Based on multidimensional poverty measurement, the Poverty Ratio (Head Count Ratio) in Tamil Nadu declined from 4.89% in 2015-16 to 1.57% in 2020-21, based on the fourth and fifth rounds of the National Family Health Survey (NFHS) data. Is this too good to believe? Maybe so. Academics have questioned the quality of NFHS data for various reasons, based on the previous four rounds of NFHS databases. Such questions may be raised against the NFHS 5 database also. But first, let us explore the poverty statistics derived from NFHS 5 data using the multidimensional poverty measurement suggested by NITI Aayog and its pointers for policy intervention. After this, we will raise questions about the quality of NFHS data with the aim of using it with caution and to improve data quality keeping the future in mind.
On the MPI
NITI Aayog, armed with a fairly large sample survey data of NFHS 4 (with more than six lakh households in India), estimated the Multidimensional Poverty Index (MPI) and published the baseline report in 2021. The rationale for the MPI was derived from the concept that poverty is the outcome of simultaneous deprivations in multiple functions such as attainments in health, education, and standard of living. The NITI Aayog identified 12 indicators in these three sectors and calculated the weighted average of deprivations in each of these 12 indicators for all men and women surveyed in NFHS 4. If an individual’s aggregate weighted deprivation score was more than 0.33, they were considered multidimensionally poor.
The non-poor may also be deprived in a few of these indicators, but not as much to be classified as multidimensionally poor. The proportion of the population with a deprivation score greater than 0.33 to the total population is defined as the Poverty Ratio or Head Count Ratio. The authors have estimated the MPI and its components for Tamil Nadu using NFHS 5 and compared it with the estimates based on NFHS 4 given by NITI Aayog.
Another interesting aspect of this approach is the estimation of the Intensity of Poverty. This is the weighted-average deprivation score of the multidimensionally poor. For instance, the Intensity of Poverty in Tamil Nadu declined from 39.97% to 38.78% during this period, indicating that the summary measure of multiple deprivations of the poor has only marginally declined in these five years, and has to be underlined for policy focus.
The MPI is a product of Head Count Ratio and Intensity of Poverty. The MPI for Tamil Nadu declined from 0.020 to 0.006. This sharp decline in MPI is largely due to a greater decline in Head Count Ratio compared to Intensity of Poverty. This gives us a clue that any further decline in MPI in Tamil Nadu should happen only by addressing all the dimensions of poverty and reducing its intensity substantially across the State.
Direction of intervention
The deprivation estimation also indicates that the overall population that has been identified as deprived in most of the indicators individually is higher than the population identified as multidimensionally poor. This once again reiterates the point that people may be deprived severely in a few functions, but may not be multidimensionally poor. This adds another aspect of public policy intervention, i.e., attacking poverty in Tamil Nadu should not only be multidimensional but also universal. Only this approach can address deprivations in all the indicators. This will also surely and squarely reduce the Intensity of Poverty in Tamil Nadu.
Also read | T.N.’s success in poverty reduction
Statistically, the Head Count Ratio and Intensity of Poverty can be calculated for each district and segregated by gender, rural and urban, and other dimensions. Therefore, the usefulness of the MPI and its components is enormous in terms of understanding poverty in its totality as well as the granular details that are essential for sectoral and spatial policy and programmatic interventions. The strength of the MPI as an instrument for data-driven public policy depends on the quality of survey data, namely the NFHS data.
Quality of NFHS data
The quality of survey data has been widely debated in academia. The National Sample Survey Organisation’s (NSSO) sample surveys have been debated among economists and statisticians, both in terms of sampling and non-sample errors, right from its initial days in the 1950s. Following several review reports on the NSSO’s methodologies, the NSSO has been attempting to improve sampling design and reduce non-sampling errors, particularly with reference to recall periods for providing consumption expenditure by households. All these are well documented.
Demographers such as K. Srinivasan, S. Irudaya Rajan, and K.S. James have written several articles on the non-sampling errors in different rounds of the NFHS data. They tested, for instance, the arbitrariness in reporting the age of the dead, differences in data quality between educated and uneducated respondents, data quality based on differences in time taken to complete a survey of different household types, etc. All these have serious implications for health data such as fertility and death rates. A market-based approach to decide the data collection process is also critiqued by demographers.
The authors have done a different kind of quality check for NFHS 5 data for Tamil Nadu. For instance, in Tamil Nadu, the NFHS data was collected in two time periods: 8,382 households (30%) in the pre-pandemic period and 19,547 households (70%) in the post-lockdown period, aggregating to 27,929 households for the State. The data collected from 19,547 households in the post-lockdown period should reflect the impact of the first wave of the COVID-19 pandemic. Let us compare pregnant women and their age distribution in the two periods for a glimpse of this. The proportion of pregnant women below the age of 19 years was 18:82; those between 19-21 years was 25:75 compared to the proportion of 32:68 for pregnant women above 21 years. The pandemic has resulted in increasing pregnancy among women below the age of 21 years, more so among teenage girls. Death per 1,000 households surveyed increased from 118.23 to 135.01 — this is clear evidence of the impact of the pandemic.
The authors have estimated the Head Count Ratios for the 12 indicators and found that such ratios were lower in the post-lockdown period than in the pre-pandemic period, leading to the inference that post-lockdown, the deprivation in several functionings was lower, implying a lower poverty ratio as well as Intensity of Poverty. In particular, the deprivation in terms of nutrition and maternal health declined, and schooling and school attendance increased in the post-lockdown period.
Substitution of dry rations for hot meals in the mid-day meal programmes and high pressures in hospitals in handling COVID-19 cases are expected to increase deprivation in nutrition and maternal health in the post-lockdown period, contrary to the decline in deprivation in nutrition and maternal health in the post-pandemic period that we derived from this database. Tamil Nadu is known to have increased enrolment and reduced the dropout rate year after year; hence, the increase in deprivation in terms of schooling should raise questions. As far as school attendance is concerned, we do not know how parents interpreted school attendance during the long period of closure of schools in the lockdown period. Therefore, combined survey data from two different time periods separated by a major pandemic have to be approached with caution while interpreting the statistics derived from the entire database.
Assuming that survey data are from a single time period, it is normal to compare the results of survey data on specific indicators, with the programmatic data derived from official records. There are claims that the deprivation indicators in terms of drinking water and sanitation are on a higher level in Tamil Nadu than the claims made by the respective State government departments. Such issues are common in survey data. For instance, consumption expenditure on foodgrain derived from NSSO data would not be in agreement with the estimation of food consumption, as per the System of National Accounts.
Data use and quality
The quality of survey data has always been a contentious issue in academic and policy debates for various (well-founded) reasons. However, this has not stopped academicians and policymakers from inferring policy directions because such data at a reasonably aggregate level (say at the level of a State), should be useful. As mentioned earlier, in Tamil Nadu, the sharp decline in Head Count Ratio and a marginal decline in Intensity of Poverty in NFHS 5 compared to NFHS 4, cannot be brushed aside.. From this, we can infer that in order to reduce the Intensity of Poverty we need to address deprivations across the entire population, that is there should be a universal approach instead of a targeted approach to addressing it.
The survey data gives us only broad policy pointers whereas programmatic interventions should be curated with ground-level realities. At the same time, continuous engagement with survey data in terms of improving the sample design and response quality has to be sustained. Analysing the data and finding the incongruence of inferences from different databases on an issue would help improve data gathering systems. Let us continue to use survey data both to derive policy conclusions (with caution) and also to help improve data quality.
R. Srinivasan is a Member of the Tamil Nadu State Planning Commission. S. Raja Sethu Durai is a Professor of Economics at University of Hyderabad. The views expressed are personal