0
$\begingroup$

I have a time-series model, with stationary variables. Testing for outocorrelation hasn't been easy: - a can't calculate DW stat because of my small number of observations; - the GB test indicates that there is autocorrelation - the Correlogram Q statistics tells me differently, indicating that there is no autocorrelation...

what can cause this contradictory results? how do i manage this?

Improve this question
$\endgroup$
1
  • $\begingroup$ Are the various results all bouncing around your cutoff for significance (ie, some p=.04, some p=.06)? $\endgroup$
    – gung - Reinstate Monica
    Commented Oct 22, 2012 at 16:30

1 Answer 1

Reset to default
-1
$\begingroup$

Are you relying on statistical significance tests? That could lead to this result.

Your language makes me think you are relying on these, since you say "is" vs. "is not" autocorrelation. There is almost certainly some autocorrelation, and the question is not whether it is statistically significant or not, it is whether it is big enough to matter.

What were the effect sizes for the two tests?

Also, you should graph your data and see the autocorrelation.

Improve this answer
$\endgroup$
9
  • $\begingroup$ the problem is that i need to rely in statistical significance tests since it is my thesis, and i can't find any other example of the same problem in the internet. How can i correct the autocorrelation? $\endgroup$
    – Cindy Sabenca
    Commented Oct 22, 2012 at 16:28
  • $\begingroup$ i can't create an AR(1) model because of the limited amount of observations... $\endgroup$
    – Cindy Sabenca
    Commented Oct 22, 2012 at 16:29
  • $\begingroup$ If you have so little data that you can't run an AR(1) model, then you probably don't have a data set suitable for time series. But you haven't told us anything about your project at all, so there's no way for us to offer help. Please read How to ask a statistics question $\endgroup$
    – Peter Flom
    Commented Oct 22, 2012 at 16:36
  • $\begingroup$ I am conducting a Study on poverty and inequality in Portugal through an OLS regression, from 2002 to 2010 (which are the only years for available data)... i am in such a stressed moment that i don't even know how to ask the right questions.. but my big problem has to do with these contradictory test solutions on the matter... thank you for trying to help... $\endgroup$
    – Cindy Sabenca
    Commented Oct 22, 2012 at 16:43
  • $\begingroup$ Have you considered a multi level model? $\endgroup$
    – Peter Flom
    Commented Oct 22, 2012 at 16:44

Your Answer

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

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