Dear Bethany Mcbride ,

Yes there is. It is a group of techniques called dimension reduction in SPSS. The technique I would recommend is called Principal Component Analysis. However, only having three variables for each of your constructs might be a problem as three is the minimum number you can combine together and sometimes your scales might not work or you need to remove variables from them.

If you save the regression scores it will also transform your data into scale variables opening up the possibility of Pearson correlation instead of Spearman provided that its assumptions are satisfied.

For more information please see my study guides on scale reliability with small samples and Pearson correlation.

Hello Bethany,

You could:

1. Sum or average the scores for the three related items into a new, "total," variable. Then evaluate correlations among the totals only.

2. Try common factor analysis or, if you can justify the more stringent requirements of principal component analysis (as suggested by Peter Samuels), you could use PCA to combine the individual questions into linear composites.

3. If you have some criterion variable that makes sense to use, you could use multiple regression (probably ordinal) to construct linear combinations of individual question scores that best account for differences on the criterion variable, then use those weightings for the related questions.

4. Use item response theory (IRT) models, such as graded response, to evaluate how well/poorly a given set of items/questions behaves to form a coherent scale.

What's best? I think it likely on your specific research question/s, and how you propose to use the subsequent composite/combined scores.

Good luck with your work.

Run a Factor Analytic check and see which item is explained most by the Research Question under bother the Rotated and Unrotated factor Matrix. But before then you may want to check initial rotation to see the fitness of the 3 items.

First, you need to determine whether combining your variables into a scale would have sufficient reliability (i.e., internal consistency across your measures). The SPSS procedure for this is coefficient alpha (Analyze > Scales > Reliability).

Then, if you have sufficient reliability (preferably .8 and above, but potentially .7), you could run a factor analysis to make sure that your results are uni-dimensional (i.e., all the variables load on a single factor) For this I recommend Principal Factors analysis rather than the default (Principal Components). As others have mentioned, three variables is the minimum requirement for factor analysis, so the result is essentially a "yes-no" question of whether all three variables do indeed load on one factor.

Bethany Mcbride, you have received some valuable advice above, though not all of it in line with each other. I'll try to assemble some thoughts in a sequence that might be immediately helpful for you.

First, if you have sufficient numbers of people in your sample, I recommend you use parallel analysis to identify the number of factors in your data.

Then, if you have more than two factors in your data, I recommend you use exploratory factor analysis rather than principal components analysis (despite the latter being the default option in SPSS), and that you probably use principal axis factoring as the method of extraction along with promax as the method of rotation if there are two or more factors in your data.

Some people might not agree with the above recommendations, and I'm quite happy for them to indicate their disagreement if so. If you need references to support what I have suggested, please let me know.

Doing the above should give you a sense of which items you could combine to form separate domains from your data. If you do have separate domains (and I agree that ideally each domain should have at least three items in it - and preferably more), you can combine the relevant items in SPSS by using the Transform / Compute procedure. There are some good YouTube sites demonstrating how that can be done. The following might get you started:

https://www.google.com/search?q=spss+create+new+variable+from+multiple+variables&rlz=1C1GCEA_enAU822AU822&oq=SPSS+crea&aqs=chrome.1.69i57j0l7.8372j0j7&sourceid=chrome&ie=UTF-8

I should add that, if you have as few as only three or four items on each of your dimensions, coefficient alpha values lower than .70 might well be associated with an adequate range of interitem correlations within your data.

I hope that's helpful - and all the best with your research.