The Effects of School Phone Bans: National Evidence from Lockable Pouches

May combine two claims.

All states administer standardized assessments in grades 6 through 8, allowing comparable achievement data across the country and over time.

“All states administer standardized assessments in these grades, allowing us to obtain comparable achievement data across the country and over time.6”

• cites [6] 6. Beneito, P., and Ó. Vicente-Chirivella. 2022. "Banning mobile phones in schools: evi- dence from regional-level policies in Spain." Applied Economic Analysis, 30(90): 153-175. sourcePartly supported

  The paper uses PISA scores to compare academic achievement across Spanish regions and over time (Section 2.1, p. 6), and describes the scores as internationally comparable (p. 2). It does not support the claim that all states administer assessments in grades 6 through 8: the data come from representative samples of 15-year-olds assessed every three years (note 13, p. 20). (LLM, conf=0.87)

  Experimental — accuracy not yet measured

Classified as probably not a substantial claim. May combine two claims.

The study uses the difference-in-differences framework developed by Callaway and Sant'Anna (2021), with a doubly robust implementation combining inverse probability tilting and outcome regression.

“We use the difference-in-differences framework developed by Callaway and Sant'Anna (2021), which is well suited to settings with staggered treatment adoption and heterogeneous treatment effects.We use the doubly robust implementation of this estimator, which combines inverse probability tilting and outcome regression.”

• cites Callaway and Sant'Anna (2021) sourcePartly supported

  The Abstract and Section 1 describe a DiD framework for staggered treatment timing and heterogeneous effects. Section 4 says its doubly robust approach combines outcome regression with inverse probability weighting, but our copy does not describe the implementation as using inverse probability tilting; that term appears in a reference title. (LLM, conf=0.88)

  Experimental — accuracy not yet measured

Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

The authors committed ex-ante to treat attendance rates as their primary outcome and present chronic absenteeism as a secondary measure.

“Because attendance rates are continuous and have been used in prior research on school phone bans (e.g., in Figlio and Ozek (2025)), we committed ex-ante to treat attendance rates as our primary outcome and present chronic absenteeism as a secondary measure.”

• cites Figlio and Ozek (2025) sourceSupported

Probably not a substantial claim: may not pass S3 (load-bearing).

The improved doubly robust implementation remains consistent if either the treatment model or the outcome model is correctly specified.

“We use the improved doubly robust implementation of this estimator, which combines inverse-probability tilting to estimate the propensity score and weighted least squares for the outcome regression, and remains consistent if either the treatment model or the outcome model is correctly specified (Sant'Anna and Zhao, 2020).”

• cites Sant'Anna and Zhao, 2020 sourceSupported

Probably not a substantial claim: may not pass S3 (load-bearing).

Using the Callaway and Sant'Anna (2021) estimator with regression adjustment only, rather than the doubly robust implementation, yields nearly identical estimates.

“Using the Callaway and Sant'Anna (2021) estimator with regression adjustment only, rather than the doubly robust implementation, yields nearly identical estimates.”

• cites Callaway and Sant'Anna (2021) sourceSupported

Probably not a substantial claim: may not pass S3 (load-bearing).

This concentration of effects in math is consistent with a broader literature in the economics of education finding that interventions often generate larger or more detectable impacts in math than in ELA.

“This concentration of effects in math is consistent with a broader literature in the economics of education finding that interventions often generate larger or more detectable impacts in math than in ELA (e.g., Dee and Jacob, 2011; Dobbie and Fryer Jr, 2011; Chetty, Friedman and Rockoff, 2014).”

• cites Dee and Jacob, 2011 sourceSupported
• cites Dobbie and Fryer Jr, 2011 sourceNot checked — no full text of the source
• cites Chetty, Friedman and Rockoff, 2014 sourceNot checked — no full text of the source

Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

With the smaller Panorama sample, the doubly robust estimator fails to compute some cohort-by-year ATTs, whereas the outcome-regression-only implementation is well behaved.

“With the smaller sample, the doubly robust estimator fails to compute some cohort-by-year ATTs, whereas the outcome-regression-only implementation is well behaved.”

• cites Callaway and Sant'Anna (2021) sourceCould not verify

  Remark 9 (Section 4, page unavailable) says limited covariate overlap may make doubly robust inference imprecise and outcome regression more informative; it does not report an estimator failing to compute. The Results (Section 5.1, page unavailable) report successful doubly robust estimation for a minimum-wage study, not a smaller Panorama sample. Searches for Panorama and computation failures found no passage establishing the claimed comparison. (LLM, conf=0.78)

  Experimental — accuracy not yet measured

Probably not a substantial claim: may not pass S3 (load-bearing).

The overlap of schools serving both middle and high school grades accounts for approximately 7.1 percent of school-year observations in the test score data.

“This overlap accounts for approximately 7.1 percent of school-year observations in the test score data.”

May combine two claims.

Average effects of Yondr adoption on test scores are close to zero over the first three years following adoption; the study can rule out improvements larger than approximately 0.008 student-level standard deviations.

“For academic achievement, average effects on test scores are consistently close to zero.”

• Also stated in: § 1 Introduction · plain

  “Turning to academic achievement, we find average effects on test scores that are close to zero over the first three years following adoption: we can rule out improvements larger than approximately 0.008 student-level standard deviations.”

• AssumptionRests on an assumption: Conditional on observed school characteristics, treated and comparison schools would have followed parallel outcome trends absent Yondr adoption.

May combine two claims.

The high school test score sample includes 656 Yondr-adopting schools, representing 69 percent of the set of potential adopters available for this analysis.

“The high school test score sample includes 656 Yondr-adopting schools, representing 69 percent of the set of potential adopters available for this analysis.”

May combine two claims.

In middle schools, Yondr adoption has small negative effects on test scores, roughly half as large in magnitude as the high school effects.

“High schools see modest positive effects, particularly in math, while middle schools see small negative effects.”

• Also stated in: § 1 Introduction · plain

  “In contrast, effects in middle schools are generally negative and even smaller in magnitude-roughly half as large.”

• AssumptionRests on an assumption: Conditional on observed school characteristics, treated and comparison schools would have followed parallel outcome trends absent Yondr adoption.

May combine two claims.

For self-reported classroom attention, there is a negative and statistically significant estimate in the second post-adoption year, but estimates for this outcome should be interpreted with caution because there is some evidence of differential pre-trends in the subsample used for this measure.

“For classroom attention, we observe a negative and statistically significant estimate in the second post-adoption year; however, estimates for this specific outcome should be interpreted with caution, as there is some evidence of differential pre-trends in the subsample used for this measure.”

• AssumptionRests on an assumption: Conditional on observed school characteristics, treated and comparison schools would have followed parallel outcome trends absent Yondr adoption.

May combine two claims.

Regardless of the specific exam, high school assessments were required for all students in a given grade or course, including ACT or SAT exams, which were typically administered to all grade 11 students in the state.

“Importantly, regardless of the specific exam, these assessments were required for all students in a given grade or course, including ACT or SAT exams, which were typically administered to all grade 11 students in the state.”

May combine two claims.

The middle school test score sample includes 1,341 Yondr-adopting schools, representing 98 percent of the set of potential adopters available for this analysis.

“The middle school test score sample includes 1,341 Yondr-adopting schools, representing 98 percent of the set of potential adopters available for this analysis.”

May combine two claims.

Effects of Yondr adoption on the school attendance rate are precisely estimated and close to zero; the study can rule out improvements larger than 0.056 percentage points.

“We find little evidence of effects on school attendance, self-reported classroom attention, or perceived online bullying.”

• Also stated in: § 1 Introduction · plain

  “Effects on the attendance rate are precisely estimated and close to zero: we can rule out improvements larger than 0.056 percentage points.”

• AssumptionRests on an assumption: Conditional on observed school characteristics, treated and comparison schools would have followed parallel outcome trends absent Yondr adoption.

May combine two claims.

Parents express support for pouches and expect improvements in academic, social, and mental health outcomes.

“Parents express support for pouches and expect improvements in academic, social, and mental health outcomes.”

May combine two claims.

Teachers in schools that adopt pouches report reduced frequency of inappropriate student phone usage in class and increased satisfaction with their schools' overall policy approach.

“Teachers in schools that adopt pouches report reduced frequency of inappropriate usage by students in class and increased satisfaction with their schools' overall policy approach.”

May combine two claims.

Teacher reports indicate that the share of students using phones in class for personal reasons falls from 61 percent to 13 percent following Yondr adoption, a drop of roughly 80 percent, substantially larger than in non-adopting schools.

“Second, teacher reports from our survey indicate large declines in student phone use following adoption.The share of students reported to be using their phones in class for personal reasons falls from 61 percent to 13 percent following Yondr adoption (a drop of roughly 80 percent), a substantially larger decline than in non-adopting schools.”

May combine two claims.

Effects of Yondr adoption on chronic absenteeism are precise and small.

“We find similarly precise and small effects for chronic absenteeism and no measurable improvements for perceived online bullying or self-reported classroom attention.”

• AssumptionRests on an assumption: Conditional on observed school characteristics, treated and comparison schools would have followed parallel outcome trends absent Yondr adoption.

May combine two claims.

Students tend to oppose pouches and expect less change in academic, social, and mental health outcomes.

“Students tend to oppose pouches and expect less change in these outcomes.”

May combine two claims.

The pooled estimate of the effect of Yondr adoption on SWB is -0.305 (SE = 0.201) school-level standard deviations, which rules out even modest positive effects.

“The pooled estimate is -0.305 (SE = 0.201) school-level standard deviations, which rules out even modest positive effects.”

• AssumptionThe authors state this as an assumption: The causal interpretation of the difference-in-differences estimates relies on the parallel trends assumption: treated and comparison schools would have followed parallel trends absent adoption, conditional on baseline characteristics.

May combine two claims.

Relative to never adopters, Yondr-adopting schools are larger, more likely to be located in urban areas, and serve student populations that have larger shares of Black and Hispanic students.

“Relative to never adopters, Yondr-adopting schools are larger, more likely to be located in urban areas, and serve student populations that have larger shares of Black and Hispanic students.”

Probably not a substantial claim: may not pass S3 (load-bearing).

The Panorama data span academic years 2019 through 2025.

“The Panorama data span academic years 2019 through 2025.”

May combine two claims.

The estimate of the effect of Yondr adoption on classroom attention is -0.088 (SE = 0.085), indicating no evidence of changes following adoption.

“The estimate is -0.088 (SE = 0.085), indicating no evidence of changes following adoption.”

• AssumptionThe authors state this as an assumption: The causal interpretation of the difference-in-differences estimates relies on the parallel trends assumption: treated and comparison schools would have followed parallel trends absent adoption, conditional on baseline characteristics.

May combine two claims.

The point estimate for the disciplinary index moves toward zero in t+1 and turns negative by t+2, though these later estimates are imprecisely estimated.

“The point estimate moves toward zero in t + 1 and turns negative by t + 2, though these later estimates are imprecisely estimated.”

• AssumptionThe authors state this as an assumption: The causal interpretation of the difference-in-differences estimates relies on the parallel trends assumption: treated and comparison schools would have followed parallel trends absent adoption, conditional on baseline characteristics.

Probably not a substantial claim: may not pass S3 (load-bearing).

Parents on average report that a policy requiring lockable pouches would increase standardized test scores (0.370).

“On a scale ranging from -1 to 1 (-1 = get worse / oppose; 0 = won't change / neutral; 1 = improve / support), parents on average report that such a policy would increase standardized test scores (0.370), improve student relationships (0.461), and improve mental health (0.597), and they also express net support for such bans (0.369).”

May combine two claims.

For chronic absenteeism, the sample includes 1,786 adopters, representing 88 percent of the potential schools.

“For chronic absenteeism, it includes 1,786 adopters, representing 88 percent of the potential schools.”

Probably not a substantial claim: may not pass S3 (load-bearing).

Parents on average report that such a policy would improve mental health (0.597).

“On a scale ranging from -1 to 1 (-1 = get worse / oppose; 0 = won't change / neutral; 1 = improve / support), parents on average report that such a policy would increase standardized test scores (0.370), improve student relationships (0.461), and improve mental health (0.597), and they also express net support for such bans (0.369).”

Probably not a substantial claim: may not pass S3 (load-bearing).

8 percent of students report using lockable pouches such as Yondr.

“16 percent of students are required to store phones in lockers and 8 percent report using lockable pouches such as Yondr.”

Probably not a substantial claim: may not pass S3 (load-bearing).

The resulting analytic samples differ across outcomes due to differences in data availability and reporting regimes.

“The resulting analytic samples differ across outcomes due to differences in data availability and reporting regimes.”

May combine two claims.

Yondr adopters have substantially lower baseline (2018) test scores in both math and ELA and are located in neighborhoods with lower income-to-poverty ratios.

“In addition, Yondr adopters have substantially lower baseline (i.e., 2018) test scores in both math and ELA and are located in neighborhoods with lower income-to-poverty ratios.”

May combine two claims.

Relative to NCES schools overall, Panorama participation is more common among larger schools, schools located in urban areas, schools serving more racially diverse student populations, and schools in somewhat higher-income neighborhoods.

“Table A1 compares baseline characteristics of schools with non-missing Panorama data for at least one of the three primary indices to the full NCES population, showing that, relative to NCES schools overall, Panorama participation is more common among larger schools, schools located in urban areas, schools serving more racially diverse student populations, and schools in somewhat higher-income ne”

May combine two claims.

Policy activity among control schools rises modestly over time but remains small in absolute terms.

“Policy activity among control schools rises modestly over time but remains small in absolute terms.”

Probably not a substantial claim: may not pass S3 (load-bearing).

Chronic absenteeism data are available for 38 states.

“For attendance rates, we assemble data for 31 states between 2018 and 2025; for chronic absenteeism, data are available for 38 states.”

May combine two claims.

The attendance sample includes 1,711 Yondr-adopting schools, representing 84 percent of the set of potential adopters available for this analysis.

“The attendance sample includes 1,711 Yondr-adopting schools, representing 84 percent of the set of potential adopters available for this analysis.”

May combine two claims.

Adoption of Yondr increased sharply beginning in the 2023 school year, with the largest number of schools adopting in 2025 and 2026.

“While a small number of schools adopted Yondr prior to the COVID-19 pandemic, adoption increased sharply beginning in the 2023 school year, with the largest number of schools adopting in 2025 and 2026.”

May combine two claims.

In 2023, roughly 1 percent of control schools adopt a strict alternative policy; the share remains below 2 percent in 2024 and below 5 percent in 2025.

“In 2023, roughly 1 percent of control schools adopt a strict alternative policy; the share remains below 2 percent in 2024 and below 5 percent in 2025.”

May combine two claims.

The Panorama file includes 62,289 school-year observations with non-missing survey data, representing 16,285 unique schools across 46 states.

“In total, the Panorama file includes 62,289 school-year observations with non-missing survey data during this period, representing 16,285 unique schools across 46 states.”

May combine two claims.

While baseline test scores do not enter the weighting procedure, they are also much closer across the two groups in the reweighted sample.

“While baseline test scores do not enter the weighting procedure, they are also much closer across the two groups in the reweighted sample.”

May combine two claims.

The estimate of the effect of Yondr adoption on attendance rate is -0.138 percentage points (SE = 0.099), which allows ruling out even modest improvements relative to the mean attendance rate of 93 percent.

“The estimate of -0.138 percentage points (SE = 0.099) allows us to rule out even modest improvements relative to the mean attendance rate of 93 percent.”

• AssumptionThe authors state this as an assumption: The causal interpretation of the difference-in-differences estimates relies on the parallel trends assumption: treated and comparison schools would have followed parallel trends absent adoption, conditional on baseline characteristics.

May combine two claims.

In all states included in the analysis, out-of-school suspensions are coded as absences, implying that attendance rates may reflect both instructional attendance and disciplinary exclusions.

“In all states included in our analysis, out-of-school suspensions are coded as absences, implying that attendance rates may reflect both instructional attendance and disciplinary exclusions.”

Probably not a substantial claim: may not pass S3 (load-bearing).

Across all school-year observations in the NCES panel during this period, 6.9 percent have at least one Panorama survey measure observed in a given year.

“Across all school-year observations in the NCES panel during this period, 6.9 percent have at least one Panorama survey measure observed in a given year.”

May combine two claims.

In the year of implementation, SWB declines sharply by approximately 0.603 school-level standard deviations and then recovers, becoming positive and statistically significant by t+2 (approximately 0.485 standard deviations).

“In the year of implementation, SWB declines sharply by approximately 0.603 school-level standard deviations and then recovers, becoming positive and statistically significant by t + 2 (approximately 0.485 standard deviations).”

• AssumptionThe authors state this as an assumption: The causal interpretation of the difference-in-differences estimates relies on the parallel trends assumption: treated and comparison schools would have followed parallel trends absent adoption, conditional on baseline characteristics.

May combine two claims.

The estimates imply an increase of roughly 0.33 percentage points on suspension rates, corresponding to about an 11 percent increase relative to baseline levels.

“The estimates imply an increase of roughly 0.33 percentage points, corresponding to about an 11 percent increase relative to baseline levels.”

May combine two claims.

The discipline sample includes 1,557 Yondr-adopting schools, representing 77 percent of the set of potential adopters available for this analysis.

“The discipline sample includes 1,557 Yondr-adopting schools, representing 77 percent of the set of potential adopters available for this analysis.”

May combine two claims.

At adoption, disciplinary incidents rise sharply, by approximately 0.085 (SE = 0.020) school-level standard deviations, corresponding to roughly 0.03 student-level standard deviations or about a 16 percent increase in suspension rates relative to baseline levels.

“At adoption, disciplinary incidents rise sharply, by approximately 0.085 (SE = 0.020) school-level standard deviations, corresponding to roughly 0.03 student-level standard deviations or about a 16 percent increase in suspension rates relative to baseline levels.”

• AssumptionThe authors state this as an assumption: The causal interpretation of the difference-in-differences estimates relies on the parallel trends assumption: treated and comparison schools would have followed parallel trends absent adoption, conditional on baseline characteristics.

May combine two claims.

This corresponds to roughly 0.199 and 0.16 student-level standard deviations in t and t+2, respectively.

“This corresponds to roughly 0.199 and 0.16 student-level standard deviations in t and t + 2, respectively.”

May combine two claims.

Attendance estimates return to zero in subsequent periods, with no evidence of sustained changes.

“The magnitude is small, however, and estimates return to zero in subsequent periods, with no evidence of sustained changes.”

• AssumptionThe authors state this as an assumption: The causal interpretation of the difference-in-differences estimates relies on the parallel trends assumption: treated and comparison schools would have followed parallel trends absent adoption, conditional on baseline characteristics.

May combine two claims.

The estimate of the effect of Yondr adoption on students' perceived likelihood of online bullying is -0.039 (SE = 0.092), which is small and not statistically significant.

“The estimate is -0.039 (SE = 0.092), which is small and not statistically significant.”

• AssumptionThe authors state this as an assumption: The causal interpretation of the difference-in-differences estimates relies on the parallel trends assumption: treated and comparison schools would have followed parallel trends absent adoption, conditional on baseline characteristics.

May combine two claims.

Average effects on standardized test scores are close to zero and precisely estimated, with similarly small and null effects on attendance, classroom attention, and perceived online bullying.

“Average effects on standardized test scores are close to zero and precisely estimated, with similarly small and null effects on attendance, classroom attention, and perceived online bullying.”

• AssumptionThe authors state this as an assumption: The causal interpretation of the difference-in-differences estimates relies on the parallel trends assumption: treated and comparison schools would have followed parallel trends absent adoption, conditional on baseline characteristics.

May combine two claims.

In the pooled sample of middle and high schools, the estimated effect of Yondr adoption on the combined test score index is close to zero (-0.004, SE = 0.010), with similarly small estimates for Math and ELA.

“In the pooled sample of middle and high schools (Panel A), the estimated effect on the combined test score index is close to zero (-0.004,SE = 0.010), with similarly small estimates for Math and ELA.”

• AssumptionThe authors state this as an assumption: The causal interpretation of the difference-in-differences estimates relies on the parallel trends assumption: treated and comparison schools would have followed parallel trends absent adoption, conditional on baseline characteristics.

May combine two claims.

In high schools, effects are concentrated in math, with an effect of approximately 0.048 school-level standard deviations, which translates to 0.024 student-level standard deviations.

“In high schools, effects are concentrated in math, with an effect of approximately 0.048 school-level standard deviations, which translates to 0.024 student-level standard deviations.”

Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

Among Yondr-adopting schools, the most common prior phone policy was 'no show' (62.01%), followed by no policy (23.16%), class collection (7.36%), locker (6.17%), centralized collection (0.97%), and no phone (0.32%).

“No show 62.01 No policy 23.16 Class collection 7.36 Locker 6.17 Centralized collection 0.97 No phone 0.32 Total 100.00”

May combine two claims.

In high schools, effects are positive for both high- and low-poverty schools, and are larger for higher-poverty schools than for lower-poverty schools, though the latter estimate is not statistically significant and the two estimates are not statistically distinguishable.

“In high schools, effects are positive for both high-and low-poverty schools, and are larger for higherpoverty schools than for lower-poverty schools, though the latter estimate is not statistically significant and the two estimates are not statistically distinguishable.”

• AssumptionThe authors state this as an assumption: The causal interpretation of the difference-in-differences estimates relies on the parallel trends assumption: treated and comparison schools would have followed parallel trends absent adoption, conditional on baseline characteristics.

May combine two claims.

The heterogeneity in test score effects is robust when focusing specifically on math test scores: high school math effects remain positive and statistically significant, while middle school math effects remain negative, smaller in magnitude, and statistically significant in most specifications.

“Table A15 further shows that this heterogeneity is robust when focusing specifically on math test scores, where effects are most pronounced: high school math effects remain positive and statistically significant, while middle school math effects remain negative, smaller in magnitude, and statistically significant in most specifications.”

May combine two claims.

Results for discipline and attendance are broadly similar across school levels, though disciplinary effects are somewhat larger and more precisely estimated in high schools.

“Results are broadly similar across school levels, though disciplinary effects are somewhat larger and more precisely estimated in high schools.”

• AssumptionThe authors state this as an assumption: The causal interpretation of the difference-in-differences estimates relies on the parallel trends assumption: treated and comparison schools would have followed parallel trends absent adoption, conditional on baseline characteristics.

May combine two claims.

Across racial and ethnic groups, estimates are most negative for White students and somewhat smaller for Black students.

“Across racial and ethnic groups, estimates are most negative for White students and somewhat smaller for Black students.”

• AssumptionThe authors state this as an assumption: The causal interpretation of the difference-in-differences estimates relies on the parallel trends assumption: treated and comparison schools would have followed parallel trends absent adoption, conditional on baseline characteristics.

May combine two claims.

While the average academic effects mask interesting heterogeneity-modest positive effects in high schools and small negative effects in middle schools-the magnitudes are relatively small in both settings.

“While the average academic effects mask interesting heterogeneity-modest positive effects in high schools and small negative effects in middle schools-the magnitudes are relatively small in both settings.”

• AssumptionThe authors state this as an assumption: The causal interpretation of the difference-in-differences estimates relies on the parallel trends assumption: treated and comparison schools would have followed parallel trends absent adoption, conditional on baseline characteristics.

May combine two claims.

This corresponds to approximately -0.002 student-level standard deviations, and the precision of the estimates allows ruling out improvements larger than approximately 0.008 student-level standard deviations.

“This corresponds to approximately -0.002 student-level standard deviations, and the precision of the estimates allows us to rule out improvements larger than approximately 0.008 student-level standard deviations.”

May combine two claims.

In subsequent years, test score estimates become slightly negative but remain statistically insignificant.

“In subsequent years, the estimates become slightly negative but remain statistically insignificant.”

• AssumptionThe authors state this as an assumption: The causal interpretation of the difference-in-differences estimates relies on the parallel trends assumption: treated and comparison schools would have followed parallel trends absent adoption, conditional on baseline characteristics.

May combine two claims.

Yondr-adopting schools differ from non-adopting schools: adopters have a lower share of White students (0.35 vs 0.54), a higher share of Black students (0.27 vs 0.14), a higher share of Hispanic students (0.29 vs 0.23), and are more urban (0.49 vs 0.25).

“Total enrollment 591 681 821 Share White 0.54 0.35 0.44 Share Black 0.14 0.27 0.22 Share Hispanic 0.23 0.29 0.24 Urban 0.25 0.49 0.42”

May combine two claims.

In middle schools, the combined test score declines by -0.024 (SE = 0.012), with a decline in math of -0.027 (SE = 0.014) and a smaller, statistically insignificant estimate for ELA.

“In middle schools, the combined score declines by -0.024 (SE = 0.012), with a decline in math of -0.027 (SE = 0.014) and a smaller, statistically insignificant estimate for ELA.”

• AssumptionThe authors state this as an assumption: The causal interpretation of the difference-in-differences estimates relies on the parallel trends assumption: treated and comparison schools would have followed parallel trends absent adoption, conditional on baseline characteristics.

May combine two claims.

In t+1, the classroom attention estimate is slightly positive and not statistically significant, while by t+2 it turns negative and statistically significant.

“In t + 1, the estimate is slightly positive and also not statistically significant, while by t + 2 it turns negative and statistically significant.”

• AssumptionThe authors state this as an assumption: The causal interpretation of the difference-in-differences estimates relies on the parallel trends assumption: treated and comparison schools would have followed parallel trends absent adoption, conditional on baseline characteristics.

May combine two claims.

Over time, disciplinary impacts fade and well-being rebounds, becoming positive in subsequent years.

“Over time, however, disciplinary impacts fade and well-being rebounds, becoming positive in subsequent years.”

• AssumptionThe authors state this as an assumption: The causal interpretation of the difference-in-differences estimates relies on the parallel trends assumption: treated and comparison schools would have followed parallel trends absent adoption, conditional on baseline characteristics.

May combine two claims.

Event-study estimates using predicted test scores based on school composition are close to zero throughout the event window, with no evidence of systematic changes following adoption.

“These estimates are close to zero throughout the event window, with no evidence of systematic changes following adoption.”

• AssumptionThe authors state this as an assumption: The causal interpretation of the difference-in-differences estimates relies on the parallel trends assumption: treated and comparison schools would have followed parallel trends absent adoption, conditional on baseline characteristics.

May combine two claims.

The post-adoption estimates suggest somewhat more positive effects in later cohorts, though these patterns should be interpreted cautiously since we observe a limited number of post-adoption periods.

“The post-adoption estimates suggest somewhat more positive effects in later cohorts, though these patterns should be interpreted cautiously since we observe a limited number of post-adoption periods.”

• AssumptionThe authors state this as an assumption: The causal interpretation of the difference-in-differences estimates relies on the parallel trends assumption: treated and comparison schools would have followed parallel trends absent adoption, conditional on baseline characteristics.

May combine two claims.

For the 2024 cohort, estimates are small and positive in both t and t+1, though not statistically significant.

“For the 2024 cohort, estimates are small and positive in both t and t + 1, though not statistically significant.”

• AssumptionThe authors state this as an assumption: The causal interpretation of the difference-in-differences estimates relies on the parallel trends assumption: treated and comparison schools would have followed parallel trends absent adoption, conditional on baseline characteristics.

May combine two claims.

In middle schools, the estimates are generally negative and small in magnitude, with larger declines in lower-poverty schools than in higher-poverty schools.

“In middle schools, the estimates are generally negative and small in magnitude, with larger declines in lower-poverty schools than in higher-poverty schools.”

• AssumptionThe authors state this as an assumption: The causal interpretation of the difference-in-differences estimates relies on the parallel trends assumption: treated and comparison schools would have followed parallel trends absent adoption, conditional on baseline characteristics.

May combine two claims.

In middle schools, effects are negative and also concentrated in math, but are even smaller-roughly half the magnitude of the high school effects.

“In middle schools, effects are negative and also concentrated in math, but are even smaller-roughly half the magnitude of the high school effects.”

May combine two claims.

For the 2025 cohort, the estimate is small, positive, and statistically significant (0.030 school-level standard deviations).

“For the 2025 cohort, for which we observe only the adoption year, the estimate is small, positive, and statistically significant (0.030 school-level standard deviations).”

• AssumptionThe authors state this as an assumption: The causal interpretation of the difference-in-differences estimates relies on the parallel trends assumption: treated and comparison schools would have followed parallel trends absent adoption, conditional on baseline characteristics.

May combine two claims.

Changes in enrollment and demographic composition are either small relative to baseline levels or statistically insignificant, with similar patterns across high schools and middle schools.

“Changes in enrollment and demographic composition are either small relative to baseline levels or statistically insignificant, with similar patterns across high schools and middle schools.”

May combine two claims.

Disciplinary incidents increase and student-reported subjective well-being declines in the year of implementation, consistent with adjustment costs as schools and students adapt to the new restriction.

“Disciplinary incidents increase and student-reported subjective well-being declines in the year of implementation, consistent with adjustment costs as schools and students adapt to the new restriction.”

• AssumptionThe authors state this as an assumption: The causal interpretation of the difference-in-differences estimates relies on the parallel trends assumption: treated and comparison schools would have followed parallel trends absent adoption, conditional on baseline characteristics.

May combine two claims.

In high schools, the combined test score increases by 0.025 (SE = 0.017), with a statistically significant increase in math of 0.048 (SE = 0.018) and a near-zero estimate for ELA.

“In high schools, the combined score increases by 0.025 (SE = 0.017), with a statistically significant increase in math of 0.048 (SE = 0.018) and a near-zero estimate for ELA.”

• AssumptionThe authors state this as an assumption: The causal interpretation of the difference-in-differences estimates relies on the parallel trends assumption: treated and comparison schools would have followed parallel trends absent adoption, conditional on baseline characteristics.

May combine two claims.

For classroom attention, estimates are negative and statistically significant in high schools and close to zero in middle schools, though these differences should be interpreted cautiously given evidence of pre-trends, particularly in high schools.

“For classroom attention, estimates are negative and statistically significant in high schools and close to zero in middle schools, though these differences should be interpreted cautiously given evidence of pre-trends, particularly in high schools.”

• AssumptionThe authors state this as an assumption: The causal interpretation of the difference-in-differences estimates relies on the parallel trends assumption: treated and comparison schools would have followed parallel trends absent adoption, conditional on baseline characteristics.

May combine two claims.

The overall ATT of predicted test scores is -0.000 (SE = 0.005) in the pooled sample, and similarly small and statistically insignificant across school levels.

“The overall ATT is -0.000 (SE = 0.005) in the pooled sample, and similarly small and statistically insignificant across school levels.”

• AssumptionThe authors state this as an assumption: The causal interpretation of the difference-in-differences estimates relies on the parallel trends assumption: treated and comparison schools would have followed parallel trends absent adoption, conditional on baseline characteristics.

May combine two claims.

The Panorama sample differs from the NCES and NTS samples: it has a lower share of White students (0.35 vs 0.53 and 0.57), a higher share of Hispanic students (0.37 vs 0.23 and 0.20), and is more urban (0.42 vs 0.27 and 0.24).

“Total enrollment 595 832 756 Share White 0.53 0.57 0.35 Share Black 0.15 0.14 0.17 Share Hispanic 0.23 0.20 0.37 Urban 0.27 0.24 0.42”

May combine two claims.

Pooled ATT of Yondr adoption on attendance rate is -0.138 (SE 0.099), not significant.

“Panel A: Pooled ATT 0.059 * * -0.138 -0.305 -0.088 -0.039 ( 0.024) ( 0.099) ( 0.201) ( 0.085) ( 0.092)”

The quote only partly states this claim.

In high schools, the estimated ATT of Yondr adoption on ELA test scores was 0.011 (SE 0.022).

“| ATT | 0.025 | 0.048 * * * | 0.011 |”

• Also stated in: § E Robustness Checks for Test Score Estimates · plain

  “| | ( 0.017) | ( 0.018) ( 0.022) | |”

• AssumptionThe authors state this as an assumption: The baseline framework assumes that once a school adopts Yondr, it remains treated.

May combine two claims.

High school ATT of Yondr adoption on standardized test scores is 0.025 (SE 0.017), not significant.

“1.Baseline -0.004 214,556 0.025 65,978 -0.024 * * 162,974 ( 0.010) ( 0.017) ( 0.012)”

Probably not a substantial claim: may not pass S3 (load-bearing).

The filtering process yielded a set of 104 distinct student survey items for further consideration.

“This filtering process yielded a set of 104 distinct student survey items for further consideration.”

May combine two claims.

Pooled ATT of Yondr adoption on standardized test scores is -0.004 (SE 0.010), not significant.

“1.Baseline -0.004 214,556 0.025 65,978 -0.024 * * 162,974 ( 0.010) ( 0.017) ( 0.012)”

Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

Among schools in Yondr administrative data, 49.10% are in the highest confidence tier, 29.85% high, 12.55% medium, 0.89% low, and 7.62% lowest.

“Highest 2262 49.10 High 1375 29.85 Medium 578 12.55 Low 41 0.89 Lowest 351 7.62 Total 4607 100.00”

May combine two claims.

Pooled ATT of Yondr adoption on discipline index is 0.059 (SE 0.024), significant at 5%.

“Panel A: Pooled ATT 0.059 * * -0.138 -0.305 -0.088 -0.039 ( 0.024) ( 0.099) ( 0.201) ( 0.085) ( 0.092)”

May combine two claims.

Phone activity is higher in months when schools are in session and lower in summer months.

“Phone activity is higher in months where schools are in session and lower in summer months”

May combine two claims. The quote only partly states this claim.

In the raw sample, Yondr-adopting schools differ significantly from never-adopting schools on total enrollment, racial composition, urbanicity, income-to-poverty ratio, and average test scores.

“| Total enrollment | 630 | 697 | 0.000 | 690 | 697 | 0.593 |”

• Also stated in: § E Robustness Checks for Test Score Estimates · plain

  “| Share White | 0.52 | 0.28 | 0.000 | 0.28 | 0.28 | 0.562 |”

• Also stated in: § E Robustness Checks for Test Score Estimates · plain

  “| Avg. test score index | 0.05 | -0.45 | 0.000 | -0.43 | -0.45 | 0.441 |”

Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

In the final analytical sample, approximately 32 percent of observations use measure (i), 61 percent use measures (ii) or (iii), and 7 percent use measures (iv) or (v).

“In the final analytical sample, approximately 32 percent of observations use measure (i), 61 percent use measures (ii) or (iii), and 7 percent use measures (iv) or (v).”

Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

Imputation affects 4,308 observations out of the 169,163 school-year observations in the final middle school test score analysis sample, and 1,469 out of 65,971 for high schools.

“In total, imputation affects 4,308 observations out of the 169,163 school-year observations in the final middle school test score analysis sample, and 1,469 out of 65,971 for high schools.”

Probably not a substantial claim: may not pass S3 (load-bearing).

In total, 152 respondents completed the survey of secondary school teachers and administrators.

“In total, 152 respondents completed the survey.”

May combine two claims.

The main findings are not sensitive to reasonable alternative choices; across definitions, including other time-windows for school-hours and other samples of schools that match the main analyses, there is the same qualitative pattern of a sustained decline in phone activity after adoption.

“The results in Tables A10 and A11 show that the main findings are not sensitive to reasonable alternative choices.”

• Also stated in: § Additional results. · plain

  “Across definitions, including other time-windows for school-hours and other samples of schools that match our main analyses (for test scores, attendance, and discipline outcomes), we find the same qualitative pattern of a sustained decline in phone activity after adoption.”

May combine two claims.

Middle school ATT of Yondr adoption on math test scores is -0.027 (SE 0.014), significant at 10%.

“1.Baseline -0.003 210,951 0.048 * * * 64,502 -0.027 * 159,048 ( 0.012) ( 0.018) ( 0.014)”

May combine two claims.

High school ATT of Yondr adoption on attention index is -0.348 (SE 0.169), significant at 5%.

“Panel B: High School ATT 0.073 * * -0.244 -0.156 -0.348 * * -0.067 ( 0.037) ( 0.161) ( 0.204) ( 0.169) ( 0.191)”

May combine two claims. The quote only partly states this claim.

In high schools, Yondr adoption increased math test scores by 0.048 standard deviations (SE 0.018), significant at the 1% level.

“| ATT | 0.025 | 0.048 * * * | 0.011 |”

• Also stated in: § E Robustness Checks for Test Score Estimates · plain

  “| | ( 0.017) | ( 0.018) ( 0.022) | |”

• AssumptionThe authors state this as an assumption: The baseline framework assumes that once a school adopts Yondr, it remains treated.

Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

If neither subject is observed nor can be imputed, the average score remains missing; this case does not occur in the analysis sample.

“If neither subject is observed nor can be imputed, the average score remains missing; this case does not occur in the analysis sample.”

May combine two claims. The quote only partly states this claim.

ATT of Yondr adoption on chronic absenteeism rate is 0.197 (SE 0.242), not significant.

“ATT 0.197 -0.100 -0.086 ( 0.242) ( 0.099) ( 0.099)”

May combine two claims.

Phone activity increases with school enrollment, yielding values of visits per student that are close to or above 1.

“increases with enrollment-yielding values of visits per student that are close to or above 1”

The quote only partly states this claim.

The estimated ATT of Yondr adoption on pooled ELA test scores was -0.002 (SE 0.012).

“| ATT | -0.004 | -0.003 | -0.002 |”

• Also stated in: § E Robustness Checks for Test Score Estimates · plain

  “| | ( 0.010) | ( 0.012) ( 0.012) | |”

• AssumptionThe authors state this as an assumption: The baseline framework assumes that once a school adopts Yondr, it remains treated.

Probably not a substantial claim: may not pass S3 (load-bearing).

The coverage restriction drops three of the seven items in the classroom attention construct.

“This drops three of the seven items in the classroom attention construct.”

May combine two claims.

The correlations between GPS-based net visits and NTS teacher-reported in-class phone misuse are positive and statistically significant for different hour-window definitions: 0.147 (p<0.001) for 8:00-15:00, 0.161 (p<0.001) for 9:00-13:00, and 0.161 (p<0.001) for 10:00 single-hour.

“For the 8:00-15:00 measure, the correlation is 0.147 (p=< 0.001); for the 9:00-13:00 measure, the correlation is 0.161 (p=< 0.001); and for the 10:00 single-hour measure, the correlation is 0.161 (p=< 0.001).”

The quote only partly states this claim.

The estimated ATT of Yondr adoption on pooled math test scores was -0.003 (SE 0.012).

“| ATT | -0.004 | -0.003 | -0.002 |”

• Also stated in: § E Robustness Checks for Test Score Estimates · plain

  “| | ( 0.010) | ( 0.012) ( 0.012) | |”

• AssumptionThe authors state this as an assumption: The baseline framework assumes that once a school adopts Yondr, it remains treated.

May combine two claims.

GPS-measured phone activity is systematically lower in schools with more restrictive phone policies (e.g., bell-to-bell) and higher in schools with weaker policies.

“GPS activity is systematically lower in schools with more restrictive phone policies (e.g., bell-to-bell) and higher in schools with weaker policies.”

The quote only partly states this claim.

The estimated ATT of Yondr adoption on the pooled attendance rate was -0.138 (SE 0.099).

“| ATT | 0.059 * * | -0.138 | -0.305 | -0.088 | -0.039 |”

• Also stated in: § E Robustness Checks for Test Score Estimates · plain

  “| | ( 0.024) | ( 0.099) | ( 0.201) | ( 0.085) | ( 0.092) |”

• AssumptionThe authors state this as an assumption: The baseline framework assumes that once a school adopts Yondr, it remains treated.

The quote only partly states this claim.

The estimated ATT of Yondr adoption on pooled combined test scores was -0.004 (SE 0.010).

“| ATT | -0.004 | -0.003 | -0.002 |”

• Also stated in: § E Robustness Checks for Test Score Estimates · plain

  “| | ( 0.010) | ( 0.012) ( 0.012) | |”

• AssumptionThe authors state this as an assumption: The baseline framework assumes that once a school adopts Yondr, it remains treated.

May combine two claims.

Middle school ATT of Yondr adoption on standardized test scores is -0.024 (SE 0.012), significant at 5%.

“1.Baseline -0.004 214,556 0.025 65,978 -0.024 * * 162,974 ( 0.010) ( 0.017) ( 0.012)”

May combine two claims.

In California, the estimated effect of 0.059 SD on suspension rates corresponds to an increase of roughly 0.33 percentage points, about 11% of the baseline suspension rate.

“In California, the school-level standard deviation of suspension rates is approximately 5.6 percentage points.Multiplying this by the estimated effect of 0.059 implies an increase of roughly 0.33 percentage points, corresponding to about 11 percent of the baseline suspension rate.”

• AssumptionThe authors state this as an assumption: The baseline framework assumes that once a school adopts Yondr, it remains treated.

The quote only partly states this claim.

The estimated ATT of Yondr adoption on pooled online bullying was -0.039 (SE 0.092).

“| ATT | 0.059 * * | -0.138 | -0.305 | -0.088 | -0.039 |”

• Also stated in: § E Robustness Checks for Test Score Estimates · plain

  “| | ( 0.024) | ( 0.099) | ( 0.201) | ( 0.085) | ( 0.092) |”

• AssumptionThe authors state this as an assumption: The baseline framework assumes that once a school adopts Yondr, it remains treated.

Probably not a substantial claim: may not pass S3 (load-bearing).

Schools enter and exit the Panorama samples depending on whether the relevant survey modules were administered in a given year.

“Schools enter and exit the sample depending on whether the relevant survey modules were administered in a given year.”

The quote only partly states this claim.

In middle schools, Yondr adoption decreased math test scores by 0.027 standard deviations (SE 0.014), significant at the 10% level.

“| ATT | -0.024 * * | -0.027 * -0.017 | |”

• Also stated in: § E Robustness Checks for Test Score Estimates · plain

  “| | ( 0.012) | ( 0.014) ( 0.013) | |”

• AssumptionThe authors state this as an assumption: The baseline framework assumes that once a school adopts Yondr, it remains treated.

May combine two claims.

High school ATT of Yondr adoption on math test scores is 0.048 (SE 0.018), significant at 1%.

“1.Baseline -0.003 210,951 0.048 * * * 64,502 -0.027 * 159,048 ( 0.012) ( 0.018) ( 0.014)”

May combine two claims.

Across all specifications, the pooled estimates remain close to zero and precisely estimated, while the pattern of positive effects in high schools and negative effects in middle schools is preserved, particularly for Mathematics.

“Across all specifications, the pooled estimates remain close to zero and precisely estimated, while the pattern of positive effects in high schools and negative effects in middle schools is preserved, particularly for Mathematics.”

The quote only partly states this claim.

The estimated ATT of Yondr adoption on the pooled classroom attention index was -0.088 (SE 0.085).

“| ATT | 0.059 * * | -0.138 | -0.305 | -0.088 | -0.039 |”

• Also stated in: § E Robustness Checks for Test Score Estimates · plain

  “| | ( 0.024) | ( 0.099) | ( 0.201) | ( 0.085) | ( 0.092) |”

• AssumptionThe authors state this as an assumption: The baseline framework assumes that once a school adopts Yondr, it remains treated.

The quote only partly states this claim.

In middle schools, Yondr adoption decreased combined test scores by 0.024 standard deviations (SE 0.012), significant at the 5% level.

“| ATT | -0.024 * * | -0.027 * -0.017 | |”

• Also stated in: § E Robustness Checks for Test Score Estimates · plain

  “| | ( 0.012) | ( 0.014) ( 0.013) | |”

• AssumptionThe authors state this as an assumption: The baseline framework assumes that once a school adopts Yondr, it remains treated.

The quote only partly states this claim.

The estimated ATT of Yondr adoption on the pooled student subjective well-being index was -0.305 (SE 0.201).

“| ATT | 0.059 * * | -0.138 | -0.305 | -0.088 | -0.039 |”

• Also stated in: § E Robustness Checks for Test Score Estimates · plain

  “| | ( 0.024) | ( 0.099) | ( 0.201) | ( 0.085) | ( 0.092) |”

• AssumptionThe authors state this as an assumption: The baseline framework assumes that once a school adopts Yondr, it remains treated.

May combine two claims.

Pooled ATT of Yondr adoption on math test scores is -0.003 (SE 0.012), not significant.

“1.Baseline -0.003 210,951 0.048 * * * 64,502 -0.027 * 159,048 ( 0.012) ( 0.018) ( 0.014)”

May combine two claims.

The classroom attention items share a common conceptual focus and were consistently rated as highly responsive to phone restrictions by educators.

“Although drawn from different modules, these items share a common conceptual focus and were consistently rated as highly responsive to phone restrictions by educators.”

May combine two claims.

Across all specifications, the estimated effects of Yondr adoption on test scores remain small and close to zero in the pooled sample, with no evidence of meaningful impacts.

“Across all specifications, the estimated effects remain small and close to zero in the pooled sample, with no evidence of meaningful impacts of Yondr adoption on test scores.”

• AssumptionThe authors state this as an assumption: The baseline framework assumes that once a school adopts Yondr, it remains treated.

Probably not a substantial claim: may not pass S3 (load-bearing).

Approximately 7.1% of schools appear in both the middle school and high school test score samples.

“Second, we address the fact that a relatively small share of schools (approximately 7.1%) appear in both the middle school and high school test score samples.”

May combine two claims.

Phone activity during school hours is higher in middle and high schools than in elementary schools, scales positively with school enrollment, and is substantially lower at night and on weekends.

“Phone activity during school hours is higher in middle and high schools than in elementary schools, scales positively with school enrollment, and is substantially lower at night and on weekends.”

• AssumptionRests on an assumption: GPS pings are a valid proxy for relative phone use across schools and periods.

May combine two claims.

The focus on Yondr allows the study to leverage detailed administrative data from the company, yielding a treatment measure with precise information on adoption timing and broad national coverage.

“First, our focus on Yondr allows us to leverage detailed administrative data from the company, yielding a treatment measure with precise information on adoption timing and broad national coverage.”

Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

As of 2026, roughly two-thirds of U.S. states and more than half of countries worldwide have enacted legislation limiting student phone access.

“Policymakers have responded by sharply restricting phone use in schools: as of 2026, roughly two-thirds of U.S. states and more than half of countries worldwide have enacted legislation limiting student phone access (D'Addio, 2025; Prothero, Langreo and Klein, 2026).”

• cites D'Addio, 2025Not matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

• cites Prothero, Langreo and Klein, 2026Not matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

May combine two claims.

The main results are robust to restricting the control group to schools reporting no recent phone policy changes, although estimates become less precise in this case.

“We show that our main results are robust to restricting the control group to schools reporting no recent phone policy changes, although our estimates become less precise in this case.”

Probably not a substantial claim: may not pass S3 (load-bearing), S4 (based).

Nearly 5000 schools were using Yondr pouches by 2026.

“The intervention has been widely adopted, with nearly 5000 schools using pouches by 2026.”

Probably not a substantial claim: may not pass S3 (load-bearing), S4 (based).

The Nationwide Teacher Survey sample includes roughly 50 percent of all middle and high schools in the country.

“The NTS sample includes roughly 50 percent of all middle and high schools in the country.”

May combine two claims.

Relative to the NCES population, schools represented in the Nationwide Teacher Survey are larger on average, more likely to serve high school grades, somewhat more suburban, less likely to be charter schools, and serve student populations from higher-income neighborhoods.

“Relative to the NCES population, schools represented in the NTS are larger on average and more likely to serve high school grades.NTS schools are also somewhat more suburban, less likely to be charter schools, and serve student populations that are from higher-income neighborhoods.”

Probably not a substantial claim: may not pass S3 (load-bearing).

The details of the analysis were publicly preregistered, including empirical specifications and outcome definitions.

“The details of our analysis were publicly preregistered, including empirical specifications and outcome definitions; the pre-analysis plan is available here.”

Classified as probably not a substantial claim.

The authors adopted a pre-specified process to identify a small number of primary outcomes prior to estimating treatment effects.

“Because Panorama's platform includes a large number of distinct survey items and because survey content varies across schools and years, we adopted a pre-specified process to identify a small number of primary outcomes prior to estimating treatment effects.”

May combine two claims.

The estimator allows for the inclusion of covariates, so identification relies on parallel trends conditional on observed characteristics.

“Importantly, the estimator allows for the inclusion of covariates, so identification relies on parallel trends conditional on observed characteristics, which is important in our setting where adopters differ substantially from non-adopters.”

Probably not a substantial claim: may not pass S3 (load-bearing).

The analysis was pre-registered.

“First, as discussed above, the analysis was pre-registered.”

May combine two claims.

While both groups are most optimistic about mental health effects, parents are substantially more positive than students across all outcomes and differ sharply on support for the ban itself.

“Thus, while both groups are most optimistic about mental health effects, parents are substantially more positive than students across all outcomes and differ sharply on support for the ban itself.”

May combine two claims.

Together with the GPS-based results, the NTS findings indicate that Yondr adoption meaningfully reduces in-school phone use, even if enforcement may be imperfect and the exact magnitude of the reduction cannot be measured directly.

“Together with the GPS-based results, these findings indicate that Yondr adoption meaningfully reduces in-school phone use, even if enforcement may be imperfect and the exact magnitude of the reduction cannot be measured directly.”

• AssumptionThe authors state this as an assumption: The causal interpretation of the difference-in-differences estimates relies on the parallel trends assumption: treated and comparison schools would have followed parallel trends absent adoption, conditional on baseline characteristics.

Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

The comparisons are descriptive and rely on retrospective teacher reports.

“We emphasize that these comparisons are descriptive and rely on retrospective teacher reports.”

Probably not a substantial claim: may not pass S3 (load-bearing).

In all specifications, standard errors are clustered at the school level.

“Standard errors are clustered at the school level.”

Probably not a substantial claim: may not pass S3 (load-bearing).

Before exploring any subgroup analysis, we pre-registered dimensions of heterogeneity that we thought were most important to test.

“Before exploring any subgroup analysis, we pre-registered dimensions of heterogeneity that we thought were most important to test.”

May combine two claims.

The Yondr schools observed in the NTS data represent only the subset of adopters that overlap with both our survey sample and the administrative adoption records and may not be fully representative of all Yondr schools nationwide.

“In addition, as discussed above, the Yondr schools observed in the NTS data represent only the subset of adopters that overlap with both our survey sample and the administrative adoption records and may not be fully representative of all Yondr schools nationwide.”

May combine two claims.

Across alternative specifications, pooled estimates remain close to zero and precisely estimated, while the pattern of positive effects in high schools and negative effects in middle schools is preserved.

“Across these specifications, pooled estimates (across middle and high schools) remain close to zero and precisely estimated, while the pattern of positive effects in high schools and negative effects in middle schools is preserved.”

Classified as probably not a substantial claim.

Because several student outcome measures are unavailable during the 2020 and 2021 academic years due to the COVID-19 pandemic, we exclude these two years from the analysis.

“Because several of the student outcome measures are unavailable during the 2020 and 2021 academic years due to the COVID-19 pandemic, we exclude these two years from the analysis.”

May combine two claims.

Panorama surveys are administered at the discretion of schools and are observed for a relatively small share of school-year observations, so these estimates reflect a potentially selected subset of schools and should be interpreted accordingly.

“As discussed in Section 3, Panorama surveys are administered at the discretion of schools and are observed for a relatively small share of school-year observations, so these estimates reflect a potentially selected subset of schools and should be interpreted accordingly.”

May combine two claims.

This corresponds to an increase of roughly 0.9 percentile points in the test score distribution (von Hippel, 2025) and about one-fifth of the gain associated with a one standard deviation increase in teacher value added (Chetty, Friedman and Rockoff, 2014).

“For comparison, this corresponds to an increase of roughly 0.9 percentile points in the test score distribution (von Hippel, 2025) and about one-fifth of the gain associated with a one standard deviation increase in teacher value added (Chetty, Friedman and Rockoff, 2014).”

• cites von Hippel, 2025Not matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

• cites Chetty, Friedman and Rockoff, 2014 sourceNot checked — no full text of the source

Probably not a substantial claim: may not pass S3 (load-bearing).

The estimate for the 2024 cohort at t + 1 is not reported because the estimator does not return an ATT(g, t) for this cell, plausibly due to limited support or insufficient comparison observations.

“The estimate for the 2024 cohort in t + 1 is not reported because the estimator does not return an ATT(g, t) for this cell, plausibly due to limited support or insufficient comparison observations.”

Probably not a substantial claim: may not pass S3 (load-bearing).

Effects of this magnitude are modest relative to many other educational interventions.

“Effects of this magnitude are modest relative to many other educational interventions.”

Classified as probably not a substantial claim.

In all event-study specifications, never-treated schools serve as the comparison group.

“Never-treated schools serve as the comparison group.”

Probably not a substantial claim: may not pass S3 (load-bearing).

To maintain a consistent event-time structure, we relabel the 2018 and 2019 school years as 2020 and 2021, respectively, so that event times t-5,...,t-1 correspond to successive pre-adoption years.

“To maintain a consistent event-time structure, we relabel the 2018 and 2019 school years as 2020 and 2021, respectively, so that event times t -5, . . ., t -1 correspond to successive pre-adoption years.”

May combine two claims.

These patterns suggest that the modest positive effects observed in high schools are relatively broad-based, though somewhat larger for high-poverty schools, while the absence of positive average effects in middle schools does not mask benefits for any particular group.

“These patterns suggest that the modest positive effects observed in high schools are relatively broad-based, though somewhat larger for high-poverty schools, while the absence of positive average effects in middle schools does not mask benefits for any particular group.”

May combine two claims.

The initial decline in SWB is driven primarily by later-adopting cohorts, while the subsequent recovery reflects earlier-adopting cohorts.