ISSN (Print) - 0012-9976 | ISSN (Online) - 2349-8846
Reader Mode
-A A +A

Burden of Child Malnutrition in India

A View from Parliamentary Constituencies

Akshay Swaminathan ( is at the Department of Statistics, Harvard University, Cambridge.Rockli Kim ( is at the Department of Social and Behavioral Sciences, Harvard TH Chan School of Public Health, Boston. Yun Xu ( and S V Subramanian ( are at the Harvard Center for Population and Development Studies, Cambridge. Jeffrey C Blossom ( is at the Center for Geographic Analysis, Harvard University, Cambridge. William Joe ( is at the Population Research Centre, Institute of Economic Growth, Delhi. R Venkataramanan ( is with the Tata Trusts, Mumbai. Alok Kumar ( is with the NITI Aayog, Government of India, New Delhi.

In India, monitoring and surveillance of health and well-being indicators have been focused primarily on the state and district levels. Analysing population data at the level of parliamentary constituencies has the potential to bring political accountability to the data-driven policy discourse that is currently based on district-level estimates. Using data from the fourth National Family Health Survey 2016, two geographic information systems methodologies have been developed and applied to provide estimates of four child malnutrition indicators (stunting, underweight, wasting, and anemia) for the 543 parliamentary constituencies in India. The results indicate that several constituencies experience a multiple burden of child malnutrition that must be addressed concurrently and as a priority.

(Appendix Figures A1–A6 and Tables A1–A4 accompanying this article are available on the EPW website.)

Alok Kumar and R Venkataramanan would like to state that interpretations made in this article do not refl ect the views of their respective affi liated institutions.

This research was supported by the Harvard Center for Population and Development Studies, and the Tata Trusts.

In India, monitoring and surveillance of health and well-being indicators have largely focused on states, and increasingly on districts. For instance, the National Institution for Transforming India (NITI) Aayog, the premier think tank of the Government of India (GoI), continues to release financial, labour, education, health, infrastructure, and poverty statistics only at the state and district levels. Since many of the most prominent sources of population data now report district-level outcomes, districts have become the unit of interest and intervention in policy discourse. For instance, the fourth National Family Health Survey 2016 (NFHS-4), conducted across India, released district-level data on a variety of health, nutrition and population indicators. The NITI Aayog Aspirational Districts Programme, which aims to prioritise 115 districts across India for intervention—identified due to their lagging development indicators—is one example of an increased focus on districts in development policy (NITI Aayog 2018). An increased policy focus on districts creates a virtuous cycle, which in turn necessitates even more data collection at the district level, resulting in discourse that is evidence-based.

Another decentralised geographical unit of substantial political influence in India is the parliamentary constituencies (PCs): the 543 geographical regions represented by the members of Parliament (MPs) of the Lok Sabha. The MPs of the Lok Sabha are elected by first-past-the-post universal adult suffrage and serve five-year terms in the lower house of the Parliament (Parliament of India 2017). Unlike districts that do not have any direct governmental accountability, a focus on PCs can bring a greater degree of accountability to policy vision and implementation since the MPs are directly responsible for the well-being of their constituents. Whereas health and development indicators and other population data are widely available at the district level, there is a notable scarcity of PC-level data. This lack of PC-level data steers policy discourse away from PCs, which in turn discourages data collection at the PC level.

PCs are relevant not only due to their direct representation of people, but also because of the resources allocated to them by the national government. In 1993, the GoI established the Members of Parliament Local Area Development Scheme (MPLADS), wherein each year every MP may receive up to ₹ 5 crore to carry out development projects in their respective PCs (MoSPI 2017a). To date, ₹ 31,833.35 crore has been disbursed to Lok Sabha MPs by the GoI as part of the MPLADS programme (MoSPI 2017b).

Determining the most appropriate interventions to fund requires understanding the local context. For example, an MP seeking to reduce the prevalence of stunting in their PC, first and foremost, requires current, accurate stunting prevalence estimates for their PC. They also require PC-specific economic, infrastructural, and demographic data in order to carry out a successful intervention to prevent stunting.

Several efforts have been recently established to aid MPs in understanding their constituents’ needs. In early 2016, for instance, a select group of MPs, along with the Swaniti Initiative and Tata Trusts, launched an initiative, “Supporting Parliamentarians in Analysis and Research in the Constituency” (SPARC), which assigns 20 young professionals to 20 MPs across India to aid in implementing PC-specific development projects (Swaniti Initiative 2017). The SPARC programme arose when a group of MPs “came together to brainstorm ways through which Parliamentarians can become more effective leaders” (Swaniti Initiative 2017). Similarly, the Parliamentary Research Service (PRS), established in 2011, is a think tank and resource base that provides PC-specific data and research to support MPs, but with a focus on legislative matters (PRS Legislative Research 2018). The PRS publishes legislation analyses, statistical reports, research notes, and has also established the PRS Legislative Assistants to Members of Parliament (LAMP) fellowship, which, similar to SPARC, pairs young professionals with MPs to aid in research and policymaking. The combination of federal funding programmes for PC development and supporting resources for MPs makes the PC an important focus for population data analysis.

In order to address the data gap for health and development indicators at the PC-level, we developed and applied two novel methodologies to generate estimates of child malnutrition at the PC level. Specifically, using the NFHS-4 data on indicators of child malnutrition, we do three things. First, we present a state-of-the-art geographic information system (GIS)-based methodology to use district-level estimates and create a “crosswalk” to generate PC-level estimates. Second, we present a method of generating PC-level estimates by directly aggregating individual data in instances where one can link individual data to their PCs. Third, we apply these methodologies to rank PCs on indicators of child malnutrition (that is, stunting, underweight, wasting, and anaemia) and assess the patterns of PC variability across these indicators. We exemplify our methods using indicators of child malnutrition in order to provide timely evidence to inform current discussion on POSHAN Abhiyaan, a three-year programme launched in March 2018 that aims to “reduce the level of stunting, undernutrition, anaemia and low birth weight babies” (PIB 2018).


We use three main sources of data for our analysis. The first is the district-level fact sheet and individual data on child malnutrition indicators from the NFHS-4. The NFHS-4 is a national survey conducted by the Ministry of Health and Family Welfare, and has collected information from 6,01,509 households on socio-demographic characteristics, water and sanitation, child health, women’s and men’s health, and other health-related variables (IIPS 2016). This survey also comes under the larger framework of the global demographic and health surveys (DHS) that are conducted across a wide range of low- and middle-income countries about every five years (DHS 2018a). Although the individual data is available for NFHS-4, using the district-level aggregate data is relevant because one of our methodologies to generate PC-level estimates only requires data at the district level. In our case, individual data is not necessary, thus making this approach widely applicable to many data sets where district-level estimates are available, but individual data are not.

The second data source is the geographic data provided by the DHS, where sampling clusters—from which households are sampled—are geo-referenced by latitude and longitude coordinates and available via special request (Burgert et al 2013). These survey cluster coordinates are mostly collected in the field using global positioning system (GPS) receivers, which are accurate to +/- 15 metres. The GPS positions are then displaced randomly to maintain respondent confidentiality. The displacement distance is a maximum of 2 kilometres (km) for urban clusters and 5 km for rural clusters. The displacement is restricted so that the displaced cluster locations stay within the districts. For clusters without GPS readings, coordinates are extracted from a paper map or a gazetteer of settlement names, or from pre-existing census data provided by the country’s census agency/ministry. We used the geographic data collected in 2015–16, which has a total of 28,526 clusters in India (DHS 2018b).

The third data source was the boundary shapefiles for PCs and districts. The “India–Map of Parliamentary Constituencies, 2014” GIS shapefile was downloaded and used as the PC boundaries for this project (Github 2014a). This data set has boundaries mapped for 543 PCs in polygon format. The “India–District Map” GIS shapefile was downloaded from GitHub, and contains 641 mapped district boundaries in polygon format, and will be referred hereafter as “Districts” (GitHub 2014b). Note that this shapefile reflects the district boundaries from the 2011 Census of India, and was published on 6 April 2016.

We considered four indicators of child malnutrition (Subramanian et al 2016; Corsi et al 2016; Balarajan et al 2011). These four indicators were: (i) the percentage of stunting for children under five years (defined as height for age below -2 standard deviation (SD) of the age- and sex-specific median according to the World Health Organization [WHO] Child Growth Standards [Onis 2006]); (ii) the percentage of underweight children under five years (weight for age < -2 SD); (iii) the percentage of wasting for children under five years (weight for height < -2 SD); and (iv) the percentage of children aged 6–59 months who are anaemic (haemoglobin concentration <11.0 g/dl) (Table 1, p 46).

Using the data sources described above, we generate PC-level estimates of child malnutrition through two distinct methodologies. The first method involves using the boundary shapefiles to build a crosswalk between districts and PCs. Using this crosswalk, district-level data can be transformed and aggregated to generate PC-level estimates. We apply this crosswalk methodology to the district-level NFHS-4 malnutrition data to generate estimates for the PCs. The second method involves directly aggregating the individual malnutrition data to the PC level. Aggregation to the PC level is possible since the randomly displaced GPS locations of the sampling clusters are provided in the data source, so that the PC of each individual data point can be determined.

Developing a Crosswalk from Districts to PCs

Methods for geographic interpolation using GIS have been previously described (Loganet al 2014; Hibbert et al 2009; Forsyth et al 2006), but their applications to public health have not been thoroughly explored. We apply these methods to interpolate PC-level child malnutrition data given district boundaries. Briefly, we superimposed the shapefiles for district and PC boundaries, revealing segments of districts contained within each PC (see Figure A1). A given district could have segments that fall in several different PCs. For each of these district segments, the proportion of the total district area and population was calculated using raster maps, allowing us to calculate area and population estimates for PCs. For example, if a PC is made up of 40% district A and 60% district B, the total area and population of the PC can be calculated by computing the weighted average of the area and population of the two districts. A data set was generated in which each row corresponded to a segment of a district, and columns included the district ID, the PC ID in which the segment fell, the proportion of the district’s population in that segment, and the proportion of the district’s area in that segment.

Using ArcGIS Pro version 2.0 (Esri, Redlands, California), the geographic area in square kilometres of each district was calculated using the Kalianpur 1975/India Zone IIa coordinate system (EPSG:24379), and saved into a field called Tot_Area. The Intersect command was used with the PCs and districts as the input shapefiles. This command creates a new shapefile, (named PC_District_Intersect) splitting the polygons where they are not identical, creating new geometric shapes. The new geometries are given the attributes of both, the overlapping PC and district. Next, the area in square kilometres of the new polygons was calculated using the Kalianpur 1975/India Zone IIa coordinate system into a field named AREA_GEO. A new field Pct_Area was calculated with the formula: “AREA_GEO/Tot_Area.” The values in this Pct_Area field represent the percentage of district area contained in each new shape. Next, all polygons with a Pct_Area value less than .0001 (less than a hundredth of a percent of the area) were deleted. These extremely small areas are “slivers” that act as noise, and are created from slight boundary inaccuracies between the district and PC shapefiles.

Then, the Zonal Statistics command was performed using the district shapefile as the “zone” and the AsiaPop2015 population raster ( as the underlying data raster. This raster contains population estimates for all of India at a resolution of 100 metres, updated as of 2015. Results from this were saved in the Tot_pop field, and represent total population in the district. Then, the Zonal Statistics command was performed on the PC_District_Intersect shapefile, using the AsiaPop2015 population raster. These results were saved in the POP field, and represent the population in each portion of the PC/district intersection. Then the percentage of population for each apportioned area contains was calculated with a formula: “POP/Tot_pop.” This was saved in the Pct_pop field. The PC_District_Intersect attribute table was exported to excel and eventually to R.

The PC_District_Intersect attribute table contained 1,530 rows, corresponding to 1,530 segments of districts. In order to calculate PC-level estimates, this attribute table was merged with the NFHS-4 district-level data, and the district malnutrition data columns were multiplied by the Tot_pop field, resulting in columns representing the estimated number of individuals in each district segment with a particular malnutrition state. The data set was then aggregated by PC ID, and the resulting malnutrition data columns represented the estimated number of individuals in each PC with a particular malnutrition state. These columns were then divided by the Tot_pop field and multiplied by 100, resulting in columns representing the estimated percent prevalence of malnutrition in each PC.

Generating Direct Estimates by Linking Clusters to PC

Direct aggregation from individual data to PCs was possible because the NFHS-4 utilised DHS sampling cluster locations, for which GPS coordinates are available. In the DHS geographic data, the centre of the populated place of each cluster is recorded with a GPS receiver. These locations are listed in degrees of latitude and longitude. We generated a GIS map of cluster points using the latitude–longitude coordinates and the ArcGIS Add Data from the XY Coordinates tool. We then combined this map with the PC boundary map using the ArcGIS Spatial Join tool. This tool determines which PC each cluster falls into and adds this information to the cluster attribute table. Then, the number of individuals in clusters that linked to each PC constitute the “sample population” for that PC, and the prevalence was computed as number of individuals with anthropometric failure divided by the total number of individuals in each PC.

Statistical Analysis

We adopted the following analytical approaches. First, summary statistics (mean, median, interquartile range) were calculated to describe the distribution of indicators of child malnutrition. Second, we present how the geographies of PC are correlated between the indicators. Then, maps were generated to visualise the geographic variation in the prevalence of child malnutrition and to identify “hotspots” of PCs with particularly high (that is, darker shades in Figure 3, p 47) or low (that is, lighter shades in Figure 3) burdens. All maps show quintiles of prevalence on indicators of child malnutrition.

We also compare the district–PC crosswalk with the direct aggregation method by repeating the above analyses using the directly aggregated data (see Tables A1, A2, A3, and Figures A2, A3, A4, A5 and A6). We also compare the PC rankings produced by both methodologies. For each malnutrition indicator, we rank the PCs using the district–PC crosswalk data, and then using the directly aggregated data. We then calculate the Pearson’s correlation coefficient between the crosswalk rankings and the direct aggregation rankings. Finally, we compare the two methods by calculating the difference between the two prevalence estimates for each PC across all indicators. All statistical analyses were conducted using R version 3.2.


While we have provided rankings of PCs using both GIS methodologies, we discuss the observed empirical patterns based on the results obtained from the district–PC crosswalk data. This is because we believe that this methodology has the potential to be more widely utilised, since the direct aggregation method was only possible given the availability of the survey cluster GPS data, which is generally not provided in most population data sets. Furthermore, the substantive empirical patterns as well as PC rankings were highly similar between the two methods, as evidenced by the results obtained from repeating all analyses using the direct aggregation method (see Tables A1, A2 and A3, Figures A2, A3, A4, A5 and A6).

Distribution and Correlation Indicators across PCs

Across India, over 20% of children under five experienced wasting, over one-third were underweight, over one-third were stunted, and nearly 60% of children aged 6–59 months were anaemic (Table 1). India’s 57% prevalence of child anaemia is a “severe” public health problem, based on the WHO cut-off of greater than or equal to 40% prevalence (WHO 2015). Across PCs, stunting ranged from 13.7% to 61.7%, underweight ranged from 10.5% to 60.9%, wasting ranged from 7.3% and 40.6%, and anaemia ranged from 19.5% to 83.0%. Stunting, underweight, and wasting were approximately symmetrically distributed across PCs, while anaemia was left-skewed (Figure 1, p 46).

Across PCs, stunting was strongly correlated with underweight (r = 0.85). PCs with higher levels of underweight were more likely to be also PCs with higher levels of wasting (r = 0.61). Anaemia showed moderate correlations with other malnutrition indicators (0.35 < r < 0.55) (Figure 2, p 46). In general, PCs with high prevalence of one form of child malnutrition are also likely to suffer from high prevalence of other forms of malnutrition.

Mapping Indicators of Child Malnutrition across PCs

Stunting: PCs in central and north-eastern India show the highest burden of stunting, particularly in Uttar Pradesh, Jharkhand, Bihar, and Madhya Pradesh (Figure 3a, p 47). Shrawasti (Uttar Pradesh, 61.7%), Kaisarganj (Uttar Pradesh, 61.4%), and Bahraich (Uttar Pradesh, 60.6%) are the PCs with the highest burden of stunting. PCs in northern and southern India, in Punjab, Himachal Pradesh, Telangana, Tamil Nadu, and Kerala, show the lowest burden of stunting. Pathanamthitta (15.9%), Kottayam (15.8%), and Idukki (13.7%) in Kerala are the PCs with the lowest burden of stunting. Karnataka, Maharashtra, and Odisha, the states with the highest interquartile range (IQR) for stunting, contain several PCs in the top two and bottom two quintiles of stunting prevalence. Of these states, Karnataka and Maharashtra have a similar distribution where northern PCs show higher prevalence of stunting than southern PCs.

Underweight: The distribution of underweight shows trends similar to that of stunting (Figure 3b, p 47). PCs in central and northeastern India show the highest burden of underweight, particularly in Uttar Pradesh, Jharkhand, Bihar, Chhattisgarh and Madhya Pradesh. The PCs with the highest prevalence of underweight are Singhbhum in Jharkhand (60.9%), Puruliya in West Bengal (58.2%), and Shahjahanpur in Uttar Pradesh (54.3%). PCs in northern, southern, and eastern India show the lowest burden of stunting, such as in Punjab, Himachal Pradesh, Uttarakhand, Telangana, Tamil Nadu, Kerala, Sikkim, Assam, Arunachal Pradesh, Nagaland, Manipur, Mizoram, and Tripura. Kottayam (12.3%), Kasaragod (11.6%), and Kannur (10.5%) in Kerala are the PCs with the lowest burden of underweight. Karnataka, Odisha, Rajasthan, Maharashtra and West Bengal, the states with the highest IQR for underweight, contain several PCs in the top two and bottom two quintiles. Of these states, Karnataka and Maharashtra have a similar distribution where northern PCs show higher prevalence of underweight than southern PCs. West Bengal and Rajasthan show the opposite pattern where southern PCs show a higher prevalence of underweight than northern PCs.

Wasting: Prevalence of wasting is highest in central and western India (Figure 3c, p 47), particularly in Madhya Pradesh, Gujarat, Maharashtra, Chhattisgarh, and Jharkhand. The PCs with the highest prevalence of wasting are Jamshedpur in Jharkhand (40.6%), Puruliya in West Bengal (34.6%), and Nandurbar in Maharashtra (34.5%). Parts of southern, eastern and northern India show the lowest rates of wasting, such as in Himachal Pradesh, Punjab, Andhra Pradesh, Sikkim, West Bengal, Assam, Arunachal Pradesh, Nagaland, Manipur, Mizoram, and Tripura. The PCs with the lowest prevalence of wasting are Kanyakumari in Tamil Nadu (9.0%), and Inner Manipur (7.6%) and Outer Manipur (7.6%). West Bengal, Odisha, Rajasthan, Karnataka, Bihar, and Tamil Nadu show high within-state, between-PC variability. Of these states, Karnataka and Tamil Nadu have a similar distribution where northern PCs show higher prevalence of wasting than southern PCs. In West Bengal, western PCs show a higher prevalence of wasting than eastern PCs.

Anaemia: The highest rates of anaemia are found mostly throughout central India, particularly in Madhya Pradesh, southern Rajasthan, Haryana, and Gujarat (Figure 3d, p 47). Madhya Pradesh and Haryana have the highest rates of anaemia, with Haryana containing PCs only in the top two quintiles. The PCs with the highest prevalence of anaemia are Singhbum in Jharkhand (83.0%), Banswara in Rajasthan (79.3%), and Khargone in Madhya Pradesh (79.1%). States in southern, eastern, and parts of northern India have PCs in the bottom two quintiles of anaemia, such as Tamil Nadu, Kerala, Maharashtra, Punjab, Chhattisgarh, Sikkim, West Bengal, Assam, Arunachal Pradesh, Nagaland, Manipur, Mizoram, and Tripura. The PCs with the lowest prevalence of anaemia are Nagaland (20.2%), Attingal (19.5%), and Kollam (19.5%) in Kerala. In Odisha, western PCs show higher rates of anaemia than eastern PCs. In Karnataka, northern PCs show higher rates of anaemia than southern PCs. In Rajasthan, southern PCs show higher rates of anaemia than northern PCs. And, in Gujarat, western PCs show higher rates of anaemia than eastern PCs.

A total of 72 PCs were in the top two quintiles of prevalence for all indicators of child malnutrition (Table 2, p 48). Of these, 12 PCs were in Jharkhand, 19 in Madhya Pradesh, 10 in Karnataka, six in Rajasthan and eight in Uttar Pradesh. Of these 72, 13 PCs were also in the top quintile of all variables (in bold in Table 2). Twenty-nine PCs were in the bottom quintile for all four outcome variables (in bold in Table 2). Of these, 12 were in Kerala, four were in West Bengal, and six in Odisha.

State-specific Variations and Deviant PCs

Kerala and Goa had the lowest median prevalence of stunting, and Bihar and Uttar Pradesh had the highest (Figure 4a, p 49). Karnataka had the highest IQR of stunting among PCs. Jammu and Kashmir, and Kerala had the lowest median prevalence of underweight, and Jharkhand and Bihar had the highest (Figure 4b, p 49). Odisha had the highest IQR of underweight among PCs. Manipur and Nagaland had the lowest median prevalence of wasting (Figure 4c, p 49). Daman and Diu, Dadra and Nagar Haveli, and Jharkhand had the highest median prevalence of wasting. West Bengal had the highest IQR of wasting among PCs. Nagaland and Manipur had the lowest median prevalence of anaemia, and Chandigarh and Daman and Diu had the highest (Figure 4d, p 49). Odisha had the highest IQR of anaemia among PCs. Overall, Nagaland, Manipur, Mizoram, and Kerala showed low median prevalence of the variables of interest. Jharkhand showed high median prevalence of all variables.

We identify positive deviant PCs, that is, PCs with low prevalence nested within states with high prevalence (Table 3). The positive deviant PCs provide an opportunity to identify best practices within states that have a high prevalence of child malnutrition. Jamnagar in Gujarat is the only positive deviant for stunting. The positive deviants for underweight are located entirely in Maharashtra and Rajasthan, those for wasting are located almost entirely in Odisha and Rajasthan, and those for anaemia are located entirely in Uttar Pradesh and Gujarat. Jaipur, Jhunjhunun, and Sikar in Rajasthan are positive deviants for both wasting and underweight. Positive deviant PCs were not found in states other than Gujarat, Maharashtra, Rajasthan, Odisha, and Uttar Pradesh. Future studies should focus on these PCs to find positive practices or characteristics that can be applied to other PCs to improve child malnutrition outcomes. We found no negative deviant PCs, that is, PCs with high prevalence nested within states with low prevalence.

Comparison with Direct Aggregation Method

In order to directly compare the results of the district–PC crosswalk method and the direct aggregation method, we first compare PC rankings. For each malnutrition indicator, the 543 PCs were ranked based on the crosswalk data and then by the directly aggregated data. The correlations between these two rankings were positive and strong for all indicators, as is reflected by r = 0.92 for stunting, r = 0.92 for underweight, r = 0.84 for wasting, and r = 0.89 for anaemia. We also compared the two prevalence estimates generated by the two methodologies for each PC across all indicators. The absolute value of the difference in prevalence estimates between the two methods had a mean of 2.9% and a SD of 2.7% for stunting, a mean of 3.1% and a SD of 3.1% for underweight, a mean of 2.6% and a SD 2.8% for wasting, and a mean of 3.8% and a SD of 4.0% for anaemia.

The overall trends described above are recapitulated in the malnutrition data generated by direct aggregation to the PC level. Summary statistics (Table A1) are similar to those obtained with the crosswalk data, although the directly aggregated data shows a wider IQR across PCs. The list of PCs in the top and bottom quintiles showed overlap with that generated by the crosswalk method (Table A2). Twenty-eight out of 57 PCs that were in the bottom two quintiles of all indicators and 44 out of 62 PCs that were in the top two quintiles of all indicators also appeared on the list generated by the crosswalk method. The direct aggregation method led to identification of notably more positive deviant PCs than the crosswalk method (Table A3). All-India box plots and correlations are very similar to those produced using the crosswalk method (Figures A2 and A4). Maps produced using the directly aggregated data show similar national and intra-state trends across PCs (Figure A4).


Our study has three salient findings. First, prevalence of child stunting, underweight, wasting, and anaemia was highly variable across PCs. State-specific analysis showed that the relative importance of the PC level may be different across states. Second, there were moderate/high correlations between malnutrition indicators at the PC level, indicating that several PCs experience a multiple burden of child malnutrition that must be addressed concurrently. Third, we found several PCs in Madhya Pradesh and Jharkhand that had high prevalence of all indicators of interest, and may represent the highest priority for health interventions. We also found PCs that show low prevalence for all indicators, and represent positive deviant PCs that should be investigated to elucidate best practices for child nutrition.

Why might one expect substantial variation at the PC level in indicators of child malnutrition? One, such variation may simply reflect how risk factors for malnutrition are distributed across PCs. For example, household poverty has been shown to be a significant risk factor for stunting, wasting, and underweight in India (Corsi et al 2016; Kim et al 2017), so certain PCs that have a large proportion of poor households may also show high prevalence of child malnutrition. A prior multilevel analysis of household poverty and health spending indeed reported significant variation at the local level (Kim et al 2016; Mohanty et al 2018). Thus, PC-level variation in malnutrition may be a result of the underlying distribution of malnutrition risk factors. This may also explain intra-state variation in states like Karnataka, Odisha and Maharashtra, which have PCs in the top two and bottom two quintiles across many indicators. Indeed, further research is required to determine the exact mechanism for intrastate malnutrition variation across PCs.

There are also factors relating to the MPs that may notably impact child malnutrition at the PC level, specifically with regards to the MPLADS. This programme has grown significantly since its inception, with the annual allotted budget per MP increasing 100-fold from ₹ 0.05 crore in 1993 to ₹ 5 crore in 2011 and onwards (MoSPI 2016). Since 2015, there have been several approved MPLADS relating to health, such as the equipment of local hospitals, purchase of ambulances and hearse vans, and the installation of outdoor gyms (MoSPI 2015a, 2015b, 2015c). Thus, the MPLADS has the potential to significantly influence PC-level health outcomes, especially since “preference is given to works relating to national priorities, such as … public health” (MoSPI 2016). Given the distinct nutritional profile of each PC, PC-level interventions should be developed considering the local context. This is precisely where programmes such as SPARC, PRS, and LAMP are needed to support MPs in understanding their constituents’ needs and in making informed decisions about policy and resource allocation.

The role of MPs in MPLADS is of particular interest in the context of political business cycles, defined by Blair (2017) as “increased spending by governments just before elections in the hope of staying in power.” During the term of the 15th Lok Sabha from 2009 to 2014, MPs vying for re-election in 2014 strategically spent the bulk of their allotted MPLADS funds towards the end of their term (Blair 2017). Additionally, the age of the MP significantly affects MPLADS spending to promote political business cycles, with younger MPs utilising funds more uniformly across time than older ones (Pal and Das 2010). Overall, there was considerable variation in MPLADS spending, with a mean and SD of 0.46 and 0.38, respectively, for the proportion of allotted funds spent from May 2004 to October 2006. It is clear that several personal attributes of MPs, such as age and intention to rerun, all significantly impact MPLADS spending and, thus, development and health at the PC level.

It may also be that constituents’ demands ultimately determine PC-level spending. Pal and Das (2010) found that constituents’ awareness and demands significantly have an impact on the implementation of MPLADS funds, with more awareness leading to more consistent usage of funds, and “[leave] less scope for the MPs to utilize funds with political motives.” This suggests that MPs may be less likely to spend funds as a means to promote political business cycles if their constituents are informed and proactive about their demands.

Our methods can be easily extended to estimate population outcomes at other geopolitical levels. One reasonable extension, especially relevant for child malnutrition, is calculating estimates for the 4,120 assembly constituencies (ACs), represented by members of the legislative assembly (MLAs) who are elected to the legislature of the state governments (ECI 2018). We found considerable intra-state variation of child malnutrition, which underscores the importance of understanding the local context when developing health interventions. Since MLAs are representatives of smaller geopolitical units than MPs, understanding AC-level outcomes may better inform local interventions and increase accountability for MLAs as well as MPs. Additionally, calculating AC-level estimates may promote the success of POSHAN Abhiyaan, since the impact of national efforts is supplemented by the support of representatives at all political levels.

In summary, the large variations across PCs suggest that one needs to focus on both describing the magnitude of differences across various health and development indicators, as well as understanding to what extent these differences are a consequence of PC-specific processes as opposed to reflecting the characteristics endogenous to the PC. In this article, we present a methodology that should enable researchers and policymakers to generate PC estimates from abundant data available at the district level from multiple sources. We believe that, in order for the policy discourse to be effective, there needs to be a complementary data-driven discourse in the political domain. Data can be empowering to the MPs as well as their constituents. The ranking of PCs on indicators of child malnutrition, we hope, will help prioritise which PCs need targeting in order to realise the goals of the national POSHAN Abhiyaan programme.


[S V Subramanian conceptualised and designed the study. Akshay Swaminathan contributed to the conceptualisation and led the data analysis, co-wrote the first draft, and led the revision. Rockli Kim contributed to the conceptualisation, analysis and interpretation of the results, and writing. Jeffrey C Blossom, William Joe, Yun Xu and S V Subramanian contributed to the data analysis, interpretation of the results and writing. Alok Kumar and R Venkataramanan contributed to critical revisions. S V Subramanian provided overall supervision to the study. All authors approved the final submission of the study.]


Balarajan, Y, U Ramakrishnan, E Özaltin, A H Shankar and SV Subramanian (2011): “Anaemia in Low-income and Middle-income Countries,” Lancet, Vol 378, No 9809, pp 2123–35.

Blair, H (2017): “Constituency Development Funds in India: Do They Invite a Political Business Cycle,” Economic & Political Weekly, Vol 52, No 31, pp 99–105.

Burgert, C R, J Colston, T Roy and B Zachary (2013): “Geographic Displacement Procedure and Georeferenced Data Release Policy for the Demographic and Health Surveys,” DHS Spatial Analysis Reports No 7, ICF International and United States Agency for International Development, Calverton, Maryland, United States.

Corsi, D J, I Mejía-Guevara and S V Subramanian (2016): “Improving Household-level Nutrition-specific and Nutrition-sensitive Conditions Key to Reducing Child Undernutrition in India,”Social Science & Medicine, Vol 157, pp 189–92.

Corsi, D J, I Mejía-Guevar and S V Subramanian (2016): “Risk Factors for Chronic Undernutrition Among Children in India: Estimating Relative Importance, Population Attributable Risk and Fractions,” Social Science & Medicine, Vol 157, pp 165–85.

DHS (2018a): “DHS Overview,” DHS Program: Demographic and Health Surveys,

— (2018b): “India Demographic and Health Surveys Geographic Dataset,” DHS Program: Demographic and Health Surveys, 13 February,

ECI (2018): “State/UT Wise Seats in the Assembly and Their Reservation Status,” Election Commission of India, Government of India, New Delhi,

Forsyth, A, K Schmitz, M Oakes, J Zimmerman and J Koepp (2006): “Standards for Environmental Measurement Using GIS: Toward a Protocol for Protocols,” Journal of Physical Activity and Health, Vol 3, No s1, pp S241–57.

Github (2014a): “India–Map of Parliamentary Constituencies, 2014,”

— (2014b): “India–District Map,”

Hibbert, J, A Liese, A Lawson, D Porter, R Puett, D Standiford et al (2009): “Evaluating Geographic Imputation Approaches for Zip Code Level Data: An Application to a Study of Pediatric Diabetes,” International Journal of Health Geographics, Vol 8, No 54, pp 1–11.

IIPS (2016): “National Family Health Survey-4: 2015–2016: India Fact Sheet,” International
Institute for Population Sciences, Mumbai.

Kim, R, I Mejía-Guevara, D J Corsi, V M Aguayo and S V Subramanian (2017): “Relative Importance of 13 Correlates of Child Stunting in South Asia: Insights from Nationally Representative Data from Afghanistan, Bangladesh, India, Nepal, and Pakistan,” Social Science & Medicine, Vol 187, pp 144–54.

Kim, R, S Mohanty and S Subramanian (2016): “Multilevel Geographies of Poverty in India,” World Development, Vol 87, pp 349–59.

Logan, J, Z Xu and B Stults (2014): “Interpolating US Decennial Census Tract Data from as Early as 1970 to 2010: A Longtitudinal Tract Database,” Professional Geographer, Vol 66, No 3, pp 412–20.

MoSPI (2015a): “MPLADS: Purchas of Equipments for Government Hospitals/Dispensaries,” Ministry of Statistics and Programme Implementation, Government of India, New Delhi,

— (2015b). “MPLADS: Installation of Garden Gym Machines under the MPLADS,” Ministry of Statistics and Programme Implementation, Government of India, New Delhi,

— (2015c): “MPLADS: Purchase of Ambulances/Hearse Vans,” Ministry of Statistics and Programme Implementation, Government of India, New Delhi,

— (2016): “Annual Report 2015–16: MPLADS,” Ministry of Statistics and Programme Implementation, Government of India, New Delhi,

— (2017a): “Members of Parliament Local Area Development Scheme,” Ministry of Statistics and Programme Implementation, Government of India, New Delhi,

— (2017b): “Summary of Funds Released by MPLADS,” Ministry of Statistics and Programme Implementation, Government of India, New Delhi,

Mohanty, S, R Kim, P K Khan and S Subramanian (2018): “Geographic Variation in Household and Catastrophic Health Spending in India: Assessing the Relative Importance of Villages, Districts, and States, 2011–2012,” Milbank Quarterly, Vol 96, No 1, pp 167–206.

NITI Aayog (2018): “Transformation of Aspirational Districts,” Governemnt of India, New Delhi,

Onis, M (2006): “WHO Child Growth Standards Based on Length/Height, Weight and Age,” Acta Paediatrica, Vol 96, pp 76–85.

Pal, R and A Das (2010): “A Scrutiny of the MP-LADS in India: Who Is It For?” Economic & Political Weekly, Vol 45, No 2, pp 63–67.

Parliament of India (2017): Lok Sabha,

PIB (2018): “POSHAN Abhiyaan to Address Malnutrition through Convergence, Use of Technology and a Targeted Approach,” Press Information Bureau, 20 March,

PRS Legislative Research (2018): “What We Do,”

Subramanian, S V, I Mejía-Guevara and A Krishna (2016): “Rethinking Policy Perspectives on Childhood Stunting: Time to Formulate a Structural and Multifactorial Strategy,” Maternal & Child Nutrition, Vol 12, No S1, pp 219–36.

Swaniti Initiative (2017): “SPARC: Supporting Parliamentarians on Analysis and Research in Constituency,”

WHO (2015): “The Global Prevalence of Anaemia in 2011,” World Health Organization, Geneva,

Updated On : 15th Jan, 2019


(-) Hide

EPW looks forward to your comments. Please note that comments are moderated as per our comments policy. They may take some time to appear. A comment, if suitable, may be selected for publication in the Letters pages of EPW.

Back to Top