1
$\begingroup$

I am running a pooled OLS regression as a benchmark model on a panel data set of online forum member activity. The aim of the model is to understand the relationship between exposure to hate speech and its adoption.

The pooled model therefore includes exposure to hate speech the previous month as the independent variable of interest, alongside variables controlling for the number of months spent on the forum and the total number of forum posts. The results of the regression can be seen below:enter image description here

Analysis of the residual errors from the pooled model suggest that the residual errors are not normally distributed, shown by a heavy-tailed Q-Q plot. While there is also evidence of heteroskedasticity and autocorrelation.

enter image description here.

To improve upon the pooled OLS model I plan on implementing a fixed effects model, controlling for time and entity (forum member) fixed effects via a process of de-meaning. This is done using the PanelOLS package from linear models in python and using robust and clustered standard errors.

Improve this question
$\endgroup$
7
  • $\begingroup$ Some of the terminology you use differs strongly between fields, so it would help if you could include more details, including the equation and/or a brief bit of code to explain the model. $\endgroup$ Commented Jul 26, 2023 at 8:24
  • 1
    $\begingroup$ Thanks for the heads up. I have edited the question slightly to include more detail - hope this helps! $\endgroup$ Commented Jul 26, 2023 at 9:20
  • 1
    $\begingroup$ Well done, and +1. I know this is not common in economics and some other fields, but have you considered partial pooling (via mixed effects models a.k.a. hierarchical models)? In my field this is considered superior to both unpooled and completely pooled models. $\endgroup$ Commented Jul 26, 2023 at 9:42
  • $\begingroup$ What's the question? $\endgroup$ Commented Jul 26, 2023 at 11:01
  • 1
    $\begingroup$ Also consider semiparametric models that do not make distributional assumptions. $\endgroup$ Commented Jul 28, 2023 at 12:09

0

Reset to default

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Start asking to get answers

Find the answer to your question by asking.

Ask question

Explore related questions

See similar questions with these tags.