1

I have been using fixed effects (from fixest package) on different type of panel data (longitudinal) models. Mainly parametric.

I was wondering if fixed effects could be apply to machine learning models like random forest?

(My background is not in stat/econometrics)

Shunrei
  • 31
  • 1
    What do you mean by "fixed effects" in machine learning? – Tim Nov 22 '22 at 14:06
  • It's widely used in econometrics for panel data. They represent time-invariant (related to time or subject) that we want to keep constant and control in our dataset in order to reduce the risk of omitted variable bias. – Shunrei Nov 22 '22 at 14:18

1 Answers1

0

Fixed vs random effects is a terminology used to describe different kinds of regression models (linear regression, generalized linear regression, etc). If you want to understand it better check threads like What is the difference between fixed effect, random effect and mixed effect models? and What is a difference between random effects-, fixed effects- and marginal model?. The terminology regards how we treat, interpret, and estimate the parameters of the models. In the case of machine learning models, some are non-parametric, e.g. $k$NN, decision trees, some technically do have parameters, but they play a different role as compared to standard regression models. So there isn't something like fixed vs random effects for machine learning models. Also, notice that the distinction is not that sharp, which is why some authors (e.g. Gelman and Hill) avoid using it alltogether.

Tim
  • 138,066
  • I will reformulate my question: Will it not be a problem in the ML model to have repeated information for each subject in the context of panel data ? I mean can it assess the net effect of the predictors on the outcome variable ? I know that ML allows to not have any assumptions but still I was wondering if it would not bias the results as in regression models we can control for time invariant characteristics. – Shunrei Nov 22 '22 at 16:28
  • @MyrTH why would it bias the results? If you are looking for machine learning models for panel data, that's a different question and we have threads on it. – Tim Nov 22 '22 at 19:05