0
$\begingroup$

I am conducting a diff-in-diff regression over two periods of time with the treatment in between. Next to the dummy variable time ('Post'), I have 3 groups for the other variable and therefore 2 dummy variables ('app', 'instagram'). I want to investigate the exposure to news media over different channels over two periods of time.

My question now is: - What is the adequate process to finding out the relevant covariates which I shall include as control variables?(e.g. age/gender/graduation/media use behavior etc)

  • And how do I include them methodically?

I tried proceeding with a crosstabulation of the covariates to the DV and looking at the eta values. I chose this procedure because I read it is better if the potential covariates are nominal. I built dummy variables for every expression, e.g. media use behavior often, media use behavior sometimes, media use behavior never). Is it right that I made binary dummy variables for a categorical variable like this (which I questioned via 5-point-likert scale? Do i just include the variables with an eta-value over 0.3 as covariates? (I thought this could be a good threshold correlation-wise)

Is there some analysis afterwards necessary (I read of Manova, but I am very unsure if sth like this is necessary)

I am sorry for the long text, but I am really looking for help. I am doing this for the first time ...

$\endgroup$
0
$\begingroup$
  1. The decision to include or exclude an explanatory variable should be based on theory. It is very important that you add variables that affect the outcome and treatment, otherwise, you will have omitted variable bias. Moreover, also add variables that only affect the outcome, as this can improve the precision of your model. Do not include variables based on descriptives only. Do not include variables that are themselves affected by the treatment.

  2. There is no general test that you can do afterward. If you have a theory that guides your inclusion and exclusion of certain variables, that theory is likely to have a number of implications that perhaps are testable.

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.