view of statistical formulas).