Many multilevel models can be estimated using mixed model procedures however the syntax will be rather different. The ml2mixed command (search ml2mixed; (see How can I use the search command to search for programs and get additional help? for more information about using search) can be used to help you convert from the multilevel model specification to the mixed syntax. The ml2mixed command will work for multilevel models with either two or three levels. For two level models ml2mixed allows for both random intercepts and random slopes at level 2 while the three level model only allows for random intercepts.
We will begin with a simple example of a 2-level unconditional model with random intercepts. The dependent variable is placed in the dep() option and the level-2 cluster variable is put into the l2id().
ml2mixed, dep(math) l2id(class)
Multilevel Model
Level 1 Model
math =
Level 2 Model -- id = class
[int] =
Stata Mixed Model -- Stata 11 notation
xtmixed math ///
|| class:
Now we can add two level-1 predictors hmwk and gender using the l1() option.
ml2mixed, dep(math) l1(hmwk gender) l2id(class)
Multilevel Model
Level 1 Model
math = hmwk gender
Level 2 Model -- id = class
[int] =
Stata Mixed Model -- Stata 11 notation
xtmixed math hmwk gender ///
|| class:
For the next example we will keep the two level-1 predictors and add a level-2 predictor of the random intercept. The level-2 intercept predictor is placed in the l2i() option. In addition, we will include the notes option which displays some hopefully helpful comments.
ml2mixed, dep(math) l1(hmwk gender) l2id(class) l2i(meanses) notes
Multilevel Model
Level 1 Model
math = hmwk gender
Level 2 Model -- id = class
[int] = meanses
Stata Mixed Model -- Stata 11 notation
xtmixed math hmwk gender meanses ///
|| class:
Stata Notes
1) Categorical predictors need the i. prefix.
2) Continuous variables in interactions need the c. prefix.
3) Use var option to get variances instead of standard deviations.
4) If outcome variable is binary use -xtmelogit- command.
5) If outcome variable is a count use -xtmepoisson- command.
Next is a model with a random slope for hmwk in addition to having a random intercept. The random slope is indicated by putting the level-1 variable inside square brackets within the l2s() option. This example also includes the sas and spss options with provide the syntax for the SAS proc mixed and for the SPSS mixed procedures respectively.
ml2mixed, dep(math) l1(hmwk) l2id(class) l2i(meanses) l2s([hmwk]) sas spss
Multilevel Model
Level 1 Model
math = hmwk
Level 2 Model -- id = class
[int] = meanses
slope[hmwk] =
Stata Mixed Model -- Stata 11 notation
xtmixed math hmwk meanses ///
|| class: hmwk , cov(unstr)
SAS Proc Mixed
proc mixed;
class class [...];
model math = hmwk meanses / solution;
random intercept hmwk / subject=class type=un;
run;
SPSS Mixed
mixed math [by...] with hmwk meanses
/print = solution
/fixed = hmwk meanses
/random = intercept hmwk | subject(class) covtype(un).
In the next model we include a predictor for the random slopes. We do this by including the variable names after the square brackets in the l2s() option. Please note that the ses#meanses term in the model is Stata 11 syntax. If you are running an earlier version of Stata you will need to create interactions using the xi or create them manually.
ml2mixed, dep(math) l1(hmwk ses) l2id(class) l2i(meanses) l2s([ses] meanses [hmwk] meanses) notes
Multilevel Model
Level 1 Model
math = hmwk ses
Level 2 Model -- id = class
[int] = meanses
slope[ses] = meanses
slope[hmwk] = meanses
Stata Mixed Model -- Stata 11 notation
xtmixed math hmwk ses meanses ses#meanses hmwk#meanses ///
|| class: ses hmwk , cov(unstr)
Stata Notes
1) Categorical predictors need the i. prefix.
2) Continuous variables in interactions need the c. prefix.
3) Use var option to get variances instead of standard deviations.
4) If outcome variable is binary use -xtmelogit- command.
5) If outcome variable is a count use -xtmepoisson- command.
The final model is a 3-level model with a level-3 variable, poverty, predicting the random intercept for hmwk. The level-3 identified is indicated with the l3id() option and the level-3 predictor variable with the l3i() option.
ml2mixed, dep(math) l1(hmwk) l2id(class) l2i(meanmath) l2s([hmwk] meanmath) l3id(school) l3i(poverty)
Multilevel Model
Level 1 Model
math = hmwk
Level 2 Model -- id = class
[int] = meanmath
slope[hmwk] = meanmath
Level 3 Model -- id = school
[int] = poverty
Stata Mixed Model -- Stata 11 notation
xtmixed math hmwk meanmath poverty hmwk#meanmath ///
|| school: || class: hmwk , cov(unstr)