# Discriminant, logistic approach to attrition analytics

I have a few variables for each employee like tenure, age, marital status, certification, working away from home, #of OThours in an year, last promotion, performance rating, etc...

I have these variables and few more for employees who had left and employees who are working. If you see my variables, few are continuous and others are categorical, is it possible to run a discriminant analysis on this and identify which variables are discriminating better between a person leaving/not leaving the company and to forecast for existing employees or should I only use logistic regression since DA may not support discrete variables in the equation, or do we have any other method to achieve my goal?

• What is "attrition analytics"? Did you mean to say "attrition analysis"? Feb 25 '15 at 4:37

You can choose the suitable technique after checking the following assumptions with your data

Assumptions ------------- Discriminant Analysis ---- Logistic Regression

Multivariate Normality ->   Yes                       No
Homoscedasticity       ->   Yes                       No
Non-Multicolinearity   ->   Yes                       Yes
Absence of Outliers    ->   Yes                       Yes
Large Sample Size*     ->    No                       Yes

(Predictor Variables
can be Categorical,
Continuous or Discrete) ->   Yes                       Yes

*Based on ten (10) independent variables, Discriminant analysis would need minimum of 50 cases and Logistic regression would need minimum of 250 cases.


Well, I have applied Logistic regression for my organization data to predict the attrition. I see very similar situation you have. You could also refer this document. Hope this helps !

• Sample size needs are the same for logistic regression and DA. Feb 25 '15 at 11:57