Hi Thomas!

What is DK/NR?

You should try another type of analysis where you can include the sociodemographics simultaneously, e.g., ordinal or linear regresson. It depends on what level you have your dependent variable (radio exposure) on an ordinal (none, low, medium, high) or continuous level (hours?)

Best

Jk

Thomas,

Looking at your data I would probably ignore NR/DK in statistical analyses. But you should report the frequency or proportion of these kinds of answer, such as how you did in the tables.

* * *

As John-Kåre suggests, some of your data is ordinal in nature, and it would be better to treat it as such.

The savvy way would be to create a single ordinal model with Exposure as the dependent variable and Age, Marital, Education, Employment, IIncome, HIncome, and Location as independent variables.

If that's too much for you, you can still treat the ordinal variables as ordinal. For example, for the Age x Exposure table, you can treat Exposure as ordinal. If you then treat Age as ordinal, you can use the linear-by-linear test. If you treat Age as nominal, you can use a variant of the Cochran-Armitage test that works on more than two categories.

* * *

Whatever DK/NR represent it may not be advisable to ignore them particularly in the individual monthly net income and the household monthly net income groups where their cell counts are not all negligible.  Ignoring them shall definitely affect the results of this analysis.