Missing values in a significant variable during the modelling process

[Shivi82]

Hi.

During the Modelling process, if we have selected some of the variables and we see that for one of the very significant variable the values are missing so is there a threshold or industry specified % for the variable to be kept in the model and ignore the missing values.

I understand that we can replace the missing values with either the mean or the median of the variable however wanted to see if statisitcally there is a threshold.

Regards, Shivi

[stat_sas · Accepted solution]

Hi,

I think there is no standard threshold for variables with missing values to be considered for modeling. Normal practice is to drop a variable having 50% missing values. If you consider variable having 50% missing values is important for modeling, then you can create an indicator variable and use that instead.

[Shivi82]

thank you.