0

I carried out White's test for a particular model to check for heteroskedasticity. I got the following results for White's test (details shown below). AS it can be seen the p-value (0.299) isn't that small so I interpreted it as not having to reject the null hypothesis. However, I also decided to do an rvfplot and I've attached a picture of how it looks as well. To me it doesn't display homoskedasticity. Are my interpretations incorrect?

enter image description here

Image of the rvfplot:

enter image description here

Kurapika
  • 23
  • 3
  • 1
    The residuals look very nicely distributed to me, I would say there is no heteroskedasticity. – user2974951 May 26 '21 at 08:03
  • I was a bit uncertain with the diagram but this helped me to clear some of the confusion. Even then if I try to re-estimate the model with heteroskedasticity-robust standard errors (despite the fact it seems unnecessary) and end up getting different values for the standard errors, test statistic and p-values then what would you conclude about the model? Does it have some heteroskedasticity in the end? – Kurapika May 26 '21 at 08:11
  • 1
    That looks absolutely fine for me, too. – Bernhard May 26 '21 at 08:46
  • 1
    @Kurapika Where you have fewer data points, the range of your data points is generally also smaller. This is why a fitted vs residuals plot isn't the best for judging heteroscedasticity on its own, but you should also inspect a scale-location plot if you suspect non-constant variance. That said, I agree with the the others that there is nothing remotely problematic judging by the residuals vs fitted plot. – Frans Rodenburg May 26 '21 at 09:53
  • 1
    Also, note that the robust and conventional standard errors will not be numerically identical in finite samples even if there is no heteroskedasticity. – Christoph Hanck May 26 '21 at 10:03

0 Answers0