Skip to main content
. 2023 Aug 2;43(31):5693–5709. doi: 10.1523/JNEUROSCI.2101-22.2023

Table 2.

Overview of computational model fits to each dataseta

Model number N model parameters per mouse Forgetting function Retrospective WM Prospective WM Side biases Distal-response biases Dataset 1
Dataset 2
Dataset 3
WAIC ΔWAIC (SE) WAIC ΔWAIC (SE) WAIC ΔWAIC (SE)
M1 3b Power-law 71 495.2 15 678.6 (362.0) 7940.5 584.8 (74.3) 3807.6 441.9 (76.1)
M2 2 Power-law 71 471.2 15 654.7 (361.1) 8491.5 1135.8 (103.5) 6753.4 3387.6 (359.0)
M3 1 NA 66 937.5 11 120.9 (370.9) 14 189.0 6833.3 (300.2) 10 405.6 7039.8 (576.7)
M4 1 NA 76 250.8 20 434.3 (422.7) 14 321.4 6965.7 (296.0) 10 502.6 7136.8 (571.0)
M5 4 Power-law 56 025.8 211.3 (65.5) 7378.2 22.5 (15.7) 3812.1 446.3 (80.9)
M6 5b Power-law 57 088.1 1271.6 (117.6) 7511.2 155.5 (36.7) 3382.2 16.5 (18.5)
M7 5b Power-law 68 261.1 12 444.5 (335.2) 7771.4 415.7 (61.4) 3707.8 342.0 (69.7)
M8 5b Power-law 58 258.4 2441.9 (148.0) 7511.9 156.2 (36.7) 3401.1 35.3 (17.4)
M9 6b Power-law 55 816.5 7355.0 3365.8

aWAIC values are presented on a deviance scale (lower values indicate better model fit). Equivalent model fits for exponential and sigmoidal forgetting functions can be found in Extended Data Tables 2-1 and 2-2, respectively. Model parameter estimates for the best-fitting model M9 can be found in Extended Data Table 2-3. Results of a further comparison of variants of M9 with two forgetting rates can be found in Extended Data Table 2-4. Results of model comparison for data from control animals only can be found in Extended Data Table 2-5.

bThere is one fewer parameter per mouse in Dataset 3 because a smaller number of separations in this dataset rendered the α parameter nonidentifiable.