Abstract
Student absenteeism has remained high following the COVID-19 pandemic, and districts need low-cost strategies to improve attendance. During the 2022–2023 and 2023–2024 school years, the National Center for Rural Education Research Networks (NCRERN) conducted a randomized trial of a personalized messaging intervention that had promising results in an early pilot with eight rural districts. NCRERN worked with a student information system to test the intervention in 47 districts in 16 states. We find that the personalized messages reduced student absences by between 1.7% (p < .05) and 4.5% (p < .05) and cost, on average, $4.07 per student. While generally cost-effective, these effects represent a modest 0.2 to 0.5 days saved. We discuss implementation challenges and heterogeneous effects that suggest caution prior to full-scale adoption.
Even before the pandemic, school leaders sought to address high rates of absenteeism, with many states including chronic absenteeism in their Every Student Succeeds Act (ESSA) accountability measures (Jordan & Miller, 2017). However, this challenge has been growing in recent years, particularly with the start of the COVID-19 pandemic (Carminucci et al., 2021). Between the 2017–2018 and 2021–2022 school years, chronic absenteeism increased 47% among rural students, 44% among students in urban districts, and 46% among students in suburban areas (Bay, 2023). Nationally, in 2024, chronic absenteeism remained 57% higher than prepandemic levels (Malkus, 2025). Such dramatic—and persistent—increases are concerning and warrant research into promising strategies to reduce absenteeism.
District leaders are confronting the challenge of rising student absenteeism as state and federal pandemic relief dollars expire in early 2026. During the pandemic, the federal government invested $190 billion in Elementary and Secondary School Emergency Relief (ESSER) funds to accelerate student learning following the pandemic. While these investments led to improvements in math and reading scores (Dewey et al., 2024), educators still need low-cost, sustainable solutions to pressing challenges such as absenteeism. In rural districts, existing challenges around funding and staff capacity further emphasize the need for solutions that are inexpensive and easily integrated into existing school practices (Best & Cohen, 2014). Rural communities are shaped by unique strengths and challenges that demand evidence tailored to inform rural praxis (Azano & Biddle, 2019). In this context, there is a clear need for rigorous causal research establishing the viability of low-cost strategies that can be feasibly implemented by rural educators to reduce student absenteeism.
In urban districts, nudge interventions (those that focus caregiver attention on their children’s absences and the importance of regularly attending school) have been shown to improve student attendance (Bergman & Chan, 2019; Heppen et al., 2020; Rogers & Feller, 2018; Rogers et al., 2017). However, research in rural districts is scarce, and the rural context may mediate the effect of nudges. In rural communities, transportation barriers may loom larger than in urban or suburban areas, as rural school districts tend to encompass larger geographic areas (Gottfried et al., 2021), which may pose an insurmountable logistical barrier regardless of whether or not caregivers’ attention is focused on attendance. On the other hand, rural districts have the advantage of maintaining close relationships between teachers, their students, and families (Barley & Beesley, 2007; Sheldon, 2007). While these relationships are potentially important for attendance (Smythe-Leistico & Page, 2018), there is limited research on whether nudges are helpful or harmful in maintaining those relationships. The efficacy of these interventions may also be blunted in rural contexts if they do not ultimately provide new information to families that they haven’t already heard from their child’s teacher.
In the 2020–2021 school year, eight rural districts in New York and Ohio, working with the National Center for Rural Education Research Networks (NCRERN) piloted a personalized messaging intervention in which caregivers were sent regular updates via text, email, or phone call about their child’s cumulative absences. The messages also asked caregivers to set a goal for attendance, offered encouragement, or provided a comparison to the grade or school, and they ended with encouragement to contact the school to strategize for better attendance. Using a classical frequentist hypothesis test, the messages were found to lead to a statistically insignificant decrease in absences. However, using a Bayesian framework with relatively weak priors, we found that the intervention likely did decrease absences (with a certainty level of 83%), with a point estimate of a 2.4% decline in absences (Swanson, 2023). Given these promising findings, NCRERN partnered with a student information system that offers a built-in messaging platform to replicate the intervention at scale and across rural contexts. In this paper, we present results from Leveraging Interactions with Families To (LIFT) Up Attendance, a 2-year national replication study of this intervention.
LIFT Up Attendance was implemented in 47 rural districts across 16 states in the 2022–2023 and 2023–2024 school years. All districts already used the same student information system (SIS) with messaging capabilities. The SIS provider, Infinite Campus, integrated the LIFT Up message templates into their platform, allowing districts to send personalized messages to a randomly assigned treatment group of students in grades K–12. Messages were sent every 4 to 6 weeks and included information about the student’s cumulative absences over the prior 4 to 6 weeks, the importance of attendance, and an invitation to connect with the school. Messages were differentiated, providing encouragement for continued high attendance for students with 0 to 1 absences and offering a goal for reduced absenteeism for students with 2 or more absences.
While LIFT Up Attendance was designed as a low-cost, easy-to-execute intervention, districts faced significant implementation challenges each year. In 2022–2023, the launch was delayed due to technical issues integrating the message templates into the SIS. In 2023–2024, the filters used to differentiate messages between students who had 0 to 1 days absent and students with 2 or more days absent in the preceding 4 to 6 weeks were inconsistently applied by district implementers. In both years, there was some treatment noncompliance, with students assigned to treatment not receiving messages and students assigned to the control condition receiving messages. Despite these implementation challenges, the intent-to-treat (ITT) estimates suggest personalized messaging reduced absences by 1.7%, an effect that is statistically significant at the 5% level. Treatment-on-treated (TOT) estimates from instrumental variables analysis indicate the effect of LIFT Up Attendance was larger, reducing absences by 4.5%, also statistically significant at the 5% level. We find evidence of heterogeneity in effects by student gender, race/ethnicity, receipt of free or reduced-price lunch (FRPL), receipt of Special Education services, and whether the student was chronically absent in the prior year. Effects were nominally larger for students receiving FRPL or Special Education services, as well as for students who were chronically absent in the prior year. We also see nominally larger effects for Indigenous students and nominally smaller effects for Black students than for White, Asian/Pacific Islander, Hispanic/Latinx, and Multiracial students. Additionally, we see that the effects of personalized messaging were greater in districts with higher implementation fidelity. Cost analyses suggest that LIFT Up Attendance was a low-cost intervention—on average, the cost to implement was $4.07 per student.
This study provides some of the first nationally representative causal evidence on the impact of caregiver nudges to reduce student absenteeism in rural areas. We discuss the challenges of implementing a new intervention, even one that leverages existing systems and is designed to require minimal effort. These results fill a critical gap in the literature about effective practices for increasing student attendance and offer key design insights for future rural research.
Relevant Literature
Causes of Absenteeism
Each day a student is out of school is the result of macro-level factors (such as local transportation infrastructure and housing), family-level factors (such as adult work schedules and relationships with teachers and school leaders), school-level factors (such as bus schedules, school climate, and disciplinary policies), and student-level factors (such as feelings of safety and engagement with school; Attendance Works, 2022). These factors are mediated by rural contexts. Given the travel distance to school in many rural communities, transportation barriers are a more prevalent challenge than in urban or suburban areas (Gottfried et al., 2021). With increases in remote work, many rural areas are experiencing rapid population growth (Davis et al., 2023), with the pace of demographic and economic changes accelerating in many communities following the pandemic. With the changing community, educators must work to build new relationships as their connections with the community are a key strength of rural education (Azano & Biddle, 2019; Sheldon, 2007) and a valuable leverage point for reducing absenteeism (Smythe-Leistico & Page, 2018).
Students may disengage from school if they are not engaged by their classes and lack meaningful relationships with their teachers and others at their school (Attendance Works, 2022). Rural schools may struggle to attract and retain teachers in specialized areas, leading to a narrower range of course offerings, limiting students’ options, and offering an incomplete match to their interests. For example, rural schools typically offer fewer career and technical education courses than non-rural schools (Arneson et al., 2020). While student disengagement is infrequently reported as the reason for absenteeism, recent national surveys suggest there has been an increase following the COVID-19 pandemic in the share of elementary students reporting they miss school because it is “boring” or because they have “given up,” though these results are not disaggregated by locale (Gee et al., 2025).
Family-level factors driving student absences include misconceptions about the school’s attendance policies, an underestimation of their child’s absences, or a lack of information about the adverse effects of absenteeism (Attendance Works, 2022). Caregiver attitudes toward absences may have changed since the COVID-19 pandemic, particularly around when an illness is severe enough to warrant keeping a child home from school. This pattern has been reported by district leaders and on some surveys of adults and youth (e.g., Diliberti et al., 2025; Gee et al., 2025). Students’ non-school responsibilities and structures may also be shifting in the wake of the pandemic, with more elementary school students reporting they have missed school because they had to care for others or had not gotten enough sleep in the 2022–2023 and 2023–2024 school years relative to pre-pandemic years (Gee et al., 2025).
Providing information to caregivers about student absenteeism improves attendance (Bergman & Chan, 2019; Heppen et al., 2020; Musaddiq et al., 2023; Rogers & Feller, 2018; Rogers et al., 2017), but these initiatives have only been evaluated in urban settings. Much of the research is also from prior to the COVID-19 pandemic, although recent work follows this trend as well (SchoolStatus, 2025). As caregivers tend to have stronger existing relationships with schools in rural areas (Azano & Biddle, 2019; Sheldon, 2007), this type of intervention may present less of a contrast with and represent a smaller value-add to existing practice in rural schools.
The resources available to schools when attempting to address challenges to student attendance vary by locale. In rural areas, staff often already feel overburdened with responsibilities, with this driving the exits of many rural teachers (Hammer et al., 2005; Lazarev et al., 2017). Rural districts also often confront shrinking budgets and high fixed costs for transportation and other student services impacted by geography (Biddle & Azano, 2016). There is also a gap in the research literature about effective strategies for reducing absenteeism in rural areas and how they might be sustainably implemented. Much of the existing literature about the factors contributing to absenteeism was conducted in urban areas, while studies aimed at understanding how these factors manifest in rural contexts focus on a specific rural community. While this analytic approach allows for a rich and nuanced exploration of that community, there is a need for broader analysis that probes the root causes of rural student absenteeism across contexts.
Promising Strategies to Reduce Absenteeism
Although student absenteeism is rooted in multiple complex, interlocking causes, prior research has found that low-cost, low-touch interventions can yield modest gains in attendance. Behavioral nudges offer support through mechanisms such as information availability and goal setting, with the aim of encouraging individuals to make positive behavioral changes (Damgaard & Nielsen, 2018).
As students form their early perceptions about the importance of school and being present in the classroom, families play a critical role in ensuring students develop positive behaviors toward school attendance and academic achievement (Houtenville & Conway, 2008). However, families often hold misconceptions about their child’s attendance relative to their peers: two-thirds of caregivers of students with above-average absences reported that their child had an absence rate lower than their fellow students (Rogers & Feller, 2018). Informational nudges integrated into schools’ existing communications systems have the potential to influence the behavior of students and families in a low-cost, low-effort way.
The appeal of low-cost interventions with flexible options for delivery has yielded a growing literature on the effectiveness of nudges in urban educational settings. In an evaluation of a program that sent personalized postcards home to families, Himmelsbach et al. (2022) found a reduction in absences of 8.3% for students in primary grades. High-frequency, information-based interventions that provide caregivers with messages containing information on their child’s academic performance and attendance, such as those studied by Bergman and Chan (2019), have increased students’ presence at school by 12%. Similar interventions have decreased rates of chronic absenteeism for students whose caregivers receive periodic mailings about their child’s school attendance (Rogers & Feller, 2018; Rogers et al., 2017). Across studies, interventions that provide regular, accurate information about students’ cumulative absences and the importance of attendance reduce student absences (Kurki et al., 2021; Musaddiq et al., 2023; Robinson et al., 2018).
Only one study has evaluated the impact of personalized messaging in rural districts (Swanson, 2023). In this trial, personalized messaging was estimated to lead to a smaller estimated decrease in absenteeism than had been previously documented, about a 2% decrease and not statistically significant, although a Bayesian analysis found there was an 83% certainty that the intervention reduced absences relative to status quo operations (Swanson, 2023). That trial was conducted in the 2021–2022 school year, when schools were facing the acute challenges of the COVID-19 pandemic, including finding ways to safely return to in-person learning. The consistent success of personalized messaging in urban settings and the promise of personalized messaging in rural districts suggest that further evidence is needed on the efficacy of such an intervention in rural districts.
We fill this gap in the literature through a national evaluation of a personalized messaging intervention in a rural setting. We also add to the literature by estimating the costs of the intervention, including the dedication of staff time to implementation. Finally, we provide detailed implementation data highlighting the challenges of starting any new initiative, even one designed to be low effort. Together, this study makes a significant contribution to the literature about how rural districts can reduce student absenteeism in an educational landscape still recovering from the COVID-19 pandemic and facing a perilous financial outlook.
LIFT Up Attendance
The personalized messaging intervention evaluated here is an informational nudge aimed at caregivers/family members of students in grades K–12. Three message templates were included as part of the intervention. First, an initial welcome message personalized with the caregiver’s name, providing an overview of the study and its objectives and saying that their student (personalized with the student’s name) had been randomly selected to participate. The welcome message was to be sent at the beginning of the school year. Next, recurring messages were sent to caregivers every 4 to 6 weeks with the exact cadence chosen by implementing districts. There were two recurring message templates. If the student had been absent 0 or 1 days in the prior 4 to 6 weeks, the caregiver received a message congratulating the student on their attendance and encouraging them to maintain their attendance in the coming period. If the student had been absent 2 or more days, caregivers received a message with five core components: (a) the student’s name; (b) the number of days the student had been absent in the preceding 4 to 6 weeks; (c) a reminder of the importance of attendance; (d) a goal for attendance in the next 4 to 6 weeks (intended to be realistic and attainable, based on the number of days the student had been absent); and (e) encouragement and contact information to connect with the school to discuss the student’s attendance further (see Supplemental Appendix C [available in the online version of this article] for the message templates used in LIFT Up Attendance).
The standard message templates were available through Infinite Campus’s messaging platform. Districts could personalize the templates with the name of the person sending the message, a reference to a school mascot or slogan, or with additional information. The templates automatically pulled in the student’s name, school/district name, number of absences in the preceding period, and the goal for absences in upcoming period (1 day if the student had been absent 2 to 4 days; 2 days if the student had been absent 5 to 7 days; or 3 days if the student had been absent 8 or more days).
Districts could send the messages by manually pressing “send” in the messaging platform or by scheduling the messages in advance for the entire school year. Importantly, for the recurring messages, districts needed to select the appropriate filters associated with the message templates to ensure that caregivers received either the message congratulating the student for missing 0 or 1 day or the message informing them of the number of days the student had been absent and setting a goal for the upcoming month. Districts could send messages as emails, texts, or robocalls; in the implementation data, we see that all districts sent the messages via email, with some adding other modalities.
We theorized that the messages would reduce absenteeism through four key mechanisms. First, by providing cumulative absence data, caregivers would have an accurate, holistic picture of their student’s attendance. This would go beyond typical school communications around absences, which tend to inform caregivers of an absence the day it occurs or when the student is nearing the threshold for truancy, rather than providing proactive, cumulative information. Cumulative information helps reinforce the magnitude of the student’s absenteeism, potentially making the issue more salient for caregivers. Reporting cumulative absences for each messaging period of 4 to 6 weeks rather than the year-to-date was intended to increase caregiver motivation to improve absences. By giving caregivers a “clean slate” each reporting period, caregivers of students with high absences earlier in the school year could receive recognition for improved attendance instead of a steady stream of messages reporting high absences, which could be demotivating. Second, by reinforcing the importance of attendance, the messages could increase caregivers’ motivation to encourage students to attend. Third, by setting a goal for the next month, caregivers may be prompted to formulate clear action steps (such as adjusting bedtime and morning routines or creating contingency transportation plans if their child misses the bus). Fourth, by explicitly welcoming the caregiver to connect with the school, the messages would help strengthen trusting relationships between caregivers and the school. Together, these three mechanisms were intended to make the problem of absenteeism more salient to caregivers, drive action, and make it easier for caregivers to request and receive additional support from the school to address barriers to attendance. Figure 1 illustrates the theory of change of personalized messaging.
Personalized messaging theory of change.
While we cannot explicitly test the mechanisms by which LIFT Up Attendance affects student absences, we evaluate here the impact of personalized messaging on our distal outcome of interest, the number of days a student is absent in the treatment period. Next, we discuss the context in which LIFT Up Attendance was implemented.
Study Context
In 2019, the federally funded NCRERN launched its first rural research network with 49 districts in New York and Ohio, with a focus on understanding and reducing student absenteeism. Partner districts in the rural research network engage in a data-driven continuous improvement process, through which researchers and practitioners identify potential areas of improvement, design an intervention based on prior literature, test the intervention, and adjust successive interventions based on findings. Through this model, districts receive coaching and collaboratively problem-solve to design, implement, and evaluate rural-specific strategies. Eight districts in the network piloted the original personalized messaging intervention in the 2021–2022 school year, with support from NCRERN staff. NCRERN launched its national replication study during the 2022–2023 school year in 21 districts in five states. After the first cohort completed participation in the study, NCRERN implemented and tested the intervention with a second cohort of 42 districts in 2023–2024. Unlike the original districts in New York and Ohio, replication network districts received minimal ongoing support from the NCRERN team.
To scale the personalized messaging intervention, NCRERN partnered with Infinite Campus, an SIS with an integrated messaging platform. We recruited rural districts based on National Center for Education Statistics (NCES) locale codes (Geverdt & Maselli, 2024) for administrative ease and cross-state consistency, while acknowledging limitations of this classification system. Specifically, rurality definitions in education research vary considerably and remain contested (Grant et al., 2024), and district-level designations may not align with individual school contexts—rural-designated districts can contain non-rural schools, and rural schools can exist in non-rural districts (Greenough & Nelson, 2015). We also limited our recruitment to districts that were already Infinite Campus Messenger users. We did so to facilitate data access (a formidable barrier given the number of rural districts needed for power) and to reduce implementation challenges, as districts would already be familiar with the messaging platform. When the replication network was launched, Infinite Campus had contracts with over 2,000 rural districts in all 50 states, representing about a quarter of the almost 8,100 districts classified as “rural” by the NCES (2023).
District recruitment was led by the study team. Study team members reached out to all rural districts that had active contracts with Infinite Campus via email and set up calls with district leaders who expressed interest in the intervention. The study team also held informational webinars and publicized the study through national channels, including the NCRERN newsletter and the National Forum to Advance Rural Education. While all data were provided by Infinite Campus, this project was done with the written consent of the districts involved, each of which signed a data use agreement with the study team. Sixty-three districts initially agreed to participate in the personalized messaging study; this study is restricted to the 47 districts in 16 states that ultimately launched the intervention. These 16 states represent diverse rural contexts, including the Midwest, South, West, and Northeastern regions of the United States. Districts range in size: the average district size is about 2,300 students, with enrollments ranging from under 50 students to almost 6,000 students. Table 1 summarizes the characteristics of participating districts by year and offers a comparison to the universe of rural districts in the United States in 2023–2024.
District Characteristics, by Year of Participation
Note. Share of ELL students unavailable for 2022–2023 study cohort. Sample includes all students in randomization. National rural absence and chronic absence rates from June 2024 NCES School Pulse Panel survey. National rural school level, race/ethnicity, and gender are from 2023–2024 Common Core of Data, restricted to regular public schools coded as “elementary, middle, high, or secondary.” National stats represent all students in a rural school (NCES locale codes 41, 42, and 43), with the number of districts representing unique Local education agencies (LEAs) with at least one rural school. National rural stats for students with disabilities and ELL students reflect shares of students in rural districts in 2019, as reported by NCES (2023); SDs not available. ELL = English language learner; NCES = National Center for Education Statistics.
As shown in Table 1, there are differences in sample composition across the two cohorts of participating districts. The 2022–2023 cohort has a higher average prior-year absence rate, a higher prior-year chronic absenteeism rate, and a substantially higher share of students of color (52% compared to 26% in 2023–2024). The 2023–2024 cohort is larger and has a higher share of Indigenous students but dramatically lower shares of Latinx/Hispanic and Black students. There is substantial variation in student characteristics across districts; for example, while 23% of students in the overall 2022–2023 sample are Black, 89% of Black students in the 2022–2023 sample are enrolled in a single district. In the 2023–2024 sample, the share of White students enrolled in districts ranges from 1.4% to 97%.
Our pooled randomization sample broadly reflects the national population of students (shown in the final column of Table 1) in rural K–12 schools in terms of race/ethnicity, gender, receipt of Special Education or English Language services, and overall attendance rate. Our sample diverges from the national population by including a higher share of high school students and a lower share of middle school students, a higher rate of FRPL receipt, and a higher chronic absenteeism rate.
Implementation
This study was designed to evaluate personalized messaging under real-world conditions, with implementation led largely independently by participating districts. Districts received written guidance about how to implement the intervention from the study team, participated in an initial 1-hour training session with the study team, were encouraged to have a one-on-one support call with Infinite Campus to set up the templates, and could request support from either the study team or Infinite Campus (which was part of districts’ existing contracts with the provider) at any point during the implementation period. Given this, the results of this study should be understood as the impacts of personalized messaging in rural districts with typical resource constraints and external support.
LIFT Up Attendance was designed as a low-effort intervention that rural districts could incorporate into their existing practices without significant additional resources. However, launching any new intervention can be a challenge, and personalized messaging was no exception. There were several threats to the fidelity of implementation, both on the technical side for the provider and on the logistical side for implementing districts.
The message templates included several new custom fields that had not been previously generated in the Infinite Campus messaging platform, including the calculation of cumulative absences, the targeting filter directing a specific message template based on the student’s prior absences, and the adaptive goal based on the student’s prior absences. Creating these fields was ultimately a more complex and lengthy process than anticipated, leading to delays in the release of the message templates to districts participating in the 2022–2023 school year. The message templates were expected in August 2022, prior to the start of the school year, but were not released until October 2022. In states with some level of review of SIS updates, districts could not access the templates until up to a month later. This led to significant delays in launching the intervention (e.g., sending the initial welcome message) and reduced the treatment period from the full school year to academic quarters two through four of 2022–2023. In the 2022–2023 cohort, welcome messages were sent between October 7, 2022, and January 30, 2023. For the 2023–2024 cohort, templates were available from the beginning of the school year, and most districts launched between August and October; 30 districts sent the welcome message between August 1, 2023, and October 27, 2023, with the remaining eight sending the first message between November 3, 2023, and January 23, 2024. School start dates vary by state and district but typically fall between early August and early September. The delayed release of the templates affected implementation fidelity in 2022–2023. The recommended cadence was to send messages every 4 to 6 weeks (e.g., between six and nine rounds of messages). On average, districts in 2022–2023 sent messages 3.7 times throughout the year, with districts sending messages between one and nine times. We classify districts that sent messages at least three times as “high implementers” for 2022–2023. Implementation was stronger in the second cohort. In 2023–2024, 30 of 38 districts sent at least six rounds of messages, with all districts sending messages between 2 and 10 messages throughout the year. We classify districts that sent messages at least six times as “high implementers” for 2023–2024.
Along with delays in the release of the message templates, there were coding errors in the message text. Initial messages for the 2022–2023 cohort included typos and pulled a student’s cumulative absences since the beginning of the school year rather than since the last message had been sent (as was referenced in the text). These errors were corrected by January 2023 and did not affect the spring semester of the first cohort or the second cohort.
On the logistical side, districts encountered two main barriers to high-fidelity implementation. First, when sending messages, districts needed to apply the targeting filter, which would ensure caregivers of students who had been absent 0 or 1 day in the prior month received a message recognizing their accomplishment and encouraging continued strong attendance, and that caregivers of students who had been absent 2 or more days in the prior month received a message informing them of their student’s cumulative absences and setting a goal for the coming month. This filter was not consistently turned on, meaning that caregivers sometimes received both messages at the same time. Often, this had to do with the way in which the message templates and filters were released to the districts—the SIS could only push the templates to each participating district’s administrator group, and if the staff member responsible for sending the messages was not part of that group, a district administrator needed to copy the message templates and filters into the appropriate user group. If the message templates were copied over without the filters, then both messages would be sent to caregivers. If a staff member tried to send messages without the filter, they would get a warning message, but could still push out the messages. This was primarily a challenge in the 2023–2024 cohort, with the share of the treatment group affected varying by month from none to 17% (in up to seven districts).
The second logistical challenge faced by districts was a lack of contact information for students’ caregivers. There were a few causes for this lack of coverage. When districts first purchase Infinite Campus’s messaging capabilities, the default is to release the platform to families as an opt-in opportunity to hear from the school. Districts can change the default to opt-out (recommended by Infinite Campus), but if they do not, they can only send messages (including personalized messaging, although not specific to this intervention) to those who have opted in. Lack of coverage can also stem from having out-of-date contact information or from families not providing any contact information. The lack of contact information meant that sometimes substantial shares of caregivers whose students were assigned to the treatment group never received personalized messages. In the 2022–2023 cohort, in four of nine districts, fewer than half of the caregivers of treatment group students received messages. In the 2023–2024 cohort, <50% of caregivers of students assigned to treatment received any message in 9 of 38 districts.
These implementation challenges may dilute the treatment effect we estimate in this study, but they offer important lessons for districts and researchers looking to evaluate the efficacy of new interventions. Regardless of how straightforward and low effort the intervention is designed to be, there are significant barriers to start-up that need to be proactively mitigated.
Data, Methods, and Analytic Approach
We evaluate LIFT Up Attendance through a student-level randomization. Here, we describe our analytic sample, randomization procedure, and approach to impact estimation.
Sample Description
We began with 41,990 K–12 students in standard learning environments (excluding those in alternative learning environments, homeschool students, etc.) across the 47 districts that participated in the LIFT Up Attendance trial. We had minimal student-level attrition from students exiting the districts: 317 students (<1%) had 0 days enrolled in a participating district during the treatment period. An additional 30 students were dropped because they did not have any attendance data for the treatment period. Finally, we dropped 175 students (less than half a percent) who had more absences recorded than days enrolled. This limits our analytic sample to 41,468 students across 47 districts.
Students are evenly distributed across grades, with about 7% to 8% of the sample in each grade kindergarten through 12th grade; this means that when we look at schooling levels (elementary school grades K–5; middle school grades 6 to 8; and high school grades 9 to 12), we have a larger share of elementary students (45%) than middle school students (24%) or high school students (31%). On average, students missed about 7% of days in the year prior to the intervention launch, and there was an 18% chronic absenteeism rate across participating districts. Just over 40% of students receive FRPL, 16% have an Individualized Education Program (IEP), and 6% have received English Language Learner services. About 66% of students in the analytic sample are White, 15% are Latinx/Hispanic, 9% are Black, 5% are Indigenous, 5% are Multiracial, and 1% are Asian, Native Hawaiian, or Pacific Islander.
Randomization
Randomization occurred through a filter connected with the message templates. Students’ treatment assignment was based on their ID in the SIS, and school staff did not know which students were assigned to the treatment or control groups. Balance was achieved through randomization, as shown in Table 2.
Balance in Analytic Sample
Note. District fixed effects included in the joint omnibus test.
An omnibus balance test across the two cohorts, including student characteristics and district fixed effects, confirms balance in our analytic sample, with an F statistic of 0.64 and a p-value of .986.
While randomization was successful, there were threats to compliance. As randomization was linked to a filter accompanying the message templates, it had to be consistently turned on with each round of messages sent by the district (the same as the targeting filter discussed above). This was not always the case. In three of nine districts in 2022–2023 and in 12 of 38 districts in 2023–2024, the randomization filter was not turned on for at least one round of messages, meaning that caregivers of students assigned to the control group also received messages in that round. There is no evidence that districts selectively turned off the randomization filter to make exceptions for specific students they thought would benefit from the intervention, but this does limit the treatment-control contrast in these districts. Table A1 in Supplemental Appendix A (available in the online version of this article) shows compliance by district.
Analytic Approach
We estimate the impact of personalized messaging on the number of days a student is absent during the treatment period. We include all absences, whether excused or unexcused, in this count. Days absent are calculated in the system based on the percentage of minutes a student was present in the school during the school day; schools enter the expected number of instructional minutes per day when they first set up the system (or, a default of 240 minutes per day is used). The number of minutes the student is in class is then recorded as the numerator; absences are rolled up to half days and reported to us by term. We sum these absences over the total treatment period to calculate the cumulative number of days absent to use as our outcome measure. As our outcome variable, cumulative days absent, is a count variable, we estimate all results using Poisson models.
ITT Effects
We first estimate ITT effects of LIFT Up Attendance. 1 We estimate ITT effects following Equation 1.
Our outcome represents the number of days student i was absent during the treatment period, using the calculation described above. The treatment period varies across the two cohorts, given the delays in launching the intervention from 2022–2023. We include all absences in quarters two through four for students in the 2022–2023 cohort and all absences in the school year for students in the 2023–2024 cohort, reflecting the timing of the start of implementation in each year. Importantly, the treatment period begins in the quarter when the first welcome message is sent, which does not include information about students’ absenteeism. Our estimates are thus conservative in that they include absences accrued before any message was sent in the first treatment quarter, as well as absences accrued before the first message providing personalized information about students’ absenteeism, which is a key theorized mechanism for impact. In an exploratory analysis, we also look at the impact of the intervention on students’ likelihood of being chronically absent in the treatment year using a Probit rather than a Poisson.
The randomized design identifies the effect of
We conduct exploratory subgroup analyses by interacting the treatment indicator with an indicator for each subgroup of interest. We show estimated ITT effects by student gender, race/ethnicity, grade level, FRPL receipt, and prior-year chronic absenteeism. We include results from post hoc tests comparing estimated effects by subgroup to each other.
TOT Effects
Given noncompliance in our sample—both students in the control group receiving treatment and students in the treatment group not receiving treatment—we also estimate TOT estimates. We use the random assignment as an instrument for determining whether a caregiver received any personalized message about their student from the school. Our measure of receipt is somewhat limited: we observe whether the message was sent to the caregiver but not whether it was opened or read. We use the same model specification as in Equation 1, substituting
Restricted Samples
Another strategy we use to address randomization noncompliance is to limit our sample to those districts with a valid randomized controlled trial. Since there is no standard threshold of how much noncompliance is too much, we test the sensitivity of our findings on different samples determined by randomization of compliance within districts. Specifically, we estimate ITT and TOT effects as described above on four subsamples of districts:
Districts in which no caregivers of students assigned to the control group received messages AND in which at least 50% of caregivers of students assigned to the treatment group received messages.
Districts in which there is at least a 5-percentage point contrast between the share of caregivers who received messages whose students were assigned to the treatment and control group, respectively.
Districts in which there is at least a 25-percentage point contrast between the share of caregivers who received messages whose students were assigned to the treatment and control group, respectively.
Districts in which there is at least a 50-percentage point contrast between the share of caregivers who received messages whose students were assigned to the treatment and control group, respectively.
We present results from the first restricted sample (none of the control group treated and at least half the treatment group treated); results from the other three samples are shown in Supplemental Appendix A (available in the online version of this article).
Robustness Checks
We test the sensitivity of our results through a few checks. First, we estimate more parsimonious specifications of our main ITT and TOT models, fitting models that include no covariates, district fixed effects only, and district fixed effects and prior-year absence splines in addition to the fully specified model. The fully specified model described above was included in our pre-registration plan and is our preferred model.
Finally, as a sanity check to our analyses, looking at how results vary based on treatment contrast, we estimate results for districts with low treatment-control contrast. We estimate ITT effects on samples of districts with less than a 50-percentage point treatment-control contrast, less than a 25-percentage point contrast, and less than a 5-percentage point contrast.
Cost Analyses
In addition to estimating the efficacy of personalized messaging, we use the ingredients method to estimate the cost of implementing the intervention. We use data on the costs of developing the templates (paid by NCRERN to the SIS and divided across all participating districts), training (time spent by district staff), setting up the messages (time spent by district staff), and sending and responding to messages (time spent by district staff). We collected data through district surveys about the time required, as well as which staff were involved. We used staff titles to benchmark compensation against state averages (salaries and a 30% fringe rate), adjusted with a national rural scalar. We use contract costs between NCRERN and Infinite Campus for development and training costs. We present total costs per site and per student.
Results
We present four estimates of the effect of personalized messaging on the number of days absent: ITT effects in our full sample, ITT effects in a sample restricted to districts with high randomization compliance, TOT effects in our full sample, and TOT effects in the restricted sample of districts with high randomization compliance. Table 3 summarizes these estimates. Coefficients are from a Poisson specification and can be interpreted as the percent change in days absent (from a base absence rate of about 7%). Negative coefficients suggest the outcome is moving in the desired direction, reducing absenteeism.
Estimated Impacts of Personalized Messaging on Total Number of Days Absent
Note. Grade and district fixed effects, absence splines, and missing indicators are not shown. The exposure term is the number of days the student was enrolled in the district in the outcome year. Standard errors clustered by student. The restricted sample comprises students in districts that did not treat any students assigned to the control group and reached at least half of the students assigned to the treatment group. Poisson coefficients shown. Boldfaced values/text indicate key explanatory variables of interest. FRPL = free or reduced-price lunch. IEP = Individualized Education Program (receiving Special Education services).
p < .1. **p < .05. ***p < .01.
As shown in Table 3, we consistently estimate small decreases in student days absent as a result of LIFT Up Attendance. Our most conservative estimate is the ITT effect of being assigned to have messages sent to a caregiver, which suggests a statistically significant 1.7% decrease in absences, saving about 0.21 days throughout a 180-day school year. 3 These results are largely robust across our sensitivity checks. Point estimates from more parsimonious models with no covariates or district fixed effects only suggest reductions of 1.2% to 1.3%, although estimates are not significant. When we add students’ prior-year absences, the estimated decline in absences is a statistically significant 1.8% (see section “Results” in Supplemental Appendix Table A2 [available in the online version of this article] across model specifications). Students’ prior-year absences are a key covariate for precision in our results and were included in our pre-registered model specification.
Given the noncompliance observed with LIFT Up Attendance, both with control group caregivers receiving messages and with treatment group caregivers not receiving messages, we also estimate TOT estimates. This analysis suggests that when caregivers did receive at least one message during the treatment period, students’ absences decreased by a statistically significant 4.5%, representing a little over half a day saved throughout a 180-day school year. These results are largely stable across specifications, again with students’ prior-year absences serving as the key covariate in the model for precise estimation of the treatment effect (see Supplemental Appendix Table A2 [available in the online version of this article]; estimated effects with no covariates and district effects only are similarly sized but insignificant).
We test the sensitivity of our findings by estimating effects on a restricted sample of districts that met two benchmarks of randomization compliance and implementation fidelity. First, districts in the restricted sample did not send any messages to caregivers of students assigned to the control group. Secondly, districts in the restricted sample sent messages to caregivers for at least half of the students assigned to the treatment group. The estimated ITT effect from this sample is slightly noisier because of the loss of sample, but we still find a statistically significant decrease in absences of 3.1% (or 0.39 days over the course of the year). The estimated TOT effect in this sample is a statistically significant decrease in absences of 3.5% (or 0.44 days over the course of the year).
We further check the sensitivity of these findings to different sample restrictions, based on the observed contrast between the treatment and control groups rather than external thresholds for treatment status compliance; specifically, we check whether results vary when restricting the sample to districts with at a 5-, 25-, and 50-percentage point contrast, respectively (shown in Table A3 in Supplemental Appendix A in the online version of the journal). All estimated effects are similarly sized, between a 1.5% and 5% decrease in absences, and statistically significant. In Figure A1 in Supplemental Appendix A (available in the online version of this article), we show a scatterplot of district treatment estimates against the treatment-control contrast in the district (i.e., the difference in the share of students assigned to the treatment and control groups who actually received a message). This shows a general pattern of increasing effects (greater declines in absences) as compliance increases, although with some variation across districts. As a sanity check, we also estimate effects specifically for low (<50 percentage points), very low (<25 percentage points), and virtually no (<5 percentage points) treatment-control contrast samples (see Supplemental Appendix Table A4 in the online version of the journal). All estimates from the low contrast samples are, as expected, statistically insignificant. Surprisingly, however, the point estimate, which starts small for the low contrast sample and decreases in the very low contrast sample, nominally increases among the virtually no contrast sample. This is because we define the treatment-control contrast based on whether students ever received any LIFT Up message throughout the year, obscuring contrasts in dosage. On average, eight more messages were sent regarding students assigned to treatment in the “virtually no contrast” than regarding students assigned to the control group (note multiple messages could be sent regarding a single student per round of message sending, either via a different mode [text, email, phone call] or point of contact [mother, father, other caregiver] so this is not a contrast in the amount of information shared). If we further restrict our sample to the three districts in which the treatment-control contrast is less than 5 percentage points and the difference in average total messages between students assigned to the treatment and control groups is less than 2, the estimated effect predictably drops to zero.
While our primary outcome of interest is students’ cumulative days absent, we also explore the impact of the messages on students’ likelihood of being chronically absent, as shown in Table A5 in Supplemental Appendix A (available in the online version of this article). Estimated effects are nominally negative (e.g., a reduced likelihood of chronic absenteeism) but are statistically insignificant and range from a decreased likelihood of 0.3 percentage points (full sample ITT) to 4.0 percentage points (restricted sample TOT).
Subgroup Estimates
We are also interested in examining any potential differential effects of personalized messaging across student populations. We estimate the effects of personalized messaging for student populations defined by demographics, services received, grade level, and prior absences. Figure 2 presents the estimated ITT effects by subgroup (point estimates reported in Supplemental Appendix B in the online version of the journal).
Estimated ITT effects of personalized messaging, by subgroup.
As shown in Figure 2, there is some evidence of nominally heterogeneous effects across populations. Starting at the top of the chart, we see estimated treatment effects by student race/ethnicity. Estimated decreases in absences are larger and statistically significant for Indigenous (a 6.6% decrease) and White students (a 2.3% decrease). There are no significant effects of personalized messaging for Hispanic/Latinx, Asian/Pacific Islander, or Multiracial students. The differences in estimated effects for Indigenous, Hispanic/Latinx, Asian American/Pacific Islander, White, and Multiracial students are not statistically significant.
We find a significant adverse effect for Black students (a 6.5% increase), which is significantly different from the estimates for Indigenous, Hispanic/Latinx, White, and Multiracial students (and is not different from the estimate for Asian American/Pacific Islander students). This finding is unexpected and concerning, as prior evaluations of similar interventions have not found adverse impacts for certain populations. This estimate is driven by the 2022–2023 school year, when the majority (89%) of Black students in our sample were enrolled in one district with low rates of contact information for caregivers, meaning most caregivers in the treatment group did not receive messages. The distribution of prior-year absences is also more skewed to the left in this district than the rest of the sample (e.g., higher shares of students missed no or quite few days); given that we see larger effects of the intervention for students with higher prior absences, this could also be a mediating factor for this result. The overall estimated effect of the intervention for this district was insignificant but positive (about an estimated 5% increase in absences overall, with a roughly 6% increase in absences for Black students), making it difficult to disentangle the mediating effect of unobserved district characteristics from our overall estimated effect of the intervention for Black students. When we estimate results for just the 2023–2024 cohort, we no longer find an adverse effect of LIFT Up Attendance for Black students, though importantly the number of Black students in the second cohort is much smaller than in the first (under 900 in the second cohort compared to over 2,900 in the first).
We find a larger estimated decrease for male students (about a 2.4% decrease) than for female students (an insignificant 0.9% decrease), although the difference in estimated effects for male and female students is not significant. Similar to the nominal differences by race/ethnicity, this is driven by the 2022–2023 cohort; we do not see any differences in estimated effects by gender when looking at the 2023–2024 cohort.
We find there is a significant decrease in absences for students receiving FRPL (a 3.7% decrease). This estimate is larger than that found for students not receiving FRPL (a 0.4% increase), with a post hoc F test significant at the 5% level. We also find a significant decrease in absences for students receiving Special Education services (a 4.7% decrease), although this is not significantly different from the smaller estimated 1% decrease for students not receiving Special Education services.
We also find larger estimated decreases in absences for students who were chronically absent in the prior year (a 3.5% decrease) than for students not chronically absent in the preceding year (a 0.8% decrease). The estimated effects for those previously chronically absent and those not previously chronically absent are not significantly different from each other. However, the large and statistically significant estimated decrease for students who were previously chronically absent suggests that personalized messaging may be effective at both decreasing absenteeism overall and particularly among students who may be at greatest risk of missing key academic progress because of high absenteeism. We explore this further by disaggregating students into quintiles based on their prior-year absence rate and estimating effects of the intervention for each quintile (see Supplemental Appendix Figure A2 in the online version of the journal). We see an increasing estimated effect (e.g., increasingly large reductions in absences) for students in higher quintiles of prior-year absences. In short, students who had missed more days of school in the prior year reaped larger benefits from the messaging: only the estimated effect for students in the highest quintile of prior absences, who on average missed 16.7% of the school year, is statistically significant at the 5% level (ITT estimated reduction of 3%; TOT estimated reduction of 8%; see Supplemental Appendix Table B2 [available in the online version of this article] for point estimates).
Finally, we see slight nominal differences in estimated effects by grade level. We find that personalized messaging reduced absences by 1.1% for students in elementary school, by 1.8% for students in middle school, and by 2.4% for students in high school. However, none of these estimates are statistically significant or different from each other.
Cost Estimates
Overall, our results suggest that personalized messaging effectively reduced student absenteeism by a small to moderate amount, with effects predictably larger when focusing on students whose caregivers actually received a message from the school. Initial cost analyses suggest that personalized messaging is likely cost-effective in most contexts. The average estimated cost per student to implement personalized messaging in the 2022–2023 school year was $3.79. Costs were slightly higher in the 2023–2024 school year, as districts sent more rounds of messages and received additional support for implementation, but still averaged just $4.20 per student. Table 4 summarizes the implementation costs of the intervention.
Average Total Annual Cost of Implementation, by Category and Cohort
Note. Personnel costs include time spent setting up messages, sending messaging, and following up with families about the messages. Training costs include the cost to deliver training and time spent by implementing staff in training sessions. NCRERN costs are program fixed costs, including the contractual cost with the SIS to develop the message templates. NCERN = National Center for Rural Education Research Networks; SIS = student information system.
Costs varied across sites depending on the number of students served, which staff were responsible for implementation, and how many staff were involved in implementation. The smallest district across the 2 years had 24 treated students in 2022–2023, with a total cost of $2,090.56 and a per-student cost of $87.11. The largest district across the 2 years reached almost 3,000 students with a total site cost of $7,807.14 and a per-student cost of just $2.64. In the 2023–2024 school year, one participating district had three school administrators (dean, vice principal, principal, and/or superintendent) implementing the intervention, which pushed the per-student cost for their roughly 400 treated students to $9.19—in contrast, a similarly sized district that relied on a guidance counselor and teacher for implementation had a per-student cost of $2.86 for the year. Fixed per-district costs varied between the 2 years, as contractual costs for message creation and individual district support were fixed, but there were 38 districts in 2023–2024 rather than the 9 in the 2022–2023 school year.
Using as a benchmark prior estimates that put the value of an additional day in school at $75, given the relationship between attendance and test scores (Aucejo & Romano, 2016) and expected returns to increased achievement (see Watts, 2020, for a recent review), this would suggest that personalized messaging is cost-effective if the per student cost is between $15 and $41.25, depending on the effectiveness estimate we use. 4 Of the 16 states included in the sample, two (California and Missouri) include average daily attendance (ADA) in their school funding formulas. In California, starting in 2022–2023, this was based on the highest ADA of the current, prior, or average of the 3 prior years, while in Missouri, this was based on the highest ADA in the current, prior, or 2-year prior school year and summer. Missouri also includes student weighting in their formula, while California’s formula includes substantial categorical/grant funding based on grade level, student population, and other characteristics. These features make it difficult to calculate the specific relationships between a dollar amount of funding, ADA unit, and individual student attendance, but ADA funding still provides a potentially useful benchmark. In California, each additional ADA unit results in $887.40 in funding (Education Commission of the States, 2024). Our most conservative estimated effect, of a 1.7% decrease in absenteeism from a base of roughly 7%, would imply a positive cost-benefit ratio at roughly $15 per student or less, while our more optimistic TOT estimate of a 4.5% reduction in absenteeism would imply a positive cost-benefit ratio at about $40 per student or less.
These estimates suggest personalized messaging is a cost-effective strategy for most districts but may not be cost-effective for very small districts. However, the costs detailed here reflect the first year of implementation for each district. In subsequent years, districts would not face the fixed costs of message development. Excluding these costs reduces the average per-student cost to $2.21 across the 2 years. If the staff member(s) implementing the intervention did not change in subsequent years, training costs would also be a one-time investment, further reducing the cost of the intervention and making personalized messaging a cost-effective strategy even for the smallest districts.
Discussion and Conclusion
We report on a national trial of a personalized messaging intervention designed to reduce student absenteeism in rural areas. The intervention was implemented by 47 rural districts across 16 states in all regions of the United States in the 2022–2023 and 2023–2024 school years. We evaluated this intervention using student-level randomization. Overall, we find that LIFT Up Attendance reduced the number of days students were absent. Our most conservative estimate is an ITT effect across all 47 districts and suggests the intervention reduced absences by 1.7%, a decrease that is statistically significant at the 5% level. We additionally estimate TOT effects given observed noncompliance with random assignment. In our full sample, this suggests that the intervention reduced absences by 4.5%, an effect that is again statistically significant at the 5% level. We further triangulate our findings by estimating effects in restricted samples, focusing on districts that met minimum benchmarks of randomization compliance and/or implementation fidelity. These estimates consistently suggest that the intervention reduced absences somewhere in the 1.7% to 4.9% range and are all statistically significant. Estimates from an initial cost analysis using the ingredients method suggest that this intervention, which on average costs roughly $4 per student for the year, is typically cost-effective, although for small districts the benefits may not exceed the costs until the second year of implementation.
The estimated ITT effect we find in this study (1.7%) is slightly smaller than that estimated in the 2020–2021 NCRERN pilot with eight districts in New York and Ohio (a 2.4% reduction), as well as some of the effects found for similar interventions in urban settings (e.g., Robinson et al., 2018; Rogers et al., 2017; Musaddiq et al., 2023). There is slightly more congruence when comparing TOT effects. For example, Musaddiq et al. (2023) report an estimated ITT reduction in absenteeism of 2% to 3%, and a TOT reduction of 3% to 4%, with two of their four districts implementing the intervention with fidelity. Rogers et al. (2017), Robinson et al. (2018), and Himmelsbach et al. (2022) report larger estimated ITT effects (from a 2.4% reduction to an 8.3% reduction) but do not report compliance rates or TOT estimates for comparison. Rogers and Feller (2018) find a 6% reduction in absences in their ITT specification but report that on average 4.2 of 5 intended messages were sent, suggesting strong implementation fidelity.
We find some heterogeneity in effects across student populations, including patterns that merit further study. We find that the intervention led to nominally larger-than-average decreases in absences among Indigenous, White, and male students, as well as students receiving FRPL and Special Education services. Estimated effects were also larger for students who had missed 10% or more of the prior school year. Further suggesting the intervention can influence students with previously high absence rates, we find increasingly large reductions in absences for students in higher quintiles of prior-year absenteeism.
Concerningly, we find an estimated adverse impact of the messages among Black students. This estimate is driven by the 2022–2023 cohort, and specifically the estimate from one district included in that sample, with low treatment compliance due to limited caregiver contact information. The estimated average treatment effect in this district—across all racial/ethnic populations—is positive (increasing absences) and insignificant, meaning that unobserved district characteristics may be contributing to the estimated effect and cannot be fully disentangled from potentially true heterogeneity by student race/ethnicity. This district also had lower baseline absence rates (7.5% vs. 9.6% in the rest of the sample, with higher shares of students with no or very few absences), further suggesting it may be unique in unobserved ways, or simply mirroring the finding that the intervention had larger effects among students with higher prior-year absence rates. The 2022–2023 cohort also showed differential effects by gender, which was not replicated in the 2023–2024 cohort or found in earlier studies, suggesting this cohort may have been unique in unmeasured ways. While there is reason to be cautious in our interpretation of the estimated adverse effect for Black students, it should not be ignored by decision-makers when deciding whether to implement a similar intervention in their context. Additional research is needed to study the effects of personalized messaging across contexts and student populations, to better identify mediators and moderators of the effect of this type of intervention for reducing student absenteeism.
An initial line of inquiry would be to explore the reactions of caregivers to these messages and the impact of the messages on relationships between caregivers and the school. Understanding how the messages may be interpreted by caregivers and the enabling contextual factors that facilitate the hypothesized mechanisms for change may help to clarify the LIFT Up Attendance theory of action. Such qualitative and localized research can point to changes in the intervention design needed to more consistently and equitably support student attendance. While our data cannot speak to this systematically, we do have some survey data from district leaders regarding caregiver reactions to the LIFT Up messages in the 2023–2024 school year. Of the 32 districts that responded, 4 heard no feedback from caregivers, 13 (41%) received negative feedback, 9 (28%) received a mix of negative and positive feedback, and 6 (19%) received positive feedback. Negative feedback included questions about why messages were sent when students “had only missed a couple of days,” rather than focusing on students with more frequent absences; some parents also had negative reactions to the inclusion of all absences, not just unexcused absences, in the count. Districts also reported that some parents did not want to receive any messages, regardless of the content. In terms of positive feedback, districts reported parents appreciated periodic and consistent communication from the school, with some parents using it as an opportunity to celebrate their child publicly after receiving the message.
There were significant implementation challenges to LIFT Up Attendance. There were unexpected technical delays in creating the message templates and releasing them to partner districts. The use of filters to differentiate message templates for students with different patterns of recent attendance increased the complexity of the message setup and of sending the messages in ways that led to some caregivers receiving conflicting messages about their students’ attendance. Districts’ prior decisions about sending messages through the platform on an opt-in or opt-out process had consequences for how many caregivers received the treatment messages. These implementation challenges were unanticipated, as the study was designed to leverage existing infrastructure; for example, all districts already had the messaging platform, and the SIS provided other messaging templates. Some of these challenges were mitigated following the first year of the study, when the study team identified critical areas for district support, but challenges related to the use of message filters persisted and affected both implementation fidelity and randomization compliance. These implementation challenges offer valuable lessons to researchers and district leaders alike. Launching any new intervention, however seemingly incremental and low effort, is challenging, given that schools are complex organizations in which staff members already face numerous competing priorities. Researchers and district leaders alike need to meaningfully plan for the implementation of any new intervention, anticipating challenges, identifying what responsibilities can be removed from implementing staff’s portfolio to free capacity for the new intervention, and continually monitoring implementation to intervene and adjust as needed.
There are a few limitations to this study. First, while we have a large sample overall, particularly for the rural setting, sample sizes get sparse when looking at effects for specific subgroups. Given the differences we see in estimated effects across student populations, districts should consider their student population and the nature of their existing relationships with families. It is possible that if relationships are currently contentious or negative, these personalized messages may feel accusatory or punitive rather than welcoming and supportive. Further research is needed to understand how the nature of current school-family relationships moderates the effectiveness of personalized messaging. Districts should begin by implementing such an intervention on a small scale and gathering feedback from families to better understand the potential benefits or unintended consequences of personalized messaging before adopting it at scale. Relatedly, while the 47 rural districts in our sample represent 16 states and all geographic regions of the state, they are all districts that were already using a specific SIS and messaging platform prior to the start of the intervention. They also represent a small share of the total number of potential participant districts (47 out of about 2,000). These districts may differ from other rural districts in ways that are moderating the effect of the intervention. Additional evidence is needed about the efficacy of personalized messaging interventions in rural contexts to better understand the generalizability of our findings here.
Finally, our identification is threatened by the randomization non-compliance observed in our sample. This noncompliance seems to have been driven by technical challenges associated with message (and randomization filter) setup, as well as prior coverage of caregiver contact information, rather than intentional compliance based on perceptions of which students would benefit from the intervention, but does extend to a substantial portion of our sample. We address this by estimating TOT effects through an instrumental variables approach as well as by restricting our sample to districts with evidence of a valid treatment-control contrast. All estimates from these robustness checks suggest the same conclusion: personalized messaging significantly reduced student absenteeism by a small to moderate amount. Future adaptations should consider providing additional support for setting up the intervention and reducing complexity as much as possible by, for example, considering using one message template instead of differentiating based on the student’s recent attendance pattern.
Despite these limitations, this study contributes to the literature by evaluating a low-cost nudge intervention in a national sample of rural districts. Our findings suggest that adopting such a strategy is likely a cost-effective way for rural districts to marginally improve student attendance. Though the magnitude of reduction is small (about one-fifth to one-half day), our results suggest that students with previously high absences—and their caregivers—can be influenced by small, low-cost interventions and that the causes of all of their absences are not intractable or as difficult to solve as transportation or illness. However, the magnitude of our effects shows that this intervention is not sufficient to return absenteeism to prepandemic levels. As districts, rural and non-rural alike, continue to grapple with worsened attendance following the COVID-19 pandemic and contracting budgets, personalized messaging may offer a promising and cost-effective strategy that can meet immediate needs to reduce absenteeism as districts and researchers continue to seek solutions that address other root causes of absenteeism, particularly systems-level barriers or shifting attitudes toward school attendance, that are needed to more substantially curb absenteeism.
Supplemental Material
sj-pdf-1-epa-10.3102_01623737261438143 – Supplemental material for Lifting Up Attendance in Rural Districts: A Multi-Site Trial of a Personalized Messaging Campaign
Supplemental material, sj-pdf-1-epa-10.3102_01623737261438143 for Lifting Up Attendance in Rural Districts: A Multi-Site Trial of a Personalized Messaging Campaign by Elise Swanson, Sativa Thompson, Jennifer Ash, Hayley Didriksen, Thomas J. Kane, Douglas O. Staiger and Lisa Sanbonmatsu in Educational Evaluation and Policy Analysis
Footnotes
Acknowledgements
The authors would like to acknowledge Atsuko Muroga for her suggestions, feedback, and work on the analyses reported here. We would also like to thank our partners at Infinite Campus for their collaboration, insights, and support of this project. Finally, we are grateful to our district partners for their participation in the study and dedication to improving outcomes through evidence.
Declaration of Conflicting Interests
The authors declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.
Funding
The authors disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: This study was funded by the Institute of Education Sciences, U.S. Department of Education, through Grant R305C190004 to Presidents and Fellows of Harvard College. The opinions expressed are those of the authors and do not represent the views of the Institute or the U.S. Department of Education.
1.
Our plan to estimate ITT effects was pre-registered on the Registry of Efficacy and Effectiveness Study under Registry ID 14780. The preregistration was for the first year of the study, estimating ITT effects. In this analysis, we have added a second cohort and included the TOT estimates as well as the ITT estimates for the samples restricted based on randomization assignment compliance. The model specification is the same as the pre-registered analysis plan, as are the student subgroups.
2.
Absences have a roughly linear relationship year over year; we split absence rate into splines to match this relationship in the Poisson model.
3.
ITT effects are larger when estimated by treatment fidelity. Districts with high fidelity (sending three or more messages) in 2022–2023 had a 5% decrease in absences (p < .05), with no significant impact in low-fidelity districts. Districts with high fidelity (six or more messages) in 2023–2024 had 2.1% decrease in absences (p < .5), with no significant decrease in absences in low-fidelity districts.
4.
A $15 per-student cost is based on our ITT estimate from the full sample, which found a 1.7% decrease in absences, or 0.21 days in a 180-day school year. A $41.25 per-student cost is based on our TOT estimate from the full sample, which found a 4.4% decreases in absences, or 0.55 days in a 180-day school year.
Authors
ELISE SWANSON, PhD, is the director of research initiatives for college and career success at the Center for Education Policy Research at Harvard University. Her research focuses on evaluating programs, policies, and practices aimed at improving students’ preparedness for and successful transitions into college and the workforce.
SATIVA THOMPSON, MS, is the associate director of data and evaluation at Partners for Rural Impact. Her research focuses on the evaluation of instructional practices to support academic achievement for rural students.
JENNIFER ASH, PhD, is the founder and principal of Edify Research and Strategy. Her research focuses on partner-centered projects to study implementation and identify effective interventions to strengthen students’ long-term success.
HAYLEY DIDRIKSEN, MS, is the senior director of data and technology at Portland Public Schools. Her research focuses on data and evidence use within K–12 systems to strengthen student outcomes.
THOMAS J. KANE, PhD, is the Walter H. Gale Professor of Education at the Harvard Graduate School of Education and Faculty Director of the Center for Education Policy Research at Harvard University. His research covers topics in K–12 and higher education, including teacher effectiveness, student absenteeism, school accountability, and achievement recovery following the COVID-19 pandemic.
DOUGLAS O. STAIGER, PhD, is the John Sloan Dickey Third Century Professor in the Department of Economics at Dartmouth. His research focuses on the economics of education and healthcare and statistical methods.
LISA SANBONMATSU, PhD, is the director of research at the Center for Education Policy Research at Harvard University. Her research focuses on student absenteeism, high-dosage tutoring, the long-term impacts of charter school enrollment, and economic mobility.
References
Supplementary Material
Please find the following supplemental material available below.
For Open Access articles published under a Creative Commons License, all supplemental material carries the same license as the article it is associated with.
For non-Open Access articles published, all supplemental material carries a non-exclusive license, and permission requests for re-use of supplemental material or any part of supplemental material shall be sent directly to the copyright owner as specified in the copyright notice associated with the article.
