I can't go to anyone at my uni, because I am the resident stats person :$

I need a way to analyse differences between groups.....

there are 20 in each group, so 40 participants

BUT each participant has up to 90 cases each, so all in all there are 1700 cases. e.g. the data set looks like

p1
p1
p1
p1
p1
p2
p2

My sup says I can't analyse it with a t-test because of there being more than one case per participant and chi square?????

I reckon this post will go unanswered, but its worth a try!

hi sneaks
is this a multiple response thingy?
love satchi

basically I have 40 participants who I've interviewed. I then code their sentences - so for each participant, there can be up to 100 sentences, I would *like* to compare them based on 2 groups e.g. gender. But the scores aren't independent, i.e. they come from the same participant - so most tests are out of the window.

I was just wondering if there was some fancy way around this?

(I can go on and complete the study, it would be nice to use the full data set though, not just averages of their scores)

hiya sneaks
do u think it is possible to test according to the sentences/theme?

oh ya you remember how good I am at statistics HAHAHAAHAHAHA
I'm just making a guess here :-) :-) :-) hehehehe

what if you coded the data according to the themes. but the problem of not independent scores is still there, perhaps maybe only a simple frequencies' analysis to say such-and-such sentences came from so-and-so.

have you also done intercoder reliability? that is some result too if there's not much tests you can do after that :-) :-) :-)

love
satchi

yep I've done interrater reliability and got descriptives from the BIG data set. Will have to run analysis on the reduced data i.e. the means unless my hubby can find a magical test - he's just got in from work so am going to quiz him!

======= Date Modified 20 May 2010 18:26:09 =======
double post

I'm not sure I fully understand what you ahve in the way of data but I'm leaning towards chi squared on the frequencies of each coding per participant maybe. What did your hubby say?

I might sound dead thick here, but could you not use multiple response analysis in SPSS: http://books.google.co.uk/books?id=t3-JPGGaNO8C&pg=PA104&lpg=PA104&dq=multiple+response+sets+groups+multiple+cases+spss&source=bl&ots=ULPCIJiSAh&sig=fsjkHMlhJDDlOnlaQEC0pychyCY&hl=en&ei=WK_1S4C6LaT20wSHq_DpBw&sa=X&oi=book_result&ct=result&resnum=4&ved=0CB0Q6AEwAzgK#v=onepage&q&f=false

Wow thanks Wal - I've never actually heard of that (and I'm the stats guru in my department :$ ) I will take a look through.

My hubby suggested dividing the groups differently - I hadn't thought of that. So do repeated measures (as my participants were paired across groups) as well as independent t-tests. I'm going to spend ALL day today sorting this out.

:-(

Right, I think I can't use multiple response sets because my data is like I've asked my participants whether they eat choc,crisps, sweets, mulitple times so my data is like..

Crisps Choc Sweets
p1 0 1 1
p1 1 0 0

(continue for 90 more goes, then on to p2) - so I have found out its called 'Clustered responses' and have found a very difficult mathematical paper that allowed you to run regression it - which may be useful. Going to get hubs to look at it later as he is a mathematician so should be able to make sense of some of it.

I'm posting, just in case someone has the same question in years to come and gets annoyed that the answer isn't here.

I *think* the answer is mixed multi-level modelling. But I have not got my copy of andy field on me, I will confirm tomorrow!