Leverage points of 1 in Multiple Regression

[Andreas Manuel]

Hey there,

I have got a dataset of about 180 observations and several explanatory and control variables. After winsorizing my variables, regressing my data and calculating the leverage, I find that there are several observations having a leverage of 1.

Can somebody explain to me the idea of having a leverage point of 1, whether this is realistic and how to find out which variable of my observation causes the leverage of 1?

Thank you!!

Andreas

[Carlo Lazzaro]

Andreas:
my temptative reply is to start out with one predictor, calculating -predict leverage_1, leverage-, then add another predictor, calculate leverage_2 and so forth until the last predictor:

reg y x1
predict leverage_1, leverage
reg y x1 x2
predict leverage_2, leverage
sum leverage_*

and see when high leverage values start to come alive (unfortunately, I am not aware of any breakpoint that split acceptable from non acceptable values).
Things would be easier if you were interested in investigating the observation(s) with higher-than-average-leverage(s), as -help lvr2plot- would do the trick.

Kind regards,
Carlo

[Joao Santos Silva]

Andreas,

It looks like you have in your model dummies that are equal to 1 for a single observation; these will have leverage equal to 1.

Joao

[Andreas Manuel]

Thank you very much for your help, Joao & Carlo. It turns out that I have some observations which are the only ones in their industry (dummy variable) and this leads to the leverage of 1.