CODAP Task Clustering

Last updated Jan 10,1997

The traditional application of hierarchical clustering in CODAP is to group together individuals who perform similar tasks for similar amounts of time. However, the capability also exists to group the most similar tasks together to form task modules. To cluster tasks, the Case Data file must be "transposed". That is, the file must be written in terms of tasks rather than cases. The program that does this is called XPOSE. The file created by XPOSE, called a Task Data file, can be input to OVRLAP, which computes the similarity matrix, based on do-don't or time spent overlap. GROUP can then create the task clustering solution.

As with case clustering, the presentation sequence can be enhanced by MPATH, and DIAGRM is used to graphically display the results of the clustering. The "KPATH" numbers for tasks are referred to as "TPATH" numbers. While a set of people is called a "group", a set of tasks is called a "module".

There are several programs available to help analyze task modules from a clustering solution. This analysis process is called module typing. Although module typing is a separate section later in this paper (see Module Typing), the following programs are also discussed here. The task clustering analog to the automatic stage selection in case clustering (JOBTYP) has recently been developed and is called MODTYP. To identify the task clusters to be used as task modules, MODTYP uses measures of within-module homogeneity (based on the average co-performance of module tasks with each other), between-module discrimination (based on the average difference in co-performance of tasks within a module and the co-performance of these tasks with the complementary module containing all other tasks), module size, and drop in "average between" in successive clusterings of task groups. For all task clusters selected, the program computes the average co-performance values of every task with the cluster. It then generates a TASSET run for analyzing the selected clusters.

For stages not selected by MODTYP but requested by a job analyst, a co-performance factor can be computed by STGFAC for the specified stages of the task clustering. TASSET can then be used to compare and analyze the task modules.

In a manner similar to "affirmation" techniques or non-hierarchical clustering, TASSET will identify tasks not assigned to a target task module which have a co-performance value greater than or equal to the lowest co-performance value of any task currently in the target task module. Although TASSET is as useful in task clustering as CORTAS is in case clustering, the two programs are not analogous. Software development plans call for the development of a CORTAS analog in task clustering (CORCAS) and for a TASSET analog in case clustering (CASSET).

If it is desired to view task factors on a PRTFAC in clustered order, it must be remembered that PRTFAC does not accept presentation sequences as does PRTVAR. Instead, the user can create a task factor containing the KPATH (or in this case, TPATH) numbers by using the program CLUSMAP. This factor can then be used as a sort factor in PRTFAC, thereby accomplishing the desired result.

Back to the CODAP home page