Malaika, your problem may be that you are doing a Kruskal Wallis via the "legacy" submenu (as Guenael mentioned it earlier).

here is the detailed procedure:

Go to the Analyze menu>non Parametric tests>independent samples.

- Objective tabs: choose custom analysis

- Enter the data to compare in the Fields tab

- In the Setting tab, choose Kruskal Wallis, and choose all Pairwise.

Click on run

In the output document, double click on the Hypothesis test summary table, a new window will opened, named “model viewer”. On the right hand side, at the bottom, there is a “view” menu where you can choose pairwise comparisons. A new table will show up with pairwise comparisons and it will show the p value for each of them.

Hope this helps!

Sure. So when you perform an ANOVA just click on the post hoc menu and choose the post hoc test you need.

http://www.youtube.com/watch?v=8DHFTd-KjjI

Greets

As Olivier stated, sure you can perform post hoc. There are several post hoc techniques you can employ in SPSS, buy I prefer using the Duncan, Turkey and at times LSD. Just try using it and see how it works.

Thank you for the answers. But how do I do with non-parametric tests?

Depends a little on what type of nonparametric test. For Kruskal-Wallis, one approach is to run pairwise comparisons (i.e., do a Mann-Whitney for each pair of groups) and adjust the alpha for each using a Bonferroni multiple comparison procedure. That's probably the only way most software is set up to run the post hocs. For chi-square, one older approach is to look at the contributions to chi-square for each cell but you can also form 2x2 subtables and, again, use Bonferroni.

It is great that SPSS has this post hoc option direct implemented into the ANOVA test. But I do not see any similar option for nonparametric tests. I am afraid that the "manual" way of pairing, as Warren has described, is the only way to solve this problem. Is this correct?

Actually, with a recent version of SPSS the post hoc comparisons are done automatically using Mann-Whitney or Wilcoxon with Bonferroni correction.

You just have to double-click on the output table and select "Pairwise Comparisons" in the field "View" of the right panel.

This only works if you don't use the "legacy" functions.

I am interested in the old fashion way of Kruskal wallis Post hoc, I believe that is the post-hoc i was taught years ago, that didn't have a name...

I do it like this: Make the KW statistics. Note the mean ranks and the mean rank differences. Take the mean rank differences and compare it to the chi-square table under df=(number of groups-1) and requested alpa. If the mean rank difference exceeds the value in the chi-square table it is significant to that level.

Does anyone recognize the method? Do You believe it is confident? Does it have a name or can it be cited somewhere?

I have SPSS version 22 installed and tried the post-hoc pairwise comparisons for Kruskal Wallis, but I do not seem to get any post-hoc output. Anyone encountered the same problem? What should I do to get the post-hoc comparisons?

Malaika, your problem may be that you are doing a Kruskal Wallis via the "legacy" submenu (as Guenael mentioned it earlier).

here is the detailed procedure:

Go to the Analyze menu>non Parametric tests>independent samples.

- Objective tabs: choose custom analysis

- Enter the data to compare in the Fields tab

- In the Setting tab, choose Kruskal Wallis, and choose all Pairwise.

Click on run

In the output document, double click on the Hypothesis test summary table, a new window will opened, named “model viewer”. On the right hand side, at the bottom, there is a “view” menu where you can choose pairwise comparisons. A new table will show up with pairwise comparisons and it will show the p value for each of them.

Hope this helps!

I'm having an issue with running the "new" parametric version of the Kruskal-Wallis. Every time I try to run it using Analyze -> Nonparametric tests -> Independent Samples, I get an obnoxious error message saying "org.xml.sax.SAXParseException: An invalid XML character (Unicode: 0xc) was found in the value of attribute "value" and element is "value". Execution of this command stops."

When I run the Kruskal-Wallis test using the legacy dialogues (K independent samples) it works just fine, but I need the post-hoc results. I have spent literally hours trying to figure out what this error means, but have come up with nothing.

I saw that Warren said "For Kruskal-Wallis, one approach is to run pairwise comparisons (i.e., do a Mann-Whitney for each pair of groups) and adjust the alpha for each using a Bonferroni multiple comparison procedure" , but I don't know how to do this. Is there a formula to get the proper alpha to use when doing the Mann-Whitney tests? I tried just running them as is, but I then get several "significant" results between groups that were not significant in the overall Kruskal-Wallis test. I assume this is because of the error multiplication in the Mann-Whitney test, but I may be wrong? Any help anyone can give me would be greatly appreciated!

What I've since been told is that the simple Bonferroni correction is to just divide the alpha level by the number of tests. I have 4 levels of my variable, which means I am running a total of 6 Mann-Whitney tests. Dividing the standard .05 alpha by 6 gives me a new value of .00833 for a significant alpha level. When I do this, however, I now find no significant differences. I believe this is due to the standard Bonferroni correction being so conservative, is that correct? And if so, is there something else to use? It seems inaccurate to report there was a significant Kruskal-Wallis test but then no significant pairwise combinations.

Dear Marie Billaud, your answer just saved my life!!!! You are the best!!!!!!

Marie Billaud, your answer saved my life too. Thanks a lot!!

@Richard Did you solve the problem with the error occuring while performing the nonparametric test? I'm having the same problem at the moment and can't figure out what to change in my datafile. 

Right after asking I solved the problem myself. So for everyone with that "invalid xml character"-error: It's seems to be a problem with string variables in the dataset. Although they are not used in the nonparametric test they produce this kind of error. Just delete the string variables and everything should work just fine.

I am running what Marrie Billaud has described above but I am gettiing the message:

Execution of this command stops.

Error: no license to run Component npar

Can anyone help me what is the problem?

Dear Louk, this happens only with this test???

yes this happens only with this test. Could you please help me?

@Louk You need to have the non-parametric module installed in order to perform Analyze menu>non Parametric tests>independent samples. If you don't you should use the legacy dialog.

Dear Richard, dear Anja,

did you run across any other solution to this problem? I'm getting the same error message when performing the Kolmogorow-Smirnow-Test. I checked my variables double and they are all numeric. 

Error message reads: "org.xml.sax.SAXParseException: An invalid XML character (Unicode: 0xc) was found in the value of attribute "value" and element is "value". Execution of this command stops."

Many thanks in advance

Caspar

Edit 22.10.2014:

I couldn't solve the problem but it does not occur when i use the older version of the NPTESTS-command: NPAR TESTS its somewhat more complex but also, as i find, clearer in the specifications and it works.

Thank you Marie for your clear post! Really useful!

this is from IBM's website, it clearly explains how to make a posthoc analysis in Kruskal-Wallis test:

'The newer NPTESTS procedure offers post hoc tests for the Kruskal-Wallis omnibus test. In the menus, select Analyze>Nonparametric Tests>Independent Samples. (Note that in versions through 21, you can only specify Scale dependent variables, while beginning with Version 22 you can specify Ordinal or Scale dependent variables.) You will get a Kruskal-Wallis test and will also get post hoc tests automatically if the omnibus test is significant if your grouping variable has more than two levels. Note that the full test results for the K-W test and the post-hoc tests are contained in the Model Viewer in the output, if you have your settings to show Model Viewer output. You need to double-click on this object in the output to see the omnibus test results, which will be on the right-hand side of the Viewer when it opens. Select the View drop down at the bottom of the screen and Pairwise Comparisons to see the post-hoc results. For the pairwise comparisons, adjusted significance levels are given by multiplying the unadjusted significance values by the number of comparisons, setting the value to 1 if the product is greater than 1.'

Hi,

I'd like to recommend Andy Field's SPSS tutorials on You Tube that has helped me a lot with my analysis, especially non-parametric tests.

Also, if you prefer to have all the output in the Output Viewer (instead of going to the Model Viewer) you can select Edit>Option and select the Output tag. In the Output Display choose "Pivot tables and charts". Best to do this before you do your analysis.

/Anna

Hey,

I am using the new modell viewer for doing the kruskal wallis test - but I am not sure which test this program is using for the pairwise comparison? u-tests?

thanks

Hi Michaela

Please try this path may be helpful!

Analyze → Nonparametric Tests → Independent samples

In Settings tag select Customize test and click Kruskall-Wallis

All pairwise in Multiple comparisons box and Run!

thanks - but which test is being used there? u-test? bonferroni? ....?

Hi Michaela, I am not an expert on this but I believe the pairwise comparisons are using the Dunn-Bonferroni approach. Perhaps this link from the IBM website might help? 

http://www-01.ibm.com/support/docview.wss?uid=swg21479073

Lauren

Hello everyone, thank you so much for your invaluable help!

I have some more question to be solved:

For one variable significant by KW, there is none post hoc pairwise significant comparision (p

I am guessing that "sig ady" is the Spanish version of adjusted significance. SPSS calculates an "adjusted" significance based on some sort of complicated concoction of sample size, dfs, etc. In it similar to r versus adjusted r. 

As for your question, I have seen this happen before. Post hoc analysis takes into account the variation that includes that contributed by other variables. It also introduces a significant degree of conservatism into the equation. If you think of the Bonferroni correction, the required value for significance drops as a function of the number of groups compared. In doing so, what would previously have been significant may "become insignificant" based on the reduced significance level. 

When interpreting these results, what you would say is that a Kruskall-Wallis test indicated that there was a significant difference between the groups all together, but there were no significant bivariate differences. In writing up the results of a regression analysis, I wrote it in this way "In the second model, while the model as a whole was significant, none of the individual predictors were significant, although family supports approached significance (β = -.23, p < . 10)." You likewise have one variable that is approaching significance (if you partake in the idea that "approaching significance is p

Thank you so much for your help Richard!!

Of course it´ll help!

Is the adjusted significance equal to Boniferroni correction procedure?

Im sorry, im encountering some problem here, it says "Pairwise Comparisons view encountered some problem(s) and will not be shown." I hope u could help. Thank you.

Yudha, have you solved your error? I have the same problem, pairwise comparison is not possible, there is some error and I don´t know what does it mean - "Pairwise Comparisons view encountered some problem(s) and will not be shown." 

I tried to find out the reason but without succes. Can anyone help me please?

Pavla, I just encountered the same problem, turned out it is a technical error. I needed to change the type of output window, then it worked. See here the link:

http://www-01.ibm.com/support/docview.wss?uid=swg1PI40127

Dear Forumers,

I have encountered the same problem and it was highly useful to find this forum chatter.

However, I have faced an issue that, as far as I see, is not addressed in this forum. Namely, is anybody aware what sort of non parametric post hoc test is exactly applied within SPSS automatically? Many of you suggest changing the settings through Edit, Options, Output, and ticking "Pivot tables and charts" instead of using the Model Viewer, and then perform a non parametric Kruskal Wallis all pairwise multiple comparison. This method provides a nice table in the Output with nominal significances as well significances adjusted for multiple comparisons, seemingly Bonferroni-type. However, others suggested the traditional method by splitting the file in different ways or selecting the subgroups for analyses and performing serial Mann-Whitney U tests for all combinations. I was interested in if the two methods give identical pairwise outputs and (non-adjusted) significances. Notably, I had to face that they indeed DO NOT!  Furthermore, I tested if the output might be from pairwise Kruskal Wallis test as "post hoc", however, such analyses provide a third different type of output. Though the numbers are similar to some extent, they are not identical. Does anyone have an idea what is going on here?

Thank you for the comments in advance!

I confirm the link that Jirina provides offers a solution to the recent display problem with SPSS v23 for the post-hoc of the Kruskal-Wallis. Finally!

To fix, do the following:

"you can go to Edit - Options - Output tab and change the Output display method from Model Viewer to Pivot Table and charts and run the analysis again.

Then you will get all tables and charts printed on viewer also for the pairwise comparisons tests"

Note that this nonparametric post-hoc test accounts for the variation in all of the groups and is more reliable than just performing Mann-Whitneys separately. It is called a Dunn-Bonferroni post-hoc

Great thanks a lot ! 

Thank you for these useful tips. I was also having this display problem. Not anymore.

If you think that Bonferroni-corrections make the tests too conservative, you could think of using some adjustments of the original Bonferroni-method. Like Hochberg-modification. This site has easily understandable instructions of how to do it  :

http://onbiostatistics.blogspot.fi/2009/08/hochberg-procedure-for-adjustment-for.html

It needs some extra "hand work" but is (much) less conservative than the original method. For instance, if all the unadjusted p-values of the pair-wise tests are

I obtained ajusted p-values following your advises but I have a doubt. What is the multiple comparisons correction that SPSS uses to give those adjusted p-values? Is it Bonferroni? Do anybody know?

This was also my question posted above recently, anybody have an idea?

 Dear Levente, 

I have just read that is Bonferroni correction. It is mentioned at the footnote.

How do you report the post hoc results in a paper?

In versions before Spss v21 only scale variable can be accepted, while in v.22 and over, ordinal variables are accepted too.

http://www-01.ibm.com/support/docview.wss?uid=swg21477370