Message/Author |
|
|
Hello, I am working on a multi-level meta analysis. The model we have has a continuous Level 1 outcome, a continuous Level 1 predictor, and a continuous Level 2 predictor. I have been able to easily fit a fixed effects model to the data in MPlus, and would like to now fit a random effects model. However, I am running into serious issues with this model, and in looking more closely at the available examples for such analyses I have noticed that they all only use Level 1 predictors. Indeed, when the Level 2 predictor is excluded, the random effects model runs just fine. I am wondering if it is possible to estimate such a model including a Level 2 predictor, and if you might be able to point me to any references regarding this type of analysis? Thank you, Sara Douglass |
|
|
You cannot use a level 2 predictor on level 1. You may be wanting a cross-level interaction. See Example 9.2 in the user's guide. |
|
|
Sorry, let me clarify: Is it possible to just include a Level 2 predictor, at Level 2, while modeling a random slope at Level 1. For instance: %WITHIN% s | x on y; %BETWEEN% s; [sw]; [x@0.0]; x@1.0; x on z; Thank you! |
|
|
I apologize for the error above - I meant [s]; not [sw]; |
|
|
The variable x must be on the WITHIN list when it is used as part of a random slope. So x cannot appear on between. See Example 9.2. |
|
|
Hello I am working on a 3 level meta-analysis. In my data, the effect sizes (= level 1) are nested under samples (= level 2) which are nested under countries (= level 3). In your handout, I found code for a two-level meta-analysis (on slide 157): https://www.statmodel.com/download/Topic9-v52%20%5BCompatibility%20Mode%5D.pdf My understanding is that in your code ID is the unique identifier for each level 2 cluster (samples in my case). I would like to apply your approach to 3 level data. In addition, I would like to test a level 3 moderator. As in your code, ID is the unique identifier for each level 2 cluster (= samples). CLUSTER is the unique identifier for each level 3 cluster (= countries): VARIABLE: Names = Cluster Id y sd A_SB; USEVARIABLE = y A_SB x; CLUSTER = Cluster Id; WITHIN = y x; BETWEEN = (Cluster) A_SB; DEFINE: y = y/sd; x = 1/sd; ANALYSIS: TYPE = THREELEVEL RANDOM; ESTIMATOR = ML; MODEL: %WITHIN% [y@0.0]; y@1.0; theta | y ON x; %BETWEEN Id% theta; %BETWEEN Cluster; theta ON A_SB; Is this correct? Thank you! |
|
|
Looks like it is on the right track. |
|
|
Dear Drs. Muthen, I am working on a 3-level meta-analysis, in which the effect sizes (level 1) are nested in samples (level 2) which are nested in countries (level 3). In our data, ID is the unique identifier for the level 2 clusters (i.e., samples) and CLUSTER is the unique identifier for the level 3 clusters (i.e, countries). We have observed predictors on level 2 (cond, task) and level 3 (cult) to be tested as moderators. Using the code below worked to get the results. VARIABLE: Names = CLUSTER ID cond task cult r vi; USEVARIABLE = cond task cult cor w2; CLUSTER = CLUSTER ID; WITHIN = cor w2; BETWEEN = (CLUSTER) cult; BETWEEN = (ID) cond task; DEFINE: w2 = SQRT(vi**(-1)); !Weight for transformation cor = w2*r; !transformed r ANALYSIS: TYPE = THREELEVEL RANDOM; ESTIMATOR = ML; MODEL: %WITHIN% [cor@0.0]; cor@1.0; f | cor ON w2; %BETWEEN Id% f ON cond task; %BETWEEN CLUSTER; f ON cult; However, we would also like to test cross-level interaction effects between level 2 and level 3 variables. Is this generally possible and, if yes, how can I implement this in the model? Many thanks, Matthias |
|
|
Yes, this cross-level interaction is possible. For the Between id level, you can add random slopes s1 and s2 s1 | f on cond; s2 | f on task; and then regress s1, s2 on a variable at the Between Cluster level. |
|
|
Many thanks for your quick response! This works nicely. |
|
Back to top |