60 cards
Test scores averaged 4 out of 10 overall due to a 48% zero-score non-compliance rate, but active participants (excluding zero-scores) averaged 8 out of 10, demonstrating that mandatory grading secured high mastery among engaged students while highlighting ongoing compliance hurdles in EdTech.
Lower secondary GPA (coefficient = -0.174, p < 0.01) and standardized entrance examination scores (coefficient = -0.003, p < 0.01) accounted for 19.0% of the total variation in upper secondary completion rates.
CUNY ASAP nearly doubled graduation rates (18 percentage point increase) at an additional cost of $16,284 per student ($42,065 total three-year program cost), whereas Stay the Course achieved a 31.5 percentage point female degree increase at $5,640 per student.
While FFT was selected as the primary rubric to approximate general district implementation, alternate observation tools demonstrated comparable predictive validity when adjusted for baseline student characteristics. However, specialized subject tools (such as MQI in math and PLATO in ELA) captured distinct domain-specific instructional practices.
Facilitated course completion and remote interaction, contributing to a 0.11 SD increase in willingness to report sexual abuse (p < 0.05) and a 0.26 SD increase in abuse reporting attitudes (p < 0.01) six months post-intervention.
ELA value-added stability coefficients between consecutive years were only 0.350 in elementary and 0.117 in middle school, compared to 0.396 and 0.512 in math. This instability stemmed from state ELA tests relying primarily on multiple-choice reading comprehension items that fail to directly capture teacher impacts on student writing.
Standardized attitude improvements decreased from 0.24 SD at 1 week post-course to 0.17 SD at 6 months post-course (p < 0.01), contrasting with factual knowledge gains which remained flat (0.37 SD to 0.38 SD).
Assigned teacher effectiveness had no statistically significant impact on student state test data availability (-0.012, SE = 0.033), 4th–8th grade MET project outcome availability (-0.070, SE = 0.143), or high school MET project outcome availability (-0.077, SE = 0.144), establishing that sample attrition was independent of treatment assignment.
Random assignment was strictly bounded within individual schools, grade levels, and subjects, preventing statistical evaluation of whether teacher value-added metrics hold predictive validity for cross-school teacher transfers or district-wide comparisons.
Qualitative surveys revealed that 80.0% of participating teachers found the training topics innovative and almost 100.0% expressed willingness to participate again, demonstrating institutional feasibility despite implementation delays.
Attending 1 to 4 CBT sessions increased 11th-grade enrollment by 7.4 percentage points (p < 0.01), while attending 5 or more sessions increased 11th-grade enrollment by 19.0 percentage points (p < 0.01) and overall student engagement by 12.1 percentage points (p < 0.01).
Patience training alone produced no significant effect on gritty task choices or perseverance after failure, whereas standalone grit training in Sample B replicated the full magnitude of treatment effects observed in Sample A's combined intervention (+19.1 vs +14.3 percentage point increase in week-ahead task commitment).
FFT classroom observation scores used as a standalone predictor yielded an IV coefficient of 0.807 (p < 0.01, SE = 0.293); while causally valid and not statistically different from 1, standalone observations exhibited roughly double the estimation error of test-based value-added measures (SE = 0.149).
80.0% of interviewed teachers confirmed that cognitive triad modeling enabled effective engagement, stressing that internal reflection was safer and more effective than forced emotional disclosure.
Participants in the top third of baseline degree propensity achieved a 65.7 percentage point increase in six-semester college enrollment (TOT, p < 0.05) and a 41.9 percentage point increase in degree completion (TOT, p < 0.10), whereas lower propensity terciles exhibited no statistically significant gains.
Service plans were collaboratively reviewed and updated every 90 days, enabling navigators to build sufficient rapport for students to disclose intimate personal crises and complete action steps across 33.8 interactions per semester.
The incremental component removed by controlling for student demographic characteristics yielded an IV coefficient of -0.025 (SE = 0.521), demonstrating that removing demographic variables does not significantly alter post-randomization outcome predictions, though large standard errors leave the policy trade-off statistically imprecise.
Evaluation is ongoing with baseline phone surveys completed prior to and during COVID-19 lockdowns, and 3-month post-intervention follow-up surveys conducted via phone [8, 10].
First-stage compliance coefficients ranged from 0.744 in Dallas and 0.517 in Charlotte-Mecklenburg down to 0.157 in New York and -0.079 (non-significant) in Denver. Low compliance in Denver stemmed from structural language certification rules requiring Spanish-speaking students to remain with Spanish-certified teachers.
Using raw value-added scores without shrinkage generated actual achievement differences following random assignment that were only 43% as large as predicted (coefficient = 0.430, SE = 0.057). Single-year math value-added required multiplication by shrinkage factors of 0.396 in elementary and 0.512 in middle school to predict future performance.
Leveraged existing labs to deliver instruction at a low marginal cost of $14.60 per student, though 3 of 46 treatment groups suffered non-implementation due to localized internet connectivity failures.
Treatment-on-the-Treated (TOT) associate degree completion for female participants tripled (31.5 percentage point increase, p < 0.05) off a 15.7% control group mean, while male outcomes showed no significant impact (ITT = -0.025, standard error = 0.040). Standalone Action: Pair male postsecondary students with gender-matched mentors and structure coaching frameworks to specifically address gender-based differences in advising receptivity.
Prior-year teacher effectiveness predicted post-randomization high school achievement across stacked subjects (math, ELA, biology) with an IV coefficient of 0.828 (p < 0.01, SE = 0.292), confirming causal validity in secondary education despite a 53% student sample attrition rate.
Even among poor households with spending near baseline levels, 88 percent spent more than the grant amount (Rs 125/year), allowing anticipated grants to relax general budget constraints and result in a 76% private expenditure offset (elasticity of -0.21, p < 0.01).
Strict promotion criteria (failing 4 or more of 8 subjects requiring complete semester repetition) yielded a 15.6% grade repetition rate, a 15.5% annual dropout rate, and an overall on-time graduation rate of only 60.0%.
Randomly assigned teacher effectiveness translated into actual teacher effectiveness at a rate of 0.293 (first stage) due to non-compliance. The IV coefficient of actual teacher effectiveness on student test achievement was 0.955 (p < 0.01, SE = 0.123), proving that a 1-unit predicted difference caused a statistically equivalent ~1-unit actual achievement gain.
The intent-to-treat reduced-form estimate yielded a coefficient of ~0.280 to 0.326, which reflected the true treatment impact (~0.955) diluted by an overall compliance rate of approximately 29.3%.
Broad expansion in low-skill formal export manufacturing jobs substantially raised the probability of quitting upper secondary school for youth on the dropout margin at age 16.
LIML estimation yielded a pooled causal effectiveness coefficient of 0.955 (p < 0.01, SE = 0.123) while maintaining a minimum first-stage F-statistic of 38 across instrument specifications, eliminating weak-instrument bias concerns.
Targeted and untargeted cash scholarship interventions produced a precisely estimated null effect (0 percentage point impact) on high school graduation rates.
Because only 29.3% of the assigned teacher effectiveness translated into actual teacher effectiveness due to student and teacher movement, the LATE IV estimate scaled the reduced-form impact (~0.28) up to 0.955 (p < 0.01, SE = 0.123), proving that the true causal impact for compliant rosters was approximately 1:1 with predicted effectiveness.
Aggregate sexual knowledge gain remained constant at 0.37 SD at 1 week post-intervention and 0.38 SD at 6 months post-intervention (p < 0.01), showing zero knowledge decay over time.
Joint hypothesis tests produced p-values of 0.163 for the full randomized sample, 0.282 for the state test sample, and 0.106 for the supplemental outcome sample, confirming that randomized assignment created statistically balanced groups.
The multi-measure composite predicted actual student achievement gains following random assignment with an IV slope coefficient of 0.955 (p < 0.01, SE = 0.123), demonstrating that composite ratings accurately match causal student achievement impacts.
42% of two-year public college starters drop out or fail to earn credentials within six years; 90% of debt defaulters leave without a degree; 71% cite full-time work requirements as the primary reason for dropping out.
A 1.0 standard deviation increase in predicted teacher effectiveness caused a 0.966 standard deviation increase in student-reported classroom enjoyment (p < 0.01, SE = 0.285), whereas effects on grit (-0.029, SE = 0.216), growth mindset (0.331, SE = 0.220), and effort (0.199, SE = 0.214) were statistically non-significant.
School enrollment increases for young girls in treatment villages (5.0 percentage points overall, p < 0.01) were equally pronounced in households where parents had zero formal education, proving demand-side labor market returns offset parental educational deficits.
Assigned teacher effectiveness showed no statistically significant relationship with peer baseline math scores (coefficient = 0.246, SE = 0.164) or other peer traits; joint hypothesis p-values were 0.405 for grades 4–8 and 0.321 for high school, proving post-randomization peer stability.
The component removed by peer effect controls predicted student achievement after random assignment with a coefficient of 1.150 (p < 0.01, SE = 0.336), proving it contains causal teacher effectiveness. In contrast, the component removed by individual baseline score controls yielded a coefficient of 0.047 (SE = 0.042), showing baseline score controls successfully remove non-teacher sorting noise.
Over 50% of public community colleges operate at counselor-to-student ratios between 1:1,500 and 1:3,500, contrasting sharply with intensive models utilizing 34:1 caseloads.
The signal standard deviation of persistent within-school student baseline sorting was 0.297 in math and 0.300 in ELA for randomized MET teachers, compared to 0.437 in math and 0.440 in ELA for non-randomized MET school teachers, proving that structural student sorting was substantial prior to experimental intervention.
Free-lunch eligible students comprised 50.6% of applicants in randomized marginal priority groups at high-performing schools compared to only 33.6% in waitlisted groups, significantly increasing low-income admission rates and counteracting strategic hedging incentives.
668 randomization blocks were successfully formed across 284 schools. Standard deviation of baseline student sorting across classrooms within these blocks was 0.382 to 0.520 prior to randomization, demonstrating that blocked randomization successfully eliminated substantial pre-existing within-school student sorting.
SURE joint hypothesis testing yielded p-values of 0.163 across the initial randomized sample, 0.282 for the state test sample, and 0.405 for subsequent peer characteristics, statistically confirming successful experimental balance across all baseline features simultaneously.
The estimated signal standard deviation of true teacher effects ranged from 0.16 to 0.18 student-level standard deviations in math and 0.10 to 0.12 in English Language Arts (ELA). ELA demonstrated lower signal variance and higher year-to-year volatility due to reliance on multiple-choice reading comprehension items without direct writing assessments.
Implementation resulted in a 97.0% zero-attendance rate among targeted students due to out-of-school scheduling friction, preventing statistical verification of math achievement gains.
No statistically significant difference in six-semester college enrollment (ITT = 0.015, standard error = 0.037) or associate degree completion (ITT = -0.012, standard error = 0.027) compared to the control group.
Attending 5 or more CBT sessions reduced the probability of retaking the COMIPEMS placement exam by 6.8 percentage points (p < 0.01), whereas non-attending treatment students saw a 2.5 percentage point increase (p < 0.05) in re-sitting the exam as part of disengagement.
Prior to randomization, the within-school standard deviation of teacher mean baseline math scores was 0.382 to 0.520. Evaluating instructors using unadjusted end-of-year student test scores yielded a severely biased predictive coefficient of 0.228 (SE = 0.042), significantly overstating true instructor impact.
Standalone Tripod student perception survey ratings predicted post-randomization student test achievement with an IV coefficient of 0.940 (p < 0.01, SE = 0.311). While causally valid, standalone survey ratings exhibited roughly double the estimation error of test-based value-added measures (SE = 0.149).
Instructors predicted to raise state test scores by 1.0 standard deviation increased student performance on supplemental conceptual assessments by 0.661 standard deviations (p < 0.01, SE = 0.153), indicating significant cross-format skill transfer at approximately two-thirds the state test effect size.
Intention-to-treat analysis demonstrated a statistically significant increase of 0.165 to 0.175 standard deviations in socio-emotional skills measured on the PLANEA scale (p < 0.05), but no statistically significant effect on math test scores or overall enrollment.
Students averaged 33.81 total interactions per semester, consisting of 4.13 in-person meetings (averaging 41.25 minutes each), 17.08 text/email exchanges, and 9.70 phone calls.
Screening out students with > 30 accumulated credit hours successfully isolated a vulnerable cohort at risk of dropping out, yielding a 31.5 percentage point increase in female associate degree completion (TOT, p < 0.05).
Averaging student residuals provided an unbiased baseline-adjusted measure of teacher performance that, when combined with shrinkage adjustments, predicted post-randomization achievement gains with an IV slope of 0.955 (SE = 0.123).
Unadjusted end-of-year score ratings yielded a predictive coefficient of 0.228 (p < 0.01, SE = 0.042), significantly overstating true teacher effectiveness differences compared to baseline-adjusted growth models (coefficient = 0.955).
In English language arts, including an extreme outlier teacher with a 1.666 SD value-added swing dropped the predictive coefficient to an imprecise 0.375 (SE = 0.264); trimming the top 1% restored the ELA coefficient to 0.697 (p < 0.01, SE = 0.213) and overall composite predictive coefficient to 0.955 (SE = 0.123).
Students were ranked based on raw scores and mapped to standard normal quantiles (mean 0, SD 1). This eliminated cross-district scale artifacts while preserving relative performance ranks across 6 distinct testing regimes.
Teacher effect estimates generated from random-effects specifications controlling for classroom section-mean covariates are highly correlated with teacher fixed-effects specifications because the vast majority of student baseline variance exists within classrooms rather than between classrooms.
Integrating structured youth engagement exercises enabled lay teacher-facilitators to build an atmosphere of trust, with qualitative surveys showing nearly 100.0% teacher willingness to participate in future implementations.