The distinct development of stimulus and response serial dependence

Abstract

Serial dependence (SD) is a phenomenon wherein current perceptions are biased by the previous stimulus and response. This helps to attenuate perceptual noise and variability in sensory input and facilitates stable ongoing perceptions of the environment. However, little is known about the developmental trajectory of SD. This study investigates how the stimulus and response biases of the SD effect develop across three age groups. Conventional analyses, in which previous stimulus and response biases were assessed separately, revealed significant changes in the biases over time. Previous stimulus bias shifted from repulsion to attraction, while previous response bias evolved from attraction to greater attraction. However, there was a strong correlation between stimulus and response orientations. Therefore, a generalized linear mixed-effects (GLME) analysis that simultaneously considered both previous stimulus and response, outperformed separate analyses. This revealed that previous stimulus and response resulted in two distinct biases with different developmental trajectories. The repulsion bias of previous stimulus remained relatively stable across all age groups, whereas the attraction bias of previous response was significantly stronger in adults than in children and adolescents. These findings demonstrate that the repulsion bias towards preceding stimuli is established early in the developing brain (at least by around 10 years old), while the attraction bias towards responses is not fully developed until adulthood. Our findings provide new insights into the development of the SD phenomenon and how humans integrate two opposing mechanisms into their perceptual responses to external input during development.

Keywords: Attractive bias; Development; Perception; Repulsive bias; Serial dependence.

Fig. 1

Example trial sequence on the orientation reproduction task. Participants viewed a Gabor patch presented in the center of a screen and subsequently reproduced the perceived orientation by adjusting the orientation of an imaginary line connecting two dots framed by a circle. ITI = intertrial interval

Fig. 2

Serial dependence calculated using conventional analysis in one example participant. A Stimuli and responses in successive trials. Relative orientation of previous stimulus: previous Gabor orientation minus current Gabor orientation. Relative orientation of previous response: previous response orientation minus current Gabor orientation. Response error: current response orientation minus current Gabor orientation. B Illustration of the response error in relation to the relative orientation of the previous stimulus for one participant. The slope of the linear function is the participant’s regression coefficient of previous stimulus. C Illustration of the response error in relation to the relative orientation of the previous response for one participant. The slope of the linear function is the participant’s regression coefficient of the previous response. (Color figure online)

Fig. 3

Results of conventional data analysis of the performance of three age groups on the orientation reproduction task. A The regression coefficients of the previous stimulus on current response errors. A positive regression coefficient indicates that the current response is biased towards the previous stimulus orientation (an attractive effect), while a negative regression coefficient indicates a bias away from the previous stimulus orientation (a repulsive effect). Also, the greater the absolute value of the regression coefficient, the greater the attractive/repulsive SD. B The regression coefficients of the previous response on current response errors. In each subgraph, the dark gray line and light gray shadow represent the mean values (e.g., mean regression coefficients of previous stimulus for Fig. 3A; mean regression coefficients of previous response for Fig. 3B) and their corresponding 95% CIs, respectively, calculated from 1,000 shuffling iterations. Error bars represent ±1 standard error. (Color figure online)

Fig. 4

Results of joint bias analysis of the performance of three age groups on the orientation reproduction task. A Average joint bias map of the averages for each age group. Response errors are plotted as functions of the relative orientation of the previous stimulus and response (compared with the current stimulus). The colors represent the value of response errors. The positive values of relative orientations and response errors are represented as counterclockwise differences and the negative values as clockwise differences. More details are given in the Methods section. B Response errors are plotted as functions of the relative orientation of the previous stimulus conditioned by the relative orientation of the previous response (left) and as functions of the relative orientation of the previous response conditioned by the relative orientation of the previous stimulus (right). The data plotted are from the area inside the red dashed line box in the joint bias maps shown in (A). A positive slope indicates that the current response is biased towards the previous stimulus (or response) orientation (an attractive effect), while a negative slope indicates a bias away from the previous stimulus (or response) orientation (a repulsive effect). Also, the greater the absolute value of the slope, the greater the bias. Error bars represent ±1 standard error. (Color figure online)

Fig. 5

Results of generalized linear mixed-effects analysis of the performance of three age groups on the orientation reproduction task. A Model comparison results. The ΔBIC of Model 1 (previous stimulus) and Model 2 (previous response) compared with that of Model 3 (previous stimulus + previous response). Positive ΔBIC values indicate the model is worse than Model 3, and negative ΔBIC values indicate the model is a better fitting model than Model 3. B The regression coefficients for the previous stimulus and response. Negative regression coefficient values represent repulsive bias, and positive values represent attractive bias. Moreover, the larger the absolute value of the regression coefficient is, the larger the bias is. In each subgraph, the dark gray line and light gray shadow represent the mean values (e.g., mean ΔBICs for Fig. 5A; mean regression coefficients for Fig. 5B) and their corresponding 95% CIs, respectively, calculated from 1,000 shuffling iterations. Error bars represent ±1 standard error. ΔBIC = delta Bayesian information criterion. (Color figure online)