# Thread: Non-parametric equivalent of a factorial anova (not repeated measures)?

1. ## Non-parametric equivalent of a factorial anova (not repeated measures)?

Hi there.

Firstly apologies for the awful spelling in the main title - obviously I meant factorial

Secondly hello I am new to this forum and very much hoping someone can help me out. I am performing analysis on a research questionnaire for my MSc and have got stuck on the last bit - searching the internet for hours has just caused more confusion!

My respondents were scored on an attitude scale after exposure to one of three information conditions).I have performed all the basic analysis but now I want to look for interactive effects of two independent variables (gender and test condition) on the dependent variable (attitude score - which is interval type).

I had planned to use factorial anova (as suggested by my supervisor) but my data is non-normal. Searching the internet initially seemed to point to Friedman test but further reading seemed to suggest this was the equivalent to a repeated measure anova - which mine is not (each respondent was only scored once).

Any suggestions would be appreciated - I am using SPSS 18 (PASW). I have never done transformations before and that would probably push my knowledge - so if there is an appropriate test anyone could suggest I would be very grateful.

Kind regards
Chegginson

2. ## Re: Non-parametric equivalent of a facotrial anova (not repeated measures)?

A great resource for any up and coming statistician/researcher is the UCLA website. HEre is a link the the UCLA's site about SPSS tests that answers your question(LINK).

Since they say it better:
Originally Posted by UCLA website
Friedman test

You perform a Friedman test when you have one within-subjects independent variable with two or more levels and a dependent variable that is not interval and normally distributed (but at least ordinal). We will use this test to determine if there is a difference in the reading, writing and math scores. The null hypothesis in this test is that the distribution of the ranks of each type of score (i.e., reading, writing and math) are the same. To conduct a Friedman test, the data need to be in a long format. SPSS handles this for you, but in other statistical packages you will have to reshape the data before you can conduct this test.
Hope this is helpful!

3. ## Re: Non-parametric equivalent of a facotrial anova (not repeated measures)?

I think that suggests what I thought that Friedman is not the right way to go. I am wondering if there is not an exact equivalent. I have found company called genespring offering a programme that can perform a non parametric option of a two way anova so looking into that.

Does any one know if there is/isnt a direct non-parametric equivalent to the two way anova? The one only I have found is for repeated measures anova.

4. ## Re: Non-parametric equivalent of a facotrial anova (not repeated measures)?

Originally Posted by chegginson
I think that suggests what I thought that Friedman is not the right way to go. I am wondering if there is not an exact equivalent. I have found company called genespring offering a programme that can perform a non parametric option of a two way anova so looking into that.

Does any one know if there is/isnt a direct non-parametric equivalent to the two way anova? The one only I have found is for repeated measures anova.

Yes, there is. If you have a 2x2 ANOVA, then it is legitimate to perform the usual OLS parametric procedure on the ranks of the original data points. If your design has more levels (than 2X2), then this complicates matters because you have to an alignment on the data before performing the rank transform procedure.

5. ## Re: Non-parametric equivalent of a facotrial anova (not repeated measures)?

Thanks for your reply but you went a bit over my head I am afraid (my masters is in Animal Welfare and I am very far from a stats expert).

I was told to do a factorial anova (or two-way) and the data I have is one dependent variable (interval scored data) and two dependent - gender and information condition (which has three ordinal categories). I have found an effect of gender and information condition on the dependent variable but was told I need to assess for interacting effects of the two. As I said my data is not of normal distribution. If you could explain in a bit simpler terms for me the test I could use and how that would be carried out that would be great!

It had taken me alot to get my head round what I have done so far so data transformations etc seem to be one step to far at the moment

Thanks again.

6. ## Re: Non-parametric equivalent of a factorial anova (not repeated measures)?

Is your data not normal? Or are the residuals from the fit non-normal? We only care if the error term in the anova is normally distributed so if the residuals look approximately normal then there isn't much of a problem.

7. ## Re: Non-parametric equivalent of a factorial anova (not repeated measures)?

As far as I am aware it is my dependent variable data that is not normal - according to a shapiro-wilks test.

8. ## Re: Non-parametric equivalent of a factorial anova (not repeated measures)?

Then try fitting the anova first - we don't actually care if the dependent variable is normal. In terms of normality the only thing we care about is the error term so check the residuals for normality.

9. ## Re: Non-parametric equivalent of a factorial anova (not repeated measures)?

Do you mean try performing the avova on the data in spss? How do I check the residuals for normaility. I had a look through the options and outputs of my previous normaility test but couldn't see anything relating to reisduals. Sorry as you can guess I am quite a novice - really appreciate all your help though.

10. ## Re: Non-parametric equivalent of a factorial anova (not repeated measures)?

I couldn't help you there - I don't use SPSS...

11. ## Re: Non-parametric equivalent of a factorial anova (not repeated measures)?

No problem. Thanks for helping anyway :O) I shall try to investigate

12. ## Re: Non-parametric equivalent of a factorial anova (not repeated measures)?

Here's a website on testing for normality of residuals in SPSS. I would recommend the use of the QQplot as the easiest means of assessing normality.

http://statistics.laerd.com/spss-tut...statistics.php

13. ## The Following 2 Users Say Thank You to trinker For This Useful Post:

chegginson (12-03-2011), Dason (12-03-2011)

14. ## Re: Non-parametric equivalent of a factorial anova (not repeated measures)?

Originally Posted by trinker
Here's a website on testing for normality of residuals in SPSS. I would recommend the use of the QQplot as the easiest means of assessing normality.

http://statistics.laerd.com/spss-tut...statistics.php
This is actually the website I used and followed to check whether my data was normally distributed. I performed the test on the raw data values for the dependent variable of each group and the results showed it was not normally distributed. I am confused with what residuals are and whether I am calculating normaility of my incorrectly.

15. ## Re: Non-parametric equivalent of a factorial anova (not repeated measures)?

Is there a way to save the residuals to the data set when you perform you anova?

16. ## Re: Non-parametric equivalent of a factorial anova (not repeated measures)?

For now residuals are the error terms (how far each point lies from the best fit line of the model) or how much the model doesn't explain the data. These need to be normally distributed. Now for SPSS let me get back to you on this one.

EDIT: here's the ucla website on examining residuals in SPSS. You may have to type syntax to do this (not sure SPSS isn't my program of choice) http://128.97.141.26/stat/spss/webbo...2/spssreg2.htm