Who became poor, who escaped poverty, and why? Developing and using a retrospective methodology in five countries

Abstract

The Stages-of-Progress methodology helps identify context-specific reasons associated with households' movements into or out of poverty. Developed in 2002, it was used over the next seven years for examining the experiences of 35,567 households in 398 diverse communities of India, Kenya, Uganda, Peru, and North Carolina. This essay looks at the reasons that motivated the development of a different methodology for exploring poverty flows, explores the steps involved, and briefly presents key results. Large numbers of households have fallen into poverty in every context examined, but large numbers have also become persistently poor. Different reasons are associated, respectively, with escaping poverty and falling into poverty. Different policies are, therefore, required to deal with each of the two poverty flows. © 2010 by the Association for Public Policy Analysis and Management.

Formulating more effective policies for poverty reduction is assisted by having answers to two sets of questions. The first set of questions is concerned with the beginnings of poverty. How do people come to be poor in the first place? How many are born poor, and how many others fall into poverty? For what reasons do people experience descents into poverty? What should be done to reduce the number of descents? The second set of questions is concerned with how poverty ends. Why are some (but not other) people in a country or region able to escape poverty? What personal characteristics, household strategies, and forms of external assistance have assisted escapes from poverty in the past?

The basis vectors of poverty—the reasons why poverty is given rise and how it is brought to an end—must be identified more precisely before programs of assistance are given shape. Unfortunately, given the data at hand and the methodologies available, it has been difficult to obtain valid answers to these questions, especially in developing countries, where the need is the greatest.

No nationally representative panel data sets containing the required information are presently available for any developing country.11 Smaller-scale panel data sets have been put together, usually covering quite short intervals of time, but they do not provide information about household-level processes and events. Looking at these data, one can tell how many households escaped from (or fell into) poverty, but it is not possible to deduce the nature of factors responsible for their rise or descent. Households' event histories have not been compiled so far by panel data studies. Usually, only consumption (or income) data are available for two points in time. Thus, because virtually “no attention is focused on the events which lead people into and out of poverty, it is very difficult [using these methods] to trace the processes whereby people may suddenly or gradually escape poverty” (Bane & Ellwood, 1986, p. 4). Other methods of examining poverty have also been implemented. For example, a great deal of useful knowledge has accrued from participatory poverty assessments and ethnographic inquiries,22 but dependable answers to questions related to the beginnings and the ends of poverty have remained elusive.33

It becomes necessary, therefore, to invest in primary data collection. Relatively long periods of time need to be considered. Households' plans for upward mobility are typically framed in terms of generational time horizons; for instance, parents invest in education so that their children's economic prospects can improve in the distant future; farmers labor for years on their agricultural fields, leveling land, digging wells, and constructing irrigation canals, so that low-value crops can eventually be replaced by higher-value ones.44

New panel data sets can be constructed in order to examine these longer-term effects, and several such efforts have commenced recently.55 Because of the need to examine longer periods of time, however, the results of such prospective studies will only become available many years in the future.

A retrospective method, collecting data through recall, can provide more immediate results. However, recall can be tricky. People can forget things, especially over longer periods of time, and memories can be selective, resulting in the production of partial or biased information. A great deal of caution is necessary, therefore, while working with a retrospective method. Not everything that is important to know about poverty can be reliably learned using such methods. But knowledge about micro-level reasons associated, respectively, with escaping poverty and becoming poor, can be gained dependably using Stages-of-Progress, a retrospective methodology that I developed beginning in 2002. The space available does not permit full development and support of the methodology and the results obtained, but the process that led to the construction of this methodology, the steps involved in implementation, and some key results are briefly discussed below.66

DEVELOPING A METHODOLOGY

Between 2001 and 2002, I spent several months in a group of villages in the state of Rajasthan in central India,77 experimenting with different survey tools and poverty measurement methods, but these tools, based as they were on measuring consumption expenditures (or incomes), were hardly ideally suited for the purpose of recall. Consumption expenditures of the distant past are hard to bear in mind—one does not usually remember how many hamburgers one consumed or how many gallons of gas one purchased during some particular fortnight 10 years ago—but assets, being lumpy and more easily distinguished, are easier to recall. (It is more likely that one remembers whether or not one owned a television set, a car, or a motorcycle 10 years ago.) Thus, in order to retrospectively track people's well-being levels over relatively long periods of time, a measure based on assets (or on Amartya Sen's notions of capabilities)88 is a handier tool.

During my extended stay in the villages of Rajasthan, I learned that people have their own robust ways of assessing poverty, based on successively acquired capabilities and assets. Putting food on the table on a regular basis is the most basic capability, followed by the ability to send children to primary school, performing essential house repairs, and retiring debts.99 Asset acquisition comes later in this progression; biologically necessary and socially incumbent needs are satisfied first. After households have gained the capacity to feed themselves, send their children to school, and repay what they owe to others, the next stages of progress involve acquiring cattle, farm equipment, additional cultivable land, and other material assets.

A clear sequence of stages of progress was related by village respondents. Physiological needs (such as food and shelter) came first, socially dictated needs (such as wearing presentable clothing and repaying debts) came next, economic development needs were third in order (including cattle and land), and discretionary—or in this context, luxury—needs (such as acquiring a television set, a motorcycle, or jewelry) came last in this order of improvement. Not all households move up the economic ladder following exactly the same stages of progress. For example, some households have no children, so the capacity to send children to school is not important for them. But these well-recognized stages of progress serve, nevertheless, as a yardstick or set of markers for assessing households' relative well-being in a community setting with the participation of the people involved.

Adjoining communities shared essentially the same progression of stages. In community meetings held successively in 36 villages of Rajasthan, the same sequence of stages was narrated by the assembled community group, and the same definition of poverty prevailed. Something underlying, something real, was being uncovered by this process. Most important, it seemed to provide a promising way of identifying a household's economic status through a process of participatory recall.

Over the next several months, a methodology, described below, was built up that draws on these robust local understandings. A number of safeguards and verification procedures were incorporated as this methodology was successively implemented in 400 diverse communities in India, Kenya, Uganda, Peru, and North Carolina (USA). Each time, I worked alongside local scholars and area-based NGOs. Together, we examined the pathways traversed by a total of 35,567 households, conducting detailed interviews with a random sample of 10,142 households.

In only one case (Kenya), where a larger pot of funds was available, did the selected sample of communities represent the entire country in a statistical sense. In other cases, specific parts of a country were examined, as reported below, and communities in these parts were selected deliberately to explore a range of variation. Small communities were studied together with larger ones; communities closer to market towns and major roads were selected along with others that are more remotely situated; and communities with mixed ethnic compositions were selected together with others where a single ethnic group dominates. Thus, these results, while illustrative and robust, are not statistically representative of the entire region, far less of the country concerned.

In each separate region we conducted investigations in the local language, so different teams of investigators were selected and trained separately. Up to three teams operated in tandem after training. Typically, each team was composed of six to eight individuals and an equal number of males and females.

Because so much depends on the quality of interviewing—and on combining carefully the results obtained separately from household interviews and community groups—training is a very important aspect of this methodology. Training for a period of two weeks or more was built in at the start of this exercise in each study region. Following three days of classroom discussions and simulations, the study teams would go out to conduct a practical exercise, first in one set of communities and then, following feedback and discussions, in a second set of communities. I remained with the study teams for additional periods of up to four weeks, working with them and watching them as they worked, and developing in discussion with them further refinements to this set of methods.

Before reviewing the steps in this methodology and the results obtained, two caveats are in order. First, the Stages-of-Progress is not equally useful for examining diverse questions related to poverty. In particular, since it works with context-specific measures of poverty, valid cross-country comparisons of the numbers in poverty cannot be generated. Second, since it utilizes an assets- and capabilities-based index of poverty, estimates of poverty produced with the help of the methodology are not directly comparable with those obtained from consumption- or income-based measures—although, in some cases, these two sets of estimates are fairly close, as we will see below.

As Blank (2008) points out, however, “there is no ‘right’ way to develop poverty thresholds or resource measures,” so instead of getting mired in fruitless debates, analysts should focus more closely on “progress (or regression) over time, and this may be more important than the precise level of poverty at any point in time” (pp. 243, 252). By focusing on the dynamic aspects of household poverty, the Stages-of-Progress methodology helps fill important gaps in current knowledge, reliably identifying context-specific factors associated with escaping poverty and becoming poor.

THE STAGES-OF-PROGRESS METHODOLOGY

This methodology has seven successive steps, followed in order each time a study is conducted within any community. Given limitations of space, these steps are outlined below, and procedures that were implemented to verify the robustness of the results are presented briefly in the following section.1010

Step 1. Assembling a representative community group. A representative community group was assembled at the start, including males and females, higher-status and lower-status community members, and older as well as younger members. Advance notice had to be given in some instances to the community and its leaders. Investigations were conducted during the lean period of the seasonal work cycle, so that community members could more easily spare the time required to participate in these discussions. In situations where women let men do all the talking in mixed groups (for example, in some parts of India), a parallel meeting was convened for the women of the community. Information obtained from these women's meetings served as a useful cross-check on the information obtained from the men's groups.

Step 2. Presenting our objectives. Next, the objectives and procedures of the study were clearly described to the community group. It was important that the group understand that no tangible benefits would accrue to anyone from participating (or not) in the study.

Step 3. Describing “poverty” collectively through identifying the stages of progress. This step was the most important one in this study. It builds on the key realization, gained earlier, about sequentially acquired material capabilities. Community groups were asked to delineate the sequence in which assets and capabilities were typically acquired as households made their ways out of extreme poverty.

“What does a household in your community typically and usually do,” the assembled community members were asked, “when it climbs out gradually from a state of acute poverty?” Which asset or capability is acquired first? “Food”—or rather, the capability to acquire food on an ongoing basis—was the answer given invariably in every community studied. What follows immediately after? “Sending children to primary school,” was the invariable response in communities of Rajasthan.

As more money flows in incrementally, what does the typical household do in the third stage, in the fourth stage, in higher-level stages? Remarkably similar stages of progress were reported by different communities of Rajasthan. Particularly at the lowest stages, when households are still desperately poor or just coming out of dire poverty, there were no differences in the sequences narrated by community groups of 36 villages. The first four stages, in particular—having food on a regular basis, sending children to primary school, possessing clothes to wear outside the house, and retiring debt in regular installments—were common not just to different villages. They were also commonly reported by the men's and women's groups that were organized and consulted separately.1111

“After crossing which stage is a household no longer considered poor,” the assembled community members were asked after the progression of stages had been drawn up. This cutoff was invariably drawn by all Rajasthan village groups after Stage 4, indicating a common social construction of poverty.

What these villagers identified as their stages of progress makes a powerful statement about what constitutes poverty in their understanding: You are poor when you cannot afford to send your children to school (even when such a school charges no tuition fee). You are poor when your creditors no longer press for payment. You are poor if when it rains you get wet inside your house. And if you are not capable of obtaining food regularly then, of course, you are the poorest of the poor, unlikely to survive for very long without outside assistance. In fact, very few households were found to be located at Stage 1.

This process of identifying successive stages of progress worked equally well within each of the other regions studied. People everywhere, including in North Carolina, had little hesitation in reconstructing their stages of progress. There were hardly any disagreements about the successive stages of progress, particularly the initial ones, even though the process of identification took longer in some community meetings.

Somewhat different stages of progress were reported by the communities studied in different regions and countries. Within each region, however, the same initial stages of progress were commonly identified across all communities. Table 1 reproduces the successive stages of progress reported by rural communities of western Kenya.1212

Table 1. Stages of progress (western Kenya).

1. Food

2. Clothing

3. Repairs to house (primarily thatched roof)

4. Primary education for children

5. Purchase chickens

6. Purchase a sheep or goat

––– Poverty Cutoff: Beyond this line, households are no longer considered poor.

7. Purchasing local cattle

8. Improvements to housing, furniture

9. Secondary education for children

10. Buy or lease land

––– Prosperity Cutoff: Beyond this line, households are considered relatively well off.

11. Purchasing dairy cattle

12. Buying land/plots

13. Constructing permanent houses

14. Investing in a business

Notice how the poverty cutoff is drawn after Stage 6 in this case. Possessing chickens or sheep and goats is not enough; some cattle must also be acquired in order that one be regarded as nonpoor by one's neighbors. Notice also that an intermediate category exists between the categories of “poor” and “prosperous.” Additional progress needs to be made—another four stages need to be covered beyond the poverty cutoff—before a household is considered to be prosperous or rich.

Table 2 reproduces the sequence of stages reported in all of the other regions studied. The constitution of poverty as locally understood—in terms of sequentially acquired assets and capabilities—was identical within each particular socio-cultural context. Across regions as well, there are considerable similarities in terms of these local depictions of poverty, with understandable differences arising in the case of communities in North Carolina.

Table 2. Initial stages of progress (before the poverty cutoff).

Stage	Peru (Cajamarca and Puno)	Western Kenya	Uganda (West and Central)	Andhra Pradesh, India	Gujarat, India	Rajasthan, India	North Carolina, USA
Note: The dotted line corresponds to the poverty cutoff in each case. Households advancing past this threshold are no longer considered poor, either by themselves or by others in their communities. For North Carolina, the first dotted line, below Stage 4, denotes the Very Poor category. The second dotted line, below Stage 8, denotes the Poor category. No separate Very Poor category was identified in the other regions.
1	Food	Food	Food	Food	Food	Food	Basic Shelter
2	Clothing	Clothing	Clothing	House repairs	Clothing	Primary education	Food
3	House repairs	House repairs	Primary education	Debt payments	Primary education	Clothing	Transportation—gas money, cheap used car
4	Purchase small animals	Primary education	House repairs	Clothing	Debt payments	Debt payments	Clothing
5	Primary education	Small animals			House repair/roof	Telephone
6	Purchase small plot of land				Renting in a small tract of land for sharecropping		In-home entertainment, usually TV
7							Better car
8							Debt payments

A commonly known and widely agreed-on understanding of poverty exists in each case, and this everyday understanding of poverty is much more real for these residents than any definition proposed from the outside. These locally constructed understandings of poverty constitute the criteria within these communities for identifying who is poor. They also give shape to the goals and the strategies of poor people: What people do in order to deal with poverty depends on what they understand to be the defining features of this state.

Step 4. Treating households of today as the unit of analysis, inquiring about households' poverty status today, 10 years ago, and 25 years ago. Based on these well-defined and clearly understood criteria, community groups could classify which households are poor at the present time and which households were poor in the past.1313 Before starting with this classification exercise, a large chart recording the identified stages of progress was prominently displayed. A complete list of all households in each community was also put on view.

Implementing a study with a longer time horizon was feasible in rural areas, where people have lived together over multiple generations and can more easily recall and verify important events transpiring in each other's households. In rural communities a recall period of 25 years was generally considered. In urban communities, which are more fluid in terms of composition, a shorter recall period of 8 to 10 years was adopted.

Referring to the shared understanding of poverty developed in the previous step and consulting the publicly displayed stages of progress, rural community groups identified each household's status in terms of the specific stage where it was located at the present time, 25 years ago, and at an intervening point, 8 or 10 years ago. In order to denote the earlier periods clearly, reference was made to some significant event that is commonly remembered. For instance, in India, reference was made to the national emergency of 1975 to 1977, which is clearly remembered, particularly by older villagers. For each household in every community, three measurements were recorded in this manner, relating to its material status, respectively, 25 years ago, 8 or 10 years ago, and at the time of study.

Step 5. Assigning households to particular categories. After ascertaining its poverty status for the present time and for a previous period, each household was assigned to one of four mutually exclusive categories:  

Category A. Poor earlier and poor now	(Remained poor);
Category B. Poor earlier but not poor now	(Escaped poverty);
Category C. Not poor earlier but poor now	(Became poor); and
Category D. Not poor earlier and not poor now	(Remained not poor).

A residual category, E, was also defined, and households that could not be placed in any of categories A through D because of lack of information were assigned to Category E. In reality, very few households, less than half of 1 percent in all, were required to be placed in this residual category. People in communities studied in developing countries performed this categorization without apparent self-consciousness or stigma. In North Carolina, where poverty has a different social connotation, people were less forthcoming with information about each other's situations. The methodology had to be adapted, as discussed below, to cope with this different social situation.

Step 6. Inquiring about reasons for escape and reasons for descent in respect to a random sample of households. Next, a random sample was drawn, consisting of 30 percent of all households from each of the four categories. Detailed inquiries were conducted with the assembled community groups about the contributory factors that were associated with each household's trajectory. A comparative perspective was adopted while inquiring about processes and events. For example, it was not sufficient that someone suggested “bad luck” or “God's will” as a reason for falling into poverty. “How did God's will manifest itself?” members of the study team probed. “What happened in this household that did not also happen in other households of this community?”

Training was central for developing this ability to record event histories accurately and comprehensively. The research assistants were trained to reconstruct events in the same sequence as they had occurred. “Did they sell their cattle first, or did the father die first?” “What happened in the beginning that put this household on a downward path? What happened next? What did they do or what befell them that led over time to their particular trajectory? What else occurred that helped improve their situation?”

Step 7. Following up by interviewing household members. After ascertaining event histories for each selected household from the community group, further verification and more complete accounts were obtained from individual members of the households concerned. The goal here was to delve in more detail into processes and events associated with this household's movement (or stability) and to cross-check the information provided by the community group.

At least two members of each household were interviewed separately in their homes. As a practice, women investigators interviewed female members of these households, and male investigators interviewed the men. A comparative perspective was retained in these interviews, which elicited through gentle probing the sequence of events.

Multiple sources of information were consulted for ascertaining the reasons associated with each selected household's experiences. Information, particularly about events and processes, was triangulated in this manner, before being recorded, coded, and analyzed. Discrepancies brought forth repeat interviews; community groups and household members were consulted multiple times until their accounts agreed, but such differences were infrequent; for instance, repeat interviews were required for fewer than 20 households among more than 1,500 interviewed in Rajasthan.

Supplementary questionnaires helped collect additional information about household characteristics and other items of interest. Asset holdings and education levels were ascertained in every region studied. Further inquiries were made in specific regions about livestock holdings, community development benefits, and participation in democratic activities.

It took the team of six to eight individuals about three days on average to complete these inquiries within one community, constituted by about approximately 175 households in rural areas and smaller numbers in urban neighborhoods. Preliminary results, developed soon after these inquiries were completed, were shared in every case with members of the communities concerned.

For ease of data entry and to facilitate more quantitative forms of data analysis, the reasons—processes and events—recorded in each case were coded using a code sheet that was developed and finalized toward the end of the training period. Examining the details of the recorded life histories provides a qualitative lens on this information, and statistical analyses using the coded data help add a quantitative dimension.

These studies, conducted across diverse regions of five separate countries on four continents, provide the first large-scale examination of household strategies and household-level processes and events. This information was collected and compiled not only for households that escaped from or fell into poverty, but also for households that have remained poor or non-poor, thereby enabling comparisons to be made.

In relation to the discussion about absolute versus relative measures of poverty,1414 Stages-of-Progress can be conceived of as providing subjective assessments of an absolute measure related to a particular context. The fact that poverty thresholds change across contexts—and could, conceivably, change over time within the same context—indicates that this measure is sensitive to general levels of wellbeing and economic progress.

The Stages-of-Progress methodology has been adapted for use by other scholars and by several practitioners.1515 Several years after Stages-of-Progress was developed and made public, the World Bank developed a methodology of its own, which it has called Ladder of Life. This latter methodology borrows extensively from Stages-of-Progress, which is gratifying; the sequence of steps and the core logic are nearly identical.1616 However, it does not appear to have incorporated the safeguards and triangulation procedures that are essential for a retrospective methodology.

VERIFICATION AND SAFEGUARDS

Observing common stages of progress across different communities of the same region provided some initial ground for believing in the validity of this methodology. Some other aspects of the methodology helped add reliability to recall.

Because community groups were asked to retrace large rather than small steps, and because each such large step was associated with a clear and visible empirical referent, community members could more easily discuss and verify among themselves the status of particular households at different points in time. Questions were not phrased in terms of “better or worse” but rather in terms of discrete and verifiable capabilities, prominently displayed as the stages of progress chart.

Biased estimates could have arisen on account of several factors, including migration, stigmatization, and elite domination. Briefly, I describe below how each of these sources of bias was dealt with and why the remaining biases are not likely to be large.

Migration has certainly occurred in the communities examined. However, relatively few entire households have moved out (or moved in). Individual members of households, particularly younger males, have left for cities in search of better economic opportunities, but very few members have left permanently, and fewer still have departed permanently along with their families. Investigations conducted in other developing countries have similarly found that when considered in terms of entire households the proportions involved in migration are much smaller than when considered in terms of individuals.1717

Household composition has been relatively stable in all rural communities studied. Still, an important assumption underlies the choice of present-day households as the units of analysis: Households that died out entirely during the period investigated and others that left the community were implicitly ignored. In a small group of communities these disappearances were investigated, and it was found that they had been undergone by roughly 2 percent of households at either end of the wealth distribution.

The possibility of stigmatization was reduced to some extent, because households were classified in terms of a particular stage (1 to 13) or a particular category (A through D), and there was no need to refer to particular individuals as “poor” or “rich.” As one community organizer involved with the study in Burke County, North Carolina, explained: “It's like you are asking a woman about her dress size: Are you a five or a four? Are you Stage 5, or are you Stage 4? People don't mind talking about these things.” She was right in this regard: People who attended these North Carolina community meetings spoke freely about their own positions along the stages of progress. They were more wary and close-mouthed, however, when someone else's situation was being discussed. It was quite different in communities of the other regions studied, where people spoke more freely and openly about each others' situations, perhaps because poverty is less often considered within developing country contexts as an indication of some individual's personal failure.

Incentives for elite domination were reduced by making it clear at the start of every community meeting that no tangible benefits would be involved. Balance was also maintained in the composition of community groups. In India, for instance, discussions did not begin until lower- and upper-caste people were both present at the community meeting.

Triangulating all information recorded by consulting multiple sources also helped. Information was obtained in the initial instance from community groups, and it was verified later with the individuals and households concerned. To the extent that fear of elites does not also extend into households' private spaces, bias arising on account of elite domination was further reduced.

Despite these and other safeguards, discussed in the larger work, of which this article forms a necessarily abbreviated part, the methodology was not extended much beyond its initial application in India until corroboration with objective data from the past had also been obtained. As part of these verification exercises, a fresh study was conducted in 2004 within a group of 61 villages in Rajasthan where I had conducted a previous study in 1997. This previous study was intended to examine social capital, economic development, and other outcomes.1818 A random sample of individuals in each of these villages had been interviewed in 1997, and information related to a number of different items, including asset ownership, had been collected. The second study, undertaken in 2004, implemented Stages-of-Progress. A seven-year recall period was implemented. Each household's stage was recorded for the time of study (2004) and for seven years ago (1997). These recalled stages were compared with the assets recorded by the 1997 survey. Table 3 presents these results.

Table 3. Stages (as recalled) vs. assets possessed 7 years ago (61 communities of Rajasthan, India).

	Assets Possessed in 1997
Stage for 1997 as recalled in 2004	Land (Bighas*)	Large Animals	Small Animals	Kaccha (Mud) House
* * A bigha is a local unit of land roughly equal to one-fourth of an acre.
Very poor (Stage 1–3)	3.6	1.8	2.8	86%
Poor (Stage 4–5)	5.5	2.5	3.7	77%
Middle (Stage 6–8)	8.1	3.1	5.1	51%
Better off (Stage 9+)	10.6	4.3	3.1	22%

There is a close match between the recall data and the data recorded seven years previously. Households' stages of progress for 1997 (as recalled in the community meetings of 2004) are closely correlated with asset holdings recorded by the 1997 survey. Whether considered in terms of agricultural land, large or small animals, or home construction quality, households that were recalled to have a lower stage of progress actually possessed fewer assets at that time.

Separately, asset holdings over the longer period of 25 years were also cross-checked by looking at official landholding records. Such records have existed unbroken for more than a century in some parts of India, although it is difficult to gain access and to knit together the entire long-period record. For a random sample of households, the landholding record was consulted for the entire period of 25 years. These results are revealing. Households classified within Category C (became poor) by the Stages-of-Progress exercises were found to have lost the largest part of their previous holdings of agricultural land. Households belonging to the other three categories have not lost anywhere near the same proportion of the land that they (or their forebears) had owned in the past. Observing this verisimilitude with objective data collected in the past helped provide additional reassurance.

SOME KEY RESULTS

Four sets of results are worth highlighting, albeit in brief:

• 1.
  In every context examined—no matter how high or low the country's economic growth rate or its per capita income—escapes from poverty and descents from poverty have occurred in parallel.
• 2.
  Most of those who fell into poverty have remained persistently poor.
• 3.
  The frequency of descents into poverty has increased. Compared to a previous period, more people have fallen into poverty recently.
• 4.
  Escapes from poverty and descents into poverty are asymmetric in terms of reasons.

Important policy consequences follow. Moving poor people out of poverty is not all that is necessary. The flow of people into poverty has to be brought under control at the same time. Two sets of poverty policies are required in parallel: one set to address the micro-level reasons associated with descents into poverty, and another set to promote the different set of factors that are associated with escapes from poverty. Reasons for escape and descent tend to vary both across space and over time, so standardized and centralized policies are unlikely to make best use of the resources available. Context-specific poverty policies are required, and these policies need to evolve over time.

I will briefly discuss each of these results below, illustrating each one with examples from some among the different regions studied. Let us begin by considering evidence establishing the ubiquitous simultaneity of escapes and descents.

Parallel Flows

Even though the total numbers of the poor may change very slowly, the composition of the poor is constantly in flux. This basic fact can be overlooked when one considers only national stocks of poverty, but having access to flow data results in changing one's outlook.

Consider the results from the Uganda and North Carolina studies. In Uganda, 36 diverse rural communities were examined that belong to two regions (central and western) out of the five in which this country is administratively divided.1919 Household poverty dynamics were examined for two consecutive periods of time, a 15-year period (1979 to 1994), and the following 10-year period (1994 to 2004). On average in these 36 villages, 45 percent of all households lived in poverty in 1979, 37 percent were poor in 1994, and 35 percent were poor in 2004.

Overall, poverty has fallen consistently over the long period examined, and the average figure for these 36 villages—35 percent in 2004—is the same as the figure provided by the official statistics for the entire country, based on the national poverty line (Deininger & Okidi, 2003). But looking at these changes in the stock of poverty does not tell the entire story.

Even as poverty fell overall by 10 percent over the 25-year period, from 45 percent in 1979 to 35 percent in 2004, a far larger number of households experienced a change in poverty status. Table 4 provides these figures.

Table 4. Trends in household poverty in 36 villages of Uganda (1979 to 2004) (percent of households).

Region	Remained Poor	Escaped Poverty	Became Poor	Remained Nonpoor	Poor 25 Years Previously	Poor at Time of Study
All 36 villages	20.4	24.0	15.0	40.6	44.4	35.4
Central (18 villages)	12.8	28.6	14.5	44.0	41.4	27.3
Western (18 villages)	28.9	18.8	15.5	36.8	47.7	44.4

Of the total of 2,631 households resident in these 36 villages, 20.4 percent were poor in 1979, and they remained poor in 2004; 40.6 percent were not poor in 1979, and they remained not poor in 2004; but 24 percent of all households escaped poverty over this period, while another 15 percent of all households fell into poverty. Poverty flows were copious in both directions: A total of 39 percent of all households moved out of or into poverty. As discussed later, most of those who fell into poverty during the first 15-year time period remained poor at the end of the following 10-year period.

A similar story of simultaneously occurring escapes and descents emerged from the study of 13 North Carolina communities.2020 A 10-year period, 1995 to 2005, was considered for this study. A total of 35 percent of all households in these communities changed their poverty situation over this 10-year period, either by falling into or by coming out of poverty. While 23 percent of households escaped poverty, another 12 percent of households fell into poverty. Overall, 11 percent fewer households were poor in 2005 compared to 1995. It is important to keep in mind, however, that more than three times as many households, 35 percent in all, moved into or out of poverty.

These figures for poverty, derived from a community-based definition, correspond closely to the numbers identified as poor by the official (income based) definition of poverty in North Carolina. Further, the median household income for households found to be in poverty according to the community-based definition was calculated as $15,000. The nationally recognized poverty threshold for a family of three is very close to this figure: $15,219.2121

Movements into and out of poverty in these North Carolina communities were considered separately for two adjoining 5-year periods. Table 5 reports these results.

Table 5. Movements in two subperiods (13 North Carolina communities) (percent of households).

	First Subperiod, 1995–2000	Second Subperiod, 2000–2005
Escaped poverty	16	13
Became poor	6	12
Remained poor	44	27
Remained not poor	34	48

Escapes out of poverty were more frequent in the first subperiod considered, and descents into poverty were more frequent in the second subperiod. Consequently, poverty reduction was faster in the first sub-period compared to the second. Relatively few instances of cycling into and out of poverty were found, quite likely because these measurements relied upon an assets- and capabilities-based index. The vast majority of households who moved out of poverty stayed out during the next subperiod, and the majority of those who fell into poverty remained persistently poor.

Each of the other Stages-of-Progress studies generated a similar conclusion about parallel and opposite poverty flows. Studies that have utilized other methodologies and measurement tools and considered different countries, including both high-income and low-income ones, have also produced quite similar results: Everywhere, poverty is simultaneously both created and reduced.2222

Producing Chronic Poverty

Researchers have directed attention toward two important analytical categories: chronic and transitory poverty, with the former category referring to those who remain poor for long periods of time and the latter category depicting shorter-term residents in the pool of poverty.2323 The Stages-of-Progress data show that the bulk of those who fell into poverty in the past have become chronically poor, and relatively few of those individuals were able to bounce back. The North Carolina data, reviewed above, provided one such indication. Similar results were obtained in the other regions studied.

A micro poverty trap, corresponding to a low-level equilibrium, tends to ensnare households who fall into poverty.2424 High health care costs or droughts force households to sell off their assets. Selling off a few cows and goats may not be so bad over the longer term, but as successively more assets are sold there comes a point when a critical threshold is crossed. Households falling below this threshold of asset ownership find themselves caught within a poverty trap. Their capacity to earn has been impaired so much that it becomes impossible to escape from the micro poverty trap.

Illustrations of micro poverty traps are provided by the studies undertaken in Uganda and Peru. Studying households in 36 Ugandan communities over two separate time periods, respectively, 1979 to 1994, and 1994 to 2004, it was found that 6 percent of all households had fallen into poverty during the first time period. Only one-third of these households (that is, 2 percent of the total number) managed to escape poverty during the second time period. The majority—two-thirds—of those had fallen into poverty during the first period continued to be poor 10 years later. Of 344 households that experienced descents between 1979 and 1994, as many as 24 percent were not capable of obtaining food and clothing on any assured basis in 2004, and another 29 percent had removed their children from schools.

A study of 40 communities in two regions of Peru (Cajamarca and Puno) uncovered similar evidence. Among households that fell into poverty during an earlier 15-year period, the majority were poor 10 years later. Many among them had been reduced to a situation from which recovery seemed difficult, if not altogether impossible.2525

Increasing Vulnerability

There is evidence to show that the incidence of descents into poverty has increased. We saw such evidence above while reviewing results from of North Carolina. Larger numbers of descents were experienced in the second five-year period examined compared to the first five-year period. Similar trends were in evidence in the other regions studied.

For instance, the study in Kenya, undertaken in 2006, selected communities using a process of stratified random sampling. A group of 71 communities (rural villages and urban neighborhoods) were selected that are nationally representative in a statistical sense. The results from this study show how vulnerability has increased appreciably in more recent times (Table 6). In every one of five separate livelihood zones of this country—corresponding to areas with common agro-ecological conditions and livelihood options—considerably more descents into poverty occurred during the later time period compared to the earlier one.

Table 6. Share of households that fell into poverty in different livelihood zones of Kenya.

Livelihood zones	1st Time Period (1990 to 1998)	2nd Time Period (1998 to 2006)
Livelihood Zone 1: Mixed farming high potential—food crops, tea, coffee, sugar cane; livestock. Districts: Nyeri, Kirinyaga, Butere Mumias, and Nandi. Livelihood Zone 2: Mixed farming lower potential—food crops, livestock, and some cash crops. Districts: Kisumu, Busia, and Migori. Livelihood Zone 3: Mixed farming marginal with livestock—food crops, livestock, palm, coffee, pyrethrum. Districts: Laikipia, Makueni, Tharaka, and Kilifi. Livelihood Zone 4: Pastoral—livestock, some food crops. Districts: Marsabit, Tana River, and Wajir.
Livelihood zone 1	9	11
Livelihood zone 2	7	14
Livelihood zone 3	7	13
Livelihood zone 4	8	22
Urban zone	10	15

It is not only in remote or left-behind communities that descents into poverty have occurred in large numbers. In all communities studied—no matter how remotely or how centrally they are located—escapes and descents have co-occurred, affecting households at different points in the wealth distribution. In capital cities and other current-day foci of economic growth, large numbers of people have fallen into poverty. In Nairobi, the capital city of Kenya, and in Mombasa, the second-largest city of this country, 10 percent of households fell into poverty over the eight-year period, 1990 to 1998. Less than one-third of these households were able to bounce back over the next eight-year period.

Evidence of increasing vulnerability has been found as well by scholars who have studied other countries. For the United States, available evidence shows that “poverty entries and exits have changed over the past two decades, with the mid-1990s seeing an increase in both entries into poverty and exits from poverty” (McKernan & Ratcliffe, 2002, p. viii). It was found that “while the gaps between the rungs on the ladder of the American economy have increased, what has increased even more quickly is how far people slip down the ladder when they lose their financial footing” (Hacker, 2006, pp. 12–13).

Large numbers of people fall into poverty with a regularity that is disturbing. There are signs that these numbers might be growing larger still, but that might not be equally true in all places. Everywhere, however, descents are occurring in parallel with escapes from poverty.

These results are at variance with some previous findings—for instance, in the U.S.—which suggest that “most individuals who encounter poverty do so for fairly short periods.” Thus, it is not only necessary that “policies be in place to provide an effective social safety net to soften the blow of poverty when it occurs” (Rank & Hirschl, 2001, pp. 749, 751). It is also necessary that descents into poverty be prevented, to the extent possible, before they occur and become persistent. Unfortunately, little preventive assistance has been available in many contexts, and households facing impending descents have generally been left to fashion their own remedies.

Asymmetric Reasons

Specific policies are required for preventing descents into poverty. Policies that assist escapes from poverty are not equally effective for preventing descents. Escape and descent are not symmetric in terms of reasons. One set of reasons is associated with the experiences of households falling into poverty, but a different set of reasons is associated with escaping poverty.

Reasons for escape and descent were identified by examining the coded event histories of different categories of households. Qualitative examinations were accompanied by quantitative ones. Logistic regression analyses helped compare across the experiences of different categories of households.2626 In general, discrete events are associated with both escapes and descents. Apart from one household characteristic—gender—other characteristics (such as household size, age of household head, and also, surprisingly, education) are not consistently associated with either escapes or descents.

Consider descents first. No single reason is usually associated with falling into poverty; multiple linked factors propel most descents. Table 7 presents these results for a selection of regions examined.

Table 7. Principal reasons for descent into poverty (percent of descending households).

Reasons	Rajasthan, India n = 364	Uganda: Central and Western n = 202	Peru: Puno and Cajamarca n = 252
Note: The numbers in each column add up to more than 100 because multiple causes are usually associated with each case.
Poor health and health-related expenses	60	71	67
Marriage/dowry/new household-related expenses	31	18	29
Funeral-related expenses	34	15	11
High interest private debt	72
Drought/irrigation failure/crop disease	18	19	11
Unproductive land/land exhaustion		8

Ill health and health-related expenses are overwhelmingly the most important reason for households' descents into poverty. These factors were associated with nearly 60 percent of all descents recorded in villages of Rajasthan and, respectively, 71 percent and 67 percent of all descents in Uganda and Peru. Evidence from many other developing countries points unambiguously to the deleterious effects of high health care costs on households' welfare.2727 More than half of all personal bankruptcies in the United States have been attributed to medical costs (Himmelstein et al., 2005).

Social and customary expenses on marriages and funerals constitute another set of factors associated with descents into poverty. Funeral expenses had more pernicious effects in some regions. Marriage-related expenses were more important in other regions. Land-related factors, including crop disease, land exhaustion, drought, and irrigation failure were also associated with a large number of descents. High-interest debt was another significant feature, particularly in India, and to some extent also in North Carolina.

Other reasons for descent, not reported in Table 7 but especially relevant in North Carolina, included the loss of a job resulting from retrenchment or retirement. A total of 66 percent of all households falling into poverty in North Carolina either lost a job or suffered a reduction in working hours. Importantly, each of these households also mentioned at least one health-related problem, such as an illness, an accident, high medical expenses, or disability. Beyond the impact on income, job loss very often implied the loss of health insurance benefits, which in turn increased the vulnerability of the household concerned. A combination of job loss, insurance loss, and serious illnesses was associated with the majority of descents into poverty in North Carolina.

Table 8 provides commonly operating reasons for escapes from poverty in three regions. Income diversification has served as the most important pathway out of poverty in these developing-country regions.

Table 8. Principal reasons for escaping poverty (percent of escaping households).

Reasons	Rajasthan, India n = 499	Uganda: Central and Western n = 398	Peru: Puno and Cajamarca n = 324
Note: The numbers in each column do not add up to 100 because multiple causes are associated with most cases.
Diversification of income (urban informal sector)	58	52	44
Diversification of income (agriculture)	39	41	69
Private sector employment	7	9	8
Public sector employment	11	6	5
Government/NGO assistance	8	3	4

Poor rural households have diversified their livelihood and income sources using two strategies. On-farm strategies include pursuing new crops, new techniques, and new methods of livestock husbandry. Diversification into nontraditional export crops (including vanilla and coffee) was important in both regions of Uganda. Cash crop diversification was also important in western Kenya and in the Cajamarca region of Peru. Off-farm strategies have included local petty trade, small businesses, and, most importantly, casual or temporary employment within the informal sector in a city.2828 In Rajasthan, irrigation and land improvement also served as important reasons for escape.

In general, regular jobs did not serve as the principal, or even a very important, pathway out of poverty in these regions. Government and private-sector jobs were involved in the cases of less than one-fifth of all households who escaped poverty in Rajasthan, and with lower proportions of households escaping poverty in Uganda (15 percent) and Peru (13 percent).

Because formal jobs were not readily available, education did not always produce an escape from poverty. Information and connections matter in addition to education, and the lucky few who have found regular jobs were assisted by having a helpful contact in some city, usually an uncle or a cousin who was formally employed.

Diverse reasons for escape and descent have operated in the different regions studied, so pinpoint rather than blanket solutions need to be devised and implemented. Some policies, such as improved health care, may be better implemented in a centralized fashion. Other reasons for escape and descent, such as a potential for irrigation or local customs related to weddings and funerals, are better addressed with the help of a decentralized response.

A sobering lesson of these studies is that both government and nongovernmental assistance and programs are not contributing substantially to households' movements out of poverty. In fewer than 10 percent of all observed escapes from poverty did the household concerned receive any direct assistance from a government agency or an NGO. Not targeting context-specific reasons for escape and descent is an important reason for this unfortunate (and avoidable) result. More effective NGO and government programs can be designed by first identifying and then targeting context-specific reasons for escape and descent.

CONCLUSION

Instead of being seen in terms of a rising tide, with all boats being lifted equally, poverty is better conceptualized in terms of a different image. Some households are lifted upward by a rising tide, but other households are simultaneously plunged into chronic poverty. Policies intended to move people out of poverty will not suffice to reduce poverty. Unless the flow into poverty is simultaneously controlled, newly impoverished people will simply replace those who move out of the pool of poverty.

Specific policies are required to prevent the growth of future poverty. It seems more fruitful to prevent the creation of poverty in the first instance than to provide assistance only after someone has become acutely and chronically poor.

Ill health and high health care expenses constitute the biggest single reason for descents into poverty. Policies to combat illnesses and the high costs of medical attention are critically important for reducing future poverty. Other region-specific policies are also required.

Localized studies are required for identifying the nature of factors that matter most within each specific context. Community-based methods of investigation, including those grounded in asset-related understandings of poverty, can be helpful for these purposes. The Stages-of-Progress is one such method. By directly examining changes occurring at the grassroots, it helps to understand processes and strategies operating at the household and community levels. Households' pathways over time can be reliably tracked using this method. Reasons for escaping poverty (and becoming poor) can be identified with confidence.

Some remaining issues will have to be tackled by developing additional steps and incorporating other safeguards. First, Stages-of-Progress will need to deal better with intra-household differences. Because it considers entire households as the units of analysis, differences between male and female members are not easily detected using this process. I continue to seek a viable means for incorporating these additional steps within Stage-of-Progress.

Another set of refinements is required for examining newly formed communities. As presently constructed, this methodology works better among longer-standing communities. Some initial refinements were developed for the study undertaken in North Carolina. First, a shorter time period was considered. Second, greater reliance was placed on household interviews. Reasons for escape and descent were ascertained entirely through interviews with multiple household members, because community groups did not possess (or did not provide) this information. Thus, one element of triangulation was not available in North Carolina. Another set of improvements is required on this account.

No one method of investigating poverty can suffice for learning all that is necessary or useful. Thus, Stages-of-Progress is helpful for ascertaining micro-level reasons for escape and descent, but other methods are better for keeping track of higher-level trends associated with international and national events. Combining methods suitably will help understand different facets of poverty. Disaggregating by trend (escape and descent), by reason (why escape and why descent), and by region and method will help us to uncover new policy-relevant facts. Critical for this process will be

a willingness to break down the hierarchical relationship between social scientific ways of knowing and other forms of expertise—to recognize, that is, the legitimacy and importance of knowledge that is grounded in practice, in activism, and in the experience not only of material deprivation but of the everyday workings of the economy … of building long-term collaborative relationships for setting as well as carrying out poverty research—a model that takes the production of knowledge out of or at least beyond traditional or expert venues and into a variety of communities. (O'Connor 2001, pp. 293–294)

Progress in poverty reduction will be better as a result.

• 1

  In their review of developing-country panel studies available at the start of the current millennium, Baulch and Hoddinott (2000, p. 18) comment on the “continuing paucity of longitudinal household studies.” In relation to higher-income countries, as well, “until recently, only a few … (for example, the U.S., Canada, Germany, and Great Britain) had either a short- or a long-term panel. Hence, official poverty statistics in both Europe and the U.S. [have been] based on cross-sectional data” (Burkhauser, 2009, p. 718).

• 2

  For examples of participatory poverty assessments, see Chambers (1995), Narayan et al. (2000), and Salmen (1987). For examples of ethnographic studies, see Attwood (1979), Moser and Felton (2007), and Van Schendel (1981).

• 3

  For instance, Moore, Choudhary, and Singh (1998, p. 6) find that participatory poverty assessments do not “clearly separate (a) the indicators of poverty identified by poor people from (b) their understandings of the causes of poverty.”

• 4

  On the need for examining longer time periods, see, for instance, Haddad and Ahmed (2003), and especially Walker and Ryan (1990).

• 5

  See, for instance, Baulch and Davis (2007) and Quisumbing (2007).

• 6

  For further development of these arguments and a detailed exposition of the results, see my larger study (Krishna, 2010). In its abbreviated form, the argument could at some points appear overly stark. Further substantiation is provided in the larger work.

• 7

  I selected Rajasthan because I have worked in this area before, first as a manager of development programs for the Indian government, and later as a scholar of village economy and politics.

• 8

  In Sen's broad understanding, capabilities are regarded to be the substantive freedoms people enjoy for leading the lives they have reason to value. See Sen (1981, 1984). My narrower and more specific understanding of this concept is explicated below.

• 9

  Housing is not such an immediate need in rural settings because relatively few people are homeless. In urban communities, shelter came immediately after food and clothing in the stages of progress.

• 10

  Readers interested in probing these arguments in more detail are referred to the larger study (Krishna, 2010).

• 11

  Some differences across communities in each region arose in relation to higher-level stages, those achieved long after poverty is overcome, when spending becomes more discretionary, and both human physiology and social imperatives are less prescriptive. Hardly any differences arose, however, at lower-level stages, up to and somewhat beyond the reported poverty cutoff in each region.

• 12

  Among communities in the capital city, Nairobi, the first few stages below the poverty cutoff were commonly reported as food, clothing, renting a small house, primary education for children, and investing in a small business. Different livelihood requirements between rural villages and cities lead to somewhat different social constructions of poverty, yet similarities also exist.

• 13

  Assets are typically owned in common by households; thus, households were selected as the units of analysis. Individuals within households can certainly differ, as I will discuss later.

• 14

  See, for example, the compilation of essays edited by Besharov and Couch (2009).

• 15

  Scholars associated with the following institutions had adapted and utilized this methodology at the time of writing: the World Agro-Forestry Center, the Government of Kenya, the International Food Poverty Research Center, the Food and Agriculture Organization of the United Nations, the International Livestock Research Institute, Uppsala University (Sweden), University of Edinburgh (U.K.), Humboldt University (Germany), and the World Bank. NGOs have also utilized this methodology for diverse tasks related to program development and evaluation, including BRAC (in Bangladesh), CARE and CIP (in Peru), and SEWA (in India).

• 16

  Although no acknowledgment is provided by the World Bank, the direction of influence is obvious. Several journal articles describing Stages-of-Progress had appeared in print well before the World Bank began developing its methodology (Krishna 2003, 2004; Krishna et al., 2004, 2005). I presented Stages-of-Progress at the World Bank in 2003 and 2004 and contributed an essay to the first volume of their thematic series on Moving out of Poverty (Krishna, 2007). The second volume in this series (Narayan, Pritchett, & Kapoor, 2009) unveiled Ladder of Life, which follows a sequence of steps very similar to Stages-of-Progress.

• 17

  See, for instance, Banerjee and Duflo (2007).

• 18

  See Krishna (2002).

• 19

  See Krishna, Lumonya, et al. (2006) for details.

• 20

  These communities were selected within the following counties: Beaufort, Burke, Gates, and Vance. See Krishna, Gibson-Davis, et al. (2006) for details.

• 21

  The average household size in the North Carolina communities was 2.79.

• 22

  For lower-income countries, see, for example, Carter and May (2001); Deininger and Okidi (2003); Dercon and Krishnan (2000); Haddad and Ahmed (2003); Narayan, Pritchett, and Kapoor (2009); and Sen (2003). For higher-income countries, see, for example, Bane and Ellwood (1986); Corcoran (1995); Hacker (2006); McKernan and Ratcliffe (2002); OECD (2001); Sawhill (1988); and Stevens (1999). For a review of recent studies of the dynamics of poverty in the United States, see Cellini, McKernan, and Ratcliffe (2008).

• 23

  See Hulme and Shepherd (2003); CPRC (2005); and Jalan and Ravallion (2000).

• 24

  For more on poverty traps of different kinds, see Bowles, Durlauf, and Hoff (2006); Carter and May (2001); and Carter and Barrett (2006).

• 25

  See Krishna, Kristjanson, et al. (2006) for details.

• 26

  Readers interested in looking at the results of these multiple regression analyses should consult Krishna, Lumonya, et al. (2006); Krishna and Lecy (2008); and Kristjanson et al. (in press).

• 27

  See, for example, EQUITAP (2005); Fabricant, Kamara, and Mills (1999); and Whitehead, Dahlgren, and Evans (2001).

• 28

  Characterized by “easy entry, little unionization, no legal minimum wages, weak safety standards at work, low physical capital inputs, low returns to labor, and mainly small (often family-based) units” (Lipton & Ravallion, 1995, p. 2601), the informal sector is made up of an array of enterprises producing a variety of goods and services. An authoritative review shows how “the bulk of new employment in recent years, particularly in developing and transition countries, has been in the informal economy” (International Labor Organization, 2002, p. 1).