SDRI - The Social Development Research Initiative

Pragmatism Papers 1: Exploring causation in poor people's lives

2010-01-14T07:20:00.000-08:00

This is the text of the first in our series of Pragmatism Papers, a collaborative venture into the philosophical foundations of development research. A fully referenced version is available here. This blog version is for comments and discussion.

Exploring Causation in Poor People's Lives: Towards a Pragmatic Approach
Peter Davis, August 2009

In this think-piece I draw on a mixed-methods study into poverty dynamics in Bangladesh that I have been involved in since 2006 with colleagues from the Chronic Poverty Research Centre and the International Food Policy Research Institute. Mixed-methods research in the social sciences is currently undergoing a revival as researchers recognise how different research approaches can complement each other in helping us understand the social world. In poverty studies in developing countries, which tend to use either a quantitative or a qualitative approach, rather than mixed methods, the potential for integrating both quantitative and qualitative methods is also now much better recognised. However in practice integrated qualitative and quantitative methods are still not particularly well developed.
In research methodology we are accustomed to thinking about ontological ideas, about what is in the social world, preceding epistemological ideas, about how we investigate the social world. Qualitative and quantitative researchers often hold similar views on the nature of poverty, but the types of indicators they use affects the way we think about poverty. So the influence goes both ways. In research practice sometimes our epistemological approach affects the ontological idea of what we are trying to understand. Quantitative researchers tend to think about poverty in relation to poverty lines and in money-metric or other quantifiable terms; qualitative researchers are drawn to exploring poverty in less measurable multidimensional and contextual terms. An integrated approach encourages researchers to consider tacitly held assumptions, particularly when disagreements arise over the assessments of poverty status of particular individuals or households.
A pragmatic approach to ontological and epistemological differences in poverty studies would encourage us to move beyond ideological differences in research methodology to choosing a mix of methods to do something. In our case to try to understand a complex problem like poverty – its causes and possible solutions – so that usable policy-relevant knowledge can be generated. Ontological and epistemological debates about poverty should be conducted while solving concrete and practical research problems, and not in detached hypothetical or abstract terms.
In poverty studies, often qualitative research is used to pilot, or provide illustrative examples for, what is seen as the real research which is based on non-anecdotal quantitative evidence. However true integrated research has more potential than having qualitative piloting/exploratory studies informing, or providing illustrations for, larger propositionally focussed quantitative research projects. In particular qualitative and quantitative research can complement each other in the challenge of attributing causative significance to events and episodes in people’s lives. Understanding causation is important in poverty dynamics studies because we are not just interested in trends (observing movements into or out of poverty) but also in understanding why movements take place – and ultimately in making informed decisions about what can be done in policy terms to support causes of improvement, and protect from causes of decline. It is here that the integration of qualitative and quantitative findings becomes particularly valuable.
Research like this can benefit from the particular strengths of both qualitative and quantitative research and minimise some of the weaknesses. In qualitative research it is often difficult to make general inferences about a population because cases are too few in number to claim representativeness (statistical or otherwise) of the wider population. Quantitative researchers, on the other hand, need to recognise the limitations on grounded understanding and explanation imposed by predefinition of categories and numerical representation of complex phenomena. For example in quantitative survey research it is impossible to change categories on a survey questionnaire once the fieldwork has started which limits its exploratory and explanatory potential. The weaknesses on both sides can be ameliorated by their combination. For example, by nesting qualitative subsamples within larger quantitative samples, information from one side can strengthen or qualify evidence from the other.
The task of explaining particular outcomes in people’s lives, or of explaining collective trends, involves making causal judgements. Some researchers choose to avoid the language of causation because statistical regularities cannot confirm causes or causal mechanisms. It is true that statistical regularities such as correlation do not imply causation, however correlation can provide evidence of a putative causal relationship, and this evidence can be strengthened in combination with other methods.
In research and policy practice often euphemisms are used to talk about causation. When we talk about ‘drivers’, ‘levers’, ‘agency’, ‘determinants’, ‘contributors’ or ‘forces’ with regard to a social phenomenon like poverty, we are talking about causation in other words. Instead of avoiding criticism by eschewing the language of cause-and-effect, pragmatist poverty researchers should address these problems head-on, and by mixing methods and triangulating between findings, improve the evidential basis of causal inferences in poverty dynamics research.
Four different epistemological approaches to understanding causation can be identified in the various parts of a research project I have been conducting in Bangladesh in collaboration with colleagues from the Chronic Poverty Research Centre, the International Food Policy Research Institute and Data Bangladesh (see Table 1 below). In what follows I outline the strengths and weaknesses of each approach and suggest that by combining these epistemological stances we are best positioned to take up the challenge of identifying cause-and-effect relationships in poverty dynamics research.

Table 1: Different epistemological approaches to causation in poverty dynamics research
[Please refer to published version for the table]

An experimental approach to exploring causation
When we talk about causation in the social sciences we are usually referring to probabilistic causation. The link between cause and effect (X and Y) is rarely deterministic nor completely necessary and/or sufficient. In the natural sciences a common way of exploring the relation between cause and effect is by controlled experiments. The nearest analogies in the social sciences are natural or quasi experiments, and randomised control trials (RCT), where a treatment or intervention of some kind (such as a conditional cash transfer) is applied to particular cases and withheld from others (controls). This is a common approach in studying health or education interventions in developing countries, but the contexts in which such experiments are possible, or ethical, are limited, particularly in the study of poverty. In poverty studies it is rarely possible, nor desirable, to have complete control over which treatments are applied to which cases.
In our study three development interventions can be seen as a quasi experiment, where sites were identified where an intervention had taken place, and other sites, where the interventions were absent, seen as controls. These interventions were microfinance, fish farming and horticulture, and educational transfers (food and cash for education). These are described in Zeller et al. (2001), Hallman et al. (2007) and Ahmed (2005), respectively. However in our study, educational transfer and microfinance interventions were rolled out over the years that followed the baselines, so control sites no longer remained without the intervention over the length of the study. This limited the possible inferences that could be made using a purely experimental epistemological stance. Fortunately other approaches could be deployed using the other techniques outlined below.
Philosophers of science and econometricians argue that an experimental approach to identifying causation, where a counterfactual (closest possible world without the intervention) is either created by the researcher, or naturally occurs, is a reliable way of identifying a causal relationship between an independent and a dependent variable, because the effect of the treatment variable can be isolated and other effects controlled. The biggest problem is that true experimental conditions (for various ethical and real world reasons) are rarely possible or desirable in poverty dynamics research.

A statistical (or covariational) approach to exploring causation
A statistical approach to the identification of causation draws from what is sometimes referred to as ‘regularity theory’ which builds on the philosopher David Hume’s idea that causation involves regularity in relations between empirically observed entities, with mechanisms not comprehensible (Marini and Singer 1988). Because causation can be either indeterministic or deterministic at the ontic level, but due to measurement error is only observed as indeterministic in actual research practice, supporting evidence of causation tends to be probabilistic in nature. That is, covariation is observed between causes (X1,2,...n) and an effect (Y) usually expressed in the form of a regression equation. This logic lies behind statistical analyses of using correlations between variables as evidence of causation. However without a ‘plausibility story’ (see Abbott 2001: 132) explaining the mechanism linking a particular variable to another in a causal sequence, and under which particular circumstances, these observations provide only a partial explanation of causal phenomena.

Using participatory counterfactual discussions to explore causation
One way of supplementing observations of regular covariation of variables is to conduct thought experiments concerning particular causes and effects where a ‘closest possible world’ is imagined with or without out the cause. When it is not possible to conduct an experiment creating an intervention situation which can be compared with a closest possible world without the intervention (a control), a counterfactual situation can be imagined. Counterfactual ideas are central to many philosophical ideas of causation. They are put into practice when we conduct qualitative life-history interviews and discuss with a participant why a particular outcome occurred, whether the same outcome would have happened, or how the outcome would have differed, without the putative cause.
For example in life-history interviews in our study we ranked crises and opportunities according to the effect they had had on a person’s present circumstances. In order to rank these we invited participants to think counterfactually, something we all do when we consider life trajectories and the effect of past events on present circumstances. When participants identified events or episodes that made a significant difference for their present situation we invited them to consider how things would be now if the particular event had not taken place. So for example a statement like, ‘if I hadn’t got the job, I wouldn’t have been able to afford medical care for my mother, and she would have died’, contains information about circumstances which draws from an insider’s perspective of the circumstances in which she lives and knows best.

Mechanismic or process tracing approaches to exploring causation
The identification of causal processes or mechanisms within a temporal analysis of events and episodes also helps to strengthen the body of evidence that makes causal inferences more plausible. The term ‘process tracing’ is used by political scientists to describe this kind of activity in political analysis. It refers to peering ‘into the box of causation’ (Gerring 2007:1). Process and mechanisms are seen as the means by which a cause (X1) is seen to produce the effect (Y). This kind of analysis also attempts to uncover the particular circumstances in which the observed causal regularities are likely to recur. For example, a statistical correlation may be observed between low income and poor health. However process-tracing case studies may clarify mechanisms linking low income to ill-health, for example via malnutrition, poor housing, or poor access to clean water, and under which circumstances, such as where cheap and effective health provision is not available. Certainly some of these intervening variables can also be measured and covariations analysed statistically, however plausible mediating links are usually uncovered after within-case analysis.
An analysis of mechanisms may also uncover instances where a causation relationship is reversed: poor health in some cases may cause low income due to physical weakness, inability to retain a job, or increased time spent on caring impeding income earning (see Deaton 2003). Process tracing approaches also allow analyses of causal fields, multiple causation, thresholds and turning points. Some causal processes seem self evident, while others are more complex, and the analysis of causal processes or mechanisms in case studies can help to uncover the most plausible explanations of observed regularities.

Concluding remarks
Within-case analyses of qualitative research (including participant insights), combined with the cross-case analyses of statistical regularities from quantitative research, provides the best chance of reliably uncovering causation in poverty dynamics research. Without case-based research, quant researchers rely on anecdotal ‘plausibility stories’ (see Abbott 2001:132) to explain regularities between variables. This becomes more problematic in quantitative poverty research when there is a division of labour between field researchers and analysts, with analysts not familiar enough with local realities to be able to judge whether their explanations of mechanisms are realistic or not.
In our study in Bangladesh the integration of quantitative and qualitative methods allowed use to uncover a number of neglected areas which could not have been explored so well using either qual or quant methods alone. One example is uncovering the extent to which medical expenses and dowry payments have a long-term impact on poor households. A household survey can pick up these lumpy expenditures to show how widespread and expensive they are and life history interviews allowed us to identify effects on depletion of productive assets and long-term wellbeing.
A pragmatic approach to making causal inferences in poverty dynamics research would recognise that by exploiting a range of epistemological stances and research methods we are most likely to create a reliable body of evidence on which causal inferences can be made. Both quantitative and qualitative research methods have comparative advantages in strengthening this evidential base for causal claims. And in the end practical and robust approaches are needed to inform public policy regarding poverty.

References
[Please see published version for list of references]

Curbing dowry should be more central in poverty reduction strategies

2010-01-14T06:32:00.000-08:00

Research by SDRI fellow Peter Davis on poverty dynamics in Bangladesh shows that dowry is main cause of impoverishment in rural areas. This research is being featured as a DFID "Discovery of the Week", which is circulated to DFID staff. Here we reproduce the text of that report.

In South Asia, dowry has long been recognised as a harmful social problem. Governments recognise this but have been unsuccessful in curbing it and the damage it causes.
Those who oppose dowry usually cite its pernicious effects on human rights, as a cause of violence against women, or a hindrance to women’s empowerment. Research from Bangladesh suggests that dowry should also be seen as a major hindrance to poverty reduction.
There is a reluctance by international development organisations to engage with the problem as it is complex, cultural, and assumed to disappear with economic development. However, CPRC research suggests that dowry may not disappear with development. Dowry was rare among Muslims in Bangladesh before the 1960s but today is ubiquitous and harmful. This is puzzling because dowry has been outlawed in Bangladesh since 1980 and has never been recognised legally or customarily as an Islamic practice.
However, it is not only the increasing frequency of dowry that is the problem, the damage dowry inflicts on the poor also seems to be increasing. Dowry and wedding expenses were identified as the most important cause of impoverishment, alongside illness and medical expenses, by the rural people interviewed. Most commonly damage is due to loss of assets and indebtedness and the impact can be so great that a household never recovers from the crisis caused. This is most likely when productive assets such as land and livestock are lost, and a downward spiral of impoverishment and illness occurs as the bride’s parents become older.
Dowry is often raised from a number of sources with money borrowed from relatives and NGOs and loans paid back by selling land or livestock. Of the loans where the source was made clear, the research showed that 27 percent were interest-bearing from development NGOs. The actual percentage is certainly higher. These loans are supposed to be supporting income-generating activities rather than extracting profit from the distress-expenditure of poor families.
For most poor families large dowries relative to their means are unavoidable if their daughters are to be married. Payment by installments is often agreed and this can leave the woman vulnerable to threats of divorce, or worse, if her parents have trouble meeting payments. New brides can be abandoned by their husbands and sent home with demands for additional dowry payments and threats of divorce. This extortion is interlocked with, and facilitated by, the vulnerability of unmarried or divorced women.
Even though dowry is a transfer between families, it tends to be regressive. More dowry can be demanded by a wealthier or more powerful family. Marrying up involves a larger investment, marrying down is the cheaper option. However, for the poorest it is not possible to marry down. So the poorest families lose out.
Another issue is that the incentive to invest in a girl’s education is reduced because the natal household will not benefit from the returns on her education, and sometimes a highly educated bride is not seen as an attractive bride.
One implication of the serious negative socio-economic impact of dowry on poor households is that public policy focusing on ameliorating the negative impact of dowry should be seen as an anti-poverty imperative, as well as a way of reducing violence against, and oppression of, women. However, a better recognition of this side of the socio-economic impact of dowry does not mean that the prohibition of dowry is the only answer, even if it were to be adequately enforced, which in Bangladesh it is not.[1]
The complex economic and social environment that allows the combination of exclusion and stigmatisation of unmarried women, with the extraction of profit from the organisation of weddings, lays down a challenge for innovative approaches to anti-dowry social policy. These are needed at local levels in civil society as well as at the national level, and with anti-poverty and anti-dowry policies working together. Overall it is clear that more needs to be done to address the problem of dowry in Bangladesh if millions of poor families are to be spared a future of long-term poverty caused by extractive dowry and wedding costs.

Davis, P. (2007) Discussions among the poor: exploring poverty dynamics with focus groups in Bangladesh. Working Paper Series, No. 73, Chronic Poverty Research Centre.
Davis, P. (2010) Social exclusion and adverse incorporation in rural Bangaldesh: learning from a mixed-methods study of poverty dynamics. Working Paper Series, (forthcoming), Chronic Poverty Research Centre.

[1] Dowry has been formally illegal in Bangladesh since 1980, however the practice of giving and receiving dowry is still widespread. Many poor families interviewed in this study reported that it would be almost impossible to arrange a marriage for their daughter if a large dowry was not given. Regional differences in dowry rates within Bangladesh were detected in this study.

About this blog

2010-01-14T06:22:00.000-08:00

This blog has been set up to accompany the Social Development Research Initiative (SDRI) website at http://www.sdri.org.uk/. We will use this blog to update news and as a venue for comment and discussion on our work and related issues. Please feel free to join in the debate!