Overstating of attendance by school authorites seems widespread in Bangladesh. To show their success school authorities often cook up the figures on attendance. Another likely reason for such exaggeration of maintaining 85 percent attendances is for the students receiving government stipends (known as 'upobrritti"). On an average, a 15.1 percentage points lower attendances in 468 schools and 17.6 percentage points lower attendances in 150 schools due to headcounts cannot be explained with a lower than normal presence during headcount or the class visits. For an average 15.1 percentage points lower attendances in 468, the difference is 18.3 percentage points lower for the control schools whereas it is 14.6 percentage points for the intervention schools. Relatively greater exaggeration of registry attendances by the control is the possibility because there are fewer counterchecks for them. For intervention schools, be it biscuit or meal, there is a third agency to match class presence with the packets of biscuits distributed or the number of meals cooked. These places pressure on them not to exaggerate much. The systematic bias in attendance exaggeration in school records conceals the impact estimation on attendances.
In the context of 468 schools under biscuit intervention, we calculate impact based on difference-in-difference of mean attendances across the intervention and the control schools. We obtain decrease in attendance by 0.8 percentage points due to biscuit intervention. The same estimate is increase in attendance by 1.9 percentage points due to biscuit intervention when we construct it based on headcounts. The impact of the intervention is 2.1 percentage points of greater attendance for the biscuit feeding schools when multivariate adjustment is made. In the context of 150 schools under meal as well as biscuit intervention, we obtain a 1.5 and 0.4 percentage points of improvement in attendance due to biscuit and meal intervention, respectively. To the contrary, we obtain 11 and 6 percentage points of improvement in attendance based on headcounts due to biscuit and meal intervention, respectively. The impacts are 8.5 and 5.7 percentage points of improvement in attendance based on headcounts due to biscuit and meal intervention, respectively, when multivariate adjustment is made.