Control variables for interaction effects.

[Jun Park]

Hello! I'm new to econometrics and this might be a dumb question.. but if I were to include an interaction term, do I also have to input the variables separately as a control variable?
For example, Y = a + beta1 + beta2 + beta1*beta2. This contains all of the variables separately as a control. Is this always the right way to go? I used to think it is, but sometimes, I get confused because some papers have a model like Y = a + beta1 + beta1*beta2. Is this bad practice?
When would some one include all of the variables separately as controls and not for others?

[Rich Goldstein]

there are special situations where one can, or even must, leave out one of the terms but in general you want to include all terms that are part of an interaction; Stata makes it easy to do this via factor variable notation; see

h fvvarlist

[Maarten Buis]

In economics I suspect that the most common special case is a fixed effects model where on of the variables is a time constant variable. The interaction changes over time, so it can be estimated, but the main effect cannot. This is not a problem, as the time constant main effect is included in the fixed effects. So this is one such special case where you include an interaction without (explicitly) including all main effects.

[Jun Park]

Maarten Buis Thank you for your reply! I am used to fixed effects, but what do you mean by "the time constant main effect is included in the fixed effects?" Did you mean the time constant variable is a separate variable in a fixed effects model? or that because we're including fixed effects, the time trend gets wiped off?

[Maarten Buis]

Neither. I mean that the fixed effects capture all time constant variables.