2. MARCO DE REFERENCIA
2.3. MARCO LEGAL
The findings in this chapter show that high-level features like control preferences and overconfidence, studied through a top-down approach, are features of choice option representations that can have an effect on behavioural outcomes. Furthermore, it is clear that these features are only relevant given the way the task goal was constructed (i.e., as a choice between the self and an advisor). Focus was placed on features concerning representation of the self since the only information that was provided about the advisor was their accuracy and the cost of their services. The accuracy of the advisor represents uncertainty in this task and was presented as a point estimate (i.e., a property of the stimulus format). Providing richer information regarding uncertainty was restricted here to allow focus on self-representation (see chapter 3 for a relaxation of this restriction on stimulus formats for uncertainty).
The results demonstrate that participants are willing to forgo rewards for the opportunity to make their own choices and hence to control their own payoffs. Thus, placing a premium on control shapes the agent’s utility function, having implications on the way they solve such a task goal. This preference was observed not only when faced with potential gains (in accord with Owens et al. 2014), but also when faced with potential loss. Moreover, the findings indicate that participants accurately assess the (sub)optimality of their delegation choices, suggesting that they are aware of the premium they are paying to maintain control.
The results could not be accounted for by participants’ overconfidence in their ability to maximise rewards and minimise losses, as their beliefs regarding their own ability were elicited and accounted for. In fact, on average participants were slightly overconfident even though they were unlikely to hold prior beliefs about their ability from “real world” experience and were given plenty of experience with the task. In this case, self-representation as being better than they really are did not have as big an effect as the construction of the exact utility
function that considered the premium placed on control. Participants were also given complete information about potential advisors, which would allow them to make rational decisions. Thus, under-delegation could not be attributed to a systematic misperception of either the participant’s expected utility nor the advisor’s expected utility. Indeed, when questioned about their ability to delegate accurately, participants’ representations of their delegation performance were surprisingly accurate.
The findings are in accord with a past study that identified a significant “control premium” in an experimental setting in which participants could bet that a partner, or instead they themselves, would answer quiz questions correctly (Owens et al. 2014). In light of participants’ elicited beliefs, participants should have bet on themselves 56.4% of the time – but in fact, did so 64.9% of the time. This suggests that representation of themselves as overconfident only explains part of the effect but that the specific control-adjusted utility function explains the rest. It follows that participants were, on average, willing to give up 8% to 15% of their expected earnings to retain control. In other words, the preference for control could not simply be explained by overconfidence or subjective beliefs; control appeared intrinsically desirable. In a way, their overconfident self-representation was aligned with their specific utility function (i.e., the task goal).
Fehr et al. (2012) also find, in a fundamentally different design, that people will sacrifice their material interest to maintain authority. They conduct an “authority game,” in which principals could choose to delegate decisions to an agent. Their central finding is that people will under-delegate, showing “a strong behavioural bias among principals to retain authority against their pecuniary interests and often to the disadvantage of both the principal and the agent.” Their major explanation is that people do not like to be overruled, and they know that if they delegate, their agent might disregard their information, or their wishes, in order to make their own selection. Using a different experimental design, Bartling et al. (2013) similarly find that decision rights have intrinsic and not merely instrumental value. There is also a relationship here between the findings and the phenomenon of “reactance,” which suggests that people rebel against choice-denying commands by defying them (Brehm and Brehm 2013).
The results support the past findings for the existence of a “control premium” and extend them by demonstrating a positive control premium not only in the domain of potential
gains, but also in the domain of potential loss. Furthermore, the results suggest that participants are aware that they are selecting to pay a premium for control. Future studies may wish to vary levels of expertise for the participants, include a measure of reliability or trust for the advisors, and perhaps include a variable that modulates effort that the participant may need to realize when he or she decides to retain control. Likewise, using this paradigm to study depressive populations could yield insights into the relationship of depression and locus of control; perhaps depressive populations would delegate more given that their locus of control is seen as driven by external forces (Benassi, Sweeney, & Dufour, 1988).
Why would people choose to under-delegate when they are seemingly aware of the material cost of under-delegation? Speculatively, this behaviour reflects a non-monetary intrinsic value for control, which is expressed via choice. The intrinsic value of choice may have emerged for several reasons. First, outcomes that we select ourselves often suit our preferences and needs more than those that have been selected for us (Beattie et al. 1994; Kray 2000; Polman 2010; Stone and Allgaier 2008; Waldfogel 1993). Thus, we have learned that environments in which we can exercise choice are usually (of course not always) more rewarding (Leotti et al. 2010; Patall et al. 2008; Rotter 1966). The frequent association between choice and reward may have led choice itself to be experienced as rewarding - something we seek and enjoy.
Second, a biologic system that provides higher intrinsic reward for things we have obtained ourselves compared to those that were simply chosen for us may be adaptive. If we learn that an action results in a reward, we can repeat that action in the future to gain more of the same. However, if we do not execute an action to obtain reward (or avoid harm), we lose the opportunity to acquire a “blueprint” of how to gain rewards (or avoid harm) again in the future. Thus, the value of outcomes we had obtained ourselves emerge both from their utility and from the information they contain for future outcomes. This means that features of self- representation, such as the intrinsic value of control, may reflect adaptive pressures from the environment.
In sum, in this controlled laboratory experiment the results show that people are willing to pay a control premium to make their own choices. With a somewhat different design, it is found that people will pay a control premium in the domain of gains and a similar premium in the domain of losses. This finding runs counter to the idea that people prefer to delegate
decisions involving unwanted outcomes in order to avoid regret (Loomes and Sugden 1982) and instead supports the notion that choice may be preferred regardless of expected valence of the outcome (Cockburn et al. 2014; Leotti and Delgado 2011, 2014; Sharot et al. 2009, 2010a). Moreover, the current study suggests that people are aware of the monetary premium they are paying to retain agency, but do so anyway, presumably for psychological benefit; impacting the construction of their utility function and the way they solve the task goal. Thus, in the normative sense, choosers can be losers, and knowingly so.
However, it is possible that the results are better explained by the fact that people have greater precision over their likelihood of picking the correct shape than over the likelihood of the advisor doing so. Aversion to ambiguity may thus contribute to the preference for control. This means that the exact representation of uncertainty can have an influence on which choice option (the self or the advisor) is opted for. Such a theme is expanded upon in the next chapter where uncertainty in the choice options is modeled under Bayesian statistical assumptions.
With this first study it is possible to see that defining a task goal imposes a cascade of constraints that interact with each other. Here the task goal was to maximise expected utility within a social domain. The specifics of the task involved a choice between two agents where the representations of these agents (choice options in this case) are relevant. Thus, the stimulus formats further constrained the study of choice option representations given the fact that the task goal was already defined. The stimulus format here consisted of verbal and numerical representations regarding the advisor and absent stimulus for the self (provided that this stimulus has already been internalised by the participating agent). In the next chapter, the study will show how modifying assumptions about the task goal (with respect to computation of uncertainty) requires relaxing stimulus formats constraints.