So, I'm having data which represent two groups, one that used tool 1 and the other that used tool 2. I asked a series of questions (10 questions) and I recorded the time as well. Afterwards I calculated the productivity based on the data and now I want to analyse if the productivity of tool 1 is significantly different than the productivity of tool 2. To achieve this, I need to run either T Test or nonparametric tests; however, in order to decide which one should I use, I need to firstly ran the normality test, which I did in SPSS. Since I have 80 participants, I used Shapiro-Wilk normality test.

The results are something like this:

QS1 ... tool1 0,000

QS1 ... tool2 0,000

as can be seen from the example, the sig. is 0,000 which means that data is not normally distributed and I should use nonparametric tests. However, what should I do in case of this:

QS2 ... tool1 0,047

QS2 ... tool2 0,586

In this case QS2 and tool 1 the sig. is < 0,05, which calls for nonparametric tests. However, QS2 and tool 2 are > 0,5, which means I should use T Tests. How should I interpret these results and which tests for comparing the means should I choose?

  • $\begingroup$ I would use the non-parametric throughout. That way you don't have to worry so much about the shape of the distribution, and I am guessing your questions have ordinal characteristics. The non-parametric test is a good one. As an aside, look into the problem of multiple tests if your p-values are not small. $\endgroup$
    – mandata
    Commented Feb 2, 2016 at 23:38
  • $\begingroup$ How does "productivty of tool" have to do with the answers to the 10 questions? What are the questions recording (/how are they represented?). Much about this question is unclear $\endgroup$
    – Glen_b
    Commented Feb 2, 2016 at 23:43
  • $\begingroup$ Excellent point, I will update the main question to make it more clear. $\endgroup$
    – uglycode
    Commented Feb 3, 2016 at 10:40

1 Answer 1


If your data are not normally distributed you should use a non parametric test.

You can use the Wilcoxon-test for paired data: here is a guide R Wilcoxon Test

For unpaired data, use the Mann-Witney U test of location in SPSS.

  • $\begingroup$ I don't think Wilcoxon-test is the right way to go, since (as far as I know) it is intended for dependent data, i.e. values PRE and values POST. However, in my case, I have two separated groups (one that used tool-1 and the other that used tool-2; one participant did not use both tools, but just one, randomly assigned). I think Mann-Whitney U Test is more appropriate here, since I have two independent groups. $\endgroup$
    – uglycode
    Commented Feb 3, 2016 at 13:18
  • $\begingroup$ Sorry didn't fully understand the situation $\endgroup$
    – GGA
    Commented Feb 3, 2016 at 13:21
  • 1
    $\begingroup$ No problem, I just hope that I understand it well enough to provide a solid argument. My original question addressed the situation where data from one group is not normally distributed whereas the other group has normally distributed data - which tests should I use? I think the basic T-Test cannot be used in this case, even though one group has normally distributed data. So, non-parametric tests are the way to go, I believe. mandata confirmed this as well in his response. $\endgroup$
    – uglycode
    Commented Feb 3, 2016 at 13:24

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.