if a & b are both continuous variables, you should center them first to reduce multicollinearity (the correlation between a & a*b, and b & a*b), and get centered_a and centered_b. Then you compute centered_a*centered_b, and finally regress Y on centered_a, centered_b and centered_a*centered_b.
Hope this helps, mate.
** sorry that I am a Chinese who have been living overseas for years.... I can read Chinese characters but cannot write/ type.
|