Using different matching variable in different year.

[Sk Moniruzzaman]

Hello,

I am using Coarsened Exact Matching (CEM) to analyze my data, which consists of repeated cross-sectional surveys conducted annually from 2015 to 2021. For most years (2015-2017, 2020-2021), I used eight covariates for matching comparison between treated and control groups. However, for the years 2018 and 2019, I only used three covariates that are highly relevant to my outcome variable. Following the matching, I applied Ordinary Least Squares (OLS) regression using the CEM weights to estimate causal effects, as recommended by Matthew Blackwell and colleagues.

My question is: Is it methodologically sound to use a different set of covariates for matching in different years within the same study? Could you please recommend some journal articles or resources that address this issue?

Thank you in advance for your assistance

[George Ford]

Does it fall apart if you apply the same matching variables in all years?

[Sk Moniruzzaman]

Thank you George Ford for your quick response. It didn't fall apart in my data set but asking for my curiosity. What will happen if I didn't have the same covariates over the years?

[George Ford]

It smells fishy to me, and I suspect that it could cause bias/inconsistency since the selection process for treated/control is changing over time.

Personally, to avoid questions and explanations, I would find a consistent approach.

You might ask Gary King or some other real expert in this area.

[Sk Moniruzzaman]

Thank you George.