Readme file
SERIES
C
 Applied
Statistics
A local-influence-based diagnostic approach to a speeded item response theory model, by
Y. Goegebeur, P. De Boeck, G. Molenberghs and G. del Pino
Journal of the Royal Statistical Society, Series
C, Applied Statistics, Volume 55
(2006), 647 - 676
Data file
-----------
simce1.txt: response profiles of 3000 examinees
Variables:
1: dependent variable, Y, coded
* 2: correct answer
* 1: wrong answer
* 0: missing value
2: item identification number (from 1 to 48)
3: examinee identification number (from 1 to 3000)
4-51: item dummy variables
SAS programs
---------------------
sasprogram_reduced_model.txt: SAS program to fit the reduced model
sasprogram_speeded_model.txt: SAS program to fit the speeded
model
FORTRAN programs
------------------------------
These programs make use of the NAG library and compute the different local
influence diagnostics.
The user needs to set the following parameters (parameter list in the
variable declaration part):
- ni: number of items,
- np: number of examinees,
- nip: number of quadrature points (max 64).
Besides the user has to supply the following information in text files:
- unit=11: a file with the response profiles of the examinees (only the responses,
stored in a single column,
grouped by examinee),
- unit=12: a file with the parameter estimates obtained from fitting the reduced
model (the order in which the
parameters need to be supplied is: beta, sigma_theta^2, mu_{xi_0}, sigma_{xi_0}^2,
c),
- unit=13: a file that contains the asymptotic covariance matrix of the latter
estimates,
- unit=14: a file that contains the correction factor L_22, see paper pg 13.
The following results are written to the file associated with unit=15:
- examinee identification number,
- local influence diagnostic,
- local influence diagnostic for particular subset of parameter vector (depends
on the program chosen).
program files:
fortran_beta.for: computes local influence for global parameter
vector and for item difficulty estimates,
fortran_sigma_theta_2.for: computes local influence for global
parameter vector and for ability variance estimate,
fortran_mu_xi_0.for: computes local influence for global parameter
vector and for mean initial propensity to omit,
fortran_sigma_xi_0_2.for: computes local influence for global
parameter vector and for variance initial propensity to omit,
fortran_c.for: computes local influence for global parameter
vector and for random guessing parameter c.
The following program computes the local influence for the global parameter
vector and the direction of maximal
curvature. The user needs to set the parameters described above
and supply the following input files (txt format):
- unit=11: a file with the response profiles of the examinees (only the responses,
stored in a single column,
grouped by examinee),
- unit=12: a file with the parameter estimates obtained from fitting the reduced
model,
- unit=13: a file that contains the asymptotic covariance matrix of the latter
estimates,
The program returns the following results
- unit=14: the direction of maximal curvature,
- unit=15: local influence diagnostic.
program file:
fortran_max_curv.for: computes local influence for global parameter
vector and the direction of maximal curvature.
Yuri Goegebeur
Department of Statistics
University of Southern Denmark
J.B. Winsløws Vej 9
Entrance B, 2nd floor
DK-5000 Odense C
Tel: +45 6550 3360
Fax: +45 6550 3345
E-mail: yuri.goegebeur@stat.sdu.dk
Web: http://www.nat.sdu.dk/users/nat%2Dsdu/yuri/
Datasets (goegebeur.zip,
size - 886KB) |
