Difference-in-Differences and Gender: Is Triple Differences the Answer?

[Ricardo Vargas]

I know that this is more an econometric question than a STATA question but i think some practicioners here can help me

I'm currently conducting an impact evaluation using panel data and employing the difference-in-differences (DD) method. My primary objective is to ascertain if there are differentiated effects between males and females. My intuition suggests change from a DD model to a triple difference (DDD) model to capture these effects.

Is this a valid approach? I've been searching for literature that explicitly discusses this aprroach, but I've found limited resources and literature about that. If anyone has experience with this or can point me to relevant literature , I would greatly appreciate it.

[Maxence Morlet]

To be honest I think we would not a lot more details to answer this question sensibly. What is your research question? The outcome? The exogenous variation you are using? Are there systematic trends in the outcome? These are just some of the relevant questions

[Ricardo Vargas]

I'm studying the impact of an agricultural training program on the implementation of good agricultural practices. The intervention was conducted at the village level, with some villages receiving the training (treated) and others not (control). My data, however, is at the farmer level, allowing me to observe individual farmer practices both before and after the intervention.

To address potential concerns about parallel trends, I've tested the parallel trend assumption using administrative data and found it to be satisfactory.

For my primary DD impact, I've estimated the effect using the following Stata command:

regress practice1 treatment1##t control_variable, vce(cluster village)

Where practice1 is the outcome variable
treatment1 is the treatment variable
t is the time variable

To explore heterogeneous effects by gender, I'm considering the following specification:

regress practice1 treatment1##t##gender control_variable, vce(cluster village)

Where gender is a dummy (1 male, 0 female)


I hope this provides a clearer picture of my research context and approach. I'd appreciate any insights or suggestions on the feasibility and appropriateness of this approach for capturing gender-specific effects.

[Mike McDonald]

Have you come up with something? I have a similar issue. Thanks!