Special Report: Quality in Education on the Move

Using Statistical Process Control to Improve Attendance

by Sharron Walker, Ed.D.

An Arizona high school tests the utility of profound
knowledge by applying SPC to an attendance problem.

There has been little operational research in education on the use of profound knowledge and statistical process control by a school principal in a public high school setting. W. Edwards Deming advocated the theory of profound knowledge, which suggests that management must understand how theories of variation, knowledge, systems and psychology interconnect to optimize organizational aims. Deming maintained that managers cannot lead organizations without knowledge of how each of these elements impacts the other.

The theory of knowledge helps managers understand the importance of prediction to the process of continuous improvement. The theory of variation helps managers understand that the principle of randomization lies at the core of continuous improvement. The theory of psychology helps managers understand the contribution that people who work in the system make to improving processes. A theory of systems helps managers understand how continuous improvement is based upon an awareness of the delicate interaction of people and processes. Profound knowledge theorizes that one element is incomplete without understanding the others.

Profound knowledge is a theory proven in action by the application of statistical process control, a methodology used to study and refine processes using statistical methods and techniques to continually improve production. SPC uses various tools like control charts to analyze a process so as to take appropriate actions to achieve and maintain a state of statistical control and to improve the process.

Because schools are not proc-ess-oriented, they have few methods and tools to help them measure how their processes function over time. Schools need a useful way of understanding, studying and measuring their processes. SPC provides an approach to validating whether school processes improve over time.

As the principal of a small high school in southern Arizona, I decided to test the utility of profound knowledge through the application of SPC to a school attendance problem. My purpose was to see if the root causes of the attendance problem could be determined by approaching the problem from a system perspective and using the people who work in the system to work on the system.

In Arizona, students must attend high school for a minimum of 175 days each year. Students who miss some of these days lose out on their own instruction and also increase the need to have teachers spend class time reviewing and reteaching. In addition, the state of Arizona bases funding on attendance.

During the last two years, our teachers began grumbling about the amount of time they spent on preparing lessons for students who were continually absent. The superintendent began complaining about lost revenue.

The vice principal, who is in charge of discipline and attendance, implemented a system of rewards and punishment to attract students to school. We awarded gift certificates for McDonald's, local record or clothing stores to students who had 100-percent attendance each month. But these incentives attracted those students who would normally come to school anyway. Students who continually missed school were called into the vice principal's office with their parents present. The students were suspended. This resulted in an even higher proportion of absenteeism. The school worked with the courts on the truancy rate, also to no avail. Figure 1 depicts the mean attendance rates for the first 100 days of school (including the mean rate by grade level) for the total school of 400 students over a six-year period. A program of rewards and punishments did not appear to increase attendance rates.

Obviously, a chronic attendance problem existed. In addition, the 100-attendance-day reports indicated that attendance had declined steadily during the last two years. Treating the chronic problem with a system of rewards and punishment only provided temporary relief to the problem. We treated the symptoms but did not search for the reasons why the problem occurred. We needed to first probe for the root causes.

Based upon Deming's principle that those who work in the system contribute the most to solving system problems, we formed an attendance task force. Teachers, students, parents and administrators comprised the task force. The team collected data to assess the current attendance situation, determined causes of nonattendance and identified potential solutions. These solutions would be presented to the superintendent and school board in the form of recommendations.

Data collection
The task force structured its data collection using the organizational framework proposed in Hoshin Planning, The Developmental Approach by Bob King. We began by asking ourselves the purpose for the data collection. We then posed a question to guide the data collection. We decided on the kind of data we would use to answer the question and called this our input.

We then determined what the output should mean. We listed the tools we would use to collect the data and wrote our results. We then analyzed the results and formed theories. These theories usually formed the basis for the next question.

The task force began with a macro question to guide the project: Why isn't the attendance rate at our school at 94 percent? A 94-percent attendance rate held the promise of less reteaching and more state aid. The first micro question would help us understand what prevents students from attending school: What influenced students from coming to school? To answer this question, we created a picture of the variables that could affect student absenteeism under the current system.

Using a brainstorming process, we drew a cause-and-effect diagram identifying these variables. The cause-and-effect diagramming process asks the people who are part of the system to identify system influences.

This creative process became our input. The cause-and-effect diagram of system influences on what determines whether students attend school became our output. We brainstormed five categories of influences: instruction and curriculum, economic environment, people, social environment and outside agency influences. We then brainstormed further detail for each category. The result was a pictorial definition of the attendance system as the committee viewed it. Because members of the team represented various segments of the school community, our analysis of the results revealed diverse factors that affected attendance rates. For example, it was suggested that students sometimes miss school because they have to wait all day to see a doctor at the local hospital.

Next we identified periods of time where high numbers of students were absent. We asked two questions. First, what are the mean attendance rates for grades 7­p;12? Second, is the variation in attendance from year to year stable? We talked about the meaning of stability. We decided that it meant that attendance rates, if stable, would evidence a random pattern over time. We examined monthly attendance data from 1988­p;1994.

The data indicated the potential number of days students could be in school each school month and the actual number of days they were in school. We used the first eight school months of the year. This accounted for about 150 days of school (out of 175 days). We used a control chart to plot mean attendance rates showing nonrandom and random points.

The team used a p chart for attributes because we asked a yes/no question: Were students in school or were they not in school during a particular period of time? The p chart would analyze whether a stable system existed. If the system was stable, the chart would graphically show random patterns of attendance within the control limits. This would tell us that we had a predictable system. If attendance rates were to be increased, the system would have to be changed. If the system was not stable, the chart would graphically show nonrandom patterns outside the control limits. The root causes of this nonrandomness would have to be determined and the instability eliminated in order to continually improve the system.

The team decided that all points above the upper control limits were positive since these points indicated higher attendance rates. Results from a p control chart revealed that 60 percent of the points were above or below the control limits, indicating a system that was not stable. The chart informed the team that attendance was at its highest point at the beginning of the school year, steadily declined until the second semester, where it spiked (although at a lower level than the beginning of the year), and declined again, spiking at increasingly lower levels.

This roller-coaster pattern of attendance perplexed the team. We speculated that the nonrandom nature of the graph over time indicated a system that contained many special causes. Because the group lived in the community, they hypothesized that the consistent roller-coaster pattern might indicate that there were common elements among students which caused them to miss school. They speculated that the nonrandom causes might be a result of events that occurred in the community.

The team split in two. The first team studied the hypothesis that the special causes were due to nonrandom events in the community. They did this by comparing the attendance patterns of our school with the two other feeder schools in our district. All three schools drew from similar populations. The second team disaggregated the data to determine if certain grade levels contributed more to the problem than others.

The first team developed p charts for attributes for the feeder schools. They reasoned that if attendance patterns at these schools were similar, the nonrandom or special causes influencing low attendance rates would be common to all schools. The p control charts they developed indicated that attendance patterns at the two other schools were primarily the same. The charts showed higher attendance at the start of school and then a general decline with small spikes in similar months at the high school. Although the mean averages were higher at the two other schools, the patterns were the same.

The team probed deeper and drew a Pareto diagram, which was used to determine the days of the week students were most likely to be absent in all three schools. They found that students were most likely to be absent on Mondays, followed by Fridays and Tuesdays. Thursdays and Wednesdays were the best-attended days. The team brainstormed reasons why Mondays, Fridays and Tuesdays would have lower attendance rates and then listed what events were occurring in the community to verify their findings. They found that on some of these days, important events occurred in the community that took a higher number of children out of school.

The first team also sought to determine the influence of certain months on attendance rates. During what months (attendance periods) do students have the highest absentee rates? Using mean attendance rates for each attendance period, they drew a Pareto diagram. The diagram, depicted in Figure 2, shows that February and March had the highest absentee rates while August and September had the lowest. Attendance spiked during the January and February period, when finals were administered and a new semester began. Similar patterns existed at the other two schools, indicating to the team that the time of the year and the day of the week influenced student attendance. Knowing when students would be out would help predict future attendance rates.

The second team, using monthly attendance rates over a six-year period, found that grades nine, eight and seven had the lowest attendance rates. In addition, the roller-coaster pattern generally appeared to repeat itself over time. The ninth grade revealed a general pattern of instability, but the pattern shifted slightly upward during the 1993/1994 school year. The team speculated that this might have resulted from a new at-risk program that had been piloted during the school year.

At grades seven and eight (middle school), new programs had been introduced during the course of the six years to attract students to school. Initially, the programs had shifted the attendance patterns upward. Gradually, however, attendance again declined. The team wondered whether the program's newness had initially attracted the students or other factors were at work. They interviewed the teachers who taught at this level. The teachers suggested teacher turnover, staff reduction and higher suspension rates (due to a gang element entering the school for the first time) all contributed to the decline in attendance during the last two years. The teams then regrouped. We had identified through the use of control charts special causes of variation and identified which months and days of the weeks contained points indicating special events. We had disaggregated the data and found the grade levels that contributed most to the problem.

The group wanted one more piece of information before determining root causes of nonattendance. The control charts helped predict future attendance based upon our current system, but they could not predict the characteristics of the student most likely to be absent. We decided to draw up a profile of a truant student for predictive purposes. We defined a truant student as a student who missed half a semester or more. We used individual student attendance records to compile data.

Using a check sheet, the team found that 21 percent of the students fit into this category. We collected data on the grade level, gender, average age, address and working status of truant students. By analyzing student-attendance data, we built a profile of a typical truant student: male, ninth grader, 16.1 years of age, lived in the city and his parents did not work. Data on retention informed the group that the ninth grade had the highest repeat level and percentage of students who eventually dropped out of school.

Determining causes of nonattendance
We now had a great deal of student-attendance data, but we didn't know what to do with it. This happens often in schools. The "where do we go from here" syndrome sets in, and we begin doubting what we are doing.

We reviewed the data we had collected and decided that even though we could predict when students missed school and had developed a profile of the student least likely to attend, we did not know why students missed school. What reasons do students give for being absent? We developed a simple survey that asked students to check reasons they were absent on a particular day. We did this for a three-week period of time. Figure 3 shows a Pareto analysis of the results.

The output showed the primary reasons students give for being absent: illness, death in the family, oversleeping, sick children, not having a babysitter, transportation problems, medical appointments, ditching, suspensions, being in jail, hospital treatment for substance abuse and pregnancy-related illnesses. Many absences appeared to be factors that the school could not control, such as death in the family or oversleeping. The Pareto diagram indicated that sick children, babysitting, medical appointments and pregnancy-related events accounted for 24 percent of absences. Disciplinary infractions comprised almost 13 percent. The team decided to investigate this further because these areas presented opportunities for improvement.

In order to determine the influence of suspensions on attendance, the team asked the vice principal for data on the percentage of students at each grade level who were suspended for discipline infractions. More than 95 percent of the suspensions occurred between grades seven through nine. These grades had the highest rate of nonattendance. The team probed further to determine if there were certain times during the day when the greatest number of infractions occurred. Using a check sheet, we found that most students committed disciplinary infractions during class changes and lunch.

The team also wanted to determine the influence of student-parent status on attendance. Student-parents were defined as students who had children or were in the process of having children. We wanted to know how the attendance of a student-parent compared to a student. Looking at attendance records of the representative groups, a simple check sheet indicated that student-parents were absent at four times the rate of nonstudent-parents. Being a parent directly influenced a student's ability to attend school. Disaggregating the data on student-parents, it was found that these students seldom had unverified absences, were ill infrequently and had no suspensions. But a high percentage of these students dropped out of school.

Identifying potential solutions
By using SPC, the team discovered that nonrandom effects caused much of the attendance problem. Through the use of control charts, cause-and-effect diagrams, Pareto analysis, graphs, check sheets and surveys, we found that a high percentage of these nonrandom effects occurred in schools all over the district. The team made a number of recommendations:
Offering alternative forms of schooling to increase the attendance rates of 16-year-olds in the ninth grade.
Creation of an in-house suspension program for students who would normally be suspended from school.
Refine the current middle school program in order to curtail the number of change periods.
Offer more lunch activities for students that age.
Hire a fully certified nurse who could attend to students' medical needs on campus.

Our next step was to prepare a cost analysis based upon our recommendations. SPC is a method of assessing system proc-esses that is typically used in the business sector. Profound knowledge is an integrated theory of understanding people, variation, systems and knowledge. It forms the foundation for SPC. This study analyzed the use of SPC, based upon Deming's theory of profound knowledge, by the principal in a high school. It hoped to test the usefulness of Deming's theory by applying it to a school attendance problem. It used the people involved in the system to study the system's variation and suggest theories of improvement.

Although we often wandered off course and sometimes didn't know what direction to take, by using data to guide our research questions, we were able to assess and analyze our current system by studying our processes, something we had never been able to do before. We learned to use new tools in conducting our assessment. Most important, we learned that the combined energies of a team of diverse people working collaboratively for a common purpose was more productive in understanding the interwoven nature of a system than individual efforts.

By working together and documenting our system, we were able to understand that variation is part of every system. By finding the root causes of the variation, we diagnosed problems in our system and made recommendations. We did not solve our attendance problem-we only took one step toward beginning to understand and improve it. We learned that a basic tenet of profound knowledge is that continuous improvement is a never-ending journey.

About the author
Sharron Walker, Ed.D., is the principal of Baboquivari Middle/Senior High School in Sells, Arizona.