Sam Stroope writes:
I’m creating county-level averages based on individual-level respondents. My question is, how few respondents are reasonable to use when calculating the average by county? My end model will be a county-level (only) SEM model.
My reply: Any number of respondents should work. If you have very few respondents, you should just end up with large standard errors which will propagate through your analysis.
P.S. I must have deleted my original reply by accident so I reconstructed something above.
Was that an intentional "blank" reply? Why county-level "only" and not multilevel modelling … ahh. the SEM thing. Using a mean looses the information about the variability within county. I guess the more the merrier. That or it takes two to make an average.
Is this a test, where we fill in the blank with your answer? I'm going to guess "Mr. P".
Judging from your reply, you believe that the correct answer to the question is "None."
(Let me explain: nothing at all follows the words, "My reply:")
…if the selected sample of respondents is non-random, then select the entire population as a minimum.
I understand why Andrew gave that reply but I'd imagine that the SEM estimates would be useless if there were so few individuals per county as to make the standard errors very big.
Also, we're assuming he has some way of incorporating the SE of the county-level estimates into his SEM instead of plugging in point estimates. When he says he has a "county level (only)" SEM model, maybe not?
Related to the above I am facing a problem which i hope someone here can answer. I am trying to regress trips by purpose against employment and population. The problem i am having is that the trip information is at the person level and the employment and population information is at traffic analysis zone (TAZ) level (similar to a census tract). Another issue is that the trips by purpose are from sample NHTS data and i have weights at the person level but the analysis needs to be done at the TAZ level. So my question is – how do i account for the weights? As far i know i cannot simply aggregate the person weights to the TAZ level and then do the regression. Any inputs from the group here would be much appreciated. I was told that it might be better to finish the estimation then develop the zonal weights but I am at a loss of how this done and my coefficients that i am getting are very small (10e-5 etc.). It is a simple linear regression model.
My data looks like this (just a sample)
TAZ
NumShopTrips
Population
TotalEmp
101
24
2446
243
102
28
3446
2433
It is possible to do a multilevel SEM, so you don't need to aggregate to the county level, you can keep the raw data, and the information in the raw data. I use Mplus for this sort of thing.
KrishnanV: I think the usual approach for this sort of data would be to just do the analysis at the person level. Why do you need to do the anlysis at the TAZ level? What is your research question?