Full page fax print

Multicollinearity- What is it, its impact, diagnostics for identification and remedies
It is a phenomena in statistics where two or more predictor variables of a multiple regression are
highly linearly related. It is a situation where the co-efficient estimates show an erratic response on
small changes in the model or data. It does not reduce the reliability or predictive power of model, it
only alters the calculations regarding individual predictors.
We get perfect multicollinearity if, the correlation among two independent variables is either 1 or -1.
It happens because of following reasons:-
a. Dummy variables are used improperly
b. Computing a variable that is calculated using other variables in equation (e.g. Profit =
revenue-total cost, and the regression includes all three variables)
Impact
 Greater the multicollinearity, greater the standard errors
 If high multicollinearity is there, confidence intervals tend to be very wide for co-efficient...