Help with heteroscedstisity

[Simeon Ivanov]

Dear all,

I am currently having to test small sample ( 20 ) of crossectional data with 7 regressors.

Currently, I have done Breush-Pegan and White test for heteroscedasticity but I'd like to run any others that might help.


I hope you can help with name and command in Stata


Thank you for the time spent!

Simeon!

[Dave Airey]

help regress postestimation
help estat hettest

make diagnostic plots

https://www3.nd.edu/~rwilliam/stats2/l25.pdf
http://stats.idre.ucla.edu/stata/web...n-diagnostics/

[Joao Santos Silva]

Dear Simeon,

Heteroskedaticity is the least of your problems: estimating 8 parameters form 20 observations is asking too much form the data and standard inference is unlikely to be valid even if you have homoskedasticity. Anyway, the heteroskedasticity tests are only asymptotically valid and they will be of little use in this context.

Best wishes,

Joao
PS: Nick Cox -whitetest- produces a result in this case but I do not think the test is feasible because there are not enough degrees of freedom. Can you clarify?

[Carlo Lazzaro]

Simeon:
welcome to the list.
As an aside to Joao's helpful recommendations, please consider that 20 observations per predictor are advised for multiple regression (Katz MH. Multivariable Analysis. Second Edtion. NY: Cambridge University Press, 2006: 81), even though 10 obs per predictor may sound wise enough.

[Joao Santos Silva]

Following up on Carlo's comment, the results in Portnoy (1988) suggest that in general we need k²/n to go to zero as n goes to infinity (where n is the sample size and k is the number of parameters). So, if the number of parameters to estimate is large enough, even 20 observations per parameter won't be enough for reliable inference.

Best wishes,

Joao

[Simeon Ivanov]

Dear all,

Thank you for the good responses!

I know that the sample size is small but I can't do anything. This is because it is part of university assignment and it is killing us to do any part of it with this small sample.

Still, I have to

Describe and use an alternative procedure which one might use to test for the
presence of heteroskedasticity in the disturbances of a regression equation

. Before that, we made a Breush-Pegantest.


So anything that you could suggest?

[Marcos Almeida]

Such a small sample "invites" to a Bayesian approach as well.

That said, under "real life" conditions, 7 predictors for 20 observations, as previously commented, is just pushing too much the envelope.

With regards to assignments, being this the case, please take a look at this FAQ , particularly the topic 4.

[Simeon Ivanov]

I know that it is not good to help with homework and this is why I do not say anything about the data or what exactly I can see in it.

I am just asking for material to read and any suggestions on what can be done

Thank you all in advance

[Carlo Lazzaro]

Simeon:
see:
-help regress postestimation plots-
-help regress postestimation##estatimtest-
Eventually, you might find the following textbook useful: http://eu.wiley.com/WileyCDA/WileyTi...HEP000911.html