Like all national poverty rates, India’s poverty rate is interpreted as the share of the population that is poor in a given year. In this post, Merfeld and Morduch argue that, in practice, India’s poverty rate is better thought of as the approximate fraction of the year that households experience poverty. They describe how this is rooted in the nature of data collection, and how it changes understandings of poverty and policy in the country.
The reduction of poverty in India since the 1980s is one of the great achievements of the Indian economy. Indeed, it is one of the great global economic achievements of the past half-century. Debates about the exact levels of poverty and rates of change in India persist, but there is no serious debate about the historical sweep of change.1 Yet, unnoticed in the public discussion is that the idea of ‘poverty’ in national statistics is not what has been assumed. In practice, if not by definition, India’s national poverty rate reflects a more encompassing notion of deprivation than official documents describe. As a result, India’s progress has likely come about thanks to a broader set of economic and social mechanisms than economists have considered.
What does the poverty rate measure?
Since the start of the 20th century, poverty rates have been defined as the fraction of the population living below poverty lines. A poverty rate of 10% is taken to mean that 10% of India is unable to afford a basic standard of living. The definition is so familiar and commonsensical that it goes unquestioned.
But, in fact, a more accurate understanding of what the national poverty rate measures, to an approximation, is the average fraction of the year that people spend below poverty lines. This is not what the poverty rate is meant to measure, but we argue that this is what is actually reflected in the official numbers tabulated by the government.2 A poverty rate of 10% then means that, on average, across all households and all quarters of the year, 10% of those quarters are ones in which households experience poverty.
Contrary to the conventional notion of poverty, India’s poverty rate captures poverty experienced temporarily by people who would not usually be considered poor. Similarly, it reflects the fact that some ‘poor’ people are not poor for the full year. In short, the de facto national poverty rate – as opposed to the textbook definition – captures the fluid nature of economic status within the year, dynamic conditions that have defined well-being on the sub-continent for hundreds of years. It is a reality in which seasons often dictate economic opportunities, and where getting through the year requires navigating between better and worse times. Sometimes the variability occurs by choice, and sometimes by circumstance.3
In some ways this de facto notion of poverty, by bringing in a temporal dimension, captures the challenges of deprivation more accurately than what is captured by the narrower, conventional notion of poverty. In other ways, the de facto poverty rate can obscure aspects of households’ choices and economic conditions.
We are not arguing from a normative, philosophical position that this is what the national poverty rate should measure. Our aim here is only to show what, in practice, the national poverty rate actually measures – to a reasonable approximation.
An “average of poverties”
Specifically, the de facto poverty rate approximates an “average of poverties.” Think of the year divided into quarters. The “average of poverties” takes each quarter for each household as an observation (assigning a 1 if households are poor in that quarter or a 0 if not poor) and creates a national average. What results is the average fraction of the year that households are poor, counting the quarterly experiences of people who are sometimes poor, always poor, and never poor.4
One way to think about it is that the conventional poverty concept imagines taking all residents and counting how many are poor based on their average consumption for the year. But, instead, imagine taking all the quarters of the year for all residents and counting how many of those are experienced as poverty. This, we argue, is what the national poverty rate approximates.
How did this happen?
No one set out to transform the national poverty rate into an approximation of the “average of poverties.” It happened without a change in the mathematical formula used to construct the poverty rate. Instead, it slipped in through the back door; the transformation arose indirectly as an approximation which holds true in large samples, and this has made the result harder to see at first.
The outcome is rooted in the nature of data collection, perhaps the least appreciated part of poverty measurement. Specifically, the result stems from a series of reasonable choices made by the National Sample Survey Office (NSSO) about how to collect data on the economic conditions of households, in which data accuracy has been prioritised for the given budget (NSSO, 2001).
There are four key choices that led to this outcome. The first is the choice to measure household resources based on what they spend and consume rather than what they earn, a choice that is common globally, with the biggest exceptions being high-income countries and countries in Latin America. The second is to ask questions about food expenditure and common consumption items with a short recall period, recognising the imperfections of memory. In India, this design choice has received a great deal of attention, and the recall period has varied between the past 7 and 30 days (with a small group of bigger items and large durables being collected with 365-day recall). Our analysis of the 2011-12 National Sample Survey (NSS) shows, for example, that 82% of overall household expenditures in the NSS were collected with 30-day recall, more than half of which were on food.
The third choice is to lower the survey budget by interviewing households only once during the year – a practice used by most statistical agencies in low- and middle-income countries (Smith et al. 2014).5 And the fourth is to compensate for the one-time interviews by choosing new random samples to interview each quarter, thereby picking up the impacts of the seasons on economic conditions across the population while maintaining representation quarter by quarter (NSSO, 2001).
In themselves, these choices are sensible and uncontroversial. In fact, the choices roughly conform to the global consensus on best practices for survey design (World Bank and Food and Agriculture Organization of the United Nations (FAO), 2019, Mancini and Vecchi 2022).6 When the choices are taken together, however, they lead to a result that is surprisingly different from the intended one.
Appreciating the result starts with the observation that, because households are interviewed once during the year with a main focus on their recent conditions, the NSSO does not in fact collect data on the annual expenditure or consumption of any individual household. To estimate that, households would have to be interviewed in each of the four quarters and totaled. Instead, the one-time interviews yield data that reflect conditions specific to the date in which the household was interviewed.7 If a rural household is interviewed at harvest time, their data will likely show them as being better off than their yearly average. And if they are interviewed in a low season, their data will reflect conditions that are worse than average. In the 2011-12 NSS, for example, there is a marked drop in the average consumption of households in the third quarter of the year, the time of the monsoon, a historically challenging season in rural India.8,9
This sampling strategy nonetheless produces a good approximation of average consumption for the entire population, putting aside other methodological issues (Scott 1992). For poverty, however, the result shifts due to the nonlinear way that consumption (expenditures in practice) maps to poverty status for each household.
How does this work? Consider a household that is not poor in the first quarter but poor in the three remaining quarters of the year. Their quarterly poverty status can be represented as {0, 1, 1, 1}, where being not poor in the quarter = 0 and being poor = 1. Given the sampling strategy, the household is interviewed in only one of those quarters, with a 75% chance they will show up as being below the poverty line in the data. The chance that they are interviewed in any quarter is equi-probable, and hence their expected contribution to the overall poverty measure is (0 + 1 + 1 + 1)/4 = 3/4. This is simply the average fraction of quarters that this household spends in poverty. It is the approximate result that would emerge if the NSSO sampled 1000 households with a similar quarterly pattern, randomly choosing in which quarter to interview each household, and then calculated the collective poverty rate.10
When the entire sample is considered in this way, the data will include some households that are always poor {1, 1, 1, 1}, never poor {0, 0, 0, 0}, and variations in between (for example, {0, 0, 1, 0}). The average across the thousands of households in the NSS will yield an approximately correct average of the fraction of quarters spent below the poverty line in the population. In other words, it is an approximation to the “average of poverties,” not a noisy measure of the conventional concept of poverty.
The quantitative implications can be large
The insights above begin with the stated methodology used by NSSO, and then take the method to its logical conclusion. We cannot, however, use the NSS to compare the conventional concept of poverty to the “average of poverties”. The problem, as above, is that the NSS does not provide an accurate measure of annual consumption for households, making the conventional poverty measure impossible to calculate.
Instead, we turn to data collected by the International Crops Research Institute for the Semi-Arid Tropics (ICRISAT) in rural south India between 2010 and 2014. Their Village Dynamics in South Asia (VDSA) survey provides longitudinal data on income and expenditure each month for a range of households. We analyse data on 1,526 households over 60 months, from July 2010 to June 2015. The households come from 30 villages across 15 districts in nine states.
The VDSA data is not collected from a random sample of households in India, so it is impossible to compare results directly to those from the NSS. Importantly, the households in the VDSA are rural and depend largely on agriculture, relative to the rest of the population. Seasonality thus marks their lives and livelihoods in ways that are more pronounced than for typical Indian households.
We can, however, use the VDSA data to demonstrate how the ideas above fit with actual data, clarifying how similar mechanisms can emerge in other data sets like the NSS. We do this in Figure 1. When we use the VDSA data to measure the conventional poverty rate (using 12 months of data for each household to calculate their average expenditure for the entire year), we get the bar on the left. In the sample, the headcount is centred just above 29%. To the right is the “average of poverties,” centred near 37%, showing the average fraction of months in poverty across the sample. The latter is 26% (or 8 percentage points) larger than the former, driven mostly by accounting for poverty in the lean season experienced by ‘non-poor’ households. Only when households completely smooth consumption across months will the two calculations converge – at 29%.
Figure 1. Comparisons of poverty measures

Notes: i) The headcount mean is the annual poverty rate when using average expenditure for the entire year. ii) The “average of poverties” mean is the annual poverty rate when using the proportion of months in the year that the household is poor. iii) The density estimate is estimated annual poverty rates when using a single, randomly selected month, across 1,000 replications.
Source: Merfeld and Morduch (2024).

The NSSO draws random samples of households from across India to interview each quarter. As noted above, because the questions depend largely on short-term spending and consumption patterns, the data in the NSS reflect conditions around the date that households were interviewed. To mimic this in the monthly data, we next estimate annual poverty rates using a single, randomly selected month for each household, which is used as the prediction of the household’s annual average consumption. We then plug that into the standard mathematical formula for poverty.11
The figure shows that the 1,000 replications converge to the “average of poverties” (the right-hand bar, at 37%). Some replications are somewhat lower and some are somewhat higher, but together they give an approximation to the true “average of poverties”. In short, even with just a single observation from a randomly-chosen period for each household, we arrive at an approximation of a measure calculated using complete longitudinal data for every household in every period. This happens thanks to the specifics of randomised data collection through the year.
The figure applies to the headcount, but the same idea can be applied to distributionally-sensitive poverty measures like those of Watts (1968) and Foster et al. (1984), as well as inequality measures. In that case, the “average of poverties” captures the times of greatest deprivation during the year, registering moments of greatest need independently. In the conventional measure, in contrast, those times with worse-than-average conditions are offset one-for-one by times with better-than-average conditions. The VDSA shows that the “average of poverties” is 40-50% higher than the conventional measure in the sample from rural south India.12
What this means for economic policy
Under the conventional notion of poverty, the only way to reduce the poverty rate is to increase average yearly resources for poor households. The focus is then on strategies that include creating better paying jobs, building human capital, increasing public benefits, and generating inclusive growth.
But the de facto poverty rate reflects more than that, and policy makers can use a broader set of tools, including financial services policy and targeted public benefits, to reduce the measure. Interventions that support households during times of scarcity will be particularly impactful on the national poverty rate. Helping households save for moments when resources are scarce (as in, for example, Somville and Vandewalle 2023), or helping household borrow in times of need can also shape the poverty rate. Similarly, policies like the Mahatma Gandhi National Rural Employment Guarantee Act (MNREGA) that create jobs in low seasons can, in principle, reduce poverty by smoothing seasonal patterns of consumption, independent of any impacts on average yearly consumption. An unwritten part of the story of India’s historic reduction of the poverty rate is bound up with these kinds of changes.
What now?
One response to the arguments here is that the NSSO should immediately redesign practices in order to accurately measure the conventional notion of poverty. The conventional poverty rate would indeed be useful to policy makers, but it is not directly measurable with the existing NSS data. For example, current poverty rates make it more difficult to talk about the ‘poor’ as a group, since the de facto measures capture a variety of experiences of poverty, many of which are transient or seasonal.
That said, maintaining the historical comparability of poverty rates requires that the NSSO continues to calculate and publish the “average of poverties”.13 Beyond its historical legacy, the “average of poverties” can be a valuable tool for policymakers once its properties are recognised. Relative to the conventional approach to poverty, the “average of poverties” captures a broader array of experiences across a broader array of households. It reflects the fundamentally dynamic nature of poverty, and it highlights that households’ ability to protect their living standards in the most difficult periods of the year can matter independently of their average resources.14
We do not prefer one poverty concept over the other – both have pros and cons that need to be considered, conceptually and logistically. One concern should be with accurate data collection on durables and bulk purchases in household consumption. Another need is to examine the role of measurement errors, some of which may be averaged out in the conventional poverty measure but not in the “average of poverties”.
Other issues hinge on households’ choices and behaviours. Some of the variability of consumption during the year is caused by instability and illiquidity due to seasonality and varying economic conditions. But some is likely due to deliberate choices made by households to improve their well-being, such as clustering purchases during festival times or paying school fees in set months (Paxson 1993). If these choices are made without constraint, it could be (in theory) that households are choosing to experience greater poverty in some periods in order to enjoy a higher level of consumption in other periods. To that extent, the main policy concern should be with whether the households have adequate resources across the year, without similar concern for when households spend.
These are all issues that arise with the current, de facto poverty rate in India, and they present challenges and opportunities. We end by noting that these are not just ideas for India; these kinds of methodological and conceptual questions also arise in many other countries (see Gibson et al. 2003, Jolliffe and Serajuddin 2018, Merfeld and Morduch 2024). By taking a clear-eyed look at India’s national poverty rate, Indian statisticians and economists can also lead in refining global practices and concepts.
Notes:
- See, for example, the 2022 e-Symposium in Ideas for India (Ghatak 2022), and the longer history of the debate (for example, Bardhan et al. 2017, Deaton and Dreze 2002, Srinivasan 2007, Subramanian 2012, Roy and Van der Weide 2022).
- We focus on the national poverty rate here, the basic headcount.
- Bardhan (1984) describes how seasonal risk has long shaped agrarian institutions, and Breza et al. (2021) provide a recent example, using a randomised experiment to reveal seasonal qualities of labour markets. Financial diaries led by Orlanda Ruthven in Uttar Pradesh and Delhi, detail how poor households cope with variability in economic conditions through the year (Collins et al. 2009), and we argue that the variability finds its way into poverty measures. In Merfeld and Morduch (2024), we apply this insight to a broad range of countries.
- In Merfeld and Morduch (2024), we define the “average of poverties” and its properties, connecting the ideas to literature on seasonality and global poverty measurement.
- The practice of the NSSO has been to have a single interview with each household, but they changed the method for the NSS Household Consumption and Expenditure Survey, 2022-23 , where households were instead visited three times over a three-month period, each time being asked a different set of questions, where the order of modules was randomly chosen (Anant 2024). This procedure is designed to improve data quality as surveys grow ever longer, and it does not change the fundamental issues we describe.
- The Minister of State, Ministry of Statistics and Programme Implementation (MoSPI), told Parliament in 2024 that NSS methodology aligns with the United Nations (UN) Handbook, Designing Household Survey Samples: Practical Guidelines (Rajora 2024). Although the UN Handbook is mute on implications of asking questions with short-recall periods coupled with stratified sampling over time, the NSS methodology aligns with references like World Bank and FAO (2019).
- As noted above, the NSSO changed the method for the NSS Household Consumption and Expenditure Survey, 2022-23, where households are interviewed three times. The three interviews, though, are restricted to a single three-month period within the year, leaving our fundamental insight unchanged – that is, most of the data reflects conditions around the interview dates, not the full year.
- While measurement error can add to the appearance of variability in poverty through the year, the systematic seasonal pattern suggests that much of the variation is rooted in households’ changing conditions.
- These concerns and observations are not new. Deaton and Grosh (2000), for example, outline the idea that collecting data with short-term recall has potential implications for the eventual measures of poverty. See also the closely-related work of Gibson et al. (2003) and Jolliffe and Serajuddin (2018). Our contribution builds on their insights by describing how data collection methods shift the meaning of what is measured as ‘poverty’ in specific, systematic ways that are interpretable and potentially useful.
- This interpretation depends on the assumption that a household’s consumption measured at the time of the interview is a reasonable approximation for their consumption during that quarter, but not necessarily for the full year.
- The NSS stratifies on quarter rather than month, but here we take advantage of monthly data.
- The Watts (1968) and Foster et al. (1984) measures are not cardinally meaningful. The large percentage increases in the measure matter to the extent that they affect rankings of distributions.
- That said, changing the reference period – as has happened in the past – qualitatively changes poverty measures for the reasons discussed here.
- Our analysis suggests that calculating the conventional poverty measure imposes a high logistical bar and requires much more data than exists. Households would need to be interviewed quarterly given how important seasonality is in economic conditions, especially for poorer households. Or perhaps economic modeling could be used to approximate living standards for the full year. Doing so would yield new numbers and a fundamentally different understanding of conditions.
Further Reading
- Anant, T (2024), ‘Consumer expenditure survey: Its new methodology is superior’, The Mint, 12 June.
- Bardhan, P (1984), Land, Labor, and Rural Poverty, Columbia University Press, New York.
- Bardhan, P, T Srinivasan and AS Bali (2017), ‘Poverty and Inequality in India: An Overview’, in A Banerjee, P Bardhan, R Somanathan and T Srinivasan (eds.), Poverty and Income Distribution in India, Juggernaut, New Delhi.
- Breza, Emily, Supreet Kaur and Yogita Shamdasani (2021), “Labor Rationing”, American Economic Review, 111(10): 3184-
- Collins, D, J Morduch, S Rutherford and O Ruthven (2009), Portfolios of the Poor: How the World’s Poor Live on $2 a Day, Princeton University Press, Princeton.
- Deaton, Angus and Jean Dreze (2002), “Poverty and Inequality in India, A Re-Examination”, Economic and Political Weekly, 37(36): 3729-3748.
- Deaton, A and M Grosh (2000), ‘Consumption’, in M Grosh and P Glewwe (eds.), Designing Household Survey Questionnaires for Developing Countries: Lessons from 15 Years of the Living Standards Measurement Study, World Bank.
- Foster, James, Joel Greer and Erik Thorbecke (1984), “A Class of Decomposable Poverty Measures”, Econometrica, 52(3): 761-
- Ghatak, M (2022), ‘Introduction to e-Symposium: Estimation of poverty in India’, Ideas for India, 10 October.
- Gibson, John, Jikun Huang and Scott Rozelle (2003), “Improving Estimates of Inequality and Poverty From Urban China’s Household Income and Expenditure Survey”, Review of Income and Wealth, 49(1): 53-68. Available here.
- Jolliffe, Dean and Umar Serajuddin (2018), “Noncomparable Poverty Comparisons”, Journal of Development Studies, 54: 533-536.
- Mancini, G and G Vecchi (2022), ‘On the Construction of a Consumption Aggregate for Inequality and Poverty Analysis’, Report, World Bank.
- Merfeld, J and J Morduch (2024), ‘Poverty at Higher Frequency’, Working Paper, KDI School and New York University.
- National Sample Survey Organisation (2001), ‘Concepts and Definitions Used in NSS. (Golden Jubilee Publication)’, Technical Report, Ministry of Statistics and Programme Implementation, Government of India. Available here.
- Paxson, Christina H (1993), “Consumption and Income Seasonality in Thailand”, Journal of Political Economy, 101(1): 39-
- Rajora, S (2024), ‘Concerns over the sampling design of national surveys not tenable: Govt’, Business Standard, 8 February.
- Roy, S and R Van der Weide (2022), ‘Poverty in India Has Declined over the Last Decade But Not As Much As Previously Thought’, World Bank Policy Research Working Paper
- Scott, Chris (1992), “Estimation of Annual Expenditure from One-month Cross-sectional Data in a Household Survey”, Inter-Stat Bulletin, 8(1): 57- Available here.
- Smith, L, O Dupriez and N Troubat (2014), ‘Assessment of the Reliability and Relevance of the Food Data Collected in National Household Consumption and Expenditure Surveys’, IHSN working paper N 008, International Household Survey Network.
- Somville, Vincent and Lore Vandewalle (2023), “Access to banking, savings and consumption smoothing in rural India”, Journal of Public Economics, 223:
- Srinivasan, TN (2007), “Poverty Lines in India: Reflections after the Patna Conference”, Economic and Political Weekly, 42(41): 4155-4165.
- Subramanian, S (2012), The Poverty Line, Oxford University Press, New Delhi.
- Watts, HW (1968), ‘An Economic Definition of Poverty’, in DP Moynihan (ed.), On Understanding Poverty, Basic Books, New York.
- World Bank and FAO (2019), ‘Food Data Collection in Household Consumption and Expenditure Surveys: Guidelines for Low- and Middle-Income Countries’, LSMS
Comments will be held for moderation. Your contact information will not be made public.