A safety committee reviews the risks in a manufacturing facility. Two risks are on the table. The first is a rare type of accident that, when it occurs, is dramatic and highly visible - a colleague was seriously injured in one last year. The second is a cumulative risk from ambient noise exposure, which affects a much larger proportion of workers over time and is estimated to cause significantly more total harm.
The committee allocates substantially more resources to addressing the first risk.
The data supports the opposite allocation. But the dramatic, recent, visible incident has produced a salience that the cumulative, invisible risk does not have. That salience is doing work that the data should be doing.
What Salience Bias Means
Salience bias is the tendency to assign disproportionate weight to information that is perceptually or emotionally prominent - vivid, dramatic, unusual, or recent - relative to information that is equally or more relevant but less immediately striking.
The information is not being ignored. In the committee example, the noise risk data is present and reviewed. But the dramatic accident generates an emotional weight that the ambient risk data does not. When forming the final judgment, the striking information exerts more influence than its objective evidential value warrants.
How It Works: The Mechanism
The mechanism is attentional enhancement followed by memory amplification. Stimuli that are high in contrast, novelty, or emotional arousal are detected faster, processed more thoroughly, and encoded more durably than stimuli that are unremarkable. They are also more readily available when recalling information for a judgment.
This produces a compounding effect: salient information is noticed first, encoded most deeply, and retrieved most readily. At every stage of processing, it accumulates advantage over less salient information. By the time a judgment is being formed, the salient input has been mentally rehearsed multiple times while the less salient input has been processed once at a shallow level.
The result is not that less salient information is absent from the judgment. It is that it is underweighted relative to information of equal or greater evidential value that happened to present more vividly.
Why This Matters
In risk assessment, salience bias causes overreaction to dramatic but rare events and underreaction to common but less visible risks. Spectacular incidents receive extensive resources; quiet, chronic harms receive insufficient attention. This creates predictable misallocation in safety, public health, and environmental policy.
In media and public opinion, salience bias means that the distribution of public attention and concern tracks the vividness of events rather than their magnitude. A single identifiable person in a dramatic situation generates more response than a statistical description of a much larger number of people facing equivalent or greater harm.
In personal decisions, vivid personal experiences carry more influence in decision-making than abstract data about base rates. Someone who has a friend who benefited from a particular treatment will rate that treatment more favourably than the statistical evidence warrants. Someone who has a vivid memory of a near-miss will overestimate the probability of that type of event occurring.
The Common Misunderstanding
Salience bias is not the same as novelty-seeking - a general preference for new experiences. It is specifically about the disproportionate influence of vivid, emotionally salient information on judgment, regardless of its actual relevance.
A second common misunderstanding is that making information more vivid is therefore always manipulative or problematic. This misses the dual nature of the effect: salience bias can be addressed by either reducing the overweighting of dramatic information or by increasing the salience of important information that is currently underweighted. Making accurate risk information more vivid, concrete, and emotionally accessible can improve decision quality when the relevant information is the underpowered input.