How do I interpret the results from a redundant variable test (EViews)? The odd thing is that it cannot reject the null of a variable with a significant coeffecient (p-value 0.0277) to be insignificant (p .1011). My guess is that removal of the variable does not alter the regression significantly. Is that correct? But still, the variable should be included anyway, right?
You likely have collinearity. Check with Variance Inflation Factors. Simply google this term to lead to how to do it or contact me. Best wishes. David
Lars -
I don't know how eviews makes its determination, but it might be a good idea for you to check documentation and find out, for now and future reference.
Regarding looking at a p-value to determine "significance," that is a misnomer and the misunderstanding is still widespread. The p-value is misleading and by itself represents incomplete information.
You might want to remove part of your data as test data, and use all other data to predict, using one regressor model with that variable, and one without it, to see which model appears to predict for the withheld y-values better, and hopefully you might have enough data to be able to withhold enough to get a reasonable check on the overall estimated variances of the prediction errors as well.
Residual analysis graphics might help, using predicted y on the x-axis where the usual one-regressor illustrations use x.
You might want to research "cross-validation."
In general, omitting an independent variable you need may bias results (omitted variable bias), and including an independent variable that you do not need tends to inflate variance. (Note, however, that I have found in practice that omitting a needed variable can increase the estimated variance of the prediction error, and I explain that by considering that it can inflate the calculated sigma estimate.)
Cheers - Jim
https://www.openforecast.org/adam/model-specification-redundant-variables.html
- Badges
- Science topic