Question about detecting influential observations in panel data

[Alex Mai]

Dear Statalist,

I have a question about how to detect influential observations (in terms of leverage and/or residual) in panel data. For cross-sectional data, Stata provides options like -rstandard-, -rstudent-, and -leverage- after -predict- to allow to detect observations with large residuals (outliers) and/or high leverage, either statistically or graphically. However, for panel data, these commands do not work. In particular, I have learnt from previous thread that for panel regression there is no agreed definition of standardized residuals, which makes it difficult to detect outliers from residual plot.

So I would like to ask for suggestions about how to detect influential observations (outliers and/or observations with high leverage) in panel data. The purpose is actually not to deal with these influential observations in estimation, but more to just find out them.

Thank you very much!

[Alex Mai]

Any suggestions would be appreciated. Many thanks!

[Phil Bromiley]

One option is to do the fixed effect estimator with i.panel in regress and then use the tools associated with regress. I don't know whether this is legit.

[Alex Mai]

One option is to do the fixed effect estimator with i.panel in regress and then use the tools associated with regress. I don't know whether this is legit.

Dear Phil,

Sorry for disturbing, but may I follow up this thread? I am reading the Stata manual on xtreg postestimation. The manual gives the description for -predict- as the follows:

predictions, residuals, influence statistics, and other diagnostic measures

It mentions "influence statistics". However, I do not find anything relevant to influence statistics (outlier and leverage) in the manual on xtreg postestimation. So may I ask if you have any clue for this "influence statistics"?

Also, do you think whether it is safe to say that there is no good method to detect influential observations for panel regression (e.g. as a response to comment)?

Thank you very much!