Incorporating public transfers into the measurement of poverty

Transcription

1 Incorporating public transfers into the measurement of poverty Anders Kjelsrud and Rohini Somanathan July, 2013

2 The Problem Poverty measures in India, and elsewhere, are based on private consumption data from NSS-type surveys. Health and education needs are either ignored or incorporated into poverty lines in various ad-hoc ways, often using actual out-of-pocket expenses, or scaling up a subsistence basket by a fixed amount. If some communities get these through public goods, whose availability varies systematically with wealth, we have a serious measurement problem. This is the case for India, where richer communities often receive higher public transfers and better public goods. Question: How can measures of poverty and inequality incorporate public transfers and public goods? These methods will also help us assess the poverty-targeting of public spending.

3 PDS and market prices across states Table: Unit values (Rupees per kg) Rice Wheat Market PDS Market PDS Andhra Pradesh Assam Bihar Chhattisgarh Gujarat Haryana Jharkhand Karnataka Kerala Madhya Pradesh Maharashtra Orissa Punjab Rajasthan Tamil Nadu Uttar Pradesh West Bengal

4 Shares consuming any PDS rice or wheat

5 Unit values per kg.

6

7

8

9

10

11 Public schools and private educational expenses Share of villages with schools (district averages), versus the median education expenses per school going child ( tuition/fees could include other thing than schooling). Share of villages with middle school Middle schools and median educ expenses KTK MAH TN ASM AP ORS BHR CHHMP JHA RAJ GUJ UP WB KRL PUN HAR Out-of-pocket education expenses Share of villages with senior school Senior schools and median educ expenses APTN MAHGUJ KTK ASM ORS CHHMP BHR JHA RAJ UP WB KRL PUN HAR Out-of-pocket education expenses

12 Public health centers and private medical expenses Median medical expenses (institutional and non-institutional) versus share of villages with primary health centers and subcenters Share of villages with PHC GUJ PHC and median medical expenses TN KTK HARAP ASM BHR RAJ WB MAH CHHORS JHA MPUP PUN KRL Share of villages with PHS GUJ PHS and median medical expenses AP TN HAR KTK RAJ WB MAH MP ORS CHH BHR ASM JHA UP PUN KRL Out-of-pocket medical expenses Out-of-pocket medical expenses

13 Indian poverty measures: early approaches All India poverty lines: 1962: 20 and 25 rupees per capita per month for rural and urban areas, respectively, in prices. 1979: calorie norms of 2400 and 2100 calories per capita per day for the rural and urban sector expenditure equivalents of these norms identified through the empirical expenditure distribution observed in the NSS survey of resulting poverty lines were 49 rupees (rural) and 57 rupees (urban). no attempt to capture differences in prices or spending across states State-wise lines: Lakdawala EG, 1993 spatial price indices had been computed for the 1960s in two previous studies based on NSS data. these series were extended using the consumer price index for agricultural labours (CPIAL) and the consumer price index for industrial workers (CPIIW) for rural and urban areas respectively. Both indices were reweighted to reflect the consumption patterns of the poor in So while health and education expenses were implicitly included in the PLB, there was no special attention to them.

14 The Tendulkar expert group: overall approach The PDS: Treated as a price effect Lumps the PDS items with the relevant market items before computing unit values = Little effect on the unit values and the state-wise price comparisons Education and Health Education and health are two out of 23 sub price indices used to construct an overall state-wise price index Derived by looking at the median household out-of-pocket expenses in each state

15 The Tendulkar expert group education prices 1. Find the number of children in the age group of 5-15 enrolled in school in each household 2. Find the household s total expenses on tuition and stationery (not just schooling) 3. Divide the total expenditure on education by the number of school going children 4. Compute the median hh expenditure on education by state and sector 5. Compute the price as the median divided by the weighted all-india average in each sector Rural Urban Andhra Pradesh Assam Bihar Chhattisgarh Gujarat Haryana Jharkhand Karnataka Kerala Madhya Pradesh Maharashtra Orissa Punjab Rajasthan Tamil Nadu Uttar Pradesh West Bengal

16 State-wise EG prices and MPCE If all hhs face the same prices and education is a normal good, the EG prices will be higher in richer states. EG's price of education CHH BHR ORS JHA MP WB UP KTK ASM TN Rural MAH AP GUJ RAJ PUN Median MPCE KRL HAR EG's price of education BHR UP JHA MPORS CHH Urban RAJ WB TN ASM HAR KTK AP PUN KRL GUJ MAH Median MPCE

17 Our approach Collect primary data on consumer expenditure and transfers through the PDS. Impute values to public education and health facilities. Arrive at a new distribution of expenditures using these imputed values. Raise poverty lines to account for median transfers- this gives us roughly the same fraction poor Study changes the overall distribution of consumer expenditure and the spatial distribution of poverty. In this presentation, we focus on the PDS and Education.

18 Primary data Field survey conducted in Bihar in the period September-December 2012 Drew 10 districts with probabilities in proportion to population size (census 2001 figures), 5 from the northern NSS region and 5 from the NSS southern region. Sampled 4 villages at random in each district. 3 parts: Household survey: 50 randomly chosen households from each village Village survey: basic village characteristics Public facility survey: visits to the main public and private school, and to main public and private health facility.

19 Map of sample villages Gaya Patna Rohtas Jamui Purnia Banka Araria Saran Bhabua Katihar Siwan Supaul Madhubani Bhojpur Nawada Buxar Nalanda Muzaffarpur Aurangabad Vaishali Samastipur Bhagalpur Pashchim Champaran Purba Champaran Sitamarhi Darbhanga Saharsa Gopalganj Begusarai Munger Kishanganj Khagaria Madhepura Jehanabad Lakhisarai Sheohar Sheikhpura

20 Table: Access to selected facilities Share Distance* Mean Min Max (1) (2) (3) (4) Schooling Government school with grades Government school with grades Private school with grades Private school with grades High school Anganwadi centre Health Government PHC Government hospital Private clinic Private hospital Other PDS shop Bus stop Train station Commercial bank Note: * Conditional on not having the particular facility within the village.

21 Construction of poverty lines Adjust the Planning Commission poverty line for Bihar in , to Sep-Dec 2012 by the CPIAL: base is jan, 2009, 22% increase between NSS data and our survey data CPIAL Bihar

22 Poverty measures Table: Poverty and inequality measures Poverty Inequality HC PG Gini GE 1 d9/d1 (1) (2) (3) (4) (5) Arwal/Jehanabad Aurangabad Begusarai Jamui Katihar Lakhisarai Nawada Pashchim Champaran Siwan Vaishali All

23 Access and average consumption (II) Siwan Share of HHs with any PDS grain cons. Begusarai Vaishali Pashchim_Champaran Katihar Jamui Aurangabad Arwal_Jehanabad Lakhisarai Nawada Head count Siwan Begusarai Conditional average p.c. cons. Pashchim_Champaran Vaishali Katihar Jamui Lakhisarai Aurangabad Nawada Arwal_Jehanabad Head count Note: The right graph displays average consumption conditional on any consumption.

24 Imputation: Subsidized grains as income transfers 1. We compute district-wise median unit values for rice and wheat, separately for market and PDS purchases and separately for BPL and Antyodaya HHs. 2. We evaluate the household specific quantity consumed from the PDS by the local market unit value. Since the PDS prices are lower than the market prices, this raises the expenditure level of households reporting PDS consumption. Note: 97% of the hhs consuming PDS rice made rice purchases in the regular market, while 90% of the hhs consuming PDS wheat bought wheat in the regular market = indicates that it is reasonable to treat the subsidy as an income transfer.

25 Median unit values market vs. the PDS (by district) Unit values rice Unit values wheat Head count Head count Market BPL Antyodaya

26 Share of PDS purchases made by Antyodaya HHs (by district) Relatively fewer Antyodaya PDS purchases in the poorest districts Siwan Begusarai Pashchim_Champaran Vaishali Katihar Jamui Aurangabad Lakhisarai Arwal_Jehanab Nawada Head count

27 Adjusted poverty lines In all the calculation when we adjust for public facilities we also adjust the poverty line as follows: 1. Look at households ± 5 per cent of the original poverty line 2. Calculate the average imputed value for the particular public facility 3. Add this amount to the poverty line and apply this new line for all households

28 Mean per capita transfer and changes in HCs Siwan Begusarai Vaishali Pashchim_Champaran Mean p.c. transfer Katihar Jamui Aurangabad Arwal_Jehanabad Lakhisarai Nawada Head count Siwan Begusarai Pashchim_Champaran Vaishali Change in HC Katihar Jamui Lakhisarai Aurangabad Nawada Arwal_Jehanabad Head count

29 Private school rates from NSS (rural) Grade level Andhra Pradesh Assam Bihar Chhattisgarh Gujarat Haryana Jharkhand Karnataka Kerala Madhya Pradesh Maharashtra Orissa Punjab Rajasthan Tamil Nadu Uttar Pradesh West Bengal All India rural

30 Private teaching and schooling shares and median (annual) expenses Private teaching could include teaching at a coaching centre, extra teaching at the school after the regular hours (for money) or teaching from a private teacher at home. Private school rates are 0.08 for grades 1-5 and 0.04 for grades 6-8. Private schooling Private teaching outside school In public school In private school Share Tuition Share Costs Share Costs Arwal/Jehanabad Aurangabad Begusarai Jamui Katihar Lakhisarai Nawada Pashchim Champaran Siwan Vaishali All

31 Private teaching vs. private schooling Villages with low private school rates generally have a higher share of kids receiving private teaching. Private school shares Private school shares Pashchim_Champaran Nawada Aurangabad Arwal_Jehanabad Katihar Begusarai Lakhisarai Jamui Siwan Vaishali Private teaching shares Private teaching shares Note: The shares are calculated based on all kids enrolled in grades 1-8.

32 Impute values for public schooling Three steps: 1. Find all students enrolled in grade levels one to eight at a public school 2. Add the imputed value of being enrolled in the public school 3. Sum over all such students in the household and convert this amount to monthly per capita expenditure Consider three methods for imputing school values: Naive: Add the median expense on tuition+school books, separately for grade 1-5 and 6-8 (1200rs and 1800rs). Similar HHs: Use the median expenses on tuition+school books from similar households. Quality: Use the median expenses on tuition+school books from private schools of similar quality.

33 Method 2: Similar HHs The MPCE numbers are biased due to the present of public facilities. We use the share of total calories from rice and wheat as an indicator of welfare. Rice and wheat calorie share mpce bandwidth =.8

34 Method 2: Similar HHs We divide households into quartiles based on the calorie shares (highest share=quartile 1 and so on). The table below displays the median expense among those enrolled in a private school within each quartile There are to few children in group 1 and 2 for grade 6-8 for a meaningful comparison. Therefore: this is for grades 1-8 combined. Table: Private schooling expenses on tuition and school books Median No of children

35 Method 3: Schools of similar quality The government and private schools are very different in nature: the government schools are larger, have more proper buildings, more students per teacher and per classroom and are less likely to offer teaching in English. Quality is not necessarily reflected by the same set of characteristics across government and private schools. Government Private mean sd n mean sd n (1) (2) (3) (4) (5) (6) Enrolled students Attendance on day of visit Students per teacher Students per classroom No of latrines per 100 student Building made of pucca Proper floor in building Serves more than 3 midday meals a week Main teaching language English Any teaching in English (all grades) Tests in math and reading (all grades)

36 Explaining village-wise variation in median private school expenses Dependent variable: median private school expenses (tuition + school books). (1) (2) (3) (4) No of latrines per 100 student (126.6) (127.6) (128.4) (129.0) School building made of pucca (556.2) (630.7) (674.3) Proper floor in building (642.4) (645.9) Any teaching in English (all grades) (796.5) Constant (360.9) (509.2) (550.1) (959.7) Observations R

37 Impute values for public schools Use the estimated coefficients from the regression and characteristics from the public schools to predict annual values.

38 Validation I: Private shool rates Lower private school rates in villages with public schools of (estimated) good quality. Private school share Predicted public school values

39 Validation II: Household evaluation of school quality Positive correlation between the households own evaluation of the local public school and our quality measure. Average school evaluation Predicted public school values Note: How would you evaluate the follow characteristics of the government school in your village? (good, mediocre or bad). In the graph we give value 1 if good, 0 otherwise. Average over the following categories: teaching quality, teaching material and classroom, drinking water, latrines and meals.

40 Distribution across villages (deciles) Across villages Schooling (Naive) PDS Note: The graph groups villages in 10 groups based on average MPCE in each village (4 villages in each group).

41 Average distribution within villages (deciles) Average within villages Schooling (Naive) PDS Note: The graph first divides households into 10 groups within each village. It then takes the average of each group across villages.

42 Distribution across all HHs (deciles) Across all households Schooling (Naive) PDS Note:The graph groups households in 10 groups based on MPCE.

43 Summing up Public spending on the PDS is largely un-targeted. Transfers through public schools are marginally progressive: better schools are located in richer villages, but within villages the poor attend these at higher rates. Results in other states may be very different because the share of public schooling, quality and the PDS varies substantially by state. Can we use these methods to get accurate poverty rates, that account for all public transfers? Hard, because of the number of public programs. Better at detecting targeting. Doing this requires micro data on government programs matched to consumption data. With universal access to high quality public goods, these sources of measurement error go down.