1
$\begingroup$

I use publicly available EU-Silc data to estimate the market price of social dwellings (subsidized dwellings). However my X variables are almost perfectly available, my Y variable rent is often missing.

Given observed Y and X variables I have very poor fit. I assume that if I would have completed or almost completed Y variables I would have a better fit. Hence I did some imputation for Y's.

Is there any reason that I should not do that?

Improve this question
$\endgroup$

1 Answer 1

Reset to default
2
$\begingroup$

Yes, there are reasons to not do that. Consider three variables $x$, $y$, and $z$. If $y$ is missing, then $x$ and $z$ are used to impute $y$. Then, if a regression equation like $y=x+x$ is used, then of course there will be great fit. I will search for the source, but one method is to do the imputation, but then delete cases where the outcome was imputed.

Edit: A method known as Multiple Imputation with Deletion is recommended by von Hippel (https://onlinelibrary.wiley.com/doi/epdf/10.1111/j.1467-9531.2007.00180.x), but it has its detractors (https://modeling.uconn.edu/wp-content/uploads/sites/1188/2016/05/Don%E2%80%99t-be-Fancy.-Impute-Your-Dependent-Variables.pdf).

Improve this answer
$\endgroup$
1
  • 1
    $\begingroup$ Thank you for your answer @Jay. The links you are referring actually providing conflicting results. But they I enjoyed to read. $\endgroup$
    – Econ_matrix
    Commented Oct 31, 2018 at 22:43

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

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