Fig. 1: Reduced working memory-dependent serial dependence in anti-NMDAR encephalitis and schizophrenia.

a In each trial, subjects were to remember a stimulus that appeared for 0.25 s at a randomly chosen circular ___location with fixed distance from the center. Delay lengths varied randomly between trials (0, 1 or 3 s). Subjects made a mouse click to report the remembered ___location and started the next trial by moving the mouse back to the screen’s center during the inter-trial-interval (ITI). b Serial dependence is measured as a systematic shift of responses towards previous target locations. Attractive effects depend on the distance θ^d between previous and current stimulus. c Precision for each subject and delay was inversely estimated as the circular s.d. of bias-corrected error distributions (“Methods”). For longer delays, participants’ responses were less precise (delay, F(2,147) = 76.87, p < 1e−16). There were no overall or delay-dependent group differences in precision (group, F(2,147) = 1.74, p = 0.18; group × delay, F(4,147) = 0.07, p = .99, all p-values from ANOVA). Error bars indicate ±s.e.m. d–f, Serial dependence by group and delay length. Serial dependence is calculated as the ‘folded’ error \(\theta ^{e\prime }\) for different θ^d (dashed lines; “Methods”). Solid lines show linear model fits (“Methods”), omitting intercepts and negative values of θ^d. Shading, ±s.e.m. across pooled trials from n = 19 healthy controls (ctrl), n = 17 patients with schizophrenia (schz), and n = 16 patients with anti-NMDAR encephalitis (enc). g–i Individual (random coefficients; dots) and group estimates of serial bias strength (fixed effects; error bars indicate mean and bootstrapped 95% C.I. of the mean) by delay. g Serial dependence was repulsive in 0 s trials (DoG(θ^d), F(1,52) = 12.67, p = 0.0008), independently of group (group × DoG(θ^d), F(2,52) = 0.46, p = 0.63). h For 1 s trials, group differences in serial dependence emerged (group × DoG(θ^d), F(2,48) = 6.52, p = 0.003) between ctrl and schz (t = 3.73, p = 7e−4, Cohen’s d = 1.28) and enc and schz (t = 2.73, p = 0.01, Cohen’s d = 0.98). i After 3 s delay, both patient groups showed reduced biases compared to ctrl (group × DoG(θ^d), F(2,50) = 15.35, p = 6e−5; ctrl vs enc, t = 4.14, p = 2e−4, Cohen’s d = 1.45; ctrl vs schz, t = 6.44, p = 2e−7, Cohen’s d = 2.21, and enc vs schz, t = 3.40, p = 0.002, Cohen’s d = 1.22). All t-tests, two-sided. In all panels, single data points show data from n = 19 healthy controls (ctrl), n = 17 patients with schizophrenia (schz), and n = 16 patients with anti-NMDAR encephalitis (enc).