SPSS QUESTION for smart pscyhology people ONLY?

Question

What would we need to do (procedurally with SPSS) if we wanted to estimate either test/retest reliability or alternate form reliability?

Answer 1

I'm a smart public administration person, so my response doesn't apply... but I'll give it anyway.

I'll start with test/retest reliability:

OK, so I'll start by assuming that you have all ready entered your data as two variables, first the test, then the retest.

I would do a Bivariate Correlation (which should be Analyze -> Correlate -> Bivariate).  Your variables should be your test and your retest.  Hit OK.

The output should display your correlation coefficient, the higher the more the test and retest correlate - the more reliable the test.  I believe the rule of thumb in the social sciences is to go by two-tenths for each degree of correlation.  0-0.2 is no/insignificant correlation, 0.2-0.4 is low correlation, 0.4-0.6 is moderate, 0.6-0.8 is good, 0.8-1.0 is high.

I think you would do alternate form reliability the same way, the only difference would be that the retest was different in structure than the initial test.  If I'm wrong, inform me and I will consult with my mentor.

Answer 2

SPSS has several models of testing the different forms of reliability associated with a scale.  If you wanted an estimate of alternate forms (coefficient of equivalence), go to analyze - then scale - the reliability analysis.  Choose the model Alternate forms.  You can also tick off any descriptives or intraclass correlations if you hit the Statistics button.  

As for test-retest (coefficient of stability), it is the same scale (just administered at a different time), so you can do the bivariate correlation as stated above.

reliability = .90 excellent, .80 good, below .60, questionable (possibility of multidimensionality).