Stepwise and Hierarchical regression - can these 2 be combined?

#1
Hi all,

I have run both hierarchical and stepwise methods on my data with 34 IVs (categorized under 7 groups of variables), using SPSS. The hierarchical regression allows these 7 groups of variables to be entered in the order that I want, and the SPSS output shows a coefficient table which presents all 34 IVs regardless of whether they are statistically significant or not. whereas the final model output of a stepwise regression presents only the IVs that are statistically significant. However, as stepwise regression enters the 34 IVs individually based on their significance, I am unable to obtain the R2 changes for the addition of each group of variables.

In other words, I would like to run a hierarchical regression in which the R2 changes are presented for the sequential addition of the 7 groups of variables (rather than each of the 34 individual variables), but at the same time I would hope to have a final model output which shows only the statistically significant IVs (like what we can get in the stepwise regression output).

I am wondering if there is a way to ask SPSS to omit the insignificant variables from the final hierarchical model output? I suppose I should not manually omit the insignificant variables on my own because removing variables can cause changes in the coefficients of other variables remained in the model.

Many thanks in advance and any help is very much appreciated,
Leanne
 

noetsi

Fortran must die
#2
When I ran hiearchical regression in SPSS last, it showed the changed in R square not for individual variables but for the group of variables you entered in a given block. Are you saying this is not the case any more?

You can always run all the variables and manually copy and paste those that are statistically significant yourself (which is common enough anyway since SPSS is not going to generate the APA style often demanded by academics). I do not believe that SPSS has an option to automatically exclude variables above or below a given signficance level.

Generally speaking publications print variables that are and are not statistically signficant at a given level with asteriks showing which are significiant. Of course it is possible that individuals removed certain variables that were not significant and reran their model, but I believe that is frowned on (if probably done anyway).
 
#3
Hi noetsi,

Many thanks for your reply! Yes, hierarchical still shows changes in R2 for groups of variables. I was referring to the stepwise method when I said changes in R2 are shown for individual variables.

I initially wanted to exclude the insignificant variables from the output because I thought listing out all 34 variables in my dissertation (including even the insignificant ones) would be too exhaustive and confusing for readers to read and interpret. However, considering SPSS has no option to exclude insignificant values I will take your advice and use asterisks.

Again, thank you so much!!! really appreciate it. :)
 

noetsi

Fortran must die
#4
Personally I would not use Stepwise regardless unless you committee requested it. There are many who are critical of it and to me their criticisms make sense.

For a dissertation unlike a journal article printing every variable should not be an issue. I assume you brought them all up in your introduction or literature anyway (although it has been a while since I did one so that may no longer be the case). You committee in any case are the right ones to ask this (or your supervisory professor).

Good luck on the dissertation.