anyone have some advice??
Hi! Stumbled over this forum recently, there's so much knowledge here I could just sit and read and read. Anyway, I'm not the strongest in stats so I thought I'd ask you guys! I'm thankful for any suggestions I can try out. (I'm new in software too, I want to use R, but have spss and jmp just in case i give up R ).
So!
I have 9 similar figures that I ask a bunch of people to rank
First I want to see if the figures are ranked differently, and to what degree, is one of them ranked significantly different from the others? (I think I can use an g-test here, is that right?)
Will be very thankful if someone can help with little or a lot!
Last edited by Dave4; 01-09-2012 at 11:20 AM.
anyone have some advice??
There are various ways to do this. One, if some intervention is occuring, is within subject ANOVA. The omnibus F test would tell you if there was a change; you could use I believe either contrast or post hoc test to see if there was a statistically signficant change (I work with between subject ANOVA and am not sure if the same methods apply, but SPSS will definitely do this).
You would have to have a measure showing their relative rankings to test at different stages (say after some interventions).
Dave4 (09-02-2011)
i looked at it again now..but i think i need things in with a spoon...could u be more specific? is there any downside by doing it this way?
The downside would be if your data does not support the test. Too few people, learning unrelated to your test (history or threats to internal validity). One thing that is not clear to me, is what exactly your are testing, that is what is causing the change in your model? Do you have a control group, a pre and post test?
ok. thank u.
I just realized that my data are non-parametric (yes I'm at that level...) so then I can't use anova can I right?
So maybe a kruskal-wallice rank sum test to see if any of the figures are looked at differently then? are there different kruskal-wallice tests? when I google to find formulas for R software I find several different ones...
Just want to mention this: data itself isn't "parametric" or "non-parametric". Data is data. The tests we do and the models we apply to the data are either parametric or non-parametric (or semi-parametric...) but data itself is just what it is.
Dave4 (10-04-2011)
thanks Dason. my mistake.
But since I can't summarize the rankings my subjects give the figures in any logical way, they can't have a normal distribution? and thus I have to use non-parametric statistics on them?
thanks Dason. my mistake.
But since I can't summarize the rankings my subjects give the figures in any logical way, they can't have a normal distribution? and thus I have to use non-parametric statistics on them?
Last edited by Dave4; 01-09-2012 at 11:21 AM.
To follow up on what Dason said, parametric methods are what you use when your data is normal (or multivariate normal). Non-parametric methods don't require normality. So your data is normal or not and based on that you chose parametric or non-parametric tests.
Unfortunately I don't know the non-parametric methods you are using so I am no help there....
You don't need the data to be normally distributed to use parametric statistics. You can set up lots of models without using the normal distribution that would still be classified as parametric.
ok...so.. I think I've found that Friedman is probably not the best test anyway. Should I look for a test that tests the individual subjects rankings in the 4 figures in round 1 to round 2 or one that tests the medians between round 1 and 2?
well, anything that assumes a distribution of any type can be classified as parametric... it is perhaps because you (along with mostly everyone, so that's a big 'you' there ) are more used to using the normal distribution than any other of the myriads of distributions out there.... plus just to add the 2-cents here, most of the time (especially in the general linear model) the distribution of normality is not on the data itself, but on the residuals of the model... in regression/ANOVA in particular, your predictors and whatnot can follow any distribution, but as long as the residuals are normally distributed, the properties for the inference will hold....
for all your psychometric needs! https://psychometroscar.wordpress.com/about/
Tweet |