1

My thesis advisor recommended me to model the monthly electricity consumption of households as a function of their state and multiple other regressors. In R, I turned the state variable into a factor and added it to the regression.

df$state <- df$state %>% as.factor
M1 <- lm(kwh ~ state + log(income) + household_members, data=df)

Is this what's known as a fixed effects model?

I'm confused because I thought fixed effects could only be applied to panel data when you subtract $y_{t+1}^i - y_t^i$ in order to get rid of them.

I am also hesitant about how to interpret each coefficient. If I'm not mistaken, the nth coefficient is the additional expected consumption of a household who lives in the nth state compared to the base level (the state when the factor level is equal to one). Is this correct?

Arturo Sbr
  • 305
  • 1
  • 12

1 Answers1

4

Yes, this is a fixed effects model, but categorical variables can be modelled as either fixed or random effects.

There's nothing about fixed effects that is specific to panel data - any model in which you do not treat any variable as a random effect is a fixed effects model (the distinction explained in much more detail here). In your case, I would advise modelling state with a random effect, though, to take advantage of the benefits of partial pooling.

Your interpretation of the coefficients is basically correct, except that you should treat the reference state (or 'base level') as having a coefficient of zero, not one.

mkt
  • 11,770
  • 9
  • 51
  • 125