• No results found

4.3 Data & Methods

4.3.5 Operationalization of Variables

From the Economic Census, the main variable of interest is the number of private schools in a district in each wave of the census. The Economic Census identifies whether an enterprise is privately or publicly owned and the industry a firm is involved in. I code all privately owned firms engaged in primary education as private primary schools. From the DISE SRC data I use the year schools were established to create a panel

DPEP Status

Non-DPEP DPEP Difference

Mean literacy pre-DPEP 40.34 31.67 -8.67

(1.04) (0.79) (1.33)

Mean literacy 2001 52.38 46.53 -5.85

(0.91) (0.73) (1.19) Change in mean literacy 1991-2001 12.04 14.86 2.82

(1.38) (1.07) (1.78)

Mean literacy 2011 61.84 55.80 -6.04

(0.75) (0.63) (0.99) Change in mean literacy 1991-2011 21.51 24.14 2.63

(1.28) (1.01) (1.66)

Notes: Standard errors in parentheses.

Table 4.6:Difference in Literacy Rates Between DPEP and Non-DPEP Districts

from Independence in 1947 to 2014 of the number of government and private schools per district. I divide the number of schools in both the economic census and DISE SRC data by the number of school-aged children in the district estimated from district-level estimates from the National Sample Survey. From the ASER data, I construct a summary index of the level that children achieve on the various elements of the survey, using a summary index designed byAnderson(2008) to aggregate related variables.

I conduct a number of analyses to overcome the limitations of each individual dataset. First, I use a difference-in-difference design using the data on school construction from the DISE SRC data. Second, I use a fuzzy regression discontinuity design using the district-level cutoff for the implementation of the Dis- trict Primary Education Programme (DPEP). Third, I conduct a time-series-cross-sectional analysis using district-years as the unit of observation in the panel.

4.3.6 Difference-in-Difference Analysis

For the difference-in-difference analysis, I use the year schools were established to create a district-year level panel of the number of schools per district. I divide districts into those that received assistance under DPEP and those that did not.55 I limit my analysis to the period between the second National Policy of Education (1986) to the introduction ofSarva Shiksha Abhiyan(SSA) in 2002 as this provides eight years of observations before and after DPEP was introduced in 1994, as well as providing a clean theoretical start and end point for analyzing the effects of DPEP. As I argue in Chapters1and2, the National Policy on Education changed the political economy of education in India where there was a greater focus on primary education, as well

SSA universalized the program across the country, which does not allow for clean identification of districts that received greater education funding.56

For the time trends in government and private school growth by DPEP and non-DPEP districts, I plot the number of government schools by the number of school-aged children in Figure4.5and the number of private schools by the number of school-aged children in4.6.57 Figure4.5confirms the original goals of DPEP: districts that received DPEP funding saw an increase in the number of government schoolsafter

DPEP and DPEP had the intended effect of nearly eliminating the government school access gap between DPEP and non-DPEP districts. There is a sharp growth in government schools post-DPEP beginning in 1996 that leads to greater government school construction for the entire DPEP period.

DPEP 0 50 100 150 Government Schools 1985 1990 1995 2000 2005 Year

DPEP District Non-DPEP District

Figure 4.5:Difference-in-Difference: Government Schools

The dashed line represents the average number of government schools per school-aged children in districts that received DPEP funding by year. The solid line represents the average number of government schools per school-aged children in districts that did not receive DPEP funding by year. The horizontal vertical line at 1994 represents the year DPEP was signed.

Moving from the growth of government schools, to private schools, DPEP also saw a “crowding-in” response from private schools. Figure4.6suggests that districts that received DPEP funding began with

56For a separate attempt to identify the effects of SSA at the sub-district level, see (Khanna,2015a). 57I provide a test of the parallel trends assumption in Appendix SectionB.1.

a lower baseline number of private schools. After receiving DPEP, however, the gap between DPEP and non-DPEP districts in the number of private schools per district narrows to nearly zero. Although the post-DPEP break is not as dramatic and slightly later, subsequent analysis shows that there was a private school response to the introduction of DPEP. The later response is to be expected as the private school response should be delayed by a couple of years after the government school response.

DPEP 0 25 50 Private Schools 1985 1990 1995 2000 2005 Year

DPEP District Non-DPEP District

Figure 4.6:Difference-in-Difference: Private Schools

The dashed line represents the average number of private schools per school-aged children in districts that received DPEP funding by year. The solid line represents the average number of private schools per school-aged children in districts that did not receive DPEP funding by year. The horizontal vertical line at 1994 represents the year DPEP was signed.

To unpack the difference-in-difference results more formally, I fit the following equation:

term that takes the value of 1 for districts that received DPEP and the observation year is after DPEP was implemented.

Our coefficient of interest isβ3that I argue should be positive for both government and private schools. With government schools, DPEP was to increase the number of schools in previously under-served districts. With respect to private schools, if my theory that private services require strong state capacity to thrive is correct, we should see a greater increase in private services in districts that received DPEP, a program specifically designed to increase state capacity.

4.3.7 Regression Discontinuity Design

I also use a fuzzy regression discontinuity (RD) design, using the literacy cutoff for DPEP eligibility as the running variable to understand the local impact of receiving DPEP funds on government and private schools in 1998 and 2005.58. The running variable for the RDD is the 1991 district-level female literacy rate that determined which districts were eligible for DPEP funds. As not all districts below the average female literacy rate were selected, and some states were allowed to include districts that were above the mean female literacy rate at their discretion, I employ a fuzzy RD design instead of a sharp RD design. The fuzzy RD design is essentially a two-stage-least-squared estimate in which the first stage is a dummy for assignment re- gressed on the literacy cutoff, and the second stage uses the predicted values from this regression to estimate the distance from the cutoff on our outcomes of interestImbens and Lemieux(2008).59.

There is little possibility that states and districts can manipulate the literacy cutoff as the literacy rates are derived from the 1991 population census conducted by the Census of India, an independent Central Government agency. Additionally, the literacy rates come from 1991, three years before the borrowing agreement for DPEP was signed, and long before the planning for DPEP began. While there was certainly discussion on the criteria for inclusion in DPEP (see for example (Ministry of Human Resource Develop- ment,1992, 37)), this was never explicitly discussed as being based on female literacy or what the precise cutoff would eventually be.

4.3.8 Time-Series-Cross-Sectional Analysis

Given that I employ a fuzzy RD design, I also use the same dataset from the fuzzy RD design to conduct a time-series-cross-sectional analysis.Imbens and Lemieux(2008) suggest that as the running variable does not perfectly predict assignment to treatment and the results from a fuzzy RD are local to the cutoff, to

58These two years correspond to the data availability in the Economic Census

use the same data to run a more general analysis on the full range of data. Here, I take the number of government schools as a predictor for the number of private schools in a district. For this I estimate the following time-series-cross-sectional equation:

Yi,t =β0+β1Government Schoolsi,t+γX0i,t+δt+θi+εi,t (4.2)

whereYi,t is the number of private schools per 10,000 school-aged children in a district, Government Schoolsi,t is the number of government primary schools in villageiat timet,X0i is a vector of controls, δt andθi are time and district fixed effects.60 The controls include controls a district level population control, district average consumption from the national sample survey, district level fertility rates, caste fractionalization,61and a lagged indicator of the number of private schools per 10,000 school-aged children.

4.4 Results

I present my results in three sections. The first stage presents results on state territoriality and shows that state territoriality has increased measured as both the reach and size of the Indian state. Second, I look at state functionality: Has the ability of the state to properly implement its policies improved? Finally, I look at the private sector response.