|
||
|
Title: Stats, stat. Post by Noke Lieu on Dec 2nd, 2014, 6:46pm So, I'm actually a little bit stuck... Perhaps you can help. My usual go-to gurus have said that it's an interesting question, and sort of left it at that. I've been running some questionnaires- and it turns out that 3/4 runs were with an obsolete form. There are, however, cross over questions- two in particular. No problem, you say- run a T test to show that "correct" 1/4 are the same as the 3/4 and generalise... Except, it being a questionnaire, the data are Likert-like 1-5 scale. That means no-can-T-test. Then perhaps you'd want to use chi squared... but there are empty cells, so... nope. Empty cells? Well then, use Fishers Exact... Can't. Not normally distributed, and heavy edges. (see the data below) Oh, then use Mann-Whitney U test... Sure guy... but can't find the significance tables for the sample sizes; haven't found the algorithm for calculating it. And every reference I know says that it starts approximating normal with big samples- which this doesn't. Just in case you're interested, the data sets (1,2,3,4,5) are S1 {0,0,4,6,23} P1 {2,9,45,106,181} S2 {0,0,4,11,21} P2{19,0,12,55,262} I'm keenly reminded that stats exist to tease out relationships that aren't immediately obvious. If you just graph these sets, you can see they're the same distribution... but I'd like to be a bit more thorough than that... Any thoughts? |
||
|
Powered by YaBB 1 Gold - SP 1.4! Forum software copyright © 2000-2004 Yet another Bulletin Board |