ASSESSING INCOME DISTRIBUTION FOR INDIA AT S ON O N A A THE DISTRICT LEVEL USING THE DISTRICT LEVEL USING NIGHTTIME SATELLITE IMAGERY Mayuri Chaturvedi, Tilottama Ghosh, Laveesh Bhandari Indicus Analytics Private Limited New Delhi India New Delhi, India 32 nd APAN Meeting
Overview Overview � Introduction: Why Income Distribution and Night Lights I t d ti Wh I Di t ib ti d Ni ht Li ht � Literature review: Potential of Nighttime Lights � Research Objective � Research Objective � Method � Data Used � State-level graphical analysis � Developing the Model � Results R l � Model estimates and diagnostics � Error maps � Error maps � Discussion � Conclusion
Why Income Distribution and Nightlights? Why Income Distribution and Nightlights? � Inclusive growth one of the major policy thrust areas in I l i th f th j li th t i the current as well as next Five-Year Plan � Income distribution data not easy to come by � Income distribution data not easy to come by � Limitations include: � Under-reporting, Over-reporting, Misreporting � Under reporting, Over reporting, Misreporting � Inappropriate sampling and/or weighting � Lack of standardized methodology across sampling organizations i i � Huge time lags between collection and publication, and low frequency of data collection q y � Coarse spatial resolution, Modifiable Areal Unit Problem � Nightlights (NL) can help circumvent these problems
Potential of Nighttime Lights Potential of Nighttime Lights • Imhoff et al 1997 • Imhoff et al 1997 Mapping urban extent and its • C. Small 2001 ecological impact • Milesi et al 2003 • P.C Sutton 1997 Estimating urban populations and Estimating urban populations and • Sutton et al 2001 intra-urban population density • Sutton et al 2003 Estimating impervious surface • Elvidge et al 2004 g area • Doll et al 2000 Mapping Green-House Gas • Oda and Maksyutov 2010 emissions • Ghosh et al 2010 Ghosh et al 2010 Mapping fire and fire-prone • Cova et al 2004 areas Estimating and mapping GDP • Ebener et al 2005 at the national and sub- • Doll et al 2006 national levels • Sutton et al 2007 • Elvidge et al 2009 El id l 2009 Creating global grids of GDP • Ghosh et al 2010 and poverty
Research Objective Research Objective In this paper, we take a look at the relationship between night lights and Income distribution, as captured by the number of households in different b f h h ld i diff t income brackets � Use multinomial regression � Use multinomial regression techniques to study the statistical relationship � Map the prediction errors to � Map the prediction errors to identify regions of maximum estimation errors � Use socio-economic insights to � Use socio economic insights to understand probable reasons behind the errors Radiance-calibrated nighttime image of India, 2004
Method I: Data used… Method I: Data used… Radiance-calibrated Nighttime Image of India, 2004 (NOAA NGDC) 2004 (NOAA, NGDC) Households Income Data, 2004 (Indicus Analytics) 2004 (Indicus Analytics) States and District Shapefiles for India Shapefiles for India (Indicus Analytics)
Method I: Data used Method I: Data used � Three categories of households Th t i f h h ld defined on the basis of annual Upper Income household income Households � Upper income households (earning U i h h ld ( i more than Rs 10 lakh per annum, or US $22,500) Middle Income Households Households � Middle income households (earning Rs � Middle income households (earning Rs 3-10 lakh per annum, or US $6,700 – US $22,500) Lower Income � Lower income households (earning less � Lower income households (earning less Households than Rs 3 lakh per annum, or US $6,700) � Sum of lights extracted for the States � Sum of lights extracted for the States and the Districts � Sum of lights and number of households in each income category households in each income category graphed at the State level
Method II: State-level graphical analysis g p y A look at the Sum of Lights and Households’ Income data
Method II: State-level graphical analysis Inferences � Nightlights inadequately capture household income data across Indian states � Some examples that highlight the need to conduct a finer spatial resolution analysis p y � Maharashtra and Andhra Pradesh (similar lights, dissimilar incomes) d ss a co es) � Madhya Pradesh and Rajasthan (similar incomes, dissimilar lights) dissimilar lights) � Uttar Pradesh in the graph and in the NL Image (variegated lighting pattern) (variegated lighting pattern) � Complex role of population highlighted
Method III: Developing the Model The relationship between Night Lights and Households data suggested a The relationship between Night Lights and Households data suggested a logarithmic model… Scatter plot for 585 districts Scatter plot for 585 districts • X- Sum of lights • X- Natural log of sum of lights • Y- No. of households in Upper • Y- Natural log of no. of households income group in Upper income group
Method III: Developing the Model …and the use of dummy variables for commercially and administratively …and the use of dummy variables for commercially and administratively important districts that are also high population zones 1.Metropolitan Districts 2.Suburbs of Delhi Metros M Mumbai b i Noida 3.Large Kolkata Thane Industrialized Chennai Towns Gurgaon Pune Agra Faridabad 4 State 4.State Bangalore Jamnagar Capitals Nagpur Ahmadabad Kanpur Hugli Patna Hyderabad Rangareddi Khordha Surat… Other Other Cuttack Mumbai Mumbai Districts Krishna… Jaipur… All the remaining…
Method III: Developing the Model Hypotheses Proposed… N Number of households f Contribution to Nightlights N Upper pp Upper Income Upper Income Income Middle Middle Income Income Income Income Lower Lower Income Income While we can have data on households in different income brackets, we can obtain � information only on total sum of lights in a region Hypothesis One: NL should be more closely associated with the richer in any given region � than with the poorer Hypothesis Two: NL will most likely tend to under-estimate the number of poor households � and over estimate the rich households and over-estimate the rich households Logarithmic multivariate regression model used for all three income categories using the � same predictor variables
Results I: Model Estimates and Diagnostics Model Co-efficients Ln Y = α + β 1 (Ln X 1 ) + β 2 X 2 + β 3 X 3 + β 4 X 4 + β 5 X 5 Natural log of number of Natural log of number of Natural log of number of Predictor Variables households in the 'lower' households in the households in the income group 'middle' income group 'upper' income group 0.45* 0.51* 0.55* Natural log of Sum of Lights, β 1 1.01* 1.60* 2.11* Dummy for Metro cities, β 2 0.37 $ 1.35* 1.67* Dummy for Suburbs of Metros, β 3 Dummy for Suburbs of Metros, β 3 0.18 # 0.55* 0.72* Dummy for Large Towns, β 4 ‐ 0.30 $ 0.46* 0.76* Dummy for Capital cities, β 5 8.08* 8 08* 4.26* 4 26* 1 86* 1.86* Intercept, α I t t 0.61 0.75 0.76 Adjusted R 2 * Significant at the 99% Confidence Interval, $ Significant at the 95% Confidence Interval, # Significant at the 90% Confidence Interval
Results I : Model Estimates and Diagnostics Inference � Tightening of relationship between NL and households’ Ti ht i f l ti hi b t NL d h h ld ’ categories as the income goes up as seen in higher adjusted R 2 values for middle and upper income category models � Magnitude of the co-efficient for NL ( β 1) increases for the more affluent segments � Most of the predictor variables significant at the 99% level of � Most of the predictor variables significant at the 99% level of significance � Co-efficients of all dummy variables also go up monotonically for higher income group � Lights are better able to estimate households in more affluent categories (Hypothesis One) categories (Hypothesis One) � Β ’s consistently highest for the Metropolitan dummy followed by dummy for Suburbs of Metros for all three models
Results I : Model Estimates and Diagnostics Scatter graphs for the three income category households with Nightlights
Results II: Error Maps From a predominant pattern of under-estimation to an emerging pattern of From a predominant pattern of under estimation to an emerging pattern of over-estimation
Results II: Error Maps Results II: Error Maps � Lower Income Households L I H h ld � Under-estimation among states: Bihar, parts of UP , Jharkhand, WB, Orissa, Assam, AP and Kerala � Over-estimation among states: Punjab, Haryana, HP , parts of Gujarat, Rajasthan, some N-E states’ districts � Middle Income Households � Under-estimation among states: UP , Bihar, Jharkhand, WB, Kerala and some NE districts � Over-estimation among states: Changlang (Arunachal), Udhampur (J&K), Senapati (Manipur), Yanam (Pondicherry) � Upper Income Households � Under-estimation among states: Kerala, parts of UP U de es a o a o g s a es: e a a, pa s o U , u jab, , Punjab, HP , Bihar, , a , Jharkhand, WB, Assam, Nagaland � Over-estimation among states: Gujarat, Rajasthan, Karnataka
Recommend
More recommend
