Seminars

Many interactions feature some degree of inequality in power between transacting parties. In some instances, inequalities of power are innocuous, while in others, they lead to monetary inequalities as well. In this paper, we investigate group cohesiveness when some members have been privileged with power over others to determine the distribution of resources amongst members. Using two Prisoner's Dilemma games (one with equal power and the other with asymmetric power), followed by a linear Voluntary Contribution Mechanism, the results show that power is used relatively benevolently towards wealth equality. When power is symmetrically distributed, the richest players free-ride significantly off poorer players, a result which is not evident when power asymmetrically distributed. In the treatment group with inequalities in power, contributions are reduced across the board with equal contributions, independent of wealth. Finally, the results suggest that lower contributions by powerless individuals are a consequence of usurped agency.

This paper provides a critical look at the evidence for prospect theory that does not rely on experimental laboratory data, but rather observational data of various sorts. It is commonly asserted that prospect theory is well confirmed and a significant part of the support comes from observational data. We argue that these data are considerably less supportive of prospect theory is claimed.

Part 1 of the paper first provides a brief overview of prospect theory and then goes on to identify some key results in philosophy of science that forms the framework for our analysis. We take it as fairly well established that:

• Testing always involves multiple or auxiliary assumptions to tie a specific hypothesis to data. This does not mean that there cannot be relatively conclusive tests of hypotheses, but it does mean that assessing the support any dataset gives a hypothesis requires careful attention to the assumptions needed to infer from data to . Background assumptions can easily be ignored or hidden in reports on observational data.
• Theories have multiple components that may need to be tested separately, and good tests eliminate situations where the same data might be produced by different elements of a theory.
• Theories in science (or models) get different interpretations in different contexts, and care is required in any empirical application to clearly specify which instantiation of a theory is at issue.

We argue that all these points apply to observational, nonexperimental evidence that to support prospect theory.

Part 2 of the paper is a critical examination of observational data allegedly supporting prospect theory. We show that:

• The data in question generally is relevant to some aspect of prospect theory, not the theory as a whole, though it is frequently claimed that prospect theory, not just some element of it, is supported.
• The data are in some cases quite weak, e.g., in measurements of risk attitudes using questionnaires.
• The results cited often could arise from different aspects of prospect theory (e.g., from loss aversion or probability weighting) thus rendering the data inconclusive, though this is frequently not acknowledged.
• Sometimes the phenomenon being cited is sufficiently distant from prospect theory that it is unclear how much support is provided for even single components of the theory and whether the phenomenon could arise from psychological mechanisms unrelated to prospect theory. Status quo bias is a case in point.

We use a frequently cited paper by Benartzi and Thaler (1995) that finds support for prospect theory in that it explains the equity premium puzzle. While certainly interesting, their account: 1) assumes it is utility loss aversion that explains the phenomenon while probability weighing, another part of prospect theory, could also explain the phenomenon but is not ruled out; 2) rests on assumptions about frequency of portfolio evaluation that are not entailed by prospect theory and which have minimal evidence; 3) depends on measures of loss aversion parameters from Tversky and Kahneman (1992) experiments which did not use real incentives and have multiple known problems (Harrison and Ross (2017)); and 4) depends on a reference point for how losses and gains are evaluated that has no direct evidence. Thus a paper that is widely cited as supporting prospect theory has numerous controversial assumptions which are largely unnoticed by those claiming it supports the model, providing evidence that the widespread support for prospect theory is not just due to the data but also to disciplinary trends, fad, and fashion. We will show that similar kinds of issues surface in key observational evidence that is claimed to support prospect theory.

Part 3 of the paper analyses the dataset of Rieger, Wang and Hens (2015) who used a hypothetical survey to elicit the risk preferences of university students in 53 countries. The authors claim to find evidence that supports features of prospect theory: specifically, risk aversion over gains, and over losses. However, a comparison of the original working paper (Rieger, Wang and Hens (2011)) and the published version (Rieger, Wang and Hens (2015)) shows that these conclusions are very sensitive to data coding and selection choices, and the types of analyses that are conducted, raising the spectre of what Gelman and Loken (2014) term “the garden of forking paths”. Furthermore, while the working paper focused on loss aversion, probability weighting, and risk aversion over gains and over losses, the publication confined its focus to the latter two components of prospect theory. In reanalyses of the we find that even the “robust” results reported in the publication do not hold up to rigorous data analysis.

In sum, this paper challenges the received wisdom in behavioural economics that prospect theory finds significant empirical support in observational data. We show that many of the claims in the literature to specific components of the theory rather than the full model itself, that the results which appear to support the model are often based on ad hoc additions to the original formulation of the theory, and that some of the findings are very sensitive to the types of analyses that are conducted.

The objective of this paper is to estimate willingness to pay for three electricity-driven vehicles –hybrid, plug-in hybrid, and electric vehicles. A discrete choice experiment is used to elicit consumer preferences for several vehicle attributes and policy incentives in Poland. Quota sampling was used to draw a representative sample of the Polish adult population in terms of several socio-demographic characteristics and a sample who intend to buy a passenger car within next three years. The survey took form of structured computer-assisted web interviews. This survey is the first on this topic and using discrete choice experiment in Central and Eastern Europe. We found that preferences of Polish consumers for hybrid and electric vehicles are significantly lower than for a conventional vehicle, even under a scenario that would implement a public program that would allow slow mode recharging an electric or plug-in hybrid vehicle in the place where people usually park their car. Charging time, availability of charging stations and driving range are currently the most important barriers to development of market for electricity driven vehicles. On average, Polish drivers are willing to pay about 1,260 zł for each additional 100 km of driving range, and about 500 zł for each hour saved for recharging. Drivers who intend to buy a second-hand car value the driving range less. Preference for AFVs markedly rose, when availability of fast-mode recharging improved from low level to medium or even high level. Providing other benefits, such as free parking and public transport, increases the probability to choose the electricity driven vehicles. Average WTP for free parking is 2,300 zł and for public transport 1,400 zł, and new car buyers are willing to pay more than second-hand buyers.

Research on using behavioural interventions to reduce energy usage in residential households in recent times have been explored however till date, proof of this for the office building context is unavailable. Coordinating and monitoring energy use among office floors is more challenging than residential energy consumption for two reasons. Unlike residential users, occupants of office buildings typically do not have any financial incentives to reduce their energy use. Secondly, while residential households typically have four members, office floors can have between 50-200 individuals, making coordination much more challenging even where the will exists. In this paper, we use behavioural insights to design a nudge – a low cost but high impact tweak to the existing system – aimed at reducing electricity consumption in 4 Dorp Street, a large office building with 24 floors, where several provincial government departments in the Western Cape are headquartered. We collected half-hourly meter readings from randomly assigned floors between the period 2015 – 2016. Results show a 9%  reduction in energy consumption from the mix of routine information and inter-floor competitions while a combination of regular information, inter-floor competitions and assigned floor advocates yielded a 14% reduction in energy use. These are some of the largest reductions in energy consumption to have been demonstrated using behavioural nudges.

The paper develops a framework for word-of-mouth learning in networks where agents strategically decide when to take an irreversible action. Agents face a trade-off between taking an irreversible action early enough to avoid the cost of waiting and waiting to receive more information to increase confidence in their choice. We characterize equilibrium exit times and establish conditions for correct learning in large societies. We show that exit times are shorter for highly connected networks and agents. The necessary conditions for correct learning are: (i) no single or small group of agents should have unbounded influence as measured by conditional in-degree; (ii) The underlying network must have a bounded diameter. Finally, we show that the presence of noise in signals prolongs exit times, and hence increases the likelihood of asymptotic learning.

One of the factors linked to South Africa’s relatively high maternal mortality ratio is late access to antenatal care (ANC). Early access is especially important in South Africa due to the high HIV prevalence amongst pregnant women. This study examined the impact of a package intervention, consisting of an incentive called the Thula Baba Box (a so-called "starter kit" for a new baby) and a community health worker (CHW) programme, on early access to ANC.  A pilot randomised controlled trial consisting of 72 women aged 18 and older was conducted in an urban area in South Africa to evaluate the impact of the package intervention. The outcomes measured are the timing of first antenatal care visit and whether they attended more than four times. Results: Women in the intervention groups sought care on average 1.35 months earlier than the control group. They were also significantly more likely to attend at least four antenatal clinic visits.

Economists concerned about potential violations of the independence axiom of Expected Utility Theory (EUT) have proposed experiments to test if subjects are instead more likely to behave as if they weight probabilities, as is proposed by Cumulative Prospect Theory and Rank Dependent Utility (RDU) theory. Generally, these experiments involve subjects making binary choices between two lotteries or gambles, structurally estimating the alternative utility models, and choosing a "winning" model among the alternatives estimated. I conduct a power analysis of the capacity of a lottery battery instrument to correctly classify experimental subjects as employing either EUT or RDU, and the effect of this classification on the accuracy of subjects' welfare surplus. For large ranges of parameter values for these models, I find that the probability of type I and type II errors in the classification process are non-trivial, and can be very costly in terms of welfare surplus. Additionally, I show that for a hypothetical population comprising subjects employing EUT or RDU, we can arrive at more accurate welfare surplus estimates in aggregate by assuming that every subject employs the RDU functional, rather than by first trying to differentiate RDU subjects from EUT subjects.

Shortly after the introduction of Expected Utility Theory, economists and psychologists began publishing results of experiments that showed choices made by experimental subjects which apparently violate one or more of the axioms of Expected Utility Theory. These responses vary from developing new theoretical models, typically ones that nest Expected Utility Theory as a special case such as Rank Dependent Utility, to critiques of experimental method and scope, to the reemergence of stochastic models of choice. Over the last several decades, experimental economists have incorporated these insights of theory and practice into their methods in an effort to refine both the hypotheses and evidence associated with these apparent violations. I conduct a power analysis of the ability of a lottery battery instrument to correctly classify experimental subjects as employing either Expected Utility Theory or Rank Dependent Utility given a popular stochastic choice model, and the effect of this classification on the accuracy of the estimates of welfare surplus for the subjects. For large ranges of parameter values for these models, I find that the probability of type I and type II errors in the classification process are not trivial, and can be very costly in terms of welfare surplus. Additionally I show that for a hypothetical population comprising subjects employing Expected Utility Theory or Rank Dependent Utility, we can arrive at more accurate welfare surplus estimates in aggregate by assuming every subject employs the Rank Dependent Utility functional rather than first trying to differentiate Utility subjects from Expected Utility Theory subjects.

In evolutionary game theory, “minigames” with reduced strategy sets are sometimes analysed in lieu of more complex models with many strategies. Are these simplified versions up to the task of explaining pertinent dynamic features of the larger models? This paper looks at the ultimatum game, in which it is known that a noisy evolutionary model leads to stable dynamic equilibriums that are far away from the game’s unique subgame perfect solution. It is argued that a naive approach is unsatisfactory and that the minigame analysis is more useful when related to the full game explicitly. A constellation of embedded minigames is identified in the full game, one for each imperfect equilibrium of the full game, with each playing out on its own conditional frequency space. It is shown that the conditional frequency dynamics applicable to these minigames have the same form as a full game’s dynamics with a reduced strategy set. While the minigames thus identified are still not two-dimensional, it is shown that two critical variables in each can be treated separately from the others, and these indeed behave like the variables in a two-dimensional standalone minigame. A graphical analysis based on selection-mutation equilibrium loci allows a clear understanding of why stable imperfect equilibriums exist and which factors tend to stabilize particular equilibriums. For example, lower-offer equilibriums are easier to stabilize, because a) proposers have more to lose by deviating from them and b) responder mutation aims at a higher target for the relevant conditional frequency.

Gambling can be traced back to some of the earliest human civilisations and it is now legal in many jurisdictions, but what are the factors that affect a person’s decision to gamble and the severity with which they gamble? I analyse the relationship between gambling behaviour and a self-reported measure of impulsivity in the 2008 National Urban Prevalence Study of Gambling Behaviour in South Africa. I adopt a novel, sophisticated statistical framework, which conforms closely to the hypothesised, latent process generating the data and corrects for sample selection problems, to explore this relationship. I find that impulsivity affects the decision to gamble, the likelihood of experiencing gambling-related problems, and the severity of these problems. These findings are relevant to theories of addiction and policy interventions aimed at treating gambling disorder and mitigating its harmful personal, familial, and social effects.

There is a dearth of studies on cooperation within the extended inter-generational family, a common living arrangement in South Africa. Cooperation impacts on families’ social and economic fortunes. This paper presents evidence on cooperation in the extended family from a framed field experiment, which adopts a Voluntary Contribution Mechanism (VCM) game that was implemented in two poor South African communities. The subjects’ contributions to the public accounts measure cooperation and the post-experimental survey provides thematic data to map out correlates of intra-household cooperation. Random Effect (RE) regression models are employed to explore predictors of cooperation. The failure of household surplus-maximisation is confirmed by limited cooperation. Discussion, as a form of communication, mitigates contribution decay in urban Bloemfontein. Cooperation between parents and grandparents is enhanced by the level of family communication in urban Bloemfontein. The endowment size effect in urban Bloemfontein imply that the benefits of an economic opportunity for one family member, to earn a wage or receive cash transfer, may cascade to all family members as a result of cooperation and sharing. Given the central role played by the family in economic, socialisation and procreation activities, cooperation in extended families and the associated investment in family public goods, in urban informal settings can be augmented by social developmental policies that enhance family communication and cohesion. Examples include preventative and developmental social work programmes, in particular training in parenting and marriage skills.

Gambling behavior is pervasive, apparently growing, and of methodological and substantive interest to economists. We examine the manner in which the population prevalence of disordered gambling has been estimated. General population surveys have deepened our knowledge of the population prevalence of gambling disorders, as well as the manner in which gambling disorder is associated with other mental health problems. However, we identify a fundamental bias in the manner in which these surveys have been used to draw inferences about the general population prevalence of gambling problems, due to a behavioral response to seemingly innocuous “trigger,” “gateway” or “diagnostic stem” questions in the design of surveys. Formal modeling of the latent sample selection behavior generated by these trigger questions leads to dramatically different inferences about population prevalence and comorbidities with other psychiatric disorders. The population prevalence of problem or pathological gambling in the United States is inferred to be 7.7%, rather than the figure of 1.3% that is commonly cited when this behavioral response is ignored. Comorbidities are inferred to be much smaller than the received wisdom, particularly when considering the marginal association with other mental health problems rather than the total association. The issues identified here apply, in principle, to every psychiatric disorder covered by these surveys, and not just gambling disorder. We discuss ways in which these behavioral biases can be mitigated in future surveys.

Cason and Plott (2014, hereafter CP) propose that sub-optimal behavior is sometimes explained by game form misconception (GFM), a failure of game form recognition, rather than by non- standard preferences or framing effects. This paper explores whether GFM can robustly explain bidding mistakes by inexperienced subjects in the Becker-DeGroot-Marschak mechanism (Becker et al., 1964, hereafter BDM). We derive two new general tests of the CP hypothesis based on comparing subject behavior on the misconceived task and on the task it is misconceived for. When we apply these tests to the BDM we do not find new evidence in support of the GFM hypothesis.

This paper experimentally tests whether violations of Savage’s (1954) sure-thing principle (STP) decrease through statistical learning. Our subjects repeatedly had to bet on the drawings from an urn with an unknown proportion of differently colored balls. The control group was thereby subjected to learning through mere thought only. In addition, the test group received more and more statistical information over the course of the experiment by observing the color of the ball actually drawn after each bet. We expected that statistical learning would decrease the decision makers’ ambiguity, thereby implying a stronger decrease of STP violations in the test than in the control group. However, our data surprisingly shows that learning by mere thought rather than statistical learning leads to a decrease in STP violations.

We examine social network effects in informal intellectual collaboration in financial economics. We make use of a novel social network connecting 7,424 economists that have collaborated formally or informally on 1,886 research papers. While formal collaboration – co-authoring a paper – is well researched, we focus on informal collaboration, which is the provision of helpful input and commonly acknowledged in special sections. We hypothesize that more central individuals have access to more information traversing the network, and collaborating with these individuals results in higher scientific impact. Treating the assignment of discussants at major finance conferences as quasi-natural experiment, we establish a network effect: Commentary from a discussant with a one standard deviation higher network centrality is associated with 11% more citations over four years for an average paper. This effect is weaker, however, for the probability of publishing a paper in one of the top three finance journals.

This study characterises the risk attitudes of a sample of individuals living in a flood-prone, urban informal settlement. A number of Cape Town’s informal settlements are vulnerable to flooding. Vulnerability to flood risk is determined, in part, by participants’ adaptation strategies. The objective of this paper is to determine the extent to which risk attitudes are correlated with choice of adaptation strategy. Risk attitudes are elicited from participants’ choices over a series of lottery tasks for real monetary prizes. The results indicate that individuals adopting more effective (and costly) adaptation strategies are more . This result implies that attempts by local authorities to encourage uptake of adaptation strategies must take into account risk attitudes. For example, city officials might supplement ongoing education initiatives with proactive measures to incentivize even risk-seeking individuals to engage with adaptation (for example, through the provision of a subsidy on building materials).

Stochastic choice models have long been employed in the structural estimation of utility functions for individuals, or pooled representative agent models. These models usefully allow for choices that apparently violate a utility theory, be it Expected Utility Theory (EUT), Utility Theory (RDU) or something else, to be rationalized as probabilistic events with likelihoods lower than choices which do not violate the theory. Most often, though not always, this is done by characterizing these apparent violations as "mistakes" or "choice errors" which are costly in terms of welfare. However, the almost exclusive focus on apparent violations of a utility theory has pushed to the background another important implication of these models; many choice patterns which are apparently consistent with a utility theory may also contain "mistakes". Furthermore, should we make certain assumptions about the distributions of preferences in a population, or estimate these distributions directly from pooled choice data, it may be the case that many apparently consistent choice patterns actually result in lower welfare than apparently inconsistent choices.

In this paper we make use of a stylized example in which simulated populations of agents make choices over the popular Holt and Laury (2002) multiple price list of lotteries. We demonstrate firstly that it is entirely feasible that the welfare of an agent who produces a choice pattern apparently consistent with a utility theory is not necessarily greater than the welfare of an agent who produces apparently inconsistent choice patterns. Secondly we demonstrate that the expected welfare of a population is primarily driven not by the distribution of preferences in the population, but the distribution of the stochastic elements of choice in the population.

The probability discounting model is a popular framework for investigating people’s instantaneous or atemporal attitudes toward risk in experimental settings. It is particularly common in studies of addiction where delay discounting data is also often obtained. The model was developed in a series of three papers by Rachlin, Logue, Gibbon and Frankel (1986), Rachlin, Castrogiovanni and Cross (1987), and Rachlin, Raineri and Cross (1991). As its name suggests, the probability discounting model draws its inspiration from models of temporal or delay discounting, where the delay to a reward is replaced with the odds against receiving a reward. This paper explains the model in language familiar to decision theorists, statisticians and economists, and it shows how the model relates to standard theories of choice under risk like expected utility theory, prospect theory, rank-dependent utility theory, and rank-dependent expected value theory. It highlights the way in which researchers typically collect probability discounting data and how they analyse it statistically. The paper then uses data from a well-cited study to show that more flexible specifications of a probability weighting function outperform the probability discounting model’s implicit probability weighting function. It concludes by arguing that researchers who are interested in investigating choice under risk should adopt theoretical, methodological and statistical tools appropriate to the task, and should therefore abandon the probability discounting model.

A principal source of interest in behavioral economics has been its advertised contributions to policies aimed at ‘nudging’ people away from allegedly natural but self-defeating behavior toward patterns of response thought more likely to improve their welfare. This has occasioned controversies among economists and philosophers around the normative limits of paternalism, especially by technical policy advisors. One recent suggestion has been that ‘boosting,’ in which interventions aim to enhance people’s general cognitive skills and representational repertoires instead of manipulating their choice environments behind their backs, avoids the main normative challenges. A limitation in most of this literature is that it has focused on relatively sweeping policy recommendations and consequently on strong polar alternatives of general paternalism and strict laissez faire. In what follows, we review a real instance, drawn from a consulting project we conducted for an investment bank, of a proposed intervention that is more typical of the kind that economists are more often actually called upon to offer. In this example, the sophistication of current tools for preference attribution, combined with philosophical externalism about the semantics of preferences that makes it less plausible to attribute their literal self-conscious representation to people as content becomes more tightly refined, blocks applicability of the distinction between nudging and boosting. This seems to call for irreducible, context-specific ethical judgment in assessing the appropriateness of the forms of paternalism that economists must actually wrestle with in going about their everyday business.