Claims 32
From the paper Distinct Neural Representation in the Dorsolateral, Dorsomedial, and Ventral Parts of the Striatum during Fixed- and Free-Choice Tasks · claims file · checked 2026-10-21: ready
- trial-counts
The analysis covers 78,107 trials in 86 recording sessions from seven rats, of which 12,185 are single fixed-choice, 22,234 double fixed-choice, 36,292 free-choice and 3505 extinction-test trials.
Here, we report the results of all 78,107 trials in 86 recording sessions, performed by seven rats, consisting of 12,185 single fixed-choice trials (16.4%), 22,234 double fixed-choice trials (30.0%), 36,292 free-choice trials (48.9%), and 3505 extinction test trials (4.7%)
Results, opening section, paragraph 2How it will be judged
The rat count must match exactly. The paper does not say which trials it counts (the four categories sum to 74,216, not 78,107). Readings are (a) all trials started, including wait-error and error trials; (b) trials with an L/R choice only. Reproduced when, under one reading, every count is within tolerance.
- trials-per-block
Completing a block took 22.56 ± 8.70 trials for single fixed-choice blocks, 47.11 ± 20.29 for double fixed-choice blocks and 41.10 ± 27.58 for free-choice blocks (mean ± SD).
Numbers of trials required to complete one block were 22.56 ± 8.70 for single fixed-choice blocks, 47.11 ± 20.29 for double fixed-choice blocks, and 41.10 ± 27.58 for free-choice blocks, (mean ± SD)
Results, opening section, paragraph 2How it will be judged
The tolerance of each mean is the stated SD (protocol); each SD is checked to 10%. The paper leaves open which trials are counted and which blocks enter. Readings are (a) choice trials only (the block rule counts "choice trials"), completed blocks only; (b) all trials including wait-error and error trials, completed blocks only; (c) as (a) with the last, possibly unfinished, block of a session included; (d) as (b) with it included. Extinction trials are not counted in any reading. Reproduced when, under one reading, all six values are within tolerance.
- fig2c-extinction-differs
During extinction tests, choice probabilities for the left and right tones differ significantly from those for the choice tone.
Choice probabilities during extinction tests were significantly different between the left and right tones and the choice tone (Fig. 2C; p < 0.0001, χ2 test)
Results, Behavioral performance, paragraph 1; Figure 2B and 2CHow it will be judged
Choices over the five trials of each extinction test (Fig. 2C legend). The contingency table of the χ2 test is not given. Readings are (a) left-tone trials against choice-tone trials after a left-optimal free-choice block, and right-tone trials against choice-tone trials after a right-optimal block, each a 2 x 2 table of left and right choices; (b) one 2 x 2 table of choices of the previously optimal action, fixed tones pooled against the choice tone pooled. Reproduced when, under every reading, every test gives p < 0.0001 with the fixed tones further from 50% than the choice tone.
- fig2b-extinction-time-course
In extinction tests the choice-tone choice probability shifts toward 50% after the first unrewarded trial, while the left- and right-tone choice probabilities stay biased after five unrewarded trials.
whereas the choice probabilities for the free-choice tone shifted toward 50% following the first nonrewarded trial (orange lines, Fig. 2B), choice probabilities for the left and right tones remained biased even after five no-reward trials
Results, Behavioral performance, paragraph 1; Figure 2BHow it will be judged
Left-choice probability at each of the five extinction trials, per tone, with the choice tone split by the optimal action of the previous free-choice block (Fig. 2B legend), with 95% confidence intervals of a binomial proportion (the interval method is not given; the readings are Wilson and Clopper-Pearson intervals). Reproduced when, under both readings, (1) for both choice-tone groups the probability at trial 2 is closer to 0.5 than at trial 1, and (2) at trial 5 the 95% interval for the left tone lies above 0.5 and that for the right tone below 0.5.
- fig2d-f-decision-trees
Choice-tone choices depend on the previous one or two trials' actions and rewards (a decision tree that expands symmetrically), while left- and right-tone choices stay biased toward left and right whatever the previous experience.
The decision tree for choice tones expands symmetrically (Fig. 2D), indicating that action selection was sensitive to past experiences ... action probabilities for the left or right tones (in both single and double fixed-choice blocks) are biased toward 1 (Fig. 2E) or 0 (Fig. 2F), respectively, indicating insensitivity to past experiences
Results, Behavioral performance, paragraph 2; Figure 2D to 2FHow it will be judged
Conditional left-choice probabilities for d = 1 and d = 2 from counts NL and NR after each experience sequence, within choice-tone trials and within left- and right-tone trials (single and double fixed-choice blocks), as the Methods define (Decision trees); 95% binomial intervals as in fig2b-extinction-time-course. Experience sequences are taken over consecutive trials of the same tone within a session (the Methods say "sequences of choice behavior in choice-tone trials"); whether they run across block borders is open, and both are readings. The paper gives no test or threshold. Reproduced when, under every reading, (1) for choice tones P(L) after L1 exceeds P(L) after R1 with non-overlapping 95% intervals, and (2) every d = 1 and d = 2 conditional probability is above 0.5 for the left tone and below 0.5 for the right tone. For (2), the readings are all sequences, and sequences with at least 10 trials (a sparse branch is not a reliable estimate; the paper gives no floor).
- fig2g-fq-variable-best
Of six DFQ-learning variants and the Markov models, the FQ-learning model with time-varying parameters predicts held-out rat choices in free-choice blocks best.
found that the FQ-learning model with time-varying parameters predicted rat behaviors best (Fig. 2G). Prediction accuracy was higher than that of Markov models
Results, Behavioral performance, paragraph 5; Figure 2GHow it will be judged
Choice-tone trials of free-choice blocks; models Q (alpha2 = kappa2 = 0), FQ (alpha1 = alpha2) and DFQ, each with constant or time-varying parameters, and Markov models of order d (Methods, Evaluation of decision-making models). Sessions split into training and test halves (the paper used 101 and 101 sessions; how the split was made is not given, so the readings are alternate sessions and a seeded random half, both shown); parameters fitted by maximum likelihood on training data, the time-varying models with a particle filter of 5000 particles. Model equations as O8 (opening comment). The normalized likelihood formula is missing; it is read as the geometric mean of per-trial choice likelihoods, which is 0.5 at chance, computed per test session for the paired Wilcoxon test across sessions. Reproduced when, under every reading, FQ (variable) has the highest normalized likelihood of test data and exceeds every Markov model.
- actor-critic-near-chance
A regular actor-critic model with constant parameters predicts test choices close to chance, with normalized likelihood 0.5080.
We also tested a regular actor-critic model with constant variables (Sutton and Barto, 1998), but the normalized likelihood of test data was close to the chance level (0.5080; data not shown)
Results, Behavioral performance, last paragraphHow it will be judged
The paper names the model only as a regular actor-critic model with constant parameters (Sutton and Barto, 1998) and gives neither its equations nor its parameters. "Constant" is read as in Fig. 2G, parameters constant across sessions, fitted by maximum likelihood on the training sessions and scored on the test sessions as in fig2g-fq-variable-best. The readings of the model are the tabular actor-critic for one state and two actions (critic V and actor preferences both updated by the reward prediction error, softmax choice with a free inverse temperature) with (a) one learning rate shared by critic and actor, (b) separate learning rates. Reproduced when every reading is within tolerance; blocked as methods_underspecified when the readings disagree.
- neuron-counts
Stable recordings were made from 260 DLS, 178 DMS and 179 VS neurons, of which 190, 105 and 119 were phasically active neurons.
Stable recordings were made from 260 neurons in DLS, 178 neurons in DMS, and 179 neurons in VS from seven rats ... Among these, 190, 105, and 119 neurons from DLS, DMS, and VS, respectively, were classified as PANs
Results, Activity patterns of phasically active neurons, paragraph 1How it will be judged
Counts of units per region; PANs by the rule of the Methods (proportion of time in ISIs longer than 1 s greater than 0.4), over the whole session. The Results also name waveforms as a basis of the PAN class, but the Methods give no waveform rule, so none is applied. If the dataset holds only curated units, the stable counts are compared as deposited and the PAN rule is applied to them. Reproduced when all six counts are within tolerance.
- fig4b-vs-phase1
More than 60% of VS PANs increase their activity as the rat approaches the center hole (Phase 1), significantly more than in DLS and DMS.
The proportion of neurons that increased their activity as a rat approached the center hole (trial Phase 1) was >60% in VS, significantly larger than in DLS and DMS (p = 0.0015 and p < 0.0001, respectively, χ2 test)
Results, Activity patterns of phasically active neurons, paragraph 3; Figure 4BHow it will be judged
Per PAN, the EASH over all trials (O3), normalized to a maximum of 1 (Fig. 4A legend), smoothing as O2. A neuron is activated in a phase when its normalized EASH reaches 0.5 during that phase (Fig. 4B legend); readings are (a) any bin at or above 0.5 in the phase, (b) the first crossing of 0.5 falls in the phase (the sort of Fig. 4A). 2 x 2 χ2 tests between regions, O4. Reproduced when, under every reading, the VS proportion exceeds 0.54 and VS exceeds DLS and DMS with p < 0.05 in both tests.
- fig4b-dms-phase5
More than 60% of DMS PANs are activated between center-hole exit and L/R-hole entry (Phase 5), significantly more than in DLS and VS.
After a rat's exit from the center hole until its entrance into the L/R hole (trial Phase 5), >60% of DMS neurons were activated, which was significantly larger than the proportions of DLS and VS (p = 0.00021 for DLS, p = 0.040 for VS, χ2 test)
Results, Activity patterns of phasically active neurons, paragraph 3; Figure 4BHow it will be judged
As fig4b-vs-phase1, for Phase 5. Reproduced when, under every reading, the DMS proportion exceeds 0.54 and DMS exceeds DLS and VS with p < 0.05 in both tests.
- fig4b-vs-phase7
During pellet retrieval (Phase 7) the proportion of activated neurons is significantly larger in VS than in DLS.
During the trial phase of receiving a sugar pellet (trial Phase 7), the proportion of activated neurons was significantly larger in VS than in DLS (p = 0.0040, χ2 test)
Results, Activity patterns of phasically active neurons, paragraph 3; Figure 4BHow it will be judged
As fig4b-vs-phase1, for Phase 7. Reproduced when, under every reading, the VS proportion exceeds the DLS proportion with p < 0.05.
- fig4c-activation-duration
VS PANs stay active longer than DLS and DMS PANs, and DLS PANs have the shortest activation, shorter than VS and DMS.
VS neurons maintained activation longer (more than half of peak activity) than DLS (p < 0.0001, Mann–Whitney U test) and DMS neurons (p < 0.0001, Mann–Whitney U test), and DLS neurons had shorter durations of activation than VS neurons and DMS neurons (p < 0.0001 for VS, p = 0.040 for DMS, Mann–Whitney U test)
Results, Activity patterns of phasically active neurons, paragraph 3; Figure 4CHow it will be judged
Per PAN, the activity ratio of each phase (time the normalized EASH is above 0.5, divided by the phase duration), averaged over the seven phases (Fig. 4C legend); EASH as in fig4b-vs-phase1 (O2, O3). Two-sided Mann-Whitney U tests between regions. Reproduced when, under every reading of O2 and O3, the region means (Fig. 4C plots the average) order VS > DMS > DLS and all three tests give p < 0.05.
- fig5b-state-info-vs
Information about the block type (fixed- or free-choice) is strongest in VS, especially during the approach to the center hole (Phase 1).
State information (fixed- or free-choice block) was strongest in VS, especially, during the approach to the center hole (trial Phase 1) (p < 0.0001 for DLS, p = 0.022 for DMS, Mann–Whitney U tests)
Results, Information coding of state, action, and reward, paragraph 2; Figure 5A and 5BHow it will be judged
Per PAN, MI between firing and block type (fixed-choice blocks, single and double, against free-choice blocks) in each 100 ms EASH bin over all trials (Methods, Mutual information; O1, O3). The test's unit of comparison is not stated; readings are per-neuron MI (a) averaged over the bins of Phase 1, (b) in each Phase 1 bin, the result read from the bins that pass. Two-sided Mann-Whitney U tests. Reproduced when, under every reading, VS exceeds DLS and DMS with p < 0.05 and the VS region average exceeds the shuffle threshold in Phase 1.
- fig5e-action-info-rises-dls
Action information (left or right choice) starts rising during cue presentation (Phase 3) only in DLS.
Action information (left or right hole choice) started increasing during the tone presentation (trial Phase 3) specifically in DLS (Fig. 5C–E)
Results, Information coding of state, action, and reward, paragraph 2; Figure 5C to 5E; Discussion, paragraph 1How it will be judged
Region averages of per-PAN action MI in 100 ms EASH bins (O1, O3), against the shuffle threshold of the Methods (100 random-event repeats, second largest, p < 0.01). Reproduced when, under every reading, the DLS average exceeds its threshold in at least one Phase 3 bin and the DMS and VS averages exceed theirs in no Phase 3 bin.
- fig5e-action-info-cue-offset-dls
In the 100 ms before cue offset, action information is significantly higher in DLS than in DMS and VS.
Action information during the 100 ms before the offset of the cue tone was significantly higher in DLS than in DMS and VS (p = 0.0175 for DMS, p = 0.00091 for VS, Mann–Whitney U tests)
Results, Information coding of state, action, and reward, paragraph 3; Figure 5EHow it will be judged
Per-PAN action MI in the 100 ms EASH bin ending at E3 (O1, O3); two-sided Mann-Whitney U tests. Reproduced when, under every reading, DLS exceeds DMS and VS with p < 0.05 in both tests.
- fig5e-action-info-center-exit
In the 100 ms before center-hole exit, action information in DMS is higher than in DLS (not significant) and VS, and is weakest in VS.
Immediately after the offset of the cue tone (trial Phase 4) action information in DMS rapidly increased and became higher than that in DLS and VS during the 100 ms before the offset from the center hole, while information in VS was weakest (p = 0.16 for DLS, p = 0.0027 for VS, p < 0.0001 for DLS vs VS, Mann–Whitney U tests)
Results, Information coding of state, action, and reward, paragraph 3; Figure 5D and 5EHow it will be judged
Per-PAN action MI in the 100 ms EASH bin ending at E4 (O1, O3); two-sided Mann-Whitney U tests. Reproduced when, under every reading, the region means order DMS > DLS > VS, and DMS against VS and DLS against VS give p < 0.05. DMS against DLS is reported; its significance is not part of the claim.
- fig5d-action-info-phase5-dms
During action execution (Phase 5), action information is highest in DMS.
During action execution (trial Phase 5), the action information was the highest in DMS (p < 0.0001 for DLS, p < 0.0001 for VS, p < 0.0001 for DLS vs VS, Mann–Whitney U tests)
Results, Information coding of state, action, and reward, paragraph 3; Figure 5DHow it will be judged
Per-PAN action MI in the 100 ms EASH bins of Phase 5 (O1, O3); the bins compared are not stated, so the readings are (a) the mean over Phase 5 bins, (b) the bin of the DMS region peak. Two-sided Mann-Whitney U tests. Reproduced when, under every reading, DMS exceeds DLS and VS with p < 0.05 in both tests. The DLS against VS test is reported.
- fig5g-reward-info
Reward information rises in all regions after L/R-hole entry and is strongest in VS, then DMS, with DLS significantly below both.
Reward information was strongest in VS, followed by DMS, and reward information in DLS was significantly less than that in VS and DMS (p = 0.012 for DLS vs DMS, p < 0.0001 for DLS vs VS, p = 0.13 for DMS vs VS, Mann–Whitney U tests)
Results, Information coding of state, action, and reward, paragraph 4; Figure 5F and 5GHow it will be judged
Per-PAN reward MI (rewarded against unrewarded choices) in 100 ms EASH bins (O1, O3). The window of the test is not stated; readings are the mean over the bins of (a) Phase 6, (b) Phases 6 and 7, (c) the bin of the VS region peak. Two-sided Mann-Whitney U tests. Reproduced when, under every reading, (1) the region average of each region first exceeds its shuffle threshold (Methods, Mutual information) after E5, in Phase 6, and (2) the means order VS > DMS > DLS and DLS is below DMS and VS with p < 0.05.
- fig6-event-coding-proportions
Proportions of event-coding neurons follow the MI results. State-coding neurons during cue presentation are most common in VS, action-coding neurons in DLS before action execution and in DMS during it, and reward-coding neurons in VS during L/R poking.
the proportion of state-coding neurons during cue presentation (trial Phase 3) was the largest in VS, and the proportion of action-coding neurons was the largest in DLS before action execution and in DMS during action execution. The proportion of reward-coding neurons was the largest in VS during the L/R poking (Fig. 6D)
Results, Information coding of state, action, and reward, paragraph 4; Figure 6DHow it will be judged
Over all trials (Fig. 6D; O3), a PAN codes an event when firing differs between the two conditions by a Mann-Whitney U test with p < 0.01 (Fig. 6 legend). Windows from the legend; action command (AC) the 500 ms before E4, action the 500 ms after E4, Reward1 the 500 ms before E6, Reward2 the 500 ms after E6. The state window is drawn only in Fig. 6A, which could not be read; readings are (a) all of Phase 3 on the EASH axis, (b) the 500 ms before E3. Reward "during the L/R poking" is read as Reward1. Reproduced when, under every reading, VS has the largest state and Reward1 proportions, DLS the largest AC proportion and DMS the largest action proportion. The paper's χ2 tests between regions (O4) are reported; which pairs it marked significant could not be read from the figure.
- fig6-above-chance
Every proportion of event-coding neurons, in every region, in fixed-choice blocks, free-choice blocks and all trials, is significantly larger than chance.
All populations are significantly larger than the chance level (binomial test, p < 0.01)
Figure 6B to 6D legendHow it will be judged
As fig6-event-coding-proportions, for fixed-choice trials (Fig. 6B), free-choice trials (Fig. 6C) and all trials (Fig. 6D); state coding exists only for all trials. The chance level is not given; it is read as 0.01, the selection threshold. One-sided binomial test per region and event. Reproduced when every test gives p < 0.01.
- fig6bc-similar-fixed-free
The proportions of action- and reward-coding neurons are similar in fixed- and free-choice blocks.
These proportions in the population were similar in fixed-choice and free-choice blocks (Fig. 6B,C)
Results, Information coding of state, action, and reward, paragraph 4; Figure 6B and 6CHow it will be judged
The paper names no test. Proportions of AC, action, Reward1 and Reward2 neurons per region, computed separately in fixed-choice and free-choice trials, as in fig6-event-coding-proportions. Reproduced when the region with the largest proportion is the same in both block types for each event. The difference per event and region, with a 2 x 2 χ2 test (O4), is reported.
- fig8a-d-action-reward-coding
Substantial proportions of neurons in all three regions code the current action before and after action execution and the reward after L/R poking, and rewards of the previous trial are coded in all phases in all regions; a significant proportion of DMS neurons codes the previous action during action execution.
Substantial proportions of neurons in DLS, DMS, and VS coded actions before and after action execution (Fig. 8A) ... a significant proportion of DMS neurons coded previous actions during action execution (trial Phase 5; Fig. 8B). Substantial proportions of neurons in all subareas coded rewards after the start of L/R hole poking (Fig. 8C), and rewards in the previous trial were also strongly coded in all trial phases in all striatal regions (Fig. 8D)
Results, Model-based analysis of action value and state value coding, paragraph 3; Figure 8A to 8DHow it will be judged
Regression with BIC selection as the Methods give it (opening comment; O6, O7, O8), coding at p < 0.01 per coefficient. Proportions per region and window against chance by a one-sided binomial test at p < 0.01 (Fig. 8 legend); the chance level is not given and is read as 0.01. Reproduced when, under every reading, (1) a(t) is coded above chance in every region in at least one window before and one after E4, (2) r(t) above chance in every region in at least one window after E5, (3) r(t-1) above chance in every region in at least one window of every phase, and (4) a(t-1) above chance in DMS in a Phase 5 window.
- fig8e-action-value-dms
The proportion of action-value-coding neurons is highest in DMS during action execution, significantly above DLS.
the proportion of action-value coding neurons was highest in DMS during action execution (Fig. 8E; p = 0.00058 for DLS, p = 0.081 for VS, Mann–Whitney U test)
Results, Model-based analysis of action value and state value coding, paragraph 4; Figure 8EHow it will be judged
Action-value coding is QL or QR coded (Fig. 8E legend), regression as in fig8a-d-action-reward-coding. Per-neuron 0/1 indicators in the Phase 5 windows compared by Mann-Whitney U tests (O5). Reproduced when, under every reading, DMS has the highest proportion of the three regions in a Phase 5 window and exceeds DLS there with p < 0.05. DMS against VS is reported; its significance is not part of the claim.
- fig8e-qr-dms
Most action-value-coding DMS neurons represent QR, and the proportion coding QR during action execution is significantly larger in DMS than in DLS.
The majority of these action value-coding neurons in DMS represented QR, the action value for right hole choice, and the proportion of neurons coding QR during action execution was significant larger in DMS than in DLS (p = 0.00017, Mann–Whitney U test)
Results, Model-based analysis of action value and state value coding, paragraph 4How it will be judged
As fig8e-action-value-dms, with QR coding alone. Reproduced when, under every reading, more than half of the DMS action-value-coding neurons in the Phase 5 window code QR and the DMS proportion coding QR exceeds the DLS proportion with p < 0.05.
- fig8f-state-value-vs
State-value-coding neurons (QL and QR coded with the same sign) are most common in VS in all phases, significantly more than in DLS and DMS in Phase 1 and in the first half of Phase 4.
The proportion of state-value coding neurons was the highest in VS in all trial phases, especially in starting phase (trial Phase 1) and action initiation phase (trial Phase 4), the proportion was significantly larger in VS than in DLS and DMS (p = 0.039 for DMS and p = 0.0027 for DLS, in trial Phase 1; p = 0.024 for DMS and p = 0.0045 for DLS, first half of Phase 4; Mann–Whitney U test)
Results, Model-based analysis of action value and state value coding, paragraph 5; Figure 8F (cited as Fig. 7F in the text)How it will be judged
State-value coding as the Fig. 8F legend defines it, regression as in fig8a-d-action-reward-coding, Mann-Whitney U tests on 0/1 indicators (O5). Reproduced when, under every reading, VS has the largest proportion in every phase (at least one window per phase) and exceeds DLS and DMS with p < 0.05 in a Phase 1 window and in the window covering the first half of Phase 4.
- no-policy-coding
No region has a significant number of policy-coding neurons (QL and QR coded with opposite signs) in any time bin.
We did not find significant numbers of policy-coding neurons (those with significant correlations with both action values with opposite signs), in any subarea or in any time bin (data not shown)
Results, Model-based analysis of action value and state value coding, paragraph 5How it will be judged
A claim of absence, on the paper's test for Fig. 8, a one-sided binomial test of the proportion against chance at p < 0.01, with chance read as 0.01 (as in fig8a-d-action-reward-coding; the paper names one chance level for all panels of Fig. 8 and does not give it). Reproduced when, under every reading of O6 to O8, no region and window gives p < 0.01; not reproduced when, under every reading, one does; blocked when they disagree. The largest proportion is reported either way.
- fig9-temporal-order-all-blocks
In every block type, action information rises first in DLS during cue presentation, then in DMS after cue offset (Phase 4), and last in VS after the onset of action execution (Phase 5); the pattern holds in free-choice blocks too.
Action information in the single and double fixed-choice blocks started to rise first in DLS during cue presentation and then in DMS after cue offset (trial Phase 4), and finally in VS after onset of action execution (trial Phase 5) (Fig. 9A,B). Contrary to expectation, this temporal pattern of action information was seen also in free-choice blocks (Fig. 9C,D)
Results, Action-command information in fixed- and free-choice blocks, paragraph 1; Figure 9A to 9DHow it will be judged
Per PAN and block type (single fixed, double fixed, free with higher probabilities (90, 50) and (50, 90), free with lower probabilities (50, 10) and (10, 50)), action MI from the last 20 trials of each block, in a sliding window of the preceding 500 ms with 50 ms steps (Fig. 9 legend), averaged per region; first time the average exceeds the p < 0.01 shuffle threshold (the triangle of Fig. 9). How the last 20 trials of several blocks of one type in a session are combined is not stated; readings are (a) pooled into one MI estimate per neuron, (b) MI per block, averaged. O1 applies. The rise of a region is the start of its sustained rise into action execution, the first crossing after which the average stays above threshold through Phase 5, since Fig. 10C reports a transient VS peak during cue presentation in single fixed-choice blocks, which the paper does not count as the VS rise. The first crossing is reported too. Reproduced when, under every reading, (1) in single and double fixed-choice blocks the rise falls in Phase 3 for DLS, in Phase 4 for DMS and in Phase 5 for VS, and (2) in both free-choice block types the rises come in the same order, DLS, then DMS, then VS (the text says only that "this temporal pattern" was seen; Fig. 10A says the DLS rise is later in free-choice blocks); their phases are reported.
- fig9-dms-never-above-dls
In no block type is the action command in DMS stronger than in DLS.
There were no conditions in which the action command in DMS was stronger than in DLS
Results, Action-command information in fixed- and free-choice blocks, paragraph 1; Figure 9How it will be judged
A claim of absence for which the paper gives no test. Action command is action MI before action execution (Fig. 9 legend, upcoming action for Phases 2 to 4), as in fig9-temporal-order-all-blocks, in the two windows the paper tests in Fig. 10 (the 500 ms before E3 and the 500 ms before E4). Judged with the test the paper uses to compare MI between regions (Results, Information coding of state, action, and reward), a Mann-Whitney U test across neurons, two-sided at p < 0.05. Reproduced when, under every reading, DMS is not significantly above DLS in any block type and window; not reproduced when, under every reading, it is somewhere; blocked when the readings disagree. The differences are reported either way. Note that fig5e-action-info-center-exit has DMS above DLS, not significantly, in the 100 ms before E4 over all trials.
- fig10a-dls-fixed-above-free
In DLS, action information is significantly stronger in fixed-choice than in free-choice blocks in Phase 3 and Phase 4; it rises right after cue onset in fixed-choice blocks, more slowly in free-choice blocks with higher rewards, and later and weaker with lower rewards.
Action information in free-choice blocks with higher rewards increased more slowly than that in fixed-choice blocks, and action information in free-choice blocks with lower rewards appeared later and was weaker. There were significant differences in action information between fixed-choice blocks and free-choice blocks in trial Phase 3 (during the 500 ms before cue presentation) and trial Phase 4 (during the 500 ms before center hole exit) (p = 0.00047 and p = 0.00037, respectively, paired-Mann–Whitney U test)
Results, Action-command information in fixed- and free-choice blocks, paragraph 2; Figure 10AHow it will be judged
Per-PAN action MI as in fig9-temporal-order-all-blocks. Windows from the text, the 500 ms windows ending at the end of Phase 3 and of Phase 4 ("500 ms before cue presentation" is read as the 500 ms before cue offset, the end of Phase 3, since the window must lie in Phase 3; the 500 ms before cue onset is shown as a variant). Fixed is SF and DF, free is FH and FL; readings of how they are combined are (a) the per-neuron mean of the two block types, (b) trials pooled. "Paired-Mann-Whitney U test" is read as (a) the Wilcoxon signed-rank test over neurons, (b) the Mann-Whitney U test unpaired. Reproduced when, under every reading, fixed exceeds free with p < 0.05 in both windows and the DLS region average crosses threshold first in SF or DF, then FH, then FL, with FL's peak in Phases 3 and 4 the lowest of the four.
- fig10b-dms-fixed-above-free
In DMS too, action information is significantly stronger in fixed-choice than in free-choice blocks in Phases 3 and 4, against the paper's prediction.
In DMS, action information was stronger in single and double fixed-choice blocks than in free-choice blocks (Fig. 10B), similar to that in DLS, with significant differences in trial Phases 3 and 4 (p = 0.040 for Phase 3, p = 0.0029 for Phase 4, paired-Mann–Whitney U test)
Results, Action-command information in fixed- and free-choice blocks, paragraph 2; Figure 10BHow it will be judged
As fig10a-dls-fixed-above-free, for DMS, without the order of crossings. Reproduced when, under every reading, fixed exceeds free with p < 0.05 in both windows.
- fig10c-vs-no-fixed-free-difference
In VS, action information shows a transient peak during cue presentation only in single fixed-choice blocks and stays low until action execution, with no significant difference between fixed- and free-choice blocks in Phases 3 and 4.
Action information in VS had a transient peak during cue presentation only in single, fixed-choice blocks and remained low until the onset of action execution (Fig. 10C) with no significant differences between fixed- and free-choice blocks in trial Phases 3 and 4 (p = 0.23 and p = 0.28, respectively, paired-Mann–Whitney U test)
Results, Action-command information in fixed- and free-choice blocks, paragraph 2; Figure 10CHow it will be judged
A claim of absence, on the paper's test as in fig10a-dls-fixed-above-free (each of its readings of the test, as the test itself is ambiguous), at p < 0.05. Reproduced when, under every reading, neither window gives p < 0.05; not reproduced when, under every reading, either does; blocked when the readings disagree. The effect sizes are reported. The SF peak is reported, as whether the SF region average crosses threshold in Phase 3 while DF, FH and FL do not.
- fig10d-dls-stay-above-switch
In free-choice blocks, DLS action information is stronger in stay trials than in switch trials in Phase 3, while DMS shows no difference between stay and switch trials.
In DLS, action information was stronger in stay trials than in switch trials in trial Phase 3 (during the 500 ms before cue presentation) (p = 0.019, Mann–Whitney U test). In DMS, however, we could not find any difference in the action-command signal between switch and stay trials
Results, Action-command information in fixed- and free-choice blocks, paragraph 3; Figure 10D and 10EHow it will be judged
Free-choice trials split by whether the action repeats the previous trial's (stay) or not (switch). The trials used are not stated; the Fig. 9 legend and the Results for Fig. 9 say the last 20 trials of each block (the Fig. 10 legend does not), so readings are (a) the last 20 trials of each free-choice block, (b) all free-choice trials. Per-PAN action MI in the Phase 3 window of fig10a-dls-fixed-above-free (O1); two-sided Mann-Whitney U test. Reproduced when, under every reading, DLS stay exceeds switch with p < 0.05 and, for DMS (a claim of absence on the same test), neither the Phase 3 nor the Phase 4 window gives p < 0.05.
Pipeline
- Paper
- done (1 of 1)
- Claims
- done
- Claims check
- done (ready)
- Attempt
- todo
- Questions
- none
- Verified
- none