[statistica] hierarchical reggression


I conducted a research in psychology, about the link between episodic memory abilities and other cognitive functions (language, executive functions, workin g memory). I would like to conduct a hierarchical linear regression in order to assess the specific contirbution of each factor, based on theorical model.
Thus, I want to enter one of my variable (e.g. working memory) asa predictors of episodic memorsy score at Step 1. Then, I want to intriduce another variable (e.g. language) to the model at Step 2, and then again with another varaible for a third step (e.g. exectuive functions).
I often use statistica to realize stepwise regression, but I don’t know how to conduct hierarchical regression with Statistica.
Could someone help me ? I’ll be more than happy ! (and the reviewer of my research paper too !)
Thanks a lot !



New Member
To perform a hierarchical regression, you need to compute a simple linear regression analysis with all of the variables in the model and specify sequential (type 1) sums of squares. All statistical software programs do this.


  1. Go to Statistics and run General Linear Models (GLM).
  2. Keep the default analysis type and click the "OK" button.
  3. Quick tab: Click the "Variables" button to select variables. Then click the "Between effects" button. On the "GLM Between Effects" dialog, select the "Use custom effects" option. You can now select variables and order them in the "Effects in between design" group box.
  4. Options tab: Select "Type 1" for sums of squares.
  5. Click the "OK" button to perform the analysis.
Thanks a lot for your answer Jen !! :)

I tried to analyze my data with the method you suggested but I have a problem ..

Usually, when conducting regression analyses, I get information that are essential for my interpretation : the r (and r squared) that inform me about the amount of information shared by my dependant variable and each of my predictor.
With the method you suggest, I know if each variable contribute significantly above the other variables previously entered in the model. But, can I have complementary information, to know the amount of variance specifically due to the variable that contribute significantly ?

Thanks again for your answer :)
and hope you'll have a solution for me :D


New Member
You're welcome :)

On the GLM Results dialog, click the "Whole model R" button for r and r-squared values.

Hope this helps...
I'm sorry but I have another question ... :) (the last, I promise !)

I can now analyse my data with the method you suggest (thanks again!). Nonetheless, if possible, I would like to enter some of my variables together; I mean that I don't want to introduce my variables in a strictly sequential way because some of them are underlined by very close processes (so I would like to introduce them as a whole).

For example, I would like to predict one score (e.g intelligence abilities), by introducing simultaneously two variables at Step 1. (memory score in a test A, memory score in a test B)
and then, by introducing three others variables at step 2 (e.g. language score in a test C, in a test D, and in a test E).

you see??

(it would be so pleasant and useful if we can have trainers like you when we go to Statistica's training!)