chi square.. i think?
Locked
 
|
hi all greetings from Ireland!
doing my thesis at the moment and im basically wondering if im using the right stats. so i have a situation where participants watch a video of a gameshow and then are asked if they would split or steal the jackpot with the opposing contestant. this is repeated 2 more times with a different video each time and at the end of each they are asked would they split or steal the jackpot from the opposing contestant. so i have V1 V2 V3 with split/steal in each, i'm looking to know how to compare if the difference between the number of people who split in v1 compared to v2, is significant, (and likewise v2-v3, and v1-v3) this is done by chi squares, as far as i can figure out ? could someone please explain how to run this on spss? finally, i have people grouped in high/low competitiveness and i want to do the same again, checking if the amount of high comp people who chose to split is significantly different between video 1,2 and 3. how do i do this on spss? any help would be so greatly appreciated thanks a lot |
Re: chi square.. i think?
|
|
Hi "pleasehelp" (it's good to have a name to be addressed, you know).
If the participants are the same for the different videos, then you have K related samples with a binary (yes/no) response: Google (or Bing) for Cochran & McNemar tests. HTH, Marta GG El 01/03/2012 19:53, pleasehelp escribió: > hi all greetings from Ireland! > > doing my thesis at the moment and im basically wondering if im using the > right stats. > > so i have a situation where participants watch a video of a gameshow and > then are asked if they would split or steal the jackpot with the opposing > contestant. > this is repeated 2 more times with a different video each time and at the > end of each they are asked would they split or steal the jackpot from the > opposing contestant. > > so i have V1 V2 V3 with split/steal in each, i'm looking to know how to > compare if the difference between the number of people who split in v1 > compared to v2, is significant, (and likewise v2-v3, and v1-v3) > > this is done by chi squares, as far as i can figure out ? could someone > please explain how to run this on spss? > > finally, i have people grouped in high/low competitiveness and i want to do > the same again, checking if the amount of high comp people who chose to > split is significantly different between video 1,2 and 3. > > how do i do this on spss? > > any help would be so greatly appreciated > > thanks a lot > > -- > View this message in context: http://spssx-discussion.1045642.n5.nabble.com/chi-square-i-think-tp5528795p5528795.html > Sent from the SPSSX Discussion mailing list archive at Nabble.com. > > ===================== > To manage your subscription to SPSSX-L, send a message to > [hidden email] (not to SPSSX-L), with no body text except the > command. To leave the list, send the command > SIGNOFF SPSSX-L > For a list of commands to manage subscriptions, send the command > INFO REFCARD > ===================== To manage your subscription to SPSSX-L, send a message to [hidden email] (not to SPSSX-L), with no body text except the command. To leave the list, send the command SIGNOFF SPSSX-L For a list of commands to manage subscriptions, send the command INFO REFCARD |
|
|
In reply to this post by pleasehelp
Marta's suggestion is certainly one way to attack this problem. Another approach would be through regression. See my message below for details:
Ok. I've tried to simulate data which represent the OP's study. For those interested, I'm a bit more detailed in how the main effect coefficients are calculated in the presence of an interaction term in the context of logistic regression. Anyway, I suggest that the OP consider employing a GEE-generalized linear model since he/she has repeated measures. I assumed an exchangeable covariance matrix given my simulation; however, an unstructured covariance matrix may better suit the OP's data. The EMEANS statements should begin to answer some of the OPs RQs, but keep in mind that the estimates are in log odds. (I do wonder if the OP randomized the order in which the videos were presented?--I'm not touching that issue for the moment).
Here you go:
set seed 98723454.
new file.
inp pro. comp video = -99.
comp group = -99. comp logit_g0v1 = -99. comp logit_g0v2 = -99. comp logit_g0v3 = -99. comp logit_g1v1 = -99. comp logit_g1v2 = -99. comp logit_g1v3 = -99. comp b0 = -99. comp b1 = -99. comp b2 = -99. comp b3 = -99. comp b4 = -99. comp b5 = -99. comp rand_eff = -99. comp subject = -99. leave video to subject.
loop subject= 1 to 5000.
comp logit_g0v1 = ln(.10 / (1 -.10)). comp logit_g0v2 = ln(.35 / (1 -.35)). comp logit_g0v3 = ln(.25 / (1 -.25)). comp logit_g1v1 = ln(.20 / (1 -.20)). comp logit_g1v2 = ln(.75 / (1 -.75)). comp logit_g1v3 = ln(.40 / (1 -.40)). comp b0 = logit_g0v1.
comp b1 = logit_g1v1 - logit_g0v1. comp b2 = logit_g0v3 - logit_g0v1. comp b3 = logit_g0v2 - logit_g0v1. comp b4 = (logit_g1v3 - logit_g0v3) - (logit_g1v1 - logit_g0v1). comp b5 = (logit_g1v2 - logit_g0v2) - (logit_g1v1 - logit_g0v1). comp group = rv.bernoulli(.50).
comp rand_eff = sqrt(.30)*rv.normal(0,1) loop video = 1 to 3. comp eta = b0 + b1*(group=1) + b2*(video=3) + b3*(video=2) + b4*(group=1)*(video=3) + b5*(group=1)*(video=2) + rand_eff. comp p = 1 / (1+exp(-eta)). comp y = rv.bernoulli(p). end case.
end loop. end loop. end file. end inp pro. exe. delete variables logit_g0v1 logit_g0v2 logit_g0v3 logit_g1v1 logit_g1v2 logit_g1v3 b0 b1 b2 b3 b4 b5 eta rand_eff p.
* Generalized Estimating Equations.
GENLIN y (REFERENCE=FIRST) BY group video (ORDER=DESCENDING) /MODEL group video group*video INTERCEPT=YES DISTRIBUTION=BINOMIAL LINK=LOGIT /EMMEANS TABLES=group SCALE=TRANSFORMED COMPARE=video CONTRAST=PAIRWISE PADJUST=LSD /EMMEANS TABLES=video SCALE=TRANSFORMED COMPARE=video CONTRAST=PAIRWISE PADJUST=LSD
/EMMEANS TABLES=group*video SCALE=TRANSFORMED COMPARE=group*video CONTRAST=PAIRWISE PADJUST=LSD /REPEATED SUBJECT=subject WITHINSUBJECT=video SORT=YES CORRTYPE=EXCHANGEABLE ADJUSTCORR=YES COVB=ROBUST MAXITERATIONS=100 PCONVERGE=1e-006(ABSOLUTE) UPDATECORR=1 /PRINT CPS DESCRIPTIVES MODELINFO FIT SUMMARY SOLUTION (EXPONENTIATED). On Thu, Mar 1, 2012 at 1:53 PM, pleasehelp <[hidden email]> wrote: hi all greetings from Ireland! |
Re: chi square.. i think?
|
|
For what it’s worth, more than likely the structure of the covariance matrix in this scenario will likely not create an interpretable difference be it unstructured
or exchangeable. While I’ve read numerous books and articles giving the specifics of the difference and when/why each should be used, my own experience, and what I’ve been told in my past coursework has been that the differences are often small in many situations.
I’ve certainly found circumstances where that is not true, but those are far more rare. In fact, one of the suggestions I received from a colleague was just to try them all and go with the best model (This is actually someone who could be referred to as having
“wrote the book” on a related set of methods, and feels we often make too big a deal out of the covariance structure). It’s also worth noting that the GEE logit model may work, making certain assumptions about the data, but depending on sample size and variables available, may
offer less power to detect significant effects than the McNemar test or in the scenario mentioned, more likely Cochran’s test. If no covariates can be included in the model, I’d suggest the Cochran test is better. If covariates can be added, and these covariates
can create a level of baseline equivalence, then that may increase the power to detect differences better.
Matthew J Poes Research Data Specialist Center for Prevention Research and Development University of Illinois 510 Devonshire Dr. Champaign, IL 61820 Phone: 217-265-4576 email:
[hidden email] From: SPSSX(r) Discussion [mailto:[hidden email]]
On Behalf Of R B Marta's suggestion is certainly one way to attack this problem. Another approach would be through regression. See my message below for details: Ok. I've tried to simulate data which represent the OP's study. For those interested, I'm a bit more detailed in how the main effect coefficients are calculated in the presence of an interaction term in the context of logistic regression. Anyway,
I suggest that the OP consider employing a GEE-generalized linear model since he/she has repeated measures. I assumed an exchangeable covariance matrix given my simulation; however, an unstructured covariance matrix may better suit the OP's data. The EMEANS
statements should begin to answer some of the OPs RQs, but keep in mind that the estimates are in log odds. (I do wonder if the OP randomized the order in which the videos were presented?--I'm not touching that issue for the moment).
Here you go: set seed 98723454. new file. comp video = -99. leave video to subject. loop subject= 1 to 5000.
comp b0 = logit_g0v1. comp group = rv.bernoulli(.50).
end case. delete variables logit_g0v1 logit_g0v2 logit_g0v3 logit_g1v1 logit_g1v2 logit_g1v3 b0 b1 b2 b3 b4 b5 eta rand_eff p. * Generalized Estimating Equations. /EMMEANS TABLES=video SCALE=TRANSFORMED COMPARE=video CONTRAST=PAIRWISE PADJUST=LSD On Thu, Mar 1, 2012 at 1:53 PM, pleasehelp <[hidden email]> wrote: hi all greetings from Ireland! |
| Free forum by Nabble | Edit this page |