A postcard that goes to families beats a portal that waits for them

What mail did in randomized trials

The cheapest piece of mail with a randomized result behind it is a $0.22 postcard. Researchers ran a single-postcard trial across 51,197 students in the School District of Philadelphia and measured what happened next. Absences fell by about 2.4 percent (Institute of Education Sciences, Regional Educational Laboratory Mid-Atlantic, REL 2017-252, February 2017). The design was a randomized controlled trial, the strongest design education research has.

A longer trial raised the dose. Rogers and Feller mailed personalized attendance information to the households of 28,080 students across a school year, again with Philadelphia as the study site (Nature Human Behaviour, April 2018). Chronic absenteeism, which means missing 10 percent or more of the school year, came in at 36.0 percent in the control group and 31.9-33.0 percent across the treatment groups, roughly 1.1 fewer days absent per student in the information treatments. The full intervention cost $6.60 per household, a figure the authors report in the paper's free supplementary information (Table 29). The authors of that trial founded an organization that helps districts implement the intervention. The paper discloses the relationship.

The result has replicated. Robinson, Lee, Dearing and Rogers ran a randomized field experiment across 10 districts in grades K-5 and decreased chronic absenteeism by 15 percent (American Educational Research Journal, 2018). No What Works Clearinghouse rating exists for any of this, because the clearinghouse has never reviewed attendance messaging at all; the evidence stands on the trials themselves. Texting can be cheaper still where the district holds a working cell number. Bergman and Chan sent families weekly automated alerts about missed assignments, grades and absences; course failures fell by 27 percent, class attendance rose by 12 percent and student retention increased, though state test scores did not move. The program sent more than 32,000 messages for $63 in variable cost, at most $0.002 per message (computed) (Journal of Human Resources 56(1), 2021).

What the portal did in a randomized trial

Districts have also invested in the opposite design: a portal that holds the information and waits for the family to come collect it. Kraft and Bolves ran the first randomized field trial of implementation supports for one, across 132 New York City schools (Education Finance and Policy 17(3), 2022). The app auto-generated 91,317 parent accounts across the 122 of 132 study schools that completed registration, so signing up was never the barrier. In the study's control-group schools, which received the app with a standard rollout, 15 percent of parents ever logged in; schools given intensive implementation support reached 18 percent. One parent in twenty in those schools used the app for any action at all: a message, a post, a click, a like or a save. In seven schools out of ten, fewer than 20 percent activated.

The parents were behaving sensibly. Using the app meant installing it, keeping a password and checking it on a schedule, and the information inside was information the school could have mailed home. The study also tested the standard remedy: schools randomized to intensive implementation support, with training, guidance, monitoring and encouragement, moved staff activation by 7 percentage points and produced no measurable improvement in perceived communication quality. In the control-group schools, the average parent registered 0.57 interactions across the year, which this piece rounds to 0.6.

The families the trials could not reach

The strongest mail trial also documents who mail misses. To enroll families, the Philadelphia study had to produce its consent form in each household's home language, and it managed nine: English, Albanian, Arabic, French, Khmer, Mandarin, Russian, Spanish and Vietnamese. The study excluded 5,536 of 82,554 otherwise-eligible students because the household's language fell outside those nine or because home language was missing from the district's records. Computed, those exclusions come to 6.7 percent of the eligible sample, one family in fifteen (REL 2017-252, 2017).

The same report explains why the study used mail in the first place: at the district, guardians' home addresses are more available and accurate than guardians' cell phone numbers. The study also excluded students without reliable mailing addresses. The families hardest to reach were structurally absent from the trial, and the authors name that limitation themselves.

The legal duty and the languages it covers

Reaching those families is a legal obligation. Districts must ensure meaningful communication with limited English proficient parents, the legal term for parents whose English is limited, in a language they can understand, and they may not rely on students, siblings, friends or untrained staff to interpret. The Department of Education's Office for Civil Rights and the Department of Justice set this out jointly in a Dear Colleague Letter on January 7, 2015 (Section J). The letter interprets authority that no guidance withdrawal can touch: Title VI of the Civil Rights Act of 1964, the Supreme Court's 1974 decision in Lau v. Nichols, and the Equal Educational Opportunities Act. Footnote 103 of the same letter cautions against web-based automated translation, requires machine output to be reviewed and edited by a qualified individual, and flags the student-privacy exposure in uploading records to a translation service.

On an ordinary Tuesday, a family liaison has a discipline notice to send home, and the family's home language is Pashto. The district keeps templates in Spanish and Arabic. She cannot hand the notice to the student to interpret for his own parents; the letter names that practice specifically. She cannot paste it into a free translation site and hit send, because footnote 103 requires a qualified reviewer, and the notice carries student records she should not upload anyway. So she looks for a qualified translator for the notice. She will repeat the search the next time a message has to go home in a language the district has no template for.

Public schools enroll 5.3 million English learners, students still acquiring English, 10.6 percent of all enrollment (National Center for Education Statistics, Condition of Education, May 2024, on fall 2021 data). Spanish accounts for 76.4 percent, followed by Arabic, Chinese and Vietnamese. States plan around their five most common English learner languages, and fifty languages or language categories appear across those top-five lists (Office of English Language Acquisition, February 2023, on 2019-20 data). How much of a state's English learner population the top five covers varies widely, from just over 12 percent in Montana to almost 90 percent in Texas. In Montana, roughly seven English learner families in eight (computed) speak one of the many languages outside the five the state plans for.

Where the absence numbers live

Nationally, 22.6 percent of students were chronically absent in 2024-25, by the longest-running tracking available (Nat Malkus, American Enterprise Institute). For anyone choosing a channel, the concentration matters more than the level. High-poverty districts went from 19 percent chronically absent in 2018 to 36 percent in 2022, while low-poverty districts went from 10 percent to 21 percent (AEI). In FutureEd's March 2026 analysis of 26 states and the District of Columbia, low-income students rose more than 17 percentage points in 2021-22 alone and remain more than 9 points above their 2018-19 baseline in 2024-25, and English learners remain about 11 points above. Where states report 2024-25 levels by income, the spread is wide: Ohio, 33 percent of low-income students against 11 percent of affluent students; Rhode Island, 30 against 12; Washington, 35 against 19 (FutureEd, March 17, 2026).

The absence numbers are concentrated in low-income households and in English learner households (AEI; FutureEd, 2026). Every randomized result cited on this page moved attendance through mail or text, and the one district that audited its own contact data found guardians' mailing addresses more reliable than their cell phone numbers (REL 2017-252, 2017). A district acting on this evidence would spend on translation, current mailing addresses and postage. The best-tested item on that list costs 22 cents.

Methodology

• 6.7 percent excluded, and one family in fifteen. 5,536 excluded students divided by 82,554 otherwise-eligible students = 0.0671, or 6.7 percent. 82,554 divided by 5,536 = 14.9, so approximately one in fifteen. Both computed from counts published in IES REL 2017-252 (2017).
• $0.002 per text message, an upper bound. $63 in variable cost divided by 32,000 messages = $0.00197 per message, rounded to $0.002. The paper reports "more than 32,000" messages, so the true per-message cost is at or below this figure. Counts and cost from Bergman and Chan, Journal of Human Resources 56(1).
• Seven English learner families in eight in Montana. The OELA fact sheet (February 2023) reports that just over 12 percent of Montana's English learners speak one of the state's top five English learner languages. 100 minus 12 = 88 percent outside the top five, which is approximately seven in eight (7/8 = 87.5 percent).
• "All other languages" bar in Figure 5. Total English learners 5,263,596 minus Spanish 4,023,289, Arabic 130,917, Chinese 95,584 and Vietnamese 75,070 = 938,736. Component counts from NCES, Condition of Education (May 2024, fall 2021 data).
• Computed gaps in Figure 7's caption. Ohio 33 minus 11 = 22 points; Rhode Island 30 minus 12 = 18 points; Washington 35 minus 19 = 16 points. Levels from FutureEd (March 17, 2026).
• "One in twenty" and "0.6" in the portal figures. Kraft and Bolves (2022) report that in control-group schools one parent in twenty used the app for any action across the school year: a message, a post, a click, a like or a save. One in twenty is 5 percent, so the funnel stage is drawn at 5 percent. The control-group mean of 0.57 interactions per parent per year is rounded to 0.6 in this piece; the pooled mean across all study schools is 0.96.
• The $6.60 per household is a published figure, restated here as reported in the free Supplementary Information of Rogers and Feller (2018), Table 29, where it is built from $1.00 per mailing across an average 4.1 mailings plus $2.50 in labor. It does not appear in the paywalled body of the paper.
• Percent against percentage points. The postcard's 2.4 percent and the AERJ trial's 15 percent are relative reductions. The Nature Human Behaviour result (36.0 percent control against 31.9-33.0 percent treated) and the poverty-gap and baseline-gap figures (AEI; FutureEd) are levels or percentage-point differences.
• Study sites. Philadelphia and New York City appear here as the sites of published federal and peer-reviewed studies. Nothing on this page grades any district, and no figure here is a named district's own reported absenteeism rate; the 36.0 percent figure is the control group of a trial sample.
• Chronic absenteeism series. The 22.6 percent national level for 2024-25 is from Nat Malkus at AEI, whose tracker compiles state-reported data under definitions that vary by state. It appears in the body once, for context; this piece does not chart the time series.
• Disclosure. The authors of the Nature Human Behaviour trial founded an organization that helps districts implement the mail intervention. The paper discloses this; so do we.
• What Works Clearinghouse. No WWC intervention report or practice guide exists for attendance messaging. No sentence here implies a WWC rating.