SERforum

 

Join Andrew Olshan, for SERforum Live!

“Career and Professional Development”

JOIN SERforum – Live
December 5th, 12pm EST

 

Do you ever find yourself struggling to figure out a question about epidemiologic methods, or other topics in epidemiology, and don’t know who to ask? The SERforum allows for individuals to answer questions that come up in our daily work around substantive and methodological topics in epidemiology.
 
All topics may be viewed, but to read and post comments, SER membership is required. If you are member, login! Not a member, join us!

You need to log in to create posts and topics.

Semi-ecologic study biases

I'm looking for a study (simulation or using real data) that highlights bias possibly introduced when using a semi-ecologic study design particularly when an ecologic or group-level variable is used for the exposure and an individual-level variable is used for the outcome. Are there any papers you could refer me to?

Reply from: Stephen Mooney

Hi Nelson,

 

First, thanks for the question.  This is an important topic, particularly as we analyze policy interventions.

 

Second, I’ve never heard of this design referred to as semi-ecologic.  That may just be my ignorance, but I’d encourage you to use “multi-level” when searching on PubMed, Google Scholar, etc.

 

Now to the more direct answer: So Ana Diez Roux wrote two great overview papers on how to think about this type of analysis back in the late 1990s:

 

Diez-Roux, A.V., 1998. Bringing context back into epidemiology: variables and fallacies in multilevel analysis. American journal of public health, 88(2), pp.216-222.

 

Diez-Roux, Ana V. "Multilevel analysis in public health research." Annual review of public health 21, no. 1 (2000): 171-192.

 

For a more statistical simulation of how size and number of ecologic groups might affect results, I like this paper a lot:

 

Theall, Katherine P., Richard Scribner, Stephanie Broyles, Qingzhao Yu, Jigar Chotalia, Neal Simonsen, Matthias Schonlau, and Bradley P. Carlin. "Impact of small group size on neighbourhood influences in multilevel models." Journal of Epidemiology & Community Health (2010): jech-2009.

 

Finally and self-promotingly, the first methods paper I ever wrote was on the bias that can arise due to using group-level measures that were aggregated from individual measures with measurement error (like % of residents living in poverty). That might be the kind of thing you’d want to think about, depending on your scenario:

 

Mooney, S.J., Richards, C.A. and Rundle, A.G., 2014. There Goes the Neighborhood Effect: Bias Due to Non-Differential Measurement Error in the Construction of Neighborhood Contextual Measures. Epidemiology (Cambridge, Mass.), 25(4), p.528.)

 

More generally, though, there’s a rich literature on multi-level studies and risks of biases that can arise in them.  I’d encourage you to read the papers cited above for background, then think about what you’re worried about in your specific scenario and dive in from there.

 

Good luck!

 

Best,
Steve