# Interpreting interaction effect between two dummy variables

I would like to double-check with you guys whether i am interpreting my results correctly, before I move on with my analysis. I have read several other questions on this topic but non seemed to apply enough.

I am investigating whether having a reservation in a restaurant (significantly) results in more spending per person. I have included several (dummy) variables in my regression analysis, as well as some continuous. I have included interaction effects between having a reservation and the other variables too, as reservations are my main interest.

dummy variables of interest for this question: "Reservation " --> Walk-in = 0, Reservation = 1 "Dinner" --> Lunch = 0, Dinner = 1

As you can see from my results, the interaction effect between having a reservation and eating dinner is significant, but having a reservation on its own is not. Is it correct if I interpret this as:

"Having a reservation only has significant influence on spending if the reservation is made for dinner (and thus not for lunch)."

Thanks!

• Relevant: What is effect coding? – Alexis Mar 15 '18 at 3:57
• In your syntax the dependent variable is called Log. Is your outcome a log-transformed variable? – Jay Schyler Raadt Oct 5 '18 at 22:04

I would look at what this is doing to your predictions! Since Reservation and Dinner are both dummy variables, the coefficient for Reservation:Dinner will only subtract from your predicted value when both variables have a "1". So the interpretation here would be: "For groups that have a reservation (specifically) for dinner, we expect the amount spent per person to be \$.0125 less on average than groups that did not have a reservation (inclusive) or were eating lunch"

Try generating a model where you only fit for Dinner and exclude Reservation, then run an anova - are the models significantly different? Without seeing your data I suspect the "effect" of Reservation + Dinner you're seeing is just the effect of Dinner. The interaction term has a lower p-value, higher standard error, and slope coefficient closer to zero than Dinner by itself, suggesting the interaction merely dampens the effect of Dinner.