Are risks over multiple attributes traded off? A case study of aid

https://doi.org/10.1016/j.jebo.2019.05.025Get rights and content

Abstract

Understanding decisions in situations involving multiple risks is vital in many contexts, including the provision of aid to developing countries. Aid typically involves trade-offs over multiple, desirable, but risky outcomes. We partner with an aid organisation and design a laboratory experiment to investigate multi-attribute risk preferences in the context of donations to risky aid projects. We find significant differences between attitudes over single risk outcomes in monetary and non-monetary domains. Our results show that individuals are generally multi-attribute risk averse and view aid outcomes as substitutes rather than complements. These preferences of donors contrast sharply with the views of experts and aid agencies, who tend to emphasise the complementary nature of aid outcomes. Finally, when individuals make risky donation decisions with their own money rather than someone else's money, the degree of risk aversion decreases markedly.

Introduction

How do we make decisions when outcomes are subject to risk? Economists have produced a rich theoretical and empirical literature on attitudes towards single risks. To date comparatively little empirical attention has been given to situations that involve multiple, risky outcomes. Yet such situations are commonplace. For example, physicians decide about health care, which consists of multiple outcomes, all of which are associated with inherently different levels of clinical risk. Similarly, development interventions are frequently designed to achieve the dual objectives of poverty reduction and natural resource protection, both of which are subject to risk. We examine decision making in the important context of aid where individuals face risks on single and multiple outcomes and in different domains.

Global spending on foreign aid is substantial and amounts to around USD 500 billion annually. The vast majority of foreign aid targets specific sectors, projects or outcomes, with only a tiny fraction spent on non-specific objectives (OECD, 2015). In addition to aid given by foreign governments to improve developmental outcomes in poor countries, private donors also contribute significantly towards these outcomes. Individual preferences matter not only for the donation decisions of individuals but also because they are reflected in the allocation of aid provided by the public sector. Importantly, an individual's preferences over aid projects may differ depending on whether private or public funds are used. This issue remains largely unexplored and is a potential reason for why we may observe preference heterogeneity amongst individuals when we study the use of public versus private funds for aid provision.

Preference heterogeneity may arise due to the risky and multidimensional nature of aid. Aid is inherently risky, in the sense that the final outcomes of aid projects depend on many factors that are outside the control of donors. Risk may or may not be explicitly acknowledged to donors, and can have a variety of causes ranging from corruption and geographical or technological challenges to adverse weather and natural disasters. Brock et al. (2013) investigate the preferences of individuals for giving in risky settings where perceptions of fairness behavior are related to ex ante comparisons of opportunities or, as in our study, ex post comparisons of outcomes. Similarly, Exley (2016) investigates the effect on contributions when the charitable outcome is subject to risk and finds significant reductions in donations even when the risk is small. In addition to investigating risk attitudes over single outcomes, we investigate preferences and behavior when more than one outcome is subject to risk.2

Aid projects commonly target many outcomes simultaneously. For example, the distribution of solar lamps in developing countries aims at improving educational outcomes by enabling children to study after dark as well as reducing emissions by replacing kerosene lamps with a solar-powered lamp (SolarAid, 2018; Stojanovski et al., 2018). In the health domain, aid agencies often promote the multipronged approach of improving water, sanitation and hygiene (WASH) jointly to combat waterborne diseases.

It is not clear, however, how individual donors trade-off different aid outcomes. The charitable giving literature has investigated the relationship of an individual's giving to multiple organizations. For example, de Oliveira et al. (2011) find that there are “giving types:” people who give in many instances and across many charities (such as charities targeted towards education, health and job training) as opposed to people who seldom or never give. In contrast, we keep the aid organization the same but have multiple outcomes that serve the purpose of improving the health and well-being of disadvantaged people. Also, in our setting these outcomes are subject to risk.

Moreover, while aid outcomes in the solar lamp example are substitutes since improvements in educational outcomes are not necessary to achieve emission reductions, aid agencies clearly promote the complementarity across WASH initiatives. It is an open question as to whether individual donors perceive aid outcome substitutability or complementarity in accordance with expert opinion and how these perceptions affect donation behaviour. To answer these questions, we design a unique experiment that allows us to disentangle multi-attribute risk attitudes from perceived outcome complementarity.

We partner with a well-established aid organisation to accurately describe individual preferences with respect to risky aid. In particular, the problem of insufficient control in the field, needed to allow one to investigate these subtle conceptual issues, is circumvented by conducting laboratory experiments that use simple treatments over well-defined risks and aid outcomes. The partner organisation has the added advantage of allowing donors to target specific aid outcomes, making it the ideal conduit to bring together our research investigation on risk and multiple outcomes.

We implement a three-part experimental, within-subject design, where in Part 1 participants decide over binary lotteries with monetary outcomes. In Parts 2 and 3 the decisions involve respectively experimenter donation and own donation decisions over binary aid lotteries where single and multiple outcomes for the beneficiaries of the aid project are subject to risk.

This design allows us to first analyze the extent to which risk preferences differ across monetary outcomes in Part 1 and non-monetary outcomes in subsequent parts. This part of the design builds on the risk elicitation method developed by Holt and Laury (2002) and contributes to the extensive analysis of single attribute risk. Second, we investigate attitudes towards the presence of multiple risks. Is the type of aid project chosen one that is likely to achieve one good outcome with near certainty at the cost of almost surely failing all others, one that might achieve just passable outcomes across all objectives, or one that either succeeds or fails in all domains? Are these risks traded off in an additive manner? Our novel experimental design in Parts 2 and 3 involves participants choosing amongst different kinds of aid projects, all of which have risk associated with them. In some cases participants choose amongst projects, targeting a single risky outcome, while other choices are over aid projects that target multiple risky outcomes. In the case of aid projects with multiple outcomes, we vary the extent to which outcome risks are correlated.

Third, the effect of using private versus public funds to administer aid on observed preferences is examined by comparing donation decisions made by the participants when using their own money in Part 3 with decisions that the same participants make in Part 2, knowing that the donation is taken out of the experimenter's fund.

In terms of preference ranking our results indicate that individuals have a clear preference for monetary outcomes to themselves over non-monetary outcomes for aid recipients. Within this latter category we find that one aid outcome is also preferred over the other. We are able to identify this aspect of preferences over different attributes of risk because we avoid reducing all attributes to their monetary equivalent. Doing so would mimic the workings of a complete market between these attributes, a convenient modelling assumption but one that we find unattractive for the applications that interest us.

Similarly, in terms of single attribute risk attitudes, we find significant differences in attitudes towards risks in monetary versus non-monetary (aid) outcomes. Across the different non-monetary outcomes, average risk attitudes are indistinguishable. We also find that individuals are on average averse to multi-attribute risk in non-monetary settings, implying that donors faced with multiple risky aid outcomes will apply an overall risk premium comprising the risk premia for each singular outcome risk as well as a premium for the joint risk. Our result that multi-attribute risk aversion is less pronounced when subjects make risky donation decisions with their own money rather than someone else's money suggests that the joint risk premium is highest when evaluating contributions to aid projects made by a third party, such as Government Aid.

These insights contribute to the growing application of models of tradeoffs over risky attributes. For example, Bommier, 2006, Bommier, 2010, Bommier, 2013) applies multi-attribute risk aversion to the idea that agents know that they have finite lives, but do not know when they will die. This is a naturally-occurring and pervasive time-dated risk, and concepts of correlation aversion then have sharp implications for the characterization of discounting, portfolio choice, and life-cycle behavior. One could easily imagine extensions of our application to consider risk management choices, in the form of investments in health or location to improve one's chances of living longer. Or extensions to consider another attribute, the quality of life, as well as longevity per se.

Section 2 provides a brief exposition of the theoretical framework tested. The experimental design and procedures are described in Section 3. The results of the data analysis are presented in Section 4, followed by a discussion of limitations and extensions of our research in Section 5. Section 6 concludes.

Section snippets

Theoretical framework

Modelling an individual's choices over risky aid projects requires a theoretical framework that is able to handle multiple outcomes that are subject to potentially correlated risks. A utility function that is non-additively separable in its multiple attributes has been shown to be sufficiently rich to allow for context-dependent risk preferences over single risks as well as aversion to multiple risks (Richard, 1975, Keeney and Raiffa, 1976, Bommier, 2007, Andersen et al., 2018). By assuming

3.1. Overview

The experiment involves donations to WaterAid, an international aid organisation that provides water, sanitation and hygiene (WASH) related services to combat waterborne diseases in developing countries. According to WaterAid's philosophy WASH services are highly complementary as each service targets a distinct root cause of waterborne diseases in developing countries. Donations to WaterAid can be made easily and verifiably online, and it is possible to specify whether the donor is an

Data

We present results in two, complementary ways. The first, in this section, is to examine the observed behavior using descriptive statistical tools, which allow us to remain agnostic about the latent structure. The second, in the next section, is to assume some latent structure consistent with our theoretical model, along with some parametric structure, to allow us to identify parameters characterizing the extent and nature of any tradeoff between risky attributes.

Table 5 shows the proportion of

Results

We now turn to the econometric results, which are organised around the three research questions outlined earlier. All risk parameters are estimated jointly for the theoretical model described by Eq. (1), which allows for the two aid outcomes to interact in a non-additive manner, while these two then interact with money in an additive manner. In effect, this is assuming a nested multi-attribute utility function, which is the natural choice, given that the amount that the subjects donate in Part

Extensions

This research contributes to our understanding of how preferences over multiple monetary and non-monetary risks may be characterised. Given the additional complexity required to model multi-attribute risk attitudes, we deliberately kept other aspects of the utility framework simple by assuming EUT. Our analysis suggests that EUT is indeed suitable to model behavior subject to multiple risks. However, this does not rule out other models of decision-making under risk, such as Rank Dependent

Conclusion

Questions about how people behave when facing decisions involving single and multiple risks are important in many public and private good contexts that involve trade-offs over multiple risky outcomes. We designed a laboratory experiment to investigate whether individuals exhibit different levels of risk aversion when choosing over risky monetary or aid outcomes; if there is evidence of multi-attribute risk aversion when several outcomes are subject to risk; and whether individuals perceive risk

References (41)

  • J. Andreoni

    Giving with impure altruism: applications to charity and Ricardian equivalence

    J. Polit. Econ.

    (1989)
  • J. Andreoni

    Impure altruism and donations to public goods: a theory of warm-glow giving

    Econ. J.

    (1990)
  • O.E. Barndorff-Nielsen et al.

    Inference and Asymptotics

    (1994)
  • A. Bommier

    Uncertain lifetime and intertemporal choice: risk aversion as a rationale for time discounting

    Int. Econ. Rev.

    (2006)
  • A. Bommier

    Risk aversion, intertemporal elasticity of substitution and correlation aversion

    Econ. Bull.

    (2007)
  • A. Bommier

    Portfolio choice under uncertain lifetime

    J. Public Econ. Theory

    (2010)
  • A. Bommier

    Life-cycle preferences revisited

    J. Eur. Econ. Assoc.

    (2013)
  • A. Botelho et al.

    Testing static game theory with dynamic experiments: a case study of public goods

    Games Econ. Behav.

    (2009)
  • J.M. Brock et al.

    Dictating the risk: experimental evidence on giving in risky environments

    Am. Econ. Rev.

    (2013)
  • T.C. Cheng et al.

    Do you have to win it to fix it? A longitudinal study of lottery winners and their health-care demand

    Am. J. Health Econ.

    (2018)
  • Cited by (0)

    The authors thank the editor and two anonymous reviewers, Paul Raschky and Tim Cason for helpful comments and suggestions. Funding from the Cooperative Research Centre for Water Sensitive Cities (CRC grant number 20110044) is acknowledged.

    1

    Harrison is also affiliated with the School of Economics, University of Cape Town.

    View full text