Good evening,

I have a study with 19 participants, 9 of whom are experts and 10 are novices chess players. Each participant is asked to identify meaningful events in an ongoing chess game, with the number of identified events being recorded as a score. This task is performed twice under two different conditions: fine segmentation (where they try to identify as many events as possible) and coarse segmentation (where they only identify the main events). As a result, I have two scores for each segmentation task for each participant (thus 4 scores).

My variables are as follows:

Segmentation Score: A continuous variable ranging from 3 to 15.

Chess Rating: A continuous variable ranging from 600 to 2200.

Level of Expertise: A dichotomous variable (1 = expert, 0 = novice).

I have been running a Pearson’s correlation analysis in three ways:

1. Using Chess Rating as the dependent variable with all scores of fine and coarse segmentation together (however, the scores of fine and of coarse have been averaged per each player, thus 4 scores became 2 scores per player)

2. Using Chess Rating and only fine segmentation scores.

3. Using Chess Rating and only coarse segmentation scores.

My hypothesis is that there is a significant positive correlation between the number of events identified in fine and coarse segmentation tasks for expert players, and this correlation is expected to be stronger for experts compared to novices.

My first question now is: Do I have to necessarily create an average of the two scores for fine and also for coarse? Because if so, my Shapiro-Wilk Test becomes significant for analysis 2 and 3 … Meaning that I can not conduct the test, as the assumption isn’t met.

My second question is: Is there maybe another analysis I could run in order to investigate my hypothesis?