Measuring poverty dynamics without (actual) panel data: Could we square the circle?

|

This page in:

Motivated by the success of the Millennium Development Goal that saw the global poverty rate in 1990 halve before 2015, the international community has multiplied its efforts to reduce poverty further. For example, the World Bank recently raised the bar by proposing that the global extreme poverty rate be reduced to 3 percent or less by 2030. This ambitious goal would no doubt require stronger efforts by all stakeholders on every battle front of poverty reduction, including the (perhaps less glorious) one of poverty measurement. 

Analyzing dynamic aspects of poverty is important for a better and more efficient design of poverty reduction policies. Without specific insights into the dynamics of poverty, the same poverty rate of, say, 20 percent in two periods could well mask dynamic processes ranging from complete stagnation (i.e., zero mobility where all households see no change in their welfare) to extreme volatility (i.e., perfect mobility where all poor households in the first period escaped poverty and were all replaced by households that had previously been non-poor in the first period). Policies to deal with these situations are clearly very different. If poverty is mostly transitory, attention should perhaps be focused on designing safety net programs to prevent non-poor but vulnerable households from falling into poverty.  On the other hand, if poverty is mostly chronic, attention could perhaps better be directed to structural and longer-term interventions such as investment in human capital and building infrastructure. Transitory and chronic poverty typically require different policy instruments and no single policy is likely to successfully address both. However, the acute shortage of panel household survey data in most developing countries has rendered such analysis very difficult to undertake. 

To overcome this obstacle, we have proposed new statistical methods (Dang, Lanjouw, Luoto, and McKenzie, (2014), Dang and Lanjouw (2013)) to construct synthetic panels to measure poverty dynamics. A major contribution of these methods is that they can be applied in a wide range of data-scarce contexts as long as two comparable rounds of cross sections are available. (I have another post on this related topic of measuring poverty when survey data are not comparable over time). Various validation exercises done by us and by other colleagues suggest that poverty estimates based on these synthetic panels approximate quite closely those based on true panel data across a number of countries in geographically diverse locations and at different income levels. In fact, synthetic panels may even provide higher data quality and more reliable estimates under certain conditions. 

These methods have also been increasingly embraced in regional flagships reports at the Bank, including a 2012 Latin America and Caribbean report on economic mobility and the middle class anda 2015 South Asia report on inequality, and other ongoing regional and country reports. Researchers outside the Bank have started to employ these methods for different purposes such as poverty measurement and program impact evaluation. The availability of synthetic panels may also offer opportunities to study the welcome dynamics in other areas besides poverty such as labor transition or international trade in the absence of panel data. 

Our recent study of poverty dynamics in Senegal—where no actual panel data exist—can provide an illustration of these methods. Estimates using two cross sections of the nationally representative household survey (ESPS 1 and 2) suggests that poverty slightly decreased from 48 percent in 2006 to 47 percent in 2011. But this change is not statistically significant, implying that poverty remained largely unchanged over time. When the distributions of household consumption for the two years are plotted against each other, they appear rather stable over time, with the distribution in 2006 being almost identical to that in 2011. Thus estimation using the cross sectional data alone may lead us to conclude that there was not much going on with households’ poverty status in Senegal over this period. 

However, analysis using the synthetic panels points to the contrary. Beneath the relatively stable poverty rates estimated with cross sectional data, there turns out to have been much “churning” out of as well as into poverty in Senegal in this same period. Poor households were somewhat more likely to escape poverty (45 percent) than non-poor households were likely to fall back into poverty (40 percent), and were more likely to have stronger consumption growth as well. Overall, the population was rather mobile with more than half (53 percent) of the population experiencing either upward or downward mobility.  In particular, more than two-thirds of the extreme (food) poor (i.e., people living under the food poverty line) moved up one or two welfare categories.

We further examine the characteristics of households in poverty transition and provide a summary of the results in two graphs below. Factors such as education achievement, having a female household head, urban residence, and formal work contract are strongly correlated with poverty reduction (Figure 1). As discussed earlier, this suggests that addressing chronic poverty might be most effectively achieved via a combination of interventions that lift long-term income prospects and that promote economic growth in sectors where the poor are active, such as agriculture. Measures that promote school attendance by children in these chronically poor families as well as building and maintaining rural roads are also important in this respect.
 
Escaping Poverty in 2011
Figure 1: Profiling of the Probability of Different Population Groups Escaping Poverty in 2011 Conditional on Being Poor in 2006, Senegal

On the other hand, the correlates that help prevent households from falling into poverty include education achievement, work contract, urban residence, having a female head, migration, belonging to other ethnic groups, working in the non-agriculture sector, social capital, having no disaster, and having no disability (Figure 2).  These factors can be included as elements of targeting programs that aim at providing social protection. 
 
Falling Back into Poverty in 2011
Figure 2: Profiling of the Probability of Different Population Groups Falling Back into Poverty in 2011 Conditional on Being Non-poor in 2006, Senegal  

Each country (or region) with its own specific socio-political economy structure may reveal an entirely different set of issues associated with poverty dynamics. Since one of the strengths of these methods is their relatively light data requirements, we hope to expand this line of research more broadly to other similar contexts.

Authors

Hai-Anh H. Dang

Economist, Poverty and Inequality Unit, Development Research Group, World Bank

Join the Conversation

Prakash Sapkota
January 20, 2015

Very robust analysis.

Anonymous
January 21, 2015

i still have a doughty how it can be possible to measure dynamic poverty impact without panel data, particularly in a cross country analysis.