Work measurement is determining how long it takes to do a job. Managing human resources requires managers to know how much work employees can do during a specific period. Otherwise they cannot plan production schedules or output. Without a good idea of how long it takes to do a job, a company will not know if it can meet customer expectations for delivery or service time. Despite the unpopularity of wage-incentive systems among some TQM proponents, they are still widely used in the United States, and work measurement is required to set the output standards on which incentive rates are based. These wage rates determine the cost of a product or service.
Work measurement has also seen a revival within the ever-growing service sector. Services tend to be labor intensive, and service jobs, especially clerical ones, are often repetitive. For example, sorting mail in a postal service, processing income tax returns in the IRS, making hamburgers at McDonald's, and inputting data from insurance forms in a computer at Prudential are all repetitive service jobs that can be measured, and standards can be set for output and wages. As a result work measurement is still an important aspect of operations management for many companies.
The traditional means for determining an estimate of the time to do a job has been the time study, in which a stopwatch is used to time the individual elements of a job. These elemental times are summed to get a time estimate for a job and then adjusted by a performance rating of the worker and an allowance factor for unavoidable delays, resulting in a standard time. The standard time is the time required by an "average" worker to perform a job once under normal circumstances and conditions.
Work measurement and time study were introduced by Frederick W. Taylor in the late 1880s and 1890s. One of his objectives was to determine a "fair" method of job performance evaluation and payment, which at that time was frequently a matter of contention between management and labor. The basic form of wage payment was an incentive piece-rate system, in which workers were paid a wage rate per unit of output instead of an hourly wage rate; the more workers produced, the more they earned. The problem with this system at the time was that there was no way to determine a "normal," or "fair," rate of output. Management wanted the normal rate high, labor wanted it low. Since management made pay decisions, the piece rate was usually "tight," making it hard for the worker to make the expected, or fair, output rate. Thus, workers earned less. This was the scenario in which Taylor introduced his time study approach to develop an equitable piece-rate wage system based on fair standard job times.
The stopwatch time study approach for work measurement was popular and widespread into the 1970s. Many union contracts in the automotive, textile, and other manufacturing industries for virtually every production job in a company were based almost entirely on standard times developed from time studies. However, the basic principle underlying an incentive wage system is that pay is the sole motivation for work. We have pointed out earlier in this chapter that this principle has been disproved. In fact, in recent years incentive wage systems have been shown to inhibit quality improvement.
However, performance evaluation represents only one use for time study and work measurement. It is still widely used for planning purposes in order to predict the level of output a company might achieve in the future.
The result of a time study is a standard time for performing a repetitious job once. Time study is a statistical technique that is accurate for jobs that are highly repetitive.
The basic steps in a time study are:
The Metro Food Services Company delivers fresh sandwiches each morning to vending machines throughout the city. Workers work through the night to prepare the sandwiches for morning delivery. A worker normally makes several kinds of sandwiches. A time study for a worker making ham and cheese sandwiches is shown in Figure 8.5. Notice that each element has two readings. Row t includes the individual elemental times, whereas the R row contains a cumulative (running) clock reading recorded going down the column. In this case the individual elemental times are determined by subtracting the cumulative times between sequential readings.
In Figure 8.5 the average element times are first computed as
For element 1 the average time is
The normal elemental times are computed by adjusting the average time, t, by the performance rating factor, RF. For element 1 the normal time is
The normal cycle time, NT, is computed by summing the normal times for all elements, which for this example is 0.387. The standard time is computed by adjusting the normal cycle time by an allowance factor,
If, for example, the company wants to know how many ham and cheese sandwiches can be produced in a 2-hour period, they could simply divide the standard time into 120 minutes:
If the Metro Food Services Company pays workers a piece rate of $0.04 per sandwich, what would an average worker be paid per hour, and what would the subject of the time study in Example 8.3 expect to be paid?
The average worker would produce the following number of sandwiches in an hour:
The hourly wage rate would thus average
Alternatively, the worker from Example 8.3 would produce at the average cycle time not adjusted by the rating factor, or 0.361 minutes. Adjusting this time by the allowance time results in a time of
This worker could be expected to produce the following number of sandwiches per hour:
The average hourly wage rate for this worker would be
or $0.40 more per hour.
In Example 8.3 the time study was conducted for ten cycles. However, was this sufficient for us to have confidence that the standard time was accurate? The time study is actually a statistical sample distribution, where the number of cycles is the sample size.
Assuming that this distribution of sample times is normally distributed (a traditional assumption for time study), we can use the following formula to determine the sample size, n, for a time study:
In Example 8.3 the Metro Food Services Company conducted a time study for 10 cycles of a job assembling ham and cheese sandwiches, which we will consider to be a sample. The average cycle time, T, for the job was 0.361 minutes, computed by dividing the total time for 10 cycles of the job, 3.61, by the number of cycles, 10. The standard deviation of the sample was 0.03 minutes. The company wants to determine the number of cycles for a time study such that it can be 95 percent confident that the average time computed from the time study is within 5 percent of the true average cycle time.
The sample size is computed using z = 1.96 for a probability of 0.95, as follows:
The time study should include 11 cycles to be 95 percent confident that the time-study average job cycle time is within 5 percent of the true average job cycle time. The 10 cycles that were used in our time study were just about right.
Workers often do not like to be the subject of a time study and will not cooperate, and rating workers can be a difficult, subjective task. Time studies can also be time-consuming and costly. As an alternative, many companies have accumulated large files of time study data over time for elements common to many jobs throughout their organization. Instead of conducting an actual time study, these elemental standard time files can be accessed to derive the standard time, or the elemental times in the files can be used in conjunction with current time study data, reducing the time and cost required for the study.
However, it can be difficult to put together a standard time without the benefit of a time study. The engineer/technician is left wondering if anything was left out or if the environment or job conditions have changed enough since the data were collected to alter the original elemental times. Also, the individuals developing the current standard time must have a great deal of confidence in their predecessor's abilities and competence.
The use of elemental standard times from company files is one way to construct a standard time without a time study, or before a task or job is even in effect yet. Another approach for developing time standards without a time study is to use a system of predetermined motion times. A predetermined motion time system provides normal times for basic, generic micromotions, such as reach, grasp, move, position, and release, that are common to many jobs. These basic motion times have been developed in a laboratory-type environment from studies of workers across a variety of industries and, in some cases, from motion pictures of workers.
To develop a standard time using predetermined motion times, a job must be broken down into its basic micromotions. Then the appropriate motion time is selected from a set of tables (or a computerized database), taking into account job conditions such as the weight of an object moved and the distance it might be moved. The standard time is determined by summing all the motion times. As might be suspected, even a very short job can have many motions; a job of only 1 minute can have more than 100 basic motions.
Several systems of predetermined motion times exist, the two most well known being methods time measurement (MTM) and basic motion time study (BMT). Table 8.3 provides an example of an MTM table for the motion move. The motion times are measured in time measurement units, or TMUs, where one TMU equals 0.0006 minutes and 100,000 TMUs equal one hour.
There are several advantages of using a predetermined motion time system. It enables a standard time to be developed for a new job before the job is even part of the production process. Worker cooperation and compliance are not required, and the workplace is not disrupted. Performance ratings are included in the motion times, eliminating this subjective part of developing standard times.
There are also disadvantages with a predetermined motion time system. It ignores the job context within which a single motion takes place--that is, where each motion is considered independently of all others. What the hand comes from doing when it reaches for an object may affect the motion time as well as the overall sequence of motion. Also, although predetermined motion times are generally determined from a broad sample of workers across several industries, they may not reflect the skill level, training, or abilities of workers in a specific company.
Work sampling is a technique for determining the proportion of time a worker or machine spends on various activities. The procedure for work sampling is to make brief, random observations of a worker or machine over a period of time and record the activity in which they are involved. An estimate of the proportion of time that is being spent on an activity is determined by dividing the number of observations recorded for that activity by the total number of observations. A work sample can indicate the proportion of time a worker is busy or idle or performing a task or how frequently a machine is idle or in use. A secretary's work can be sampled to determine what portion of the day is spent word processing, answering the telephone, filing, and so on. It also can be used to determine the allowance factor that was used to calculate the standard time for a time study. (Recall that the allowance factor was a percentage of time reflecting worker delays and idle time for machine breakdowns, personal needs, and so on.)
The primary uses of work sampling are to determine ratio delay, which is the percentage of time a worker or machine is delayed or idle, and to analyze jobs that have nonrepetitive tasks--for example, a secretary, a nurse, or a police officer. The information from a work sample in the form of the percentage of time spent on each job activity or task can be useful in designing or redesigning jobs, developing job descriptions, and determining the level of work output that can be expected from a worker for use in planning.
The steps in work sampling are summarized as follows:
The Northern Lights Company is a retail catalog operation specializing in outdoor clothing. The company has a pool of 28 telephone operators to take catalog orders during the business hours of 9:00 A.M. to 5:00 P.M. (The company uses a smaller pool of operators for the remaining 16 off-peak hours.) The company has recently been experiencing a larger number of lost calls because operators are busy and suspects it is because the operators are spending around 30 percent of their time describing products to customers. The company believes that if operators knew more about the products instead of having to pull up a description screen on the computer each time a customer asked a question about a product, they could save a lot of operator time, so it is thinking about instituting a product awareness training program. However, first the company wants to perform a work sampling study to determine the proportion of time operators are answering product-related questions. The company wants the proportion of this activity to be accurate within ±2 percent, with a 95 percent degree of confidence.
First determine the number of observations to take, as follows:
This is a large number of observations, however; since there are 28 operators, only 2,017/28, or 72, visits to observe the operators need to be taken. Actually, the observations could be made by picking up a one-way phone line to listen in on the operator-customer conversation. The "conversation" schedule was set up using a two-digit random number table (similar to Table S12.2). The random numbers are the minutes between each observation, and since the random numbers ranged from 00 to 99, the average time between observations is about 50 minutes. The study was expected to take about 8 days (with slightly more than 9 observations per day).
In fact, after 10 observation trips and a total of 280 observations, the portion of time the operators spent answering the customers' product-related questions was 38 percent, so the random sample size was recomputed:
This number of observations is 246 more than originally computed, or almost 9 additional observation trips, resulting in a total of 81. (As noted previously, it is beneficial periodically to recompute the sample size based on preliminary results in order to ensure that the final result will reflect the degree of accuracy and confidence originally specified.)
Work sampling is an easier, cheaper, and quicker approach to work measurement than time study. It tends to be less disruptive of the workplace and less annoying to workers, because it requires much less time to sample than time study. Also, the "symbolic" stopwatch is absent. A disadvantage is the large number of observations needed to obtain an accurate sample estimate, sometimes requiring the study to span several days or weeks.
A learning curve, or improvement curve, is a graph that reflects the fact that as workers repeat their tasks, they will improve performance. The learning curve effect was introduced in 1936 in an article in the Journal of Aeronautical Sciences by T. P. Wright, who described how the direct labor cost for producing airplanes decreased as the number of planes produced increased. This observation and the rate of improvement were found to be strikingly consistent across a number of airplane manufacturers. The premise of the learning curve is that improvement occurs because workers learn how to do a job better as they produce more and more units. However, it is generally recognized that other production-related factors also improve performance over time, such as methods analysis and improvement, job redesign, retooling, and worker motivation.
Figure 8.6 illustrates the general relationship defined by the learning curve; as the number of cumulative units produced increases, the labor time per unit decreases. Specifically, the learning curve reflects the fact that each time the number of units produced doubles, the processing time per unit decreases by a constant percentage.
The decrease in processing time per unit as production doubles will normally range from 10 to 20 percent. The convention is to describe a learning curve in terms of 1, or 100 percent, minus the percentage rate of improvement. For example, an 80 percent learning curve describes an improvement rate of 20 percent each time production doubles, a 90 percent learning curve indicates a 10 percent improvement rate, and so forth.
The learning curve in Figure 8.6 is similar to an exponential distribution. The corresponding learning curve formula for computing the time required for the nth unit produced is
Paulette Taylor and Maureen Becker, two undergraduates at State University, produce customized personal computer systems at night in their apartment (hence the name of their enterprise, PM Computer Services). They shop around and purchase cheap components and then put together generic personal computers, which have various special features, for faculty, students, and local businesses. Each time they get an order, it takes them a while to assemble the first unit, but they learn as they go, and they reduce the assembly time as they produce more units. They have recently received their biggest order to date from the statistics department at State for 36 customized personal computers. It is near the end of the university's fiscal year, and the computers are needed quickly to charge them on this year's budget. Paulette and Maureen assembled the first unit as a trial and found that it took them 18 hours of direct labor. To determine if they can fill the order in the time allotted, they want to apply the learning curve effect to determine how much time the 9th, 18th, and 36th units will require to assemble. Based on past experience they believe their learning curve is 80 percent.
The time required for the 9th unit is computed using the learning curve formula:
The times required for the 18th and 36th units are computed similarly:
Learning curves are useful for measuring work improvement for nonrepetitive, complex jobs requiring a long time to complete, such as building airplanes. For short, repetitive, and routine jobs, there may be little relative improvement, and it may occur in a brief time span during the first (of many) job repetitions. For that reason, learning curves can have limited use for mass production and assembly line type jobs. A learning curve for this type of operation sometimes achieves any improvement early in the process and then flattens out and shows virtually no improvement, as reflected in Figure 8.7.
Learning curves help managers project labor and budgeting requirements in order to develop production scheduling plans. Knowing how many production labor hours will be required over time can enable managers to determine the number of workers to hire. Also, knowing how many labor hours will eventually be required for a product can help managers make overall product cost estimates to use in bidding for jobs and later for determining the product selling price. However, product or other changes during the production process can negate the learning curve effect.
The Excel spreadsheet for Example 8.7 is shown in Exhibit 8.2. Notice that cell C8 is highlighted and the learning curve formula for computing the time required for the 9th unit is shown on the toolbar at the top of the screen. This formula includes the learning curve coefficient in cell D4, the time required for the first unit produced in cell D3, and the target unit in B8.
Excel OM also has a learning curve macro that can be accessed from the "OM" menu on the toolbar at the top of the spreadsheet after the Excel OM program has been loaded. The Excel OM spreadsheet for Example 8.7 is shown in Exhibit 8.3. Notice that the learning curve time is computed for each of the first 36 units in cells B16:B51. However, since all these cell values could not be shown in Exhibit 8.3, we "froze" the spreadsheet at row 27 so that after the 12th unit is shown the next unit shown is the 36th. (The spreadsheet is frozen by clicking on "Window" at the top of the spreadsheet and then selecting "Freeze" from the resulting menu.)
POM for Windows, used elsewhere in this text, also has a module for learning curves that will provide a learning curve graph.
8-23. Describe the steps involved in conducting a time study, and discuss any difficulties you might envision at various steps.
8-24. A traditional performance rating benchmark (or guideline) for "normal" effort, or speed, is dealing 52 cards into four piles, forming a square with each pile 1 foot apart, in 0.50 minute. Conduct an experiment with one or more fellow students in which one deals the cards and the others rate the dealer's performance, and then compare these subjective ratings with the actual time of the dealer.
8-25. What are some of the criticisms of work measurement, in general, and time study, specifically, that have caused its popularity to wane in recent years?
8-26. Compare the use of predetermined motion times for developing time standards instead of using time study methods and discuss the advantages and disadvantages.
8-27. When conducting a work sampling study, how can the number of observations required by the study be reduced?
8-28. When is work sampling a more appropriate work measurement technique than time study?
8-29. Describe the steps involved in conducting a work sample.
8-30. Select a job that you are conveniently able to observe, such as a secretary, store clerk, or custodian, and design a work sampling study for a specific job activity. Indicate how the initial estimate of the proportion of time for the activity would be determined and how the observation schedule would be developed. (However, do not conduct the actual study.)
*8-31. How do learning curves affect productivity? How should we take learning curves into account?
8-32. For what type of jobs are learning curves most useful?
8-33. What does a learning curve specifically measure?
8-34. Discuss some of the uses and limitations of learning curves.
*These exercises require a direct link to a specific Web site. Click Internet Exercises for the list of internet links for these exercises.