**The internal representation of stimuli is imperfect and subject to bias. Noise introduced at initial encoding and during maintenance degrades the precision of representation. Stimulus estimation is also biased away from recently encountered stimuli, a phenomenon known as adaptation. Within a Bayesian framework, greater biases are predicted to result from poor precision. We tested for this effect on individual difference measures. Through an online experiment, 202 subjects contributed data. During separate face and color blocks, they performed three different tasks: an immediate stimulus match, a delayed match-to-sample, and a delayed match following 5 s of adaptation. The stimulus spaces were circular, and subjects entered their responses on a color/face wheel. Bias and precision of responses were extracted while accounting for the probability of random guesses. We found that the adaptation manipulation induced the expected bias in responses, and the magnitude of this bias varied reliably and substantially between subjects. Across subjects, there was a negative correlation between mean precision and bias. This relationship was replicated in a new experiment with 192 subjects. This result is consistent with a Bayesian observer model, in which the precision of perceptual representation influences the magnitude of perceptual bias.**

*SD*= $0.51).

*near*the target value. The parameters fitted by this procedure correspond to the probability of guesses, the precision of responses (the inverse of the standard deviation of the best-fitting Gaussian distribution), and the bias, which is the mean of the von Mises distribution centered near the target value (Figure 2b).

*ε*=

*θ*

_{target}−

*θ*

_{response}), and we fitted the distribution of error values for each subject using a superposition of distributions defined over a circular support (−180° <

*θ*< +180°; Zhang & Luck, 2008; Bays, Catalao, & Husain, 2009). This procedure simultaneously estimates the precision of the error distribution (i.e., the inverse of the standard deviation of the best-fitting Gaussian distribution; Jammalamadaka & Sengupta, 2001) and the probability of random guesses (Figure 2a).

*ε*) with the distribution of reflected error values (−

*ε*) from trials where the adaptor was located at +45° from the target. In these experiments, the fitting procedure also estimates two additional parameters: the mean of the error distribution (i.e., the bias induced by the paradigm) and the probability that the selected response matches the adaptor stimulus and not the target stimulus (Figure 2b and 2c).

*repulsive*aftereffects, in which subject responses to target stimuli tend to be biased

*away*from the preceding, adapting stimulus (Table 4).

*F*(1, 201) = 0.0160,

*p*= 0.9, suggesting that the magnitude of adaptation biases is a stable individual characteristic for the duration of the experiment. For that reason, we again fitted the response bias using data from both blocks of each experiment. We then performed a repeated-measures ANOVA to identify the sources of variability in adaptation bias. We observed a small but significant effect of stimulus class,

*F*(1, 201) = 34.9,

*p*< 0.001, explaining only 6% of the total variance. The effect of mean subject bias was more substantial,

*F*(201, 201) = 1.56,

*p*< 0.001, accounting for 57% of the total variance in bias values. The distribution of individual subject bias was well fitted with a Gaussian with a mean of 8.0° for colors (95% confidence interval [CI] [0.2°, 18.8°]) and 5.3° for faces (95% CI [−4.9°, 17.1°]). These results are consistent with an individual difference in adaptation bias that is present across face and color stimuli.

*F*(1, 201) = 0.86,

*p*= 0.36, we combined data from both blocks of each experiment. We then performed a repeated-measures ANOVA to identify the sources of variability in response precision. We observed a significant effect of stimulus class,

*F*(1, 201) = 170.1,

*p*< 0.001, accounting for 16.5% of the total variance, with response precision being higher for colors. We also observed a significant effect of mean subject precision,

*F*(201, 201) = 3.28,

*p*< 0.001, accounting for 64% of the total variance. Furthermore, response precision estimated from stimulus-match trials was well correlated with precision estimated from the adaptation experiment (color:

*r*= 0.45,

*p*< 0.001; faces:

*r*= 0.44,

*p*< 0.001; Figure 4a and 4b). This correlation is possibly due to shared sources of variability between experiments. Note, however, that the variance associated with reporting the target on stimulus-match trials is smaller than that associated with reporting the target in the adaptation experiment, possibly due to the fact that the latter is presented briefly and involves a short-term-memory component.

*ρ*= −0.26, jackknife resampling 95% CI [−0.28, −0.25],

*p*< 0.001; faces:

*ρ*= −0.13, jackknife resampling 95% CI [−0.15, −0.12],

*p*< 0.001 (Figure 4c and 4d). This suggests that subjects with lower representation precision exhibit larger biases away from the adapting stimulus, consistent with predictions from a Bayesian observer model.

*r*= 0.22, jackknife resampling 95% CI [−0.21, −0.23],

*p*< 0.001. A possible source for this relationship is the shared effect of individual differences in precision. We cannot, however, be certain that this is the entire explanation, as the correlation between face and color bias values remains in partial correlations that attempt to account for individual variation in precision (partial Pearson correlation),

*r*= 0.18, jackknife resampling 95% CI [−0.17, −0.19],

*p*< 0.001. We considered the possibility that individual differences in reaction time could account for individual differences in adaptation magnitude. The magnitude of perceptual bias decays following cessation of the adaptor (Greenlee, Georgeson, Magnussen, & Harris, 1991). Perhaps subjects who respond quickly tend to experience more bias due to less decay. However, this prediction was not confirmed. For faces, we found no significant relationship between response bias and average response time (Spearman's rank correlation),

*ρ*= 0.07,

*p*= 0.30. For colors, the opposite pattern was found: Subjects who responded quickly experienced less bias (Spearman's rank correlation),

*ρ*= 0.20,

*p*= 0.0047. We do not have a specific mechanism (neural or otherwise) to offer as the basis for this small correlation in induced adaptation bias between materials across subjects.

*F*(5, 950) = 3.72,

*p*= 0.002, and subject explained 53.5%,

*F*(190, 950) = 5.873,

*p*< 0.001. We then analyzed the sources of variability in adaptation bias. Similarly, experimental block explained 0.8% of variance,

*F*(5, 950) = 2.75,

*p*= 0.02, and subject explained 47.1%,

*F*(190, 950) = 4.51,

*p*< 0.001. To investigate whether the inverse relationship between representation precision and adaptation biases was replicated, we extracted the bias and precision values for each subject using data from all blocks and calculated the correlation between these two quantities. We again observed that, across subjects, there was a negative correlation between mean precision and bias—color:

*ρ*= −0.22, jackknife resampling 95% CI [−0.25, −0.20],

*p*< 0.001; faces:

*ρ*= −0.19, jackknife resampling 95% CI [−0.22, −0.17],

*p*< 0.001 (Figure 6d and 6e).

*, 5 (8), 617–629.*

*Nature Reviews Neuroscience**, 370–418.*

*Philosophical Transactions of the Royal Society of London, 53**, 9 (10): 7, 1–11, https://doi.org/10.1167/9.10.7. [PubMed] [Article]*

*Journal of Vision**, 22 (7), 622–626.*

*Current Biology**, 6 (3), 136–143.*

*Trends in Cognitive Sciences**, 17 (5), 738–743.*

*Nature Neuroscience**. Cambridge, UK: Cambridge University Press.*

*Statistical analysis of circular data**, 20 (5), 453–467.*

*Journal of Experimental Psychology**, 31 (2), 223–236.*

*Vision Research**. City, ST: World Scientific.*

*Topics in circular statistics*(Vol. 5)*. City, ST: Cambridge University Press.*

*Perception as Bayesian inference**, 25 (13), R545–R546.*

*Current Biology**, 16 (7), 779–781.*

*Vision Research**, 40 (1), 49–70.*

*International Journal of Computer Vision**, 9 (4), 578–585.*

*Nature Neuroscience**, 18, 1289–1296.*

*Advances in Neural Information Processing Systems**, 13 (9), 403–409.*

*Trends in Cognitive Sciences**, 50 (15), 1452–1461.*

*Vision Research**. New York, NY: Dover. (Original work published 1866)*

*Treatise on physiological optics. Vol. III**, 18 (10), 1509–1517.*

*Nature Neuroscience**, 3, e1058.*

*PeerJ**, 453 (7192), 233–235.*

*Nature*