lvr2plot average leverage

[Barbara Heerkens]

Hi,

I need to perform some regression diagnostics for my thesis (I follow https://stats.idre.ucla.edu/stata/we...n-diagnostics/).
Hereby, I created a lvr2plot showing below. When simply looking far away far out of the red line, I do not really see trouble. However, in the chapter mentioned above they say that you should worry when leverage is above (2k+2)/n, which is .12820513 in my case, so the mean value (red line) is way above this number....
What do I do with that?

Thank you!

[Phil Bromiley]

You didn't get quick response. You'll increase your chances of a useful answer by following the FAQ on asking questions - provide Stata code in code delimiters, readable Stata output, and sample data using dataex.

The entire area of outliers is filled with alternative views - some think they should be largely ignored, others prefer different techniques. It is generally an area in which guidance from theory is hard to find.

I often prefer cook's d to simple leverage. The regress postestimation documentation has some guidance on what suggests a problematic observation. I would ignore the leverage guideline you mention - it makes no sense to imagine such a large proportion of your observations are problematic.

The first thing to do is always to double check that the extreme observations are correct observations - no error in transcription, no error in calculation, or whatever.

One way to view this is that you don't want your results to be dominated by a very small number of observations. As a robustness check, I'd see if dropping the top 1 or 4 observations for leverage and/or the top 6 for largest residual substantially change your results. I might also check dropping a few observations with highest cook's d.