Motivated beliefs, self-serving recall and data omission: Economic implications and experimental evidence
- Autores: Adrián Caballero Castillejo
- Directores de la Tesis: Carmen Arguedas Tomás (dir. tes.), Raúl López Pérez (dir. tes.)
- Lectura: En la Universidad Autónoma de Madrid ( España ) en 2021
- Idioma: inglés
- Número de páginas: 211
- Tribunal Calificador de la Tesis: Nicolas Treich (presid.), Agnes Pinter (secret.), Eli Spiegelman (voc.)
- Programa de doctorado: Programa de Doctorado en Economía y Empresa por la Universidad Autónoma de Madrid
- Materias:
- Enlaces
- Tesis en acceso abierto en: Biblos-e Archivo
- Resumen
This Thesis is composed of four essays that provide some economic models and experimental evidence to answer several questions regarding the determinants and the consequences of data omission and biased beliefs due to cognitive limitations and preferences over beliefs. While related, each chapter can be read independently. A summary of each chapter is presented below.
In Chapter 1, we explore the consequences of self-deception in strategic settings where the consumers’ behavior may entail uncertain risks for themselves or others. This includes plenty of real-life situations. For example, most consumers are not fully aware of the specific impact of recycling plastic, flying from Madrid to Boston, or donating to a certain NGO. We present a new model in which consumers (i) may experience anticipatory utility (Caplin and Leahy, 2001; Kőszegi, 2010; Loewenstein, 1987) -i.e. their current utility may depend on their expectations about their future utility- and (ii) can choose their beliefs regarding the risks associated to a hazardous product. In contrast to the existing literature on motivated reasoning and self-deception (see, for example, Akerlof and Dickens, 1982; Bénabou, 2013; Brunnermeier and Parker, 2005), we focus on strategic settings and the relations of interdependence both between the consumers and between these and the firm(s). Thus, not only prices determine optimal beliefs and choices, but consumer’s preferences over beliefs also alter the firms’ incentives.
Consumers, for example, may alleviate their anxiety by underestimating the risk associated to the hazardous product. This, however, has consequences for the market equilibrium, since optimistic consumers are more willing to pay for that product than consumers with more realistic beliefs, other things equal. In this sense, our model provides some interesting results: when self-deception is not possible, for example, those individuals that feel more anxious or excited about their future are relatively more prone to take action in the present in order to improve their future situation. In contrast, when individuals can choose their beliefs, anticipatory utility may trigger self-deception rather than action. Also, our model predicts that self-deception is more likely to arise when there is uncertainty about the externalities associated to a product (or action), when individual behavior has a negligible impact on the economy, or when the potential consequences will realize only in a distant future. This seems particularly relevant, for example, regarding environmental problems. Interestingly, however, the existence of individuals that underestimate the risks associated to the hazardous product does not necessarily lead to a larger aggregated consumption of it in the economy. The specific result will depend on the characteristics of the economy, including the intensity of the anticipatory utility among individuals. Overall, we believe that the model presented in this chapter provides some new insights that may prove useful to deal with a large range of phenomena, including belief polarization, social responsibility or climate change denial.
In Chapter 2 we present an experimental design to test the implications of some models of motivated beliefs and optimism. All these models assume that individuals have preferences over the possible states of nature and they derive utility directly from keeping beliefs in line with these preferences (for example, in the form of anticipated utility). On the other hand, forming and keeping unrealistic beliefs may entail some costs. We focus on two families of theories that differ mainly in the factors that prevent individuals’ beliefs from departing too much from reality. According to some models (e.g. Akerlof and Dickens, 1982; Brunnermeier and Parker, 2005), optimism is more likely when the material costs associated to biased beliefs are relatively low. On the contrary, individuals may be more reflective and form more accurate beliefs in situations where biases lead to relatively costly or hazardous decisions. Alternatively, models by Rabin (1994) and Bracha and Brown (2012) stress the fact that self-deception is costly also in the sense that the individuals must selectively look for favorable information, while avoiding unfavorable evidence or restraining challenging thoughts. Remarkably, while there is abundant evidence of situations in which people seem optimistic, tests about the specific implications of these models are much less common, and far from conclusive. On account of this, we contribute to the existing literature with a novel experimental design.
Succinctly, the experiment consists of an estimation task in which each participant faces a virtual urn with 100 balls, each containing a boy or girl name. After observing a series of random draws, the participant must estimate the actual share of female balls in the urn. Importantly, monetary incentives are designed to induce a preference for that proportion to be as large as possible. Overall, the results of our experiment find scarce support for these models, although models by Rabin (1994) and Bracha and Brown (2012) fit our data relatively better. Still, we do not find systematic optimism among our participants. Moreover, the sign of the bias seems more related to the specific characteristic of the observed sample rather than individual characteristics or material incentives. The estimation task presented above was followed by a memory task, in which the participants are incentivized to recall as many boy and girl names in their urn as possible.
Using data from both tasks, we explore in Chapter 3 the connection between biased recall and optimism. In the last years, it has been suggested that the optimistic bias may be caused by self-serving recall, i.e. people often recall favorable information better than negative evidence (Epley and Gilovich, 2016; Bénabou and Tirole, 2016). Based on this literature, we provide a model of inference with self-serving recall and test experimentally its main predictions. In a nutshell, our model is based in two ideas: (i) people extrapolate from the evidence they recall; and (ii) people are more likely to recall information that is favorable given their preferences, i.e. when it supports their preferred states of nature. Our results provide some support for the latter: in our experiment, the participants are more likely to recall girl names -which are associated to a larger payoff-. In addition, they do not seem aware of this bias: in fact, they expect to recall bad news (i.e. male balls) relatively better than good news. Nevertheless, several results suggest that self-serving recall does not induce optimism in our experimental design. First, our subjects do not systematically overestimate the proportion of female balls. Further, we find no correlation between optimism and biased recall: people who provide inflated estimations are not relatively more likely to recall girl names better. In this sense, the participants’ estimations are better fitted by the Bayesian model than by a model assuming that people track the proportion of female balls in the recalled sample. Overall, our results suggest that the link between memory tasks and estimation tasks is not straightforward, and that people may infer from a different sample than the recalled one. We consider this result to be particularly relevant for experimental research, where these tasks are widely used.
Finally, Chapter 4 explores other sources of bias regarding belief formation. Specifically, we focus on data omission in contexts where there is no preference for any state of the world. To that end, we propose both an analytical framework and a lab experiment in which participants face a quite simple problem of inference. Succinctly, each participant observes a series of random draws with replacement from one urn containing red and blue balls. The subject knows that the specific rate of red balls her urn is randomly determined with uniform probability from a set of three rates. Based on the evidence observed, the participant is asked to estimate the true rate in an incentive-compatible manner. In contrast to the experimental design from Chapters 2 and 3, the participant’s payoff depends exclusively on the accuracy of her estimate and not on the rate per se. Based on the experimental evidence, we explore some relevant questions regarding data omission. First, we consider heterogeneity among the participants. For example, some individuals may omit more data than others when elaborating their estimates. On this matter we find that, while the estimates of most participants fit the Bayesian model relatively well, a non-negligible portion of them seem to rely on very small subsamples. We find this quite striking given the simplicity of the problem. Further, our evidence suggests that differences in data omissions are more likely to do with differences in attention than memory constraints. Finally, we find experimental support for the hypothesis that experience and incentives can alleviate data omission in our experimental design. Yet, the extent of the improvement seems modulated by the complexity of the problem.
