In my data the dependent variable ‘vocabulary uptake’ is measured as scores ranging from 0-10 and the independent variable is yielded from a ‘reading intervention’ with 4 different reading conditions (glossed, no glosses, etc.). There are about 80 participants in each of the four groups. The data is not normally distributed (I tried to transform the data). Given the fact that there is a natural limit (0), this is probably not surprising. I want to find out what the best way is to determine whether these groups are different or not.
I am sure that the independent is a ‘nominal’ variable, but I am in doubt about the dependent – I cannot decide whether to treat it as ‘ratio’ data or as ‘ordinal’. The latter would make sense, as the scores can be ranked, but ‘discrete ratio’ makes sense as the values are measured, have numerical value, the difference between, e.g. 1 and 2 is the same as between 3 and 4, and a score of 0 means that there is none of that variable. ‘Ordinal’ data would be easy, but if I categorize the data as ‘discrete ratio’, my problem is that – to my knowledge – the standard nonparametric tests only work with continuous data. What can I do?
What test + post-hoc test could I use to find out if there are significant differences between the groups? When working with a nominal dep. and a discrete ratio data indep. variable I would violate the assumptions for most nonparametric standard tests, which assume dichotomous independents and continuous/ordinal dep. variables (Mann-Whitney U, Wilcoxon) or categorical data for the independent variable and continuous/ordinal data for the independent variable (Kruskal-Wallis, Friedman’s test). Also there is a problem with tied values.
I am working with SPSS.
Thanks for any suggestions! Your help would be GREATLY appreciated!