Compare

Roitman & Shadlen 2002 (two macaques) and Palmer-Huk-Shadlen 2005 (six human observers) both ran the same signed motion coherence 2AFC at processed-trial level. They use the same canonical axis but different species and different report modalities (saccade vs button-press), so a slope difference is a candidate signature of perceptual sensitivity at the species level.

Both Roitman & Shadlen 2002 and Palmer-Huk-Shadlen 2005 also report median response time as a function of absolute coherence. The chronometric speed-accuracy tradeoff has been used to argue for shared bounded-accumulation dynamics across species, but the absolute time scales differ.

No 4-parameter logistic fit applies (curve type does not support it, or fits failed). Visual comparison only.

Brunton et al. 2013 trained five rats on the rat auditory clicks task and the London 2018 Mendeley release ran the same Poisson clicks accumulation paradigm in humans (DBS off baseline shown here for like-for-like comparison). Both pin to the same canonical axis — signed click-count difference (right minus left) → p_right — so this is one of the cleanest cross-species comparisons in the atlas. Differences in slope or lapse rate cannot be attributed to differing stimulus families.

The IBL trainingChoiceWorld biased variant cycles through three blocks of leftward prior probability (0.2 / 0.5 / 0.8). A bias-shift across blocks is the textbook signature of prior integration; the slope (σ) should remain stable.

Walsh et al. 2024 manipulate prior probability via cues that either match the upcoming target side (cue=valid), give no information (cue=neutral), or actively mislead (cue=invalid). Two canonical ways the cue could enter a DDM are (a) shifting the starting point z toward the favored boundary, or (b) adding a constant drift offset v0 in the favored direction. We fit both four-parameter variants to each pooled per-cue psychometric and compare AICs. Lower AIC = better-supported variant for that cue.

No 4-parameter logistic fit applies (curve type does not support it, or fits failed). Visual comparison only.

Roitman and Palmer fit the same canonical psychometric + chronometric curves on the random-dot motion task, on the same x-axis (signed motion coherence in percent), in different species. Vanilla DDM fits recover three parameters per (paper, subject) — drift per unit evidence (k), boundary separation (a), and non-decision time (t0). This comparison aggregates all 10 fits (Roitman: 1 pooled + 2 macaques; Palmer: 1 pooled + 6 humans) so the spread of k across species is directly visible.

No 4-parameter logistic fit applies (curve type does not support it, or fits failed). Visual comparison only.

Walsh et al. 2024 manipulate trial-by-trial prior cues (valid / neutral / invalid). The prediction is that valid cues produce a bias toward the cued side without changing slope, and invalid cues produce the opposite shift.