‘Edwin Mansfield and Harold H. Wein A Managerial Application of Cost Functions by a Railroad ‘There have been numerous studies where ithe relationship between cost and output has ‘een estimated for manufacturing plants, ‘These studies fail generally into’ two cate- gories. In the first category, the chief pur- [pose of the work has been to accumulate -evidence concerning various cost curves that occupy an important place in econ- ‘omics, In the second category, the work has ‘been done primarily with a view toward ‘the direct usefulness of the relationship of. cost and output to industrial managers. This paper falls into the second category. ‘We give an illustration here of how such a relationship may be used to formulate a simple control chart for costs, a topic which should be of interest to many applied statisticians and industrial man- agers. The problem of controlling per- formance and costs is important in almost every sector of industry and trade. Our data, and hence our specific results, per- -- tain to the American railroad industry, but the statistical techniques that are used ‘may be applicable to other industries as well. FREIGHT YARDS AND COST CONTROL Freight yards differ greaily in size, layout, and type, but they have certain physical characteristics and functions in common. Physical Characteristics. AU yards contain. sets of tracks. In large yards they are generally of three types: receiving tracks where incoming freight cars are stored, classification tracks where cars are switched, and outbound tracks where cars that were situated on a classification track are stored until a locomotive hauls them away as a train, ' Functions. Freight yatds switch cars. That is, they sort incoming cars by putting them on the appropriate classification tracks, and in this way they break up incoming trains to form new trains. Most yards also deliver-and pick up cars. Engines are as- signed to deliver cars to industrial sidings and other yards and to pick them up there. Finally, many yards bill and inspect freight cars and perform such ancillary services as maintenance, repair, and storage. This case is taken ftom an article in Applied Statistics. Reprinted with permission. 151CASE FOUR The importance of freight yards to a railroad is illustrated by the fact that about one-third of its total operating costs may arise in these yards. In view of this, it is clearly important that proper control be maintained over the performance of the yards. However, this problem of main: taining adequate control is made difficult by their number and their distribution over a large area..(For example, in the railroad we studied, about 200 freight yards are scattered along approximately 12,000 miles of track.) It is virtually impossible for any team of managers to have each day a rea- sonably complete knowledge of what hap- pened at each yard. They must examine selected data concerning the performance of the yards during the day, and from these data they must somehow evaluate a yard’s performance. In evaluating performance, ‘one piece of information that is used is the costs. incurred in the yard during the day. At present, the data and techniques used by most railroads do not seem well suited for their purpose. Judging by the opinions of the railroad management with which we worked, their purpose is to detect those days when the costs at a yard are unusu- ally:high for the output produced and those seties of days when the costs are repeatedly higher than would be expected. Detection of either of these would result in an inquiry concerning the causes ofthe apparent deteri- oration of yard performance. In addition, the management is interested in detecting days when costs are unusually low or when they are repeatedly lower than would be expected. In this case there would also be an inquiry, but the intention would be to encourage the responsible factors rather than to remedy them, In this paper we discuss a control chart based on the relationship between cost and output that may be useful for these pur- poses. Before discussing the chart, it seems worthwhile to describe the measures of freight-yard output and costs that are used. The two most important services performed at a yard are switching and de- 152 livery; and it seems reasonable to use the number of “cuts” switched and the number of cars delivered during a particular period: as a measure of output. A “cut” is a group of cars that rolls as a unit on to the same... : classification track; it is often used as measure of switching output. The number of cars delivered includes both the cars ” delivered to sidings and other yards and’ those that are picked up. This output measure is not ideal, one difficulty being that it conceals considerable heteroge- neity. For example, two groups of cars. may be delivered but one may be hauled.a greater distance than the other. Some of this heterogeneity could be eliminated b3 further refinements in the output measure, but the extra complexity with regard to: data collection and computation migh result in a loss of feasibility. The costs used here include all money’ costs incurred in the yard except fixed + charges, repair costs, maintenance and stor- age costs, and vacation costs. Only money.” costs are included; the costs that may be.” imputed to car delay are not taken into account. Fixed charges are excluded, but some of the included costs are essentially. fixed in the very short run, THE CONTROL CHART ‘The control chart contains the deviation 6 actual cost from the cost that would be ex: pected on the basis of the average relation ship between cost and output. ‘These de ations are used to detect days when costs. are suspiciously high or low. The model. that underlies the chart is as follows: for a particular yard, the expected cost on the’. i'th day (C)) is assumed to be a linear function of the number of cuts switched on the i’th day (S;) and the number of cars = delivered on the i’th day (D,). * When the railroad management refer to unusually. high or low costs, it seems clear” that they mean costs that are unusual if the cost-output relationship and effects of «: ' numerous small disturbances remain at their previous, satisfactory levels. That is, they are interested in detecting those C; that are unusually high or low if the average jrelationship between cost and output is unchanged. Similarly, when they refer to a sequence of days when costs are higher or lower than would be expected, it seems clear that they mean a run of the C; that is unlikely if this relationship is unchanged. If the model is adequate and if the aver- age relationship between cost and output is known, it is a simple matter to set up a control chart that will aid the manage- ment. Each day, the deviation of actual cost from the cost that would be expected on the basis of this relationship can be plotted on a chart that has two sets of con- trol limits. The outer control limits can be set so that, if this relationship remains the same, the probability that a point lies out- side them is small; The inner control firnits can be set so that, if this relationship re- mains the same, the probability that two consecutive points lie outside them (in one direction) is small, When a point lies out- side the outer limits or a pair of points lies outside the inner limits, there is evidence that the relationship may have changed and that an inquiry should be made, SETTING UP THE CHART ‘A control chart was set up at a freight yard located at Toledo, Ohio. This yard consti- tutes one of the largest and most impor- tant links in the railroad we studied. The chief types of freight that pass through the yard are livestock, perishables, coal, and automobiles. Table 1 shows the number of cars switched, the number of cuts switched, the number of cars delivered, the number of crews employed, the number of engine hours used, and the costs at the yard, for a sample one-week period. The first step-in setting up the chart was to gather historical data concerning cost and output. Data similar to those in Table 1 were collected for sixty-one days, and the average relationship between cost and output was estimated. The resulting relationship was C,=4,914 + 0.425; + 2.44D; aw ‘The second step was to test some of the assumptions underlying the chart, Some of these tests are quite similar to those used in quality control to determine if the process is “in control.” Taken together, the results of these tests did not cast any great sus- TABLE 1 . Output, Cost, and Employment, Freight Yard, Toledo, 7 Days* ITEM FRI-SAT. SUN. MON. = TUES. WED. THURS. Number of cuts switched 869 792 762 586 669 732 © 659 Number of cars switched 2534 2303. «2521 «1849 «2090 2114 1979 Number of cars delivered* 1015 1003 820. 548-877, 706 1038 Number of crews used. 45 45 40 38 46 46 46 Number of engine hours 372 369 329 309 385 381 386 Money costs ($) 7523 7464 6932 6550 7606 7757 7701 *Taken from records of cooperating railroad tincludes number of cars picked up, 153°,‘CASE FOUR picion on the assumptions underlying the chart. Indeed, the results seemed to be quite compatible with these assumptions. The third step was to draw the inner and outer control limits on the chart. The outer control limits were set at $804, and the inner control limits were set at + $410, ‘These limits (designated by ICL and OCL) are included in Figure 1. If there were no errors in the assumptions, the probability would be 0.05 that a point would lie out- side the outer limits if the relationship remained fixed. Similarly, the probability would be about 0.03 that two consecutive points would lie outside the-same inner control limit. After setting the control limits, an attempt was made to determine if any assignable cause could be found for the days FIGURE 1 Deviation of Actual Cost from Expected Cost Based on Average Relationship © Between Cost and Output: Freight Yard, 61 Days Source: Records of cooperating railroad. 154 that were “out of control.” None could be found, and it was assumed that they were due to “chance.” The number of such days was almost precisely what one would expect on a chance basis. PERFORMANCE OF THE CHART This section describes the performance of. the chart during a six-week period that was several months subsequent to the time when the chart was set up. The results apply to the» freight yard described above. On each day. during the period data were collected con- cerning the money costs (C), the number’ of cuts switched (S), and the number of cars: delivered (D) on the previous day. Thert the deviation of actual cost from expected DayDeviation ($) ————r FIGURE 2 Control Chart for Costs: Freight Yard, 42 Days Source: Records of cooperating railroad. cost based on the average relationship in equation (1) was computed and plotted on the chart.’ The deviations that were plotted are shown in Figure 2. During the six-week period, four days seemed to be out of control, and in every case there seemed to be an assignable cause. One of these days was Labor Day. The exceptionally high costs on Labor Day can be attributed to the punitive wage-rates that were paid because it was a holiday. On the other days that were out of control (all of them Sundays) the exceptionally low costs can be attributed in part to the fol- Sowing circumstances: (1) Cars delivered to other yards constituted a large propor- ~ tion of all cars delivered, and since such cars are relatively easy to deliver, costs were depressed. (2) More efficient methods were used to handie incoming cars. (34Some work ordinarily performed by the yard was done by another yard; hence costs were reduced somewhat. ‘The basic data were’ collected by rail- road employees and the points on the chart were computed by officials of the railroad. From their evaluation of the performance of the yard it appeared that the chart pro- vided a reasonably faithful picture of the level of performance. Moreover, the chart stimulated some inquiries that indicated: where improvements in yard performance might be made. 4 Actually, the relationship was recomputed as mote and more data became available. 155,1. Do managers often find it difficult to minimize costs? If so, why? 2. Can control charts of the sort de- scribed in this case help managers to minimize costs? If so, how? 3. Is equation 1 a short-run or long-run cost function? Why? 156 . What is the marginal cost of delivering © .. Does the average cost per cut switched. . In what ways can these estimates of acar? exceed the marginal cost of switching acut? marginal and average cost be useful to railroad managers? Explain.