I have a set of traits that are measured twice: the first result is just a percentile (between 0 and 1) while the second result is a raw score (between 1 and 7) with mean and SD. Thus, i can covert the second measurement into percentiles, too. Now i want to know if the first measurement differs significantly from the second - if it is a predictor for the second measurement. Can i use a paired t-test (normal distribution assumed) or the Wilcoxon Rank test (non-parametric) or isn't it allowed to compare percentile scores that way?
Thanks!
Since your second variable can be assumed to be categorical, I assume that you may use one-way ANOVA to test for equality of means (percentages in your case) for the categories 1 to 7.
Dear Stamatis,
thank you for the answer. Maybe i was not clear enough: The first variable is a quantile, the second is normally treated as continuous. So i don't want to conduct an ANOVA on ONE variable (there are no categories) - i want to compare the first and the second assessment variable. Thus, both have to be the samr scale: either percentiles or z-scores.
Hello Stefan Poier
For proportional data (0;1) as response variable you can use beta-regression (
https://rcompanion.org/handbook/J_02.html
https://cran.r-project.org/web/packages/betareg/vignettes/betareg.pdf
).
The paired t-test analog in this case is beta-regression with random effect look like:
fit
- Badges
- Science topic