The Power of Absence Streaks in Attendance Forecasting

Executive Summary

Panorama Education’s Data Science and Research Team has uncovered a key insight in the attendance space that allows us to predict the future absenteeism of a student. Students who miss school across many separate absence events are more likely to continue missing school than students who miss the same number of days in a single continuous stretch. This information powers our Attendance Forecast feature, and can be used to flag students likely to be chronically absent more accurately than methods currently deployed by most districts.

What is the Streak Concept?

The Streak Concept uses two simple metrics to identify attendance patterns that help forecast future absenteeism.

• Count of streaks: This is the number of distinct absence sets. It counts how many separate times a student has missed one or more consecutive days of school. 

• Longest streak: This is the length of the longest consecutive absence set. It measures the maximum number of school days a student has missed in a single, continuous period.

How the Streak Metric is Calculated

The streak metrics are straightforward to compute, which facilitates easily implementable differentiation among students with comparable absence rates.

• The calculation is based on individual student attendance records, utilizing only full-day absences tracked on weekdays. 

• Every absence is part of a single streak and a single day absence is a streak on its own.

• Multiple single day absences will increase a projected future absence rate.

• Partial absences or the excused/unexcused status are not considered in this approach. 

• The metrics can be calculated at any time during the year.

Example:

Student A misses 5 consecutive days. This student has 5 absences and 1 absence streak.

Student B misses Monday and Thursday of one week and Wednesday of the following week. This student has 3 absences and 3 absence streaks since none of their absences were continuous.

Effectiveness and Impact on Forecasting

The number of streaks proves to be a consistently impactful feature across all modeling techniques used to predict future absence rates, in contrast to less useful metrics like the day-of-week rate. The key insight is that attendance patterns provide additional predictive information beyond total absences alone.

Key Findings:

• Educators can use the ‘number of streaks’ indicator alongside simple current absence rate data to identify students likely to miss a significant portion of the school year with higher accuracy. 

  • This improved ability to differentiate students with similar rates makes the streak metric a valuable tool for early intervention.

• Holding total absences constant, students with more absence streaks are more likely to miss school during the remainder of the year. In the table below we can see how this looks at 30 days.

• Students with longer continuous absence streaks tend to have lower future absences rates than students with the same number of absences spread across multiple streaks.

  • This inverse relationship was included in our model that also included absence rate and number of streaks. In this context, the strength of this relationship is less powerful than the number of streaks. 

  • There is also a strong inverse relationship between the length of the longest absence and the number of streaks. Despite this, our model found that the inclusion of both metrics improves accuracy of the predictions so it is best practice to include both.

• Using streak features alongside the overall absence rate yields a higher R-squared value (0.508) in a regression forecasting the future absence rate. 

  • This can be used to rank students’ risk of future absence with a higher accuracy than just the absence rate.

• This method outperforms even models that utilize day of week absence information.

  • While Panorama also built a model that broke out absences by which day of the week they occurred, when we account for streaks this did not meaningfully improve the prediction accuracy in terms of R-squared.

Example:

Why This Matters

Even though Student A has more absences, students with attendance patterns like Student B missed an average of 13.3% of the remaining school year, compared to 7.65% for students with patterns like Student A.

Table: Future Absence Rate as of 30th Day of School Year with Streaks Breakdown. These are actual full-year rest-of-year absence rates calculated from a national dataset.

*An observation is a single unit of data collected. In this research, each observation represents one student in one school year. If a student appears in multiple school years, they contribute one observation for each year.