This paper aims to clarify the confusion over estimates of poverty in Pakistan1
. The paper highlights the root causes of
the confusion in the existing literature, which are based on estimates from the “nationally representative” data collected by
the Federal Bureau of Statistics Household Integrated Economic Surveys (HIES). The paper uses the latest available HIES
2010-11 to illustrate and clarify these issues.
The poverty estimates for Pakistan have shown a remarkable and consistent decline in the headcount of the poor with
official – not formally released – estimates showing the incidence at 12 per cent for 2010-11 – down from 34.5 per cent in
2001-02. This decline in poverty is inconsistent with the observed lack of improvement in various non-money metric
measures of welfare and the generally poor macroeconomic performance.
Increasingly governments and especially multilateral lenders measure economic performance in terms of poverty reduction.
Lenders use these as key criteria for continued lending.2 More importantly, the formulation of social protection and other
anti-poverty programs and assessment of their implementation and eventual impact depend upon these numbers.
The official poverty line for Pakistan, which was estimated in 2001-02 based on data from the HIES of 1998-99, has
been adjusted for inflation using a CPI-based extrapolation for the subsequent years. This method suffers from five problems:
1) outdated sampling frames underlying the HIES and estimation of national poverty line; 2) changes in the underlying
consumption basket due to price fluctuations, which are not adequately represented in the CPI and hence in poverty estimation;
3) sensitivity of poverty estimates to the choice of consumption basket on which the poverty line is based; 4) differences
in consumption patterns across rural and urban areas of all provinces and poverty estimation; 5) sensitivity of estimates to
caloric threshold underlying the poverty lines.
The first problem arises from the sample frame that draws on the 1998 Population Census, with some subsequent but
not publically known adjustments, and on which the HIES surveys are based. This leads to a serious underestimation of the
overall population. Moreover, according to its own report the HIES does not cover transitory populations, presumably also
excluding peri-urban localities (where the majority of the poor are now living in Pakistan), military cantonment areas, Azad
Jammu Kashmir, the FATA and FANA. There are thus serious concerns about the representativeness of these numbers.
The second problem results from the likely change in the consumption basket due to the rising trends in food and fuel
prices; especially following the price hike of 2006-083
. The CPI does not capture these changes adequately because of two
reasons: 1) it is based only on urban prices; and 2) the share of food groups is based on the Family Budget Survey 2007-08
which understates by 9 percent the overall weight of the food category in total consumption as compared to the HIES
for the same year. Details of the issues with the current CPI of Pakistan are discussed in a separate discussion paper (Malik
et al., 2014).
The third problem is the sensitivity of poverty estimates to the choice of consumption basket. The selection of consumption
basket to estimate poverty lines is not uniform across studies conducted for Pakistan. For example, while the official
poverty line is based on the consumption basket of households that fall in the bottom three quintiles (i.e., bottom 60 percent
households) (Government of Pakistan, 2003; 2009), Jamal (2012) uses the consumption basket of the households that
belong to the bottom expenditure quartile only (i.e., the bottom 25 percent households). Both sources do not make this
assumption very clear in their write-up, thereby seriously complicating comparison of estimates across these sources.
The fourth problem is that the existing poverty estimates assume that consumption patterns across the country are similar.
However, the data of HIES show significant differences in consumption patterns, not only across provinces, but also
across urban and rural areas of these provinces. These consumption patterns are reflected in the dietary intake and hence in
calorie consumption that is an important indicator to measure poverty. The fifth problem arises from the minimum threshold level of calories4 selected to estimate the poverty line. Several
studies conducted for Pakistan before 1998-99 used a level of 2550 per day per adult equivalent, recommended by the
Nutrition Cell of the Planning Division of Pakistan (Khan and Khan, 1990). However, in 2002, the Planning Commission
reduced the threshold to 2350 per day per adult equivalent5
. In 2011, based on the dietary guidelines of FAO/WHO, the
Nutrition Division of the Planning Commission recommended a least cost consumption basket providing minimum average
energy of 2150 calories and about 65 gm of protein per day per person. This consumption basket was used in the Five Year
Plan 2001-05 and Medium Term Development Framework (MTDF) 2005-10 (for details, see Government of Pakistan, 2011).
Poverty estimates are highly sensitive to the minimum threshold level of calories and can change the estimated values
significantly.
Using the latest available primary survey data from the HIES for 2010-11, this paper tests the sensitivity of the poverty
estimates to the choice of consumption basket, the minimum threshold level of calories, and the assumptions based on
which these threshold calories are estimated. In addition, it demonstrates the spatial differences in poverty estimates across
rural and urban areas.
The paper is divided into six sections. The trends in official poverty estimates and a comparison of these estimates with
those from Jamal (2012) are presented in Section 2. Section 3 is divided into five parts based on the econometric testing of
the five problems identified above. Discussion on the outdated sampling frames of HIES and poverty estimation is presented
in section 3.1. Section 3.2 describes how underestimation of CPI leads to an underestimation of the poverty line. Section 3.3
presents analyses of the sensitivity of poverty estimates to the choice of consumption basket. A detailed analysis of the
differences in consumption patterns across rural and urban areas of all provinces and poverty estimation is presented in
section 3.4 and section 3.5 measures the sensitivity of estimates to caloric threshold underlying the poverty lines. Conclusions
and policy recommendations are given in the final section.