Chronic absenteeism in Philadelphia schools

1. Headline takeaways

A convention for the figures below: annual district totals and subgroup totals sum eligible student records across schools (a student enrolled at least 10 days at two schools counts at each), so “students” in those sentences means student records. School medians and percentiles instead summarize school-level rates. Figure 8 uses the district’s separate monthly file and its within-month eligibility rule. The charter comparison uses the state’s file and its own rules, described where it appears.

• Improvement since the 2021-22 peak has nearly stalled. In 2021-22 — the worst year in the data, when Omicron disrupted schools — 42.1% of student records at Philadelphia district schools showed more than 10% of school days missed (a student enrolled at least 10 days at two schools counts at each). That share fell to 39.0% the next year, then 38.1%, then 37.7% in 2024-25: each year’s improvement smaller than the last, and the latest barely different from the year before. After three years, the rate has come down from 42.1% to 37.7%, about a tenth of the way to zero from its worst point. One caution: the district’s school-level data starts in 2020-21, so these figures have no pre-pandemic “normal” to compare against. The district’s own research office has reported that attendance was considerably better before the pandemic — in 2018-19, 97 of 241 district and alternative schools had more than three-quarters of their students attending 90% or more of days, against 48 in 2021-22 (its June 2023 report, cited on the methodology page; that report’s universe includes alternative schools, unlike this page’s) — and the state’s measure in Figure 9 reaches back to 2017-18. [Fig 1 · proof]
• At the typical school, about four in ten students are chronically absent — and at more than a quarter of schools, it’s a majority. At the median school, 39.0% of students missed more than 10% of days in 2024-25, down from 44.1% in 2021-22. And 62 of the district’s 216 schools (28.7%) still had more than half their students chronically absent — down only modestly from 83 schools at the peak. [Fig 1, Fig 6 · proof]
• Elementary schools are improving; high schools are not. Chronic absenteeism at Philadelphia’s elementary schools fell from 41.0% of students in 2021-22 to 33.0% in 2024-25, and at K-8 schools from 41.7% to 35.6%. High schools barely moved: 48.6% then, 47.8% now — nearly half of high schoolers, essentially unchanged in three years. New York City appears to have seen the opposite: on its own analysis page, its high-school grades improved the most (from 43.0% of students in 2021-22 to 34.1% in 2024-25), a bigger recovery than its elementary grades (38.8% to 31.7%). Two cautions about that comparison: the New York figures are the city education department’s own chronic-absenteeism rate, with whole-school rates grouped by each school’s grade band (elementary, middle, high) under New York’s category rules rather than Philadelphia’s school types, and the New York page still carries an unverified warning of its own — in particular, its two publishers disagree about whether grades 1–8 improved or worsened in 2024-25 — and the checks behind this page do not extend to it. [Fig 3 · proof]
• Absenteeism is lowest in the late-elementary grades and far higher in high school. In 2024-25, 36.5% of kindergartners missed more than 10% of school days, and 36.6% of first graders. The rate is generally lower through the elementary grades, reaching its lowest point in grade 5 (31.0%), and is much higher among 9th graders (44.2%) than 8th graders (33.8%). Through high school it stays high, with small dips between grades; nearly half of all seniors (49.3%) were chronically absent. These compare different students in the same year, not one group of students as they age. The improvement since 2021-22 is concentrated in kindergarten through 8th grade (kindergarten fell from 44.0% to 36.5%, and grades 6 through 8 improved by a similar amount as several elementary grades); grades 9–12 have barely moved. [Fig 2 · proof]
• Two ways of measuring attendance point the same way; we lead with the one that counts students, not days. The district also publishes each school’s average daily attendance — the share of enrolled days students actually attended. School by school, the two measures rank schools almost identically: schools with lower average daily attendance have more chronically absent students, with almost no exceptions. They answer different questions, though. Average daily attendance counts every missed day; chronic absenteeism counts how many students crossed a line. Since 2021-22, average daily attendance (averaged across schools by their student counts) improved from 87.7% to 89.2%, while chronic absenteeism fell from 42.0% to 37.7%. The threshold measure moved more, as it happened to here; that is a feature of how students were spread around the 90% line in these years, not a fixed conversion between the two. [Fig 4 · proof]
• Black students are chronically absent far more often than White students, and part of that gap is inside the same schools. District-wide in 2024-25, 45.5% of Black students missed more than 10% of days, against 23.6% of White students. That comparison mixes two things: how each group fares, and which schools each group attends (White students are concentrated in lower-absence schools, and the district withholds the White figure at about half of its schools). Narrowing to the 111 schools that report both groups, it’s 39.3% of Black students against 24.0% of White students, pooling each group’s students across those schools. Comparing the two groups school by school instead — giving each of the 111 schools equal weight — the average school’s Black rate was 37.7% and its White rate 29.0%, against 43.5% and 35.1% in 2021-22, so that within-school gap has barely changed. Weighting each school by its Black and White student records gives 35.2% against 29.0% in 2024-25 and 41.4% against 34.4% in 2021-22. On the 97 schools that report both groups in every in-person year, the equal-weight pair is 36.7% and 27.8% (2024-25) against 42.3% and 33.2% (2021-22). These are descriptive comparisons, not estimates adjusted for other differences between students or schools. The district’s file does not carry attendance broken down by family income, English-learner status, or special education; the state’s file offers a narrower version of those comparisons (Figure 10). [Fig 5 · proof]
• School composition tracks chronic absenteeism. Group the district’s schools into thirds by the share of students in special education: at the third with the fewest, 27.9% of students were chronically absent in 2024-25; at the third with the most, 54.3% — nearly double. Schools in the third with the fewest White students had a rate of 50.9%, against 26.6% at the third with the most. Schools with many English learners look no different from schools with few (37.1% vs 37.6%). These are descriptive patterns, not explanations — the full tables are in §3. [§3 tables · proof]
• A school’s chronic absenteeism rate doesn’t change much from year to year. Between 2023-24 and 2024-25, half of all schools saw their rate move by less than 2.3 on the 0-to-100 scale, and schools largely kept their standing: the ones with the highest rates one year had the highest rates the next. A meaningful minority did move — 52 of 216 schools by more than 5, and 12 by more than 10 — so a school’s rate is persistent, not fixed. Small schools move the most: among schools with fewer than 300 students, the spread of year-to-year changes (their standard deviation) is 2.4 times that at schools with 600 or more, and the typical (median) move is 1.8 times as large. Changes of a given size are simply more common among smaller schools in these data; read a one-year change alongside the school’s longer trend and its student population, because this size comparison does not say how much of a change is measurement variation and how much is real. [Fig 7 · proof]
• 2025-26: the best September–November in the monthly file, then record-high December, January, and March. District-wide, the average of the September, October, and November within-month shares — each the share of students enrolled at least 10 days that month who attended less than 90% of its school days — was 24.1% in 2025-26, the lowest of the six years in the file (the others range from 25.6% to 36.5%). Month by month, September, October, and November were the lowest among the fully in-person years (the mostly virtual 2020-21 year recorded a lower November and is excluded from this in-person-only month comparison). Then it reversed: December, January, and March were the highest for those months in the file — December at 44.9% against 33.2% a year earlier, January at 47.4% (a small margin over 47.1%), and March at 37.9%. February was not a record. Averaged over September through March, 2025-26 came out at 33.9% vs 32.3% a year earlier — the first time that comparison has worsened since 2022-23. Three cautions: these monthly figures aren’t comparable to the full-year rates above (a student can fall below 90% in one bad month but finish the year above it); the window averages are averages of monthly shares weighted by each month’s student count, with the same student counted in several months, not the share of students chronically absent over the window; and a month’s figure depends on how many school days it had (in a 20-day month a student can miss two days and stay at 90%, but in a shorter month two absences fall below it), so same-month comparisons are the safer reading. [Fig 8 · proof]
• Among Philadelphia schools matchable to the 2024-25 directory, the median reported share of eligible students below 90% attendance on the state’s measure was higher in 2023-24 than in 2018-19: 27.0% versus 21.1% at charter schools, and 38.5% versus 25.0% at district-run schools. The same broad pattern holds in a fixed panel of schools. Everything above comes from the district’s own file, which has no charter schools. Pennsylvania publishes a similar measure for every public school — the share of each school’s students, among those enrolled at least 90 days, attending 90% or more of days — so on that measure we can compare sectors. The state reports each year’s attendance in the following year’s file, so its latest figures describe 2023-24. The state publishes no student counts for this measure, so these are comparisons of school rates, not the share of all students in a sector: the median school is the cleanest summary, and our separate average that weights each school by its 2024-25 enrollment shows the same pattern (26.8% charter and 35.7% district-run in 2023-24, against 19.8% and 24.1% in 2018-19). Both sectors have improved since 2021-22 (enrollment-weighted averages of 33.7% charter and 38.5% district-run that year). The schools contributing change a little from year to year; holding the set fixed to the 72 charters and 214 district-run schools with a value in every year gives 19.4% to 26.4% and 24.1% to 35.3%, the same picture. These are schools that can be matched to the current directory; schools that closed, or that changed operator or state code (two Renaissance charters that returned to the district in 2022, for example), are missing for the years they can’t be matched (§5). The sector gap is not a measure of what charter schools do. Most charters (63 of 81) enroll by citywide lottery, while the other 18 are mostly Renaissance charters — former district schools handed to charter operators that still serve neighborhood catchments; students who leave a charter mid-year may return to district-run schools; the state’s 90-day rule leaves out any enrollment spell shorter than 90 days, so the most mobile students are only partly represented; and the state lets each school system decide how partial-day absences count, which we have not verified is done the same way across sectors. None of these has been measured here. Contracted and alternative programs (19 schools) are outside both sectors. And the sectors differ in shape: 53 stand-alone high schools are district-run (48 of them with a state value in 2023-24) against 9 charter, and 31 charters serve high-school grades in all — those 9 plus 22 that combine high-school grades with younger grades, which the like-for-like comparison in Figure 9’s note leaves out. [Fig 9, Fig 10 · proof]

2. Figures

3. Chronic absenteeism by school characteristics

For each characteristic below, the district’s schools are split into thirds — the 72 schools with the lowest share, the middle 72, and the highest 72 — and each third’s chronic absenteeism rate is the share of its student records (summed across its schools; a student counts at each school where enrolled 10 or more days) that showed less than 90% attendance in 2024-25. One characteristic we can’t use: family income. The district records 188 of its 216 schools as exactly 100% economically disadvantaged, so that measure can’t tell schools apart; the district’s file does not document how the figure is derived.

How these tables were made →

These are descriptive groupings, not explanations. In particular, school type is tangled up with all of them — high schools have both the highest absenteeism and different student populations than elementary schools — so none of these tables says what causes what.

5. How we checked these numbers

The Philadelphia figures on this page are generated by a single script from the publishers’ files into a committed data packet (the New York figures come from that city’s separate analysis), and two checks follow. Inside the script, the district-wide series is recomputed from the exact published student counts and compared with the version built from the published percentages: 15 comparisons (the rate, the school count, and the student count for each year), all 15 within tolerance (the largest rate difference was 0.0003, against a tolerance of 0.01). Separately, a second program written in a different language re-reads every attendance cell from the publishers’ files and recomputes a specified list of the packet’s values — the series behind each figure and the fields the wording is composed from; its results are in the status strip at the top and in the QA report, which also lists the inputs and comparisons it does not check. The checks also compare which schools appear in the district’s file against which appear in our database — that comparison caught 2 school-years missing from our database (a school that has since closed), which this page’s figures now include.

Two publishers, two numbers. For the 216 district-run schools that appear in both the district’s file and the state’s, we compared the two 90%-attendance figures for the same year, 2023-24 (the state’s comes from its 2024-25 file, because the state reports each year’s attendance a year later). They track each other closely school by school: 27.8% of schools agree within 1 on the 0-to-100 scale, 97.2% within 5, and 0 differ by more than 10. At the typical school the state’s figure sits 1.7 below the district’s (fewer students counted below 90%), though individual schools fall on both sides. One plausible reason is that the state counts only students enrolled at least 90 days while the district counts students enrolled at least 10; we have not measured how much of the difference that explains, and differences in how the two count partial days are another candidate. The same comparison for every overlapping year from 2020-21 on gives a median difference of similar size and the same sign each time; the large break an earlier version of this page reported disappears once the years are aligned. That removes the evidence for the alleged break; it does not audit every local attendance policy, and the check covers district-run schools only. (An earlier version of this page matched the state’s file years to the district’s attendance years — different school years — and wrongly reported a much larger gap in 2020-21 and 2021-22 and a change in the state’s rules; that finding is withdrawn, and the correction is logged on the methodology page.) The page still treats them as two measures: Figures 1–8 use the district’s file; Figures 9–10 use the state’s. Full check: cross-publisher attendance report.

How the checks work → · Every served number on this site has its own trace: proof trees