|
Results: Computation times
|
| ADMB-RE |
30 seconds |
| aML |
10-15 seconds |
| SAS |
50-60seconds |
| MLwiN |
10-15 seconds |
| HLM 5 |
50-60 seconds |
| MIXOR |
5-10 seconds |
Computation times for other packages taken from here.
|
|
|
Model description
The SOCATT data set is used in the comparison of different software packages for multilevel modelling, conducted by
the Centre for Multilevel Modelling.
The SOCATT data consist of the responses to a set of dichotomous items on a woman's right to have an abortion.
The outcome variable (yij) is a score constructed from these items ranging from 1 to 7, with a higher score
corresponding to stronger support for abortion. Each of 264 respondents was asked the same set of questions on four occasions,
1983-1986, and yij denotes the response from individual i in year j.
We consider one categorical covariate (religion) with 4 categories. A random intercept ordered logit model was fitted:
logit[ P(yij < s) ] = ks +
b1xi1 + b2*xi2 + b3*xi3 + ui,
ui ~ N(0,s2)
where x1i, x2i and x3i are dummy variables coding for
the different levels of the categorical covariates, and ks are threshold
parameters. A full description of the model can be found here: socatt.pdf
Results
The following table shows the parameter estimates obtained with ADMB-RE.
| t1 |
t2 |
t3 |
t4 |
t5 |
t6 |
s |
b1 |
b2 |
b3 |
| -4.13 |
-2.39 |
0.4 |
1.34 |
2.23 |
3.27 |
2.23 |
1.95 |
0.68 |
2.78 |
Results for other software packages can be obtained
here.
|