Entering interaction terms

[aosborn]

I am new to this site, but have high hopes that I can find some relief.
I have a question regarding entering interaction terms (moderation) in SPSS.
Is there a conceptual difference between entering interaction terms and main effects via direct entry (at the same time) vs hierarchically (in blocks)? I am thinking about this in terms of regression models (specifically logistic regression).
I will be entering two 3-way interactions and their respective lower ordered terms (I have a large enough sample size).

I can only assume that this may impact how the variance is explained, although I can't seem to wrap my head around the distinction.

If i go the route of Multinomial Logistic Regression, it is my assumption the only way to enter all of my interaction terms (and lower ordered terms) is by entering them all at the same time. Is this inaccurate?

Any help would be most appreciated! Also, any references or citations you could provide would also be very helpful!

 

[noetsi]

Well one thing is that it is generally accepted that you must enter the main effects if you enter the interaction term. When I have seen the terms entered in blocks it was either because you were using a F change test to decide whether to add additional terms, whether they add explantory value, or based on theory. Obviously the first makes little sense if you have to enter all these terms anyhow. I have not seen any theory tied to adding main effects before interaction terms, or the other way around, and would do so only if there is something in the pertinant substantive literature that says you should (which I doubt there will be).

As an unrelated aside don't call entering in blocks "hiearchical regression." That is a uniquely SPSS terminology and it will confuse posters here (who sometimes use this for multilevel analysis an entirely different concept).

 

[aosborn]

Yes, that part I am at least familiar with. That in order to enter a 3-way interaction (or any interaction) that all lower ordered terms must be included in the model (all possible two ways and their main effects) at least for continuous variables. But thank you, knowing that someone hasn't come across any information related to order of entering is still very useful!

 

[noetsi]

The one thing is certain, because this is inherent in the logic of blocks, is that there has to be a theoreticaL reason to enter in a specific order. So if you have none you don't do this

For example..

Hierarchical: Variables entered in blocks.
Blocks should be based on past research, or theory being tested.

http://www.google.com/url?sa=t&rct=...rYGIBA&usg=AFQjCNGuZbh44iYVSoG_feQnp_czJ9vBow

 

[hlsmith]

Blocks should be based on past research, or theory being tested.

I agree with this, and have not seen any rules or protocols for entry.

 

[noetsi]

I think hlsmith this is because the system was designed specifically for testing theory. There is no real reason to create protocol because the user is assumed to be entering the variables in the order dictated by the theory they are testing. Unfortunately SPSS, and SAS which does not emphasize this option to regression as much as SPSS does, never really explains this and (based on my own experience) few writers in the social sciences actually enter variables this way. Or if they do they don't discuss it in detail.

I found a few papers where variables with interaction were added in blocks. In all these cases they entered the main effects first. But I don't have anyway to tell if this small sample size meant anything. There may not be formal rules for this.

 

[hlsmith]

It is an art form!