Fanciful Numbers and Fictitious Intrigues
- EPW Commentary November 1, 2003
Fanciful Numbers and Fictitious Intrigues
Surjit Bhalla claims that India's poverty rate is now lower than
China's, contradicting the estimates published by the World Bank,
UNDP and others. However, while there are some puzzles in the data,
particularly for India, Bhalla's calculations do not stand up to
India is at the centre of an important ongoing debate within the
develop- ment community about the extent of poverty and inequality in
the world. In an article entitled `Lack of account ability or
accountability?' (Business Standard, July 26, 2003) Surjit Bhalla
claims to expose glaring inconsistencies in the poverty data for
China and India provided by the World Bank and the United Nations
Development Programme (in its 2003 Human Development Report). Both
institutions reckon that the incidence of poverty by a comparable
standard is roughly twice as high in India than China in the late
1990s. Bhalla quotes the estimates from the World Bank's 2003 World
Development Indicators giving headcount indices of 16 per cent and 35
per cent for China and India respectively for a common poverty line
of approximately $1/day at purchasing power parity.
Yet with simple arithmetic "back of the napkin calculations" is how
he puts it Bhalla claims to show that the agencies' own data
contradict their published poverty numbers. Indeed, when he
calculates mean consumption of the poorest 20 per cent or 40 per cent
from the purportedly same data sources he finds the opposite: that
Indians are poorer than the Chinese. Mean incomes of the poor are
about the same.
Bhalla attributes this discrepancy to serious "accounting errors" by
the international agencies. How could the agencies' researchers be so
inept? Bhalla's answer is that the agencies "are a research monopoly,
and suffer from the arrogance, and drawbacks, of every monopoly". By
Bhalla's view, poverty statistics are "big business and like with all
such big businesses, there is politics and intrigue". Bhalla's
readers are left to ponder for themselves what sort of politics and
intrigue within (and presumably between) these agencies has led them
to deceive the world so much, and at the risk of such easy exposure
by observers such as Bhalla.
But wait a minute, let's look more closely at Bhalla's numbers. While
he does not give details on his data sources, it is not too hard to
figure out where his supposed "accounting errors" come from. Without
telling his readers, Bhalla has mixed national accounts data with
distributional data on shares of consumption from household surveys.
To get his number for "mean income of the poorest 20 per cent" he has
multiplied the survey-based consumption share of the poorest 20 per
cent by GDP per capita (and then multiplying by five). To get
his "mean consumption of the poorest 20 per cent" he multiplies the
share by private consumption per capita from the national accounts.
If we consider for a moment how data on consumption shares are
generated we realise that Bhalla is not using an internally
consistent method of estimating mean consumption or income of the
poor. The share data he uses were obtained solely from household
surveys of consumption. To get the consumption share of the poorest
20 per cent (say), the World Bank's researchers have added up all the
consumptions of the poorest 20 per cent ranked by household
consumption per person and divided by total consumption across all
households in the survey. In other words, the data used by Bhalla
were created by dividing mean consumption of the poorest 20 per cent
by the overall survey mean (and dividing again by five). That is how
it should be.
So all we can reasonably claim to know is how aggregate household
consumption as measured in the surveys is distributed. The data
on "shares of the poorest 20 per cent and 40 per cent" used by Bhalla
are not shares of GDP, or share of consumption as measured in the
national accounts, but shares of aggregate household consumption as
obtained from the surveys. To retrieve the mean consumption of the
poor one should be multiplying by mean consumption from the same
survey, not consumption from national accounts.
It is not clear what sense can be made of Bhalla's method of using
instead national accounts data to retrieve the mean consumption or
income of the poor. The numbers he gives for "mean income of the
poorest" are certainly not that. GDP in the national accounts clearly
includes a lot more than household income. On top of that problem, we
simply do not know how income is distributed in India as a whole
since this is not asked in the National Sample Surveys. (We do have
national household income distributions for China, though Bhalla
chooses not to use these data.)
The numbers Bhalla calculates for "mean consumption of the poor" are
arguably no less fanciful. Private consumption per capita from the
national accounts and mean household consumption from a survey are
not measuring the same thing, and should not be expected to agree.
Some of the discrepancy is due to obvious differences in coverage.
Surveys do not include imputed housing rents or the imputed services
to households of financial intermediaries, which the national
accounts do include. These are a growing share of total consumption
in India.1 And what is called "private consumption" in the national
accounts includes things like the expenditures of India's entire non-
governmental non-profit sector, in addition to household consumption.
This is probably also a growing share of total consumption.
There are also measurement errors in both sources. For example, there
is likely to be under-reporting of consumption in the surveys and
problems in the weighting of the survey data, due to out-of-date
sample frames or selective non-compliance in sample assignments. And
there are large errors in the national accounts, as evident from the
substantial revisions that are made to India's consumption
The poverty numbers from the international agencies are based on
surveys not national accounts. Surveys are used almost universally by
those who measure poverty within and outside the World Bank and UNDP.
Indeed, household surveys originated in part from the desire to
measure poverty (starting from Seebohm Rowntree's study of living
conditions in York, England, in the 1890s). Surveys are the only
quantitative data source that can directly tell us about the living
standards of the poor. It has long been recognised that national
accounts are singularly ill-suited for this purpose.
When the surveys are used to measure poverty relative to a common
international poverty line we find that the incidence of poverty is
substantially higher in India than China. Finding a common poverty
line is itself a problematic exercise, as it requires international
price cost-of-living comparisons, as done through Purchasing Power
Parity exchange rates. The underlying price data are not ideal for
either China or India, and there are well known problems in making
index number comparisons when consumption patterns differ. Bhalla
does not appear to be concerned about these issues so I shall not go
into them further here.
We can now understand why the World Bank's numbers differ so much
from those provided by Bhalla. There is no political intrigue here;
the Bank's methods are standard and have been documented in detail in
published papers and reports since its global poverty monitoring work
began. (Better documented I might add than Bhalla's methods.)
However, underlying Bhalla's fanciful numbers there lies a more
important puzzle about India's data that is worth flagging. The
puzzle is why the ratio of the survey mean consumption to consumption
in the national accounts is so much lower for India than China.
Indeed, the discrepancy is larger for India than most other
That is the real puzzle in Bhalla's numbers: why does household
consumption from India's National Sample Surveys only account for
about 60 per cent of private consumption in India's national accounts
in the 1990s, while for developing countries as a whole the ratio is
closer to 95 per cent on average. This is hardly the product
of "politics and intrigue" at the World Bank and UNDP. It is
something coming out of the data produced by India. But it is
something we need to understand better.
However, this discrepancy between India's national accounts and
India's sample surveys does not mean that the World Bank's poverty
numbers are wrong. As I have argued, there is no basis whatsoever for
Bhalla's assumption that the national accounts are right about mean
household consumption and the surveys are wrong.4 And even if one
accepted his implicit assumption, there is no valid basis for then
replacing the survey mean by the national accounts mean, and then
recalculating poverty using the consumption shares from the same
surveys. If the surveys get the mean so horribly wrong, how could
they get the shares right? How could the surveys be accurate for
measuring inequality but way off the mark for measuring poverty? The
only case in which that would happen is if the discrepancy is due to
underestimation of consumption in surveys and that the proportionate
error is the same at all levels of living. Yet this is unlikely from
what we know about the sources of the discrepancies between these
data sources. More plausibly, the surveys are underestimating
inequality, which throws a spanner into Bhalla's simple arithmetic.
None of this is news to the experts within the international agencies
and the academics and other researchers who work on this topic,
including in India. Nor is it something that has been ignored or
hidden as part of some conspiratorial intrigue. There are economists
and statisticians doing serious work on the problems in surveys,
national accounts and price data. It remains to be seen whether that
work suggests that there is less poverty in India than we thought.
But that is certainly not established by Surjit Bhalla's statistical
[These are the views of the author, and should not be attributed to
the World Bank or any affiliated organization. Address for
1 See `Estimates of Food Consumption from Household Surveys and
National Accounts' by A C Kulshreshtha and A Kar, at
2 For further discussion see K Sundaram and Suresh Tendulkar, `NAS-
NSS Estimates of Private Consumption for Poverty Estimation',
Economic and Political Weekly, January 25, 2003.
3 See my paper, `Measuring Aggregate Welfare in Developing Countries:
How Well do National Accounts and Surveys Agree?', Review of
Economics and Statistics, Vol LXXXV, August 2003, pp 645-52.
4 This point is discussed further in my article, `Have We Already Met
the Millennium Development Goal for Poverty?' Economic and Political
Weekly, November 16, 2002, Vol 37, No 46.