The Planning Commission prepares State Specific Poverty lines and Poverty levels (head count ratios) by rural and urban sectors. These poverty rates are now available for the 2000-01 NSS 55th round. Poverty rates at a more dis-aggregated level are only available at the level of NSS region for 1993-94 the previous large sample NSS survey. As the NSS does not provide district level data, poverty estimates are not available at the district level.
This paper uses an econometric methodology to provide district level NSS estimates using the region wise poverty data. These estimates are then used to rank the districts by degree of poverty (from highest to lowest). The districts falling outside one standard deviation on the upper side of the ranking can be said to be districts with relatively high poverty, while those outside two standard deviations on the upper side can be said to be those with very high poverty. The list of such districts can be used identify special pockets of poverty for which a geographical approach to economic growth, infrastructure development and poverty reduction is the most appropriate. To the extent that so many of our development programs are directed towards removing the causes and/or consequences of poverty, this can also contribute a better understanding of the micro-determinants of poverty.
This paper uses NSS region data on available district level variables such as urbanisation, labour productivity in agriculture (Value of agricultural output per male agricultural worker), proportion of SC/ST in the district population and infrastructure variables to derive estimates for the district. The district level variables are first averaged across each NSS region. Cross sectional effect of urbanisation, agricultural labour productivity, caste and infrastructure variables on poverty are then estimated. This regression is then used to derive district level poverty based on the variation of the district level variables from their average for the region.
Determinants of PovertyThe estimated equation is as follows (numbers in bracket are t statistics):
(1) Poverty = 60.224 – 0.8232 POPurban – 1.1181 Yag/Lml + 0.4648 POPsc
(13.7) (-11.0) (-10.4) (4.1)
0.3747 POPst + 0.1820 POPmuslim – 0.3843 PostalFac – 0.00003 PHsC
(7.2) (2.4) (-6.7) (-4.9)
Multiple R2 = 0.772, R2 = 0.596, R2 (adjusted) = 0.5235.
Where Yag/Lml is the Value of agricultural output per male agricultural worker (Bhalla et al) POPurban is the proportion of urban population, and POPsc POPst & POPmuslim are respectively the SC, ST & muslim populations as a proportion of the total (1991 census) averaged over the NSS region. PostalFac and PHsC are the proportion of villages with postal facility and Primary Health sub-Centers respectively. Poverty rates are for 1993-94.
The equation shows that agricultural productivity and urbanisation play a strong role in reducing poverty. The former is the key to productive employment in rural areas and the latter plays both a direct role in generating urban jobs and an indirect role in providing the market for agricultural produce. The association of SC and ST population with poverty is also very strong. A one percent point increase in the SC and ST population increases the poverty rate by 0.47 percent point and 0.37 percent point respectively. The effect of Muslim population is smaller (0.18 percent point) though still significant at the 5% level. The effect of unemployment rate on poverty which was significant in the absence of the urbanisation variable is now insignificant.
It has been known for some time that ill health is an important factor in pushing people below the poverty line. Our results show that the existence of Primary health sub-centres in the rural areas has a significant affect on poverty reduction. This is rather remarkable given the absenteeism and poor quality of health care provided in government health centres and the large proportion of consumer expenditures on private health practitioners and facilities. There is generally one Primary health sub-centre per 5000 people. These sub-centers are staffed by an Auxiliary Nurse mid-wife and a male health worker and there primary objective is preventive health. It is known that the attendance record of the latter is not much better than that of doctors posted at Primary health centres. We understand, however that the attendance record of the Auxiliary Nurse Mid-wife (ANM) is the polar opposite, bordering on the perfect. This is perhaps an explanation for the effectiveness in reducing ill health and and poverty.
The only other variable with a significant effect on cross-regional poverty are postal facilities. Postal facilities can be channel for communications between market participants or for the flow of information and money orders from migrant labour. Different aspects could be important in different situations.
Other infrastructure variables like proportions of villages with electricity, pucca roads, drinking water, or primary schools, the density of railway station or bank branches and gross irrigated area have no additional effect on cross-regional poverty (i.e. are not significant when entered in this equation). An important reason for this is the high degree of multi- co linearity among many of these variable with the variables that turn out to be significant.
Using the same regional data we can also find out the effect of poverty and the available district level variables on un-employment.
(2) Yag/Lml = -6.7528 + 0.0472 UrbanPop + 0.1178 GrIrArea + 0.0977 POPsc
(-5.9) (3.2) (10.8) (4.4)
+0.0839 POPst + 1.4074 BankBrnch + 0.026 ElectVil – 0.0324 DrnkWtr
(4.2) (14.8) (2.3) (-4.8)
Multiple R2 = 0.791, R2 = 0.626, R2 (adjusted) = 0.619.
Numbers in brackets are t statistics. GrIrArea is Gross Irrigated area, BankBrnch is the number of bank branches per lakh population, ElectVil, DrnkWtr and PuccRd are the proportion of villages that are electrified, have drinking water supply and have Pucca road connection respectively.
Some answers are provided by an examination of the determinants of differences in productivity across regions. Equation (2) shows that urbanisation, gross irrigated area, density of bank branches and electrification have a significant effect on agricultural productivity. The slightly lower level of significance of the last variable is probably due to the fact that the mere existence of electricity connections does not reflect either the amount or quality of electricity supply, and when absent it is probably substituted by use of hydro-carbon fuels. Better irrigation and electricity therefore influence poverty indirectly through their effect on agricultural productivity.
Pucca roads however turn out have no effect on agricultural productivity or poverty. It is possible that the quality of ‘pucca’ roads is not much different from that of ‘kutcha’ ones, while the absence of critical bridges etc. means that this variable is not reflective of good road connectivity. The connectivity provided by roads is probably captured jointly by the agricultural productivity and urbanisation variables, in that surplus agricultural produce arising from high labour productivity has to be transported to and sold in urban areas. The fact that density of bank branches is significant in the agricultural productivity regression but not in the poverty one, reflects the importance of payment mechanisms and perhaps credit in the supply of surplus agricultural produce to urban areas.
The only plausible explanation for the negative effect of drinking water supply on agriculture is that it is capturing the effect of low natural water supply. In other words for any given level of irrigation the lower the natural rainfall the greater is the need for piped drinking water supply. If provision of drinking water to villages has been need based, then this variable will be a good proxy for low rainfall (i.e. inverse correlation). We would expect that rainfall has a positive effect on agricultural productivity and consequently that water supply proportion is negatively co-related with agricultural productivity.
District Poverty Rates
Using the co-efficients estimated in equation (1) we can estimate district level poverty rates for all districts within a given region as follows:
(3) Poverty (district) = Poverty (region) – 1.1181 [Yag/Lml (district) – Yag/Lml (region)] - 0.8232 [POPurban (district) – POPurban (region)] + 0.4648 [POPsc (district) – POPsc (region)] + 0.3747 [POPst (district) – POPst (region)] + 0.182 [POPmuslim (district) – POPmuslim (region)] – 0.3843 [PostalFac (district) – PostalFac (region)] – 0.00003 [PHsC (district) – PHsC (region)]
The results of applying equation (3) to derive the district level poverty estimates for 1993-94 are presented in Table 1. The districts are ordered starting from those with the highest poverty rate to the lowest poverty rate. The mean poverty rate in these districts is 37.3% and the standard deviation is 19.8%. Poverty rates falling within one standard deviation of the mean are within the normal range of variability, and only those lying outside this range on the upper side can be classified as being abnormally high. There are 66 districts with a poverty rate higher than 57.2% (mean plus standard deviation). These can be classified as having a high poverty rate. Of these 66 high poverty districts almost half were in (undivided) Bihar (32). Currently 19 (about 28%) of these will be in Bihar and 13 (20%) in Jharkhand. Undivided Madhya Pradesh had 13 (20%) of the high poverty districts of which 9 fall in current Madhya Pradesh and 5 in Chattisgarh. The remaining high poverty districts are in Uttar Pradesh (9), Orissa (5), Maharashtra (4), West Bengal (2) and Tamil Nadu (1).
There were only four district with a poverty rate higher than two standard diviation above the mean (i.e. 77% in 1993-94). These are Gumla and Lhardaga in Jharkhand and West Nimar and Barwani in Madhya Pradesh.
Bihar (undivided) had high poverty in 58% of its 55 districts. These were situated in the northern, central and southern (Jharkhand) NSS regions. Orissa came next with high poverty in 38% of its 13 districts with the highest concentration in the Southern and Western regions. Madhya Pradesh (udivided) had high poverty in 20% of its 34 districts with the highest concentration in the South Western and Chattisgarh regions. Maharashtra with 12% of its 34 districts, UP with 11% of its 79 districts and West Bengal with also 11% of its 18 districts were in the middle. Tamil Nadu’s single high poverty district constituted 5% of its 18 districts. UP’s high poverty districts were concentrated in the Southern region with a few in the Eastern region , while the high poverty districts of West Bengal were in the Himalayan region. A regional approach could be adopted in most of these states with contiguous high poverty districts developed together in an integrated state sub-plan (s).
We can compare these proportions with the 1993-94 State Poverty rates. The ordering of the States by State poverty level is identical to the ordering of states by the proportion of their districts that have high poverty levels as defined above. Thus there is perfect rank co-relation between the two.
The paper shows that agricultural productivity and urbanisation have a very significant role in reducing poverty. Districts with higher SC, ST and Muslim population have a larger proportion of poor.