Multilevel (HLM) analysis in MIXED
Locked
Multilevel (HLM) analysis in MIXED
 
|
We have data from a quasi-experiment to analyze. For the sake of this inquiry, say we have one DV, call it Happy. We measured this DV on three occasions (T1, T2, T3). It's a field study with the measurement by voluntary survey participation. Sample size is not a problem (at least not by psychology research standards), as a total of 1750 people responded at least once. the problem is that some people responded once only, some twice, and some three times, and of course we have all possible combinations of when the people did or did not participate. In addition, some participants were in offices with one type of lighting (L1), and others had a different kind (L2). Thus, Time is a repeated measures variable and Lighting is a between-groups variable. However, not every individual participated on every time. If I were to diagram my data in a table where each S stands for a Happy score from one person (and the dashes are to preserve the table appearance), it could look like: -------------T0------------T1-----------T2 L1 -------------S1------------S1-----------S1 -------------S2------------S2------------- ---------------------------S3-----------S3 ----------------------------------------S4 ---------------------------S5------------ -------------S6--------------------------- [and so on, for a total of several hundred in L1] L2 -------------S101---------S101---------S101 -------------S102---------S102----------- --------------------------S103---------S103 ---------------------------------------S104 --------------------------S105------------- -------------S106------------------------- [and so on, for a total of several hundred in L2] I want to be able to make a powerful comparison between Lighting conditions, using data from as many participants as possible. I think I can do this with multilevel modelling, but I haven't previously used this technique, and haven't previously had such unbalanced data. First question is, am I correct in thinking that MIXED can handle this design? Assuming I am correct about that, I will be doing this analysis using SPSS MIXED but am unclear about how to organize the data file and how to write the syntax. I have extensive experience in using GLM and MANOVA to run balanced designs for both repeated measures and mixed between-within designs, but this is stumping me. For instance, in GLM, one specifies repeated measures with all the data for one case (one person) on one line; is that also true for MIXED? The alternative to a multilevel analysis would be quite messy: separate between-groups ANOVA within each survey time; or to delete the people who didn't participate all three times and do a balanced 3 (Time) x 2 (Lighting) mixed within-between ANOVA using GLM. I don't like either option, but will do that if I can't figure this out in the next week (because a project deadline looms). Thanks!
--------------------------------------------------------------------------------------------------------------------------------------------- t +1-613-993-9671 | f +1-613-954-3733 | [hidden email] Government of Canada
|
Re: Multilevel (HLM) analysis in MIXED
|
|
Jennifer -
This
probably won't fully answer your question. Yes, MIXED would be appropriate
to use. Data should be in a person-period (long) file structure.
Each observation should be on a seperate line. If a subject has 3
assessments, they will have 3 lines of data. There is a Data Restructure
Wizard that can help if your data is currently in a person-level format (1 line
for each subject).
Peter
Link
VA San
Diego Healthcare System
|
| Free forum by Nabble | Edit this page |