Skip to main page content
U.S. flag

An official website of the United States government

Dot gov

The .gov means it’s official.
Federal government websites often end in .gov or .mil. Before sharing sensitive information, make sure you’re on a federal government site.

Https

The site is secure.
The https:// ensures that you are connecting to the official website and that any information you provide is encrypted and transmitted securely.

Access keys NCBI Homepage MyNCBI Homepage Main Content Main Navigation
. 2018 Sep 18;115(38):E8825-E8834.
doi: 10.1073/pnas.1805224115. Epub 2018 Aug 30.

Gain control explains the effect of distraction in human perceptual, cognitive, and economic decision making

Vickie Li  1 , Elizabeth Michael  2 , Jan Balaguer  3 , Santiago Herce Castañón  3   4 , Christopher Summerfield  3
Affiliations

Gain control explains the effect of distraction in human perceptual, cognitive, and economic decision making

Vickie Li et al. Proc Natl Acad Sci U S A. .

Abstract

When making decisions, humans are often distracted by irrelevant information. Distraction has a different impact on perceptual, cognitive, and value-guided choices, giving rise to well-described behavioral phenomena such as the tilt illusion, conflict adaptation, or economic decoy effects. However, a single, unified model that can account for all these phenomena has yet to emerge. Here, we offer one such account, based on adaptive gain control, and additionally show that it successfully predicts a range of counterintuitive new behavioral phenomena on variants of a classic cognitive paradigm, the Eriksen flanker task. We also report that blood oxygen level-dependent signals in a dorsal network prominently including the anterior cingulate cortex index a gain-modulated decision variable predicted by the model. This work unifies the study of distraction across perceptual, cognitive, and economic domains.

Keywords: anterior cingulate cortex; cognitive control; decoy effects; gain control; tilt illusion.

PubMed Disclaimer

Conflict of interest statement

The authors declare no conflict of interest.

Figures

Fig. 1.
Fig. 1.
The effect of distraction across perceptual, cognitive, and economic domains. (A, i) Participants were asked to discriminate the tilt (relative to horizontal) of a central Gabor surrounded by tilted distracters. (A, ii) Participants were biased to report the target as more clockwise when the flankers were counterclockwise, and vice versa (the “tilt illusion”). Panels i and ii republished with permission of Royal Society, from ref. ; permission conveyed through Copyright Clearance Center, Inc. (A, iii) Simulation of the adaptive gain model replicates the qualitative tilt illusion pattern in ref. and replicates the current human data both qualitatively and quantitatively (SI Appendix, Fig. S7). It further predicts that the magnitude of the bias is modulated by flanker variance; colored lines reflect flanker variabilities from low (red) to high (blue). (B, i) In the Eriksen flanker task, participants respond with a key press to a central letter while ignoring the flankers. (B, ii) RTs are the fastest on CO trials, then the SI trials, and slowest on RI trials. Panel ii republished with permission of MIT Press - Journals, from ref. ; permission conveyed through Copyright Clearance Center, Inc. See Methods for details of how CO, RI, and SI trials were defined. (B, iii) The adaptive gain model predicts the same pattern of reaction time across the three conditions. (C, i) Participants chose the most preferred of three food items. (C, ii) Increasing the value of the least-preferred item reduces the choice efficiency (i.e., probability of choosing the highest-valued target) as the normalized distractor value increases, shown by logistic slope from fitting logistic choice functions on humans choice. Panel ii adapted with permission from ref. . (C, iii) Using the adaptive gain model, we simulated the normalized subjective difference between two options X and Y (blue and red line) as a function of a third decoy Z (x axis). The subjective difference is first reduced and the increased in a qualitatively similar fashion.
Fig. 2.
Fig. 2.
The value difference between two equally preferred options (X and Y, red and blue dots) as a function of a third decoy option, Z (x and y axes). The axes correspond to the attribute values (e.g., inverse price [P] and quality [Q]). X and Y fall on the indifference line (dashed line). The red dot signals option X (where XP=15,XQ=10), and the blue dot indicates option Y (where YP=10,YQ=15). The colored surface shows the model-predicted subjective value difference X^−Y^ that is, the extent to which X is preferred over Y (red regions), and conversely the blue region in which the subjective value of X is smaller than Y, and thus Y is preferred. We have superimposed example decoy options (green dots) that produce the three context effects (compromise, C; attraction, A; and similarity, S; subscript corresponds to a preference for either option X or option Y). a.u., arbitrary units.
Fig. 3.
Fig. 3.
Relative choice shares of the target for each pair decoy types. (A–D) Reprinted with permission from ref. . (E–H) Simulations from the adaptive gain model. (E) An illustration of the simulated 2D attribute space with target and decoys. The target (red dot) is set to be {15,10}, and the competitor (gray dot) is {10,15}. Both lie on the indifference line (dashed line). The three types of decoy and their values are plotted in three different colored circles (green: similarity decoys, value = {8.5,16.5}; magenta: attraction decoy, value = {13.5,8.5}; cyan: compromise decoys, value = {20,5}). DTC refers to the distance between the target and the competitor. (F–H) Pairwise correlations among the strength of effect for each decoy type across the simulated cohort. Each unique colored dot corresponds to one simulated participant. The color corresponds to the tuning width σmax (red = low tuning width σmax, blue = high tuning width). The black line on each panel corresponds to the best least-squares fit. F shows a positive correlation between the compromise effect and the attraction effect across the population. G and H show a negative correlation between the similarity effect and the attraction or the compromise effect across the population. Across panels, we can see that participants who display a strong attraction effect also display a strong compromise effect but a weaker similarity effect, whereas those who display a weaker similarity effect display stronger compromise and attraction effects. This is consistent with the data displayed in B–D from ref. .
Fig. 4.
Fig. 4.
Relative choice preference for the target, competitor, and decoy options under time pressure. (A) The target and competitor option is presented with an attraction decoy. (B) The target and competitor option is presented with a compromise decoy. (Left) Subjects’ choice pattern under different deliberation time conditions. A and B, Left panels adapted with permission from ref. . (Right) Simulations from the adaptive gain model. Decreasing noise (comparable to increasing deliberation time) would increase the preference for the target option relative to the competitor option.
Fig. 5.
Fig. 5.
Adaptive gain model of the Eriksen flanker task. (A) Illustration of the adaptive gain model as applied to Exp. 1. The model assumes that the population width tuning envelope is governed by the trial flanker mean (Xj¯) and trial flanker variability (sXj). Congruent targets (red circles) receive higher neural gain than incongruent targets (blue circles), leading to faster RTs. The model further predicts an interaction between congruency and flanker variability. Congruent targets receive higher neural gain under low flanker variability trials (large circles) than under high flanker variability trials (small circles), while incongruent targets under the two flanker variability trials received the same low level of neural gain, meaning that flanker variability has no influence on RT for incongruent targets. (B) Task schematics. Participant first saw a fixation dot, followed by an array that contained a central target and six surrounding flankers. They responded whether the central target was clockwise or counterclockwise to the vertical axis, receiving feedback after each response. (C) Simulation of RT (1/|Xi^|) using the adaptive gain model for three levels of flanker variance (sXj, in colored lines) and congruency (x axis). (D and E) Human mean reaction time pattern (lines) under three levels of flanker variance and congruency condition in Exp. 1a (D) and Exp. 1b (E). Both interactions (congruency × flanker variance) are significant at P < 0.001. Colored circles represent the fitted mean reaction time from the adaptive gain model. Error bars show the SEM.
Fig. 6.
Fig. 6.
Model predictions and human data for Exps. 2a and 2b. (A) Surface plots showing the mean RT pattern in humans under different conditions (three levels of target orientation × three levels of flanker mean orientation × congruency × three levels of flanker variability). Warmer colors correspond to longer RTs. There is an overall cost when the target is close to the category boundary (top row of each surface plot). There was an additional cost when these targets were flanked by congruent flankers that are further from the boundary (top left corner of each subplot). The introduction of higher flanker variability reduces these additional costs in those conditions (overall faster RTs across surface plots). (B) Fitted mean RT pattern from the adaptive gain model. (C) Mean (±SEM) RT in humans were cross-plotted against the fitted mean (±SEM) model RT for each condition. Warm colors (red to yellow) correspond to levels of mean orientation from congruent flankers. Cold colors (blue to cyan) correspond to levels of mean orientation from incongruent flankers |Xj¯|. Different shapes correspond to three levels of target decision variable |Xi|.
Fig. 7.
Fig. 7.
Effect of target orientation on BOLD signal. (A) Brain areas correlating negatively with the absolute target decision variable, rendered onto a template brain in sagittal (Left), coronal (Middle), and axial (Right) slices. Images were generated with an uncorrected threshold of P < 0.0001. (B) Parametric modulators Xi and Xj¯ were each split into four quartiles (lines for Xi, x axis and shaded area for Xj¯). Mean predictions (±SEM) from different models: reciprocal gain-modulated decision variable 1/X^i (Left). Reciprocal no-gain modulated target decision variable 1/X^i (Middle). Estimated conflict was plotted for each quartile (Right). Colored lines correspond to four levels of Xi. Shaded colored background (groupings on x axis) corresponds to four levels of Xj¯. (C) Mean BOLD (±SEM) signal beta values for each level of Xi and Xj¯ from the quartile bins. (Left) dACC, (Middle) AIC, and (Right) SPL ROIs.
Fig. 8.
Fig. 8.
(A) Model comparisons across GLM models. Expected frequencies for each GLM and each ROI were computed using random effects Bayesian model selection. Gray dotted line displayed the chance level see SI Appendix, SI Materials and Methods for the definition of the models. (B) ROI analysis on dACC, AIC, and SPL on the absolute flanker mean decision variable |Xj¯|. Bar plots correspond to the mean (±SEM) beta estimates for this predictor from GLM4. (C) Simulated (negative) model output as a function of the value of chosen and unchosen option (see Methods for details).
See this image and copyright information in PMC

Similar articles

See all similar articles

Cited by

See all "Cited by" articles

References

    1. Luce RD. Individual Choice Behavior: A Theoretical Analysis. Wiley; New York: 1959.
    1. von Neumann J, Morgenstern O. Theory of Games and Economic Behavior. Princeton Univ Press; Princeton: 1944.
    1. Savage LJ. The Foundations of Statistics. Wiley; New York: 1954.
    1. Wald A, Wolfowitz J. Bayes solutions of sequential decision problems. Proc Natl Acad Sci USA. 1949;35:99–102. - PMC - PubMed
    1. Ma WJ, Beck JM, Latham PE, Pouget A. Bayesian inference with probabilistic population codes. Nat Neurosci. 2006;9:1432–1438. - PubMed

Publication types

LinkOut - more resources