Editor’s note: The first installment of Scott Cornish’s series on process improvement appeared in the July 2006 issue of Newspapers & Technology. In this, the 10th installment, Cornish continues to discuss the role control charts play in defining process improvement.

In the last two articles, we went over three of the four variable data control charts in the process improvement toolkit: the Individuals and Moving Range chart, the X-bar and R chart and the X-bar and s chart.

In this installment, we will briefly go over the fourth, the median chart, and then begin to review attribute control charts, beginning with the p chart.

The median chart, for all of you who remember math, reflects the median — or middle value of a data set — as opposed to the average, or mean, of the sample data.

The median chart is very rarely if ever used nowadays. In the past, one noted advantage of a median chart is that you don’t have to calculate an average. You just order the data from lowest to highest, count up the number of measures in your sample, and then choose the value in the middle or at 50 percent of the data. Given the power and ease of use in Microsoft Excel and other software packages, this is no longer an advantage.

What’s more, if your data exhibits a normal distribution, the mean and median should be either the same or extremely close.

That said, when would you want to use a median chart? It’s used when “outliers,” or data that measures well beyond what is normally expected, occur. Take this as an example. Let’s say Bill Gates moved into your neighborhood. Would that mean that your average net worth is now north of $40 billion? That, my dear reader, is an outlier and here is a case where a median chart would be appropriate.

Other than that scenario, it’s doubtful you’ll ever have to use a median chart. But in the event you do, here’s how to do it:

Enter your values and begin to prepare the data set as if you were to plot an X-bar and R chart.

If you are using QI Macros (or any other technique using Excel), replace the AVERAGE (##:##) function with MEDIAN (##:##) in all the appropriate places. MEDIAN is a standard function in Excel. If you use the QI Macros, you can plot a standard X-bar and R chart. In the created workbook for the chart, replace and fill down in the two sets of rows where the AVERAGE is calculated for the charts.

The control limits for the median chart are calculated in the usual way.

Relabel any points or text on the chart that indicates AVERAGE or “X-bar” with MEDIAN or “P50” so not to confuse anyone.
 

Attribute data

Now let’s move onto the second type of data: attribute, or discrete, data.

Given that newspaper production operations involve measured data, attribute charts aren’t as widely used as often as variable charts. The reason for that is that much of the data we have or have access to is from measurements. We’ve used black solid ink density in the examples for our home delivery before 6 a.m. project, but there are many others.

Discrete data comes from counting the occurrence of events. Some general examples include the number of scraped negatives or plates per shift, day or week; the number of copies wasted per press run startup; or the number of copies torn or damaged by an inserter.

The four red boxes above the upper control limit in this p-chart signify where further action is required.
 

Before we plunge into attribute control charts, I want to note one more thing. A commonly used and very accurate phrase is that “you can’t manage what you can’t measure.” To that end, it is desirable to move from discrete data to continuous data control charts since the later are more sensitive to process change.

On the other hand, some advanced process improvement techniques are only effective with discrete data.

We can find out quite a bit about how effective a process is by using one or more of the four attribute control charts:

•The p chart for proportion nonconforming.

•The np chart for number nonconforming.

•The u chart for nonconformities per unit.

*The c chart for nonconformities.
 

Begin with p

So let’s start with a p chart. Our home delivery by 6 a.m. committee has continued its work and is now digging deeper into all the various processes in the production chain, from prepress through distribution. In the earlier stages of the project, one of the areas the team looked at was late press starts.

There was a hodgepodge of data from the various departments as the group looked into each set of data to determine possible causes.

In the platemaking area, one cause was remade plates. The team decided to look at this issue, assessing why some days resulted in higher remakes than other days.
 

No spot color

A few factors to keep in mind: The paper uses two doublewide, two-around presses and always runs collect. For every page, two plates are required. For every 4-color page, three (C-M-Y) plates are added. No spot color ran during the period.

During the sample time period, any second edition and/or pressroom related remakes were not included in the count.

Page dummy plates were not included in the count either.

So after a few weeks, here’s a table of data that the production manager prepared (see above).

 Remember, be careful about using historical data unless you are certain about the conditions under which the data was collected.

In this case, operators collected the data per instructions from the production manager. Upon presenting the data, she noted a few factors about the plateroom.

First, for Monday and Tuesday issues, the plateroom is staffed by only one operator, and only for a portion of the shift.

For Wednesday and Saturday issues, a single operator who is in the area during the entire shift staffs the plateroom. He is quite busy, however, making plates for the paper as well as advanced and commercial press runs.

During the middle of the month, the plateroom had an exposure inconsistency problem with CTP unit No. 1. Later in the month, it experienced the same problem with unit No. 2.

 In both cases, there was a significant increase in remakes for those issue dates.

One member plotted a p-chart of the data (see page 22).

 

Two large spikes

As would be expected from the production manager’s caveats, we see two large spikes, one in the middle of the month and another at the end. 

We also see three other less drastic spikes during the month. The team investigated further and found that each of the three spikes was for a Tuesday issue. During the discussion about this, the production manager repeated that for Tuesday issues, the plateroom was staffed only part of the shift.

Another team member noted that Tuesday staffing was no different than Monday staffing. Upon reflection, the production manager remembered that while the Monday operator had more than eight years’ experience, the Tuesday operator was a new hire. She made a note to follow up to determine if the Tuesday operator needed additional training.

It is important to note that this chart is slightly different from the attribute charts we looked at in previous articles.

 

Factors to watch

First, there is just one chart showing the average. There is no second chart, depicting a moving range. That is the standard format for a p-chart.

Second, there are four red squares above the upper control limit — three in a row during the middle of the month. These are “significant process shifts” and require further action.

Third, note that the UCL, lower control limit and +/- 1 and 2 sigma lines “shift” for each point on the chart. That is because some software packages, such as the QI Macros app used in the example, recompute control limits each time the sample size changes.

Since our sample size changed each issue day because of increased or decreased paging and numbers of color pages, these values were recalculated. While technically correct, it can be confusing.

 

Quick note

Finally a quick note on the difference between a sample and a population. A sample is a subset of a population and is what we plotted here. But you might say that we plotted the entire data set for the shift, which would make it a population.

Not exactly. We excluded second edition, replates/PostScripts, advanced press runs and commercial work. If we included data for all these products, then we would have a population and our statistical approach would change.

Next month, we’ll continue our review of attribute control charts with coverage of the np chart, c chart and the u chart. Then we’ll review the “significant process shifts” rules.