Reducing Parent–School Information Gaps and Improving Education Outcomes

Notes: “Text messages sent” refers to the cumulative number of text messages sent to student’s parents. “Text messages received” refers to the cumulative number of text messages with a confirmed delivery status. Columns 2–5 report the T_ic coefficient of Equation 1 with the annual number of each type of text message as the dependent variable. Column 1 adds all types of text messages. Attendance, grades, and classroom behavior text messages were sent only to the treatment group. General text messages were sent to all treatment and control individuals. All models include the baseline math grade, attendance rate as control variables, classroom (randomization strata), and year fixed effects. If baseline values of baseline math grade/attendance were missing, we imputed them using the classroom-level mean and added an indicator variable for these imputed observations. Standard errors are clustered at the classroom level (shown in brackets). Significance: *p < 0.1, **p < 0.05, ***p < 0.01.

Notes: Panel A shows the intention-to-treat (T) estimates and its corresponding standard error estimated using Equation 1 using OLS. Panel B adds the interaction with the student-level at-risk index. At-risk index is a simple average of standardized baseline attendance, math grades, and negative behavioral notes. All models include the baseline math grade, attendance rate as control variables, classroom (randomization strata), and year fixed effects. If baseline values of math grade/attendance were missing, we imputed them using the classroom-level mean and added an indicator variable for these imputed observations. Models in Panel B additionally include the at-risk index variable as a control. Columns 1 and 5 report results on outcomes that were standardized so that the mean among the control students is zero and the standard deviation is one. Standard errors are clustered at the classroom level (shown in brackets). Significance: *p < 0.1, **p < 0.05, ***p < 0.01.

Notes: Each row shows the intention-to-treat estimates and its corresponding standard error estimated using Equation 2 using OLS. T refers to the randomized individual-level treatment (equal to one if parents were sent text messages and zero otherwise). High-share refers to the randomized classroom-level treatment (equal to one for high-share classrooms and zero for low-share classrooms). All models include the baseline math grade, attendance rate as control variables, classroom (randomization strata), and year fixed effects. If baseline values of math grade/attendance were missing, we imputed them using the classroom-level mean and added an indicator variable for these imputed observations. Columns 1 and 5 report results on outcomes that were standardized so that the mean among the control students is zero and the standard deviation is one. Standard errors are clustered at the classroom level (shown in brackets). Significance: *p < 0.1, **p < 0.05, ***p < 0.01.

Notes: Panel A shows intention-to-treat (T) estimates and its corresponding standard error estimated using Equation 1 using OLS. Panel B adds the interaction with the student-level at-risk index. At-risk index is a simple average of standardized baseline attendance, math grades, and negative behavioral notes. All models include the baseline math grade, attendance rate as control variables, classroom (randomization strata), and year fixed effects. If baseline values of baseline math grade/attendance were missing, we imputed them using the classroom-level mean and added an indicator variable for these imputed observations. Models in Panel B additionally include the at-risk index variable as a control. Column outcomes are indicator variables constructed by contrasting responses in parent surveys with those of student surveys or administrative records (shown in parentheses). Column 1 measures parental misinformation on all absenteeism (with and without parent permission in the previous two weeks) contrasting the responses of parents with those from students. Parents are classified as misinformed if they do not answer at least one of the questions or if at least one of the answers (in bracket days) provided by students and parents do not match. Column 2 measures misinformation on all absenteeism (with and without permission) contrasting parent responses with classroom books. The ends of original bracket days in absences with and without permission are added to construct new bracket days. Parents are classified as misinformed if they do not answer at least one of the questions or if classroom books’ records of absences over the previous two weeks do not fall in the range. Column 3 contrasts parent and student responses, and parents are classified as misinformed if they do not answer, or if reported grades’ brackets do not match. Column 4 measures parental misinformation regarding all grades by contrasting parent responses about the student’s last end-of-year grades with school records. Parents are treated as misinformed if they do not answer or if the absolute difference between reported and actual grades is greater than 0.5. Columns 5 and 6 measure misinformation about student misbehavior by contrasting parent answers with student answers and with information from classroom books, respectively. Using a four-value scale, parents and students were asked about the degree of agreement with the student’s misbehavior statements. For Column 5, parents are classified as misinformed if they do not answer at least one of the questions or if the average absolute difference between parent and student answers is larger than the median (0.8). For Column 6 parents are treated as misinformed if they do not answer; if the parent’s average answer is equal to or larger than the median (2), and student did not misbehave according to classroom books; or if the parent’s average answer is less than the median answer and student misbehaved in class according to books. Standard errors are clustered at the classroom level (shown in brackets). Significance: *p < 0.1, **p < 0.05, ***p < 0.01.

↵^a Number of observations vary by column because of survey and item nonresponse.

Notes: Panel A and Panel B show intention-to-treat (T) estimates on subjects not targeted by the intervention. Panel A shows the effect on grades and Panel B on parental misinformation regarding those grades. Point estimates and standard error were estimated using Equation 1 using OLS. All models include the baseline math grade, attendance rate as control variables, classroom (randomization strata), and year fixed effects. If baseline values of math grade/attendance were missing, we imputed them using the classroom-level mean and added an indicator variable for these imputed observations. Columns 1–3 of Panel A report results on outcomes that were standardized so that mean among the control students is zero and the standard deviation is one. Columns 1–3 of Panel B measure parental misinformation each subject grade. Parents are treated as misinformed if they do not answer or if the answered grade bracket does not match to the actual grade from administrative data. Standard errors are clustered at the classroom level (shown in parentheses). Significance: *p < 0.1, **p < 0.05, ***p < 0.01.

Notes: Panel A and Panel B shows intention-to-treat (T) estimates on parent and student standardized scales (means are zero and standard deviations are one for the control group for all scales), respectively, and its corresponding standard error estimated using Equation 1 using OLS. Outcomes are scales built with answers to surveys (see Online Appendix Tables F.2 and F.3 for details). All models include the baseline math grade, attendance rate and outcome scales as control variables, and classroom (randomization strata) and year and fixed effects. If baseline values of baseline math grade/attendance or baseline outcomes were missing, we imputed them using the classroom-level mean and added an indicator variable for these imputed observations. Standard errors are clustered at the classroom level (shown in brackets). Significance: *p < 0.1, **p < 0.05, ***p < 0.01.

↵^a Number of observations vary by column because of survey and item nonresponse; see Online Appendix VII for details.