Which stats test? please help!

I have a paper of a study i have to critique which is fine...there is plenty to get your teeth into however I am struggling on the type of test they perhaps should have used to analyse the data....

...there were originally 83 subjects (a cluster sample I think) taken from infantry recruits. All of which completed a questionnaire. then their arch index was graded and the subjects were put into 1 of 3 groups. One subject was eliminated at this point. The questionnaires relate to the retrospective data.

During their basic training period any ankle sprains were recorded. A further 2 were excluded due to lack of data and 16 subject withdrew leaving 64 subjects which contributed to the prospective data. Of the 16 subjects who withdrew from training after less than 3 weeks only their retrospective data was used.

The study refers to a statistical package SPSS and states that stats analysis was performed using alpha level 0.05. Then they go on to say the relative risk (RR) was determined with RR of 50% or more considered clinically significant and RR of 25% to 50% as probably significant. Then later in the results it refers to eg. RR 10.3 p less than 0.05 and it continues to decribe the data like this and then in percentages too.

I thought that as the data came from a cluster sample of young females it was unlikely to follow a normal distribution. Based on this, I concluded the data would be 'non-parametric' and as the both sets of data came from one set of subjects I concluded that the test they should have used was the Wilcoxon signed rank test.

However I have been doubting myself as the retrospective data was derived from a questionnaire asking them to grade past ankle sprains and based upon 83 or 80 subjects (there is some ambiguity. Whilst the prospective data which it doesnt say how this was collected but appears to have based on sprains recorded whilst training is based on the 64 remaining subjects. Is this still classed as paired data?

I am tearing my hair out here and would appreciate any help anyone is willing to impart. I should have known at 43 I should have stayed home to bake cakes!

Many thanks in anticipation.



Global Moderator
First of all, I get the impression that they are not very clear on there methodology in this paper. Don’t be afraid to ask for clarity, if you don’t understand what they have done it is not your fault but theirs. This is the general consensus, if readers don’t understand then it is the fault of the author, ask for clarity!

That is basically what I’m going to, How did they analyze there data? What tests did they use? How’s their data structured (e.g. continuous, discrete/ordinal). I also have no idea what an Arch index is. Did they first do an retrospective analysis with 83 samples and then a prospective analysis with 60? What is it exactly that they compared, previous risk to future risk? What did they use as a proxy for risk? If this is not clear from there paper then I would not accept their conclusions at all!

I will need a bit more information before I can help, however maybe someone else is more at home in your field.

So if you are pulling out your hair because of their unclear writings. Don’t worry it’s not your fault!