heteroskedasticity and normality in the error terms

[Philip Murphy]

Dear all,

I have run a fixed effects regression using robust standard errors but I wanted to look visually for heteroskedasticity, how would i do this ? as it does not allow me to view the residuals against their fitted values? when i use the command 'rvfplot STATA says that the last estimates were not found

Furthermore why is it important for the error terms to be normally distributed and homoskedastic in a regression?

Many thanks

[Carlo Lazzaro]

Philip:
did you run -regress- or -xtreg-?
Please, post what you typed and what Stata gave you back (as per FAQ). Thanks.

[Philip Murphy]

Hi Carlo,
sorry my command was

xtreg lnRelativeRevenue  lnYearRanking l.lnYearRanking l2.lnYearRanking  i.LeagueNumber  ib3.PR l.ib3.PR  , fe robust

with my data set being declared as time series

xtset TeamNumbers SeasonStart

and stata returned

Would I also be able to test for serial correlation in the error term from this?

many thanks

[Nick Cox]

rvfplot is only supported after some regression commands.

You need to run predict to put your idea of residual and your idea of fitted in new variables and then fire up scatter.

[Carlo Lazzaro]

Phil:
thanks for the clarifications.
if you're working with a large N, small T panel dataset (as you should, given that you typed -xtreg-), robust/clustered SEs deal with both autocorrelation and/or heteroskedasticity.

PS: crossed in the cyberspace with Nick's helpful reply.

[Philip Murphy]

Thanks Carlo and Nick,
Could I just ask how the cluster command would apply here? as I am not sure what variables shouldbe used with the cluster command?
and what does STATA mean when it says in the top right hand corner of my output that the standard error has been adjusted for 91 clusters?

Thanks so much for your help

Many thanks

[Carlo Lazzaro]

Philip:
Stata tells you that SE have been adjusted for the 91 Group of observations that exist in your dataset.
The variable use for clusering is the -panelid- -TeamNumbers-.
Pedantic corner: I would refresh my knowledge about those issues, which are covered in any decent econometrics textbook.