1
$\begingroup$

I am having a hard time deciding which modeling approach I should take. I have a survey data from a random sample from New York City. I want to explore the effect of NYC's number of crime incidence (city level) on people's perception of safety (individual level). I have data from 2 time points (year 2014, 2018). Here I wonder if I should either cluster standard errors or take multilevel modeling approach to deal with the clustered nature of the dataset (individuals clustered in NYC) when I have only one group at the group level (I have only NYC data).

Or maybe I should just do a typical pooled-OLS?

Please help.

Improve this question
$\endgroup$

1 Answer 1

Reset to default
1
$\begingroup$

Because you've already conditioned on group membership (i.e., by only looking at NYC residents), you don't need to further adjust for this clustering. Conditional on group membership, there is no induced correlation by group. That said, there is very likely clustering at the neighborhood level that would induce correlation, so that should be addressed (ideally using a model that explains the correlation rather than just with MLM or robust SEs). You also need to account for the fact that you are measuring the same individual at two time points. There are a few ways to do this but computing difference scores and modeling them is a good way.

Improve this answer
$\endgroup$
2
  • $\begingroup$ 1. so, you mean I am confining my "population" to NYC, right (intuitively it sounds like it)? 2. I don't have individuals' neighborhood data so that is not possible with my data set. 3. also it is a repeated cross-sectional, not panel so I guess I should just run pooled OLS. $\endgroup$
    – Kang Inkyu
    Sep 19, 2019 at 19:18
  • 1
    $\begingroup$ 1. Yes, that's what your description made me think. 2. Okay, I just didn't want to say you didn't need clustering at all when you actually might at another level you didn't think about. If neighborhood is relevant but you didn't collect it, you could run into some problems. You would have to justify why that wouldn't be an issue if you don't think it is. 3. Yes, regular OLS would be fine I suppose. $\endgroup$
    – Noah
    Sep 19, 2019 at 19:58

Your Answer

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

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