Need a clarification regarding the Difference in Difference analysis

[Fathima Sirasa]

Hi, I am Fathima, a PhD candidate, doing my last minute data analysis of my research project.
I did an intervention to preschool children to improve their healthy food knowledge and preferences.
My study included three study groups (control, intervention type 1 and intervention type 2).
The baseline characteristics of my study groups differ by ethnicity, caregiver education level and caregiver income levels (each has 3 categories).
But the baseline and post assessment values of outcome variables (means were compared using KruskallWallis test) did not differ between the categories of all three covariates mentioned above.
This is a 6 week intervention, so these covariates are time invariant, hence not sure do I need to consider them for my model.

I found codes for DID analysis with as well as without covariates. But those with dichotomous covariates or continuous covariates.
But for covariates, do I need to make dummy variables? I am not sure. Searched a lot, can't find.

Please advice.

Thank you


Fathima

[Clyde Schechter]

Concerning your covariates, if they are time-independent and you are running a fixed-effects estimation, there is no point in including them in the analysis: they will be colinear with the fixed effects and will be omitted anyway.

If you are doing some other kind of analysis, then you need to enter them using factor variable notation. (Read -help fvvarlist-). If you just enter them in a regression variable list without the appropriate i. prefix, then they will be treated in the analysis as if they were continuous variables. Now, for caregiver education and income, which are, in any case, ordinal variables, this might not be terrible. But for ethnicity it would be a serious mis-specification. So learn about factor-variable notation and use that.

[Fathima Sirasa]

Hi Clyde,

Thank you so much for the reply.

My study has three groups and I am trying to find the impact of intervention using difference-in-difference model. This is a new statistical method for me, as I used to work with SPSS mostly.
Dependent variable is a score. Data is not normal.
I think, I need to learn about factor variable notation and use it.
Please correct me if I am wrong.

Thank you

Fathima

[Clyde Schechter]

Don't worry about non-normality. It seldom really matters--just in very small samples. Focus on a model that fits the data well.

Yes, you need to learn about factor variable notation and use it. It's not difficult, and the explanation at -help fvvarlist- is clearly written.