Subjective theories on learning success in adult education
Supporting document for thesis. Summary of Klafki's multiparadigmatic approach to educational science (1998). (in German language)
My 28-item questionnaire (Likert scale 1 - 5) consists of five constructs. For one particular construct (9 items) I get an alpha of 0.687, with a sample of n = 301. When I select subsets of cases on one specific explanatory variable I get alpha: 0.673 (n= 41), alpha: 0.646 (n= 121), alpha: 0.787 (n= 49), alpha: 0.682 (n= 90). This shows me that the explanatory variable I use to select subsets does have an impact on reliability, i.e., data from the respondents in the group of n=49 has a higher reliability than data from other groups. As it is generally assumed that alpha increases with sample size I wanted to find out the alpha of a smaller, randomly extracted, sample to compare with the above subsets. I randomly extracted five times 100 cases of the original sample size of n= 301, and received the following alphas: 0.626; 0.595; 0.685; 0.711; 0.677. I am happy with the lower alphas, as this is what I expected. But how can I get 0.711 from n = 100, compared with 0.687 for the total sample of n = 301? There are a few demographic items (such as gender, age) that may have an explanatory effect. But when I select subsets of cases on them alpha hardly varies, certainly not above 0.7. I am using SPSS 24 for the calculations. Is it possible that any other effect exists to explain the alpha of 0.711? Am I wrong in assuming that alpha varies with sample size? Many thanks in advance for your thoughts.