too difficult for me ...

#1
Dear all,

I am doing some research in psychology/neuroscience/philosophy and I have just finished one of my experiment. Before doing this experiment I thought that the statistical analysis would be 'relatively' easy to perform, but I am not sure any more :shakehead

Here is the experiment :
The subject see two successive pictures on the screen, alternating roughly 30 times. In 80% of the cases, there is a change between the 2 pictures and 20% there is not.
The subject has to find the change if there is one. If he did not find the change after the 30 repetitions, he has to evaluate if he thinks there was a change during the presentation by clicking from 1 (sure there was no change) to 5 (sure there was a change even if he did not perceive it).

This task is repeated 100 times and 18 subjects participated.

Now comes the problems :rolleyes: :

What I want to know is if the subjects clicked more often on 1 when there was effectively no change and on 5 when there was one that they didn t perceived.
As a bad scientist :D the first thing I did was to perform a two way repeated measures ANOVA with (pictures with change or no change) and (number of times the subject pressed on 1,2,3,4,5) as factors... And the interaction I am looking for is significant. Yes !

But now I realise that s not so easy and that s why I need your help.
First, can I do this type of anova when one of the factors is a scale ???

The second problem is even more important but also more complicated, I hope I ll be clear enough to be understandable. Whereas the number of times the subjects as to judge if there was a change when there wasn t remained constant (20% * 100 = 20 there was almost none false alarms) the number of times subjects as to judge if there was a change when there was one depends on their performance. So the number is different from 20 for all the subjects but also highly different between them.
To solve this problem I had the idea to perform an ANOVA not on the numbers of times but on the frequency the subjects pressed 1,2,3,4 or 5.
But in that case, the different values of the scale become related and I don t know if an Anova becomes possible.

Last problem I think, the distribution of responses is not gaussian. For exemple in the case there was no change, the subjects chose most exclusively the key 1.

If somebody managed to read this loooong text and understood what I said in my baaad english, I like to know is opinion before doing some analyses that may just be nonsense.

Thank you !