A tale of two theories: A meta-analysis of the attention set and load theories of inattentional blindness.

Abstract

Inattentional blindness (IB), the failure to notice something right in front of you, offers cognitive scientists and practitioners alike a unique means of studying the nature of visual perception. The present meta-analysis sought to provide the first synthesis of the two leading theories of IB—attention set and load theory. We aimed to estimate the magnitude of the effect of each, how they interact, and how task parameters moderate the magnitude of IB summary estimates. We further sought to address several theoretical issues that have persisted within this broad literature. A total of 317 effect sizes from 81 studies that had manipulated attention set or load were synthesized in a multilevel meta-analysis. Results indicated no significant difference between the attention set summary estimate (odds ratio [OR] = 3.26, 95% confidence interval [95% CI] [2.33, 4.57]) and the load summary estimate (OR = 1.75, 95% CI [1.10, 2.79]). Theoretical moderators included a difference between feature attention sets (OR = 5.02, 95% CI [2.95, 8.55]), semantic attention sets (OR = 2.64, 95% CI [1.64, 4.25]), and inherent sets (OR = 1.90, 95% CI [1.35, 2.68]), while perceptual load (OR = 2.55, 95% CI [1.66, 3.92]) and cognitive load (OR = 1.67, 95% CI [1.14, 2.44]) were more comparable. The primary task was found as a key task parameter that moderated summary estimates. The attention set summary estimate was moderated by the number of targets and distractors, whereas the load summary estimate was moderated by the full attention (FA) trial exclusion criterion. Analyses indicated any potential publication bias were overall not likely to impact our conclusions. We discuss the implications of results for a conceptual understanding of IB and how the phenomenon can be more reliably studied in future.

An investigation of inattentional blindness using gaze and frequency tagging.

“…In the classic study of Simons and Chabris (1999), for example, observers are considered to have failed to notice the gorilla because they were too busy counting ball passes. Although recent meta-analytic findings suggest that the primary task is a key moderator of IB (Hutchinson et al, 2022), the primary task is not sufficient for the phenomenon to occur. This is readily observed in almost any IB study: participants tend to perceive the unexpected object immediately following the critical trial, when they have some expectation that the stimulus will appear yet continue to perform the primary task (during the divided attention trial).…”

“…Allocating attention based on features or nonspatial properties of the visual display is also a common manipulation in IB studies. Feature-based attention is the most robust means of eliminating IB, in that the largest difference in noticing rates occurs between conditions when the unexpected object shares features with objects participants are attending to (targets) compared with when it shares features with objects participants are ignoring (distractors) (Hutchinson et al, 2022). Yet sharing features of task properties that observers allocate their attention toward does not eliminate IB entirely, as the phenomenon persists even when the unexpected object featurally matches with targets.…”

“…Adding to its unclear nature, three recent meta-analyses all concluded that the unexpected object is processed with a considerable degree of depth during IB, even when observers fail to notice it (Hutchinson et al, 2022; Kreitz et al, 2020; Nobre et al, 2020). The most notable evidence in support of this conclusion includes that (a) correct guessing of features of the unexpected object is above chance in observers who are considered inattentionally blind (Kreitz et al, 2020; Nobre et al, 2020), (b) unexpected objects can be preferentially boosted into awareness based purely on their semantic properties (even under high load, e.g., Koivisto & Revonsuo, 2009), and (c) IB under some conditions occurs via selective inhibition (Hutchinson et al, 2022).…”

“…Adding to its unclear nature, three recent meta-analyses all concluded that the unexpected object is processed with a considerable degree of depth during IB, even when observers fail to notice it (Hutchinson et al, 2022; Kreitz et al, 2020; Nobre et al, 2020). The most notable evidence in support of this conclusion includes that (a) correct guessing of features of the unexpected object is above chance in observers who are considered inattentionally blind (Kreitz et al, 2020; Nobre et al, 2020), (b) unexpected objects can be preferentially boosted into awareness based purely on their semantic properties (even under high load, e.g., Koivisto & Revonsuo, 2009), and (c) IB under some conditions occurs via selective inhibition (Hutchinson et al, 2022). Not only do these findings demonstrate that some degree of implicit processing of the unexpected object occurs during IB, but also they imply that whatever processing depth the unexpected object reaches is at least partly modulated by top-down attention, given that the object’s semantic properties and selective inhibition are both governed—and hence modifiable—by the observer’s attention set (Hutchinson et al, 2022).…”

Addressing Dependency in Meta-Analysis: A Companion to Assink and Wibbelink (2016)

“…As deciding on the strength of the association(s) between within-study outcomes is challenging, it can be wise to conduct sensitivity analyses with variance-covariance matrices that are based on different estimates of the within-study outcome correlation(s) (see, for instance, Hutchinson et al, 2022;Li et al, 2022;Oliveira et al, 2022, for examples). Building on the outlined example above, the meta-analyst could compute additional matrices based on lower (e.g., r = .2) and higher (e.g., r = .8) constant sampling correlations than the correlation that was initially used in constructing the variance-covariance matrix (i.e., r = .47).…”