Automated work sampling with unbiased variance estimates

Abstract Work sampling is a work measurement tool that was over simplified during the slide rule era by assuming a binomial distribution for observation data totals. The binomial variance was then used to calculate the variance of the proportion estimate(s) and to determine the necessary number of observations needed to attain a certain accuracy. Now a variety of computer programs are being written to perform work sampling calculations incorporating the same quick, easy, but often biased variance estimates. When a sampling study extends over a relatively long period of time better variance estimates for the proportions are available which are essentially sample variances constructed from observation-round proportions, work shift proportions, or daily proportions. These alternate variance estimates will yield different accuracy claims and different determination of the number necessary additional study days, sometimes larger and sometimes smaller than the classic estimates. In the very practical case of unequal number of observations per round and/or per day these calculations are extremly time without a program but relatively simple in any computer language. A Fortran program is presented which calculates three different accuracies and numbers of additional study days (the classic estimates and two others), marking the one prefered based on the amount of data in the system at that point. Properly operated the program will periodically update a data file with additional study data until the operator is satisfied and terminates the work sampling study. Pace rating of work categories, when appropriate, can also be incorporated by the program to modify proportion estimates.