Hello. I am working on a thesis in public health science, and am stuck until this current problem i solved. Your help will be highly appreciated!

I want to find out what the effect of pollution level is on cardiovascular disease (CVD). However this can be confounded by type of building one lives in (potentially as a measure of socio-economic status). I want to controll for this confounder but the data reveals a potential problem:

The problem I have come across is that one category of the confounder (type of building) does not have observations in the highest two exposure categories. Specifically, one type of building (villas) are not at all exposed to the highest two pollution levels.

Is this a problem? Or can I worryless control for the potential confounder? The results look fine in the output of SPSS, but are they really?

I suspect that stratification by the confounding variable can solve the problem, but I cannot do that as it would compromise my statistical power.

Schematic presentation of problem:

Outcome: Cardiovascular disease (CVD) (dichotomous coded 0,1)

Exposure: Air pollution level (ordinal level,categorized in 5 (1,2,3,4,5)

Confounder: Type of building (three categories: villa, high-rise apartment, low-rise apartment).

(Other confounders are included as well.)

Problem: The confounder does not have observations in exposure category 4 and 5. Is this really a problem? The results look fine, but are they really?



I have searched the endless web and statistical text books without finding a solution so far.

It would be a great relief to have any opnions on this matter.

If none have the right answer at first, inputs will still hopefully bring us to a sollution.

Thank you!