Hierarchical regression building support

#1
Hello all,

I am building some regression models to look at the contribution effect of stigma on various pain outcomes. I have a large sample size (N=974). I've done my normality and collinearity testing as recommended. I have two demographic factors to include in block 1 (age and duration of pain) and stigma will be entered in block 3 as my main variable of interest. As well as pain functioning, my IV's are depression, pain intensity and loneliness.

So an example, to look at the effect on pain functioning (one of the IV's):
Block 1-demographic factors
Block 2- pain intensity (another IV and known variable related to pain functioning and stigma)
Block 3- stigma

That is a simple model recommended by my supervisor. My question is, when I look at a different IV instead of pain functioning e.g. loneliness, How would block 2 look then? Would I include all the other IV's in block 2? And if so, how would I be able to tell their unique contributions?

Many thanks
 

hlsmith

Less is more. Stay pure. Stay poor.
#2
By block do you mean level? For clarity we are talking about data clustered in units, right?
 

Karabiner

TS Contributor
#5
My question is, when I look at a different IV instead of pain functioning e.g. loneliness, How would block 2 look then? Would I include all the other IV's in block 2?

Personally, I am not quite sure what you want us to tell you. Whether you investigate one
variable at a time, or all variables simultaneously instead, depends on your very own research
question(s).

With kind regards

Karabiner
 
#6
My question is, when I look at a different IV instead of pain functioning e.g. loneliness, How would block 2 look then? Would I include all the other IV's in block 2?

Personally, I am not quite sure what you want us to tell you. Whether you investigate one
variable at a time, or all variables simultaneously instead, depends on your very own research
question(s).

With kind regards

Karabiner
My main research question is whether stigma contributes anything unique to the relationship between pain functioning and related variables (intensity, loneliness and depression).
 

noetsi

No cake for spunky
#7
Unfortunately the word block is used different ways and hierarchy. To some this means multilevel models, where one effect is nested inside another. But SPSS uses this to mean adding variables at different points in the analysis. First you add some, then you add more, and you run a test if the newly added variables add anything to the model. Multilevel models have their own version of this, but I am not sure these tests relate to the SPSS version which is not dealing with the concept of nesting as best I can tell.

Adding variables at various stages in the way SPSS does, which I have rarely seen outside classes, should be based on theory. I would, in the absence of theory, just add all the variables you think are important at one time. I think this is the norm in analysis. Regression inherently tells you if there is an independent effect (unless you have reason to suspect interaction or moderation effects).