What if interaction wipes out my direct effects in regression?
We have a question regarding a regression we are working on. We have two dummy variables, net income (If net income this year is higher than last year it equals 1, and 0 if opposite) and cash bonus (1 if cash bonus, 0 if no cash bonus). The dependent variable is derivatives use (1= use of derivatives, 0= no use of derivatives). We also have several other independent variables (both categorical and continuous), but we have made an interaction term by using the two dummy variables. When including all three components (the interaction term and dummy variables), none turns out significant. However, when we only include the interaction term it is significant, and also when including all other independent variables (exept from the two dummies that the interaction term is constructed from). Would it be okay to only include the interaction term in the model? Also, there is a high degree of collinearity between the two dummies and the interaction term. We are interested in finding out what factors may cause firms to use derivatives. We see from other similar questions that some argue that it is okay and other who argue that it is not okay. We would really apprectiate someones help on this or if someone could refer us to litterature where it describes when it is okay to only include the interaction term.