The main variable I'm investigating is a dummy variable that is 1 after a certain date and 0 before, denoting when a certain law passed. When I include date, date is significant and the dummy is not, indicating the change in response after the dummy date can explained by the existing time trend. But when I take out the date variable, the dummy variable is significant.
There is no obvious reason that date/time should be having an effect.
Should I keep date in the model?