I've got a serious problem when trying to analyse the data of my psychological experiment. Here are the details of the experiment:
In each trial, subjects were presented with lateralized colored circles. They had to press a right button when the circle was red, or a left button when the circle was blue. The compatibility condition determines if the location of the circle and the side of the response correspond (compatible) or not (incompatible). The saturation of the colors were also manipulated (6 saturation levels: 15, 25, 35, 45, 60, 80%).
So I used a 2 (compatibility condition: compatible, incompatible) x 6 (color saturation level: 15, 25, 35, 45, 60, 80%) within-subjects design. 13 subjects participated in the experiment. For each subject, there were 224 observations per experimental condition. I would like to analyse the relation between the mean and variance of each experimental condition for each individual. Classicaly, the relation is linear, leading to very high correlations for each subject (r > .85).
However, in my experiment, the r values range from 0.15 to 0.95 with a mean of 0.56. The majority of subjects shows an r < .75 (11 out of 13 subjects), which is far from the traditional finding (> .85). So I decided to switch to an other analysis. I think that the data is better described by two regression lines: one for each compatibility condition (but this is just a graphical feeling). My guess is to perform an ANCOVA for each subject with std as the dependant variable, mean as covariate and compatibility as factor. But I don't know if it is the right analysis, and if it could be considered as a within-subjects ANCOVA.
Any help would be appreciated!