0
$\begingroup$

I m currently doing some social science research using panel data to determine the impact of budget cuts on financial vulnerability. I have decided to use a fixed effects model to determine this association. When I use raw data, however, my results are insignificant (there are major outliers in my data). In order to reduce the impact of these outliers I have scaled the data using min-max scaler and this results in significant results.

However, I have seen some posts which advise against scaling panel datasets and I would like to confirm if sclaing for fixed effects regression is good practice?

Thank you!

$\endgroup$

1 Answer 1

0
$\begingroup$

If the outliers are wrong data, that is, if they are there as a result of an accident, mistake, error, or a malfunctioning device, then consider deleting them, or marking them as missing and then using a principled method to handle missing data such as multiple imputation

On the other hand, if the outliers are just extreme data points, then it would be bad practice to delete them. As for resclaing them, I don't even know how that would help at all, since the model should be invariant to rescaling.

If it is not be possible to know whether a data point is wrong or extreme, then in the first instance, common sense should prevail: for example, a measurement of 3.5m for an adult person's height is clearly an error since it is implausible. But a measurement of 2.5m, athough extreme, is nevertheless plausible. In such cases, one approach is to perform a sensitivity analysis.

Finally, try not to worry to much about statistical significance.

$\endgroup$
1
  • $\begingroup$ @ExploringDate Does this answer your question ? If so please consider marking it as the accepted answer. If not, please let us know why. Also, if you haven't already, please consider upvoting it $\endgroup$ Aug 21, 2021 at 18:30

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.