Estimating Parameters for the BNM ModelA handful of programs are available for estimating parameters for unidimensional, polytomous models. The DOS version of MULTILOG, available from Scientific Software, is discussed in the examples that follow.It is often necessary to estimate parameters for a large number of scales or experimental groups. For that purpose, we prefer the DOS version of MULTILOG because it is very easy to modify the input files and create MS-DOS batch files to conduct multiple runs. The simplest way to estimate parameters for your data using MULTILOG is to copy your data set into the directory where your copy of MULTILOG resides. Data File Your data file must be saved in ASCII format (using Notepad or a DOS editor for example). An ID number should be provided in the data set for each respondent. To see how to prepare data for MULTILOG please click here. Running MULTILOG Running MULTILOG to estimate item and person parameters involves a multi-step process, which is described throughly in the user's manual. Because this procedure is time consuming particularly if you have many scales to calibrate, we have developed an alternative method using DOS batch files. Below is an example of a DOS batch file that estimates parameters for our example dataset, AGR1_CAL.dat. Note that this batch file requirs two additional text files with extensions *.in1 and *.in2 that must be edited by the user before running MULTILOG. These file will be described below. Batch File For Estimating Item and Person Parameters Description of the batch file The first six lines The next line The next line The next six lines perform several tasks
The last three lines estimate person parameters using input file *.in2 which has commands similar to those in *.in1. Below is a sample *.in1 file. If you are following our continuing on-line example with the agreeableness scale, what follows below is the syntax in the agr1_cal.in1 file. Additional input files for our examples are provided here. IN = individual responses recorded as vectors in the data file (one line represents one individual's responses to items 1 through k). Alternatively, one could use PA for a response pattern matrix or TA for a fixed-effects table of counts instead of a matrix of individual responses. A discussion of these alternative data files are discussed in the MULTILOG manual. If you are using the MS DOS Editor (activated by typing "edit [filename]" at the DOS prompt corresponding to the appropriate directory), the above file should look like this:  To follow the strategy outlined above, the *.in2 files need to be edited to provide person parameters from MULTILOG. Below is an example that you may download here for our on-going example with the agreeableness scale: The SIXTH LINE is blank because INFORLOG is looking for a format statement for these parameters; however, none is needed if these parameters come from another set of output from MULTILOG. In the MS DOS Editor, this file appears as follows:  Running MULTILOG Now we are finished with editing the two input files that our batch file requires. Note that your file names cited in the batch file will differ depending on your data, we have chosen to use the "agr1_cal" data set for this example. To run the batch file, ensure that it is saved in the same directory as MULTILOG. Simply type the name of that batch file at the prompt, e.g., "agr1_cal". OUTPUT FILES If everything runs correctly, the following output files are provided (note that additional files may have been created and deleted following the commands issued in the batch file): agr1_cal.log agr1_cal.lis agr1_cal.sav Agr1_cal.ls1 Agr1_cal.ls2 Agr1_cal.ten Agr1_cal.prm *.LOG This file merely provides a "log" of the commands submitted to INFORLOG and should be exactly the same as what was in the *.in2 file (an earlier *.log file was created following the submission of *.in1 to INFORLOG, but deleted before INFORLOG was run with *.in2). This file is of little significance. *.LIS This file provides details of the last (second) run of MULTILOG, the first portion is provided to assist troubleshooting (e.g., "was the data read in correctly"). The second portion, provides the values used by MULTILOG in the scoring of the individuals. Finally, the last portion of this file provides person parameters. Note that the *.lis files are renamed as *.ls1 and *.ls2, discussed later. *.SAV This file is a tab delimited (data separated by a tab) matrix of the person parameters (estimated theta, standard error, and respondent ID). *.LS1 The batch file provided renames the first *.lis file as *.ls1. This file is useful in ensuring that the data were read in and the model specified correctly. View of the agr1_cal.ls1 file in the MS DOS Editor:  The *.ls1 file also provides a report of the estimation procedure with the estimated item parameters. View of the agr1_cal.ls1 file in the MS DOS Editor:  Note that it is important to check that the estimation procedure converged (i.e., the largest change in any parameter in the last cycle was smaller than the convergence criterion). If convergence was not successful, the user may wish to increase the number of cycles permitted, specify a prior distribution, or eliminate problematic items (as identified by CTT). In this case, the estimation procedure did not converge and a second run must be completed with an increase in the number of cycles permitted (e.g., agr1_bn2.in1). These parameters can be plotted with the MODFIT program (explained later in this tutorial). Remember, MULTILOG's estimates for aik and cik contain a scaling factor of 1.7. Therefore, the user must divide the aik and cik parameters by 1.7, then compute bik by dividing the new cik by the new aik and multiplying by -1. Below are the plots for the Agreeableness scale.  Plots of Agreeableness items Previous Next Below are the plots for the Conscientiousness scale. 
Plots of Conscientiousness items Previous Next |