standardized residuals and outliers

[iliana wouters]

hello everybody,

In my textbook of quantitative methos is said that before running a logistic regression I have to check for different factors first. I have to check for multicollinearity, quasi- or complete separation AND outliers.
Here is said that we can talk of an outlier if the standard residual (ZResid in SPSS) is >2 or <-2.
I calculated the standard residual as followed,
run logit
predict resid, rstandard

However, I think this is not the way. More, I hope this is not the way because the maximum value of resid is 100.45...

Can someone help me with this?

Kind regards,
Iliana Wouters

[iliana wouters]

this is my outcome when i plot it.. What should I do with the numbers higher than 3?

[Marcos Almeida]

Your query remained answered so far, and I presume it was due to scarce information.

That said, it shall be underlined that, under a logistic regression, the outcome variable is supposed to have either 0 or 1 values.

Please read the FAQ on how to post a highly informative query. Thank you.

[iliana wouters]

Dear Marcos,

Yes my output is a binary variable.
The output above is my output of the 2 commands: predict r, rstandard & stem r that I runned after my logit model.
I have red that i have to delete all the residuals that are greater than 3. But I don't know is this is the right thing to do.

[Nick Cox]

That advice appears to make little sense. If there is a case for deleting outliers, and I think there usually isn't, it's to be carried out according to raw data, not according to residuals after a model.

Marcos Almeida is unfortunately right here. There is little precise information in your post on your data and your exact commands so that people are very unclear about what you have done. If you can't post your raw data there is still advice in the FAQ on showing your problem with one of Stata's datasets or simplified fake data.

[Marcos Almeida]

Hello Illiana, in your previous message, it seems you meant "outcome" when you wrote "output", in the first paragraph.

That said, as Nick underlined, there is no clue in terms of command and data set.

I believe there must be something , well, wrong, with them.