This is the Base regression:
Y = b0*X1+b1*X2+error term
I am using an indicator variable Z which takes value 1 for years 4 to 6 and value 0 for years 1 to 3.
I am trying to analyse the impact of Z.
when I partition the data as two sets - years 1 to 3 and years 4 to 6 --I am able to proved with Adj Rsq that Z has indeed improved the model-fit.
But, when I run a pooled regression for years 1 to 6, ie,
Y = b0*X1+b1*X2+b3*Z+b4*X1*Z+b5*X2*Z+error term, I get the b4 and b5 as insignificant..
The expectation is that If in the partitioned data, adj R2 is higher in years 4 to 6 than in years 1 to 3, then it means that Z has improved the model fit and hence b4 and b5 should be positive and significant in pooled model.
But, my pooled regression shows b4 and b5 as insignificant?
Can anyone help me how to interpret the result?
Shruti R Ashok
You need to regress your equation through robust estimates and also check the vif (variance inflation factor). Sometimes a simple regression may not give desired results due to the problem of multicollinearity and heteroscedasticity as well. A robust estimate normally will resolve this issue. If the issue still persists then you need to increase the number of observations. Sometimes limited observations in your dataset may also cause a variable to become insignificant.
If the mean of your dependent variable is not zero, you must insert a constant into the regression equation. In addiction, you consider the mean of whole sample as well as those of the sub-sets. For instance, if the mean of your full sample is zero, but the means of your sub-sets is, you then should consider specifying appropriate intercepts.
... For instance, if the mean of your full sample is zero, but the means of your sub-sets is not, ...
- Badges
- Science topic
- Similar topics
- Mathematics
- Statistics
- Regression