How to test Accuracy significantly different from chance?

I

ilaria

New Member
#1
#1
Hi All,

So I have more than 60 subjects who participated in an experiment.

I would like to exclude those who performed at chance level.
The problem is where should I set my cut off? Is 55% not chance? And 57?

So my prof said I should test if the suspect cases significantly differ from chance.
But how do I do that?

I would very much appreciate your help with that!!
Best
Ilaria
 
Karabiner

Karabiner

TS Contributor
#2
#2
Could you explain this a little bit? What is the research
topic, what is the research question, how was the experiment
designed, what was measured?

With kind regards

K.
 
I

ilaria

New Member
#3
#3
Thank you for your quick reply!
It's a delayed-match to sample task; So the subjs saw an item to remember and after a delay a different or identical item was presented and their task was to make a match ("this is the same item I saw before") vs no-match ("it's a different one") judgement;

The response is given by pressing one button for match and another for no-match. Half of the trials were match and half no-match. I measure accuracy (% of correct responses = hits + correct rejection ) and RTs. These are the details I can give ;), I hope it can help

Best
I.
 
Karabiner

Karabiner

TS Contributor
#4
#4
Take each participant as an experiment of his own,
with k trials. If you perform a binomial test, then
the Null hypothesis is that a particpant's rate of
correct responses is = 0.5 (chance level). The
question is, which rate of correct responses over
k trials will lead to rejection of the Null hypothesis.
You could use e.g. an online calculator for binomial
test to find this out.

With kind regards

K.
 
You must log in or register to reply here.