samsam2610/experiments-database
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

feat(matlab): add Satterthwaite DF + honest random-slope test to LME reports

Browse Source 
Every fitlme-based report (lme_*, paper_*, phase_*, and the variations'
analyze.m) now shows, per effect: residual-DF p, Satterthwaite-DF p, and --
for the interaction -- an HONEST test from a per-animal random-SLOPE model
(day|rat), whose Satterthwaite DF collapses toward the animal count.

New: tdcs_random_slope_interaction.m (shared helper). Wired into tdcs_lme,
tdcs_paper_lme, tdcs_phase_lme, variation_analyze; SUMMARY.csv gains
interaction_p_satt / interaction_p_rs. Regenerated all results/, variations/,
matched-effort outputs.

Key point this surfaces: Satterthwaite ~= residual on the random-INTERCEPT
model (the slope's error is at session level), so it does NOT fix
pseudoreplication; the random-slope model does. Effect: full-range mergeA2
interaction 0.009 -> 0.75 (collapses); unmerge_d0_5 0.015 -> 0.13 (n.s.);
the pooled-control early windows survive honestly (naive_a2_d0_5 0.001 ->
0.028; naive_boxa_d0_5 0.003 -> 0.036). Suite 42/42.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
This commit is contained in:
Experiments DB Dev
2026-07-22 15:45:19 -04:00
parent 5fcd97f81a
commit 8a18c894dd
77 changed files with 2066 additions and 345 deletions
Download Patch File Download Diff File Expand all files Collapse all files
analysis/matlab/results/lme_mergeA2_d0_10.txt
+11 -5
View File
@@ -54,11 +54,17 @@ Group: Error
{'Res Std'} 13.04 11.2 15.183
effect t (df) F (df1) p
----------------------------------------------------------------------
days x tDCS (interaction) t(79)= 1.51 F(1)= 2.292 p=0.134
days (learning) t(79)= 12.43 F(1)= 154.593 p=2.759e-20
tDCS (main, at Day 1) t(79)= 0.80 F(1)= 0.635 p=0.428
effect t(df) / F(df1) p (resid) Satterthwaite: p (df)
------------------------------------------------------------------------------
days x tDCS (interaction) t(79)= 1.51 F(1)= 2.292 p=0.134 p=0.1338 (df=83)
days (learning) t(79)= 12.43 F(1)=154.593 p=2.759e-20 p=1.192e-20 (df=83)
tDCS (main, at Day 1) t(79)= 0.80 F(1)= 0.635 p=0.428 p=0.4278 (df=83)
HONEST LME -- per-animal random slope (day|subject): interaction F(1,20.3)=1.35, p=0.258
Note: Satterthwaite DF on the random-INTERCEPT model above stays ~= residual
(the slope's error is at the session level), so it does NOT fix pseudoreplication.
Letting each animal have its OWN slope collapses the interaction DF toward the animal
count -- this, and the per-animal slope test, are the honest learning-rate inference.
INTERPRETATION
- days x tDCS interaction: n.s. (p=0.1340, slope diff=1.38) -> slopes are parallel (no differential change over training).
analysis/matlab/results/lme_mergeA2_d0_13.txt
+11 -5
View File
@@ -54,11 +54,17 @@ Group: Error
{'Res Std'} 14.123 12.278 16.245
effect t (df) F (df1) p
----------------------------------------------------------------------
days x tDCS (interaction) t(94)= 1.36 F(1)= 1.844 p=0.1777
days (learning) t(94)= 13.14 F(1)= 172.537 p=5.403e-23
tDCS (main, at Day 1) t(94)= 1.12 F(1)= 1.264 p=0.2637
effect t(df) / F(df1) p (resid) Satterthwaite: p (df)
------------------------------------------------------------------------------
days x tDCS (interaction) t(94)= 1.36 F(1)= 1.844 p=0.1777 p=0.1776 (df=98)
days (learning) t(94)= 13.14 F(1)=172.537 p=5.403e-23 p=2.463e-23 (df=98)
tDCS (main, at Day 1) t(94)= 1.12 F(1)= 1.264 p=0.2637 p=0.2636 (df=98)
HONEST LME -- per-animal random slope (day|subject): interaction F(1,16.2)=1.37, p=0.2593
Note: Satterthwaite DF on the random-INTERCEPT model above stays ~= residual
(the slope's error is at the session level), so it does NOT fix pseudoreplication.
Letting each animal have its OWN slope collapses the interaction DF toward the animal
count -- this, and the per-animal slope test, are the honest learning-rate inference.
INTERPRETATION
- days x tDCS interaction: n.s. (p=0.1777, slope diff=1.00) -> slopes are parallel (no differential change over training).
analysis/matlab/results/lme_mergeA2_full.txt
+11 -5
View File
@@ -57,11 +57,17 @@ Group: Error
{'Res Std'} 18.954 16.597 21.644
effect t (df) F (df1) p
----------------------------------------------------------------------
days x tDCS (interaction) t(105)= -2.66 F(1)= 7.094 p=0.008953
days (learning) t(105)= 9.82 F(1)= 96.388 p=1.573e-16
tDCS (main, at Day 1) t(105)= 3.43 F(1)= 11.778 p=0.0008585
effect t(df) / F(df1) p (resid) Satterthwaite: p (df)
------------------------------------------------------------------------------
days x tDCS (interaction) t(105)= -2.66 F(1)= 7.094 p=0.008953 p=0.008906 (df=109)
days (learning) t(105)= 9.82 F(1)= 96.388 p=1.573e-16 p=1.113e-16 (df=109)
tDCS (main, at Day 1) t(105)= 3.43 F(1)= 11.778 p=0.0008585 p=0.000848 (df=109)
HONEST LME -- per-animal random slope (day|subject): interaction F(1,9.7)=0.11, p=0.746
Note: Satterthwaite DF on the random-INTERCEPT model above stays ~= residual
(the slope's error is at the session level), so it does NOT fix pseudoreplication.
Letting each animal have its OWN slope collapses the interaction DF toward the animal
count -- this, and the per-animal slope test, are the honest learning-rate inference.
INTERPRETATION
- days x tDCS interaction: SIGNIFICANT (p=0.0090, slope diff=-2.18) -> the tDCS (Box-B2) group improves SLOWER -- groups CONVERGE (Box-B2 is ahead early, the gap narrows).
analysis/matlab/results/lme_mergeB2_d0_10.txt
+11 -5
View File
@@ -54,11 +54,17 @@ Group: Error
{'Res Std'} 11.955 10.184 14.034
effect t (df) F (df1) p
----------------------------------------------------------------------
days x tDCS (interaction) t(79)= 1.91 F(1)= 3.632 p=0.06032
days (learning) t(79)= 10.55 F(1)= 111.334 p=9.547e-17
tDCS (main, at Day 1) t(79)= 0.07 F(1)= 0.004 p=0.9478
effect t(df) / F(df1) p (resid) Satterthwaite: p (df)
------------------------------------------------------------------------------
days x tDCS (interaction) t(79)= 1.91 F(1)= 3.632 p=0.06032 p=0.06034 (df=79)
days (learning) t(79)= 10.55 F(1)=111.334 p=9.547e-17 p=8.241e-17 (df=80)
tDCS (main, at Day 1) t(79)= 0.07 F(1)= 0.004 p=0.9478 p=0.9482 (df=21)
HONEST LME -- per-animal random slope (day|subject): interaction F(1,30.6)=3.00, p=0.09319
Note: Satterthwaite DF on the random-INTERCEPT model above stays ~= residual
(the slope's error is at the session level), so it does NOT fix pseudoreplication.
Letting each animal have its OWN slope collapses the interaction DF toward the animal
count -- this, and the per-animal slope test, are the honest learning-rate inference.
INTERPRETATION
- days x tDCS interaction: n.s. (p=0.0603, slope diff=1.74) -> slopes are parallel (no differential change over training).
analysis/matlab/results/lme_mergeB2_d0_13.txt
+11 -5
View File
@@ -54,11 +54,17 @@ Group: Error
{'Res Std'} 13.148 11.362 15.215
effect t (df) F (df1) p
----------------------------------------------------------------------
days x tDCS (interaction) t(94)= 1.95 F(1)= 3.812 p=0.05385
days (learning) t(94)= 9.91 F(1)= 98.208 p=2.855e-16
tDCS (main, at Day 1) t(94)= 0.29 F(1)= 0.082 p=0.7757
effect t(df) / F(df1) p (resid) Satterthwaite: p (df)
------------------------------------------------------------------------------
days x tDCS (interaction) t(94)= 1.95 F(1)= 3.812 p=0.05385 p=0.05377 (df=97)
days (learning) t(94)= 9.91 F(1)= 98.208 p=2.855e-16 p=1.981e-16 (df=98)
tDCS (main, at Day 1) t(94)= 0.29 F(1)= 0.082 p=0.7757 p=0.7776 (df=24)
HONEST LME -- per-animal random slope (day|subject): interaction F(1,29.7)=3.25, p=0.08138
Note: Satterthwaite DF on the random-INTERCEPT model above stays ~= residual
(the slope's error is at the session level), so it does NOT fix pseudoreplication.
Letting each animal have its OWN slope collapses the interaction DF toward the animal
count -- this, and the per-animal slope test, are the honest learning-rate inference.
INTERPRETATION
- days x tDCS interaction: n.s. (p=0.0538, slope diff=1.54) -> slopes are parallel (no differential change over training).
analysis/matlab/results/lme_mergeB2_full.txt
+11 -5
View File
@@ -57,11 +57,17 @@ Group: Error
{'Res Std'} 19.354 16.948 22.101
effect t (df) F (df1) p
----------------------------------------------------------------------
days x tDCS (interaction) t(105)= -1.43 F(1)= 2.043 p=0.1559
days (learning) t(105)= 6.89 F(1)= 47.494 p=4.234e-10
tDCS (main, at Day 1) t(105)= 2.45 F(1)= 6.021 p=0.01578
effect t(df) / F(df1) p (resid) Satterthwaite: p (df)
------------------------------------------------------------------------------
days x tDCS (interaction) t(105)= -1.43 F(1)= 2.043 p=0.1559 p=0.1558 (df=109)
days (learning) t(105)= 6.89 F(1)= 47.494 p=4.234e-10 p=3.734e-10 (df=109)
tDCS (main, at Day 1) t(105)= 2.45 F(1)= 6.021 p=0.01578 p=0.01572 (df=109)
HONEST LME -- per-animal random slope (day|subject): interaction F(1,11.8)=0.02, p=0.9027
Note: Satterthwaite DF on the random-INTERCEPT model above stays ~= residual
(the slope's error is at the session level), so it does NOT fix pseudoreplication.
Letting each animal have its OWN slope collapses the interaction DF toward the animal
count -- this, and the per-animal slope test, are the honest learning-rate inference.
INTERPRETATION
- days x tDCS interaction: n.s. (p=0.1559, slope diff=-1.47) -> slopes are parallel (no differential change over training).
analysis/matlab/results/lme_unmerged_d0_10.txt
+11 -5
View File
@@ -54,11 +54,17 @@ Group: Error
{'Res Std'} 12.508 10.37 15.086
effect t (df) F (df1) p
----------------------------------------------------------------------
days x tDCS (interaction) t(57)= 1.49 F(1)= 2.208 p=0.1428
days (learning) t(57)= 10.08 F(1)= 101.556 p=2.832e-14
tDCS (main, at Day 1) t(57)= 0.68 F(1)= 0.466 p=0.4974
effect t(df) / F(df1) p (resid) Satterthwaite: p (df)
------------------------------------------------------------------------------
days x tDCS (interaction) t(57)= 1.49 F(1)= 2.208 p=0.1428 p=0.1428 (df=57)
days (learning) t(57)= 10.08 F(1)=101.556 p=2.832e-14 p=1.931e-14 (df=59)
tDCS (main, at Day 1) t(57)= 0.68 F(1)= 0.466 p=0.4974 p=0.5037 (df=17)
HONEST LME -- per-animal random slope (day|subject): interaction F(1,13.0)=1.46, p=0.2479
Note: Satterthwaite DF on the random-INTERCEPT model above stays ~= residual
(the slope's error is at the session level), so it does NOT fix pseudoreplication.
Letting each animal have its OWN slope collapses the interaction DF toward the animal
count -- this, and the per-animal slope test, are the honest learning-rate inference.
INTERPRETATION
- days x tDCS interaction: n.s. (p=0.1428, slope diff=1.56) -> slopes are parallel (no differential change over training).
analysis/matlab/results/lme_unmerged_d0_13.txt
+11 -5
View File
@@ -54,11 +54,17 @@ Group: Error
{'Res Std'} 14.676 12.436 17.32
effect t (df) F (df1) p
----------------------------------------------------------------------
days x tDCS (interaction) t(66)= 1.65 F(1)= 2.721 p=0.1038
days (learning) t(66)= 9.09 F(1)= 82.593 p=3.035e-13
tDCS (main, at Day 1) t(66)= 0.95 F(1)= 0.910 p=0.3436
effect t(df) / F(df1) p (resid) Satterthwaite: p (df)
------------------------------------------------------------------------------
days x tDCS (interaction) t(66)= 1.65 F(1)= 2.721 p=0.1038 p=0.1035 (df=70)
days (learning) t(66)= 9.09 F(1)= 82.593 p=3.035e-13 p=1.823e-13 (df=70)
tDCS (main, at Day 1) t(66)= 0.95 F(1)= 0.910 p=0.3436 p=0.3434 (df=70)
HONEST LME -- per-animal random slope (day|subject): interaction F(1,12.8)=1.64, p=0.2227
Note: Satterthwaite DF on the random-INTERCEPT model above stays ~= residual
(the slope's error is at the session level), so it does NOT fix pseudoreplication.
Letting each animal have its OWN slope collapses the interaction DF toward the animal
count -- this, and the per-animal slope test, are the honest learning-rate inference.
INTERPRETATION
- days x tDCS interaction: n.s. (p=0.1038, slope diff=1.55) -> slopes are parallel (no differential change over training).
analysis/matlab/results/lme_unmerged_full.txt
+11 -5
View File
@@ -54,11 +54,17 @@ Group: Error
{'Res Std'} 14.163 11.93 16.813
effect t (df) F (df1) p
----------------------------------------------------------------------
days x tDCS (interaction) t(67)= 1.02 F(1)= 1.041 p=0.3112
days (learning) t(67)= 9.20 F(1)= 84.665 p=1.671e-13
tDCS (main, at Day 1) t(67)= 1.04 F(1)= 1.086 p=0.3012
effect t(df) / F(df1) p (resid) Satterthwaite: p (df)
------------------------------------------------------------------------------
days x tDCS (interaction) t(67)= 1.02 F(1)= 1.041 p=0.3112 p=0.3111 (df=70)
days (learning) t(67)= 9.20 F(1)= 84.665 p=1.671e-13 p=1.04e-13 (df=71)
tDCS (main, at Day 1) t(67)= 1.04 F(1)= 1.086 p=0.3012 p=0.3116 (df=18)
HONEST LME -- per-animal random slope (day|subject): interaction F(1,8.8)=0.61, p=0.456
Note: Satterthwaite DF on the random-INTERCEPT model above stays ~= residual
(the slope's error is at the session level), so it does NOT fix pseudoreplication.
Letting each animal have its OWN slope collapses the interaction DF toward the animal
count -- this, and the per-animal slope test, are the honest learning-rate inference.
INTERPRETATION
- days x tDCS interaction: n.s. (p=0.3112, slope diff=0.94) -> slopes are parallel (no differential change over training).
analysis/matlab/results/paper_mergeA2.txt
+10 -5
View File
@@ -57,13 +57,18 @@ Group: Error
effect t (df) F (df1) p
------------------------------------------------------------------
stim x day (interaction) t(105)= -2.66 F(1)= 7.094 p=0.008953
day (learning) t(105)= 9.82 F(1)= 96.388 p=1.573e-16
stim (main, Day 1) t(105)= 3.43 F(1)= 11.778 p=0.0008585
effect t(df) / F(df1) p (resid) Satterthwaite: p (df)
----------------------------------------------------------------------------
stim x day (interaction) t(105)= -2.66 F(1)= 7.094 p=0.008953 p=0.008906 (df=109)
day (learning) t(105)= 9.82 F(1)= 96.388 p=1.573e-16 p=1.113e-16 (df=109)
stim (main, Day 1) t(105)= 3.43 F(1)= 11.778 p=0.0008585 p=0.000848 (df=109)
interaction 95% CI: [-3.80, -0.56]
HONEST LME -- per-rat random slope (day|rat): interaction F(1,9.7)=0.11, p=0.746
Satterthwaite DF on the random-INTERCEPT model above stays ~= residual (the
slope's error is at session level), so it does NOT fix pseudoreplication. A per-animal
random slope collapses the interaction DF toward the animal count -- the honest test.
INTERPRETATION
- stim x day interaction: SIGNIFICANT (p=0.0090, slope diff=-2.18) -> tDCS (Box-B2) improves SLOWER -- groups converge.
- stim main effect on Day 1 (our day 0): SIGNIFICANT (p=0.0009) -> groups already DIFFER on Day 1.
analysis/matlab/results/paper_mergeB2.txt
+16 -6
View File
@@ -1,7 +1,8 @@
==============================================================================
PAPER LME REPLICATION -- mergeB2
PAPER LME REPLICATION -- mergeB2 (full range)
==============================================================================
model: behavior ~ stim + day + stim:day + (1|rat) [stim: Electrode-Box-B2=1 vs Electrode-Box-A2=0]
behavior = successful reaches (COUNT per session)
N = 8 rats, 109 sessions (day raw; day 0 = paper "Day 1")
day coverage: stim(B2) 0..22, control(A2) 0..13
** WARNING: unequal day coverage -- the full-range stim:day interaction
@@ -56,12 +57,21 @@ Group: Error
effect t (df) F (df1) p
------------------------------------------------------------------
stim x day (interaction) t(105)= -1.43 F(1)= 2.043 p=0.1559
day (learning) t(105)= 6.89 F(1)= 47.494 p=4.234e-10
stim (main, Day 1) t(105)= 2.45 F(1)= 6.021 p=0.01578
effect t(df) / F(df1) p (resid) Satterthwaite: p (df)
----------------------------------------------------------------------------
stim x day (interaction) t(105)= -1.43 F(1)= 2.043 p=0.1559 p=0.1558 (df=109)
day (learning) t(105)= 6.89 F(1)= 47.494 p=4.234e-10 p=3.734e-10 (df=109)
stim (main, Day 1) t(105)= 2.45 F(1)= 6.021 p=0.01578 p=0.01572 (df=109)
interaction 95% CI: [-3.50, +0.57]
HONEST LME -- per-rat random slope (day|rat): interaction F(1,11.8)=0.02, p=0.9027
Satterthwaite DF on the random-INTERCEPT model above stays ~= residual (the
slope's error is at session level), so it does NOT fix pseudoreplication. A per-animal
random slope collapses the interaction DF toward the animal count -- the honest test.
INTERPRETATION
- stim x day interaction: n.s. (p=0.1559, slope diff=-1.47) -> slopes parallel -- no differential learning rate over this window.
- stim main effect on Day 1 (our day 0): SIGNIFICANT (p=0.0158) -> groups already DIFFER on Day 1.
Paper (N=24): interaction t(227)=2.68, F(1)=7.12, p=0.008; day t(227)=9.64,
F(1)=267.64, p=1.2e-18; stim t(227)=0.23, F(1)=0.053, p=0.81.
analysis/matlab/results/paper_mergeNaive.txt
+16 -6
View File
@@ -1,7 +1,8 @@
==============================================================================
PAPER LME REPLICATION -- mergeNaive
PAPER LME REPLICATION -- mergeNaive (full range)
==============================================================================
model: behavior ~ stim + day + stim:day + (1|rat) [stim: Electrode-Box-B2=1 vs Electrode-Box-A2=0]
behavior = successful reaches (COUNT per session)
N = 11 rats, 170 sessions (day raw; day 0 = paper "Day 1")
day coverage: stim(B2) 0..22, control(A2) 0..26
** WARNING: unequal day coverage -- the full-range stim:day interaction
@@ -56,12 +57,21 @@ Group: Error
effect t (df) F (df1) p
------------------------------------------------------------------
stim x day (interaction) t(166)= 2.70 F(1)= 7.299 p=0.007614
day (learning) t(166)= 10.58 F(1)= 111.990 p=2.508e-20
stim (main, Day 1) t(166)= 1.27 F(1)= 1.611 p=0.2062
effect t(df) / F(df1) p (resid) Satterthwaite: p (df)
----------------------------------------------------------------------------
stim x day (interaction) t(166)= 2.70 F(1)= 7.299 p=0.007614 p=0.007598 (df=170)
day (learning) t(166)= 10.58 F(1)=111.990 p=2.508e-20 p=2.011e-20 (df=170)
stim (main, Day 1) t(166)= 1.27 F(1)= 1.611 p=0.2062 p=0.217 (df=23)
interaction 95% CI: [+0.43, +2.74]
HONEST LME -- per-rat random slope (day|rat): interaction F(1,11.4)=0.92, p=0.3565
Satterthwaite DF on the random-INTERCEPT model above stays ~= residual (the
slope's error is at session level), so it does NOT fix pseudoreplication. A per-animal
random slope collapses the interaction DF toward the animal count -- the honest test.
INTERPRETATION
- stim x day interaction: SIGNIFICANT (p=0.0076, slope diff=+1.58) -> tDCS (Box-B2) improves FASTER -- benefit accumulates over training.
- stim main effect on Day 1 (our day 0): n.s. (p=0.2062) -> groups are comparable on Day 1.
Paper (N=24): interaction t(227)=2.68, F(1)=7.12, p=0.008; day t(227)=9.64,
F(1)=267.64, p=1.2e-18; stim t(227)=0.23, F(1)=0.053, p=0.81.
analysis/matlab/results/paper_mergeNaive_d0_10.txt
+10 -5
View File
@@ -56,13 +56,18 @@ Group: Error
effect t (df) F (df1) p
------------------------------------------------------------------
stim x day (interaction) t(111)= 1.97 F(1)= 3.880 p=0.05134
day (learning) t(111)= 15.57 F(1)= 242.567 p=1.081e-29
stim (main, Day 1) t(111)= 1.32 F(1)= 1.742 p=0.1895
effect t(df) / F(df1) p (resid) Satterthwaite: p (df)
----------------------------------------------------------------------------
stim x day (interaction) t(111)= 1.97 F(1)= 3.880 p=0.05134 p=0.05149 (df=105)
day (learning) t(111)= 15.57 F(1)=242.567 p=1.081e-29 p=3.188e-29 (df=107)
stim (main, Day 1) t(111)= 1.32 F(1)= 1.742 p=0.1895 p=0.2003 (df=22)
interaction 95% CI: [-0.01, +3.25]
HONEST LME -- per-rat random slope (day|rat): interaction F(1,10.1)=2.00, p=0.1875
Satterthwaite DF on the random-INTERCEPT model above stays ~= residual (the
slope's error is at session level), so it does NOT fix pseudoreplication. A per-animal
random slope collapses the interaction DF toward the animal count -- the honest test.
INTERPRETATION
- stim x day interaction: n.s. (p=0.0513, slope diff=+1.62) -> slopes parallel -- no differential learning rate over this window.
- stim main effect on Day 1 (our day 0): n.s. (p=0.1895) -> groups are comparable on Day 1.
analysis/matlab/results/paper_mergeNaive_d0_13.txt
+10 -5
View File
@@ -56,13 +56,18 @@ Group: Error
effect t (df) F (df1) p
------------------------------------------------------------------
stim x day (interaction) t(134)= 2.66 F(1)= 7.068 p=0.008801
day (learning) t(134)= 15.01 F(1)= 225.182 p=1.766e-30
stim (main, Day 1) t(134)= 1.28 F(1)= 1.629 p=0.204
effect t(df) / F(df1) p (resid) Satterthwaite: p (df)
----------------------------------------------------------------------------
stim x day (interaction) t(134)= 2.66 F(1)= 7.068 p=0.008801 p=0.008831 (df=130)
day (learning) t(134)= 15.01 F(1)=225.182 p=1.766e-30 p=2.418e-30 (df=132)
stim (main, Day 1) t(134)= 1.28 F(1)= 1.629 p=0.204 p=0.2142 (df=24)
interaction 95% CI: [+0.46, +3.14]
HONEST LME -- per-rat random slope (day|rat): interaction F(1,54.8)=6.46, p=0.01387
Satterthwaite DF on the random-INTERCEPT model above stays ~= residual (the
slope's error is at session level), so it does NOT fix pseudoreplication. A per-animal
random slope collapses the interaction DF toward the animal count -- the honest test.
INTERPRETATION
- stim x day interaction: SIGNIFICANT (p=0.0088, slope diff=+1.80) -> tDCS (Box-B2) improves FASTER -- benefit accumulates over training.
- stim main effect on Day 1 (our day 0): n.s. (p=0.2040) -> groups are comparable on Day 1.
analysis/matlab/results/paper_unmerged.txt
+18 -6
View File
@@ -1,9 +1,12 @@
==============================================================================
PAPER LME REPLICATION -- unmerged
PAPER LME REPLICATION -- unmerged (full range)
==============================================================================
model: behavior ~ stim + day + stim:day + (1|rat) [stim: Electrode-Box-B2=1 vs Electrode-Box-A2=0]
behavior = successful reaches (COUNT per session)
N = 6 rats, 71 sessions (day raw; day 0 = paper "Day 1")
day coverage: stim(B2) 0..14, control(A2) 0..13
(Equal day coverage -- the stim:day interaction over this window is NOT
confounded by the full-range coverage imbalance.)
==============================================================================
FULL MODEL SUMMARY -- fitlme: behavior ~ stim + day + stim:day + (1|rat)
@@ -53,12 +56,21 @@ Group: Error
effect t (df) F (df1) p
------------------------------------------------------------------
stim x day (interaction) t(67)= 1.02 F(1)= 1.041 p=0.3112
day (learning) t(67)= 9.20 F(1)= 84.665 p=1.671e-13
stim (main, Day 1) t(67)= 1.04 F(1)= 1.086 p=0.3012
effect t(df) / F(df1) p (resid) Satterthwaite: p (df)
----------------------------------------------------------------------------
stim x day (interaction) t(67)= 1.02 F(1)= 1.041 p=0.3112 p=0.3111 (df=70)
day (learning) t(67)= 9.20 F(1)= 84.665 p=1.671e-13 p=1.04e-13 (df=71)
stim (main, Day 1) t(67)= 1.04 F(1)= 1.086 p=0.3012 p=0.3116 (df=18)
interaction 95% CI: [-0.90, +2.77]
HONEST LME -- per-rat random slope (day|rat): interaction F(1,8.8)=0.61, p=0.456
Satterthwaite DF on the random-INTERCEPT model above stays ~= residual (the
slope's error is at session level), so it does NOT fix pseudoreplication. A per-animal
random slope collapses the interaction DF toward the animal count -- the honest test.
INTERPRETATION
- stim x day interaction: n.s. (p=0.3112, slope diff=+0.94) -> slopes parallel -- no differential learning rate over this window.
- stim main effect on Day 1 (our day 0): n.s. (p=0.3012) -> groups are comparable on Day 1.
Paper (N=24): interaction t(227)=2.68, F(1)=7.12, p=0.008; day t(227)=9.64,
F(1)=267.64, p=1.2e-18; stim t(227)=0.23, F(1)=0.053, p=0.81.
analysis/matlab/results/phase_mergeA2.txt
+10 -8
View File
@@ -3,15 +3,17 @@ PHASED days x tDCS LME -- mergeA2 (Box-B2 vs Box-A2)
==============================================================================
model per phase: success ~ dayp*tDCS + (1|subject) [dayp = day - phaseStart]
phase N A2slope B2slope day p tDCS(lvl) p interaction p slopeDiff [95% CI]
--------------------------------------------------------------------------------------------
0-5 8 10.54 15.16 1.5e-10 0.693 0.014 +4.62 [+1.00, +8.24]
6-10 7 3.17 4.83 0.035 0.127 0.390 +1.66 [-2.22, +5.53]
6-13 7 2.92 4.47 0.00032 0.125 0.118 +1.55 [-0.41, +3.52]
phase N A2slp B2slp day p tDCS p intP(res) intP(Satt) intP(RS) slopeDiff [95% CI]
--------------------------------------------------------------------------------------------------------
0-5 8 10.54 15.16 1.5e-10 0.693 0.014 0.014 0.089 +4.62 [+1.00,+8.24]
6-10 7 3.17 4.83 0.035 0.127 0.390 0.390 0.466 +1.66 [-2.22,+5.53]
6-13 7 2.92 4.47 0.00032 0.125 0.118 0.118 0.303 +1.55 [-0.41,+3.52]
Note: the interaction p (and CI) use fitlme observation-level DF and are
ANTICONSERVATIVE at these small subject counts (see tdcs_power_sim). The early
phase carries the Box-B2 faster-acquisition signal; late phases converge.
Columns: intP(res)=observation-level DF (anticonservative); intP(Satt)=Satterthwaite
DF (~= residual on this random-intercept model); intP(RS)=per-animal random-slope
(dayp|subject) model, the honest test (DF collapses toward the animal count; 'n/a' if it
did not converge). The early phase carries the Box-B2 faster-acquisition signal; late
phases converge.
==============================================================================
FULL MODEL SUMMARY -- phase 0-5: success ~ dayp*tDCS + (1|subject)
==============================================================================
analysis/matlab/results/phase_mergeB2.txt
+10 -8
View File
@@ -3,15 +3,17 @@ PHASED days x tDCS LME -- mergeB2 (Box-B2 vs Box-A2)
==============================================================================
model per phase: success ~ dayp*tDCS + (1|subject) [dayp = day - phaseStart]
phase N A2slope B2slope day p tDCS(lvl) p interaction p slopeDiff [95% CI]
--------------------------------------------------------------------------------------------
0-5 8 9.86 14.65 2.9e-08 0.399 0.013 +4.79 [+1.06, +8.53]
6-10 7 2.40 4.80 0.18 0.142 0.253 +2.40 [-1.80, +6.60]
6-13 7 1.71 4.42 0.083 0.127 0.018 +2.71 [+0.50, +4.93]
phase N A2slp B2slp day p tDCS p intP(res) intP(Satt) intP(RS) slopeDiff [95% CI]
--------------------------------------------------------------------------------------------------------
0-5 8 9.86 14.65 2.9e-08 0.399 0.013 0.013 0.088 +4.79 [+1.06,+8.53]
6-10 7 2.40 4.80 0.18 0.142 0.253 0.254 0.328 +2.40 [-1.80,+6.60]
6-13 7 1.71 4.42 0.083 0.127 0.018 0.018 0.024 +2.71 [+0.50,+4.93]
Note: the interaction p (and CI) use fitlme observation-level DF and are
ANTICONSERVATIVE at these small subject counts (see tdcs_power_sim). The early
phase carries the Box-B2 faster-acquisition signal; late phases converge.
Columns: intP(res)=observation-level DF (anticonservative); intP(Satt)=Satterthwaite
DF (~= residual on this random-intercept model); intP(RS)=per-animal random-slope
(dayp|subject) model, the honest test (DF collapses toward the animal count; 'n/a' if it
did not converge). The early phase carries the Box-B2 faster-acquisition signal; late
phases converge.
==============================================================================
FULL MODEL SUMMARY -- phase 0-5: success ~ dayp*tDCS + (1|subject)
==============================================================================
analysis/matlab/results/phase_mergeNaive.txt
+10 -8
View File
@@ -3,15 +3,17 @@ PHASED days x tDCS LME -- mergeNaive (Box-B2 vs Box-A2)
==============================================================================
model per phase: success ~ dayp*tDCS + (1|subject) [dayp = day - phaseStart]
phase N A2slope B2slope day p tDCS(lvl) p interaction p slopeDiff [95% CI]
--------------------------------------------------------------------------------------------
0-5 11 8.26 15.16 1.6e-10 0.852 0.000 +6.90 [+3.40, +10.41]
6-10 10 5.97 4.83 3.9e-05 0.007 0.584 -1.14 [-5.31, +3.03]
6-13 10 2.83 4.47 0.00071 0.012 0.196 +1.65 [-0.87, +4.17]
phase N A2slp B2slp day p tDCS p intP(res) intP(Satt) intP(RS) slopeDiff [95% CI]
--------------------------------------------------------------------------------------------------------
0-5 11 8.26 15.16 1.6e-10 0.852 0.000 0.000 0.013 +6.90 [+3.40,+10.41]
6-10 10 5.97 4.83 3.9e-05 0.007 0.584 0.585 0.643 -1.14 [-5.31,+3.03]
6-13 10 2.83 4.47 0.00071 0.012 0.196 0.196 0.333 +1.65 [-0.87,+4.17]
Note: the interaction p (and CI) use fitlme observation-level DF and are
ANTICONSERVATIVE at these small subject counts (see tdcs_power_sim). The early
phase carries the Box-B2 faster-acquisition signal; late phases converge.
Columns: intP(res)=observation-level DF (anticonservative); intP(Satt)=Satterthwaite
DF (~= residual on this random-intercept model); intP(RS)=per-animal random-slope
(dayp|subject) model, the honest test (DF collapses toward the animal count; 'n/a' if it
did not converge). The early phase carries the Box-B2 faster-acquisition signal; late
phases converge.
==============================================================================
FULL MODEL SUMMARY -- phase 0-5: success ~ dayp*tDCS + (1|subject)
==============================================================================
analysis/matlab/results/phase_unmerged.txt
+10 -8
View File
@@ -3,15 +3,17 @@ PHASED days x tDCS LME -- unmerged (Box-B2 vs Box-A2)
==============================================================================
model per phase: success ~ dayp*tDCS + (1|subject) [dayp = day - phaseStart]
phase N A2slope B2slope day p tDCS(lvl) p interaction p slopeDiff [95% CI]
--------------------------------------------------------------------------------------------
0-5 6 9.86 15.22 1.6e-07 0.689 0.015 +5.36 [+1.11, +9.62]
6-10 5 2.40 4.17 0.23 0.098 0.486 +1.77 [-3.41, +6.95]
6-13 5 1.71 3.98 0.12 0.101 0.097 +2.27 [-0.44, +4.97]
phase N A2slp B2slp day p tDCS p intP(res) intP(Satt) intP(RS) slopeDiff [95% CI]
--------------------------------------------------------------------------------------------------------
0-5 6 9.86 15.22 1.6e-07 0.689 0.015 0.015 0.133 +5.36 [+1.11,+9.62]
6-10 5 2.40 4.17 0.23 0.098 0.486 0.486 0.555 +1.77 [-3.41,+6.95]
6-13 5 1.71 3.98 0.12 0.101 0.097 0.097 0.138 +2.27 [-0.44,+4.97]
Note: the interaction p (and CI) use fitlme observation-level DF and are
ANTICONSERVATIVE at these small subject counts (see tdcs_power_sim). The early
phase carries the Box-B2 faster-acquisition signal; late phases converge.
Columns: intP(res)=observation-level DF (anticonservative); intP(Satt)=Satterthwaite
DF (~= residual on this random-intercept model); intP(RS)=per-animal random-slope
(dayp|subject) model, the honest test (DF collapses toward the animal count; 'n/a' if it
did not converge). The early phase carries the Box-B2 faster-acquisition signal; late
phases converge.
==============================================================================
FULL MODEL SUMMARY -- phase 0-5: success ~ dayp*tDCS + (1|subject)
==============================================================================