Journal of Public Economics 145 (2017) 116–135 Contents lists available at ScienceDirect Journal of Public Economics journal homepage: www.elsevier.com/locate/jpube The fiscal cost of weak governance: Evidence from teacher absence in India Karthik Muralidharan a, * , Jishnu Das b , Alaka Holla b , Aakash Mohpal c a University of California, San Diego, CA, United States b The World Bank Group, Washington, DC, United States c University of Michigan, Ann Arbor, MI, United States A R T I C L E I N F O A B S T R A C T Article history: The relative return to strategies that augment inputs versus those that reduce inefficiencies remains a key Received 25 September 2015 open question for education policy in low-income countries. Using a new nationally-representative panel Received in revised form 6 November 2016 dataset of schools across 1297 villages in India, we show that the large public investments in education over Accepted 7 November 2016 the past decade have led to substantial improvements in input-based measures of school quality, but only a Available online 17 November 2016 modest reduction in inefficiency as measured by teacher absence. In our data, 23.6% of teachers were absent during unannounced school visits, and we estimate that the salary cost of unauthorized teacher absence is JEL classification: $1.5 billion/year. We find two robust correlations in the nationally-representative panel data that corrobo- H4 rate findings from smaller-scale experiments. First, reductions in student-teacher ratios are correlated with H52 I21 increased teacher absence. Second, increases in the frequency of school monitoring are strongly correlated M54 with lower teacher absence. Using these results, we show that reducing inefficiencies by increasing the fre- O15 quency of monitoring could be over ten times more cost effective at increasing the effective student-teacher ratio than hiring more teachers. Thus, policies that decrease the inefficiency of public education spending Keywords: are likely to yield substantially higher marginal returns than those that augment inputs. Education © 2016 The Authors. Published by Elsevier B.V. This is an open access article under the CC BY license Teacher absence (http://creativecommons.org/licenses/by/4.0/). Teacher absenteeism India Governance State capacity Monitoring 1. Introduction In this paper, we study one striking measure of public sector inefficiency - teacher absences - with panel data collected 7 years Determining the optimal level and composition of public educa- apart in India at a time of sharp increases in education spending. tion spending is a key policy question in most low-income countries. A large portion of this increase was accounted for by the salary Many education advocates believe that low-income countries need cost of hiring teachers to reduce the student-teacher ratio in pub- substantial increases in public education spending to meet enroll- lic schools. As a policy alternative to hiring more teachers, we ment and learning goals (UNESCO, 2014); others argue that pub- show that reducing teacher absences by increasing school moni- lic sector inefficiencies leave considerable room for improvement toring could be over ten times more cost effective at reducing the within existing education budgets, and that fiscal constraints make effective student-teacher ratio (net of teacher absence). Thus, while it imperative to improve the efficiency of public expenditure (World the default approach to improving education in low-income coun- Bank, 2010). However, the data to assess the relative importance of tries is input-augmentation, our results suggest that investing in these contentions remains sparse, in part, due to the difficulty in reducing inefficiencies may yield much greater returns. detecting and measuring inefficiencies in public spending. India presents a particularly salient setting for our analysis. It has the largest primary education system in the world, cater- ing to over 200 million children. Further, over the past decade, * Corresponding author. the Government of India has invested heavily in primary educa- E-mail addresses: kamurali@ucsd.edu (K. Muralidharan), jdas1@worldbank.org tion under the Sarva Shiksha Abhiyan (SSA) or “Education for All (J. Das), aholla@worldbank.org (A. Holla), amohpal@umich.edu (A. Mohpal). Campaign.” Partly financed by a special education tax, this national http://dx.doi.org/10.1016/j.jpubeco.2016.11.005 0047-2727/ © 2016 The Authors. Published by Elsevier B.V. This is an open access article under the CC BY license (http://creativecommons.org/licenses/by/4.0/). K. Muralidharan et al. / Journal of Public Economics 145 (2017) 116–135 117 program sought to correct historical inattention to primary educa- Our panel-data analysis, where we correlate changes in village- tion and led to a substantial increase in annual spending on primary level teacher absence with changes in teacher and school char- education across several major categories of inputs including school acteristics, and administrative and community-level monitoring, infrastructure, teacher quality, student-teacher ratios, and school yields two robust correlations. First, reductions in the school-level feeding programs.1 student-teacher ratio (STR) are correlated with an increase in teacher However, the public education system in India also faces substan- absence, suggesting that the potential benefits from investing in tial governance challenges that may limit the extent to which this more teachers and lower STR may be partly offset by an increase additional spending translates into improved education outcomes. in teacher absence. Second, better top-down administrative moni- Our indicator of systemic inefficiency - teacher absence - presents toring is strongly correlated with lower teacher absence. Absence a particularly striking indicator of weak governance. A nationally- rates were 6.5 percentage points lower in villages with regular pub- representative study of over 3000 public primary schools across 19 lic school inspections relative to those without, which is a 25% major Indian states found that over 25% of teachers were absent reduction in overall absence and a 40% decline in unauthorized from work on a typical working day in 2003 (Kremer et al., 2005). absence.4 Although administrative data from the government’s official records One way to estimate the cost of teacher absence is to calculate the suggest that SSA has led to an improvement in various input-based salary cost paid by the government to teachers for days of work that measures of school quality, there is little evidence on whether these they did not attend. We estimate this fiscal cost to be over $1.5 billion investments have translated into improvements in education sys- per year, which is around 60% of the entire revenue collected from tem performance, both with respect to intermediate metrics such as the special education tax used to fund SSA in 2010.5 Teacher salaries teacher absence and final outcomes such as test scores.2 typically account for over 80% of non-capital education spending Our study of this nationwide campaign to improve school qual- (Dongre et al., 2014), and the most expensive component of the ity in India uses a new nationally-representative panel dataset of recently passed Right to Education (RtE) Act in India is a commit- education inputs and outcomes that we collected in 2010. We con- ment to reduce STR from 40:1 to 30:1, by hiring more teachers at structed this dataset by revisiting a randomly-sampled subset of an additional cost of $5 billion/year. Using the most conservative the villages originally surveyed in 2003 (see Kremer et al. (2005)) panel-data estimates of the correlations between increased monitor- and collecting detailed data on school facilities, teachers, commu- ing and reduced teacher absence, we estimate that improving school nity participation, monitoring visits by officials, and teacher absence governance (by hiring more supervisory staff) could be over ten times rates. Thus, in addition to reporting updated estimates of teacher more cost effective at increasing effective student-teacher ratio (net absence, and independently-measured summary statistics on input- of teacher absence) than hiring more teachers. These calculations based measures of school quality, we are able to correlate changes suggest that the marginal returns to investing in an inefficiency- in input-based measures of school quality with changes in teacher reduction strategy (through better monitoring and governance of absence. The panel data help mitigate concerns arising from fixed the education system) are likely to be much higher than a typical unobserved heterogeneity at the village-level, and let us study how input-augmentation strategy. the sharp increases in public education spending over the last decade This paper makes several contributions to the literature on pub- have affected school quality. lic economics in low-income countries. First, teacher absence is now We find significant improvements in almost all input-based mea- widely used as a governance indicator in education in low- and sures of school quality between 2003 and 2010. The fraction of middle-income countries.6 We update estimates of teacher absence schools with toilets and electricity more than doubled, and the in rural India from 2003 and show that despite substantial increases fraction serving mid-day meals nearly quadrupled. There were sig- in education spending over the last decade, improvements on this nificant increases in the fraction of schools with drinking water, key measure of governance have been more modest. While cor- libraries, and a paved road nearby. The fraction of teachers with col- ruption in education spending has been shown to hurt learning lege degrees increased by 41%, and student-teacher ratios (STR) fell outcomes (Ferraz et al., 2012), our results highlight the importance of by 16%. The fraction of teachers not paid on time fell from 51 to 22%, also focusing on governance issues that lead to significant amounts of and the fraction of teachers reporting the existence of teacher recog- ‘passive’ waste and inefficiency on an ongoing annual basis, but may nition programs increased from 50 to 81%. Finally, the frequency not obtain as much media attention as one-off corruption scandals of school inspections and parent-teacher association (PTA) meetings (Bandiera et al., 2009; World Bank, 2010). increased significantly. Second, the fact that decreases in STRs are correlated with However, reductions in teacher absence rates were more mod- increased teacher absence underscores the importance of distin- est. The all-India weighted average teacher absence in rural areas guishing between average and marginal rates of corruption and fell from 26.3 to 23.6%. 3 While increased teacher hiring brought waste in public spending. Niehaus and Sukhtankar (2013) propose the STR down from 47 to below 40, the effective STR (ESTR), after accounting for teacher absences was still over 50 (having reduced from 64 in 2003 to 52 in 2010). The variation in teacher absence across states remains high. At one end, several top performing states 4 have teacher absence rates below 15%, while at the other end, the These point estimates are significant and similar in both individual and multiple regressions, and in specifications with no fixed effects, with state fixed effects, and poorest performing state, Jharkhand, has a teacher absence rate of with district fixed effects. However, even with the use of panel data, we cannot rule 46%. out the possibility of time-varying omitted variables at the village-level that are cor- related with village-level changes in inspections or STR. To assess the likely bias due to unobserved heterogeneity, we show using the technique developed by Altonji et al. (2005) that the ratio of unobservable to observable correlates of changes in teacher absence would have to be over 10 for our results to be completely explained by omit- 1 In the year 2004–2005, India’s education budget was Rs.1528 billion ($25 billion) ted variables. We argue that this is unlikely given our rich data on observable changes and it more than doubled to Rs.3783 billion ($60 billion) in 2009–2010 (Pratham, in school-quality (see Section 4.2.2). 5 2010). See http://indiabudget.nic.in/budget2012-2013/ub2012-13/rec/tr.pdf. 2 6 Official records were obtained from the “District Information System for The World Bank’s World Development Report 2004 provided estimates of provider Education” data (commonly known as the DISE data). absence in both health and education for a sample of low-income countries 3 The all-India weighted average teacher absence estimated in 2003 was 25.2%; the (Chaudhury et al., 2006; World Bank, 2003). These numbers have been widely cited corresponding figure for the rural sample was 26.3%. The panel survey only covered in policy discussions, and reduction in provider absence rates is often included as an the rural sample. objective in aid agreements between donors and aid recipients. 118 K. Muralidharan et al. / Journal of Public Economics 145 (2017) 116–135 this terminology in the context of wages paid in a public-works 2. Data and analytic framework program in India and find that marginal rates of leakage are much higher than average rates. We find the same result in the con- The nationally-representative sample used for the 2003 surveys, text of teachers and show that the effective absence rate of the which our current study uses as a base, covered both urban and rural marginal teacher hired is considerably higher than the average areas across the 19 most populous states of India, except Delhi. This absence (because of the increased absence among existing teachers). represented over 95% of the country’s population. The 2010 sam- This result, from a large all-India sample, mirrors smaller-sample ple covered only rural India. The sampling strategy in 2010 aimed experimental findings in multiple settings. Duflo et al. (2015), and to maintain representativeness of the current landscape of schools Muralidharan and Sundararaman (2013) present experimental evi- in rural India, and to maximize the size of the panel. We met these dence (from Kenya and India) showing that provision of an extra twin objectives by retaining the villages in the original sample to the teacher to schools led to an increase in the absence rate of exist- extent possible, while re-sampling schools from the full universe of ing teachers in both settings. In other words, additional spending on schools in these villages in 2010, and conducting the panel analysis school inputs (of which teacher salaries are the largest component) at the village level.8 was correlated with increased inefficiency of spending. Enumerators first conducted school censuses in each village, Third, improvements in top-down administrative monitoring from which we sampled up to three schools per village for the (inspections) are more strongly correlated with reduced teacher absence surveys. During fieldwork, enumerators made three sepa- absence than improvements in bottom-up community monitoring rate visits to each sampled school over a period of 10 months from (PTA meetings), consistent with experimental evidence on the rela- January–October 2010.9 Data on school infrastructure and accessibil- tive effectiveness of administrative and community audits on reduc- ity, finances, and teacher demographics were collected once for each ing corruption in road construction in Indonesia (Olken, 2007). More school (typically during the first visit, but completed in later visits if broadly, a growing body of experimental evidence points to the necessary), while data on time-varying metrics such as teacher and effectiveness of audits and monitoring (accompanied by rewards or student attendance and dates of the most recent inspections and PTA sanctions) in improving the performance of public-sector workers meetings were collected in each of the three visits. We also assessed and service providers (including Olken (2007) in Indonesia; Duflo student learning with a test administered to a representative sam- et al. (2012) in India; and Zamboni and Litschig (2016) in Brazil). ple of fourth grade students in sampled schools. See Appendix A and Our panel-data estimates using data from an “as is” nationwide Tables A1 – A3 for further details on sampling and construction of increase in monitoring of schools provide complementary evidence the village-level panel data set. to smaller-scale experiments and suggest that investing in better Teacher absence was measured by direct physical verification of governance and monitoring of service providers may be an impor- teacher presence within the first fifteen minutes of a survey visit. tant component of improving state capacity for service delivery in Data collected during the school census were used to pre-populate low-income countries (Besley and Persson, 2009; Muralidharan et al., teacher rosters for the sampled schools, so that enumerators could 2016). look for teachers and record their attendance and activity imme- Finally, recent research has pointed to ‘misallocation’ of capi- diately after their arrival at the school.10 Once teacher attendance tal and labor in low-income countries as an important contributor was recorded, all other data were collected using interviews of head to lower total factor productivity (TFP) in these settings (Hsieh teachers and individual teachers.11 and Klenow, 2009), and has also documented that a plausible rea- We record teachers as absent on a given visit if they were not son for this misallocation is that ‘management quality’ is poorer found anywhere in the school in the first fifteen minutes after enu- in low-income countries, and that public-sector firms are managed merators reached a school. We consider all the teachers in the school especially poorly (Bloom and Van Reenen, 2010). Our results pro- to be absent if the school was closed during regular working hours on vide a striking example of weak management and misallocation in a school day, and respondents near the school did not know why the publicly-produced primary education in India (a sector that accounts school was closed or mentioned that the school was closed because for over 3% of GDP in spending). In particular, our estimates suggest no teacher had arrived or they had all left early.12 To be conservative that reallocating a portion of the $5 billion/year increase in educa- in our measure of absence, we exclude all school closures due to bad tion spending budgeted for hiring more teachers towards measures weather, school construction/repairs, school functions and alterna- focused on reducing teacher absence (for instance, by hiring more tive uses of school premises (for instance, elections). We also exclude supervisory staff) may be a much more cost effective way of increas- ing effective teacher-student contact time. Thus, misallocation is likely to be a first-order issue in this setting, and reallocating educa- 8 This is also why the 2010 wave did not include urban areas. Since school-level tion spending towards better governance may substantially increase identifiers from the 2003 survey were not preserved (for confidentiality reasons), the TFP in publicly-produced education.7 panel needed to be constructed at the town/village level. However, since the fraction of urban schools covered in 2003 (relative to the total number of schools in the sam- The rest of this paper is organized as follows: Section 2 dis- pled towns) was very small, it was not possible to construct a credible panel-data cusses our empirical methods and analytical framework. Section 3 estimate of school quality in towns. In rural areas, this was not a concern because we reports summary statistics on school inputs and teacher absence. typically covered all the public schools in a village (in 84.2% of the cases) and had a Section 4 presents the cross-sectional and panel regression results. mean coverage rate of 82.7% of public schools in the sampled villages. 9 While the school year is not identical across states, it typically runs from mid- Section 5 discusses the fiscal costs of weak governance and compares June to mid-April. The three visits therefore spanned two academic years, with the the returns to investing in better monitoring with that from hiring first visit being made during January–March 2010, the second visit being made during more teachers. Section 6 discusses policy implications, and Section 7 June–August, and the third visit during August–October 2010. 10 concludes. This was important given the widespread possession of cell phones among teach- ers, which would allow them to call up absent colleagues on seeing external visitors in the school measuring teacher absence. 11 Not all interviews could be completed. Most non-responses were at the teacher as opposed to the school level (since absent teachers could not be interviewed, whereas school data could be obtained from either the head teacher or any other senior 7 Such misallocation in education spending is also seen in other low-income coun- teacher). These non-responses are unlikely to affect the analysis in this paper because tries. An even more striking example is provided by de Ree et al. (2015) who the panel-data analysis will focus on aggregated data at the village level as opposed to experimentally study the intensive-margin impacts of an Indonesian policy reform the individual data at the teacher level. 12 that doubled teacher pay across the board (at a similar cost of $5 billion/year) and find Field teams obtained lists of state and national school holidays in advance of that the teacher pay increase had no impact on student learning. creating the field plans and ensured that no visits were conducted on these days. K. Muralidharan et al. / Journal of Public Economics 145 (2017) 116–135 119 all part-time teachers, teachers who were transferred or reassigned Table 1 elsewhere, or teachers reportedly on a different shift. Changes in key variables between 2003 and 2010, village-level data. We construct a school infrastructure index by adding binary indi- (1) (2) (3) cators for the presence of drinking water, toilets, electricity, and a Summary statistics Difference library. We construct a remoteness index by taking the average of nine normalized indicators of distance to various amenities including Year Year (Ho: No diff) a paved road, bus station, train station, public health facility, private 2003 2010 health clinic, university, bank, post-office and Ministry of Education A. Teacher variables office. A lower score on the remoteness index represents a better Have bachelors degree 0.41 0.58 0.174*** Have teacher training 0.77 0.68 −0.085*** connected school. During each survey visit, enumerators referred to Are contract teachers 0.06 0.30 0.233*** written school records to note the date of the most recent school Are paid regularly 0.49 0.78 0.285*** inspection, and the date of the most recent parent-teacher associ- Recognition scheme exists 0.50 0.81 0.309*** ation (PTA) meeting. Average parental education of children in a school is computed from the basic demographic data collected for the B. School variables Student-teacher ratio (STR) 47.19 39.80 −7.388*** sample of fourth-grade students chosen for assessments of learning Mid-day meals 0.22 0.79 0.576*** outcomes. Infrastructure index (0–4) 2.14 3.35 1.205*** For most of the analysis in this paper, we use the village as our Has drinking water 0.80 0.96 0.160*** unit of analysis and examine mean village-level indicators of both Has toilets 0.40 0.84 0.440*** inputs and outcomes because a large number of new schools had Has electricity 0.22 0.45 0.236*** Has library 0.51 0.69 0.183*** been constructed between 2003 and 2010, including in villages that already had schools. This school construction resulted from a policy C. Monitoring and community variables designed to improve school access by ensuring that every habitation Road is within 1 km 0.69 0.78 0.092*** with over 30 school-age children had a school within a distance of Probability of inspection in last 3 months 0.38 0.56 0.176*** Probability of inspection in last 2 months 0.31 0.50 0.189*** one kilometer. Thus, to ensure that our sample was representative in Probability of inspection in last 1 month 0.22 0.38 0.155*** 2010, and at the same time amenable to panel data analysis relative Probability of PTA meeting in last 3 months 0.30 0.45 0.153*** to 2003, we constructed the panel at the village level, with a new rep- Mean parental education (1–7 scale) 2.03 2.43 0.394*** resentative sample of schools drawn in the sampled villages.13 All the State per-capita GDP (thousands of Rs.) 14.74 30.21 15.473*** results reported in this paper are population weighted and are thus D. Absence variables representative of the relevant geographic unit (i.e., state or all-India). Teacher absence rate (%) 26.29 23.64 −2.64*** Effective student-teacher ratio (ESTR) 64.02 52.13 −11.89† Source:Authors’ calculations; Central Statistical Organization, India. Notes: Summary 3. Summary statistics statistics (except Student-teacher ratio) are weighted by rural population of Socio- Cultural Regions (SCRs) in Census 2001. Student-teacher ratio is weighted by SCR 3.1. Changes in inputs school enrolment. Data for number of days since inspection and truncated at 99th per- centile. State per-capita GDP figures are in 2004–2005 prices. Absence figures for 2003 differ slightly from the figures in the Kremer et al. (2005) paper. This is because the The data show considerable improvements in school inputs urban schools are removed from the sample. between 2003 and 2010 along three broad categories - teacher qual- † We do not conduct inference on the changes in “Effective Student-Teacher Ratio” ifications and working conditions, school facilities, and monitoring because the data on total number of teachers are obtained from administrative (DISE) (Table 1 - Panels A–C). The fraction of teachers with a college degree data. *** significant at 1%, ** significant at 5%, * significant at 10%. increased from 41 to 58%, the fraction reporting that they were get- ting paid regularly rose from 49 to 78%, and the fraction reporting summary index of school infrastructure improved by 0.9 standard the existence of teacher recognition schemes rose from 50 to 81%. deviations.14 The fraction of teachers who report a formal teaching credential fell We also find improvements in both ‘top-down’ administrative from 77 to 68%, largely due to a significant increase in the hiring and ‘bottom-up’ community monitoring of schools over this period. of contract teachers (who are not required to have teaching creden- The fraction of schools inspected in the three months prior to a tials) in several large states. In our data, the fraction of teachers on a survey visit increased from 38% to 56%. The extent of community temporary contract or ‘contract teachers’ increased from 6 to 30%. oversight of schools, measured by the frequency of PTA meetings School facilities and infrastructure improved on almost every also increased: The probability that a PTA meeting took place during measure. The fraction of schools with toilets and electricity more the three months prior to a survey visit increased from 30% to 45%. than doubled (from 40% to 84% for toilets and 22% to 45% for Overall, Table 1 (Panel A–C) confirms that the Government of India’s electricity); the fraction of schools with functioning mid-day meal increased focus on primary education in the past decade did lead to programs nearly quadrupled (from 22% to 79%); the fraction of significant improvements in input-based measures of school quality, schools with a library increased by over 35 %(from 51% to 69%), as well as administrative and community monitoring. and almost all schools now have access to drinking water (96%). Initiatives outside the education ministry to increase road construc- 3.2. Changes in teacher absence tion have also led to increased proximity of schools to paved roads increasing the accessibility of schools for teachers who choose to We now turn to changes in teacher absence. Table 1 (Panel live farther away. Relative to the distribution observed in 2003, a D) shows that the population-weighted national average teacher absence rate for rural India fell from 26.3 percent% to 23.6%, a reduc- tion of 10%. Since students receive reduced teacher attention when teachers are absent, we divide the STR by “1 - teacher absence rate” 13 Even in the absence of school construction, the survey firm did not retain school 14 and teacher level identifiers from the 2003 survey (complying with data protec- We construct an index of school infrastructure by adding indicators for the exis- tion norms), which would have made it difficult to construct a school-level panel tence of four items: drinking water, toilets, electricity, and a library. Table 1 provides (especially for villages with multiple schools). summary statistics for each indicator and the overall index. 120 K. Muralidharan et al. / Journal of Public Economics 145 (2017) 116–135 to obtain the effective student teacher ratio (ESTR). Although the Table 2 all-India STR had been reduced to below 40 in this period, the effec- Teacher activity and reasons for absence (%). tive STR after accounting for teacher absence was still over 52. We (1) (2) present state-level data on teacher absence rates and ESTR for 2003 Year 2003 Year 2010 and 2010 in Table A4.15 Chaudhury et al. (2006) find a strong negative correlation A. Physical verification: Absent 26.29 23.64 School closed 6.08 6.60 between GDP/capita and teacher absence rates (both across Official teaching related duties 5.93 5.21 countries and within Indian states). Hence, one way to interpret the (trainings, meetings, etc.) magnitude of these changes is to compare them with the expected Official non-teaching duties (education, 0.95 0.93 reduction in teacher absence that may be attributed simply to the health campaigns, etc.) Official other duties (panchayat 0.31 0.29 economic growth that has taken place in this period. Using a growth meetings, political meetings, etc.) accounting (as opposed to causal) framework, we can decompose the Authorized leave 7.62 5.91 change in teacher absence into a component explained by changes No reason 5.40 4.70 in GDP/capita (as a proxy for ‘inputs’) and one explained by a change in governance (a proxy for TFP). Cross-sectional estimates from the B. Physical verification: Present 73.71 76.36 In classroom, actively teaching 42.93 53.08 2003 data suggest that a 10 percent increase in GDP/capita is asso- In classroom, passively teaching 5.56 4.16 ciated with a 0.6 percentage point reduction in teacher absence.16 In classroom, not teaching 15.88 8.96 In the period between 2002 and 2010, real GDP/capita in India Found outside classroom 9.35 10.15 had grown by 38%. Thus, growth in GDP/capita over this period should have by itself contributed to a reduction in teacher absence C. Logbook records Present today 80.93 84.06 of 2.4%. Our estimate of the change in teacher absence rate is Present last working day 89.76 exactly in this range, and suggests that the reduction of teacher Source: Authors’ calculations. Notes: All figures are weighted by SCR’s rural popu- absence we document is consistent with a proportional increase lation. In 0.37% of cases, respondents said that a log-book was not maintained in in ‘inputs’ into education, but a limited improvement in TFP in the school, 0.23% refused to show log-book. In the year 2003, logbook records for this period. We discuss the policy implications of this result in the previous working day were not collected. The full list of activities under for not teach- conclusion. ing are - doing administrative/paper work, talking to/accompanying the surveyor, chatting/talking (with teachers, others), reading magazines/newspapers, sleeping, watching TV/listening to radio, doing other personal work, idle. Reasons for school 3.3. Stated reasons for absence, teaching activity, and official records closed are - opening hours but no one has arrived yet, opening hours but everyone left, and no reason. In cases where a teacher was not found in the school, enumer- ators asked the head teacher (or senior-most teacher present) for the cases of teacher absence (these results are unchanged from the reason for absence. These stated reasons are summarized in 2003). Table 2 (Panel A). Two categories of clearly unauthorized absence In cases where the teacher was present, enumerators recorded (school closure during working hours and no valid reason for the activity that the teacher was engaged in at the point of obser- absence) account for just under half the cases of teacher absence vation: 53% of teachers on the payroll were found to be actively (48%), which provides a lower bound on the extent of unautho- teaching, and another 4% were coded as passively teaching (defined rized absences of 11.3 percentage points. The two other categories of as minding the class while students do their own work). Just stated absence (authorized leave and official duties) that account for over 19% of teachers were in school but were either not in the 52% of the observed absence are potentially legitimate but cannot be classroom or not engaged in any teaching activity while in the verified. classroom (Table 2 - Panel B). Thus a total of 42% of teachers on While head teachers may overstate the extent of official duties the payroll were either absent or not teaching at the time of direct to shield absent colleagues, they should have no reason to under- observations.17 state it. We can, therefore, reasonably treat the stated reasons Finally, enumerators also recorded whether a teacher had been for absence as an upper bound for duty-induced absence. This marked as present in the log-books on the day of the visit and also yields the important finding that one commonly cited reason for on the previous day, and we see in Table 2 - Panel C that going teacher absence - namely, that teachers are often asked to perform by these records would suggest a much lower teacher absence rate non-teaching duties such as conducting censuses and monitoring of 16% using the same day’s records, or as low as 10.2% using the elections - is a very small contributor to the high rate of observed previous day’s records (this was not collected in 2003).18 These teacher absence. Table 2 - Panel A shows that official non-teaching data highlight the importance of measuring teacher absence by duties account for less than 1% of observations and under 4% of direct physical verification as opposed to official records on log books. 15 We find large variation in teacher absence rates across states ranging from 12.9% in Tamilnadu to 45.8% in Jharkhand. Teacher absence rates declined in 14 out of 19 17 states with significant reductions in 12 states, and five states having teacher absence This is almost surely a lower-bound estimate because in many cases it is rates below 15%. However, the ESTR in 2010 in three of India’s most educationally easy for a teacher who may not have been teaching to pick up a book and look backward states (Bihar, Jharkhand, and Uttar Pradesh) was as high as 97, 79, and 69. like he or she is actively teaching when it is known that someone is visiting Thus, teacher absence can sharply increase the effective STR experienced by students the school (see Muralidharan and Sundararaman (2010) for evidence documenting relative to the STR calculated using state-level figures on enrollment and number of this). 18 teachers. Note that teachers sign the log-books when they come in and there is typically 16 The cross-sectional relationship is estimated by regressing village-level teacher no roll call where a head teacher records them as absent if they are not in school at absence on the log of district-level per-capita consumption (from the National Sam- a given time. Thus, the log-books record ‘presence’ rather than ‘absence’. This may ple Survey) in the 2003 survey. Estimates without state fixed effects are larger (and explain the higher recorded presence on the previous day than on the day of the visit, equal −1.17) whereas estimates with state fixed effects are smaller but still significant since teachers arriving late will sign themselves as present though they may not have (and equal to −0.63). Our default estimate is based on using state-fixed effects since arrived during the time the enumerators reached school. It is also not uncommon for cross-state variation in per-capita income is much more likely to be correlated with teachers to retrospectively sign log-books recording themselves as ‘present’ on days unmeasured governance quality. Tables are available on request. that they were absent. K. Muralidharan et al. / Journal of Public Economics 145 (2017) 116–135 121 4. Cross-section and panel regression results results with state and district fixed effects are least likely to be con- founded with time-invariant and time-variant omitted variables.19 4.1. Correlates of teacher absence in 2010 However, it is also worth noting that such a specification biases us against detecting small effects. First, first-differencing leaves us with Table 3 presents village-level cross-sectional regressions between less variation in the explanatory variables, which will increase stan- indicators of school quality and teacher absence in 2010. Column 1 dard errors. Second, to the extent there is measurement error in shows the mean level of each covariate in the sample, columns 2– the explanatory variables, first differencing would also increase the 4 present the coefficients on each indicator in individual regressions attenuation bias. This is why we focus our discussion and interpre- with the dependent variable being teacher absence, while columns tation of the results on the ones that are robustly significant and do 5–7 do so in multiple regressions that include all the variables shown not treat lack of evidence of significant effects as strong evidence in in Table 1 as regressors. favor of null effects. We first show the regressions with no fixed effects, then with Nevertheless, the results in Table 4 suggest that several plausible state fixed effects, and finally with district fixed effects. The com- narratives for the reasons for teacher absence seen in the cross- parison of results with and without state fixed effects is important sectional data reported in Kremer et al. (2005) are not supported for interpretation. Many indicators of school quality vary consider- in the panel data regressions. In particular, unlike in Kremer et al. ably across states in a manner that is likely to be correlated with (2005), we find no correlation between changes in school infras- other measures of governance and development as well as the his- tructure or proximity to a paved road and teacher absence. We also tory of education investments in these states. On a similar note, find no correlation between changes in teacher professional quali- while primary education policy is typically made at the state level, fications or professional conditions (such as regularity of pay) and there is often important variation across districts within a state changes in teacher absence.20 based on historical as well as geographical factors (Banerjee and We find two robust relationships in the panel regressions, where Iyer, 2005; Iyer, 2010). Thus, specifications with district fixed effects we define ‘robust’ as correlations that are significant in both individ- that are identified using only within-district variation are least likely ual and multiple regressions; significant in all three main specifica- to be confounded by omitted variables correlated with historical or tions (no fixed effects, state fixed effects, and district fixed effects) geographical factors. However, there may still be important fixed and consistent across all specifications (we cannot reject that the omitted variables across villages (such as the level of interest in estimates are the same across specifications). We discuss these two education in the community) that are correlated with both mea- results below. sured quality of schools and teachers as well as teacher absence. We therefore present the cross-sectional regressions in Table 3 for 4.2.1. Reductions in STR are correlated with increased teacher absence completeness and focus our discussion on the village-level panel First, villages that saw a reduction in student-teacher ratio (STR) regressions presented in Table 4. Overall, there are few robust cor- have significantly higher rates of teacher absence. A 10% reduction relations across all specifications except that schools that have been in STR is correlated with a 0.5% increase in average teacher absence, inspected recently have lower rates of absence. One important result and these estimates remain stable when we include state and district in the correlations is that there appears to be no significant relation- fixed effects and are unchanged when we include a full set of controls ship between teacher salary and the probability of teacher absence. (also measured in changes). Since salary data were not collected in the 2003 survey, this variable Changes in STR reflect changes in enrollment as well as in the is not included in the panel analysis below. number of teachers, and a higher STR may affect teacher absence through both enrollment and number of teachers. First, having more 4.2. Correlates of changes in teacher absence between 2003 and 2010 students enrolled may increase the cost to teachers of being absent since there are more students (and parents) who may complain. Sec- The main identification challenge in the cross-sectional regres- ond, the most common outcome for students when their teacher is sions presented in Table 3 (and in Kremer et al. (2005)) is that we absent is that they are combined with other classes/grades whose cannot rule out the possibility that the results are confounded with teachers are present.21 Thus, having more teachers in the school may village-level omitted variables. The use of panel data helps miti- make it easier for teachers to be absent (since other teachers can gate these concerns since our correlations are now identified using handle their class).22 changes in village-level measures of school inputs. Table 4 (columns These correlations should not be interpreted as causal (for 4–6) presents results from the following regression: instance, student enrolment may decline in response to increased teacher absence), but they are consistent with a causal relationship DAbsi = b0 + b1 • DTi + b2 • DSi + b3 • DMi + bZi • Zi + 4i (1) where DAbsi is the change in the mean teacher absence rate in gov- 19 Another way of interpreting the specifications is that the one with no fixed effects ernment schools in village i between 2003 and 2010, DTi is the is using all the variation in the nationwide changes over time in left and right-hand change in village-level means of measures of teacher attributes, DSi is side variables, and the ones with state and district fixed effects are estimated using the change in village-level means of measures of school facilities, and within-state and within-district variation in the changes respectively. 20 However, note that the introduction of teacher recognition schemes appears to DMi is the change in village-level means of measures of school mon- be correlated with lower teacher absence; with a significant negative correlation in itoring and supervision. Zi represents different levels of fixed effects four of six specifications (Columns 1–4) and unchanged point estimates (though not (state or district) and 4i is the error term. Since changes in the mea- significant) in the other two (Columns 5–6). 21 sures of school quality included above may be correlated, we report Doing so does not deviate from the norm in the context of rural Indian government-run primary schools because our data show that close to 80% of schools both individual regressions with only covariate at a time (columns practice multi-grade teaching (where one teacher simultaneously teaches students 1–3) as well as multiple regressions that include all of these covari- across multiple grades at the same time in the same classroom) in any case. ates (columns 4–6). 22 In further analysis, we find support for both these channels. Decomposing changes Since Eq.(1) differences away fixed unobserved heterogeneity at in STR into changes in enrollment and changes in number of teachers, we find that the village level (and therefore at the state and district level as well), the former are positively correlated with changes in teacher absence and the latter are negatively correlated (results available on request). However, we focus our discus- the inclusion of state and district fixed effects in the specification sion on the STR because the policy goals for teacher hiring are stated in terms of STR, controls for average state and district specific changes over time in and because changes in the number of teachers are highly correlated with changes in both the left-hand and right-hand side variables. Thus our panel enrollment. 122 K. Muralidharan et al. / Journal of Public Economics 145 (2017) 116–135 Table 3 Cross-section OLS regressions results, village level, 2010 data (dependent variable: teacher absence rate (%)). (1) (2) (3) (4) (5) (6) (7) Summary statistics Individual regressions Multiple regressions Year 2010 No fixed w/ State w/ District No fixed w/ State w/ District effects fixed effects fixed effects effects fixed effects fixed effects Teacher variables Have bachelors degree 0.58 −1.03 −6.20*** −7.51*** −1.96 −5.78** −6.84*** (0.32) (1.94) (2.39) (2.57) (1.76) (2.45) (2.59) Have teacher training 0.68 −11.95*** −3.48 −2.92 −2.39 −2.43 −2.09 (0.31) (2.38) (2.39) (2.73) (2.81) (2.69) (2.87) Are contract teachers 0.30 10.97*** 0.46 −1.12 −2.25 −0.27 −2.32 (0.30) (2.37) (2.48) (2.97) (2.83) (2.71) (3.21) Are paid regularly 0.78 −7.72*** −1.51 −1.24 −2.53 −1.10 −0.60 (0.39) (1.95) (1.92) (2.20) (2.00) (1.95) (2.17) Recognition scheme exists 0.81 −6.53*** −1.43 −1.72 −2.25 −0.19 −0.94 (0.37) (2.12) (1.86) (2.07) (2.08) (1.81) (2.01) Log of salary 9.25 −3.70*** −0.58 −0.30 0.43 −0.18 −0.15 (0.62) (1.08) (0.88) (0.96) (1.01) (0.94) (0.99) School variables Log student-teacher ratio 3.50 1.88 −2.31** −4.07*** −2.42** −1.65* −3.29*** (0.59) (1.26) (1.15) (1.40) (1.10) (0.99) (1.24) Mid-day meals 0.79 0.77 0.57 2.62 0.49 0.47 2.01 (0.38) (1.74) (1.80) (2.07) (1.70) (1.77) (2.03) Infrastructure index (0–4) 3.35 −3.44*** −0.23 −0.31 −0.89 0.07 0.07 (1.30) (0.56) (0.70) (0.80) (0.68) (0.69) (0.77) Remoteness index (normalized) 0.04 0.26 0.58 0.76 0.19 0.17 0.14 (0.95) (0.68) (0.59) (0.64) (0.64) (0.61) (0.65) Monitoring and community variables Probability of inspection in last 3 months 0.56 -10.47*** −7.87*** −7.63*** −6.64*** −6.32*** −6.20*** (0.29) (2.07) (2.08) (2.39) (1.90) (2.04) (2.37) Probability of PTA meeting in last 3 months 0.45 −6.72*** −2.80** -3.22** −2.59* −1.77 −2.13 (0.48) (1.51) (1.17) (1.32) (1.33) (1.13) (1.32) Mean parental education (1–7 scale) 2.43 −3.16*** 0.37 −0.46 −0.90 0.64 −0.82 (0.74) (1.00) (0.97) (1.08) (1.00) (0.95) (1.07) Log state per-capita GDP 3.29 −11.01*** −9.27*** (0.49) (1.51) (2.50) Regression statistics Constant 74.58*** (11.76) R-squared 0.139 0.231 0.394 Adjusted R-squared 0.126 0.211 0.273 F-statistic (Inspected = PTA met) 3.186* 3.450* 2.024 Number of villages 1,555 1,555 1,555 Source: Authors’ calculations. Notes: In summary statistics, standard deviations are in parentheses; in bivariate and multiple regressions, robust standard errors clustered at the district-level are in parentheses. In individual regressions (Columns 2–4), each cell is a separate regression of the row variables with the dependent variable being the change in teacher absence rate in percentage points at the village-level. In multiple regressions (Columns 5–7), each column is a single regression on all row variables. Infrastructure index variable uses availability of four items (drinking water, toilets, electricity, and library) with higher values representing better infrastructure; similarly remoteness index uses distances to nine sets of facilities, with higher values representing more remote villages. Summary statistics and regressions are weighted by SCR’s population. *** Significant at 1%, ** significant at 5%, * significant at 10%. between increased teacher hiring and increased absence of exist- visits) to one (all the schools in the village were inspected in the prior ing teachers that has been established experimentally in India three months in all of the three visits). We find that villages where (Muralidharan and Sundararaman, 2013) and other low-income the probability of inspection in the past three months increased from countries such as Kenya (Duflo et al., 2015). Our results provide zero to one had a reduction in average teacher absence of between complementary evidence and greater external validity to these 6.4 and 8.2 percentage points (a 27–35 percent reduction in teacher experimental results, and suggest that the benefits of additional absence).23 While these results are based on correlations, we present teacher hiring to reduce STR may be attenuated by increased teacher several pieces of evidence consistent with a causal effect of increased absence (in contexts with weak governance of education systems). school inspections on reduced teacher absence. First, we look at the categories of stated reasons for absence (official duty, authorized leave, and unauthorized absence), and find 4.2.2. Increasing monitoring is correlated with reduced teacher absence that increases in inspection probability are correlated only with The second robust result in the panel data estimates is the reductions in unauthorized teacher absence, but not with reduc- strong negative correlation between improved school monitoring tions in teacher absence due to either official duty or authorized and teacher absence. In each of the three visits to a school, enu- leave (Table 5). Second, we examine the extent to which changes in merators recorded the date of the most recent inspection, and we average across the three visits across all the sampled schools in the village to construct the variable “Probability of being inspected in 23 We also consider two alternative constructions: “Probability of being inspected in last 3 months”, which ranges from zero (none of the schools in the last 2 months” and “Probability of being inspected in last 1 month.” Results are similar village were inspected in the prior three months in any of the three and available upon request. K. Muralidharan et al. / Journal of Public Economics 145 (2017) 116–135 123 Table 4 Panel OLS regression results, village-level (dependent variable: percentage points change in teacher absence). (1) (2) (3) (4) (5) (6) Individual regressions Multiple regressions No fixed w/ State w/ District No fixed w/ State w/ District effects fixed effects fixed effects effects fixed effects fixed effects Changes in teacher variables Have bachelors degree −0.42 −1.69 −3.69 −1.68 −2.31 −4.71 (2.55) (2.52) (2.91) (2.51) (2.57) (3.04) Have teacher training 1.10 1.12 0.52 1.08 0.79 1.53 (2.51) (2.76) (3.12) (2.81) (2.85) (3.19) Are contract teachers −4.89 −3.39 −0.86 −5.26 −3.84 −0.83 (3.20) (3.41) (3.52) (3.37) (3.60) (4.03) Are paid regularly −0.18 −0.83 −1.47 −0.28 −0.97 −0.56 (1.70) (1.81) (2.11) (1.67) (1.77) (2.24) Recognition scheme exists −3.87** −3.34* −3.69** −3.06* −2.03 −3.34 (1.76) (1.75) (1.87) (1.71) (1.69) (2.23) Changes in school variables Log student-teacher ratio −5.33*** −4.89*** −4.48** −5.56*** −4.95*** −4.69*** (1.83) (1.68) (1.91) (1.81) (1.57) (1.78) Mid-day meals 1.31 1.81 4.19 1.62 0.95 2.14 (1.73) (2.09) (2.59) (1.73) (2.08) (2.85) Infrastructure index (0–4) −1.10* −0.97 −1.01 −0.97 −0.68 −0.96 (0.66) (0.69) (0.76) (0.66) (0.66) (0.78) Remoteness index (normalized) −1.16 −0.93 −0.55 −1.25 −1.04 −0.81 (1.05) (1.06) (1.08) (1.00) (0.95) (1.13) Changes in monitoring and community variables Probability of inspection in last 3 months −8.23*** −7.31*** −6.60*** −7.35*** −6.56*** −6.41*** (1.94) (1.98) (1.91) (1.83) (1.83) (2.01) Probability of PTA meeting in last 3 months −1.65 −3.18* −3.80** -1.71 −2.08 −2.96 (1.74) (1.63) (1.72) (1.67) (1.64) (2.02) Mean parental education (1–7 scale) −1.29 −0.09 0.48 −1.13 −0.46 0.51 (1.40) (1.38) (1.44) (1.29) (1.32) (1.46) Log state per-capita GDP −4.69 −6.18 (7.39) (7.18) Regression statistics Constant 3.43 (5.50) R-squared 0.071 0.143 0.346 Adjusted R-squared 0.054 0.115 0.188 F-statistic (Inspected = PTA met) 4.419** 2.921* 1.268 Number of villages 1,297 1,297 1,297 Source: Authors’ calculations. Notes: In summary statistics, standard deviations are in parentheses; in bivariate and multiple regressions, robust standard errors clustered at the district-level are in parentheses. In individual regressions (Columns 1–3), each cell is a separate regression of the row variables with the dependent variable being the change in teacher absence rate in percentage points at the village-level. In multiple regressions (Columns 4–6), each column is a single regression on all row variables. Infrastructure index variable uses availability of four items (drinking water, toilets, electricity, and library) with higher values representing better infrastructure; similarly remoteness index uses distances to nine sets of facilities, with higher values representing more remote villages. Regressions are weighted by SCR’s population. *** Significant at 1%, ** significant at 5%, * significant at 10%. inspection frequency can be explained by other observable factors, teacher attendance (Duflo et al., 2012). This experimental study, and find that there are no correlations between changes in inspec- however, was carried out in a small sample of informal schools in one tion frequency and changes in other observable measures of school district in India. Thus, our estimates using nationally-representative quality that are significant across our three standard specifications panel data of rural public schools across 190 districts provide com- (Table A5). Third, we use the technique developed by Altonji et al. plementary evidence that improved ‘top down’ administrative mon- (2005) to show that the ratio of unobservable to observable corre- itoring may have a substantial impact on reducing unauthorized lates of changes in teacher absence would have to be over a factor of teacher absence. 10 for these results to be completely explained by omitted variables In contrast, there is less evidence that increases in ‘bottom (Table A6). Given the very rich data we have on observable changes up’ monitoring by the community (measured by whether the PTA in school quality, and the fact that our estimates are unchanged even had met in the past 3 months) are correlated with reductions in after including state and district fixed effects, this is unlikely to be teacher absence (Table 4). This is consistent with the experimental the case.24 results reported in Olken (2007) on the impacts of monitoring Finally, these results are also consistent with experimental evi- corruption in Indonesia. These results should not be interpreted as dence from India that finds significant reduction in teacher absence suggesting that bottom-up monitoring cannot be effective, since it in response to improved monitoring and rewards linked to better is also likely that they reflect differences in the effective authority over teachers possessed by administrative superiors (high) versus parents (low). PTAs in India typically do not have authority to 24 However, since we cannot completely rule out this possibility, our policy recom- appoint or retain regular civil-service teachers, and they cannot mendations use a decision-theoretic approach to expanding school monitoring that sanction teachers for absence or non-performance (Banerjee et accounts for this uncertainty (see Section 6). al., 2010). 124 K. Muralidharan et al. / Journal of Public Economics 145 (2017) 116–135 Table 5 Correlation between inspection frequency and teacher absence by reason (panel analysis: year 2003 and year 2010 data). (1) (2) (3) (4) (5) (6) Individual regressions Multiple regressions No fixed w/ State w/ District No fixed w/ State w/ District effects fixed effects fixed effects effects fixed effects fixed effects Panel A: Change in teacher absence due to official duty Change in probability of inspection −1.77* −1.05 −1.45 −1.43 −1.00 −1.49 (0.92) (0.85) (0.97) (0.91) (0.83) (0.96) Panel B: Change in teacher absence due to authorized leave Change in probability of inspection 0.77 0.42 0.59 0.59 0.33 0.50 (0.83) (0.84) (0.91) (0.85) (0.84) (0.91) Panel C: Change in teacher absence due to unauthorized leave Change in probability of inspection −7.22*** −6.68*** −5.74*** −6.51*** −6.07*** −5.41*** (1.69) (1.86) (1.78) (1.66) (1.79) (1.75) Source: Authors’ calculations. Notes: Robust standard errors clustered at the district-level are in parenthesis. Regressions are weighted by SCR’s population. Multiple regressions include full set of controls as Table 3, coefficients not shown for brevity. *** Significant at 1%, ** significant at 5%, * significant at 10%. Inspectors and administrative superiors, on the other hand, pos- was vacant (Centre for Policy Research, 2012).26 Further, there is sess considerable authority over teachers. Their powers include the high turnover in the education administration (the average DEO had ability to demand explanations for absence, to issue verbal or written a tenure in office of just one year) creating periods when the posi- warnings, to make adverse entries in teachers’ performance records, tions are vacant during transitions. The lack of supervisory staff at the to recommend against a pay increment, to suspend a teacher, and in block-level is even more acute, as 32% of these positions were esti- extreme cases to initiate proceedings to fire a teacher (see Ministry mated to be vacant in 2010 (the year of our survey) even by an official of Education (1964-1966) for a detailed discussion of the design government report (13th JRM Monitoring Report, 2011). Our inter- of the Indian school inspection system and the powers it provides views suggest that these staffing gaps at the block and cluster level inspectors). While it is rare for teachers in India to actually get are the most important source of variation in inspection frequency fired for absence (Kremer et al., 2005), and also true that politically- within districts, since blocks and clusters without supervisory staff connected teachers can evade sanctions for absence (De and Dreze, are much less likely to get inspected. 1999; Kingdon et al., 2014), the teacher service rules include several The second source of variation in inspections is the diligence of provisions that make it possible for inspectors to significantly raise the concerned supervisory officer. Even if all the positions of super- the costs of teacher absence and thereby reduce it. A striking recent visory staff were filled, there would be variation in the zealousness example of how a motivated school inspector in India was able to with which these officers visited villages/schools, which might lead reduce teacher absence is provided by Anand (Feb. 19, 2016).25 to some areas being inspected more often than others based on In interpreting the result on school inspections, it is useful to con- whether they were in the coverage area of a more diligent officer sider why there might be variation in the frequency of inspections or not. However, since supervisors are typically assigned a coverage across villages and what this would imply for a causal interpretation. area of clusters or blocks that comprise many villages, variation in One obvious explanation is that inspectors are more likely to visit monitoring frequency that is driven by supervisor-level unobserv- more accessible villages, but the data do not support this hypothe- able characteristics is unlikely to be correlated with other village- sis since there is no correlation between changes in the remoteness level characteristics that are also correlated with absence. Of course, index and changes in inspection rates (Table A5). this source of variation has implications for thinking about the likely District-level interviews on school governance in India suggest effectiveness of hiring new supervisory staff (some of whom may be two important reasons for the variation in inspection frequency. The less diligent). We discuss these in Section 5.3. first is staffing. Districts are broken down further into administra- tive blocks, and schools within blocks are organized into clusters. 4.3. Teacher absence and student learning outcomes School supervision is typically conducted by “block education offi- cers” and “cluster resource coordinators”. We find that a significant Teacher absence reduces the effective student-teacher ratio fraction of these posts are often unfilled. For instance, in 19% of the (ESTR) for any given STR. To study the relationship between changes cases (where we have data) even the position of the “District Edu- in teacher absence between 2003 and 2010 and changes in student cation Officer (DEO)”, the senior-most education official in a district, learning outcomes in this period we first estimate: DTS = b1 D log(ESTR) + bz Controls + 4 (2) 25 In addition to the possibility of formal disciplinary action against absent teach- ers, an additional channel for the deterrence effect of increased inspections on teacher where changes in village-level mean normalized math test scores are absence may stem from the possibility that inspectors can extract side payments from regressed on changes in village-level ESTR. We find that reductions absent teachers in return for not making a formal adverse entry on their service record (World Bank, 2003). Social norms would make it difficult to ‘extort’ such payments from teachers who are actually present, but it would be much easier to demand a pay- 26 ment from an absent teacher in return for not initiating formal action. Thus, even if This module was designed to complement the school surveys by allowing us to the costs of initiating formal disciplinary action are high (and the incidence of such create quantitative measures of district-level education governance. Unfortunately, action is low), there may be other informal channels through which more frequent the non-completion rate for these interviews was very high (over 40%) due to non- inspections serve as a disincentive for teacher absence. We also test to see if increased availability, and non-response of district-level administrators. Since this non-response inspections are only correlated with reduced absence rates for contract teachers (who is clearly not random, we do not use the quantitative measures in regressions. have less job security), and find that this is not the case. Increased inspections are sig- Nevertheless, important qualitative insights can be obtained from these interview nificantly correlated with reductions in absence for both regular and contract teachers, transcripts. These results are summarized in a companion policy report (Centre for and there is no significant difference between the two. Policy Research, 2012). K. Muralidharan et al. / Journal of Public Economics 145 (2017) 116–135 125 in ESTR are significantly correlated with increased student test scores fiscal outlay (Reinikka and Svensson, 2004;Reinikka and Svensson, (Table 6 - columns 1 and 4).27 2005;Niehaus and Sukhtankar, 2013;Muralidharan et al., 2016).29 Reductions in ESTR can be achieved through reducing STR as well We follow a similar approach here by first quantifying the salary as by reducing teacher absence. Rewriting Eq. (2), we have: cost of absence as an estimate of ‘leakage’ in education spending, and then using these costs as the metric to evaluate alternate policy STR approaches to reducing ESTR. DTS = b1 D log + bz Controls + 4 (3) 1 − Absence Calculating the cost of teacher absence requires us to estimate and exclude the extent of legitimate absence from our calculations. If we relax the constraint of equal coefficients on the numerator As part of the institutional background work for this project, we and denominator in Eq. (3), we can rewrite it as: obtained teacher policy documents from several states across India. Analysis of these documents indicates that the annual allowance for DTS = b1 D log(STR) − b2 D log(1 − Absence) + bz Controls + 4 (4) personal and sick leave is 5% on average across states. This is close to the survey estimate of 5.9%(Table 2), but we use the official data While not very precise, the estimates in Table 6 (column 5) since the stated reasons may be over-reported. suggest that both reductions in log(STR), and reductions in log(1- Estimating the extent of legitimate absence due to ‘official duty’ Absence) matter equally for improved test scores. Column 6 of Table 5 (outside the school) is more difficult because there are no standard shows that once log(ESTR) is controlled for there is no independent figures for the ‘expected’ level of teacher absence for official duties. effect of teacher absence on learning outcomes, suggesting that the Policy norms prescribe minimal disruption to teachers during the main mechanism by which teacher absence affects learning out- school day and stipulate that meetings and trainings be carried out comes is through increasing the ESTR. The stronger relationship on non-school days or outside school hours. Since we are not able between teacher absence and student learning outcomes seen in to verify the claim that teachers were on official duty, and there is columns 2 and 3 (that do not include state or district fixed effects) evidence that head teachers try to cover up for teacher absences by suggests that teacher absence is likely correlated with other mea- claiming that these are due to ‘official duties’, our default estimate sures of education governance at the state and district levels, and treats half of these cases as legitimate. This gives us a base case of highlights why our preferred specifications are the ones with district legitimate absence of 8% (5% authorized leave, and 3% official duty). fixed effects. We also consider a more conservative case where the legitimate rate Our data, which are collected seven years apart and have only of absence is 10%. This 8–10% range of legitimate absence also makes mean village-level test scores, are not ideal for studying the impact sense because the fraction of teacher observations that are classified of teacher absence or other school characteristics on test scores (the as either ‘authorized leave’ or ‘official duty’ is in this range for the five ideal specifications would use annual panel data on student test states with the lowest overall absence rates - even treating the stated scores matched to these characteristics and estimate value-added reasons for absence as being fully true (tables available on request). models of student learning). But it allows us to present suggestive To estimate the cost of teacher absence, we use teacher salary evidence on the negative correlations between teacher absence and data from our surveys and use administrative (DISE) data on the student learning outcomes that are consistent with other studies number of primary school teachers by state.30 We provide three using better data that find similar results.28 The results in Table 6 estimates of the fiscal cost of teacher absence based on assuming also help illustrate that teacher absence can attenuate the benefits the rate of legitimate teacher absence to be 8, 9, and 10 percent% of reducing STR, and that reducing effective STR can be done both respectively, and these calculations suggest that the annual fiscal by reducing STR and by reducing teacher absence. We consider the cost of teacher absence is around Rs.81 –93 billion, which is around relative cost effectiveness of these approaches in the next section. US$1.4–1.6 billion/year at 2010 exchange rates (Table 7 - Panel A). 5.2. Calculating the returns to better governance in education 5. The fiscal cost of weak governance Using the results in Table 4, we calculate the returns to a marginal 5.1. The fiscal cost of teacher absence increase in the probability of a school being inspected. We make the following assumptions: (a) enough supervisory staff are hired High levels of teacher absence translate into considerable waste to increase the probability of a school being inspected in the past of public funds since teacher salaries are the largest component of 3 months by 10 percentage points (relative to a current probability education spending in most countries, including India. One way of of 56%); (b) increasing inspection probability by 10 percentage points estimating these costs is to calculate the total salary cost paid to teachers for days of work that they were expected to attend, but do not. Note that this is not a cost that would be saved if teacher absence 29 were to be reduced (since the full teacher salaries would be paid in Note that teacher absence per se does not entail an economic cost because it is either case). However, it is standard in the corruption literature to simply a transfer of resources from the tax payer to absent teachers (just like leakage is a transfer from taxpayers to corrupt officials). Thus, the economic cost of teacher measure the cost of corruption by the amount of public expenditure absence is the long-term cost of poor service delivery (such as lower long-term human that does not reach its intended goal (often referred to as ‘leakage’), capital and earnings). In practice, it is difficult to quantify these costs. Further, since and to measure the impact of interventions to reduce corruption by voters and tax-payers place intrinsic value on not wasting their money, there is con- quantifying the reduction in leakage, even if there is no reduction in siderable policy interest in reducing leakage. This is why the corruption literature has typically focused on estimating ‘leakage’ both to quantify corruption, and to use as a metric to study the impacts of interventions to reduce it. 30 Detailed state-level figures on the number of teachers and their average salaries 27 are presented in Table A7. We augment the salary figures by 10% to reflect govern- Regressions in Table 6 include controls for changes in all characteristics reported in Table 4, but those coefficients are not shown since that is not the focus of our ment contributions to pensions. This is a conservative estimate since most of the older analysis. Results are similar without the controls. cohorts of teachers are covered by a more expensive defined benefits pension plan. No 28 Duflo et al. (2012) show experimentally that lower teacher absence raises test adjustment is made for medical benefits. We use the total number of primary school scores, while Muralidharan (2012) shows this in value-added estimates with five years teachers by state because the DISE data provides only the total number of teachers of annual panel data on test scores in the state of Andhra Pradesh matched with the by state, and not the urban-rural breakdown. Moreover, Kremer et al. (2005) report absence rate of the teacher of each student that year. Das et al. (2007) show that high very similar teacher absence rates across urban and rural schools (24.8% versus 26.3%) teacher absences in Zambia lead to significantly lower student test score gains. See and so we use the 2010 state-level teacher absence rates for our calculations, with the (Muralidharan, 2013) for a review of this evidence with a focus on India. caveat that these are for rural areas. 126 K. Muralidharan et al. / Journal of Public Economics 145 (2017) 116–135 Table 6 Panel OLS regression results, village-level (dependent variable: change in normalized math score). (1) (2) (3) (4) (5) (6) Multiple regressions Multiple regressions No fixed effects District fixed effects Change in Log ESTR −0.199*** −0.097 −0.147** −0.142* (0.069) (0.083) (0.071) (0.082) Change in log STR −0.100 −0.149* (0.083) (0.083) Change in log (1-absence) 0.369*** 0.127 (0.106) (0.115) Change in absence rate −0.005** −0.000 (0.002) (0.002) Controls yes yes yes yes yes yes F-statistic and p-value: 4.35 0.03 dlogSTR = -dlog(1-Absence) (0.0381) (0.8707) R-squared 0.053 0.058 0.060 0.432 0.433 0.432 Number of villages 1149 1150 1149 1149 1150 1149 Source: Authors’ calculations. Notes: Robust standard errors clustered at the district level are in parenthesis. All regressions are weighted by SCR population. Regressions include the full set of controls as Table 3, coefficients not shown for brevity. *** Significant at 1%, ** Significant at 5%, * Significant at 10%. Table 7 Rs.448 million/year (see Table A8 for state-level calculations). How- The fiscal cost of absence (year 2010). ever, the reduction in wasted salary from this investment in terms of (1) reduced teacher absence amounts to Rs.4.5 billion/year, suggesting Panel A: Fiscal cost of absence that investing in better monitoring would lead to a reduction in ‘leak- Average monthly salary (Rs). 11,368 age’ of teacher salaries (defined as salary payments for days when Number of teachers 3,949,338 teachers do not attend work) that is around ten times greater than Total loss due to absence (Rs. millions) the cost of increasing monitoring by hiring more supervisory staff. Allowed absence (8%) 92,699 Allowed absence (9%) 86,773 Allowed absence (10%) 80,847 5.3. Input augmentation versus inefficiency reduction Panel B: Marginal returns to investing in governance To compare the relative cost effectiveness of hiring more teach- Student teacher ratio (STR) 31.7 Effective student teacher ratio (ESTR) 41.5 ers (input augmentation) versus hiring more supervisors to reduce Effect of increase inspection probability by 10 percentage points teacher absence (inefficiency reduction) as a way of reducing the Annual cost (Rs. millions) 448.0 ESTR, we calculate the salary cost of hiring more teachers to achieve Annual savings from reduced teacher absence (Rs. millions) 4509.6 the same reduction in ESTR that we estimate would be obtained by Expected effective student teacher ratio 41.1 increasing the inspection probability by 10 percentage points. We Cost to produce equal effect through teacher hiring 5742.0 estimate this to be Rs. 5.7 billion/year (Table 7 - Panel B; Table A8 Source: Authors’ calculations; DISE. Notes: All figures are in 2010 prices. Teacher provides detailed state-level calculations), and see that increasing salaries data are from Teacher Long and School Census Data. Data on number of teach- the probability of inspection would be 12.8 times more cost effec- ers, number of schools, and enrollment are from DISE State Report Cards. Simulation assumes that one inspection every 3 months reduces absence linearly by 6.4 percent- tive at reducing ESTR than doing so by hiring more teachers (on the age points. Inspector costs are assumed to be two times teacher salaries, travel costs current margin).32 are assumed to be 80% of monthly salary, and an inspector is assumed to work 200 The difference in the relative cost effectiveness of the two pol- days a year and inspect two schools every day. Detailed calculations are available in icy options is large enough that hiring more supervisors rather than appendix tables A9 and A10. teachers is likely to be a more cost effective way of reducing ESTR (on the current margin) even if the supervisors were to work less efficiently than assumed in these calculations. For instance, if super- would reduce mean teacher absence across the schools in a village by visors were absent at the same rate as teachers (say 25 %), allocating 0.64 percentage points (the most conservative estimate of the corre- marginal funds to hire an additional supervisor would still be nearly lation between increased inspection probability and reduced teacher ten times more cost effective at reducing ESTR than using those funds absence from Table 4); (c) the full cost (salary and travel) of a super- to hire an additional teacher.33 visor is 2.8 times that of a teacher; (d) a supervisor works 200 days per year and can cover 2 schools per day.31 The results of this estimation are presented in Table 7 (Panel B) 32 Note that the estimated cost of achieving a given ESTR reduction through hir- and we see that the cost of hiring enough supervisors to increase the ing more teachers is higher than the cost of achieving a proportional STR reduction, probability of a school being inspected by 10 percentage points is because our estimates suggest that reducing STR will increase the absence rates of the existing teachers (we use the most conservative estimate from Table 4 for this calcu- lation). In other words, this figure accounts for the fact that we estimate that reducing STR is correlated with increased teacher absence rates, suggesting that increased 31 We use DISE data on the number of schools in each state to calculate the num- spending on hiring teachers is correlated with an increase in inefficiency as seen from ber of supervisors who will be required to increase the probability of inspections in the discussion in Section 4.2.1. 33 a 3-month interval by 10 percentage points. The cost estimates are conservative and Note that the economic benefit to reducing teacher absence may also include a assume that the salary costs are double that of a teacher and that the travel costs are reduction in student absence. However, if we assume that any reduction in student equal to 80% of a full months’ salary (which is higher than the typical travel and daily absence in response to a lower ESTR will be the same regardless of the specific policy allowance provided to education department employees to travel to/from a village to by which the reduction in ESTR is brought about, then our assessment of the relative district headquarters). cost effectiveness of different policies to reduce ESTR will not be affected. K. Muralidharan et al. / Journal of Public Economics 145 (2017) 116–135 127 6. Policy implications Using village-level panel data, we find two robust correlations in the panel data that provide external validity in nationally- The main caveat to using our results to recommend a universal representative data to results established in smaller-scale experi- policy of hiring more supervisors to scale up the frequency of school ments. First, reductions in student-teacher ratios are strongly cor- inspections is that our estimates are based on correlations and may related with increased teacher absence, suggesting that increased not be convincing enough to warrant a universal scale up. Never- spending on hiring additional teachers was accompanied by theless, it is worth noting that both our key results - the correlation increased inefficiency, which may limit the extent to which addi- between increased monitoring and reduced teacher absence, and tional spending may improve outcomes. Second, increases in the the correlation between lower STR and increased teacher absence - frequency of inspections are strongly correlated with lower teacher are consistent with experimental evidence from smaller-scale, which absence, suggesting that of all the investments in improving school increases our confidence in their validity. Further, our estimates are quality, the one that was most effective in reducing teacher absence based on an expansion of existing system of inspections, and use was improved administrative monitoring of schools and teach- nationwide panel data (which mitigates omitted variables concerns) ers. We calculate that the fiscal cost of teacher absence is over representing close to a billion people, and complement results from $1.5 billion per year, and estimate that investing in improved gover- smaller-scale randomized experiments warranting them greater nance by increasing the frequency of monitoring would be over ten external validity for several reasons. times more cost effective at increasing student-teacher contact time First, while our results support results from smaller randomized than doing so by hiring additional teachers. experiments, there is evidence that experimentally-estimated posi- In interpreting our results, it may be useful to think of the per- tive results of interventions that are implemented by NGOs may not formance of the education system (measured by the level of teacher be replicated when the programs are implemented by governments absence) as comprising two components - ‘inputs’ into the produc- (Banerjee et al., 2008). Second, there is also evidence of site-selection tion of education that expand with income growth (such as school bias where implementing partners are more likely to be willing to infrastructure, class size, and teacher salaries), and the efficiency of rigorously evaluate programs in locations where they are more likely the use of these inputs (which would correspond to the TFP of educa- to be successful (Allcott, 2015). Finally, even in the absence of such tion production). Our results show that the Indian education system a bias, most experiments are conducted in very few sites, and may has made significant progress on the former, but made less progress yield imprecise treatment effects (for inference over a larger pop- on the latter. They also suggest that pivoting public expenditure ulation) in a setting where unobserved site-specific covariates may away from simply augmenting inputs towards policies that increase interact with the treatment (Pritchett and Sandefur, 2013).34 the efficiency of inputs may considerably increase the productivity Thus, even if small-scale experiments are unbiased within sam- of education spending, and thereby enable achievement of improved ple, they may be biased and also imprecise for population-level human capital outcomes at any given level of per-capita income. inference. In other words, there is likely to be a trade-off between One promising way of reducing inefficiency is improving school the potential omitted variable bias in our panel-data estimates on governance and achieving such a reallocation of resources would be one hand, and the advantages of greater precision, “as is” implemen- to expand the existing system of administrative monitoring of teach- tation, and unbiased site selection on the other. We do not attempt ers and schools by hiring more supervisory staff. Our calculations to quantify this trade-off in this paper since we have no objective indicate that such a marginal expansion could (on the current mar- basis of doing so. However, one way of reconciling this trade-off is gin) have a significant impact on reducing teacher absence, and that to conduct a substantial nationwide expansion of school inspections this would be highly cost effective in terms of reducing the fiscal by hiring more staff in the context of a large experimental evalu- cost of weak governance. More broadly, our results suggest that the ation. From a decision-theoretic perspective, our results are strong returns to investing in state capacity to better monitor the imple- enough to support such a policy even if there is only a 1% chance mentation of social programs in low-income countries may be quite that our estimates are causal. In Appendix B, we formally show that, high, and that at the very least there is a strong case for expand- barring extreme priors, a policy-maker interested in lowering effec- ing such programs in the context of large experimental evaluations tive student-teacher ratio will find it cost-effective to invest in or of “as is” implementation to obtain more precise estimates of their scale-up monitoring of teachers. benefits.35 7. Conclusions Acknowledgments The central and state governments in India have considerably We thank Julie Cullen, Gordon Dahl, Deon Filmer, Roger Gordon, increased spending on primary education over the past decade. We Gordon Hanson, Michael Kremer, Paul Niehaus, and Adam Wagstaff contribute towards understanding the impact of these substantial for their useful comments. We thank the Bill and Melinda Gates nationwide investments in primary education in India by construct- Foundation for financial support for the data collection and analysis ing a unique nationally-representative panel data set on education through grant number OPP59728. Additional funds for data collec- quality in rural India. We find that there has been a substantial tion were made available by the Governance Partnership Facility improvement in several measures of school quality including infras- grant provided through the Human Development Network of the tructure, student-teacher ratios, and monitoring. However, teacher absence rates continue to be high, with 23.6% of teachers in public schools across rural India being absent during unannounced visits to schools. 35 Muralidharan et al. (2016) is an example of just such an experimental evalua- tion, in the context of an ambitious initiative by the Government of Andhra Pradesh (AP) to improve governance in public welfare programs through biometric payments 34 The largest education experiments to date that we know of have been conducted infrastructure. Working with the government of AP, they randomize the rollout of over five districts in one state of India (Muralidharan and Sundararaman, 2010, 2011, the new biometric payments infrastructure over a potential universe of 20 million 2013). While these experiments feature random assignment in representative samples beneficiaries, and estimate that the program reduced ‘leakage’ in the rural employ- of schools (in a state with over 80 million people), they still come from just one state, ment guarantee scheme by an amount that was nine times the cost of the program. compared to the estimates in this paper that use panel data from 190 districts across Interestingly, this effect is of a similar magnitude to the returns that we estimate to 19 states. investing in better monitoring of teachers in this paper. 128 K. Muralidharan et al. / Journal of Public Economics 145 (2017) 116–135 World Bank. We are grateful to Pratap Bhanu Mehta and the Center provide state-wise counts of rural PSUs that were sampled in the for Policy Research, New Delhi for hosting the project and providing 2003 study. After removing PSUs in the three replaced districts alto- logistical support and infrastructure. We thank Sreela Dasgupta, gether and all other urban PSUs from the 2003 study, the maximum Anvesha Khandelwal, and L. Ravi for the project management panel size we could draw, including the REDS villages was 1,668. We support, and Monisha Ashok, Jack Liebersohn, Prerna Mukharya, sampled a 2003 village by default as long as the village had a popu- Suzanne Plant, and Anand Shukla for their outstanding research lation between 250 and 10,000 as per the 1991 Census, and we could assistance. The project would not have been possible without the locate the village in the 2001 Census.36 In districts where we had efforts of Charu Sheela, Trilok Sisodiya, AV Surya, K. Venugopal, and more than 8 rural PSUs in 2003, we sampled 8 PSUs randomly. The other staff of the Social and Rural Research Institute (SRI) in New lower cutoff on population was based on the Government of India’s Delhi who oversaw the field work and primary data collection. The mandate that all rural habitations exceeding 250 people should have findings, interpretations, and conclusions expressed in this paper are a school with 1 km. Since villages and hamlets can be absorbed into those of the authors and do not necessarily represent the views of expanding cities over time, we match the originally sampled 1991 any of the organizations that the authors are affiliated with, or the village to the villages in the 2001 Census to make sure that the view of the World Bank, its Executive Directors or the countries they sampled village still exists. represent. From the 2003 list of 1668 villages, we had to remove 249 from the 2010 sampling frame for reasons we discuss below (see Columns 5 through 9 of Table A1 for the distribution of these villages across states). 69 villages were dropped because they fall in districts that Appendix A. Sampling and construction of village-level had more than 8 villages in the 2003 round. A further 129 villages panel dataset were removed either because their population was below 250, or had far exceeded 10,000 in the 2001 Census (20,000 for Kerala). A total of The original survey in 2003 covered the 19 largest states of India 36 villages could not be located in the 2001 Census (suggesting that by population (except Delhi). Within each state, 10 districts were they had either been depopulated or absorbed into nearby towns). sampled using Probability Proportional to Size (PPS) and within Finally, 15 villages were replaced due to safety, logistical and acces- each district, 10 primary sampling units (PSUs, which could be vil- sibility reasons. Thus, our sample consists of 1419 villages from 2003 lages or towns) were sampled by PPS, thereby yielding a nationally (Table A1 - column 3). representative sample of 1,900 PSUs across 190 districts (includ- In districts where we had fewer than 8 villages in the 2003 sam- ing towns and villages). The exception is Uttar Pradesh where 11 ple (recall that the rural/urban sampling within districts was done on districts were sampled and Uttaranchal where 9 districts were sam- the basis of population ratios, and thus districts where over 25% of pled (since Uttaranchal had only 9 districts, and Uttar Pradesh is the the population in 1991 was urban would have fewer than 8 villages), largest state in India). Additionally, to account for the considerable we sample more villages as required to reach a minimum sample geographic diversity within Indian states, the sample was stratified size of 8 villages per district for the 2010 survey. The new villages by geographic socio-cultural region (SCRs), and the 10 districts in were sampled PPS from the universe of eligible villages in the 2001 each state were allocated to SCRs proportional to the population of Census that were not already sampled. The cross-section sample the SCRs. Similarly, the 10 PSUs within each district were allocated (including REDS villages) thus consists of 1,650 villages (Table A1 - to villages/towns proportional to the rural/urban population split in column 2). the district. All sampling was done on the basis of the 1991 census, Of the 1650 villages that comprise our 2010 sample, data from since that was the latest Census data available at the time of the 1555 villages were included in the analysis presented in this paper study. (Table A2 - column 2). First, we found that 29 of the 1650 villages The 2003 sample was augmented to include 241 villages from have no schools in the village. A large proportion of these villages the REDS survey (Foster and Rosenzweig, 1996). Since the REDS vil- (12 out of 29) are in Himachal Pradesh, which is a sparsely populated lages are drawn as a representative sample within districts, including mountainous state, with many small habitations. Another 39 villages these villages does not change the representativeness of the sample. did not have a public school within the village, but did have a private If a REDS district was in our main sample, the REDS villages were school. Since this paper focuses on changes in public schools, these included (typically 2 to 4 per REDS district) and additional villages villages are not included in the analysis. In Kerala, we lose another were sampled randomly to make up the total desired sample size. If 12 villages, because all schools in the village refused to be allowed a REDS district was not in our sample, those villages were covered in to be surveyed.37 Finally, we drop 15 more villages from our anal- addition to our core sample. Including these villages provides more ysis because in these villages, schools were either not functional or precise estimates of outcomes in the SCRs where they are located. closed in all three visits, which means we were unable to complete All analysis is weighted by SCR populations, so the final estimates surveys. A state-level breakdown of these 95 villages is provided continue to be nationally-representative on a population weighted in Columns 4–7 of Table A2. The decline in the cross-section sam- basis. ple size for reasons we discussed above, also reduces the number The final sample in 2003 comprised of 2141 rural and urban of villages for which we have panel data. After accounting for the PSUs across 19 states of India. In 2010, since the survey only cov- above 95 villages and 53 villages in 2003 for which we have no data ered rural areas, the sample size was reduced from 10 to 8 villages (for similar reasons as outlined for the 2010 survey round), our final per district. All districts in the 2003 sample were retained in the panel size is 1297 villages. These 1297 villages form the core of our 2010 study, with three exceptions where full-urban districts sam- analysis. pled in 2003 were replaced with a new PPS sampled district from the same SCR. The three replaced districts are Hyderabad in Andhra Pradesh, Ahmedabad in Gujarat, and Greater Bombay in Maharash- tra, which are highly urban districts containing their respective state capitals. 36 As we highlight in the paper, to meet our objective to maintain The exception to this is Kerala, which has a much higher population density, where the upper cut-off was 20,000. both representativeness of the current landscape of schools in rural 37 Permission to survey was refused in spite of the survey team possessing the India and to maximize the size of the panel, we retain villages from required permission documents. Kerala has a history of strong unions and it was not the 2003 study to the extent possible. In Column 1 of Table A1, we possible for the field teams to overcome this opposition. Table A1 Description of sample: Panel construction. (1) (2) (3) (4) (5) (6) (7) (8) (9) Number of villages Reasons for reduction in panel size K. Muralidharan et al. / Journal of Public Economics 145 (2017) 116–135 Year 2003 Year 2010 Panel Reduction in More than 8 panel Village population Village population Village not found Other reasons panel size villages in district less than 250 more than 10,000 in Census 2001 Andhra Pradesh 81 87 73 8 3 0 4 1 0 Assam 98 87 77 21 5 3 0 10 3 Bihar 94 84 84 10 10 0 0 0 0 Chattisgarh 85 80 76 9 1 0 1 2 5 Gujarat 82 88 74 8 2 2 2 0 2 Haryana 81 81 75 6 3 1 1 1 0 Himachal Pradesh 89 80 60 29 2 22 0 4 1 Jharkhand 87 84 73 14 7 4 0 1 2 Karnataka 91 89 84 7 2 3 2 0 0 Kerala 83 83 43 40 0 0 40 0 0 Madhya Pradesh 88 90 81 7 3 1 2 1 0 Maharastra 85 91 80 5 2 0 3 0 0 Orissa 92 87 79 13 4 5 1 3 0 Punjab 78 82 75 3 0 0 1 2 0 Rajasthan 91 98 85 6 1 1 0 4 0 Tamilnadu 84 87 69 15 5 0 6 4 0 Uttar Pradesh 114 113 104 10 9 1 0 0 0 Uttaranchal 80 72 57 23 6 14 1 2 0 West Bengal 85 87 70 15 4 3 5 1 2 India 1,668 1,650 1,419 249 69 60 69 36 15 Source: Authors’ calculations. Notes: The upper population cutoff for all states was 10,000 as per the 1991 census, except Kerala where the cutoff was 20,000. The category others include: replaced because high Naxalite activity (6 villages), replaced because duplicate in 2003 sample (2 villages), replaced because district was replaced (2 villages) replaced because village too remote (1 village), replaced because name missing in 2003 list (1 village), replaced because of floods in village (2 village), replaced because village could not be located (1 village). 129 130 Table A2 Description of sample: Data and attrition. (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12) K. Muralidharan et al. / Journal of Public Economics 145 (2017) 116–135 Year 2010 sample Reasons for attrition (Year 2010) Reasons for attrition (Year 2010) Reasons for attrition Sampled Included in Attrition No school No public school School(s) refused Other reasons Sampled Included in Attrition No data for No data for analysis in village in village to survey Analysis year 2010 year 2003 Andhra Pradesh 87 86 1 0 0 0 1 73 70 3 1 2 Assam 87 83 4 1 3 0 0 77 72 5 3 2 Bihar 84 81 3 1 1 0 1 84 77 7 3 4 Chattisgarh 80 75 5 2 1 0 2 76 69 7 4 3 Gujarat 88 85 3 0 3 0 0 74 71 3 3 0 Haryana 81 80 1 0 1 0 0 75 63 12 0 12 Himachal Pradesh 80 59 21 16 5 0 0 60 43 17 16 1 Jharkhand 84 81 3 2 1 0 0 73 58 15 3 12 Karnataka 89 88 1 0 1 0 0 84 82 2 1 1 Kerala 83 65 18 0 5 12 1 43 31 12 8 4 Madhya Pradesh 90 88 2 0 1 0 1 81 78 3 2 1 Maharastra 91 83 8 1 3 0 4 80 73 7 7 0 Orissa 87 83 4 2 1 0 1 79 73 6 3 3 Punjab 82 80 2 1 1 0 0 75 71 4 2 2 Rajasthan 98 94 4 1 2 0 1 85 83 2 2 0 Tamilnadu 87 79 8 1 5 0 2 69 62 7 5 2 Uttar Pradesh 113 111 2 0 2 0 0 104 100 4 2 2 Uttaranchal 72 67 5 1 3 0 1 57 52 5 4 1 West Bengal 87 87 0 0 0 0 0 70 69 1 0 1 India 1650 1555 95 29 39 12 15 1419 1297 122 69 53 Source: Authors’ calculations. Notes: The category others include: high Naxalite activity, village not reachable, schools not functional, and schools closed in all three visits. In 2003, if a village did not have any schools, surveyors went to the neighboring village. In 2010, the village was simply recorded as having no school. K. Muralidharan et al. / Journal of Public Economics 145 (2017) 116–135 131 Table A3 Description of sample: Final sample. (1) (2) (3) (4) (5) (6) (7) (8) Year 2010 sample Panel Number of Number of Number of Number of Number of Number of Number of Number of villages schools teachers villages schools in 2003 schools 2010 in 2003 teachers in 2010 Andhra 86 130 509 70 107 107 372 405 Pradesh Assam 83 150 525 72 122 134 437 473 Bihar 81 124 757 77 112 119 341 731 Chattisgarh 75 100 450 69 94 92 259 412 Gujarat 85 119 944 71 101 98 419 798 Haryana 80 105 520 63 85 83 386 395 Himachal 59 70 270 43 44 51 172 205 Pradesh Jharkhand 81 132 493 58 76 94 244 374 Karnataka 88 120 572 82 117 112 598 530 Kerala 65 105 608 31 57 50 353 307 Madhya 88 146 476 78 116 133 367 427 Pradesh Maharastra 83 98 495 73 96 88 441 451 Orissa 83 114 483 73 88 101 295 439 Punjab 80 88 469 71 75 76 355 417 Rajasthan 94 141 671 83 132 121 497 565 Tamilnadu 79 96 445 62 124 75 455 363 Uttar 111 135 616 100 131 119 442 542 Pradesh Uttaranchal 67 73 207 52 61 57 177 151 West 87 151 668 69 108 121 331 531 Bengal India 1555 2197 10,178 1297 1846 1831 6941 8516 Source: Authors’ calculations. Table A4 Absence rate of teachers & student-teacher ratios in rural public schools by state by year. (1) (2) (3) (4) (5) (6) (7) (8) (9) Absence rates(%) Student- teacher ratio Effective student-teacher ratio Year 2003 Year 2010 Change Year 2003 Year 2010 Change Year 2003 Year 2010 Change† Andhra 23.38 21.48 −1.90 27.51 25.79 −1.71 35.90 32.85 −3.05 Pradesh Assam 36.15 26.26 −9.89*** 28.21 36.07 7.86*** 44.18 48.92 4.74 Bihar 39.42 28.69 −10.73*** 72.44 69.01 −3.43 119.57 96.78 −22.79 Chattisgarh 30.47 14.20 −16.28*** 42.12 33.05 −9.07*** 60.59 38.52 −22.07 Gujarat 17.92 16.14 −1.77* 40.42 31.94 −8.48*** 49.24 38.09 −11.15 Haryana 21.07 17.75 −3.31** 34.40 36.34 1.94 43.58 44.18 0.60 Himachal 22.67 30.74 8.07*** 18.04 21.73 3.69** 23.33 31.38 8.04 Pradesh Jharkhand 43.50 45.84 2.34 52.30 42.84 −9.47*** 92.57 79.09 −13.48 Karnataka 22.60 23.93 1.33 29.07 23.62 −5.45*** 37.56 31.05 −6.51 Kerala 19.60 15.79 −3.81*** 24.84 24.49 −0.36 30.90 29.08 −1.82 Madhya 18.19 26.34 8.16*** 37.19 46.57 9.39*** 45.45 63.23 17.78 Pradesh Maharastra 15.43 14.12 −1.31 34.54 28.66 −5.88*** 40.84 33.38 −7.47 Orissa 21.69 14.24 −7.46*** 47.01 36.63 −10.38*** 60.04 42.72 −17.32 Punjab 36.66 13.54 −23.13*** 30.80 31.43 0.63 48.63 36.36 −12.28 Rajasthan 25.13 22.72 −2.42* 38.91 32.05 −6.86*** 51.97 41.47 −10.50 Tamilnadu 20.43 12.92 −7.51*** 29.56 25.85 −3.71** 37.15 29.69 −7.47 Uttar 26.72 31.21 4.49*** 69.37 47.40 −21.97*** 94.66 68.90 −25.76 Pradesh Uttaranchal 32.29 21.02 −11.27*** 24.49 31.02 6.54** 36.17 39.28 3.12 West 26.41 20.97 −5.44*** 58.23 41.61 −16.62*** 79.12 52.65 −26.47 Bengal India 26.29 23.64 −2.64*** 47.19 39.80 −7.39*** 64.02 52.13 −11.89 Source: Authors’ calculations; DISE Notes: All figures are weighted by SCR’s rural population. Absence figures for 2003 differ from the figures in the Kremer et al. (2005) paper. This is because the urban schools are removed from the sample. We do not conduct inference on the changes in “Effective student-teacher ratio” because the data on total number of teachers are obtained from administrative (DISE) data. †We do not conduct inference on the changes in “Effective student-teacher ratio” because the data on total number of teachers are obtained from administrative (DISE) data. *** Significant at 1%, ** significant at 5%, * significant at 10%. 132 K. Muralidharan et al. / Journal of Public Economics 145 (2017) 116–135 Table A5 Panel OLS regression results, village-level(dependent variable: Change in probability of inspection in past 3months). (1) (2) (3) (4) (5) (6) Individual regressions Multiple regressions No fixed w/state w/district No fixed w/state w/district effects fixed effects fixed effects effects fixed effects fixed effects Changes in teacher variables Have bachelors degree −0.003 0.042 0.039 0.006 0.037 0.030 (0.046) (0.053) (0.050) (0.046) (0.051) (0.055) Have teacher training 0.041 0.054 0.085 0.029 0.046 0.064 (0.056) (0.057) (0.054) (0.053) (0.055) (0.061) Are contract teachers 0.055 0.063 −0.040 0.108* 0.088 −0.009 (0.053) (0.073) (0.069) (0.059) (0.070) (0.082) Are paid regularly −0.036 −0.010 −0.010 −0.037 −0.005 −0.004 (0.030) (0.035) (0.035) (0.031) (0.035) (0.041) Recognition scheme exists 0.069** 0.062** 0.020 0.067** 0.060* 0.023 (0.028) (0.031) (0.032) (0.028) (0.031) (0.037) Changes in school variables Log student-teacher ratio 0.055* 0.032 0.029 0.049 0.024 0.012 (0.031) (0.032) (0.034) (0.030) (0.031) (0.037) Mid-day meals 0.007 −0.008 −0.024 0.018 −0.008 −0.017 (0.032) (0.041) (0.046) (0.034) (0.042) (0.050) Infrastructure index (0-4) 0.010 0.011 0.005 0.006 0.011 0.004 (0.012) (0.013) (0.015) (0.013) (0.013) (0.015) Remoteness index (normalized) −0.023 −0.026 −0.032 −0.024 −0.024 -0.028 (0.022) (0.022) (0.020) (0.021) (0.021) (0.024) Changes in monitoring and community variables Probability of PTA meeting in last 3 months 0.018 0.052** 0.068** 0.033 0.053** 0.070** (0.023) (0.024) (0.029) (0.023) (0.024) (0.027) Mean parental education (1–7 scale) −0.03 −0.04 −0.04** −0.04 −0.04* −0.05** (0.026) (0.026) (0.022) (0.023) (0.024) (0.025) Log state per-capita GDP −4.69 0.40** (7.392) (0.167) Regression statistics Constant −0.13 (0.138) R-squared 0.051 0.093 0.315 Adjusted R-squared 0.034 0.065 0.152 Number of villages 1300 1300 1300 Source: Authors’ calculations. Notes: Robust standard errors clustered at the district-level are in parentheses. Infrastructure index variable uses availability of four items (drinking water, toilets, electricity, and library) with higher values representing better infrastructure; similarly remoteness index uses distances to nine sets of facilities, with higher values representing more remote villages. Regressions are weighted by SCR’s population. *** Significant at 1%, ** significant at 5%, * significant at 10%. Table A6 Selection on observables and selection on unobservables (An application of Altonji, Elder and Taber (2005)). (1) (2) (3) Dependent variable: Percentage change in absence Treatment variable: Increase in inspection probability Coefficient on treatment Unconstrained coefficient Estimate of bias Implied ratio [(1)/(2)] Base specification (no fixed effects) −5.560*** −2.298 2.598 (1.551) State fixed effects −5.343*** −0.856 6.176 (1.499) District fixed effects −5.118*** −0.502 10.189 (1.765) Source: Authors’ calculations. Notes: Robust standard errors clustered at the district-level are in parenthesis. Regressions include full set of controls as Table 3, coefficients not shown for brevity. We discretize the main variable of interest - Change in probability of inspection. Villages where inspection rates increased between 2003 and 2010 are coded as 1, and 0 otherwise. 52% of villages experienced an increase in inspection, and inspection rates fell or did not change in the remaining 48%. *** Significant at 1%, ** significant at 5%, * significant at 10%. K. Muralidharan et al. / Journal of Public Economics 145 (2017) 116–135 133 Table A7 The fiscal cost of absence (year 2010). (1) (2) (3) (4) (5) Average monthly Number of Total loss due to absence (millions of Rs.) teacher salary (Rs.) teachers Allowed absence: Allowed absence: Allowed absence: 8% 9% 10% Andhra Pradesh 10,299 347,875 6374 5901 5428 Assam 9567 167,161 3855 3644 3433 Bihar 8645 336,359 7942 7559 7175 Chattisgarh 8290 155,573 1055 885 715 Gujarat 15,804 198,584 3374 2960 2546 Haryana 16,236 77,980 1630 1463 1296 Himachal Pradesh 12,199 48,507 1776 1698 1620 Jharkhand 9734 135,690 6598 6423 6249 Karnataka 10,897 195,929 4489 4207 3925 Kerala 10,751 54,976 608 529 451 Madhya Pradesh 9294 267,846 6027 5698 5370 Maharastra 17,246 288,914 4025 3367 2710 Orissa 9382 192,119 1484 1246 1008 Punjab 12,654 105,930 980 803 626 Rajasthan 14,165 271,205 7463 6956 6448 Tamilnadu 18,489 150,820 1811 1443 1075 Uttar Pradesh 10,370 491,455 15,615 14,942 14,269 Uttaranchal 17,155 45,782 1350 1246 1143 West Bengal 10,555 416,633 7527 6946 6366 India 11,368 3,949,338 92,699 86,773 80,847 Source: Authors’ calculations; DISE. Notes: 2010 teacher salaries are from Teacher Long and School Census Data. Data on total number of teachers are from DISE State Report Cards. All figures are in 2010 prices. Table A8 Marginal returns to investing in governance. (1) (2) (3) (4) (5) (6) Student-teacher ratio (2009–2010) Effect of increasing probability of inspection in past 3 Cost to produce equal effect months by 10 percentage points through teacher hiring Student-teacher ratio Effective student-teacher ratio Annual cost Annual savings from reduced Expected effective Annual cost (Rs. millions) (Rs. millions) teacher absence (Rs. millions) student-teacher ratio Andhra Pradesh 17.8 22.7 31.0 350.8 22.5 433.5 Assam 24.5 33.2 15.9 154.5 33.0 204.2 Bihar 58.2 81.6 21.2 273.6 80.8 374.9 Chattisgarh 24.5 28.5 13.9 120.1 28.3 135.0 Gujarat 29.8 35.5 19.1 291.8 35.3 336.2 Haryana 26.8 32.5 8.8 118.9 32.3 139.8 Himachal Pradesh 15.4 22.2 6.8 56.0 22.0 79.2 Jharkhand 41.3 76.2 14.8 127.9 75.3 236.3 Karnataka 23.6 31.0 18.5 201.6 30.8 257.7 Kerala 19.6 23.2 2.0 56.3 23.1 64.5 Madhya Pradesh 39.8 54.0 40.6 250.9 53.5 332.1 Maharastra 25.7 29.9 45.0 486.8 29.7 546.8 Orissa 29.4 34.3 20.5 177.5 34.1 199.7 Punjab 20.5 23.7 10.2 137.4 23.5 153.2 Rajasthan 26.2 33.9 40.0 361.6 33.6 454.5 Tamilnadu 28.3 32.5 24.6 264.9 32.3 293.2 Uttar Pradesh 40.1 58.2 58.4 489.4 57.7 697.1 Uttaranchal 20.6 26.0 10.7 73.3 25.8 90.0 West Bengal 32.3 40.8 30.1 409.4 40.5 502.5 India 31.7 41.5 448.0 4509.6 41.1 5742.0 Source: Authors’ calculations; DISE. Notes: Number of schools, number of teachers, and enrollment figures are from administrative (DISE) data. Simulation assumes that one inspection every 3 months reduces absence linearly by 6.4 percentage points. Inspector costs are assumed to be two times teacher salaries, travel costs are assumed to be 80% of monthly salary, and an inspector is assumed to work 200 days a year and inspect two schools every day. 134 K. Muralidharan et al. / Journal of Public Economics 145 (2017) 116–135 To ensure a representative sample of schools, enumerators first Table A8 suggest that the marginal cost of {1} would be $33 million conducted a full mapping of all public and private schools in each and that the marginal benefit would be $331 million (using our panel sampled village. Enumerators conducted “Participatory Resource data estimates).38 Thus, if our estimates are true, q would be around Assessments” with households at multiple locations (at least three) $300 million/year, and using a discount rate of 10%, the net present within each village to obtain a list of all primary schools within the value of moving to {1} would be $3 billion. Now suppose there is boundary of the village. All enumerated schools were administered a only a 1% chance that the causal impacts of inspections on teacher short survey that included questions on school administration such absence are as great as the panel data estimates presented here and as management (public or private), enrollment, infrastructure etc. that there is a 99% chance that the causal impacts of inspection are Enumerators also collected a list of all teachers in the school and not significantly different from zero (i.e. p = 0.01). Even then, we q their demographic characteristics. This school listing in each sam- see that (1+ r) ∗ pq is $30 million. pled village provided the frame for school sampling. We sampled up On the cost side, we conservatively estimate (using data from to three schools per village. If the village had three or fewer schools, our own field costs) that a highly-powered trial would have C{data} all schools were sampled. If the village had more than three schools, in the range of $1 million. A trial with an N of 0.06 would be a we stratified the schools by management type and randomly sam- very large trial and could cover a nationally-representative sample pled two public schools and one private school to the extent possible. across all major Indian states, but would only cost $1 million/year.39 In the event that there were only one public school and two or more Thus, even including all costs of data collection, the upper bound private schools, one government and two private schools were sam- of the costs of such a trial would be $2 million compared to a pled. Table A3 provides the state-level breakdown of the number of likely lower-bound expected benefit of $30 million.40 An expansion schools and teachers in the final (public school) sample used in this of school inspections in the context of an experimental evaluation paper (both cross section and panel). would therefore make sense even if there was only a 1% chance of the true effects being the same as our panel-data estimates. Appendix B. A decision-theoretic case for scale-ups of If we use a medical ethics perspective in this setting, we also need monitoring with an RCT to consider the costs of not providing a treatment that is known (or highly likely) to be effective. In this case, that would be the foregone Formally, consider a simple binary policy regarding the num- one-period benefit of scaling up the treatment immediately (which ber of supervisors to be hired that can take the values {0, 1}, where we estimate to be around $300 million). Thus, depending on their the current policy is {0} and {1} represents a ‘new’ policy of hiring prior beliefs, and the extent to which our panel data estimates shift enough supervisors to ensure that all schools are inspected once in these priors, some policy makers may choose to switch the policy three months. The costs of the new policy are the additional salary regime from {0} to {1} immediately. However, the point of our exer- and operational costs of hiring supervisors, and the benefits are the cise above is to show that policy makers, depending on their beliefs, reduced fiscal cost of teacher absence. Denote these by C{1} and B{1} should either implement {1} immediately or do a large expansion in respectively, and assume that it is optimal to implement the policy the context of an RCT as described above, but it would only be under if B{1} > C{1}. However, while C{1} is known, there is uncertainty an extreme set of beliefs (that there is less than a 1% chance of our around B{1} and a randomized controlled trial (RCT) in the context of panel-data estimates being truly causal) that a policy maker would a policy movement towards {1} would reduce the uncertainty around do nothing based on our results. B{1}. Suppose that after the trial, the likelihood that the optimal policy switches from {0} to {1} is p and that the expected per-period net References benefit of such a switch is q. Let cost of data collection and analysis of a trial be C{data} and the discount rate be r. Let the period of the 13th JRM Monitoring, 2011. 13th joint review mission (jrm) report of Sarva Shiksha trial be one year and the fraction of the population participating in Abhiyan. Allcott, H., 2015. Site selection bias in program evaluation. Q. J. Econ. 130, 1117–1165. the trial be N. Half of those in the trial are allocated to a treatment Altonji, J.G., Elder, T.E., Taber, C.R., 2005. Selection on observed and unobserved group and the other half to a control group. Since data collection will variables: Assessing the effectiveness of catholic schools. J. Polit. Econ. 113, be based on a representative sample of trial sites, we assume that 151–184. Anand, G., Feb. 19, 2016. Fighting Truancy among India’s Teachers, with a Pistol and a C{data} does not vary with the size of the trial. The one period cost of Stick. The New York Times. the trial is then C{data} + (N/2) ∗ C{1}. The benefits of the trial are the Bandiera, O., Prat, A., Valletti, T., 2009. Active and passive waste in government expected one-period benefit of the new policy (during the trial) and spending: evidence from a policy experiment. Am. Econ. Rev. 99, 1278–1308. Banerjee, A., Banerjee, R., Duflo, E., Glennerster, R., Khemani, S., 2010. Pitfalls of partic- the discounted benefits of switching to a new policy (in perpetuity), ipatory programs: evidence from a randomized evaluation in education in india. weighted by the probability that the trial will lead to a switch in the Am. Econ. J. Econ. Pol. 2, 1–30. policy. Thus, the trial should be conducted as long as: pq 1 C {data} + (N/2) ∗ C {1} < (N/2) ∗ B{1} + ∗ (B.1) 38 The estimates in Table A8 are based on hiring enough supervisors to increase the r 1+r probability of a school being inspected in the previous 3 months by 10 percentage points. Since the current probability of a school being inspected in the previous 3 To focus on the benefits of learning if the optimal policy should months is 56% (Table 1), we scale up the estimates in Table A8 by a factor of 4.4 since be {1} instead of {0}, we abstract away from the benefit of the policy moving to {1} would imply that the other 44% of schools should also be inspected. We during the trial period and the one-period delay in implementing the use an exchange rate of US$1 = 60 Indian Rupees. 39 India has around 600,000 villages, 44% of which would be 264,000 villages. An N of new policy (if found to be optimal), in which case the trial should be 0.06 with half the sample getting the treatment would imply that an additional 7900 conducted as long as: villages would be treated (3% of 264,000), which would be a very large trial by the standards of most experiments. Since covering all the remaining 264,000 villages is pq estimated to cost $33 million, the cost of covering 3 of the villages would be $1 million. C {data} + (N/2) ∗ C {1} < (B.2) 40 Note that we use extremely conservative estimates for p assigning only a 1% prob- r ability of true estimates as large as our panel-data based estimates and assigning the rest of the 99% probability to finding a zero effect. If we were to assign a uniform dis- Using our results to calibrate these quantities, it is straightfor- tribution of likely point estimates between zero and our panel-data estimates (this is ward to see that the expected benefits of a trial are very large also conservative because we would not assign any probability to the true estimate even under extremely conservative assumptions. The estimates in being larger than the panel-data estimate), the expected benefit would be even larger. K. Muralidharan et al. / Journal of Public Economics 145 (2017) 116–135 135 Banerjee, A., Duflo, E., Glennerster, R., 2008. Putting a band-aid on a corpse: incentives Ministry of Education, 1964–1966. Education commission, 1964 report. Government for nurses in the indian public health care system. J. Eur. Econ. Assoc. 6, 487–500. of India. Banerjee, A., Iyer, L., 2005. History, institutions, and economic performance: the legacy Muralidharan, K., 2012. Long term Effects of TeacherPerformance Pay: Experimental of colonial land tenure in india. Am. Econ. Rev. 95, 1190–1213. Evidence from India. UC San Diego. http://econweb.ucsd.edu/~kamurali/papers/ Besley, T., Persson, T., 2009. The origins of state capacity: property rights, taxation, and Working%20Papers/Long%20Term%20Effects%20of%20Teacher%20Performance politics. Am. Econ. Rev. 99, 1218–1244. %20Pay.pdf. Bloom, N., Van Reenen, J., 2010. Why do management practices differ across firms and Muralidharan, K., 2013. Priorities for primary education policy in India’s 12th five-year countries? J. Econ. Perspect. 24, 203–224. plan. India Policy Forum 9, 1–46. Centre for Policy Research, 2012. Quality of education in rural India (queri): Muralidharan, K., Niehaus, P., Sukhtankar, S., 2016. Building state capacity: evidence Governance Report. Technical Report. Centre for Policy Research, New Delhi. from biometric smartcards in India. Am. Econ. Rev. 106, 2895–2929. Chaudhury, N., Hammer, J., Kremer, M., Muralidharan, K., Rogers, F.H., 2006. Missing Muralidharan, K., Sundararaman, V., 2010. The impact of diagnostic feedback to in action: teacher and health worker absence in developing countries. J. Econ. teachers on student learning: experimental evidence from India. Econ. J. 120, Perspect. 20, 90–116. F187–F203. Das, J., Dercon, S., Habyarimana, J., Krishnan, P., 2007. Teacher shocks and student Muralidharan, K., Sundararaman, V., 2011. Teacher performance pay: experimental learning: evidence from Zambia. J. Hum. Resour. 42, 820–862. evidence from india. J. Polit. Econ. 119, 39–77. De, A., Dreze, J., 1999. Public Report on Basic Education in India (PROBE). Oxford Muralidharan, K., Sundararaman, V., 2013. Contract teachers: experimental evidence University Press. from india. Working Paper 19440. National Bureau of Economic Research. de Ree, J., Muralidharan, K., Pradhan, M., Rogers, F.H., 2015. Double for nothing: Niehaus, P., Sukhtankar, S., 2013. The marginal rate of corruption in public programs: experimental evidence on the impact of an unconditional teacher salary increase evidence from India. J. Public Econ. 104, 52–64. on student performance in Indonesia. NBER Working Paper No. 21806. Olken, B., 2007. Monitoring corruption: evidence from a field experiment in Indonesia. Dongre, A.A., Kapur, A., Tewary, V., 2014. How much does India Spend per Student on J. Polit. Econ. 115, 200–249. Elementary Education? Accountability Initiative, Centre for Policy Research, New Pratham, 2010. Annual Status of Education Report. Delhi. Pritchett, L., Sandefur, J., 2013. Context Matters for Size: Why External Validity Claims Duflo, E., Dupas, P., Kremer, M., 2015. School governance, teacher incentives, and and development practice don’t Mix. Working Paper 336, Center for Global pupil-teacher ratios: experimental evidence from Kenyan primary schools. J. Development. Public Econ. 123, 92–110. Reinikka, R., Svensson, J., 2004, May. Local capture: evidence from a central govern- Duflo, E., Hanna, R., Ryan, S., 2012. Incentives work: getting teachers to come to school. ment transfer program in uganda. Q. J. Econ. 119, 678–704. Am. Econ. Rev. 102, 1241–1278. Reinikka, R., Svensson, J., 2005, 04/05. Fighting corruption to improve schooling: Ferraz, C., Finan, F., Moreira, D.B., 2012. Corrupting learning: evidence from missing evidence from a newspaper campaign in uganda. J. Eur. Econ. Assoc. 3, 259–267. federal education funds in Brazil. J. Public Econ. 96, 712–726. UNESCO, 2014. Teaching and Learning: Achieving Quality for All. UNESCO, Paris, Foster, A.D., Rosenzweig, M.R., 1996. Technical change and human-capital returns and France. investments: evidence from the green revolution. Am. Econ. Rev. 86, 931–953. World Bank, 2003. World Development Report 2004: Making Services Work for Poor Hsieh, C.-T., Klenow, P.J., 2009. Misallocation and manufacturing tfp in China and India. People. Oxford University Press for the World Bank, Washington DC. Q. J. Econ. 124, 1403–1448. World Bank, 2010. Silent and Lethal: How Quiet Corruption Undermines Africa’s Iyer, L., 2010. Direct versus indirect colonial rule in India: long-term consequences. Development Efforts. World Bank, Washington DC. Rev. Econ. Stat. 92, 693–713. Zamboni, Y., Litschig, S., 2016. AuditRisk and Rent Extraction: Evidence from Kingdon, G.G., Little, A., Aslam, M., Rawal, S., Moe, T., Patrinos, H., Beteille, T., Banerjee, a Randomized Evaluation in Brazil. BGSE Working Paper No. 554. Uni- R., Parton, B., Sharma, S.K., 2014. A rigorous review of the political economy of versitat Pompeu Fabra, Available at: http://www3.grips.ac.jp/~s-litschig/pdfs/ education systems in developing countries. Institute of Education, University of Zamboni_and_Litschig_2016.pdf. London. Kremer, M., Muralidharan, K., Chaudhury, N., Rogers, F.H., Hammer, J., 2005. Teacher absence in India: a snapshot. J. Eur. Econ. Assoc. 3, 658–667.