I have counts of people with a condition, by age (5-year groups), sex, and area (100-200 areas).

For each area, I have:

  • estimated population by age (5-year groups) and sex,
  • estimated population percent living in city, small town or rural (sum to 100%),
  • estimated population in high, middle and low socioeconomic status (sum to 100%), and
  • estimated population proportion treated (via food or water supply fortification).

I know nothing about individual-level correlation between rurality and socioeconomic status (or even if it's the same across areas). I expect proportion treated is near 100% in cities, lower in small towns, much lower in rural areas.

Researcher wants adjusted rate (or odds) ratios for having the condition (and p-values) for rural vs cities, low vs high SES, untreated vs treated, young vs older.

Can I produce valid estimates?

If so, how should I model it?


1 Answer 1


Multilevel analysis based on aggregate data leads to ecological fallacy. Here is an interesting paper to read: http://www.stanford.edu/class/ed260/freedman549.pdf

  • 3
    $\begingroup$ Could you perhaps summarize the main points of that paper so that this reply remains self-contained. $\endgroup$
    – chl
    Jul 17, 2012 at 9:56
  • $\begingroup$ @Omar So are you saying no I cannot produce valid estimates? Or are you saying I cannot produce valid estimates from multilevel analysis? Is there any way i can produce valid estimates? $\endgroup$
    – timbp
    May 25, 2018 at 13:10

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.