Analyses - Confused
Locked
 
|
CONTENTS DELETED
The author has deleted this message.
|
|
|
You could do a mixed model repeated measures analysis or an analysis of covariance. Which you choose mostly depends on the correlation between pre and posttest and the degree of missingness in the data.
Paul R. Swank, Ph.D. Professor, Department of Pediatrics Medical School Adjunct Professor, Health Promotions and Behavioral Sciences School of Public Health University of Texas Health Science Center at Houston -----Original Message----- From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of StatsUser Sent: Monday, August 13, 2012 2:05 PM To: [hidden email] Subject: Analyses - Confused Hello, I am looking at whether percpetions of trust increases if you have information about someone prior to meeting them. The design looks like this: Pre-test Post-test Information about another Baseline trust trust after information No information about another Baseline trust trust after information There is one IV:information (have information of another, no information - control) and is between subjects (25 in each group). The DV is trust perceptions pre and post. The 25 participants in the information about another complete the same trust perception pre and post information; whereas a different 25 participants in the no information (control) complete the same trust perception pre and post).I am interested in knowing if having information about someone increases trust in them vs having no information. I think it makes sense to do a mixed ANOVA but am not sure, as have been reading about it being just a straight between subjects anova. Please if any one can give their two cents. Thanks, -- View this message in context: http://spssx-discussion.1045642.n5.nabble.com/Analyses-Confused-tp5714671.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 StatsUser
I would suggest that with only pre and post, RMANOVA wouldn't actually be feasible, as you don't have the middle 3rd data point to test the correlation correctly, and it would thus be mathematically equivalent to a basic t-test. The standard for pre-post data would be more of a cross-sectional approach which I call covariance model for change (It seems to go by a lot of names depending on the field). This would mean that the pre-test score would be the covariate in the model, and the post test score the DV. The predictor would be the treatment indicator, and a significant intercept for a group reflects change from the start, and a significant difference T to C reflects differential change. In this scenario, there isn't an ability to analyze the within person variation in change, but with only pre-post, there is no way to calculate that either, so I would say a mixed model isn't needed unless you have some other nesting structure. Since this is not a longitudinal change !
model, and there has been no indication of some other form of nesting, I'm not sure the value. You might argue that the treatment variable should be treated as a random effect, and then a mixed model would allow that, but why? I'd want to see a strong argument for why you think the treatment condition varies randomly by person to such an extent that you need to account for that, and how that improves your final interpretation. I would argue that doing a mixed model in which you have treatment and pre-post as variables in the model, and then the pre and post scores as the Y variable, would net you essentially the same coefficients as in the above proposed model. I say this because it was actually an issue of discussion recently where we discussed the difference between moving the intercept to the final year of an intervention to get a final net program effect, vs. doing the covariate change model in the first place, and that they were statistically equivalent. I argued that would make either just as valid, though the general feeling of the group was that only the covariate change model was valid since the longitudinal model with the reset intercept was considered pointless. 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] -----Original Message----- From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of StatsUser Sent: Monday, August 13, 2012 2:05 PM To: [hidden email] Subject: Analyses - Confused Hello, I am looking at whether percpetions of trust increases if you have information about someone prior to meeting them. The design looks like this: Pre-test Post-test Information about another Baseline trust trust after information No information about another Baseline trust trust after information There is one IV:information (have information of another, no information - control) and is between subjects (25 in each group). The DV is trust perceptions pre and post. The 25 participants in the information about another complete the same trust perception pre and post information; whereas a different 25 participants in the no information (control) complete the same trust perception pre and post).I am interested in knowing if having information about someone increases trust in them vs having no information. I think it makes sense to do a mixed ANOVA but am not sure, as have been reading about it being just a straight between subjects anova. Please if any one can give their two cents. Thanks, -- View this message in context: http://spssx-discussion.1045642.n5.nabble.com/Analyses-Confused-tp5714671.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 |
|
|
CONTENTS DELETED
The author has deleted this message.
|
|
|
An RMANOVA will actually have to be an RMANCOVA, but with only 2 data points, will effectively only be an ANCOVA. What you want is an ANCOVA. RMANOVA can only be used with 3 or more data points. SPSS will let you do it, but it's not an RMANOVA, its an ANOVA then. Treatment/Control is based on a dummy variable, and the covariate should be the pre-test (as I indicated). You could create a pre-post variable, and a treatment-control variable, and stack the data (each person has two records then), but the former approach I mentioned is accepted as a superior approach. The reason is because it allows you to associate that variable with the individual. The latter approach is literally comparing it as if it were 4 separate groups, when in fact, its only 2 groups and 2 time points.
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] -----Original Message----- From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of StatsUser Sent: Tuesday, August 14, 2012 11:09 AM To: [hidden email] Subject: Re: Analyses - Confused Thank you, those are useful points. My confusion lies in the fact that yes I can do a paired t test, but must do a split file on the IV (information 2 levels - control, experimental) each have a pre post test on trust, in order to determine if there are significant differences on pre post trust measure in the experimental group alone (and control group alone). Does it makes sense to do a series of paired t tests, split filed on the IV? Also, I am not only interested in the difference within each group (experimental and control pre post trust measure) but also across the groups (sig diff between experimental and control). Hence, I am leaning toward RMANOVA, but would like to get other perspectives/rationales. -- View this message in context: http://spssx-discussion.1045642.n5.nabble.com/Analyses-Confused-tp5714671p5714688.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 StatsUser
Repeated measures MANOVA would be the same as a univariate repeated measures model, which would be basically the same as a t test on the difference between pre and post. ANCOVA will tend to be more powerful than that unless the correlation between the pre and posttest is quite high. There are two advantages to a mixed models approach. One is you get a more varied choice of variance covariance structures. In your case, this would be using a heterogeneous variance model if the pre and post variances were different. If not, there is no advantage. The second advantage is that there is no listwise deletion of missing data. Of course, if a case is missing the pre or posttest, it can't contribute to the assessment of the difference but it can contribute to the assessment of the variances. All in all, unless the assumption of equal regressions is not met, the ANCOVA model is likely to be the best choice, all things considered, although it will drop cases with mossing data. So it wou!
ld be good to know how much missing data there is. Paul R. Swank, Ph.D. Professor, Department of Pediatrics Medical School Adjunct Professor, Health Promotions and Behavioral Sciences School of Public Health University of Texas Health Science Center at Houston -----Original Message----- From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of StatsUser Sent: Tuesday, August 14, 2012 11:09 AM To: [hidden email] Subject: Re: Analyses - Confused Thank you, those are useful points. My confusion lies in the fact that yes I can do a paired t test, but must do a split file on the IV (information 2 levels - control, experimental) each have a pre post test on trust, in order to determine if there are significant differences on pre post trust measure in the experimental group alone (and control group alone). Does it makes sense to do a series of paired t tests, split filed on the IV? Also, I am not only interested in the difference within each group (experimental and control pre post trust measure) but also across the groups (sig diff between experimental and control). Hence, I am leaning toward RMANOVA, but would like to get other perspectives/rationales. -- View this message in context: http://spssx-discussion.1045642.n5.nabble.com/Analyses-Confused-tp5714671p5714688.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 StatsUser
CONTENTS DELETED
The author has deleted this message.
|
|
|
Then you have no statistical test of the difference between treatment and control, so no, you don't want to do that.
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] -----Original Message----- From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of StatsUser Sent: Tuesday, August 14, 2012 11:30 AM To: [hidden email] Subject: Re: Analyses - Confused Thank you. I will look into the ancova. Would it not make sense to do a split file on IV, and paired t test? I am wondering if it is better to use the most parsimonious analysis. There is no missing data. -- View this message in context: http://spssx-discussion.1045642.n5.nabble.com/Analyses-Confused-tp5714671p5714689.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 StatsUser
Matthew is correct. Using split file will only tell you if the pretest and posttest are different for each group separately. A repeated measures ANOVA would tell you if there is a change over time (on average across the two groups), whether there is a difference between groups (on average across times) and whether or not the difference between pretest and posttest is the same for both groups (interaction). The ANCOVA will tell you whether the posttest scores (adjusted for the pretest) differ by group. This would be like a residualized change score.
Paul R. Swank, Ph.D. Professor, Department of Pediatrics Medical School Adjunct Professor, Health Promotions and Behavioral Sciences School of Public Health University of Texas Health Science Center at Houston -----Original Message----- From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of StatsUser Sent: Tuesday, August 14, 2012 11:30 AM To: [hidden email] Subject: Re: Analyses - Confused Thank you. I will look into the ancova. Would it not make sense to do a split file on IV, and paired t test? I am wondering if it is better to use the most parsimonious analysis. There is no missing data. -- View this message in context: http://spssx-discussion.1045642.n5.nabble.com/Analyses-Confused-tp5714671p5714689.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 |
«
Return to SPSSX Discussion
|
1 view|%1 views
| Free forum by Nabble | Edit this page |