
Finally, I realize Multidimensional Adaptive Testing system by myself. It is the first comparison of simulations between Random Administration (RA) and Information Gain selection rules. The simulation include 6 dimensions and 9 item banks (totally 1140 items). 600 examinee samples are created for calibration, and another 200 are for verification.Record the milestone of my work myself, and keep going ahead!!!

The experiment result by repeating Dr. Wang W.C.'s simulation on MAT.
Progress in MAT
Dr. Wang's article "MAT Implementation" really help me a lot. I wrote programs to verify the simulation done by him, and the results are listed as follow.
This is the correlation matrix among the 6 dimensions of latent traits specified in advance
1.0000 0.8000 0.8000 0.3000 0.3000 -0.4000
0.8000 1.0000 0.7000 0.2000 0.2000 -0.3000
0.8000 0.7000 1.0000 0.1000 0.2000 -0.2000
0.3000 0.2000 0.1000 1.0000 0.7000 -0.2000
0.3000 0.2000 0.2000 0.7000 1.0000 -0.2000
-0.4000 -0.3000 -0.2000 -0.2000 -0.2000 1.0000
I create 1000 examinees whose ability distribution and correlation satisfy the matrix above; also, I create 9 tests as Wang (160, 160, 160, 20, 20, 20, 200, 200, 200). Then, I produce response vectors by Segall (1996)'s way, and 1000 x 6 data samples be gotten.
Correlation coefficients are calculated based on the 1000 x 6 samples to verify if they follow the matrix listed above. The result is shown below:
1.0000 0.8228 0.8296 0.2621 0.2925 -0.3929
0.8228 1.0000 0.7724 0.1970 0.1985 -0.2852
0.8296 0.7724 1.0000 0.1207 0.2224 -0.2302
0.2621 0.1970 0.1207 1.0000 0.6769 -0.1953
0.2925 0.1985 0.2224 0.6769 1.0000 -0.2007
-0.3929 -0.2852 -0.2302 -0.1953 -0.2007 1.0000
Very good simulation! Now, I can continue my job ....
My recent work on simulation
Recently, I focus on simulation design and data analysis. By using simulation, I can produce any size of item bank and sample data for calibration and verification, which gives me chance to look at the object from several points of view. Many meaningful finds come out, and I would like to post them here after a while when I am more confidence with them.