Attribute Control Charts

Introduction
Attribute data are data based on counts, or the number of times we observe a particular event. The events may be:
the number of nonconforming items; the fraction of nonconforming items; the number of defects; or any other distinct occurrence that is operationally defined. defective or conforming; go or no-go; acceptable or not acceptable; or number of defects per unit.

Attribute data may include such classifications as:

Types of Attribute Control Charts

There are two basic types of attribute control charts:
classification charts count charts.

Types of Attribute Control Charts

Classification Charts Classification charts deal with
either the fraction of items or the number of items in a series of subgroups that have a particular characteristic.

p Chart. The p chart is used to control the fraction of items with the characteristic. Subgroup sizes in a p chart may remain constant or may vary. A p chart might be used to control defective versus conforming, go versus no-go, or acceptable versus not acceptable.
np Chart. The np chart serves the same function as the p chart except that it is used to control the number rather than the fraction of items with the characteristic and is used only with constant subgroup sizes.

 Count Charts
c Chart. A c chart is used to control the number of times a particular characteristic appears in a constant area of opportunity.
For example:
defects per air conditioner; accidents per workweek in a factory; and, deaths per week in a city.

 u Chart. A u chart serves the same basic function as a c chart, but it is used when the area of opportunity changes from subgroup to subgroup. For example, we may examine:
varying square footage of paper selected from rolls for blemishes; or, carloads of lumber for damage when the contents of the rail cars varies.

The p-chart for Constant Subgroup Sizes The Centerline and Control Limits

Construction of a p-chart Example

As an illustration, consider the case of an importer of decorative ceramic tiles. Some tiles are cracked or broken before or during transit, rendering them useless scrap. The fraction of cracked or broken tiles is naturally of concern to the firm. Each day a sample of 100 tiles is drawn from the total of all tiles received from each tile vendor.

Centerline (p) = 183/3000 = 0.061

 Examining the chart, we find a process that is out of control.

On day 1, the mean fraction value is above the upper control limit. The sample mean for day 14 is also above the upper control limit, another indication of lack of control.

 Further study reveals that on both day 1 and day 14 the regular delivery truck operator was absent because of illness. Another employee loaded and drove the delivery truck on those days. That individual had never been instructed in the proper care of the product, which requires special handling and treatment. To solve this problem and eliminate this special cause of variation, management created and implemented a training program using the regular driver's experience for three other employees. Any one of these three employees can now properly fill in and perform satisfactorily. Thus the system has been changed to eliminate this special cause of variation.

 After the process has been changed so that special causes of variation have been removed, the out-of-control points are removed from the data. The points are removed from the control chart, and the graph merely skips over them. Removing these points also changes the process average and standard error. Therefore the centerline, control limits, and zone boundaries must be recalculated.

 The new centerline and control limits are: = 154/2800 = 0.055

Boundary between upper zones B and C =

Boundary between lower zones B and C =

Boundary between upper zones A and B = Boundary between lower zones A and B =

 The process now appears to be stable and in a state of statistical control. Management may now look for ways to reduce the overall process average of the number of cracked or broken tiles to raise the usable number of tiles per shipment and effectively increase the process output.

 Iterative Reevaluations
It is possible -- and not at all uncommon -that by changing the process, removing points that were out of control, and recomputing the control limits and zone boundaries, points that initially exhibited only common variation will now indicate a lack of control. If and when this happens, the system must again be reevaluated to eliminate the newly revealed special causes of variation.

 Subgroup Size
When constructing a p-chart, the subgroup size is much larger than that required for variables control charts. This is because the sample size must be large enough that some nonconforming items are likely to be included in the subgroup. If, for example, a process produces 1.0 percent defectives, sample subgroups of size 10 will only occasionally contain a nonconforming item. As a general rule of thumb, control charts based on classification count data should have sample sizes large enough so that the average count per subgroup is at least 2.00. This allows the A, B, and C zones to be wide enough to provide a reasonable working region into which data points may fall for analysis. This is true for both the p-chart and the np-chart, which we discuss later in this chapter.

 Subgroup Frequency
Every process goes through physical cycles, such as shifts and ordering sequences. p-chart and np-chart calculations must be based on a sufficient number of subgroups to encompass all of the cycles of a process to include all possible sources of variation. Subgroup data should be collected at a frequency greater than the frequency at which the process can change. This frequency is determined by a process expert.

 Number of Subgroups
As a rule of thumb, the number of subgroups should be at least 25 for pcharts and np-charts.

The p-Chart for Variable Subgroup Sizes

Sometimes subgroups vary in size. This makes the construction of a manual p-chart somewhat more tedious, although circumstances may make this situation unavoidable. Common among these is when data initially collected for some purpose other than the creation of a control chart are later used to construct a control chart. The standard error, , varies inversely with the sample size. That is, as the sample size increases, the standard error decreases, and vice versa. Control limits and zone boundaries are calculated based on the standard error. Consequently, as the sample size changes so will the control limits and the zone boundaries.

 Using Average Subgroup Size: An Example

Consider, for example, the case of a highway toll barrier with two types of toll collection mechanisms: automatic and manned. The automatic lanes require exact change or a transponder while the manned lanes do not. The fraction of vehicles arriving with exact change or a transponder is examined using a control chart for a series of rush hour intervals on consecutive weekdays. As the number of vehicles passing through the toll barrier varies, the control limits change day-to-day. Onehour periods (7:30 to 8:30 am) for 20 consecutive weekdays yield the data shown.

 Using Average Subgroup Size: An Example

Centerline(p) = = 2569/6421=0.400 We can also calculate the UCL, LCL, and zone boundaries for the first data point,

Boundary between lower zones A and B = Boundary between lower zones B and C = Boundary between upper zones A and B = Boundary between upper zones B and C =

 Using Average Subgroup Size: An Example

The np-Chart
Classification data can sometimes be more easily understood if the data appear as counts rather than fractions. This is especially true when using attribute control charts to introduce control charting and encountering reluctance by some members of the affected community to deal with fractions rather than whole numbers, such as the number of defects. The quantity np is the number of units in the subgroup with some particular characteristic, such as the number of nonconforming units. Traditionally, npcharts are used only when subgroup sizes are constant. As the information used is the same as for p-charts with constant subgroup sizes, these two charts are interchangeable.

 Constructing the np-Chart

 Constructing the np-Chart

Centerline (np) =

 7.6.1 Constructing the np-Chart

Boundary between upper zones B and C = Boundary between lower zones B and C =

The upper boundary between zones B and C for this example is given by
and the lower boundary between zones B and C is given by

Boundary between upper zones A and B = Boundary between lower zones A and B = The results for this example are:

Count Charts
A defective item is a nonconforming unit. It must be discarded, reworked, returned, sold, scrapped, or downgraded. It is unusable for its intended purpose in its present form. A defect, on the other hand, is an imperfection of some type that is undesirable, although it does not necessarily render the entire good or service unusable. One or more defects may not make an entire good or service defective. For example, we would not scrap or discard a computer, a washing machine, or an air conditioner because of a small scratch in the paint.

 Conditions for Use

If we are to use the c-charts or u-charts, the events we are studying must be describable as discrete events; these events must occur randomly within some well-defined area of opportunity; they should be relatively rare; and they should be independent of each other. Exact conformance to these conditions is not always easy to verify. Usually, it is not too difficult to tell whether the events are discrete and whether there is some well-defined area of opportunity. But whether the events are relatively rare is somewhat subjective and requires process knowledge and experience. The issue of independence is generally revealed by the control chart. That is, if the events are not random and independent, they will tend to form the identifiable patterns.

c-Charts
Areas of opportunity that are constant in size are easier to manage than those that vary, in much the same way as constant subgroup sizes in a p-chart are easier to manage than those that vary. Constant areas of opportunity might be such things as:
one unit of a particular model of a TV set; a particular type of hospital room; one printed circuit board; one purchase order; one aircraft canopy; five square feet of paper board; five linear feet of wire.

When all conditions for an area of opportunity chart are met, and when the subgroup sizes remain constant, a c-chart is used.

 The number of events in an area of opportunity is denoted by c, the count for each area of opportunity. The sequence of successive c values, taken over time, is used to construct the control chart. The centerline for the chart is the average number of events observed. It is calculated as:

 Counts, Control Limits, and Zones

Consider a firm that has decided to use a c-chart to help keep track of the number of telephone requests received daily for information on a given product. Each day represents an area of opportunity. Over a 30-day period, 1,206 requests are received, or an average of 40.2 per day.

A count of 59 is within the control limits, while a count of 60 is beyond the UCL.

 Counts, Control Limits, and Zones

Boundary between lower zones B and C = Boundary between lower zones A and B = Boundary between upper zones A and B = Boundary between upper zones B and C =

Because the actual counts are whole numbers, the observation would fall into zones as follows:

The zones each contain a reasonable number of whole numbers and are close enough in size to be workable.

 Counts, Control Limits, and Zones

Consider the problem that would have been encountered if the process average had been = 2.4. Here we would get:
Boundary between lower zones B and C = Boundary between lower zones A and B = Boundary between upper zones A and B = Boundary between upper zones B and C =

Zone
Upper A Upper B Upper C Lower C Lower B Lower C 6, 7 4, 5 3 1, 2 0 0

Counts

 Counts, Control Limits, and Zones

As a rule of thumb, the zone boundaries should not be used for c charts with average counts of less than 20.0. As the observable count shrinks, the use of variables control charts must be instituted for continued process improvement.

 Construction of a c-chart: An Example

Consider the output of a paper mill: the product appears at the end of a web and is rolled onto a spool called a reel. Every reel is examined for blemishes, which are imperfections. Each reel is an area of opportunity.

 Construction of a c-chart: An Example

Centerline(c) =

 Small Average Counts

When average counts are small, data appearing as counts will tend to be asymmetric. This may lead to overadjustment (false alarms) or underadjustment (too little sensitivity).

 Stabilizing a Process: An Example

An industrial washing machine manufacturer inspects completed units for defects.

The process is not in control. Special causes of variation are present. Let us assume that the local operators responsible for the final inspection act so that the special causes of variation for points 9, 10, 14, 19, 20, 22, and 23 are identified and the appropriate corrections are made. The data for points affected by known special causes that have been eliminated are deleted from the data set, and the centerline and control limits are recomputed:

The new limits are so close to the old limits that the old limits are used for the next 24 machines produced.

 The first five data points for these next 24 machines are well below the lower control limit. Investigation by the local operators reveals that a substitute for the regular inspector counted the defects on those five machines. The substitute was not properly trained and did not identify all the defects correctly. The operators informed management, and management made appropriate changes in policy so that this situation would not recur. These points can now be eliminated from the data. Beginning with machine number 30, all counts are below the process average. Local operators decided that the process has been changed, so a revised control chart is constructed beginning with point number 30.

Construction of a c-chart: Another Example

Consider the case of a mill with a constant work force of 450 employees that has a sign posted at the employee entrance reading "SAFETY IS BETTER THAN COMPENSATION." Informal conversations with employees reveal they consider the sign a reminder to be careful. But management has not simultaneously made the work environment safer. There are still cluttered aisles, and spills and leaks of liquids on the floor are not attended to rapidly. The workers know this but have long ago stopped their fruitless efforts at getting management to allocate the resources necessary to create a safer workplace.

 Note that as the number of labor-hours per month remains constant, the area of opportunity is considered constant month-to-month.

C Chart of Accidents
4 UCL=4

Sample Count

_ C=1

0 Mar Jun Sep Dec Mar Month Jun Sep Dec

LCL=0

 As there are no indications of any special variation, we can conclude that the process is stable and in a state of statistical control. The company may not realize it, but it is in the business of producing accidents at the stable rate of one per month. It will continue to do so until some effort is made to change the underlying process. If no change in the process is made, accidents will continue to be produced at this rate. Consequently, the "SAFETY IS BETTER THAN COMPENSATION" sign is unfair: employees are not empowered to make system changes that would lower the average number of accidents per month; the sign subtly and unjustly shifts the burden of responsibility for safety from management to the employees.

 7.8.6 Construction of a c-chart: Another Example

A national company opens a sales office with six salesmen in Cleveland, Ohio. The office has been open for slightly more than six months. All salesmen have the same responsibilities and opportunities. The number of new accounts generated by each salesman in the first six months of operation is recorded on the table. You have $54,000 of bonus money to allocate to the sales force. How much would you give to each salesman? Why?

Name of Sales Person Allan Fred Mark David John Phil Total

Number of New Accounts 27 36 28 24 29 30 174

Examining the above data from the perspective of the System of Profound Knowledge and statistical thinking results in the following c-chart statistics:

According to the above calculations, the number of new accounts generated by a salesman in a six month period can be predicted to be between 13 and 45 new accounts. This large amount of variation is attributable to the sales system. All salesmen are in the same sales system, and they all deserve the percentage pay raise. No one should be rewarded, and no one should be punished. Sales management should focus their attention on improvement of he sales system, not on rewarding and punishing salesmen.

If Fred had generated 66 new accounts (instead of 36) the sales systems statistics would be

In this scenario, Fred is outside the sales system on the high side (he is above 51). Investigation by Freds manager led to the realization that Fred had developed a better telephone procedure for screening potential customers. Fred should receive special recognition because he is outside the sales system on the high side and because his efforts provide guidance for improvement for all salesmen all in the sales system.

u-Charts
In some applications the areas of opportunity vary in size. Generally, the construction and interpretation of control charts are easier when the area of opportunity remains constant, but from time to time variations may be unavoidable. For example, samples taken from a roll of paper may need to be manually torn from rolls, so that the areas sampled -the areas of opportunity -- will vary.

 The characteristic used for the control chart, u, is the ratio of the number of events to the area of opportunity in which the events may occur. For observation i, we call the number of events (such as imperfections) the observed ci, and the area of opportunity ai. Thus, ui is the ratio for each point.

 Construction of a u-chart: An Example

Consider the case of the manufacture of a certain grade of plastic. The plastic is produced in rolls, with samples taken five times daily. Because of the nature of the process, the square footage of each sample varies from inspection lot to inspection lot.

 7.9.1 Construction of a u-chart: An Example

The control limits are different for each of the subgroups and must be computed individually for each subgroup

 No points indicate a lack of control, so there is no reason to believe that any special variation is present. If sources of special variation were detected, we would proceed as we did with the c-chart - that is, we would identify the source or sources of the special variation, eliminate them from the system if detrimental, or incorporate them into the system if beneficial; drop the data points from the data set; and reconstruct and reanalyze the control chart.

Limitations of Attribute Control Charts

As processes improve and defects or defectives become rarer, the number of units that must be examined to find one or more of these events increases. If we consider a p-chart where the average fraction of nonconforming items is 0.005, then on average we'd need to examine 200 units to have an average count of just 1.00. In the extreme, to maintain a reasonable average count as the area of opportunity grows, 100 percent inspection becomes the rule.

 This implies inspecting all of the items and sorting those that conform to some specification from those that don't. Not only is this inspection costly, but it's equivalent to accepting the fact that the process is producing a constant fraction of its output as defective and will continue to do so. Hence, attribute control charts are limited in terms of the level of process improvement they make possible. Additional process improvement is possible with variables control charts.