Paramedic Research - Need help!

Hello all,

I have come across your forum after trying to source some statistic knowledge. I am currently planning a research project which I am hoping to have published in the early part of 2014. The one element of the project I am struggling with is the statistical analysing of the data I have collected.

The project is relatively simple, so I am hoping the method of which to test my hypothesis will be too.

I have 20-30 student paramedics listening to 7 sounds repeated twice in random order. They are asked to correctly identify the sound they hear. They can either get it right or wrong. There is no in between.

What is the best way I can statistically do this?

Any help is appreciated!!




TS Contributor
Are you trying to estimate the probability that a randomly selected paramedic student guesses the sound correctly, or are you comparing the first and the second sound probabilities, or are you comparing correct responses among categories of paramedic students?

I am trying to look at whether paramedics can correctly identify the sounds. In a clinical setting it's imperative that they can correctly identify the sounds. The reason I decided to play sounds twice was to ensure it wasn't fluke they guessed it?

Does that make sense?




TS Contributor
I am trying to look at whether paramedics can correctly identify the sounds.
So you want to know whether their rate of correct responses
is larger than expected by chance? Or you want to know
whether they are 100% correct (but in that case you wouldn't
need a statistical test, just 1 incorrect response would reject
the 100% assumption)? Or is there some other threshold you
want to test your sample data against?

With kind regards



TS Contributor
Sorry, I did not assume as a hypothesis that all subjects
were unable to identify any of the sounds and therefore
would completely resort to guessing. Rejecting such
a hypothesis of complete and total inabilty in all subjects
maybe could be considered a bit trivial by some.

Anyway, since seemingly it is not possible to calculate the
expected rates for correct answers due to guessing, you
could perhaps just calculate % correct for each sound, and in
addition the (95%) confidence intervals for these proportions.

In addition or alternatively, you could count number of correct
responses for each subject, calculate the mean for this across all
subjects, and also calculate a confidence interval for that mean.

Just my 2pence



TS Contributor
seems like the complete inability to guess (p=0%) would be significant in and of itself.

i think what karabiner said is pretty much the way it works. Basically use a 'one sample test of proportions' and set the null to p=50%, which is what is expected under random guessing.

The thing to keep in mind in this type of study is that there is some correlation among repeated tests done on the same subject. As you are probably aware, some students are just better prepared than others. I think the easiest approach would then be a 'cochran mantel hanzel' test, stratified by subject. Sounds scary but is not.