The hindsight bias may not be as robust as previously believed. Also known as the “knew-it-all-along effect” (Fischhoff, 1975) the hindsight bias refers to the inability for individuals to remember their previous state of knowledge after learning an outcome. Researchers have found people fall victim to hindsight biases in a variety of domains, including general knowledge (Fischoff, 1977; Wood, 1978), medical decisions (Arkes, Wortmann, Saville, & Harkness, 1981), and political outcomes (Fischhoff & Beyth, 1975). Multiple failures to debias, with methods including informing participants about the bias (Fischhoff, 1977) and manipulating participant perspective (Wood, 1978) further grew the reputation for the strength of this bias. However, we must interpret older results in the behavioral sciences with new perspective on the importance of sample size and statistical power (Ioannidis, 2005). This dissertation proposes to investigate when and why hindsight bias fails to appear, and what it means for the psychological processes underlying the hindsight bias, and more broadly, the interpretation of older psychological research.

People often report confidence in two stages: first, they select an answer; second, they report how confident they are that that answer is correct. But research suggests an alternative: one- stage elicitations ask for a belief distribution over the entire set of answers. This dissertation compares these two elicitation methods. In so doing, I shed light on to the psychology of confidence and provide guidance for survey design.I find across seven studies that one-stage elicitation formats lead to higher average peak confidence than two-stage elicitation formats and that difficulty moderates this relationship, such that the effect is stronger when the question posed is difficult. Familiarity with elicitation format does not attenuate the effect. The effect remains stable across different stimuli, and when participant decisions are incentive compatible. In examining why this effect occurs, I find that it is driven by two-stage responses that violate rationality. Across three of the four pre-registered studies, I find that the two-stage responses that indicate less than 50% confidence in the selected answer are driving the main result. When these responses are filtered out, the difference in peak confidence between elicitation formats disappears, as does the moderating effect of question difficulty. Our confidence in our ability to select the right option has enormous consequences. When we are sure, we can commit without hesitation. When we are unsure, then it might be wise to hedge our bets, to delay, or to gather more information. By gaining insights into how people form and communicate confidence in their judgments and identifying effective confidence elicitation techniques, decision-makers can make more informed and effective decisions, leading to better outcomes for organizations and individuals alike.

Algorithms, scripts for sequences of mathematical calculations or procedural steps, are powerful. Even though algorithms often outperform human judgment, people appear resistant to allowing a numerical formula to make decisions for them (Dawes, 1979). Nevertheless, people are increasingly dependent on algorithms to inform their decisions on a day-to-day basis. In eight experiments, I tested whether aversion to algorithms is as straightforward a story as past work suggests. The results shed light on the important questions of when people rely on algorithmic advice over advice from people and have implications for the use of algorithms within organizations.