Verbal and Numeric Probabilities Differentially Shape Decisions

Abstract

Decision-makers in diverse fields ranging from medicine to law to public policy depend on the clear and unbiased communication of probabilities. Expert judges often express probabilities verbally (e.g., unlikely) rather than numerically (e.g., 25% chance). Researchers have criticized the practice because verbal probabilities are vague and imprecise. However, little work has investigated the pragmatic consequences of their usage. In three online experiments (Ns = 201, 439, 435), we showed that the directionality of verbal probabilities leaked information about experts’ expectations. Receivers interpret these inferred expectations as recommendations. The inferred recommendation, in turn, influences their decisions. Compared to numeric probabilities, the directionality of verbal probabilities makes their interpretation more ambiguous due to pragmatic implicatures. We discuss how senders could improve communication fidelity by using numeric probabilities as a principal form of communicating probabilistic information, as their directionality and interpretation are less ambiguous and more consistent.

Meta‐informational cue inconsistency and judgment of information accuracy: Spotlight on intelligence analysis

“…The same applies to the credibility scale. Directionality is a pragmatic characteristic of linguistic quantifiers (Budescu et al, 2003;Teigen & Brun, 1995, 1999 that has implications for inferred recommendations and decisions (Collins et al, 2022) and for perceived trust and confidence in advisors (Collins & Mandel, 2019;Jenkins et al, 2018) levels such that levels A-C and levels 1-3 belong to a positive group of ratings and levels D and E and levels 4 and 5 belong to a negative group of ratings. Therefore, we predicted that the largest gap in judged accuracy between adjacent levels of both scales will lie between the third and fourth levels (Hypothesis 2).…”

Communicating uncertainty in national security intelligence: Expert and nonexpert interpretations of and preferences for verbal and numeric formats

“…Numeric probabilities can also convey directionality, but even for probabilities below the 50% midpoint, they tend to convey positive directionality (Teigen & Brun, 2000). Therefore, numeric probabilities are less likely than verbal probabilities to be used to imply recommendations for action (Collins & Mandel, 2019;Collins et al, 2022) and, accordingly, numeric probabilities are less likely to undermine the requirement for policy neutrality (Kent, 1951). Simply put, the use of verbal forms of uncertainty communication can undermine communication fidelity where the primary aim of such communication is to convey the level of uncertainty associated with events or hypotheses being judged Friedman, 2019;Mandel & Irwin, 2021c;Morgan, 1998).…”

Meta-informational Cue Inconsistency and Judgment of Information Accuracy: Spotlight on Intelligence Analysis

“…In the present context, directionality appears to impose a grouping of scale levels such that A-C and 1-3 belong to a positive group and D and E and 4 and 5 belong to a negative group with "cannot be judged" representing a directionally neutral zone between these groups. Since negatively directional terms are often interpreted as "recommendations against" (Collins et al, 2022), labeling sources of information with negatively directional terms may prompt the receiver to infer that the sender is recommending not to use that source or that information. This potential implication ought to be addressed in future research.…”

“…The same applies to the credibility scale. Directionality is a pragmatic characteristic of linguistic quantifiers (Budescu et al, 2003;Teigen & Brun, 1995, 1999 that has implications for inferred recommendations and decisions (Collins et al, 2022) and for perceived trust and confidence in advisors (Collins & Mandel, 2019;Jenkins et al, 2018) levels such that levels A-C and levels 1-3 belong to a positive group of ratings and levels D and E and levels 4 and 5 belong to a negative group of ratings. Therefore, we predicted that the largest gap in judged accuracy between adjacent levels of both scales will lie between the third and fourth levels (Hypothesis 2).…”