If I am following, I think you can use a t-test for the whole thing.
Your latter question was looking at a particular combination - how many times was each dolphin tested on each combination?
Basically the research we are doing involves measuring the time taken for two dolphin subjects to make their decision in a Matching-to-Sample study. What this means is that the dolphin is first presented with a sample object, let’s say a square, and then the dolphin is presented with two alternative objects, a circle and a square. The dolphin is supposed to pick the alternative that is identical to the sample in this case the square. The time taken for the dolphin to make his/her selection is considered decision time (DT).
There are 6 objects, resulting in 60 combinations in total.
1. 1)What statistical test can I use to compare an individual dolphin’s performance (in terms of DT) across the 6 objects? Would you suggest ANOVA?
2. 2)For normality test can the Shapiro-wilk’s test be used?
3. 3)What statistical test can I use to compare both dolphins’ performance to see if they performed similarly across the same object combinations? Would it make sense to test whether the differences between the average DT of the two dolphins on each object are significantly different by using a t-test?
If I am following, I think you can use a t-test for the whole thing.
Your latter question was looking at a particular combination - how many times was each dolphin tested on each combination?
Stop cowardice, ban guns!
Nithya (07-09-2015)
Hi,
I think you could just do a two-way ANOVA , one factor being the shape and the other factor the dolphin. This way you have a chance of captueing interaction effects as well. e.g. if one dolphin is better at a particular shape then the other.
regards
Nithya (07-09-2015)
Yes, I missed a part. When you said dolphin was time, I forgot that dolphin needed to be in the model.
Thanks rogojel. Side note, I watched Black Sea and one of the actors playing a Russian looked just like your avatar.
Stop cowardice, ban guns!
The problem is I don't think that my data is normally distributed so I think I cannot use a t-test.
And my sample size is very small as well.
For each combination, I have a sample size of 4 for one dolphin and 8 for the other
For each object, I have a sample size of 10 for one dolphin and 20 for the other.
Basically I am trying to see if I can prove whether an individual dolphin has significantly different DT across different objects and different combinations. Then I want to compare the DT of the two dolphins against each other to see if a similar trend emerges in terms of DT/object and DT/combination.
http://pixlr.com/express/?method=pos...binations.jpg#
hi,
the normality in itself should not be of concern, you could still try a two way ANOVA and analyse the residuals afterwards - the distribution of the residuals is that what is important.
The small sample size could be an issue, but there is no statistical method to help that, you might simply need more data. In your place I would still try the analysis first, just to see .
Still, if non-normality worries you, take a look at this article :
http://terpconnect.umd.edu/~leahkf/p...kTransform.pdf
regards
regards
Tweet |