Search code examples
regressioninteractionstata

Stata: Adding an interaction term to a regression

I'm currently trying to fit a linear regression in Stata as follows: xi:reg Dependent IV_Rating IV_Size

I aim to see if the impact (coefficient) of IV_Rating on the dependent variable is significantly different for the small size, compared to the large size (both derived from IV_Size). I've made two dummies (0= false and 1= true) for IV_Size called Small and Large (medium is excluded). I've ran the following: xi:reg Dependent IV_Rating IV_Size IV_Rating#Small IV_Rating#Large

Error: factor variables may not contain negative values

I found the following fix: I add c. (continuous variable) although the IV_Rating variable only goes from -3 to +3. xi:reg Dependent IV_Rating IV_Size c.IV_Rating#Small c.IV_Rating#Large

The P values for both interactions are not significant, which is as expected (both IVs themselves are still significant). But I also read you can use ## instead of #, and I'm starting to get really confused, among other things. Am I doing this right?

Solution

  • Just change the coding to positive integers (perhaps using recode):

    x   x2
-3  1
-2  2
-1  3
1   4
2   5
3   6

    Also, you can use factor variable notation directly (instead of xi):

    reg y i.x2##i.z

    This will include main effects for the two categorical variables as well as their interaction.