How can you model the interaction RB*age when RB is the dependent variable?

Why are "age", "adults" and "older adults" all variables in your model? That makes no sense. Either you use age as a continuous predictor or as a categorical predictor (ee.g. with the two levels "adults" and "older adults"). In either case there can only be a single coefficient for "age" in your model (I still don't understand how you will get the interaction with the response variable RB...).

Either you explained it misleadingly or you misspecified the model. Could you please write down the model formula? This might resolve some issues.

Dr Jochen Wilhelm,

thanks for your attention. I believe that I was not clear. I will add more details about my doubt.

I have two datasets from two countries: Country X (n = 450) and Country Y (450). I am interested in knowing if there is difference in religious beliefs (RB) and age between the two countries (X versus Y). RB is a numerical variable and age is categorical variable (adults 65 years).

Informations:

RB (dependent variable); Country (fixed factor); Age (covariate).

The findings were: Country*Age (F=2.850; p = .750); Country_X*age (B=.760; p = .150) and Country_Y*age (B=5.250; p=.008).

My doubt is: How to interpret these finds? For in the multivariate analysis Country*Age was not SS, however Country_Y*age was SS.

Thanks very much. 

Maybe I miss something, but I do not understand how you get Country*Age and Country_X*age and Country_Y*age in one model, I do not even understand what Country_X*age or Country_Y*age mean, and I am not aware of a B statistic. I can only guess that these are the effects of age for country X and Y, respectively.

You write: " I am interested in knowing if there is difference in religious beliefs (RB) and age between..." - Is it really that you want to test the age-difference in the two countries (i.e. is age a response variable)? Or is age a covariable used to adjust difference of RB between countries (this is what I think, as it seems plausible to assume that religious believes are age-dependent)?

If RB is the only response variable, you would usually define the model as

RB = b0 + b1*I(age) + b2*I(country) + b3*I(age)*I(country) + error

where I() is an indicator function returning either 0 or 1 depending on the argument, b0, is the intercept, and b1, b2, b3 are the coefficients of the two main effects and their interaction.

Give I(age=adluts) = 0 and I(country=X) = 0, the b's will have following meanings:

b0: expected RB for adults of country X,

b1: change in expected RB for older adults vs. adults in country X

b2: change in expected RB for country Y vs. X in adults

b3: change in expected RB for older adults vs. adults in country Y vs. X

In such a model one would first have a look at the interaction (b3). If this can not be neglected, then you would need to compare the RB between the two countries separately at each level of age (the difference in adults is b2, the difference in older adults is b2+b3).

If the interaction is negligible, the difference in RB between the countries can be analyzed in a simpler model that adjusts for age:

RB = b0 + b1*I(age) + b2*I(country) + error

This model still assumes that there can be an effect of age, but this effect is assumed identical in both countries. Here, b2 would give you the difference in RB between the countries, adjusted for age; and b1 is difference in RB between the age groups, adjusted for country.

NB: I would be careful considering p=0.008 SS for such a large sample! I strongly suggest to interpret the estimates (i.e. the values obtained for the b's) to see if the differences might be of any relevance.

Dr Jochen Wilhelm,

thank you very much for the explanation. It was very useful. 

All the best. Luciano·