Statistics

[Jachym]

Hi guys,
do we have any experts on statistics amongst us?

I need to analyze differences between two data samples, basically I have a list of values for about 20 people (before/after) and I want to find out if there is a statistically significant difference between before and after values.

I originally did a paired T-test, but Im now wondering if Wilcoxon test would be more appropriate. Im also a bit unsure about using one- or two-tailed test. The values can change both up and down so I used the two-tailed one, but correct me if Im wrong

[kocher]

I speak as an amateur statistics.

I believe that, in this case, the correct test is:
T-dependent test for paired samples

When the distribution of data is asymmetric (that is, the distribution is biased) or the distribution has very large tails, then the Wilcoxon ranks test (also known as the Mann-Whitney U-test) may have three to four times more statistical power than a t test (Student t test,)

[Jachym]

Hi,
thanks, I used t-test because it is built-in to Excel and easier to use, but Im now looking at the data and the problem is that there is only 20 values.

What would you use if you wanted to test results before/after, where the data looks like this (in the same order):

Before:

Before:

96 77 82 70 94 61 96 97 93 95 65 85 77 97 91 86 96 77 94 84

After:

94 88 86 79 95 62 94 93 95 96 76 82 86 98 94 77 100 82 91 91

[kocher]

Give me a little time, I'll study it

[Jachym]

Thanks Javier! I really appreciate your help. I want to have someone else's opinion, you know what I mean.

Basically I have test scores like this from 20 people before and after. And I want to see if the values are different after.

Im then also thinking about doing a different statistic, where I would compare male and female scores, in this case however I would compare them in general - i.e. if there was a difference between initial scores boys/girls and that would then be an unpaired test - again, not sure it T-Test is ideal for this...

[kocher]

The calculations were performed with MedCalc software (an excellent statistical program, especially focused on biostatistics).



I performed the Student T and the Wilcoxon T for paired samples

T-test:



Wilcoxon-Test:



I think you can state that there is no difference between the two samples, since p>0.05

[Jachym]

Hi Javier,
yes I understand, thanks a lot, but I actually have a lot more samples, I can do the tests myself (I dont have such fancy SW ), but I was just thinking if you could tell me which one you would say is better to use for this, more appropriate. I currently have everything calculated with t-test, but Im thinking that maybe there is not enough data and not normally distributed, in which case the WIlcoxon would be better

[kocher]

Hi Javier,
yes I understand, thanks a lot, but I actually have a lot more samples, I can do the tests myself (I dont have such fancy SW ), but I was just thinking if you could tell me which one you would say is better to use for this, more appropriate. I currently have everything calculated with t-test, but Im thinking that maybe there is not enough data and not normally distributed, in which case the WIlcoxon would be better

I agree, the appropriate statistical would, in this case, the Wilcoxon test

[Jachym]

Thanks!

I tried to make my life easier (easier to do a T-test in excel), but then my ambition for perfection when it comes to statistics stepped in and I changed my mind

[kocher]

Thanks!

I tried to make my life easier (easier to do a T-test in excel), but then my ambition for perfection when it comes to statistics stepped in and I changed my mind

An interesting attitude