Wilcoxon 'signrank' test zval is not calculating

Hello everyone,

I have a matrix for statistical analysis.

I performed the normal distribution test with the shapiro wilks test and observed that the data in my hand were not normally distributed.

According to this matrix, I want to analyze the pre- and post-intervention data of 15 people with signrank. When I do this analysis, the zval value cannot be determined as output. I couldn't understand why this happened? If the zval value cannot be calculated, is there an alternative test you can recommend?

I attaced my code and examples of variables.

load('ExampleData.mat'); 
[p,h,stats]=signrank(Var11(:,1),Var11(:,2));
stats

stats = struct with fields:
    signedrank: 36

[Frederik Bergmann]

Same problem here. and I tried both methods ('exact' and 'approximate')

Answer 1

[Star Strider]

The ‘zval’ is only returned when the 'method' is 'approximate'.

There is no statistical difference between the vectors:

[p,h,stats]=signrank(Var11(:,1),Var11(:,2), 'method','approximate')

p =

0.30029

h =

logical

0

stats =

struct with fields:

zval: -1.0358

signedrank: 36

.

[Mert Dogan]

Thanks a lot.

[Star Strider]

As always, my pleasure!

[Maria Lazzaroni]

However, if I perform the signrank with 'method' 'approximate' instead of 'exact' I have different p-value results. Is it correct? Which ones I have to present as the results of the statistical analysis?

thank you

[Star Strider]

@Maria Lazzaroni — See the method section in the signrank documentation. It tells you everything you need to know about how best to use either option, depending on the number of observations in your study.

[Maria Lazzaroni]

@Star Strider actually it says: 'approximate'= Normal approximation while computing the p-value, p. Default value for more than 15 observations in x, x – m, or x – y when 'method' is unspecified because the exact method can be slow on large samples.

I have more than 15 observations, but if I run signrank without specified the method the p value is equal to 3.0518e-04, and if I specified 'method' 'approximate' p value is equal to 0.0011. Which ones I have to choose?

[Star Strider]

Use whatever you want. Your data are statistically significantly different (p < 0.05) regardless of the method you use. If you want to report it in a paper and really impress your referees and readers, use the 'exact' result.

[Maria Lazzaroni]

but if I use the 'exact' result, can I use the z value that I find when running the function signrank with 'method' = 'approximate'?

I do not understand why Matlab does not return the z value when running signrank with the method 'exect'.

Thanks a lot for the help ;)

[Star Strider]

No.

That’s probably because the 'exact' method doesn’t use the normal distribution to calculate the probability. (The documentation doesn’t go into detail about that or how it calculates the probability for the 'exact' method. The references at the end of that page probably provide those details. I don’t have free access to them.)

My pleasure!