ugh, any SPSS gurus out there?

A

ok, so I have yet another SPSS question. Im currently doing the most massive results chapter ever and in an attempt to keep it concise I need to do an ANOVA for my data. Although, it's all non-parametric stuff, so it shouldn't be ANOVA. The book I have says do Mann Whitney instead of ANOVA, but I have 4 groups I want to compare, not 2. I also don't want to transform my data as some of it is count data which I've got a paper that says shouldn't be transformed due to having lots of zero values and the problems with that. Also, I've used this as a reason for doing loads of other tests for other stuff, and I really don't want to have to go back to redo them all with transformed data.
So, the question is, what can I do? I have data from 2 months, split into 2 groups for each month. Ie. M!, group 1, M1 group 2, M2 group 1, M2 group 2. I want to see if there are any significant differences in the groups, i.e. between months and within groups within months. Anyone know what test I can use?!

A very tired thanks in advance...

A

ok i got it, kind of. I need to use a Friedmans test, which I can do in SPSS, but there is no post hoc test function in version 17 which I have, only 18.
So I can do it in R...off I go...

A

Hey AQ, I don't really use SPSS, i'm a mac user so got prism, but if you can't transform to do an ANOVA and its non parametric can you do a Kruskal Wallis? This is similar to Man Whitney but allows you to test 3 or more groups?

Don't quote me on it, but you could have a read up on it online and see if it fits what you want? Also, if you do post testing, like Newman-Keuls then this will test each variable against eachother individually if it finds that the original test is significant.

Not sure if that makes sense - I know what I mean, but i'm not good at explaining stats! if you need more help let me know and I'll PM you, my partner is a stats man, and although he's away will have email on him.

Good luck,
AL x

A

ah, ok forget that, they are repeated measures. But if thats the case, if you didn't use ANOVA style, would it be wilcoxon matched or man-whitney? ARGHHH, I hate stats!!!!

it's amazing what i've eventually started to pick up though just through waaaaay too many hours of data analysis!!! :)

Hope R works out for ya, and you get the results you need.

A

Hey AL! thanks for that! They are all Wilcoxon tests, it's the same sites samples in 2 different months and they have the same treatments for each month. But I've just found out that these tests aren't really what I want after all. They will tell me if there is a difference in the groups, and the post hoc says which group is different, but what I really need is to find out which of the many variables I've measured for each group is causing the difference. So it's back to factorial and ordination methods methinks. I can do CCA stuff in R so I'll give that a shot. Although I'm not sure there is anything apart from super advanced stats that will show me what I need... but thanks for the very kind offer of stats help! My boyf is a stats man too but although he's sitting in the same room, he's more interested in his computer game than anything else... pfff.

A

wow, you lost me a bit there, i'm lucky having really one variable, progression of disease. I know the feeling, Starcraft came out and I struggle to get anything from him! :) He's been going to read my thesis for me for months and now its kinda too late, but hey ho, he's got to have his chill time and my thesis ain't exactly the most thrilling read! :)

Good luck, hope these work for you.

C

Alpacalover, you know they make SPSS for Mac, right? That's what I use ;-)

A

yeah, I know, but I got prism free and it does me fine! :) Cheers though!

Avatar for sneaks

Why do you want to use a non-parametric test? because the data isn't normally distributed??

If so then ANOVA is robust to deviations in normality as long as sample sizes are fairly even.

If you download the trial of SPSS 18 (unless you have it already) it gives you about 21 days to use it with all the functions. This can allow you to perform bootstrapping in any post hoc analysis which you can use with non-normal data because it works by taking multiple small samples from your data, so it doesn't matter as much if its non-normal. - you literally just click a button in SPSS to apply this on post hoc ANOVAs and T-tests.

P

======= Date Modified 08 Aug 2010 16:31:54 =======
I think because you've got a repeated measures design and you have 1 group with 3/4 levels (ie per month), I think you would use a Friedman test if your data violates parametric assumptions. The parametric equivalent would be a repeated measures ANOVA.

I've used Mann Whitney Tests for exploring differences between a group with 2 conditions (ie minorty versus majoirtiy) and I'm i've used Kruskal Wallis test for exploring differences between a group with 3 conditions (ie ethnic 1, ethnic 2, ethnic 3). I've only used independent groups tests as I've only collected data at one time point and I'm looking at differences between group rather than repeated measures/pre- post designs

Whilst parametric tests are robust enough to cope with not normally distributed statistics, I've used non parametric stats due to small samples and also violating assumptions of normality as evident from performing Kolomonov-Smirnoff tests.

Avatar for sneaks

what field is your PhD in Pineapple? sounds similar to mine :-)

A

Whoo thanks for the replies guys!
I used non-parametric tests cos most of my data is not normally distributed, and I mean really really not normal! I don't think the ANOVA or Friedman test will give me the info I need, so I have just done with the Wilcoxon Signed Rank test for a few things. I know there are differences in my groups and I have an idea where they are, but what I really need to find out is what environmental variables are causing the differences. And I've got about 28 variables.... I have been told that a GLM will be able to sort it but I don't know what the non-parametric equivalent is. Also I haven't got much time to spend a few days trying to figure things out, and I'm quite bad with stats so it takes ages to get to know a new test!
Good idea about the SPSS 18 though Sneaks!

P

Hi Sneaks, my PhD is a mixed methods exploration of vulnerable adults experiences within  care services. I basically want to explore whether minority groups were disadvantaged in some way.........(given research that has explored double discrimination, triple jeopardy for people from minority cultural groups from vulnerable adult groups etc etc)

15697