Multilevel linear Model (hierachical linear models) - repeated?
Locked
 
|
Hi,
based on the study Miller, G., Tybur, J. M., & Jordan, D. B. (2007). Evolution and Human Behavior, 28, 375–381. and Andy Fields explanation (Discovering SPSS; http://www.google.de/url?sa=t&rct=j&q=&esrc=s&source=web&cd=1&cad=rja&uact=8&ved=0CCUQFjAA&url=http%3A%2F%2Fwww.sagepub.com%2Ffield4e%2Fstudy%2Flabcoatleni%2Fchapter20.pdf&ei=Io7oU6u0KPHS4QSvr4HgCg&usg=AFQjCNF8NjwWAfHgwAx0An7HmiMg5AljlQ&bvm=bv.72676100,d.d2k I tried to run a multilevel linear model for my own analysis, but I am not sure if it is correct. I have a 2x2 within subject repeated measurement design. Dependent variable is money bid. Factors are A and B with 2 levels each. Participants see 25 pictures in each factor level (all in all 100 pictures, 25 pictures in A1B1, 25 in A2B1 and so on). Data are balanced, all participants see 25 pictures in each factor level. Upper level should be subject ID, dependent variable is amount of money bid. I tried to follow the example of Andy Field, with Analyze--mixed models linear. subjects: Subject ID; repeated: none (is that correct???) and dependent variable bid with factors A and B. Fixed effects were added (main effect A, main effect B, interaction AxB), ID is added as a random effect (only intercept, because of the assumption that subjects vary in their amount of bid, some bid always little, some bid always higher). Is this procedure OK or is there something wrong? I get dfs of more than 3000 (but I only have 33 subjects)... or do I have to somehow include somehting in repeated? If so, how? And how do I get values for each factor level (i.e., values for A1B1, A2B1, A1B2, A2B2) in the table "estimates for fixed effects"? Thanks a lot! Laura |
Re: Multilevel linear Model (hierachical linear models) - repeated?
|
|
You're running from the menu it sounds like. Somewhere in there is a button for controlling what is printed. Click on that button and select "Solution". DFs of 3000+ (nearer 3300, I'd bet) sound right because you have 4 condition, 25 cases per condition and 33 subjects. Your data file has 3300 records. In addition to a random intercept, it would be useful to investigate whether the slopes differ across persons.
Your current syntax statement ought to be Mixed y by a b/fixed a b a*b/print=solution/Random=intercept | subject(id) covtype(id). You, of course, may have other options specified in the print subcommand. Ryan Black has posted multiple times on using mixed to analyze repeated measures data. Look at the archives for posts from him. Also note example 9 in the documentation. Gene Maguin -----Original Message----- From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of LauraE Sent: Monday, August 11, 2014 5:51 AM To: [hidden email] Subject: Multilevel linear Model (hierachical linear models) - repeated? Hi, based on the study Miller, G., Tybur, J. M., & Jordan, D. B. (2007). Evolution and Human Behavior, 28, 375–381. and Andy Fields explanation (Discovering SPSS; http://www.google.de/url?sa=t&rct=j&q=&esrc=s&source=web&cd=1&cad=rja&uact=8&ved=0CCUQFjAA&url=http%3A%2F%2Fwww.sagepub.com%2Ffield4e%2Fstudy%2Flabcoatleni%2Fchapter20.pdf&ei=Io7oU6u0KPHS4QSvr4HgCg&usg=AFQjCNF8NjwWAfHgwAx0An7HmiMg5AljlQ&bvm=bv.72676100,d.d2k I tried to run a multilevel linear model for my own analysis, but I am not sure if it is correct. I have a 2x2 within subject repeated measurement design. Dependent variable is money bid. Factors are A and B with 2 levels each. Participants see 25 pictures in each factor level (all in all 100 pictures, 25 pictures in A1B1, 25 in A2B1 and so on). Data are balanced, all participants see 25 pictures in each factor level. Upper level should be subject ID, dependent variable is amount of money bid. I tried to follow the example of Andy Field, with Analyze--mixed models linear. subjects: Subject ID; repeated: none (is that correct???) and dependent variable bid with factors A and B. Fixed effects were added (main effect A, main effect B, interaction AxB), ID is added as a random effect (only intercept, because of the assumption that subjects vary in their amount of bid, some bid always little, some bid always higher). Is this procedure OK or is there something wrong? I get dfs of more than 3000 (but I only have 33 subjects)... or do I have to somehow include somehting in repeated? If so, how? And how do I get values for each factor level (i.e., values for A1B1, A2B1, A1B2, A2B2) in the table "estimates for fixed effects"? Thanks a lot! Laura -- View this message in context: http://spssx-discussion.1045642.n5.nabble.com/Multilevel-linear-Model-hierachical-linear-models-repeated-tp5726927.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 LauraE
Thanks for your post!
But I still have the question, whether it is OK to run this analysis without specifying repeated (that somehow doesnt work, with my format at least) since I have 4 conditions with 25 measurements each. Do I have to specify that somehow, that there are 25 entries per condition? For the "post-hoc" tests, I do not have to do any multiple comparisons adjustments (like Bonferoni, I get the same results as for LSD since i only have a 2 by 2 design, correct)? Thanks a lot for your help! Laura |
Re: Multilevel linear Model (hierachical linear models) - repeated?
|
|
I looked at the Andy Field explanation, which I assume you are using as a model. His example analysis does not use the repeated subcommand. If you intend to follow his example analysis, you won't use a repeated subcommand either. I am confused as to why you think you need to have a repeated subcommand. Please explain.
Simple question: Do you have your data in long format, 3300 records total? >> Do I have to specify that somehow, that there are 25 entries per condition? No. the combination of person id and the values of factors A and B will define the 25 entries per condition. >>For the "post-hoc" tests, I do not have to do any multiple comparisons adjustments (like Bonferoni, I get the same results as for LSD since i only have a 2 by 2 design, correct)? Tell us about what you want to do with the post hoc tests. What comparisons do you want to make? Gene Maguin -----Original Message----- From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of LauraE Sent: Friday, August 15, 2014 6:47 AM To: [hidden email] Subject: Re: Multilevel linear Model (hierachical linear models) - repeated? Thanks for your post! But I still have the question, whether it is OK to run this analysis without specifying repeated (that somehow doesnt work, with my format at least) since I have 4 conditions with 25 measurements each. Do I have to specify that somehow, that there are 25 entries per condition? For the "post-hoc" tests, I do not have to do any multiple comparisons adjustments (like Bonferoni, I get the same results as for LSD since i only have a 2 by 2 design, correct)? Thanks a lot for your help! Laura -- View this message in context: http://spssx-discussion.1045642.n5.nabble.com/Multilevel-linear-Model-hierachical-linear-models-repeated-tp5726927p5726973.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 LauraE
Hi,
thanks a lot for your message. Yes, I am using that Andy Field model as an example, because it explains the steps I have to take very detailed. I am not familar with mixed models at all. I just know that usually, for a normal analysis without hierachy, I would use a repeated measurement ANOVA (factors: A and B, using mean values of the subjects for each condition, e.g. a value for A1B1, A2B1 per subject etc). So, now I try to use subject ID as the upper level. Every subject sees 25 pictures in each factor level: I have a 2x2 repeated design, so I show 25 pictures in A1B1, 25 pictures in A2B1, 25 in A1B2, 25 in A2B2 to each subject. The goal of the analysis is to compare A1 versus A2 and B1 versus B2 (main effects) and their interaction A x B. So I do not know if that requires a repeated subcommand or not, just from a statistical point of view. yes, I have my data in a long format, with all in all 3300 rows! I want to make the following post-hoc tests: A1B1 versus A1B2 and A2B1 versus A2B1. Thanks for your help, and sorry if I didnt specify everything in the first place. I hope its clearer now... Laura |
Re: Multilevel linear Model (hierachical linear models) - repeated?
|
|
My suggestion is to analyze your data per example 9, which is a repeated measures analysis, in the mixed command documentation. Or, restructure your data to wide format and analyze it using GLM repeated. It sounds like you are interested in picture differences between conditions. Thus your main interest is in the between-within interactions and simple effects for those.
Others may have better suggestions. Gene Maguin -----Original Message----- From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of LauraE Sent: Friday, August 15, 2014 11:26 AM To: [hidden email] Subject: Re: Multilevel linear Model (hierachical linear models) - repeated? Hi, thanks a lot for your message. Yes, I am using that Andy Field model as an example, because it explains the steps I have to take very detailed. I am not familar with mixed models at all. I just know that usually, for a normal analysis without hierachy, I would use a repeated measurement ANOVA (factors: A and B, using mean values of the subjects for each condition, e.g. a value for A1B1, A2B1 per subject etc). So, now I try to use subject ID as the upper level. Every subject sees 25 pictures in each factor level: I have a 2x2 repeated design, so I show 25 pictures in A1B1, 25 pictures in A2B1, 25 in A1B2, 25 in A2B2 to each subject. The goal of the analysis is to compare A1 versus A2 and B1 versus B2 (main effects) and their interaction A x B. So I do not know if that requires a repeated subcommand or not, just from a statistical point of view. yes, I have my data in a long format, with all in all 3300 rows! I want to make the following post-hoc tests: A1B1 versus A1B2 and A2B1 versus A2B1. Thanks for your help, and sorry if I didnt specify everything in the first place. I hope its clearer now... Laura -- View this message in context: http://spssx-discussion.1045642.n5.nabble.com/Multilevel-linear-Model-hierachical-linear-models-repeated-tp5726927p5726981.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 LauraE
Laura, I haven't followed the thread all that closely (and honestly do not have time) but if I am correct that you have two within-subjects factors, the following code might suffice: MIXED Y BY A B /FIXED = A B A*B /REPEATED = A*B | SUBJECT(ID) COVTYPE(UN) /PRINT SOLUTION. You could then employ the EMMEANS or TEST statement(s) to construct various contrasts of interest. For the linear MIXED model code above to run you would need to structure your data in vertical format as follows: ID A B Y 1 1 1 23 1 1 2 25
1 2 1 22 1 2 2 35 2 1 1 21
2 1 2 16 2 2 1 18 2 2 2 20 .
. . Best wishes,
Ryan
On Mon, Aug 11, 2014 at 5:50 AM, LauraE <[hidden email]> wrote: Hi, |
«
Return to SPSSX Discussion
|
1 view|%1 views
| Free forum by Nabble | Edit this page |