Consolidation of items for regression analysis & inter-construct correlation analysis

Hello everyone!

I am a stats newbie and have a very basic question. Please forgive me if the question may seem too foolish.

I wanted to conduct a regression analysis between the 2 variables, USE (dependent) and ATTITUDE (independent) in SPSS. I have measured USE and ATTITUDE with 4 items each (USE_1, USE_2, USE_3, & USE_4; ATT_1, ATT_2, ATT_3 & ATT_4). I have a sample of 25 respondents.

To proceed with regression analysis in SPSS I will have to input the two variables USE and ATTITUDE. So my query now. Should I take the mean of the 4 USE items to determine the final USE variable (to be used in regression) for each of the 25 respondents. And do the same for ATTITUDE?

And do I follow the same logic to generate the correlation matrix for all the variables in the model, which are measured using multiple items?

Re: Consolidation of items for regression analysis

The question you are asking is not as simple as it may seem for a beginner. Considering you have multiple items (i.e., USE_1, USE_2, USE_3, USE_4 and ATT_1, ATT_2, ATT_3, ATT_4) for each variable, which in this case are called latent variables, you have a couple options. The first one, would be to run the factor analysis on each set of items and then use the emerging factor as a single variable for the analysis. I.e., you would have a factor for USE and a factor for ATTITUDE, and then run a regression with these two variables. The second option would imply the usage of SEM (structural equation modeling), though I assume this would be slightly more complicated for you at this stage (besides I believe SPSS does not have it by default).

Even though factor analysis may seem to complicate things, in reality it would provide you a better picture of the validity (and reliability, too, if you add internal consistency analysis option -- Chronbach's alpha) of the items you are using to measure each latent variable.

Having said this, your "most convenient" approach would be: (1) Google for factor analysis with SPSS -- its relatively simple, (2) save the resulting factors --> there will be two new variables created in your data set, and (3) run the regression using these two factors (you can also run correlation the same way).

Note, taking a mean value of the items is also a palusible option, yet it is less methodologically robust, in my point of view.