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

analysis(matlab): rate + count outputs, variation batch 5

Browse Source 
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
This commit is contained in:
Experiments DB Dev
2026-07-24 02:32:53 -04:00
parent 1606eb698b
commit fd40b6fb54
48 changed files with 1247 additions and 595 deletions
Download Patch File Download Diff File Expand all files Collapse all files
analysis/matlab/variations/unmerge_d0_13/analyze.m
+63 -39
View File
@@ -1,30 +1,58 @@
% Variation analysis -- the paper's linear mixed model on the successful-reach % Variation analysis -- the paper's linear mixed model on this folder's data,
% COUNT, fit on this folder's curated data subset. % for BOTH metrics:
% metric = count : behavior = # successes -> result.txt
% metric = rate : behavior = success / attempts -> result_rate.txt
% (rate uses only sessions with attempts > 0)
% %
% model: behavior ~ stim + day + stim:day + (1|rat) % model: behavior ~ stim + day + stim:day + (1|rat)
% behavior = successful reaches (count per session) % stim = 1 treatment / 0 control; day = training day within window (0 =
% stim = 1 for the treatment group(s), 0 for the control group(s) % first analyzed day); rat = subject (random intercept).
% day = training day within this window (0 = first analyzed day) % For the interaction we report residual DF, Satterthwaite DF, and the honest
% rat = subject (random intercept) % per-animal random-slope test. Self-contained: reads data.csv beside this
% % script. Run headless with: matlab -batch "analyze"
% Self-contained: reads data.csv beside this script and writes result.txt. % (Copy of analysis/matlab/variation_analyze.m; see make_variations.m.)
% Run headless from this folder with: matlab -batch "analyze"
% (This is a copy of analysis/matlab/variation_analyze.m; see make_variations.m.)
here = fileparts(mfilename('fullpath')); here = fileparts(mfilename('fullpath'));
if isempty(here); here = pwd; end if isempty(here); here = pwd; end
vname = regexprep(here, '.*[/\\]', ''); % folder name = variation id vname = regexprep(here, '.*[/\\]', '');
D = readtable(fullfile(here, 'data.csv'), 'TextType', 'string'); D = readtable(fullfile(here, 'data.csv'), 'TextType', 'string');
tbl = table(D.success, D.day - min(D.day), double(D.stim), categorical(D.subject), ... Rc = localAnalyze(D, 'count', here, vname);
Rr = localAnalyze(D, 'rate', here, vname);
% Machine-readable handoff for SUMMARY.csv (count drives it; rate appended).
VARRESULT = struct('name', vname, 'nRats', Rc.nRats, 'nObs', Rc.nObs, ...
'interP', Rc.interP, 'interEst', Rc.interEst, ...
'interPsatt', Rc.interPsatt, 'interPrs', Rc.interPrs, ...
'stimP', Rc.stimP, 'dayP', Rc.dayP, 'covEqual', Rc.covEqual, ...
'interPrate', Rr.interP, 'interEstRate', Rr.interEst, 'interPrsRate', Rr.interPrs);
% ------------------------------------------------------------------ helper
function R = localAnalyze(D, metric, here, vname)
if strcmp(metric, 'rate')
D = D(D.total > 0, :);
beh = D.success ./ D.total;
mlabel = 'success RATE (success/attempts)'; suffix = '_rate';
else
beh = D.success;
mlabel = 'success COUNT'; suffix = '';
end
R = struct('interP', NaN, 'interEst', NaN, 'interPsatt', NaN, 'interPrs', NaN, ...
'stimP', NaN, 'dayP', NaN, 'nRats', numel(unique(D.subject)), ...
'nObs', height(D), 'covEqual', false);
if numel(unique(D.stim)) < 2 || numel(unique(D.day)) < 2
localWrite(sprintf('VARIATION: %s [metric: %s]\nInsufficient data for this metric.\n', ...
vname, mlabel), here, suffix);
return
end
tbl = table(beh, D.day - min(D.day), double(D.stim), categorical(D.subject), ...
'VariableNames', {'behavior', 'day', 'stim', 'rat'}); 'VariableNames', {'behavior', 'day', 'stim', 'rat'});
m = fitlme(tbl, 'behavior ~ stim + day + stim:day + (1|rat)'); m = fitlme(tbl, 'behavior ~ stim + day + stim:day + (1|rat)');
C = m.Coefficients; A = anova(m); ci = coefCI(m); C = m.Coefficients; A = anova(m); ci = coefCI(m);
As = anova(m, 'DFMethod', 'satterthwaite'); % Satterthwaite denominator DF As = anova(m, 'DFMethod', 'satterthwaite');
% Honest test: refit with a per-animal random SLOPE so the interaction DF
% collapses toward the animal count (guarded -- may not converge in short windows).
rsP = NaN; rsDf = NaN; rsF = NaN; rsOk = false; rsP = NaN; rsDf = NaN; rsF = NaN; rsOk = false;
wst = warning('off', 'all'); wst = warning('off', 'all');
try try
@@ -43,6 +71,7 @@ row = @(nm, t) sprintf('%-26s t(%d)=%6.2f F(%d)=%7.3f p=%.4g p=%.4g (df=%.0f
C.DF(gi(t)), C.tStat(gi(t)), A.DF1(ga(t)), A.FStat(ga(t)), C.pValue(gi(t)), ... C.DF(gi(t)), C.tStat(gi(t)), A.DF1(ga(t)), A.FStat(ga(t)), C.pValue(gi(t)), ...
As.pValue(gs(t)), As.DF2(gs(t))); As.pValue(gs(t)), As.DF2(gs(t)));
ii = gi('day:stim'); pI = C.pValue(ii); eI = C.Estimate(ii);
maxT = max(D.day(D.stim == 1)); minT = min(D.day(D.stim == 1)); maxT = max(D.day(D.stim == 1)); minT = min(D.day(D.stim == 1));
maxC = max(D.day(D.stim == 0)); minC = min(D.day(D.stim == 0)); maxC = max(D.day(D.stim == 0)); minC = min(D.day(D.stim == 0));
if abs(maxT - maxC) > 2 if abs(maxT - maxC) > 2
@@ -50,20 +79,14 @@ if abs(maxT - maxC) > 2
else else
cov = '(equal day coverage over this window)'; cov = '(equal day coverage over this window)';
end end
if pI >= 0.05; verdict = 'n.s. -- slopes parallel (no differential learning rate)';
ii = gi('day:stim'); pI = C.pValue(ii); eI = C.Estimate(ii); elseif eI > 0; verdict = 'SIGNIFICANT positive -- treatment improves FASTER (benefit accumulates)';
if pI >= 0.05 else; verdict = 'SIGNIFICANT negative -- treatment improves SLOWER (groups converge)'; end
verdict = 'n.s. -- slopes parallel (no differential learning rate)';
elseif eI > 0
verdict = 'SIGNIFICANT positive -- treatment improves FASTER (benefit accumulates)';
else
verdict = 'SIGNIFICANT negative -- treatment improves SLOWER (groups converge)';
end
bar = repmat('=', 1, 78); bar = repmat('=', 1, 78);
raw = regexprep(evalc('disp(m)'), '</?strong>', ''); raw = regexprep(evalc('disp(m)'), '</?strong>', '');
s = sprintf('%s\nVARIATION: %s\n%s\n', bar, vname, bar); s = sprintf('%s\nVARIATION: %s [metric: %s]\n%s\n', bar, vname, mlabel, bar);
s = [s sprintf('model: behavior ~ stim + day + stim:day + (1|rat) (behavior = success COUNT)\n')]; s = [s sprintf('model: behavior ~ stim + day + stim:day + (1|rat) (behavior = %s)\n', mlabel)];
s = [s sprintf('day = training day within window (0 = first analyzed day)\n')]; s = [s sprintf('day = training day within window (0 = first analyzed day)\n')];
s = [s sprintf('treatment (stim=1): %s\n', strjoin(cellstr(unique(D.group(D.stim == 1))), ', '))]; s = [s sprintf('treatment (stim=1): %s\n', strjoin(cellstr(unique(D.group(D.stim == 1))), ', '))];
s = [s sprintf('control (stim=0): %s\n', strjoin(cellstr(unique(D.group(D.stim == 0))), ', '))]; s = [s sprintf('control (stim=0): %s\n', strjoin(cellstr(unique(D.group(D.stim == 0))), ', '))];
@@ -74,7 +97,7 @@ s = [s sprintf('%-26s %-18s %-12s %s\n%s\n', 'effect', 't(df) / F(df1)', 'p (res
s = [s row('stim x day (interaction)', 'day:stim')]; s = [s row('stim x day (interaction)', 'day:stim')];
s = [s row('day (learning)', 'day')]; s = [s row('day (learning)', 'day')];
s = [s row('stim (main, window start)', 'stim')]; s = [s row('stim (main, window start)', 'stim')];
s = [s sprintf('interaction 95%% CI: [%+.2f, %+.2f]\n', ci(ii, 1), ci(ii, 2))]; s = [s sprintf('interaction 95%% CI: [%+.4g, %+.4g]\n', ci(ii, 1), ci(ii, 2))];
if rsOk if rsOk
s = [s sprintf('HONEST LME (per-animal random slope, day|rat): interaction F(1,%.1f)=%.2f, p=%.4g\n', rsDf, rsF, rsP)]; s = [s sprintf('HONEST LME (per-animal random slope, day|rat): interaction F(1,%.1f)=%.2f, p=%.4g\n', rsDf, rsF, rsP)];
else else
@@ -82,17 +105,18 @@ else
end end
s = [s sprintf([' (Satterthwaite DF ~= residual on this random-intercept model; the random-slope\n' ... s = [s sprintf([' (Satterthwaite DF ~= residual on this random-intercept model; the random-slope\n' ...
' model above is the honest learning-rate test -- DF collapses toward the animal count.)\n'])]; ' model above is the honest learning-rate test -- DF collapses toward the animal count.)\n'])];
s = [s sprintf('INTERPRETATION: stim x day interaction %s (p=%.4g, slope diff=%+.2f)\n', verdict, pI, eI)]; s = [s sprintf('INTERPRETATION: stim x day interaction %s (p=%.4g, slope diff=%+.4g)\n', verdict, pI, eI)];
s = [s sprintf('Paper (N=24): interaction t(227)=2.68, F(1)=7.12, p=0.008.\n')]; s = [s sprintf('Paper (N=24, count): interaction t(227)=2.68, F(1)=7.12, p=0.008.\n')];
localWrite(s, here, suffix);
R = struct('interP', pI, 'interEst', eI, 'interPsatt', As.pValue(gs('day:stim')), ...
'interPrs', rsP, 'stimP', C.pValue(gi('stim')), 'dayP', C.pValue(gi('day')), ...
'nRats', numel(unique(D.subject)), 'nObs', height(D), 'covEqual', abs(maxT - maxC) <= 2);
end
function localWrite(s, here, suffix)
fprintf('%s', s); fprintf('%s', s);
fid = fopen(fullfile(here, 'result.txt'), 'w'); fid = fopen(fullfile(here, ['result' suffix '.txt']), 'w');
fprintf(fid, '%s', s); fprintf(fid, '%s', s); fclose(fid);
fclose(fid); end
% Machine-readable handoff for the summary table (see make_variations.m).
VARRESULT = struct('name', vname, 'nRats', numel(unique(D.subject)), ...
'nObs', height(D), 'interP', pI, 'interEst', eI, ...
'interPsatt', As.pValue(gs('day:stim')), 'interPrs', rsP, ...
'stimP', C.pValue(gi('stim')), 'dayP', C.pValue(gi('day')), ...
'covEqual', abs(maxT - maxC) <= 2);
analysis/matlab/variations/unmerge_d0_13/learning_curve.png
BIN
View File
Binary file not shown.
Side by Side Swipe Overlay

Before

Width:  |  Height:  |  Size: 44 KiB

After

Width:  |  Height:  |  Size: 44 KiB

analysis/matlab/variations/unmerge_d0_13/learning_curve_rate.png
BIN
View File
Binary file not shown.
Side by Side

After

Width:  |  Height:  |  Size: 41 KiB

analysis/matlab/variations/unmerge_d0_13/logpower_result.txt
+5 -6
View File
@@ -1,7 +1,7 @@
============================================================================== ==============================================================================
LOG-DAY MODEL + COHEN'S f + POWER -- unmerge_d0_13 LOG-DAY MODEL + COHEN'S f + POWER -- unmerge_d0_13 [metric: # successes (count)]
============================================================================== ==============================================================================
model: behavior ~ stim + log(day) + stim:log(day) + (1|rat) (success COUNT) model: behavior ~ stim + log(day) + stim:log(day) + (1|rat) (behavior = # successes (count))
log(day) uses 1-indexed training day (our day 0 = paper "Day 1") log(day) uses 1-indexed training day (our day 0 = paper "Day 1")
observed groups: stim n=3, control n=3 nrep=120, alpha=0.05 observed groups: stim n=3, control n=3 nrep=120, alpha=0.05
@@ -10,7 +10,7 @@ stim x log(day) interaction: F(1,66)=8.712 p(resid)=0.004373 p(Satt)=0.004357
honest per-animal random slope (log-day): F(1,12.9)=6.96 p=0.02062 honest per-animal random slope (log-day): F(1,12.9)=6.96 p=0.02062
Cohen's f (interaction, partial eta^2=0.015) = 0.124 (small-medium; f: .10 small, .25 medium, .40 large) Cohen's f (interaction, partial eta^2=0.015) = 0.124 (small-medium; f: .10 small, .25 medium, .40 large)
--- power simulation (log-day ground truth: stim:logday=+11.06, ratSD=5.68, resSD=11.36) --- --- power simulation (log-day ground truth: stim:logday=+11.06, ratSD=5.676, resSD=11.36) ---
true stim:log(day) = +11.06 (100% of observed) true stim:log(day) = +11.06 (100% of observed)
N/group | per-animal power | LME power N/group | per-animal power | LME power
@@ -22,7 +22,7 @@ Cohen's f (interaction, partial eta^2=0.015) = 0.124 (small-medium; f: .10 smal
16 | 1.00 | 1.00 16 | 1.00 | 1.00
24 | 1.00 | 1.00 24 | 1.00 | 1.00
true stim:log(day) = +5.53 (50% of observed) true stim:log(day) = +5.528 (50% of observed)
N/group | per-animal power | LME power N/group | per-animal power | LME power
------------------------------------------ ------------------------------------------
3 | 0.18 | 0.38 <- observed 3 | 0.18 | 0.38 <- observed
@@ -32,5 +32,4 @@ Cohen's f (interaction, partial eta^2=0.015) = 0.124 (small-medium; f: .10 smal
16 | 0.97 | 0.98 16 | 0.97 | 0.98
24 | 0.98 | 0.98 24 | 0.98 | 0.98
Read the per-animal column as the honest power; the LME column matches the Read per-animal as the honest power; LME matches the paper's power code (optimistic).
paper's power code (anova interaction p, observation-level DF) and is optimistic.
analysis/matlab/variations/unmerge_d0_13/logpower_result_rate.txt
+35
View File
@@ -0,0 +1,35 @@
==============================================================================
LOG-DAY MODEL + COHEN'S f + POWER -- unmerge_d0_13 [metric: success RATE]
==============================================================================
model: behavior ~ stim + log(day) + stim:log(day) + (1|rat) (behavior = success RATE)
log(day) uses 1-indexed training day (our day 0 = paper "Day 1")
observed groups: stim n=3, control n=3 nrep=120, alpha=0.05
--- fitted on real data ---
stim x log(day) interaction: F(1,66)=12.243 p(resid)=0.0008416 p(Satt)=0.0008388 (df=66)
honest per-animal random slope (log-day): F(1,29.1)=11.46 p=0.00205
Cohen's f (interaction, partial eta^2=0.030) = 0.176 (small-medium; f: .10 small, .25 medium, .40 large)
--- power simulation (log-day ground truth: stim:logday=+0.08873, ratSD=0.04364, resSD=0.07673) ---
true stim:log(day) = +0.08873 (100% of observed)
N/group | per-animal power | LME power
------------------------------------------
3 | 0.71 | 0.97 <- observed
5 | 1.00 | 1.00
8 | 1.00 | 1.00
12 | 1.00 | 1.00
16 | 1.00 | 1.00
24 | 1.00 | 1.00
true stim:log(day) = +0.04437 (50% of observed)
N/group | per-animal power | LME power
------------------------------------------
3 | 0.26 | 0.52 <- observed
5 | 0.57 | 0.72
8 | 0.85 | 0.90
12 | 0.95 | 0.96
16 | 1.00 | 1.00
24 | 1.00 | 1.00
Read per-animal as the honest power; LME matches the paper's power code (optimistic).
analysis/matlab/variations/unmerge_d0_13/logpowersim.m
+38 -39
View File
@@ -1,50 +1,52 @@
% Variation log-day analysis + Cohen's f + power simulation. % Variation log-day analysis + Cohen's f + power simulation, for BOTH metrics:
% metric = count : behavior = # successes -> logpower_result.txt
% metric = rate : behavior = success / attempts -> logpower_result_rate.txt
% %
% The paper's power code models behavior against LOG training day, not raw day: % The paper's power code models behavior against LOG training day:
% behavior ~ stim + log(day) + stim:log(day) + (1|rat). % behavior ~ stim + log(day) + stim:log(day) + (1|rat).
% Their day is 1-indexed (1..10); our data.csv day is 0-indexed (day 0 = paper % Their day is 1-indexed; our data.csv day is 0-indexed, so log(day + 1)
% "Day 1"), so log(day + 1) reproduces their transform exactly. % reproduces their transform (our day 0 = paper "Day 1"). For each metric this
% % refits that model, reports the interaction (residual / Satterthwaite / honest
% This script (a) refits that log-day model on data.csv, (b) reports the % per-animal random-slope DF) and Cohen's f (partial-eta^2 effect size), then
% interaction (residual DF, Satterthwaite DF, and the honest per-animal % runs the Monte-Carlo power sim (per-animal cluster-honest + LME power).
% random-slope test) and Cohen's f -- the partial-eta^2 effect size of the % Run: matlab -batch "logpowersim"
% interaction, var(fitted_full) - var(fitted_no_interaction) over var(behavior)
% -- and (c) runs the Monte-Carlo power simulation on the log-day model,
% scoring per-animal (cluster-honest) and LME power across N.
% Writes logpower_result.txt. Run: matlab -batch "logpowersim"
% (Copy of analysis/matlab/variation_logpower.m; see make_variation_logpower.m.) % (Copy of analysis/matlab/variation_logpower.m; see make_variation_logpower.m.)
here = fileparts(mfilename('fullpath')); here = fileparts(mfilename('fullpath'));
if isempty(here); here = pwd; end if isempty(here); here = pwd; end
vname = regexprep(here, '.*[/\\]', ''); vname = regexprep(here, '.*[/\\]', '');
D = readtable(fullfile(here, 'data.csv'), 'TextType', 'string'); D = readtable(fullfile(here, 'data.csv'), 'TextType', 'string');
FORMULA = 'behavior ~ stim + day + stim:day + (1|rat)'; % 'day' column = log(day+1)
NS = [3 5 8 12 16 24];
EFFMULS = [1 0.5];
NREP = 120;
warnState = warning('off', 'all'); localLogPower(D, 'count', here, vname);
rng(1); localLogPower(D, 'rate', here, vname);
logday = log(D.day + 1); % 0-indexed day -> their log(1-indexed day) % ---------------------------------------------------------------- per metric
tbl0 = table(D.success, logday, double(D.stim), categorical(D.subject), ... function localLogPower(D, metric, here, vname)
NS = [3 5 8 12 16 24]; EFFMULS = [1 0.5]; NREP = 120;
FORMULA = 'behavior ~ stim + day + stim:day + (1|rat)'; % 'day' = log(day+1)
if strcmp(metric, 'rate')
D = D(D.total > 0, :); beh = D.success ./ D.total; mlabel = 'success RATE'; suffix = '_rate';
else
beh = D.success; mlabel = '# successes (count)'; suffix = '';
end
warnState = warning('off', 'all'); rng(1);
logday = log(D.day + 1);
tbl0 = table(beh, logday, double(D.stim), categorical(D.subject), ...
'VariableNames', {'behavior', 'day', 'stim', 'rat'}); 'VariableNames', {'behavior', 'day', 'stim', 'rat'});
nStim = numel(unique(D.subject(D.stim == 1))); nStim = numel(unique(D.subject(D.stim == 1)));
nCtrl = numel(unique(D.subject(D.stim == 0))); nCtrl = numel(unique(D.subject(D.stim == 0)));
bar = repmat('=', 1, 78); bar = repmat('=', 1, 78);
s = sprintf('%s\nLOG-DAY MODEL + COHEN''S f + POWER -- %s\n%s\n', bar, vname, bar); s = sprintf('%s\nLOG-DAY MODEL + COHEN''S f + POWER -- %s [metric: %s]\n%s\n', bar, vname, mlabel, bar);
s = [s sprintf('model: behavior ~ stim + log(day) + stim:log(day) + (1|rat) (success COUNT)\n')]; s = [s sprintf('model: behavior ~ stim + log(day) + stim:log(day) + (1|rat) (behavior = %s)\n', mlabel)];
s = [s sprintf('log(day) uses 1-indexed training day (our day 0 = paper "Day 1")\n')]; s = [s sprintf('log(day) uses 1-indexed training day (our day 0 = paper "Day 1")\n')];
s = [s sprintf('observed groups: stim n=%d, control n=%d nrep=%d, alpha=0.05\n', nStim, nCtrl, NREP)]; s = [s sprintf('observed groups: stim n=%d, control n=%d nrep=%d, alpha=0.05\n', nStim, nCtrl, NREP)];
if nStim < 2 || nCtrl < 2 || numel(unique(tbl0.day)) < 2 if nStim < 2 || nCtrl < 2 || numel(unique(tbl0.day)) < 2
s = [s sprintf('\nInsufficient data for this analysis (need >=2 animals/group and >=2 days).\n')]; s = [s sprintf('\nInsufficient data for this analysis.\n')];
localFinish(s, here); warning(warnState); return localFinish(s, here, suffix); warning(warnState); return
end end
% ---- fitted model on the real data ----
full = fitlme(tbl0, FORMULA); full = fitlme(tbl0, FORMULA);
An = anova(full); Asatt = anova(full, 'DFMethod', 'satterthwaite'); An = anova(full); Asatt = anova(full, 'DFMethod', 'satterthwaite');
ii = strcmp(An.Term, 'day:stim'); is = strcmp(Asatt.Term, 'day:stim'); ii = strcmp(An.Term, 'day:stim'); is = strcmp(Asatt.Term, 'day:stim');
@@ -52,7 +54,6 @@ reduced = fitlme(tbl0, 'behavior ~ stim + day + (1|rat)');
eta2part = max((var(fitted(full)) - var(fitted(reduced))) / var(tbl0.behavior), 0); eta2part = max((var(fitted(full)) - var(fitted(reduced))) / var(tbl0.behavior), 0);
cohenf = sqrt(eta2part / (1 - eta2part)); cohenf = sqrt(eta2part / (1 - eta2part));
% honest per-animal random-slope interaction
rsP = NaN; rsDf = NaN; rsF = NaN; rsOk = false; rsP = NaN; rsDf = NaN; rsF = NaN; rsOk = false;
try try
mr = fitlme(tbl0, 'behavior ~ stim + day + stim:day + (day|rat)'); mr = fitlme(tbl0, 'behavior ~ stim + day + stim:day + (day|rat)');
@@ -76,19 +77,18 @@ end
s = [s sprintf('Cohen''s f (interaction, partial eta^2=%.3f) = %.3f (%s; f: .10 small, .25 medium, .40 large)\n', ... s = [s sprintf('Cohen''s f (interaction, partial eta^2=%.3f) = %.3f (%s; f: .10 small, .25 medium, .40 large)\n', ...
eta2part, cohenf, mag)]; eta2part, cohenf, mag)];
% ---- power simulation on the log-day ground truth ----
cn = full.CoefficientNames; be = full.fixedEffects; cn = full.CoefficientNames; be = full.fixedEffects;
b0 = be(strcmp(cn, '(Intercept)')); bStim = be(strcmp(cn, 'stim')); b0 = be(strcmp(cn, '(Intercept)')); bStim = be(strcmp(cn, 'stim'));
bDay = be(strcmp(cn, 'day')); bInt = be(strcmp(cn, 'day:stim')); bDay = be(strcmp(cn, 'day')); bInt = be(strcmp(cn, 'day:stim'));
psi = covarianceParameters(full); sRat = sqrt(psi{1}); sRes = sqrt(full.MSE); psi = covarianceParameters(full); sRat = sqrt(psi{1}); sRes = sqrt(full.MSE);
days = unique(tbl0.day); % the log(day) grid days = unique(tbl0.day);
s = [s sprintf('\n--- power simulation (log-day ground truth: stim:logday=%+.2f, ratSD=%.2f, resSD=%.2f) ---\n', ... s = [s sprintf('\n--- power simulation (log-day ground truth: stim:logday=%+.4g, ratSD=%.4g, resSD=%.4g) ---\n', ...
bInt, sRat, sRes)]; bInt, sRat, sRes)];
for eMul = EFFMULS for eMul = EFFMULS
bI = bInt * eMul; bI = bInt * eMul;
s = [s sprintf('\n true stim:log(day) = %+.2f (%.0f%% of observed)\n', bI, eMul * 100)]; %#ok<AGROW> s = [s sprintf('\n true stim:log(day) = %+.4g (%.0f%% of observed)\n', bI, eMul * 100)];
s = [s sprintf(' %-8s | per-animal power | LME power\n %s\n', 'N/group', repmat('-', 1, 42))]; %#ok<AGROW> s = [s sprintf(' %-8s | per-animal power | LME power\n %s\n', 'N/group', repmat('-', 1, 42))];
for N = NS for N = NS
sigPA = 0; sigL = 0; sigPA = 0; sigL = 0;
for r = 1:NREP for r = 1:NREP
@@ -102,19 +102,18 @@ for eMul = EFFMULS
end end
star = ''; star = '';
if N == nStim || N == nCtrl; star = ' <- observed'; end if N == nStim || N == nCtrl; star = ' <- observed'; end
s = [s sprintf(' %-8d | %5.2f | %5.2f%s\n', N, sigPA / NREP, sigL / NREP, star)]; %#ok<AGROW> s = [s sprintf(' %-8d | %5.2f | %5.2f%s\n', N, sigPA / NREP, sigL / NREP, star)];
end end
end end
s = [s sprintf(['\nRead the per-animal column as the honest power; the LME column matches the\n' ... s = [s sprintf('\nRead per-animal as the honest power; LME matches the paper''s power code (optimistic).\n')];
'paper''s power code (anova interaction p, observation-level DF) and is optimistic.\n'])]; localFinish(s, here, suffix);
localFinish(s, here);
warning(warnState); warning(warnState);
end
% ---------------------------------------------------------------- helpers % ---------------------------------------------------------------- helpers
function localFinish(s, here) function localFinish(s, here, suffix)
fprintf('%s', s); fprintf('%s', s);
fid = fopen(fullfile(here, 'logpower_result.txt'), 'w'); fid = fopen(fullfile(here, ['logpower_result' suffix '.txt']), 'w');
fprintf(fid, '%s', s); fclose(fid); fprintf(fid, '%s', s); fclose(fid);
end end
analysis/matlab/variations/unmerge_d0_13/plotcurve.m
+29 -19
View File
@@ -1,11 +1,12 @@
% Variation learning-curve plot, in the style of the paper: % Variation learning-curve plots, in the style of the paper:
% "Lines indicate mean (and SEM) across animals in the anodal (red) and % "Lines indicate mean (and SEM) across animals in the anodal (red) and
% control (blue) groups." % control (blue) groups."
% Plots mean +/- SEM successful reaches per training day for the treatment / % Produces TWO figures from this folder's data.csv:
% anodal group (stim = 1, red) and the control group (stim = 0, blue), reading % learning_curve.png # successes (count) per training day
% this folder's data.csv and saving learning_curve.png. The per-group N is read % learning_curve_rate.png success rate (success/attempts) per training day
% from the data (each variation pools different groups), so the legend shows the % anodal / treatment = stim 1 (red); control = stim 0 (blue). Per-group N is
% actual counts. Training day is 1-indexed (our day 0 = the paper's "Day 1"). % read from the data. Training day is 1-indexed (our day 0 = paper "Day 1") and
% the x-axis tick labels are drawn vertically.
% Run: matlab -batch "plotcurve" % Run: matlab -batch "plotcurve"
% (Copy of analysis/matlab/variation_plot.m; see make_variation_plot.m.) % (Copy of analysis/matlab/variation_plot.m; see make_variation_plot.m.)
@@ -14,15 +15,20 @@ if isempty(here); here = pwd; end
vname = regexprep(here, '.*[/\\]', ''); vname = regexprep(here, '.*[/\\]', '');
D = readtable(fullfile(here, 'data.csv'), 'TextType', 'string'); D = readtable(fullfile(here, 'data.csv'), 'TextType', 'string');
days = unique(D.day); % 0-indexed days = unique(D.day);
xd = days + 1; % plot as 1-indexed training day (paper axis) xd = days + 1; % plot as 1-indexed training day (paper axis)
red = [0.85 0.10 0.10]; red = [0.85 0.10 0.10];
blue = [0.10 0.30 0.85]; blue = [0.10 0.30 0.85];
[Ma, Sa, na] = localCurve(D, 1, days); % anodal / treatment (stim = 1) localPlot(D, days, xd, 'count', '# successes', ...
[Mc, Sc, nc] = localCurve(D, 0, days); % control (stim = 0) fullfile(here, 'learning_curve.png'), vname, red, blue);
localPlot(D, days, xd, 'rate', 'success rate', ...
fullfile(here, 'learning_curve_rate.png'), vname, red, blue);
% ------------------------------------------------------------------ helpers
function localPlot(D, days, xd, metric, ylab, outFile, vname, red, blue)
[Ma, Sa, na] = localCurve(D, 1, days, metric); % anodal / treatment
[Mc, Sc, nc] = localCurve(D, 0, days, metric); % control
fig = figure('Visible', 'off', 'Color', 'w', 'Position', [100 100 560 460]); fig = figure('Visible', 'off', 'Color', 'w', 'Position', [100 100 560 460]);
hold on hold on
e1 = errorbar(xd, Ma, Sa, '-o', 'Color', red, 'MarkerFaceColor', red, 'LineWidth', 2); e1 = errorbar(xd, Ma, Sa, '-o', 'Color', red, 'MarkerFaceColor', red, 'LineWidth', 2);
@@ -30,26 +36,30 @@ e2 = errorbar(xd, Mc, Sc, '-o', 'Color', blue, 'MarkerFaceColor', blue, 'LineWid
hold off hold off
legend([e1 e2], {sprintf('anodal, N = %d', na), sprintf('control, N = %d', nc)}, ... legend([e1 e2], {sprintf('anodal, N = %d', na), sprintf('control, N = %d', nc)}, ...
'Location', 'northwest', 'Box', 'off'); 'Location', 'northwest', 'Box', 'off');
xlabel('training day'); xlabel('training day'); ylabel(ylab);
ylabel('# successes');
title(vname, 'Interpreter', 'none'); title(vname, 'Interpreter', 'none');
set(gca, 'XTick', xd, 'FontName', 'Arial', 'FontSize', 13, 'LineWidth', 1.5, 'Box', 'off'); set(gca, 'XTick', xd, 'FontName', 'Arial', 'FontSize', 13, 'LineWidth', 1.5, 'Box', 'off');
xtickangle(90); % vertical x-axis tick labels
outFile = fullfile(here, 'learning_curve.png');
exportgraphics(fig, outFile, 'Resolution', 150); exportgraphics(fig, outFile, 'Resolution', 150);
close(fig); close(fig);
fprintf('%s: wrote learning_curve.png (anodal N=%d, control N=%d)\n', vname, na, nc); fprintf('%s: wrote %s (anodal N=%d, control N=%d)\n', vname, outFile, na, nc);
end
% ------------------------------------------------------------------ helper function [M, S, n] = localCurve(D, stimVal, days, metric)
function [M, S, n] = localCurve(D, stimVal, days) %LOCALCURVE Per-day mean and SEM across the animals in a group, for a metric.
%LOCALCURVE Per-day mean and SEM of successes across the animals in a group.
subs = unique(D.subject(D.stim == stimVal)); subs = unique(D.subject(D.stim == stimVal));
n = numel(subs); n = numel(subs);
X = nan(numel(days), n); X = nan(numel(days), n);
for j = 1:n for j = 1:n
for i = 1:numel(days) for i = 1:numel(days)
r = D.subject == subs(j) & D.day == days(i); r = D.subject == subs(j) & D.day == days(i);
if any(r); X(i, j) = mean(D.success(r)); end if ~any(r); continue; end
if strcmp(metric, 'rate')
tot = sum(D.total(r));
if tot > 0; X(i, j) = sum(D.success(r)) / tot; end
else
X(i, j) = mean(D.success(r));
end
end end
end end
M = mean(X, 2, 'omitnan'); M = mean(X, 2, 'omitnan');
analysis/matlab/variations/unmerge_d0_13/power_result.txt
+6 -7
View File
@@ -1,11 +1,11 @@
============================================================================== ==============================================================================
POWER SIMULATION -- unmerge_d0_13 POWER SIMULATION -- unmerge_d0_13 [metric: # successes (count)]
============================================================================== ==============================================================================
model: behavior ~ stim + day + stim:day + (1|rat) (success COUNT; day within-window) model: behavior ~ stim + day + stim:day + (1|rat) (behavior = # successes (count); day within-window)
observed groups: stim n=3, control n=3 nrep=120, alpha=0.05 observed groups: stim n=3, control n=3 nrep=120, alpha=0.05
ground truth: stim:day=+1.55/day, rat SD=0.00, residual SD=14.68, days=14 ground truth: stim:day=+1.547/day, rat SD=0, residual SD=14.68, days=14
true stim:day interaction = +1.55 (100% of observed) true stim:day interaction = +1.547 (100% of observed)
N/group | per-animal power | LME power N/group | per-animal power | LME power
------------------------------------------ ------------------------------------------
3 | 0.22 | 0.56 <- observed 3 | 0.22 | 0.56 <- observed
@@ -15,7 +15,7 @@ ground truth: stim:day=+1.55/day, rat SD=0.00, residual SD=14.68, days=14
16 | 0.98 | 0.98 16 | 0.98 | 0.98
24 | 1.00 | 1.00 24 | 1.00 | 1.00
true stim:day interaction = +0.77 (50% of observed) true stim:day interaction = +0.7733 (50% of observed)
N/group | per-animal power | LME power N/group | per-animal power | LME power
------------------------------------------ ------------------------------------------
3 | 0.12 | 0.22 <- observed 3 | 0.12 | 0.22 <- observed
@@ -25,5 +25,4 @@ ground truth: stim:day=+1.55/day, rat SD=0.00, residual SD=14.68, days=14
16 | 0.64 | 0.71 16 | 0.64 | 0.71
24 | 0.78 | 0.74 24 | 0.78 | 0.74
Read the per-animal column as the honest power. At the observed N this study Read the per-animal column as the honest power; LME is optimistic (obs-level DF).
is typically underpowered; per-animal power reaches ~0.8 only at larger N.
analysis/matlab/variations/unmerge_d0_13/power_result_rate.txt
+28
View File
@@ -0,0 +1,28 @@
==============================================================================
POWER SIMULATION -- unmerge_d0_13 [metric: success RATE]
==============================================================================
model: behavior ~ stim + day + stim:day + (1|rat) (behavior = success RATE; day within-window)
observed groups: stim n=3, control n=3 nrep=120, alpha=0.05
ground truth: stim:day=+0.01125/day, rat SD=0.03898, residual SD=0.0934, days=14
true stim:day interaction = +0.01125 (100% of observed)
N/group | per-animal power | LME power
------------------------------------------
3 | 0.28 | 0.66 <- observed
5 | 0.67 | 0.78
8 | 0.92 | 0.94
12 | 1.00 | 1.00
16 | 1.00 | 1.00
24 | 1.00 | 1.00
true stim:day interaction = +0.005624 (50% of observed)
N/group | per-animal power | LME power
------------------------------------------
3 | 0.15 | 0.25 <- observed
5 | 0.23 | 0.25
8 | 0.48 | 0.52
12 | 0.56 | 0.60
16 | 0.75 | 0.79
24 | 0.87 | 0.83
Read the per-animal column as the honest power; LME is optimistic (obs-level DF).
analysis/matlab/variations/unmerge_d0_13/powersim.m
+35 -34
View File
@@ -1,45 +1,49 @@
% Variation power simulation -- Monte-Carlo power for the paper's stim x day % Variation power simulation -- Monte-Carlo power for the paper's stim x day
% interaction, using THIS folder's data as the ground truth. % interaction, using THIS folder's data as the ground truth, for BOTH metrics:
% metric = count : behavior = # successes -> power_result.txt
% metric = rate : behavior = success / attempts -> power_result_rate.txt
% %
% Ground truth: fitlme(behavior ~ stim + day + stim:day + (1|rat)) on data.csv % Ground truth: fitlme(behavior ~ stim + day + stim:day + (1|rat)) on data.csv
% (success COUNT; day within-window). Its fixed effects, per-rat intercept SD, % (day within-window). Its fixed effects, per-rat intercept SD, and residual SD
% and residual SD generate NREP synthetic datasets at each rats-per-group N and % generate NREP synthetic datasets at each rats-per-group N and each true-effect
% each true-effect multiplier (1 = observed slope, 0.5 = half). Each dataset is % multiplier (1 = observed, 0.5 = half). Each is scored at alpha=0.05 by:
% scored at alpha = 0.05 two ways: % per-animal : Welch t on per-rat behavior~day slopes (cluster-honest power)
% per-animal : Welch t on per-rat behavior~day slopes (cluster-honest -- the
% honest power, matching the random-slope / per-animal inference)
% LME : the fitlme stim:day p (observation-level DF -- optimistic) % LME : the fitlme stim:day p (observation-level DF -- optimistic)
% Writes power_result.txt beside this script. Run: matlab -batch "powersim" % Writes power_result[_rate].txt. Run: matlab -batch "powersim"
% (Copy of analysis/matlab/variation_power.m; see make_variation_power.m.) % (Copy of analysis/matlab/variation_power.m; see make_variation_power.m.)
here = fileparts(mfilename('fullpath')); here = fileparts(mfilename('fullpath'));
if isempty(here); here = pwd; end if isempty(here); here = pwd; end
vname = regexprep(here, '.*[/\\]', ''); vname = regexprep(here, '.*[/\\]', '');
D = readtable(fullfile(here, 'data.csv'), 'TextType', 'string'); D = readtable(fullfile(here, 'data.csv'), 'TextType', 'string');
localPower(D, 'count', here, vname);
localPower(D, 'rate', here, vname);
% ---------------------------------------------------------------- per metric
function localPower(D, metric, here, vname)
NS = [3 5 8 12 16 24]; EFFMULS = [1 0.5]; NREP = 120;
FORMULA = 'behavior ~ stim + day + stim:day + (1|rat)'; FORMULA = 'behavior ~ stim + day + stim:day + (1|rat)';
NS = [3 5 8 12 16 24]; if strcmp(metric, 'rate')
EFFMULS = [1 0.5]; D = D(D.total > 0, :); beh = D.success ./ D.total; mlabel = 'success RATE'; suffix = '_rate';
NREP = 120; else
beh = D.success; mlabel = '# successes (count)'; suffix = '';
warnState = warning('off', 'all'); end
rng(1); warnState = warning('off', 'all'); rng(1);
day0 = min(D.day); day0 = min(D.day);
tbl0 = table(D.success, D.day - day0, double(D.stim), categorical(D.subject), ... tbl0 = table(beh, D.day - day0, double(D.stim), categorical(D.subject), ...
'VariableNames', {'behavior', 'day', 'stim', 'rat'}); 'VariableNames', {'behavior', 'day', 'stim', 'rat'});
nStim = numel(unique(D.subject(D.stim == 1))); nStim = numel(unique(D.subject(D.stim == 1)));
nCtrl = numel(unique(D.subject(D.stim == 0))); nCtrl = numel(unique(D.subject(D.stim == 0)));
bar = repmat('=', 1, 78); bar = repmat('=', 1, 78);
s = sprintf('%s\nPOWER SIMULATION -- %s\n%s\n', bar, vname, bar); s = sprintf('%s\nPOWER SIMULATION -- %s [metric: %s]\n%s\n', bar, vname, mlabel, bar);
s = [s sprintf('model: %s (success COUNT; day within-window)\n', FORMULA)]; s = [s sprintf('model: %s (behavior = %s; day within-window)\n', FORMULA, mlabel)];
s = [s sprintf('observed groups: stim n=%d, control n=%d nrep=%d, alpha=0.05\n', nStim, nCtrl, NREP)]; s = [s sprintf('observed groups: stim n=%d, control n=%d nrep=%d, alpha=0.05\n', nStim, nCtrl, NREP)];
if nStim < 2 || nCtrl < 2 || numel(unique(tbl0.day)) < 2 if nStim < 2 || nCtrl < 2 || numel(unique(tbl0.day)) < 2
s = [s sprintf('\nInsufficient data for a power simulation (need >=2 animals/group and >=2 days).\n')]; s = [s sprintf('\nInsufficient data for a power simulation.\n')];
localFinish(s, here); warning(warnState); return localFinish(s, here, suffix); warning(warnState); return
end end
lme = fitlme(tbl0, FORMULA); lme = fitlme(tbl0, FORMULA);
@@ -49,14 +53,13 @@ bDay = be(strcmp(cn, 'day')); bInt = be(strcmp(cn, 'day:stim'));
psi = covarianceParameters(lme); sRat = sqrt(psi{1}); sRes = sqrt(lme.MSE); psi = covarianceParameters(lme); sRat = sqrt(psi{1}); sRes = sqrt(lme.MSE);
days = (0:max(tbl0.day))'; days = (0:max(tbl0.day))';
s = [s sprintf('ground truth: stim:day=%+.2f/day, rat SD=%.2f, residual SD=%.2f, days=%d\n', ... s = [s sprintf('ground truth: stim:day=%+.4g/day, rat SD=%.4g, residual SD=%.4g, days=%d\n', ...
bInt, sRat, sRes, numel(days))]; bInt, sRat, sRes, numel(days))];
for eMul = EFFMULS for eMul = EFFMULS
bI = bInt * eMul; bI = bInt * eMul;
s = [s sprintf('\n true stim:day interaction = %+.2f (%.0f%% of observed)\n', bI, eMul * 100)]; %#ok<AGROW> s = [s sprintf('\n true stim:day interaction = %+.4g (%.0f%% of observed)\n', bI, eMul * 100)];
s = [s sprintf(' %-8s | per-animal power | LME power\n', 'N/group')]; %#ok<AGROW> s = [s sprintf(' %-8s | per-animal power | LME power\n %s\n', 'N/group', repmat('-', 1, 42))];
s = [s sprintf(' %s\n', repmat('-', 1, 42))]; %#ok<AGROW>
for N = NS for N = NS
sigPA = 0; sigL = 0; sigPA = 0; sigL = 0;
for r = 1:NREP for r = 1:NREP
@@ -70,20 +73,18 @@ for eMul = EFFMULS
end end
star = ''; star = '';
if N == nStim || N == nCtrl; star = ' <- observed'; end if N == nStim || N == nCtrl; star = ' <- observed'; end
s = [s sprintf(' %-8d | %5.2f | %5.2f%s\n', N, sigPA / NREP, sigL / NREP, star)]; %#ok<AGROW> s = [s sprintf(' %-8d | %5.2f | %5.2f%s\n', N, sigPA / NREP, sigL / NREP, star)];
end end
end end
s = [s sprintf('\nRead the per-animal column as the honest power; LME is optimistic (obs-level DF).\n')];
s = [s sprintf(['\nRead the per-animal column as the honest power. At the observed N this study\n' ... localFinish(s, here, suffix);
'is typically underpowered; per-animal power reaches ~0.8 only at larger N.\n'])];
localFinish(s, here);
warning(warnState); warning(warnState);
end
% ---------------------------------------------------------------- helpers % ---------------------------------------------------------------- helpers
function localFinish(s, here) function localFinish(s, here, suffix)
fprintf('%s', s); fprintf('%s', s);
fid = fopen(fullfile(here, 'power_result.txt'), 'w'); fid = fopen(fullfile(here, ['power_result' suffix '.txt']), 'w');
fprintf(fid, '%s', s); fclose(fid); fprintf(fid, '%s', s); fclose(fid);
end end
analysis/matlab/variations/unmerge_d0_13/result.txt
+4 -4
View File
@@ -1,5 +1,5 @@
============================================================================== ==============================================================================
VARIATION: unmerge_d0_13 VARIATION: unmerge_d0_13 [metric: success COUNT]
============================================================================== ==============================================================================
model: behavior ~ stim + day + stim:day + (1|rat) (behavior = success COUNT) model: behavior ~ stim + day + stim:day + (1|rat) (behavior = success COUNT)
day = training day within window (0 = first analyzed day) day = training day within window (0 = first analyzed day)
@@ -60,9 +60,9 @@ effect t(df) / F(df1) p (resid) Satterthwaite: p (df)
stim x day (interaction) t(66)= 1.65 F(1)= 2.721 p=0.1038 p=0.1035 (df=70) stim x day (interaction) t(66)= 1.65 F(1)= 2.721 p=0.1038 p=0.1035 (df=70)
day (learning) t(66)= 9.09 F(1)= 82.593 p=3.035e-13 p=1.823e-13 (df=70) day (learning) t(66)= 9.09 F(1)= 82.593 p=3.035e-13 p=1.823e-13 (df=70)
stim (main, window start) t(66)= 0.95 F(1)= 0.910 p=0.3436 p=0.3434 (df=70) stim (main, window start) t(66)= 0.95 F(1)= 0.910 p=0.3436 p=0.3434 (df=70)
interaction 95% CI: [-0.33, +3.42] interaction 95% CI: [-0.3252, +3.419]
HONEST LME (per-animal random slope, day|rat): interaction F(1,12.8)=1.64, p=0.2227 HONEST LME (per-animal random slope, day|rat): interaction F(1,12.8)=1.64, p=0.2227
(Satterthwaite DF ~= residual on this random-intercept model; the random-slope (Satterthwaite DF ~= residual on this random-intercept model; the random-slope
model above is the honest learning-rate test -- DF collapses toward the animal count.) model above is the honest learning-rate test -- DF collapses toward the animal count.)
INTERPRETATION: stim x day interaction n.s. -- slopes parallel (no differential learning rate) (p=0.1038, slope diff=+1.55) INTERPRETATION: stim x day interaction n.s. -- slopes parallel (no differential learning rate) (p=0.1038, slope diff=+1.547)
Paper (N=24): interaction t(227)=2.68, F(1)=7.12, p=0.008. Paper (N=24, count): interaction t(227)=2.68, F(1)=7.12, p=0.008.
analysis/matlab/variations/unmerge_d0_13/result_rate.txt
+68
View File
@@ -0,0 +1,68 @@
==============================================================================
VARIATION: unmerge_d0_13 [metric: success RATE (success/attempts)]
==============================================================================
model: behavior ~ stim + day + stim:day + (1|rat) (behavior = success RATE (success/attempts))
day = training day within window (0 = first analyzed day)
treatment (stim=1): Electrode-Box-B2
control (stim=0): Electrode-Box-A2
N = 6 rats, 70 sessions raw day coverage: treat 0..13, control 0..13
(equal day coverage over this window)
==============================================================================
FULL MODEL SUMMARY -- fitlme
==============================================================================
Linear mixed-effects model fit by ML
Model information:
Number of observations 70
Fixed effects coefficients 4
Random effects coefficients 6
Covariance parameters 2
Formula:
behavior ~ 1 + day*stim + (1 | rat)
Model fit statistics:
AIC BIC LogLikelihood Deviance
-114.71 -101.22 63.353 -126.71
Fixed effects coefficients (95% CIs):
Name Estimate SE tStat DF pValue
{'(Intercept)'} 0.25633 0.036798 6.9659 66 1.8623e-09
{'day' } 0.032331 0.0047766 6.7686 66 4.1691e-09
{'stim' } 0.03233 0.051391 0.6291 66 0.53146
{'day:stim' } 0.011248 0.0061725 1.8223 66 0.07294
Lower Upper
0.18286 0.3298
0.022794 0.041868
-0.070275 0.13493
-0.0010756 0.023572
Random effects covariance parameters (95% CIs):
Group: rat (6 Levels)
Name1 Name2 Type Estimate
{'(Intercept)'} {'(Intercept)'} {'std'} 0.038979
Lower Upper
0.016492 0.092125
Group: Error
Name Estimate Lower Upper
{'Res Std'} 0.093402 0.07855 0.11106
effect t(df) / F(df1) p (resid) Satterthwaite: p (df)
----------------------------------------------------------------------------
stim x day (interaction) t(66)= 1.82 F(1)= 3.321 p=0.07294 p=0.07278 (df=68)
day (learning) t(66)= 6.77 F(1)= 45.814 p=4.169e-09 p=3.37e-09 (df=70)
stim (main, window start) t(66)= 0.63 F(1)= 0.396 p=0.5315 p=0.5377 (df=17)
interaction 95% CI: [-0.001076, +0.02357]
HONEST LME (per-animal random slope, day|rat): interaction F(1,41.8)=3.20, p=0.08108
(Satterthwaite DF ~= residual on this random-intercept model; the random-slope
model above is the honest learning-rate test -- DF collapses toward the animal count.)
INTERPRETATION: stim x day interaction n.s. -- slopes parallel (no differential learning rate) (p=0.07294, slope diff=+0.01125)
Paper (N=24, count): interaction t(227)=2.68, F(1)=7.12, p=0.008.
analysis/matlab/variations/unmerge_d0_5/analyze.m
+63 -39
View File
@@ -1,30 +1,58 @@
% Variation analysis -- the paper's linear mixed model on the successful-reach % Variation analysis -- the paper's linear mixed model on this folder's data,
% COUNT, fit on this folder's curated data subset. % for BOTH metrics:
% metric = count : behavior = # successes -> result.txt
% metric = rate : behavior = success / attempts -> result_rate.txt
% (rate uses only sessions with attempts > 0)
% %
% model: behavior ~ stim + day + stim:day + (1|rat) % model: behavior ~ stim + day + stim:day + (1|rat)
% behavior = successful reaches (count per session) % stim = 1 treatment / 0 control; day = training day within window (0 =
% stim = 1 for the treatment group(s), 0 for the control group(s) % first analyzed day); rat = subject (random intercept).
% day = training day within this window (0 = first analyzed day) % For the interaction we report residual DF, Satterthwaite DF, and the honest
% rat = subject (random intercept) % per-animal random-slope test. Self-contained: reads data.csv beside this
% % script. Run headless with: matlab -batch "analyze"
% Self-contained: reads data.csv beside this script and writes result.txt. % (Copy of analysis/matlab/variation_analyze.m; see make_variations.m.)
% Run headless from this folder with: matlab -batch "analyze"
% (This is a copy of analysis/matlab/variation_analyze.m; see make_variations.m.)
here = fileparts(mfilename('fullpath')); here = fileparts(mfilename('fullpath'));
if isempty(here); here = pwd; end if isempty(here); here = pwd; end
vname = regexprep(here, '.*[/\\]', ''); % folder name = variation id vname = regexprep(here, '.*[/\\]', '');
D = readtable(fullfile(here, 'data.csv'), 'TextType', 'string'); D = readtable(fullfile(here, 'data.csv'), 'TextType', 'string');
tbl = table(D.success, D.day - min(D.day), double(D.stim), categorical(D.subject), ... Rc = localAnalyze(D, 'count', here, vname);
Rr = localAnalyze(D, 'rate', here, vname);
% Machine-readable handoff for SUMMARY.csv (count drives it; rate appended).
VARRESULT = struct('name', vname, 'nRats', Rc.nRats, 'nObs', Rc.nObs, ...
'interP', Rc.interP, 'interEst', Rc.interEst, ...
'interPsatt', Rc.interPsatt, 'interPrs', Rc.interPrs, ...
'stimP', Rc.stimP, 'dayP', Rc.dayP, 'covEqual', Rc.covEqual, ...
'interPrate', Rr.interP, 'interEstRate', Rr.interEst, 'interPrsRate', Rr.interPrs);
% ------------------------------------------------------------------ helper
function R = localAnalyze(D, metric, here, vname)
if strcmp(metric, 'rate')
D = D(D.total > 0, :);
beh = D.success ./ D.total;
mlabel = 'success RATE (success/attempts)'; suffix = '_rate';
else
beh = D.success;
mlabel = 'success COUNT'; suffix = '';
end
R = struct('interP', NaN, 'interEst', NaN, 'interPsatt', NaN, 'interPrs', NaN, ...
'stimP', NaN, 'dayP', NaN, 'nRats', numel(unique(D.subject)), ...
'nObs', height(D), 'covEqual', false);
if numel(unique(D.stim)) < 2 || numel(unique(D.day)) < 2
localWrite(sprintf('VARIATION: %s [metric: %s]\nInsufficient data for this metric.\n', ...
vname, mlabel), here, suffix);
return
end
tbl = table(beh, D.day - min(D.day), double(D.stim), categorical(D.subject), ...
'VariableNames', {'behavior', 'day', 'stim', 'rat'}); 'VariableNames', {'behavior', 'day', 'stim', 'rat'});
m = fitlme(tbl, 'behavior ~ stim + day + stim:day + (1|rat)'); m = fitlme(tbl, 'behavior ~ stim + day + stim:day + (1|rat)');
C = m.Coefficients; A = anova(m); ci = coefCI(m); C = m.Coefficients; A = anova(m); ci = coefCI(m);
As = anova(m, 'DFMethod', 'satterthwaite'); % Satterthwaite denominator DF As = anova(m, 'DFMethod', 'satterthwaite');
% Honest test: refit with a per-animal random SLOPE so the interaction DF
% collapses toward the animal count (guarded -- may not converge in short windows).
rsP = NaN; rsDf = NaN; rsF = NaN; rsOk = false; rsP = NaN; rsDf = NaN; rsF = NaN; rsOk = false;
wst = warning('off', 'all'); wst = warning('off', 'all');
try try
@@ -43,6 +71,7 @@ row = @(nm, t) sprintf('%-26s t(%d)=%6.2f F(%d)=%7.3f p=%.4g p=%.4g (df=%.0f
C.DF(gi(t)), C.tStat(gi(t)), A.DF1(ga(t)), A.FStat(ga(t)), C.pValue(gi(t)), ... C.DF(gi(t)), C.tStat(gi(t)), A.DF1(ga(t)), A.FStat(ga(t)), C.pValue(gi(t)), ...
As.pValue(gs(t)), As.DF2(gs(t))); As.pValue(gs(t)), As.DF2(gs(t)));
ii = gi('day:stim'); pI = C.pValue(ii); eI = C.Estimate(ii);
maxT = max(D.day(D.stim == 1)); minT = min(D.day(D.stim == 1)); maxT = max(D.day(D.stim == 1)); minT = min(D.day(D.stim == 1));
maxC = max(D.day(D.stim == 0)); minC = min(D.day(D.stim == 0)); maxC = max(D.day(D.stim == 0)); minC = min(D.day(D.stim == 0));
if abs(maxT - maxC) > 2 if abs(maxT - maxC) > 2
@@ -50,20 +79,14 @@ if abs(maxT - maxC) > 2
else else
cov = '(equal day coverage over this window)'; cov = '(equal day coverage over this window)';
end end
if pI >= 0.05; verdict = 'n.s. -- slopes parallel (no differential learning rate)';
ii = gi('day:stim'); pI = C.pValue(ii); eI = C.Estimate(ii); elseif eI > 0; verdict = 'SIGNIFICANT positive -- treatment improves FASTER (benefit accumulates)';
if pI >= 0.05 else; verdict = 'SIGNIFICANT negative -- treatment improves SLOWER (groups converge)'; end
verdict = 'n.s. -- slopes parallel (no differential learning rate)';
elseif eI > 0
verdict = 'SIGNIFICANT positive -- treatment improves FASTER (benefit accumulates)';
else
verdict = 'SIGNIFICANT negative -- treatment improves SLOWER (groups converge)';
end
bar = repmat('=', 1, 78); bar = repmat('=', 1, 78);
raw = regexprep(evalc('disp(m)'), '</?strong>', ''); raw = regexprep(evalc('disp(m)'), '</?strong>', '');
s = sprintf('%s\nVARIATION: %s\n%s\n', bar, vname, bar); s = sprintf('%s\nVARIATION: %s [metric: %s]\n%s\n', bar, vname, mlabel, bar);
s = [s sprintf('model: behavior ~ stim + day + stim:day + (1|rat) (behavior = success COUNT)\n')]; s = [s sprintf('model: behavior ~ stim + day + stim:day + (1|rat) (behavior = %s)\n', mlabel)];
s = [s sprintf('day = training day within window (0 = first analyzed day)\n')]; s = [s sprintf('day = training day within window (0 = first analyzed day)\n')];
s = [s sprintf('treatment (stim=1): %s\n', strjoin(cellstr(unique(D.group(D.stim == 1))), ', '))]; s = [s sprintf('treatment (stim=1): %s\n', strjoin(cellstr(unique(D.group(D.stim == 1))), ', '))];
s = [s sprintf('control (stim=0): %s\n', strjoin(cellstr(unique(D.group(D.stim == 0))), ', '))]; s = [s sprintf('control (stim=0): %s\n', strjoin(cellstr(unique(D.group(D.stim == 0))), ', '))];
@@ -74,7 +97,7 @@ s = [s sprintf('%-26s %-18s %-12s %s\n%s\n', 'effect', 't(df) / F(df1)', 'p (res
s = [s row('stim x day (interaction)', 'day:stim')]; s = [s row('stim x day (interaction)', 'day:stim')];
s = [s row('day (learning)', 'day')]; s = [s row('day (learning)', 'day')];
s = [s row('stim (main, window start)', 'stim')]; s = [s row('stim (main, window start)', 'stim')];
s = [s sprintf('interaction 95%% CI: [%+.2f, %+.2f]\n', ci(ii, 1), ci(ii, 2))]; s = [s sprintf('interaction 95%% CI: [%+.4g, %+.4g]\n', ci(ii, 1), ci(ii, 2))];
if rsOk if rsOk
s = [s sprintf('HONEST LME (per-animal random slope, day|rat): interaction F(1,%.1f)=%.2f, p=%.4g\n', rsDf, rsF, rsP)]; s = [s sprintf('HONEST LME (per-animal random slope, day|rat): interaction F(1,%.1f)=%.2f, p=%.4g\n', rsDf, rsF, rsP)];
else else
@@ -82,17 +105,18 @@ else
end end
s = [s sprintf([' (Satterthwaite DF ~= residual on this random-intercept model; the random-slope\n' ... s = [s sprintf([' (Satterthwaite DF ~= residual on this random-intercept model; the random-slope\n' ...
' model above is the honest learning-rate test -- DF collapses toward the animal count.)\n'])]; ' model above is the honest learning-rate test -- DF collapses toward the animal count.)\n'])];
s = [s sprintf('INTERPRETATION: stim x day interaction %s (p=%.4g, slope diff=%+.2f)\n', verdict, pI, eI)]; s = [s sprintf('INTERPRETATION: stim x day interaction %s (p=%.4g, slope diff=%+.4g)\n', verdict, pI, eI)];
s = [s sprintf('Paper (N=24): interaction t(227)=2.68, F(1)=7.12, p=0.008.\n')]; s = [s sprintf('Paper (N=24, count): interaction t(227)=2.68, F(1)=7.12, p=0.008.\n')];
localWrite(s, here, suffix);
R = struct('interP', pI, 'interEst', eI, 'interPsatt', As.pValue(gs('day:stim')), ...
'interPrs', rsP, 'stimP', C.pValue(gi('stim')), 'dayP', C.pValue(gi('day')), ...
'nRats', numel(unique(D.subject)), 'nObs', height(D), 'covEqual', abs(maxT - maxC) <= 2);
end
function localWrite(s, here, suffix)
fprintf('%s', s); fprintf('%s', s);
fid = fopen(fullfile(here, 'result.txt'), 'w'); fid = fopen(fullfile(here, ['result' suffix '.txt']), 'w');
fprintf(fid, '%s', s); fprintf(fid, '%s', s); fclose(fid);
fclose(fid); end
% Machine-readable handoff for the summary table (see make_variations.m).
VARRESULT = struct('name', vname, 'nRats', numel(unique(D.subject)), ...
'nObs', height(D), 'interP', pI, 'interEst', eI, ...
'interPsatt', As.pValue(gs('day:stim')), 'interPrs', rsP, ...
'stimP', C.pValue(gi('stim')), 'dayP', C.pValue(gi('day')), ...
'covEqual', abs(maxT - maxC) <= 2);
analysis/matlab/variations/unmerge_d0_5/learning_curve.png
BIN
View File
Binary file not shown.
Side by Side Swipe Overlay

Before

Width:  |  Height:  |  Size: 36 KiB

After

Width:  |  Height:  |  Size: 36 KiB

analysis/matlab/variations/unmerge_d0_5/learning_curve_rate.png
BIN
View File
Binary file not shown.
Side by Side

After

Width:  |  Height:  |  Size: 37 KiB

analysis/matlab/variations/unmerge_d0_5/logpower_result.txt
+5 -6
View File
@@ -1,7 +1,7 @@
============================================================================== ==============================================================================
LOG-DAY MODEL + COHEN'S f + POWER -- unmerge_d0_5 LOG-DAY MODEL + COHEN'S f + POWER -- unmerge_d0_5 [metric: # successes (count)]
============================================================================== ==============================================================================
model: behavior ~ stim + log(day) + stim:log(day) + (1|rat) (success COUNT) model: behavior ~ stim + log(day) + stim:log(day) + (1|rat) (behavior = # successes (count))
log(day) uses 1-indexed training day (our day 0 = paper "Day 1") log(day) uses 1-indexed training day (our day 0 = paper "Day 1")
observed groups: stim n=3, control n=3 nrep=120, alpha=0.05 observed groups: stim n=3, control n=3 nrep=120, alpha=0.05
@@ -10,7 +10,7 @@ stim x log(day) interaction: F(1,32)=4.849 p(resid)=0.03498 p(Satt)=0.03414 (d
honest per-animal random slope (log-day): F(1,7.2)=3.65 p=0.09636 honest per-animal random slope (log-day): F(1,7.2)=3.65 p=0.09636
Cohen's f (interaction, partial eta^2=0.039) = 0.200 (small-medium; f: .10 small, .25 medium, .40 large) Cohen's f (interaction, partial eta^2=0.039) = 0.200 (small-medium; f: .10 small, .25 medium, .40 large)
--- power simulation (log-day ground truth: stim:logday=+16.48, ratSD=0.00, resSD=13.58) --- --- power simulation (log-day ground truth: stim:logday=+16.48, ratSD=0, resSD=13.58) ---
true stim:log(day) = +16.48 (100% of observed) true stim:log(day) = +16.48 (100% of observed)
N/group | per-animal power | LME power N/group | per-animal power | LME power
@@ -22,7 +22,7 @@ Cohen's f (interaction, partial eta^2=0.039) = 0.200 (small-medium; f: .10 smal
16 | 1.00 | 1.00 16 | 1.00 | 1.00
24 | 1.00 | 1.00 24 | 1.00 | 1.00
true stim:log(day) = +8.24 (50% of observed) true stim:log(day) = +8.239 (50% of observed)
N/group | per-animal power | LME power N/group | per-animal power | LME power
------------------------------------------ ------------------------------------------
3 | 0.09 | 0.24 <- observed 3 | 0.09 | 0.24 <- observed
@@ -32,5 +32,4 @@ Cohen's f (interaction, partial eta^2=0.039) = 0.200 (small-medium; f: .10 smal
16 | 0.69 | 0.69 16 | 0.69 | 0.69
24 | 0.90 | 0.92 24 | 0.90 | 0.92
Read the per-animal column as the honest power; the LME column matches the Read per-animal as the honest power; LME matches the paper's power code (optimistic).
paper's power code (anova interaction p, observation-level DF) and is optimistic.
analysis/matlab/variations/unmerge_d0_5/logpower_result_rate.txt
+35
View File
@@ -0,0 +1,35 @@
==============================================================================
LOG-DAY MODEL + COHEN'S f + POWER -- unmerge_d0_5 [metric: success RATE]
==============================================================================
model: behavior ~ stim + log(day) + stim:log(day) + (1|rat) (behavior = success RATE)
log(day) uses 1-indexed training day (our day 0 = paper "Day 1")
observed groups: stim n=3, control n=3 nrep=120, alpha=0.05
--- fitted on real data ---
stim x log(day) interaction: F(1,32)=10.951 p(resid)=0.002322 p(Satt)=0.00244 (df=30)
honest per-animal random slope (log-day): F(1,8.5)=8.07 p=0.02043
Cohen's f (interaction, partial eta^2=0.109) = 0.350 (medium; f: .10 small, .25 medium, .40 large)
--- power simulation (log-day ground truth: stim:logday=+0.1553, ratSD=0.04445, resSD=0.08516) ---
true stim:log(day) = +0.1553 (100% of observed)
N/group | per-animal power | LME power
------------------------------------------
3 | 0.54 | 0.90 <- observed
5 | 0.93 | 0.97
8 | 1.00 | 1.00
12 | 1.00 | 1.00
16 | 1.00 | 1.00
24 | 1.00 | 1.00
true stim:log(day) = +0.07765 (50% of observed)
N/group | per-animal power | LME power
------------------------------------------
3 | 0.13 | 0.38 <- observed
5 | 0.44 | 0.54
8 | 0.72 | 0.73
12 | 0.81 | 0.88
16 | 0.94 | 0.97
24 | 1.00 | 1.00
Read per-animal as the honest power; LME matches the paper's power code (optimistic).
analysis/matlab/variations/unmerge_d0_5/logpowersim.m
+38 -39
View File
@@ -1,50 +1,52 @@
% Variation log-day analysis + Cohen's f + power simulation. % Variation log-day analysis + Cohen's f + power simulation, for BOTH metrics:
% metric = count : behavior = # successes -> logpower_result.txt
% metric = rate : behavior = success / attempts -> logpower_result_rate.txt
% %
% The paper's power code models behavior against LOG training day, not raw day: % The paper's power code models behavior against LOG training day:
% behavior ~ stim + log(day) + stim:log(day) + (1|rat). % behavior ~ stim + log(day) + stim:log(day) + (1|rat).
% Their day is 1-indexed (1..10); our data.csv day is 0-indexed (day 0 = paper % Their day is 1-indexed; our data.csv day is 0-indexed, so log(day + 1)
% "Day 1"), so log(day + 1) reproduces their transform exactly. % reproduces their transform (our day 0 = paper "Day 1"). For each metric this
% % refits that model, reports the interaction (residual / Satterthwaite / honest
% This script (a) refits that log-day model on data.csv, (b) reports the % per-animal random-slope DF) and Cohen's f (partial-eta^2 effect size), then
% interaction (residual DF, Satterthwaite DF, and the honest per-animal % runs the Monte-Carlo power sim (per-animal cluster-honest + LME power).
% random-slope test) and Cohen's f -- the partial-eta^2 effect size of the % Run: matlab -batch "logpowersim"
% interaction, var(fitted_full) - var(fitted_no_interaction) over var(behavior)
% -- and (c) runs the Monte-Carlo power simulation on the log-day model,
% scoring per-animal (cluster-honest) and LME power across N.
% Writes logpower_result.txt. Run: matlab -batch "logpowersim"
% (Copy of analysis/matlab/variation_logpower.m; see make_variation_logpower.m.) % (Copy of analysis/matlab/variation_logpower.m; see make_variation_logpower.m.)
here = fileparts(mfilename('fullpath')); here = fileparts(mfilename('fullpath'));
if isempty(here); here = pwd; end if isempty(here); here = pwd; end
vname = regexprep(here, '.*[/\\]', ''); vname = regexprep(here, '.*[/\\]', '');
D = readtable(fullfile(here, 'data.csv'), 'TextType', 'string'); D = readtable(fullfile(here, 'data.csv'), 'TextType', 'string');
FORMULA = 'behavior ~ stim + day + stim:day + (1|rat)'; % 'day' column = log(day+1)
NS = [3 5 8 12 16 24];
EFFMULS = [1 0.5];
NREP = 120;
warnState = warning('off', 'all'); localLogPower(D, 'count', here, vname);
rng(1); localLogPower(D, 'rate', here, vname);
logday = log(D.day + 1); % 0-indexed day -> their log(1-indexed day) % ---------------------------------------------------------------- per metric
tbl0 = table(D.success, logday, double(D.stim), categorical(D.subject), ... function localLogPower(D, metric, here, vname)
NS = [3 5 8 12 16 24]; EFFMULS = [1 0.5]; NREP = 120;
FORMULA = 'behavior ~ stim + day + stim:day + (1|rat)'; % 'day' = log(day+1)
if strcmp(metric, 'rate')
D = D(D.total > 0, :); beh = D.success ./ D.total; mlabel = 'success RATE'; suffix = '_rate';
else
beh = D.success; mlabel = '# successes (count)'; suffix = '';
end
warnState = warning('off', 'all'); rng(1);
logday = log(D.day + 1);
tbl0 = table(beh, logday, double(D.stim), categorical(D.subject), ...
'VariableNames', {'behavior', 'day', 'stim', 'rat'}); 'VariableNames', {'behavior', 'day', 'stim', 'rat'});
nStim = numel(unique(D.subject(D.stim == 1))); nStim = numel(unique(D.subject(D.stim == 1)));
nCtrl = numel(unique(D.subject(D.stim == 0))); nCtrl = numel(unique(D.subject(D.stim == 0)));
bar = repmat('=', 1, 78); bar = repmat('=', 1, 78);
s = sprintf('%s\nLOG-DAY MODEL + COHEN''S f + POWER -- %s\n%s\n', bar, vname, bar); s = sprintf('%s\nLOG-DAY MODEL + COHEN''S f + POWER -- %s [metric: %s]\n%s\n', bar, vname, mlabel, bar);
s = [s sprintf('model: behavior ~ stim + log(day) + stim:log(day) + (1|rat) (success COUNT)\n')]; s = [s sprintf('model: behavior ~ stim + log(day) + stim:log(day) + (1|rat) (behavior = %s)\n', mlabel)];
s = [s sprintf('log(day) uses 1-indexed training day (our day 0 = paper "Day 1")\n')]; s = [s sprintf('log(day) uses 1-indexed training day (our day 0 = paper "Day 1")\n')];
s = [s sprintf('observed groups: stim n=%d, control n=%d nrep=%d, alpha=0.05\n', nStim, nCtrl, NREP)]; s = [s sprintf('observed groups: stim n=%d, control n=%d nrep=%d, alpha=0.05\n', nStim, nCtrl, NREP)];
if nStim < 2 || nCtrl < 2 || numel(unique(tbl0.day)) < 2 if nStim < 2 || nCtrl < 2 || numel(unique(tbl0.day)) < 2
s = [s sprintf('\nInsufficient data for this analysis (need >=2 animals/group and >=2 days).\n')]; s = [s sprintf('\nInsufficient data for this analysis.\n')];
localFinish(s, here); warning(warnState); return localFinish(s, here, suffix); warning(warnState); return
end end
% ---- fitted model on the real data ----
full = fitlme(tbl0, FORMULA); full = fitlme(tbl0, FORMULA);
An = anova(full); Asatt = anova(full, 'DFMethod', 'satterthwaite'); An = anova(full); Asatt = anova(full, 'DFMethod', 'satterthwaite');
ii = strcmp(An.Term, 'day:stim'); is = strcmp(Asatt.Term, 'day:stim'); ii = strcmp(An.Term, 'day:stim'); is = strcmp(Asatt.Term, 'day:stim');
@@ -52,7 +54,6 @@ reduced = fitlme(tbl0, 'behavior ~ stim + day + (1|rat)');
eta2part = max((var(fitted(full)) - var(fitted(reduced))) / var(tbl0.behavior), 0); eta2part = max((var(fitted(full)) - var(fitted(reduced))) / var(tbl0.behavior), 0);
cohenf = sqrt(eta2part / (1 - eta2part)); cohenf = sqrt(eta2part / (1 - eta2part));
% honest per-animal random-slope interaction
rsP = NaN; rsDf = NaN; rsF = NaN; rsOk = false; rsP = NaN; rsDf = NaN; rsF = NaN; rsOk = false;
try try
mr = fitlme(tbl0, 'behavior ~ stim + day + stim:day + (day|rat)'); mr = fitlme(tbl0, 'behavior ~ stim + day + stim:day + (day|rat)');
@@ -76,19 +77,18 @@ end
s = [s sprintf('Cohen''s f (interaction, partial eta^2=%.3f) = %.3f (%s; f: .10 small, .25 medium, .40 large)\n', ... s = [s sprintf('Cohen''s f (interaction, partial eta^2=%.3f) = %.3f (%s; f: .10 small, .25 medium, .40 large)\n', ...
eta2part, cohenf, mag)]; eta2part, cohenf, mag)];
% ---- power simulation on the log-day ground truth ----
cn = full.CoefficientNames; be = full.fixedEffects; cn = full.CoefficientNames; be = full.fixedEffects;
b0 = be(strcmp(cn, '(Intercept)')); bStim = be(strcmp(cn, 'stim')); b0 = be(strcmp(cn, '(Intercept)')); bStim = be(strcmp(cn, 'stim'));
bDay = be(strcmp(cn, 'day')); bInt = be(strcmp(cn, 'day:stim')); bDay = be(strcmp(cn, 'day')); bInt = be(strcmp(cn, 'day:stim'));
psi = covarianceParameters(full); sRat = sqrt(psi{1}); sRes = sqrt(full.MSE); psi = covarianceParameters(full); sRat = sqrt(psi{1}); sRes = sqrt(full.MSE);
days = unique(tbl0.day); % the log(day) grid days = unique(tbl0.day);
s = [s sprintf('\n--- power simulation (log-day ground truth: stim:logday=%+.2f, ratSD=%.2f, resSD=%.2f) ---\n', ... s = [s sprintf('\n--- power simulation (log-day ground truth: stim:logday=%+.4g, ratSD=%.4g, resSD=%.4g) ---\n', ...
bInt, sRat, sRes)]; bInt, sRat, sRes)];
for eMul = EFFMULS for eMul = EFFMULS
bI = bInt * eMul; bI = bInt * eMul;
s = [s sprintf('\n true stim:log(day) = %+.2f (%.0f%% of observed)\n', bI, eMul * 100)]; %#ok<AGROW> s = [s sprintf('\n true stim:log(day) = %+.4g (%.0f%% of observed)\n', bI, eMul * 100)];
s = [s sprintf(' %-8s | per-animal power | LME power\n %s\n', 'N/group', repmat('-', 1, 42))]; %#ok<AGROW> s = [s sprintf(' %-8s | per-animal power | LME power\n %s\n', 'N/group', repmat('-', 1, 42))];
for N = NS for N = NS
sigPA = 0; sigL = 0; sigPA = 0; sigL = 0;
for r = 1:NREP for r = 1:NREP
@@ -102,19 +102,18 @@ for eMul = EFFMULS
end end
star = ''; star = '';
if N == nStim || N == nCtrl; star = ' <- observed'; end if N == nStim || N == nCtrl; star = ' <- observed'; end
s = [s sprintf(' %-8d | %5.2f | %5.2f%s\n', N, sigPA / NREP, sigL / NREP, star)]; %#ok<AGROW> s = [s sprintf(' %-8d | %5.2f | %5.2f%s\n', N, sigPA / NREP, sigL / NREP, star)];
end end
end end
s = [s sprintf(['\nRead the per-animal column as the honest power; the LME column matches the\n' ... s = [s sprintf('\nRead per-animal as the honest power; LME matches the paper''s power code (optimistic).\n')];
'paper''s power code (anova interaction p, observation-level DF) and is optimistic.\n'])]; localFinish(s, here, suffix);
localFinish(s, here);
warning(warnState); warning(warnState);
end
% ---------------------------------------------------------------- helpers % ---------------------------------------------------------------- helpers
function localFinish(s, here) function localFinish(s, here, suffix)
fprintf('%s', s); fprintf('%s', s);
fid = fopen(fullfile(here, 'logpower_result.txt'), 'w'); fid = fopen(fullfile(here, ['logpower_result' suffix '.txt']), 'w');
fprintf(fid, '%s', s); fclose(fid); fprintf(fid, '%s', s); fclose(fid);
end end
analysis/matlab/variations/unmerge_d0_5/plotcurve.m
+29 -19
View File
@@ -1,11 +1,12 @@
% Variation learning-curve plot, in the style of the paper: % Variation learning-curve plots, in the style of the paper:
% "Lines indicate mean (and SEM) across animals in the anodal (red) and % "Lines indicate mean (and SEM) across animals in the anodal (red) and
% control (blue) groups." % control (blue) groups."
% Plots mean +/- SEM successful reaches per training day for the treatment / % Produces TWO figures from this folder's data.csv:
% anodal group (stim = 1, red) and the control group (stim = 0, blue), reading % learning_curve.png # successes (count) per training day
% this folder's data.csv and saving learning_curve.png. The per-group N is read % learning_curve_rate.png success rate (success/attempts) per training day
% from the data (each variation pools different groups), so the legend shows the % anodal / treatment = stim 1 (red); control = stim 0 (blue). Per-group N is
% actual counts. Training day is 1-indexed (our day 0 = the paper's "Day 1"). % read from the data. Training day is 1-indexed (our day 0 = paper "Day 1") and
% the x-axis tick labels are drawn vertically.
% Run: matlab -batch "plotcurve" % Run: matlab -batch "plotcurve"
% (Copy of analysis/matlab/variation_plot.m; see make_variation_plot.m.) % (Copy of analysis/matlab/variation_plot.m; see make_variation_plot.m.)
@@ -14,15 +15,20 @@ if isempty(here); here = pwd; end
vname = regexprep(here, '.*[/\\]', ''); vname = regexprep(here, '.*[/\\]', '');
D = readtable(fullfile(here, 'data.csv'), 'TextType', 'string'); D = readtable(fullfile(here, 'data.csv'), 'TextType', 'string');
days = unique(D.day); % 0-indexed days = unique(D.day);
xd = days + 1; % plot as 1-indexed training day (paper axis) xd = days + 1; % plot as 1-indexed training day (paper axis)
red = [0.85 0.10 0.10]; red = [0.85 0.10 0.10];
blue = [0.10 0.30 0.85]; blue = [0.10 0.30 0.85];
[Ma, Sa, na] = localCurve(D, 1, days); % anodal / treatment (stim = 1) localPlot(D, days, xd, 'count', '# successes', ...
[Mc, Sc, nc] = localCurve(D, 0, days); % control (stim = 0) fullfile(here, 'learning_curve.png'), vname, red, blue);
localPlot(D, days, xd, 'rate', 'success rate', ...
fullfile(here, 'learning_curve_rate.png'), vname, red, blue);
% ------------------------------------------------------------------ helpers
function localPlot(D, days, xd, metric, ylab, outFile, vname, red, blue)
[Ma, Sa, na] = localCurve(D, 1, days, metric); % anodal / treatment
[Mc, Sc, nc] = localCurve(D, 0, days, metric); % control
fig = figure('Visible', 'off', 'Color', 'w', 'Position', [100 100 560 460]); fig = figure('Visible', 'off', 'Color', 'w', 'Position', [100 100 560 460]);
hold on hold on
e1 = errorbar(xd, Ma, Sa, '-o', 'Color', red, 'MarkerFaceColor', red, 'LineWidth', 2); e1 = errorbar(xd, Ma, Sa, '-o', 'Color', red, 'MarkerFaceColor', red, 'LineWidth', 2);
@@ -30,26 +36,30 @@ e2 = errorbar(xd, Mc, Sc, '-o', 'Color', blue, 'MarkerFaceColor', blue, 'LineWid
hold off hold off
legend([e1 e2], {sprintf('anodal, N = %d', na), sprintf('control, N = %d', nc)}, ... legend([e1 e2], {sprintf('anodal, N = %d', na), sprintf('control, N = %d', nc)}, ...
'Location', 'northwest', 'Box', 'off'); 'Location', 'northwest', 'Box', 'off');
xlabel('training day'); xlabel('training day'); ylabel(ylab);
ylabel('# successes');
title(vname, 'Interpreter', 'none'); title(vname, 'Interpreter', 'none');
set(gca, 'XTick', xd, 'FontName', 'Arial', 'FontSize', 13, 'LineWidth', 1.5, 'Box', 'off'); set(gca, 'XTick', xd, 'FontName', 'Arial', 'FontSize', 13, 'LineWidth', 1.5, 'Box', 'off');
xtickangle(90); % vertical x-axis tick labels
outFile = fullfile(here, 'learning_curve.png');
exportgraphics(fig, outFile, 'Resolution', 150); exportgraphics(fig, outFile, 'Resolution', 150);
close(fig); close(fig);
fprintf('%s: wrote learning_curve.png (anodal N=%d, control N=%d)\n', vname, na, nc); fprintf('%s: wrote %s (anodal N=%d, control N=%d)\n', vname, outFile, na, nc);
end
% ------------------------------------------------------------------ helper function [M, S, n] = localCurve(D, stimVal, days, metric)
function [M, S, n] = localCurve(D, stimVal, days) %LOCALCURVE Per-day mean and SEM across the animals in a group, for a metric.
%LOCALCURVE Per-day mean and SEM of successes across the animals in a group.
subs = unique(D.subject(D.stim == stimVal)); subs = unique(D.subject(D.stim == stimVal));
n = numel(subs); n = numel(subs);
X = nan(numel(days), n); X = nan(numel(days), n);
for j = 1:n for j = 1:n
for i = 1:numel(days) for i = 1:numel(days)
r = D.subject == subs(j) & D.day == days(i); r = D.subject == subs(j) & D.day == days(i);
if any(r); X(i, j) = mean(D.success(r)); end if ~any(r); continue; end
if strcmp(metric, 'rate')
tot = sum(D.total(r));
if tot > 0; X(i, j) = sum(D.success(r)) / tot; end
else
X(i, j) = mean(D.success(r));
end
end end
end end
M = mean(X, 2, 'omitnan'); M = mean(X, 2, 'omitnan');
analysis/matlab/variations/unmerge_d0_5/power_result.txt
+6 -7
View File
@@ -1,11 +1,11 @@
============================================================================== ==============================================================================
POWER SIMULATION -- unmerge_d0_5 POWER SIMULATION -- unmerge_d0_5 [metric: # successes (count)]
============================================================================== ==============================================================================
model: behavior ~ stim + day + stim:day + (1|rat) (success COUNT; day within-window) model: behavior ~ stim + day + stim:day + (1|rat) (behavior = # successes (count); day within-window)
observed groups: stim n=3, control n=3 nrep=120, alpha=0.05 observed groups: stim n=3, control n=3 nrep=120, alpha=0.05
ground truth: stim:day=+5.36/day, rat SD=4.85, residual SD=10.70, days=6 ground truth: stim:day=+5.362/day, rat SD=4.853, residual SD=10.7, days=6
true stim:day interaction = +5.36 (100% of observed) true stim:day interaction = +5.362 (100% of observed)
N/group | per-animal power | LME power N/group | per-animal power | LME power
------------------------------------------ ------------------------------------------
3 | 0.35 | 0.73 <- observed 3 | 0.35 | 0.73 <- observed
@@ -15,7 +15,7 @@ ground truth: stim:day=+5.36/day, rat SD=4.85, residual SD=10.70, days=6
16 | 1.00 | 1.00 16 | 1.00 | 1.00
24 | 1.00 | 1.00 24 | 1.00 | 1.00
true stim:day interaction = +2.68 (50% of observed) true stim:day interaction = +2.681 (50% of observed)
N/group | per-animal power | LME power N/group | per-animal power | LME power
------------------------------------------ ------------------------------------------
3 | 0.11 | 0.24 <- observed 3 | 0.11 | 0.24 <- observed
@@ -25,5 +25,4 @@ ground truth: stim:day=+5.36/day, rat SD=4.85, residual SD=10.70, days=6
16 | 0.79 | 0.81 16 | 0.79 | 0.81
24 | 0.95 | 0.94 24 | 0.95 | 0.94
Read the per-animal column as the honest power. At the observed N this study Read the per-animal column as the honest power; LME is optimistic (obs-level DF).
is typically underpowered; per-animal power reaches ~0.8 only at larger N.
analysis/matlab/variations/unmerge_d0_5/power_result_rate.txt
+28
View File
@@ -0,0 +1,28 @@
==============================================================================
POWER SIMULATION -- unmerge_d0_5 [metric: success RATE]
==============================================================================
model: behavior ~ stim + day + stim:day + (1|rat) (behavior = success RATE; day within-window)
observed groups: stim n=3, control n=3 nrep=120, alpha=0.05
ground truth: stim:day=+0.04902/day, rat SD=0.04642, residual SD=0.07858, days=6
true stim:day interaction = +0.04902 (100% of observed)
N/group | per-animal power | LME power
------------------------------------------
3 | 0.52 | 0.89 <- observed
5 | 0.92 | 0.97
8 | 1.00 | 1.00
12 | 1.00 | 1.00
16 | 1.00 | 1.00
24 | 1.00 | 1.00
true stim:day interaction = +0.02451 (50% of observed)
N/group | per-animal power | LME power
------------------------------------------
3 | 0.12 | 0.34 <- observed
5 | 0.40 | 0.48
8 | 0.62 | 0.72
12 | 0.82 | 0.87
16 | 0.94 | 0.95
24 | 0.99 | 0.99
Read the per-animal column as the honest power; LME is optimistic (obs-level DF).
analysis/matlab/variations/unmerge_d0_5/powersim.m
+35 -34
View File
@@ -1,45 +1,49 @@
% Variation power simulation -- Monte-Carlo power for the paper's stim x day % Variation power simulation -- Monte-Carlo power for the paper's stim x day
% interaction, using THIS folder's data as the ground truth. % interaction, using THIS folder's data as the ground truth, for BOTH metrics:
% metric = count : behavior = # successes -> power_result.txt
% metric = rate : behavior = success / attempts -> power_result_rate.txt
% %
% Ground truth: fitlme(behavior ~ stim + day + stim:day + (1|rat)) on data.csv % Ground truth: fitlme(behavior ~ stim + day + stim:day + (1|rat)) on data.csv
% (success COUNT; day within-window). Its fixed effects, per-rat intercept SD, % (day within-window). Its fixed effects, per-rat intercept SD, and residual SD
% and residual SD generate NREP synthetic datasets at each rats-per-group N and % generate NREP synthetic datasets at each rats-per-group N and each true-effect
% each true-effect multiplier (1 = observed slope, 0.5 = half). Each dataset is % multiplier (1 = observed, 0.5 = half). Each is scored at alpha=0.05 by:
% scored at alpha = 0.05 two ways: % per-animal : Welch t on per-rat behavior~day slopes (cluster-honest power)
% per-animal : Welch t on per-rat behavior~day slopes (cluster-honest -- the
% honest power, matching the random-slope / per-animal inference)
% LME : the fitlme stim:day p (observation-level DF -- optimistic) % LME : the fitlme stim:day p (observation-level DF -- optimistic)
% Writes power_result.txt beside this script. Run: matlab -batch "powersim" % Writes power_result[_rate].txt. Run: matlab -batch "powersim"
% (Copy of analysis/matlab/variation_power.m; see make_variation_power.m.) % (Copy of analysis/matlab/variation_power.m; see make_variation_power.m.)
here = fileparts(mfilename('fullpath')); here = fileparts(mfilename('fullpath'));
if isempty(here); here = pwd; end if isempty(here); here = pwd; end
vname = regexprep(here, '.*[/\\]', ''); vname = regexprep(here, '.*[/\\]', '');
D = readtable(fullfile(here, 'data.csv'), 'TextType', 'string'); D = readtable(fullfile(here, 'data.csv'), 'TextType', 'string');
localPower(D, 'count', here, vname);
localPower(D, 'rate', here, vname);
% ---------------------------------------------------------------- per metric
function localPower(D, metric, here, vname)
NS = [3 5 8 12 16 24]; EFFMULS = [1 0.5]; NREP = 120;
FORMULA = 'behavior ~ stim + day + stim:day + (1|rat)'; FORMULA = 'behavior ~ stim + day + stim:day + (1|rat)';
NS = [3 5 8 12 16 24]; if strcmp(metric, 'rate')
EFFMULS = [1 0.5]; D = D(D.total > 0, :); beh = D.success ./ D.total; mlabel = 'success RATE'; suffix = '_rate';
NREP = 120; else
beh = D.success; mlabel = '# successes (count)'; suffix = '';
warnState = warning('off', 'all'); end
rng(1); warnState = warning('off', 'all'); rng(1);
day0 = min(D.day); day0 = min(D.day);
tbl0 = table(D.success, D.day - day0, double(D.stim), categorical(D.subject), ... tbl0 = table(beh, D.day - day0, double(D.stim), categorical(D.subject), ...
'VariableNames', {'behavior', 'day', 'stim', 'rat'}); 'VariableNames', {'behavior', 'day', 'stim', 'rat'});
nStim = numel(unique(D.subject(D.stim == 1))); nStim = numel(unique(D.subject(D.stim == 1)));
nCtrl = numel(unique(D.subject(D.stim == 0))); nCtrl = numel(unique(D.subject(D.stim == 0)));
bar = repmat('=', 1, 78); bar = repmat('=', 1, 78);
s = sprintf('%s\nPOWER SIMULATION -- %s\n%s\n', bar, vname, bar); s = sprintf('%s\nPOWER SIMULATION -- %s [metric: %s]\n%s\n', bar, vname, mlabel, bar);
s = [s sprintf('model: %s (success COUNT; day within-window)\n', FORMULA)]; s = [s sprintf('model: %s (behavior = %s; day within-window)\n', FORMULA, mlabel)];
s = [s sprintf('observed groups: stim n=%d, control n=%d nrep=%d, alpha=0.05\n', nStim, nCtrl, NREP)]; s = [s sprintf('observed groups: stim n=%d, control n=%d nrep=%d, alpha=0.05\n', nStim, nCtrl, NREP)];
if nStim < 2 || nCtrl < 2 || numel(unique(tbl0.day)) < 2 if nStim < 2 || nCtrl < 2 || numel(unique(tbl0.day)) < 2
s = [s sprintf('\nInsufficient data for a power simulation (need >=2 animals/group and >=2 days).\n')]; s = [s sprintf('\nInsufficient data for a power simulation.\n')];
localFinish(s, here); warning(warnState); return localFinish(s, here, suffix); warning(warnState); return
end end
lme = fitlme(tbl0, FORMULA); lme = fitlme(tbl0, FORMULA);
@@ -49,14 +53,13 @@ bDay = be(strcmp(cn, 'day')); bInt = be(strcmp(cn, 'day:stim'));
psi = covarianceParameters(lme); sRat = sqrt(psi{1}); sRes = sqrt(lme.MSE); psi = covarianceParameters(lme); sRat = sqrt(psi{1}); sRes = sqrt(lme.MSE);
days = (0:max(tbl0.day))'; days = (0:max(tbl0.day))';
s = [s sprintf('ground truth: stim:day=%+.2f/day, rat SD=%.2f, residual SD=%.2f, days=%d\n', ... s = [s sprintf('ground truth: stim:day=%+.4g/day, rat SD=%.4g, residual SD=%.4g, days=%d\n', ...
bInt, sRat, sRes, numel(days))]; bInt, sRat, sRes, numel(days))];
for eMul = EFFMULS for eMul = EFFMULS
bI = bInt * eMul; bI = bInt * eMul;
s = [s sprintf('\n true stim:day interaction = %+.2f (%.0f%% of observed)\n', bI, eMul * 100)]; %#ok<AGROW> s = [s sprintf('\n true stim:day interaction = %+.4g (%.0f%% of observed)\n', bI, eMul * 100)];
s = [s sprintf(' %-8s | per-animal power | LME power\n', 'N/group')]; %#ok<AGROW> s = [s sprintf(' %-8s | per-animal power | LME power\n %s\n', 'N/group', repmat('-', 1, 42))];
s = [s sprintf(' %s\n', repmat('-', 1, 42))]; %#ok<AGROW>
for N = NS for N = NS
sigPA = 0; sigL = 0; sigPA = 0; sigL = 0;
for r = 1:NREP for r = 1:NREP
@@ -70,20 +73,18 @@ for eMul = EFFMULS
end end
star = ''; star = '';
if N == nStim || N == nCtrl; star = ' <- observed'; end if N == nStim || N == nCtrl; star = ' <- observed'; end
s = [s sprintf(' %-8d | %5.2f | %5.2f%s\n', N, sigPA / NREP, sigL / NREP, star)]; %#ok<AGROW> s = [s sprintf(' %-8d | %5.2f | %5.2f%s\n', N, sigPA / NREP, sigL / NREP, star)];
end end
end end
s = [s sprintf('\nRead the per-animal column as the honest power; LME is optimistic (obs-level DF).\n')];
s = [s sprintf(['\nRead the per-animal column as the honest power. At the observed N this study\n' ... localFinish(s, here, suffix);
'is typically underpowered; per-animal power reaches ~0.8 only at larger N.\n'])];
localFinish(s, here);
warning(warnState); warning(warnState);
end
% ---------------------------------------------------------------- helpers % ---------------------------------------------------------------- helpers
function localFinish(s, here) function localFinish(s, here, suffix)
fprintf('%s', s); fprintf('%s', s);
fid = fopen(fullfile(here, 'power_result.txt'), 'w'); fid = fopen(fullfile(here, ['power_result' suffix '.txt']), 'w');
fprintf(fid, '%s', s); fclose(fid); fprintf(fid, '%s', s); fclose(fid);
end end
analysis/matlab/variations/unmerge_d0_5/result.txt
+4 -4
View File
@@ -1,5 +1,5 @@
============================================================================== ==============================================================================
VARIATION: unmerge_d0_5 VARIATION: unmerge_d0_5 [metric: success COUNT]
============================================================================== ==============================================================================
model: behavior ~ stim + day + stim:day + (1|rat) (behavior = success COUNT) model: behavior ~ stim + day + stim:day + (1|rat) (behavior = success COUNT)
day = training day within window (0 = first analyzed day) day = training day within window (0 = first analyzed day)
@@ -60,9 +60,9 @@ effect t(df) / F(df1) p (resid) Satterthwaite: p (df)
stim x day (interaction) t(32)= 2.57 F(1)= 6.588 p=0.01515 p=0.0155 (df=30) stim x day (interaction) t(32)= 2.57 F(1)= 6.588 p=0.01515 p=0.0155 (df=30)
day (learning) t(32)= 6.67 F(1)= 44.528 p=1.571e-07 p=2.162e-07 (df=30) day (learning) t(32)= 6.67 F(1)= 44.528 p=1.571e-07 p=2.162e-07 (df=30)
stim (main, window start) t(32)= -0.40 F(1)= 0.163 p=0.6888 p=0.6906 (df=19) stim (main, window start) t(32)= -0.40 F(1)= 0.163 p=0.6888 p=0.6906 (df=19)
interaction 95% CI: [+1.11, +9.62] interaction 95% CI: [+1.107, +9.617]
HONEST LME (per-animal random slope, day|rat): interaction F(1,6.0)=3.02, p=0.1331 HONEST LME (per-animal random slope, day|rat): interaction F(1,6.0)=3.02, p=0.1331
(Satterthwaite DF ~= residual on this random-intercept model; the random-slope (Satterthwaite DF ~= residual on this random-intercept model; the random-slope
model above is the honest learning-rate test -- DF collapses toward the animal count.) model above is the honest learning-rate test -- DF collapses toward the animal count.)
INTERPRETATION: stim x day interaction SIGNIFICANT positive -- treatment improves FASTER (benefit accumulates) (p=0.01515, slope diff=+5.36) INTERPRETATION: stim x day interaction SIGNIFICANT positive -- treatment improves FASTER (benefit accumulates) (p=0.01515, slope diff=+5.362)
Paper (N=24): interaction t(227)=2.68, F(1)=7.12, p=0.008. Paper (N=24, count): interaction t(227)=2.68, F(1)=7.12, p=0.008.
analysis/matlab/variations/unmerge_d0_5/result_rate.txt
+68
View File
@@ -0,0 +1,68 @@
==============================================================================
VARIATION: unmerge_d0_5 [metric: success RATE (success/attempts)]
==============================================================================
model: behavior ~ stim + day + stim:day + (1|rat) (behavior = success RATE (success/attempts))
day = training day within window (0 = first analyzed day)
treatment (stim=1): Electrode-Box-B2
control (stim=0): Electrode-Box-A2
N = 6 rats, 36 sessions raw day coverage: treat 0..5, control 0..5
(equal day coverage over this window)
==============================================================================
FULL MODEL SUMMARY -- fitlme
==============================================================================
Linear mixed-effects model fit by ML
Model information:
Number of observations 36
Fixed effects coefficients 4
Random effects coefficients 6
Covariance parameters 2
Formula:
behavior ~ 1 + day*stim + (1 | rat)
Model fit statistics:
AIC BIC LogLikelihood Deviance
-62.201 -52.7 37.1 -74.201
Fixed effects coefficients (95% CIs):
Name Estimate SE tStat DF pValue
{'(Intercept)'} 0.2186 0.042385 5.1574 32 1.2572e-05
{'day' } 0.043366 0.010845 3.9986 32 0.0003516
{'stim' } -0.053758 0.059942 -0.89683 32 0.37651
{'day:stim' } 0.049022 0.015337 3.1962 32 0.0031273
Lower Upper
0.13226 0.30494
0.021275 0.065457
-0.17586 0.06834
0.01778 0.080263
Random effects covariance parameters (95% CIs):
Group: rat (6 Levels)
Name1 Name2 Type Estimate
{'(Intercept)'} {'(Intercept)'} {'std'} 0.046423
Lower Upper
0.019948 0.10804
Group: Error
Name Estimate Lower Upper
{'Res Std'} 0.078581 0.061014 0.10121
effect t(df) / F(df1) p (resid) Satterthwaite: p (df)
----------------------------------------------------------------------------
stim x day (interaction) t(32)= 3.20 F(1)= 10.216 p=0.003127 p=0.00327 (df=30)
day (learning) t(32)= 4.00 F(1)= 15.989 p=0.0003516 p=0.0003833 (df=30)
stim (main, window start) t(32)= -0.90 F(1)= 0.804 p=0.3765 p=0.3834 (df=16)
interaction 95% CI: [+0.01778, +0.08026]
HONEST LME (per-animal random slope, day|rat): interaction F(1,7.1)=6.49, p=0.0379
(Satterthwaite DF ~= residual on this random-intercept model; the random-slope
model above is the honest learning-rate test -- DF collapses toward the animal count.)
INTERPRETATION: stim x day interaction SIGNIFICANT positive -- treatment improves FASTER (benefit accumulates) (p=0.003127, slope diff=+0.04902)
Paper (N=24, count): interaction t(227)=2.68, F(1)=7.12, p=0.008.
analysis/matlab/variations/unmerge_d6_10/analyze.m
+63 -39
View File
@@ -1,30 +1,58 @@
% Variation analysis -- the paper's linear mixed model on the successful-reach % Variation analysis -- the paper's linear mixed model on this folder's data,
% COUNT, fit on this folder's curated data subset. % for BOTH metrics:
% metric = count : behavior = # successes -> result.txt
% metric = rate : behavior = success / attempts -> result_rate.txt
% (rate uses only sessions with attempts > 0)
% %
% model: behavior ~ stim + day + stim:day + (1|rat) % model: behavior ~ stim + day + stim:day + (1|rat)
% behavior = successful reaches (count per session) % stim = 1 treatment / 0 control; day = training day within window (0 =
% stim = 1 for the treatment group(s), 0 for the control group(s) % first analyzed day); rat = subject (random intercept).
% day = training day within this window (0 = first analyzed day) % For the interaction we report residual DF, Satterthwaite DF, and the honest
% rat = subject (random intercept) % per-animal random-slope test. Self-contained: reads data.csv beside this
% % script. Run headless with: matlab -batch "analyze"
% Self-contained: reads data.csv beside this script and writes result.txt. % (Copy of analysis/matlab/variation_analyze.m; see make_variations.m.)
% Run headless from this folder with: matlab -batch "analyze"
% (This is a copy of analysis/matlab/variation_analyze.m; see make_variations.m.)
here = fileparts(mfilename('fullpath')); here = fileparts(mfilename('fullpath'));
if isempty(here); here = pwd; end if isempty(here); here = pwd; end
vname = regexprep(here, '.*[/\\]', ''); % folder name = variation id vname = regexprep(here, '.*[/\\]', '');
D = readtable(fullfile(here, 'data.csv'), 'TextType', 'string'); D = readtable(fullfile(here, 'data.csv'), 'TextType', 'string');
tbl = table(D.success, D.day - min(D.day), double(D.stim), categorical(D.subject), ... Rc = localAnalyze(D, 'count', here, vname);
Rr = localAnalyze(D, 'rate', here, vname);
% Machine-readable handoff for SUMMARY.csv (count drives it; rate appended).
VARRESULT = struct('name', vname, 'nRats', Rc.nRats, 'nObs', Rc.nObs, ...
'interP', Rc.interP, 'interEst', Rc.interEst, ...
'interPsatt', Rc.interPsatt, 'interPrs', Rc.interPrs, ...
'stimP', Rc.stimP, 'dayP', Rc.dayP, 'covEqual', Rc.covEqual, ...
'interPrate', Rr.interP, 'interEstRate', Rr.interEst, 'interPrsRate', Rr.interPrs);
% ------------------------------------------------------------------ helper
function R = localAnalyze(D, metric, here, vname)
if strcmp(metric, 'rate')
D = D(D.total > 0, :);
beh = D.success ./ D.total;
mlabel = 'success RATE (success/attempts)'; suffix = '_rate';
else
beh = D.success;
mlabel = 'success COUNT'; suffix = '';
end
R = struct('interP', NaN, 'interEst', NaN, 'interPsatt', NaN, 'interPrs', NaN, ...
'stimP', NaN, 'dayP', NaN, 'nRats', numel(unique(D.subject)), ...
'nObs', height(D), 'covEqual', false);
if numel(unique(D.stim)) < 2 || numel(unique(D.day)) < 2
localWrite(sprintf('VARIATION: %s [metric: %s]\nInsufficient data for this metric.\n', ...
vname, mlabel), here, suffix);
return
end
tbl = table(beh, D.day - min(D.day), double(D.stim), categorical(D.subject), ...
'VariableNames', {'behavior', 'day', 'stim', 'rat'}); 'VariableNames', {'behavior', 'day', 'stim', 'rat'});
m = fitlme(tbl, 'behavior ~ stim + day + stim:day + (1|rat)'); m = fitlme(tbl, 'behavior ~ stim + day + stim:day + (1|rat)');
C = m.Coefficients; A = anova(m); ci = coefCI(m); C = m.Coefficients; A = anova(m); ci = coefCI(m);
As = anova(m, 'DFMethod', 'satterthwaite'); % Satterthwaite denominator DF As = anova(m, 'DFMethod', 'satterthwaite');
% Honest test: refit with a per-animal random SLOPE so the interaction DF
% collapses toward the animal count (guarded -- may not converge in short windows).
rsP = NaN; rsDf = NaN; rsF = NaN; rsOk = false; rsP = NaN; rsDf = NaN; rsF = NaN; rsOk = false;
wst = warning('off', 'all'); wst = warning('off', 'all');
try try
@@ -43,6 +71,7 @@ row = @(nm, t) sprintf('%-26s t(%d)=%6.2f F(%d)=%7.3f p=%.4g p=%.4g (df=%.0f
C.DF(gi(t)), C.tStat(gi(t)), A.DF1(ga(t)), A.FStat(ga(t)), C.pValue(gi(t)), ... C.DF(gi(t)), C.tStat(gi(t)), A.DF1(ga(t)), A.FStat(ga(t)), C.pValue(gi(t)), ...
As.pValue(gs(t)), As.DF2(gs(t))); As.pValue(gs(t)), As.DF2(gs(t)));
ii = gi('day:stim'); pI = C.pValue(ii); eI = C.Estimate(ii);
maxT = max(D.day(D.stim == 1)); minT = min(D.day(D.stim == 1)); maxT = max(D.day(D.stim == 1)); minT = min(D.day(D.stim == 1));
maxC = max(D.day(D.stim == 0)); minC = min(D.day(D.stim == 0)); maxC = max(D.day(D.stim == 0)); minC = min(D.day(D.stim == 0));
if abs(maxT - maxC) > 2 if abs(maxT - maxC) > 2
@@ -50,20 +79,14 @@ if abs(maxT - maxC) > 2
else else
cov = '(equal day coverage over this window)'; cov = '(equal day coverage over this window)';
end end
if pI >= 0.05; verdict = 'n.s. -- slopes parallel (no differential learning rate)';
ii = gi('day:stim'); pI = C.pValue(ii); eI = C.Estimate(ii); elseif eI > 0; verdict = 'SIGNIFICANT positive -- treatment improves FASTER (benefit accumulates)';
if pI >= 0.05 else; verdict = 'SIGNIFICANT negative -- treatment improves SLOWER (groups converge)'; end
verdict = 'n.s. -- slopes parallel (no differential learning rate)';
elseif eI > 0
verdict = 'SIGNIFICANT positive -- treatment improves FASTER (benefit accumulates)';
else
verdict = 'SIGNIFICANT negative -- treatment improves SLOWER (groups converge)';
end
bar = repmat('=', 1, 78); bar = repmat('=', 1, 78);
raw = regexprep(evalc('disp(m)'), '</?strong>', ''); raw = regexprep(evalc('disp(m)'), '</?strong>', '');
s = sprintf('%s\nVARIATION: %s\n%s\n', bar, vname, bar); s = sprintf('%s\nVARIATION: %s [metric: %s]\n%s\n', bar, vname, mlabel, bar);
s = [s sprintf('model: behavior ~ stim + day + stim:day + (1|rat) (behavior = success COUNT)\n')]; s = [s sprintf('model: behavior ~ stim + day + stim:day + (1|rat) (behavior = %s)\n', mlabel)];
s = [s sprintf('day = training day within window (0 = first analyzed day)\n')]; s = [s sprintf('day = training day within window (0 = first analyzed day)\n')];
s = [s sprintf('treatment (stim=1): %s\n', strjoin(cellstr(unique(D.group(D.stim == 1))), ', '))]; s = [s sprintf('treatment (stim=1): %s\n', strjoin(cellstr(unique(D.group(D.stim == 1))), ', '))];
s = [s sprintf('control (stim=0): %s\n', strjoin(cellstr(unique(D.group(D.stim == 0))), ', '))]; s = [s sprintf('control (stim=0): %s\n', strjoin(cellstr(unique(D.group(D.stim == 0))), ', '))];
@@ -74,7 +97,7 @@ s = [s sprintf('%-26s %-18s %-12s %s\n%s\n', 'effect', 't(df) / F(df1)', 'p (res
s = [s row('stim x day (interaction)', 'day:stim')]; s = [s row('stim x day (interaction)', 'day:stim')];
s = [s row('day (learning)', 'day')]; s = [s row('day (learning)', 'day')];
s = [s row('stim (main, window start)', 'stim')]; s = [s row('stim (main, window start)', 'stim')];
s = [s sprintf('interaction 95%% CI: [%+.2f, %+.2f]\n', ci(ii, 1), ci(ii, 2))]; s = [s sprintf('interaction 95%% CI: [%+.4g, %+.4g]\n', ci(ii, 1), ci(ii, 2))];
if rsOk if rsOk
s = [s sprintf('HONEST LME (per-animal random slope, day|rat): interaction F(1,%.1f)=%.2f, p=%.4g\n', rsDf, rsF, rsP)]; s = [s sprintf('HONEST LME (per-animal random slope, day|rat): interaction F(1,%.1f)=%.2f, p=%.4g\n', rsDf, rsF, rsP)];
else else
@@ -82,17 +105,18 @@ else
end end
s = [s sprintf([' (Satterthwaite DF ~= residual on this random-intercept model; the random-slope\n' ... s = [s sprintf([' (Satterthwaite DF ~= residual on this random-intercept model; the random-slope\n' ...
' model above is the honest learning-rate test -- DF collapses toward the animal count.)\n'])]; ' model above is the honest learning-rate test -- DF collapses toward the animal count.)\n'])];
s = [s sprintf('INTERPRETATION: stim x day interaction %s (p=%.4g, slope diff=%+.2f)\n', verdict, pI, eI)]; s = [s sprintf('INTERPRETATION: stim x day interaction %s (p=%.4g, slope diff=%+.4g)\n', verdict, pI, eI)];
s = [s sprintf('Paper (N=24): interaction t(227)=2.68, F(1)=7.12, p=0.008.\n')]; s = [s sprintf('Paper (N=24, count): interaction t(227)=2.68, F(1)=7.12, p=0.008.\n')];
localWrite(s, here, suffix);
R = struct('interP', pI, 'interEst', eI, 'interPsatt', As.pValue(gs('day:stim')), ...
'interPrs', rsP, 'stimP', C.pValue(gi('stim')), 'dayP', C.pValue(gi('day')), ...
'nRats', numel(unique(D.subject)), 'nObs', height(D), 'covEqual', abs(maxT - maxC) <= 2);
end
function localWrite(s, here, suffix)
fprintf('%s', s); fprintf('%s', s);
fid = fopen(fullfile(here, 'result.txt'), 'w'); fid = fopen(fullfile(here, ['result' suffix '.txt']), 'w');
fprintf(fid, '%s', s); fprintf(fid, '%s', s); fclose(fid);
fclose(fid); end
% Machine-readable handoff for the summary table (see make_variations.m).
VARRESULT = struct('name', vname, 'nRats', numel(unique(D.subject)), ...
'nObs', height(D), 'interP', pI, 'interEst', eI, ...
'interPsatt', As.pValue(gs('day:stim')), 'interPrs', rsP, ...
'stimP', C.pValue(gi('stim')), 'dayP', C.pValue(gi('day')), ...
'covEqual', abs(maxT - maxC) <= 2);
analysis/matlab/variations/unmerge_d6_10/learning_curve.png
BIN
View File
Binary file not shown.
Side by Side Swipe Overlay

Before

Width:  |  Height:  |  Size: 36 KiB

After

Width:  |  Height:  |  Size: 36 KiB

analysis/matlab/variations/unmerge_d6_10/learning_curve_rate.png
BIN
View File
Binary file not shown.
Side by Side

After

Width:  |  Height:  |  Size: 36 KiB

analysis/matlab/variations/unmerge_d6_10/logpower_result.txt
+5 -6
View File
@@ -1,7 +1,7 @@
============================================================================== ==============================================================================
LOG-DAY MODEL + COHEN'S f + POWER -- unmerge_d6_10 LOG-DAY MODEL + COHEN'S f + POWER -- unmerge_d6_10 [metric: # successes (count)]
============================================================================== ==============================================================================
model: behavior ~ stim + log(day) + stim:log(day) + (1|rat) (success COUNT) model: behavior ~ stim + log(day) + stim:log(day) + (1|rat) (behavior = # successes (count))
log(day) uses 1-indexed training day (our day 0 = paper "Day 1") log(day) uses 1-indexed training day (our day 0 = paper "Day 1")
observed groups: stim n=3, control n=2 nrep=120, alpha=0.05 observed groups: stim n=3, control n=2 nrep=120, alpha=0.05
@@ -10,7 +10,7 @@ stim x log(day) interaction: F(1,21)=0.581 p(resid)=0.4545 p(Satt)=0.4549 (df=
honest per-animal random slope (log-day): F(1,5.0)=0.44 p=0.5382 honest per-animal random slope (log-day): F(1,5.0)=0.44 p=0.5382
Cohen's f (interaction, partial eta^2=0.011) = 0.106 (small-medium; f: .10 small, .25 medium, .40 large) Cohen's f (interaction, partial eta^2=0.011) = 0.106 (small-medium; f: .10 small, .25 medium, .40 large)
--- power simulation (log-day ground truth: stim:logday=+16.98, ratSD=6.08, resSD=8.72) --- --- power simulation (log-day ground truth: stim:logday=+16.98, ratSD=6.081, resSD=8.718) ---
true stim:log(day) = +16.98 (100% of observed) true stim:log(day) = +16.98 (100% of observed)
N/group | per-animal power | LME power N/group | per-animal power | LME power
@@ -22,7 +22,7 @@ Cohen's f (interaction, partial eta^2=0.011) = 0.106 (small-medium; f: .10 smal
16 | 0.47 | 0.50 16 | 0.47 | 0.50
24 | 0.63 | 0.66 24 | 0.63 | 0.66
true stim:log(day) = +8.49 (50% of observed) true stim:log(day) = +8.489 (50% of observed)
N/group | per-animal power | LME power N/group | per-animal power | LME power
------------------------------------------ ------------------------------------------
3 | 0.07 | 0.11 <- observed 3 | 0.07 | 0.11 <- observed
@@ -32,5 +32,4 @@ Cohen's f (interaction, partial eta^2=0.011) = 0.106 (small-medium; f: .10 smal
16 | 0.17 | 0.17 16 | 0.17 | 0.17
24 | 0.26 | 0.26 24 | 0.26 | 0.26
Read the per-animal column as the honest power; the LME column matches the Read per-animal as the honest power; LME matches the paper's power code (optimistic).
paper's power code (anova interaction p, observation-level DF) and is optimistic.
analysis/matlab/variations/unmerge_d6_10/logpower_result_rate.txt
+35
View File
@@ -0,0 +1,35 @@
==============================================================================
LOG-DAY MODEL + COHEN'S f + POWER -- unmerge_d6_10 [metric: success RATE]
==============================================================================
model: behavior ~ stim + log(day) + stim:log(day) + (1|rat) (behavior = success RATE)
log(day) uses 1-indexed training day (our day 0 = paper "Day 1")
observed groups: stim n=3, control n=2 nrep=120, alpha=0.05
--- fitted on real data ---
stim x log(day) interaction: F(1,21)=0.027 p(resid)=0.8704 p(Satt)=0.8705 (df=20)
honest per-animal random slope (log-day): F(1,5.0)=0.01 p=0.9117
Cohen's f (interaction, partial eta^2=0.001) = 0.023 (small; f: .10 small, .25 medium, .40 large)
--- power simulation (log-day ground truth: stim:logday=+0.02357, ratSD=0.03992, resSD=0.05586) ---
true stim:log(day) = +0.02357 (100% of observed)
N/group | per-animal power | LME power
------------------------------------------
3 | 0.03 | 0.03 <- observed
5 | 0.03 | 0.06
8 | 0.06 | 0.07
12 | 0.08 | 0.09
16 | 0.05 | 0.06
24 | 0.08 | 0.06
true stim:log(day) = +0.01179 (50% of observed)
N/group | per-animal power | LME power
------------------------------------------
3 | 0.03 | 0.08 <- observed
5 | 0.07 | 0.07
8 | 0.05 | 0.05
12 | 0.07 | 0.07
16 | 0.07 | 0.07
24 | 0.06 | 0.04
Read per-animal as the honest power; LME matches the paper's power code (optimistic).
analysis/matlab/variations/unmerge_d6_10/logpowersim.m
+38 -39
View File
@@ -1,50 +1,52 @@
% Variation log-day analysis + Cohen's f + power simulation. % Variation log-day analysis + Cohen's f + power simulation, for BOTH metrics:
% metric = count : behavior = # successes -> logpower_result.txt
% metric = rate : behavior = success / attempts -> logpower_result_rate.txt
% %
% The paper's power code models behavior against LOG training day, not raw day: % The paper's power code models behavior against LOG training day:
% behavior ~ stim + log(day) + stim:log(day) + (1|rat). % behavior ~ stim + log(day) + stim:log(day) + (1|rat).
% Their day is 1-indexed (1..10); our data.csv day is 0-indexed (day 0 = paper % Their day is 1-indexed; our data.csv day is 0-indexed, so log(day + 1)
% "Day 1"), so log(day + 1) reproduces their transform exactly. % reproduces their transform (our day 0 = paper "Day 1"). For each metric this
% % refits that model, reports the interaction (residual / Satterthwaite / honest
% This script (a) refits that log-day model on data.csv, (b) reports the % per-animal random-slope DF) and Cohen's f (partial-eta^2 effect size), then
% interaction (residual DF, Satterthwaite DF, and the honest per-animal % runs the Monte-Carlo power sim (per-animal cluster-honest + LME power).
% random-slope test) and Cohen's f -- the partial-eta^2 effect size of the % Run: matlab -batch "logpowersim"
% interaction, var(fitted_full) - var(fitted_no_interaction) over var(behavior)
% -- and (c) runs the Monte-Carlo power simulation on the log-day model,
% scoring per-animal (cluster-honest) and LME power across N.
% Writes logpower_result.txt. Run: matlab -batch "logpowersim"
% (Copy of analysis/matlab/variation_logpower.m; see make_variation_logpower.m.) % (Copy of analysis/matlab/variation_logpower.m; see make_variation_logpower.m.)
here = fileparts(mfilename('fullpath')); here = fileparts(mfilename('fullpath'));
if isempty(here); here = pwd; end if isempty(here); here = pwd; end
vname = regexprep(here, '.*[/\\]', ''); vname = regexprep(here, '.*[/\\]', '');
D = readtable(fullfile(here, 'data.csv'), 'TextType', 'string'); D = readtable(fullfile(here, 'data.csv'), 'TextType', 'string');
FORMULA = 'behavior ~ stim + day + stim:day + (1|rat)'; % 'day' column = log(day+1)
NS = [3 5 8 12 16 24];
EFFMULS = [1 0.5];
NREP = 120;
warnState = warning('off', 'all'); localLogPower(D, 'count', here, vname);
rng(1); localLogPower(D, 'rate', here, vname);
logday = log(D.day + 1); % 0-indexed day -> their log(1-indexed day) % ---------------------------------------------------------------- per metric
tbl0 = table(D.success, logday, double(D.stim), categorical(D.subject), ... function localLogPower(D, metric, here, vname)
NS = [3 5 8 12 16 24]; EFFMULS = [1 0.5]; NREP = 120;
FORMULA = 'behavior ~ stim + day + stim:day + (1|rat)'; % 'day' = log(day+1)
if strcmp(metric, 'rate')
D = D(D.total > 0, :); beh = D.success ./ D.total; mlabel = 'success RATE'; suffix = '_rate';
else
beh = D.success; mlabel = '# successes (count)'; suffix = '';
end
warnState = warning('off', 'all'); rng(1);
logday = log(D.day + 1);
tbl0 = table(beh, logday, double(D.stim), categorical(D.subject), ...
'VariableNames', {'behavior', 'day', 'stim', 'rat'}); 'VariableNames', {'behavior', 'day', 'stim', 'rat'});
nStim = numel(unique(D.subject(D.stim == 1))); nStim = numel(unique(D.subject(D.stim == 1)));
nCtrl = numel(unique(D.subject(D.stim == 0))); nCtrl = numel(unique(D.subject(D.stim == 0)));
bar = repmat('=', 1, 78); bar = repmat('=', 1, 78);
s = sprintf('%s\nLOG-DAY MODEL + COHEN''S f + POWER -- %s\n%s\n', bar, vname, bar); s = sprintf('%s\nLOG-DAY MODEL + COHEN''S f + POWER -- %s [metric: %s]\n%s\n', bar, vname, mlabel, bar);
s = [s sprintf('model: behavior ~ stim + log(day) + stim:log(day) + (1|rat) (success COUNT)\n')]; s = [s sprintf('model: behavior ~ stim + log(day) + stim:log(day) + (1|rat) (behavior = %s)\n', mlabel)];
s = [s sprintf('log(day) uses 1-indexed training day (our day 0 = paper "Day 1")\n')]; s = [s sprintf('log(day) uses 1-indexed training day (our day 0 = paper "Day 1")\n')];
s = [s sprintf('observed groups: stim n=%d, control n=%d nrep=%d, alpha=0.05\n', nStim, nCtrl, NREP)]; s = [s sprintf('observed groups: stim n=%d, control n=%d nrep=%d, alpha=0.05\n', nStim, nCtrl, NREP)];
if nStim < 2 || nCtrl < 2 || numel(unique(tbl0.day)) < 2 if nStim < 2 || nCtrl < 2 || numel(unique(tbl0.day)) < 2
s = [s sprintf('\nInsufficient data for this analysis (need >=2 animals/group and >=2 days).\n')]; s = [s sprintf('\nInsufficient data for this analysis.\n')];
localFinish(s, here); warning(warnState); return localFinish(s, here, suffix); warning(warnState); return
end end
% ---- fitted model on the real data ----
full = fitlme(tbl0, FORMULA); full = fitlme(tbl0, FORMULA);
An = anova(full); Asatt = anova(full, 'DFMethod', 'satterthwaite'); An = anova(full); Asatt = anova(full, 'DFMethod', 'satterthwaite');
ii = strcmp(An.Term, 'day:stim'); is = strcmp(Asatt.Term, 'day:stim'); ii = strcmp(An.Term, 'day:stim'); is = strcmp(Asatt.Term, 'day:stim');
@@ -52,7 +54,6 @@ reduced = fitlme(tbl0, 'behavior ~ stim + day + (1|rat)');
eta2part = max((var(fitted(full)) - var(fitted(reduced))) / var(tbl0.behavior), 0); eta2part = max((var(fitted(full)) - var(fitted(reduced))) / var(tbl0.behavior), 0);
cohenf = sqrt(eta2part / (1 - eta2part)); cohenf = sqrt(eta2part / (1 - eta2part));
% honest per-animal random-slope interaction
rsP = NaN; rsDf = NaN; rsF = NaN; rsOk = false; rsP = NaN; rsDf = NaN; rsF = NaN; rsOk = false;
try try
mr = fitlme(tbl0, 'behavior ~ stim + day + stim:day + (day|rat)'); mr = fitlme(tbl0, 'behavior ~ stim + day + stim:day + (day|rat)');
@@ -76,19 +77,18 @@ end
s = [s sprintf('Cohen''s f (interaction, partial eta^2=%.3f) = %.3f (%s; f: .10 small, .25 medium, .40 large)\n', ... s = [s sprintf('Cohen''s f (interaction, partial eta^2=%.3f) = %.3f (%s; f: .10 small, .25 medium, .40 large)\n', ...
eta2part, cohenf, mag)]; eta2part, cohenf, mag)];
% ---- power simulation on the log-day ground truth ----
cn = full.CoefficientNames; be = full.fixedEffects; cn = full.CoefficientNames; be = full.fixedEffects;
b0 = be(strcmp(cn, '(Intercept)')); bStim = be(strcmp(cn, 'stim')); b0 = be(strcmp(cn, '(Intercept)')); bStim = be(strcmp(cn, 'stim'));
bDay = be(strcmp(cn, 'day')); bInt = be(strcmp(cn, 'day:stim')); bDay = be(strcmp(cn, 'day')); bInt = be(strcmp(cn, 'day:stim'));
psi = covarianceParameters(full); sRat = sqrt(psi{1}); sRes = sqrt(full.MSE); psi = covarianceParameters(full); sRat = sqrt(psi{1}); sRes = sqrt(full.MSE);
days = unique(tbl0.day); % the log(day) grid days = unique(tbl0.day);
s = [s sprintf('\n--- power simulation (log-day ground truth: stim:logday=%+.2f, ratSD=%.2f, resSD=%.2f) ---\n', ... s = [s sprintf('\n--- power simulation (log-day ground truth: stim:logday=%+.4g, ratSD=%.4g, resSD=%.4g) ---\n', ...
bInt, sRat, sRes)]; bInt, sRat, sRes)];
for eMul = EFFMULS for eMul = EFFMULS
bI = bInt * eMul; bI = bInt * eMul;
s = [s sprintf('\n true stim:log(day) = %+.2f (%.0f%% of observed)\n', bI, eMul * 100)]; %#ok<AGROW> s = [s sprintf('\n true stim:log(day) = %+.4g (%.0f%% of observed)\n', bI, eMul * 100)];
s = [s sprintf(' %-8s | per-animal power | LME power\n %s\n', 'N/group', repmat('-', 1, 42))]; %#ok<AGROW> s = [s sprintf(' %-8s | per-animal power | LME power\n %s\n', 'N/group', repmat('-', 1, 42))];
for N = NS for N = NS
sigPA = 0; sigL = 0; sigPA = 0; sigL = 0;
for r = 1:NREP for r = 1:NREP
@@ -102,19 +102,18 @@ for eMul = EFFMULS
end end
star = ''; star = '';
if N == nStim || N == nCtrl; star = ' <- observed'; end if N == nStim || N == nCtrl; star = ' <- observed'; end
s = [s sprintf(' %-8d | %5.2f | %5.2f%s\n', N, sigPA / NREP, sigL / NREP, star)]; %#ok<AGROW> s = [s sprintf(' %-8d | %5.2f | %5.2f%s\n', N, sigPA / NREP, sigL / NREP, star)];
end end
end end
s = [s sprintf(['\nRead the per-animal column as the honest power; the LME column matches the\n' ... s = [s sprintf('\nRead per-animal as the honest power; LME matches the paper''s power code (optimistic).\n')];
'paper''s power code (anova interaction p, observation-level DF) and is optimistic.\n'])]; localFinish(s, here, suffix);
localFinish(s, here);
warning(warnState); warning(warnState);
end
% ---------------------------------------------------------------- helpers % ---------------------------------------------------------------- helpers
function localFinish(s, here) function localFinish(s, here, suffix)
fprintf('%s', s); fprintf('%s', s);
fid = fopen(fullfile(here, 'logpower_result.txt'), 'w'); fid = fopen(fullfile(here, ['logpower_result' suffix '.txt']), 'w');
fprintf(fid, '%s', s); fclose(fid); fprintf(fid, '%s', s); fclose(fid);
end end
analysis/matlab/variations/unmerge_d6_10/plotcurve.m
+29 -19
View File
@@ -1,11 +1,12 @@
% Variation learning-curve plot, in the style of the paper: % Variation learning-curve plots, in the style of the paper:
% "Lines indicate mean (and SEM) across animals in the anodal (red) and % "Lines indicate mean (and SEM) across animals in the anodal (red) and
% control (blue) groups." % control (blue) groups."
% Plots mean +/- SEM successful reaches per training day for the treatment / % Produces TWO figures from this folder's data.csv:
% anodal group (stim = 1, red) and the control group (stim = 0, blue), reading % learning_curve.png # successes (count) per training day
% this folder's data.csv and saving learning_curve.png. The per-group N is read % learning_curve_rate.png success rate (success/attempts) per training day
% from the data (each variation pools different groups), so the legend shows the % anodal / treatment = stim 1 (red); control = stim 0 (blue). Per-group N is
% actual counts. Training day is 1-indexed (our day 0 = the paper's "Day 1"). % read from the data. Training day is 1-indexed (our day 0 = paper "Day 1") and
% the x-axis tick labels are drawn vertically.
% Run: matlab -batch "plotcurve" % Run: matlab -batch "plotcurve"
% (Copy of analysis/matlab/variation_plot.m; see make_variation_plot.m.) % (Copy of analysis/matlab/variation_plot.m; see make_variation_plot.m.)
@@ -14,15 +15,20 @@ if isempty(here); here = pwd; end
vname = regexprep(here, '.*[/\\]', ''); vname = regexprep(here, '.*[/\\]', '');
D = readtable(fullfile(here, 'data.csv'), 'TextType', 'string'); D = readtable(fullfile(here, 'data.csv'), 'TextType', 'string');
days = unique(D.day); % 0-indexed days = unique(D.day);
xd = days + 1; % plot as 1-indexed training day (paper axis) xd = days + 1; % plot as 1-indexed training day (paper axis)
red = [0.85 0.10 0.10]; red = [0.85 0.10 0.10];
blue = [0.10 0.30 0.85]; blue = [0.10 0.30 0.85];
[Ma, Sa, na] = localCurve(D, 1, days); % anodal / treatment (stim = 1) localPlot(D, days, xd, 'count', '# successes', ...
[Mc, Sc, nc] = localCurve(D, 0, days); % control (stim = 0) fullfile(here, 'learning_curve.png'), vname, red, blue);
localPlot(D, days, xd, 'rate', 'success rate', ...
fullfile(here, 'learning_curve_rate.png'), vname, red, blue);
% ------------------------------------------------------------------ helpers
function localPlot(D, days, xd, metric, ylab, outFile, vname, red, blue)
[Ma, Sa, na] = localCurve(D, 1, days, metric); % anodal / treatment
[Mc, Sc, nc] = localCurve(D, 0, days, metric); % control
fig = figure('Visible', 'off', 'Color', 'w', 'Position', [100 100 560 460]); fig = figure('Visible', 'off', 'Color', 'w', 'Position', [100 100 560 460]);
hold on hold on
e1 = errorbar(xd, Ma, Sa, '-o', 'Color', red, 'MarkerFaceColor', red, 'LineWidth', 2); e1 = errorbar(xd, Ma, Sa, '-o', 'Color', red, 'MarkerFaceColor', red, 'LineWidth', 2);
@@ -30,26 +36,30 @@ e2 = errorbar(xd, Mc, Sc, '-o', 'Color', blue, 'MarkerFaceColor', blue, 'LineWid
hold off hold off
legend([e1 e2], {sprintf('anodal, N = %d', na), sprintf('control, N = %d', nc)}, ... legend([e1 e2], {sprintf('anodal, N = %d', na), sprintf('control, N = %d', nc)}, ...
'Location', 'northwest', 'Box', 'off'); 'Location', 'northwest', 'Box', 'off');
xlabel('training day'); xlabel('training day'); ylabel(ylab);
ylabel('# successes');
title(vname, 'Interpreter', 'none'); title(vname, 'Interpreter', 'none');
set(gca, 'XTick', xd, 'FontName', 'Arial', 'FontSize', 13, 'LineWidth', 1.5, 'Box', 'off'); set(gca, 'XTick', xd, 'FontName', 'Arial', 'FontSize', 13, 'LineWidth', 1.5, 'Box', 'off');
xtickangle(90); % vertical x-axis tick labels
outFile = fullfile(here, 'learning_curve.png');
exportgraphics(fig, outFile, 'Resolution', 150); exportgraphics(fig, outFile, 'Resolution', 150);
close(fig); close(fig);
fprintf('%s: wrote learning_curve.png (anodal N=%d, control N=%d)\n', vname, na, nc); fprintf('%s: wrote %s (anodal N=%d, control N=%d)\n', vname, outFile, na, nc);
end
% ------------------------------------------------------------------ helper function [M, S, n] = localCurve(D, stimVal, days, metric)
function [M, S, n] = localCurve(D, stimVal, days) %LOCALCURVE Per-day mean and SEM across the animals in a group, for a metric.
%LOCALCURVE Per-day mean and SEM of successes across the animals in a group.
subs = unique(D.subject(D.stim == stimVal)); subs = unique(D.subject(D.stim == stimVal));
n = numel(subs); n = numel(subs);
X = nan(numel(days), n); X = nan(numel(days), n);
for j = 1:n for j = 1:n
for i = 1:numel(days) for i = 1:numel(days)
r = D.subject == subs(j) & D.day == days(i); r = D.subject == subs(j) & D.day == days(i);
if any(r); X(i, j) = mean(D.success(r)); end if ~any(r); continue; end
if strcmp(metric, 'rate')
tot = sum(D.total(r));
if tot > 0; X(i, j) = sum(D.success(r)) / tot; end
else
X(i, j) = mean(D.success(r));
end
end end
end end
M = mean(X, 2, 'omitnan'); M = mean(X, 2, 'omitnan');
analysis/matlab/variations/unmerge_d6_10/power_result.txt
+6 -7
View File
@@ -1,11 +1,11 @@
============================================================================== ==============================================================================
POWER SIMULATION -- unmerge_d6_10 POWER SIMULATION -- unmerge_d6_10 [metric: # successes (count)]
============================================================================== ==============================================================================
model: behavior ~ stim + day + stim:day + (1|rat) (success COUNT; day within-window) model: behavior ~ stim + day + stim:day + (1|rat) (behavior = # successes (count); day within-window)
observed groups: stim n=3, control n=2 nrep=120, alpha=0.05 observed groups: stim n=3, control n=2 nrep=120, alpha=0.05
ground truth: stim:day=+1.77/day, rat SD=6.11, residual SD=8.63, days=5 ground truth: stim:day=+1.767/day, rat SD=6.107, residual SD=8.629, days=5
true stim:day interaction = +1.77 (100% of observed) true stim:day interaction = +1.767 (100% of observed)
N/group | per-animal power | LME power N/group | per-animal power | LME power
------------------------------------------ ------------------------------------------
3 | 0.05 | 0.10 <- observed 3 | 0.05 | 0.10 <- observed
@@ -15,7 +15,7 @@ ground truth: stim:day=+1.77/day, rat SD=6.11, residual SD=8.63, days=5
16 | 0.44 | 0.42 16 | 0.44 | 0.42
24 | 0.57 | 0.58 24 | 0.57 | 0.58
true stim:day interaction = +0.88 (50% of observed) true stim:day interaction = +0.8833 (50% of observed)
N/group | per-animal power | LME power N/group | per-animal power | LME power
------------------------------------------ ------------------------------------------
3 | 0.07 | 0.11 <- observed 3 | 0.07 | 0.11 <- observed
@@ -25,5 +25,4 @@ ground truth: stim:day=+1.77/day, rat SD=6.11, residual SD=8.63, days=5
16 | 0.13 | 0.17 16 | 0.13 | 0.17
24 | 0.22 | 0.23 24 | 0.22 | 0.23
Read the per-animal column as the honest power. At the observed N this study Read the per-animal column as the honest power; LME is optimistic (obs-level DF).
is typically underpowered; per-animal power reaches ~0.8 only at larger N.
analysis/matlab/variations/unmerge_d6_10/power_result_rate.txt
+28
View File
@@ -0,0 +1,28 @@
==============================================================================
POWER SIMULATION -- unmerge_d6_10 [metric: success RATE]
==============================================================================
model: behavior ~ stim + day + stim:day + (1|rat) (behavior = success RATE; day within-window)
observed groups: stim n=3, control n=2 nrep=120, alpha=0.05
ground truth: stim:day=+0.002261/day, rat SD=0.03997, residual SD=0.0557, days=5
true stim:day interaction = +0.002261 (100% of observed)
N/group | per-animal power | LME power
------------------------------------------
3 | 0.03 | 0.04 <- observed
5 | 0.03 | 0.06
8 | 0.05 | 0.05
12 | 0.08 | 0.07
16 | 0.05 | 0.07
24 | 0.06 | 0.06
true stim:day interaction = +0.00113 (50% of observed)
N/group | per-animal power | LME power
------------------------------------------
3 | 0.03 | 0.08 <- observed
5 | 0.07 | 0.07
8 | 0.05 | 0.04
12 | 0.06 | 0.07
16 | 0.07 | 0.07
24 | 0.06 | 0.04
Read the per-animal column as the honest power; LME is optimistic (obs-level DF).
analysis/matlab/variations/unmerge_d6_10/powersim.m
+35 -34
View File
@@ -1,45 +1,49 @@
% Variation power simulation -- Monte-Carlo power for the paper's stim x day % Variation power simulation -- Monte-Carlo power for the paper's stim x day
% interaction, using THIS folder's data as the ground truth. % interaction, using THIS folder's data as the ground truth, for BOTH metrics:
% metric = count : behavior = # successes -> power_result.txt
% metric = rate : behavior = success / attempts -> power_result_rate.txt
% %
% Ground truth: fitlme(behavior ~ stim + day + stim:day + (1|rat)) on data.csv % Ground truth: fitlme(behavior ~ stim + day + stim:day + (1|rat)) on data.csv
% (success COUNT; day within-window). Its fixed effects, per-rat intercept SD, % (day within-window). Its fixed effects, per-rat intercept SD, and residual SD
% and residual SD generate NREP synthetic datasets at each rats-per-group N and % generate NREP synthetic datasets at each rats-per-group N and each true-effect
% each true-effect multiplier (1 = observed slope, 0.5 = half). Each dataset is % multiplier (1 = observed, 0.5 = half). Each is scored at alpha=0.05 by:
% scored at alpha = 0.05 two ways: % per-animal : Welch t on per-rat behavior~day slopes (cluster-honest power)
% per-animal : Welch t on per-rat behavior~day slopes (cluster-honest -- the
% honest power, matching the random-slope / per-animal inference)
% LME : the fitlme stim:day p (observation-level DF -- optimistic) % LME : the fitlme stim:day p (observation-level DF -- optimistic)
% Writes power_result.txt beside this script. Run: matlab -batch "powersim" % Writes power_result[_rate].txt. Run: matlab -batch "powersim"
% (Copy of analysis/matlab/variation_power.m; see make_variation_power.m.) % (Copy of analysis/matlab/variation_power.m; see make_variation_power.m.)
here = fileparts(mfilename('fullpath')); here = fileparts(mfilename('fullpath'));
if isempty(here); here = pwd; end if isempty(here); here = pwd; end
vname = regexprep(here, '.*[/\\]', ''); vname = regexprep(here, '.*[/\\]', '');
D = readtable(fullfile(here, 'data.csv'), 'TextType', 'string'); D = readtable(fullfile(here, 'data.csv'), 'TextType', 'string');
localPower(D, 'count', here, vname);
localPower(D, 'rate', here, vname);
% ---------------------------------------------------------------- per metric
function localPower(D, metric, here, vname)
NS = [3 5 8 12 16 24]; EFFMULS = [1 0.5]; NREP = 120;
FORMULA = 'behavior ~ stim + day + stim:day + (1|rat)'; FORMULA = 'behavior ~ stim + day + stim:day + (1|rat)';
NS = [3 5 8 12 16 24]; if strcmp(metric, 'rate')
EFFMULS = [1 0.5]; D = D(D.total > 0, :); beh = D.success ./ D.total; mlabel = 'success RATE'; suffix = '_rate';
NREP = 120; else
beh = D.success; mlabel = '# successes (count)'; suffix = '';
warnState = warning('off', 'all'); end
rng(1); warnState = warning('off', 'all'); rng(1);
day0 = min(D.day); day0 = min(D.day);
tbl0 = table(D.success, D.day - day0, double(D.stim), categorical(D.subject), ... tbl0 = table(beh, D.day - day0, double(D.stim), categorical(D.subject), ...
'VariableNames', {'behavior', 'day', 'stim', 'rat'}); 'VariableNames', {'behavior', 'day', 'stim', 'rat'});
nStim = numel(unique(D.subject(D.stim == 1))); nStim = numel(unique(D.subject(D.stim == 1)));
nCtrl = numel(unique(D.subject(D.stim == 0))); nCtrl = numel(unique(D.subject(D.stim == 0)));
bar = repmat('=', 1, 78); bar = repmat('=', 1, 78);
s = sprintf('%s\nPOWER SIMULATION -- %s\n%s\n', bar, vname, bar); s = sprintf('%s\nPOWER SIMULATION -- %s [metric: %s]\n%s\n', bar, vname, mlabel, bar);
s = [s sprintf('model: %s (success COUNT; day within-window)\n', FORMULA)]; s = [s sprintf('model: %s (behavior = %s; day within-window)\n', FORMULA, mlabel)];
s = [s sprintf('observed groups: stim n=%d, control n=%d nrep=%d, alpha=0.05\n', nStim, nCtrl, NREP)]; s = [s sprintf('observed groups: stim n=%d, control n=%d nrep=%d, alpha=0.05\n', nStim, nCtrl, NREP)];
if nStim < 2 || nCtrl < 2 || numel(unique(tbl0.day)) < 2 if nStim < 2 || nCtrl < 2 || numel(unique(tbl0.day)) < 2
s = [s sprintf('\nInsufficient data for a power simulation (need >=2 animals/group and >=2 days).\n')]; s = [s sprintf('\nInsufficient data for a power simulation.\n')];
localFinish(s, here); warning(warnState); return localFinish(s, here, suffix); warning(warnState); return
end end
lme = fitlme(tbl0, FORMULA); lme = fitlme(tbl0, FORMULA);
@@ -49,14 +53,13 @@ bDay = be(strcmp(cn, 'day')); bInt = be(strcmp(cn, 'day:stim'));
psi = covarianceParameters(lme); sRat = sqrt(psi{1}); sRes = sqrt(lme.MSE); psi = covarianceParameters(lme); sRat = sqrt(psi{1}); sRes = sqrt(lme.MSE);
days = (0:max(tbl0.day))'; days = (0:max(tbl0.day))';
s = [s sprintf('ground truth: stim:day=%+.2f/day, rat SD=%.2f, residual SD=%.2f, days=%d\n', ... s = [s sprintf('ground truth: stim:day=%+.4g/day, rat SD=%.4g, residual SD=%.4g, days=%d\n', ...
bInt, sRat, sRes, numel(days))]; bInt, sRat, sRes, numel(days))];
for eMul = EFFMULS for eMul = EFFMULS
bI = bInt * eMul; bI = bInt * eMul;
s = [s sprintf('\n true stim:day interaction = %+.2f (%.0f%% of observed)\n', bI, eMul * 100)]; %#ok<AGROW> s = [s sprintf('\n true stim:day interaction = %+.4g (%.0f%% of observed)\n', bI, eMul * 100)];
s = [s sprintf(' %-8s | per-animal power | LME power\n', 'N/group')]; %#ok<AGROW> s = [s sprintf(' %-8s | per-animal power | LME power\n %s\n', 'N/group', repmat('-', 1, 42))];
s = [s sprintf(' %s\n', repmat('-', 1, 42))]; %#ok<AGROW>
for N = NS for N = NS
sigPA = 0; sigL = 0; sigPA = 0; sigL = 0;
for r = 1:NREP for r = 1:NREP
@@ -70,20 +73,18 @@ for eMul = EFFMULS
end end
star = ''; star = '';
if N == nStim || N == nCtrl; star = ' <- observed'; end if N == nStim || N == nCtrl; star = ' <- observed'; end
s = [s sprintf(' %-8d | %5.2f | %5.2f%s\n', N, sigPA / NREP, sigL / NREP, star)]; %#ok<AGROW> s = [s sprintf(' %-8d | %5.2f | %5.2f%s\n', N, sigPA / NREP, sigL / NREP, star)];
end end
end end
s = [s sprintf('\nRead the per-animal column as the honest power; LME is optimistic (obs-level DF).\n')];
s = [s sprintf(['\nRead the per-animal column as the honest power. At the observed N this study\n' ... localFinish(s, here, suffix);
'is typically underpowered; per-animal power reaches ~0.8 only at larger N.\n'])];
localFinish(s, here);
warning(warnState); warning(warnState);
end
% ---------------------------------------------------------------- helpers % ---------------------------------------------------------------- helpers
function localFinish(s, here) function localFinish(s, here, suffix)
fprintf('%s', s); fprintf('%s', s);
fid = fopen(fullfile(here, 'power_result.txt'), 'w'); fid = fopen(fullfile(here, ['power_result' suffix '.txt']), 'w');
fprintf(fid, '%s', s); fclose(fid); fprintf(fid, '%s', s); fclose(fid);
end end
analysis/matlab/variations/unmerge_d6_10/result.txt
+4 -4
View File
@@ -1,5 +1,5 @@
============================================================================== ==============================================================================
VARIATION: unmerge_d6_10 VARIATION: unmerge_d6_10 [metric: success COUNT]
============================================================================== ==============================================================================
model: behavior ~ stim + day + stim:day + (1|rat) (behavior = success COUNT) model: behavior ~ stim + day + stim:day + (1|rat) (behavior = success COUNT)
day = training day within window (0 = first analyzed day) day = training day within window (0 = first analyzed day)
@@ -60,9 +60,9 @@ effect t(df) / F(df1) p (resid) Satterthwaite: p (df)
stim x day (interaction) t(21)= 0.71 F(1)= 0.503 p=0.486 p=0.4864 (df=20) stim x day (interaction) t(21)= 0.71 F(1)= 0.503 p=0.486 p=0.4864 (df=20)
day (learning) t(21)= 1.24 F(1)= 1.547 p=0.2273 p=0.2279 (df=20) day (learning) t(21)= 1.24 F(1)= 1.547 p=0.2273 p=0.2279 (df=20)
stim (main, window start) t(21)= 1.73 F(1)= 3.008 p=0.09752 p=0.1098 (df=11) stim (main, window start) t(21)= 1.73 F(1)= 3.008 p=0.09752 p=0.1098 (df=11)
interaction 95% CI: [-3.41, +6.95] interaction 95% CI: [-3.414, +6.947]
HONEST LME (per-animal random slope, day|rat): interaction F(1,5.0)=0.40, p=0.5549 HONEST LME (per-animal random slope, day|rat): interaction F(1,5.0)=0.40, p=0.5549
(Satterthwaite DF ~= residual on this random-intercept model; the random-slope (Satterthwaite DF ~= residual on this random-intercept model; the random-slope
model above is the honest learning-rate test -- DF collapses toward the animal count.) model above is the honest learning-rate test -- DF collapses toward the animal count.)
INTERPRETATION: stim x day interaction n.s. -- slopes parallel (no differential learning rate) (p=0.486, slope diff=+1.77) INTERPRETATION: stim x day interaction n.s. -- slopes parallel (no differential learning rate) (p=0.486, slope diff=+1.767)
Paper (N=24): interaction t(227)=2.68, F(1)=7.12, p=0.008. Paper (N=24, count): interaction t(227)=2.68, F(1)=7.12, p=0.008.
analysis/matlab/variations/unmerge_d6_10/result_rate.txt
+68
View File
@@ -0,0 +1,68 @@
==============================================================================
VARIATION: unmerge_d6_10 [metric: success RATE (success/attempts)]
==============================================================================
model: behavior ~ stim + day + stim:day + (1|rat) (behavior = success RATE (success/attempts))
day = training day within window (0 = first analyzed day)
treatment (stim=1): Electrode-Box-B2
control (stim=0): Electrode-Box-A2
N = 5 rats, 25 sessions raw day coverage: treat 6..10, control 6..10
(equal day coverage over this window)
==============================================================================
FULL MODEL SUMMARY -- fitlme
==============================================================================
Linear mixed-effects model fit by ML
Model information:
Number of observations 25
Fixed effects coefficients 4
Random effects coefficients 5
Covariance parameters 2
Formula:
behavior ~ 1 + day*stim + (1 | rat)
Model fit statistics:
AIC BIC LogLikelihood Deviance
-55.071 -47.758 33.535 -67.071
Fixed effects coefficients (95% CIs):
Name Estimate SE tStat DF pValue
{'(Intercept)'} 0.52619 0.041587 12.653 21 2.7251e-11
{'day' } 0.0099391 0.012456 0.79797 21 0.43382
{'stim' } 0.11742 0.053689 2.187 21 0.040203
{'day:stim' } 0.0022606 0.01608 0.14058 21 0.88954
Lower Upper
0.4397 0.61267
-0.015964 0.035842
0.0057634 0.22907
-0.03118 0.035701
Random effects covariance parameters (95% CIs):
Group: rat (5 Levels)
Name1 Name2 Type Estimate
{'(Intercept)'} {'(Intercept)'} {'std'} 0.039966
Lower Upper
0.016761 0.095298
Group: Error
Name Estimate Lower Upper
{'Res Std'} 0.055703 0.040859 0.075939
effect t(df) / F(df1) p (resid) Satterthwaite: p (df)
----------------------------------------------------------------------------
stim x day (interaction) t(21)= 0.14 F(1)= 0.020 p=0.8895 p=0.8896 (df=20)
day (learning) t(21)= 0.80 F(1)= 0.637 p=0.4338 p=0.4343 (df=20)
stim (main, window start) t(21)= 2.19 F(1)= 4.783 p=0.0402 p=0.05066 (df=11)
interaction 95% CI: [-0.03118, +0.0357]
HONEST LME (per-animal random slope, day|rat): interaction F(1,5.0)=0.01, p=0.9219
(Satterthwaite DF ~= residual on this random-intercept model; the random-slope
model above is the honest learning-rate test -- DF collapses toward the animal count.)
INTERPRETATION: stim x day interaction n.s. -- slopes parallel (no differential learning rate) (p=0.8895, slope diff=+0.002261)
Paper (N=24, count): interaction t(227)=2.68, F(1)=7.12, p=0.008.
analysis/matlab/variations/unmerge_d6_13/analyze.m
+63 -39
View File
@@ -1,30 +1,58 @@
% Variation analysis -- the paper's linear mixed model on the successful-reach % Variation analysis -- the paper's linear mixed model on this folder's data,
% COUNT, fit on this folder's curated data subset. % for BOTH metrics:
% metric = count : behavior = # successes -> result.txt
% metric = rate : behavior = success / attempts -> result_rate.txt
% (rate uses only sessions with attempts > 0)
% %
% model: behavior ~ stim + day + stim:day + (1|rat) % model: behavior ~ stim + day + stim:day + (1|rat)
% behavior = successful reaches (count per session) % stim = 1 treatment / 0 control; day = training day within window (0 =
% stim = 1 for the treatment group(s), 0 for the control group(s) % first analyzed day); rat = subject (random intercept).
% day = training day within this window (0 = first analyzed day) % For the interaction we report residual DF, Satterthwaite DF, and the honest
% rat = subject (random intercept) % per-animal random-slope test. Self-contained: reads data.csv beside this
% % script. Run headless with: matlab -batch "analyze"
% Self-contained: reads data.csv beside this script and writes result.txt. % (Copy of analysis/matlab/variation_analyze.m; see make_variations.m.)
% Run headless from this folder with: matlab -batch "analyze"
% (This is a copy of analysis/matlab/variation_analyze.m; see make_variations.m.)
here = fileparts(mfilename('fullpath')); here = fileparts(mfilename('fullpath'));
if isempty(here); here = pwd; end if isempty(here); here = pwd; end
vname = regexprep(here, '.*[/\\]', ''); % folder name = variation id vname = regexprep(here, '.*[/\\]', '');
D = readtable(fullfile(here, 'data.csv'), 'TextType', 'string'); D = readtable(fullfile(here, 'data.csv'), 'TextType', 'string');
tbl = table(D.success, D.day - min(D.day), double(D.stim), categorical(D.subject), ... Rc = localAnalyze(D, 'count', here, vname);
Rr = localAnalyze(D, 'rate', here, vname);
% Machine-readable handoff for SUMMARY.csv (count drives it; rate appended).
VARRESULT = struct('name', vname, 'nRats', Rc.nRats, 'nObs', Rc.nObs, ...
'interP', Rc.interP, 'interEst', Rc.interEst, ...
'interPsatt', Rc.interPsatt, 'interPrs', Rc.interPrs, ...
'stimP', Rc.stimP, 'dayP', Rc.dayP, 'covEqual', Rc.covEqual, ...
'interPrate', Rr.interP, 'interEstRate', Rr.interEst, 'interPrsRate', Rr.interPrs);
% ------------------------------------------------------------------ helper
function R = localAnalyze(D, metric, here, vname)
if strcmp(metric, 'rate')
D = D(D.total > 0, :);
beh = D.success ./ D.total;
mlabel = 'success RATE (success/attempts)'; suffix = '_rate';
else
beh = D.success;
mlabel = 'success COUNT'; suffix = '';
end
R = struct('interP', NaN, 'interEst', NaN, 'interPsatt', NaN, 'interPrs', NaN, ...
'stimP', NaN, 'dayP', NaN, 'nRats', numel(unique(D.subject)), ...
'nObs', height(D), 'covEqual', false);
if numel(unique(D.stim)) < 2 || numel(unique(D.day)) < 2
localWrite(sprintf('VARIATION: %s [metric: %s]\nInsufficient data for this metric.\n', ...
vname, mlabel), here, suffix);
return
end
tbl = table(beh, D.day - min(D.day), double(D.stim), categorical(D.subject), ...
'VariableNames', {'behavior', 'day', 'stim', 'rat'}); 'VariableNames', {'behavior', 'day', 'stim', 'rat'});
m = fitlme(tbl, 'behavior ~ stim + day + stim:day + (1|rat)'); m = fitlme(tbl, 'behavior ~ stim + day + stim:day + (1|rat)');
C = m.Coefficients; A = anova(m); ci = coefCI(m); C = m.Coefficients; A = anova(m); ci = coefCI(m);
As = anova(m, 'DFMethod', 'satterthwaite'); % Satterthwaite denominator DF As = anova(m, 'DFMethod', 'satterthwaite');
% Honest test: refit with a per-animal random SLOPE so the interaction DF
% collapses toward the animal count (guarded -- may not converge in short windows).
rsP = NaN; rsDf = NaN; rsF = NaN; rsOk = false; rsP = NaN; rsDf = NaN; rsF = NaN; rsOk = false;
wst = warning('off', 'all'); wst = warning('off', 'all');
try try
@@ -43,6 +71,7 @@ row = @(nm, t) sprintf('%-26s t(%d)=%6.2f F(%d)=%7.3f p=%.4g p=%.4g (df=%.0f
C.DF(gi(t)), C.tStat(gi(t)), A.DF1(ga(t)), A.FStat(ga(t)), C.pValue(gi(t)), ... C.DF(gi(t)), C.tStat(gi(t)), A.DF1(ga(t)), A.FStat(ga(t)), C.pValue(gi(t)), ...
As.pValue(gs(t)), As.DF2(gs(t))); As.pValue(gs(t)), As.DF2(gs(t)));
ii = gi('day:stim'); pI = C.pValue(ii); eI = C.Estimate(ii);
maxT = max(D.day(D.stim == 1)); minT = min(D.day(D.stim == 1)); maxT = max(D.day(D.stim == 1)); minT = min(D.day(D.stim == 1));
maxC = max(D.day(D.stim == 0)); minC = min(D.day(D.stim == 0)); maxC = max(D.day(D.stim == 0)); minC = min(D.day(D.stim == 0));
if abs(maxT - maxC) > 2 if abs(maxT - maxC) > 2
@@ -50,20 +79,14 @@ if abs(maxT - maxC) > 2
else else
cov = '(equal day coverage over this window)'; cov = '(equal day coverage over this window)';
end end
if pI >= 0.05; verdict = 'n.s. -- slopes parallel (no differential learning rate)';
ii = gi('day:stim'); pI = C.pValue(ii); eI = C.Estimate(ii); elseif eI > 0; verdict = 'SIGNIFICANT positive -- treatment improves FASTER (benefit accumulates)';
if pI >= 0.05 else; verdict = 'SIGNIFICANT negative -- treatment improves SLOWER (groups converge)'; end
verdict = 'n.s. -- slopes parallel (no differential learning rate)';
elseif eI > 0
verdict = 'SIGNIFICANT positive -- treatment improves FASTER (benefit accumulates)';
else
verdict = 'SIGNIFICANT negative -- treatment improves SLOWER (groups converge)';
end
bar = repmat('=', 1, 78); bar = repmat('=', 1, 78);
raw = regexprep(evalc('disp(m)'), '</?strong>', ''); raw = regexprep(evalc('disp(m)'), '</?strong>', '');
s = sprintf('%s\nVARIATION: %s\n%s\n', bar, vname, bar); s = sprintf('%s\nVARIATION: %s [metric: %s]\n%s\n', bar, vname, mlabel, bar);
s = [s sprintf('model: behavior ~ stim + day + stim:day + (1|rat) (behavior = success COUNT)\n')]; s = [s sprintf('model: behavior ~ stim + day + stim:day + (1|rat) (behavior = %s)\n', mlabel)];
s = [s sprintf('day = training day within window (0 = first analyzed day)\n')]; s = [s sprintf('day = training day within window (0 = first analyzed day)\n')];
s = [s sprintf('treatment (stim=1): %s\n', strjoin(cellstr(unique(D.group(D.stim == 1))), ', '))]; s = [s sprintf('treatment (stim=1): %s\n', strjoin(cellstr(unique(D.group(D.stim == 1))), ', '))];
s = [s sprintf('control (stim=0): %s\n', strjoin(cellstr(unique(D.group(D.stim == 0))), ', '))]; s = [s sprintf('control (stim=0): %s\n', strjoin(cellstr(unique(D.group(D.stim == 0))), ', '))];
@@ -74,7 +97,7 @@ s = [s sprintf('%-26s %-18s %-12s %s\n%s\n', 'effect', 't(df) / F(df1)', 'p (res
s = [s row('stim x day (interaction)', 'day:stim')]; s = [s row('stim x day (interaction)', 'day:stim')];
s = [s row('day (learning)', 'day')]; s = [s row('day (learning)', 'day')];
s = [s row('stim (main, window start)', 'stim')]; s = [s row('stim (main, window start)', 'stim')];
s = [s sprintf('interaction 95%% CI: [%+.2f, %+.2f]\n', ci(ii, 1), ci(ii, 2))]; s = [s sprintf('interaction 95%% CI: [%+.4g, %+.4g]\n', ci(ii, 1), ci(ii, 2))];
if rsOk if rsOk
s = [s sprintf('HONEST LME (per-animal random slope, day|rat): interaction F(1,%.1f)=%.2f, p=%.4g\n', rsDf, rsF, rsP)]; s = [s sprintf('HONEST LME (per-animal random slope, day|rat): interaction F(1,%.1f)=%.2f, p=%.4g\n', rsDf, rsF, rsP)];
else else
@@ -82,17 +105,18 @@ else
end end
s = [s sprintf([' (Satterthwaite DF ~= residual on this random-intercept model; the random-slope\n' ... s = [s sprintf([' (Satterthwaite DF ~= residual on this random-intercept model; the random-slope\n' ...
' model above is the honest learning-rate test -- DF collapses toward the animal count.)\n'])]; ' model above is the honest learning-rate test -- DF collapses toward the animal count.)\n'])];
s = [s sprintf('INTERPRETATION: stim x day interaction %s (p=%.4g, slope diff=%+.2f)\n', verdict, pI, eI)]; s = [s sprintf('INTERPRETATION: stim x day interaction %s (p=%.4g, slope diff=%+.4g)\n', verdict, pI, eI)];
s = [s sprintf('Paper (N=24): interaction t(227)=2.68, F(1)=7.12, p=0.008.\n')]; s = [s sprintf('Paper (N=24, count): interaction t(227)=2.68, F(1)=7.12, p=0.008.\n')];
localWrite(s, here, suffix);
R = struct('interP', pI, 'interEst', eI, 'interPsatt', As.pValue(gs('day:stim')), ...
'interPrs', rsP, 'stimP', C.pValue(gi('stim')), 'dayP', C.pValue(gi('day')), ...
'nRats', numel(unique(D.subject)), 'nObs', height(D), 'covEqual', abs(maxT - maxC) <= 2);
end
function localWrite(s, here, suffix)
fprintf('%s', s); fprintf('%s', s);
fid = fopen(fullfile(here, 'result.txt'), 'w'); fid = fopen(fullfile(here, ['result' suffix '.txt']), 'w');
fprintf(fid, '%s', s); fprintf(fid, '%s', s); fclose(fid);
fclose(fid); end
% Machine-readable handoff for the summary table (see make_variations.m).
VARRESULT = struct('name', vname, 'nRats', numel(unique(D.subject)), ...
'nObs', height(D), 'interP', pI, 'interEst', eI, ...
'interPsatt', As.pValue(gs('day:stim')), 'interPrs', rsP, ...
'stimP', C.pValue(gi('stim')), 'dayP', C.pValue(gi('day')), ...
'covEqual', abs(maxT - maxC) <= 2);
analysis/matlab/variations/unmerge_d6_13/learning_curve.png
BIN
View File
Binary file not shown.
Side by Side Swipe Overlay

Before

Width:  |  Height:  |  Size: 41 KiB

After

Width:  |  Height:  |  Size: 41 KiB

analysis/matlab/variations/unmerge_d6_13/learning_curve_rate.png
BIN
View File
Binary file not shown.
Side by Side

After

Width:  |  Height:  |  Size: 39 KiB

analysis/matlab/variations/unmerge_d6_13/logpower_result.txt
+4 -5
View File
@@ -1,7 +1,7 @@
============================================================================== ==============================================================================
LOG-DAY MODEL + COHEN'S f + POWER -- unmerge_d6_13 LOG-DAY MODEL + COHEN'S f + POWER -- unmerge_d6_13 [metric: # successes (count)]
============================================================================== ==============================================================================
model: behavior ~ stim + log(day) + stim:log(day) + (1|rat) (success COUNT) model: behavior ~ stim + log(day) + stim:log(day) + (1|rat) (behavior = # successes (count))
log(day) uses 1-indexed training day (our day 0 = paper "Day 1") log(day) uses 1-indexed training day (our day 0 = paper "Day 1")
observed groups: stim n=3, control n=2 nrep=120, alpha=0.05 observed groups: stim n=3, control n=2 nrep=120, alpha=0.05
@@ -10,7 +10,7 @@ stim x log(day) interaction: F(1,30)=3.007 p(resid)=0.09315 p(Satt)=0.0931 (df
honest per-animal random slope (log-day): F(1,12.8)=2.47 p=0.1406 honest per-animal random slope (log-day): F(1,12.8)=2.47 p=0.1406
Cohen's f (interaction, partial eta^2=0.025) = 0.161 (small-medium; f: .10 small, .25 medium, .40 large) Cohen's f (interaction, partial eta^2=0.025) = 0.161 (small-medium; f: .10 small, .25 medium, .40 large)
--- power simulation (log-day ground truth: stim:logday=+22.97, ratSD=7.14, resSD=7.81) --- --- power simulation (log-day ground truth: stim:logday=+22.97, ratSD=7.136, resSD=7.808) ---
true stim:log(day) = +22.97 (100% of observed) true stim:log(day) = +22.97 (100% of observed)
N/group | per-animal power | LME power N/group | per-animal power | LME power
@@ -32,5 +32,4 @@ Cohen's f (interaction, partial eta^2=0.025) = 0.161 (small-medium; f: .10 smal
16 | 0.71 | 0.74 16 | 0.71 | 0.74
24 | 0.90 | 0.93 24 | 0.90 | 0.93
Read the per-animal column as the honest power; the LME column matches the Read per-animal as the honest power; LME matches the paper's power code (optimistic).
paper's power code (anova interaction p, observation-level DF) and is optimistic.
analysis/matlab/variations/unmerge_d6_13/logpower_result_rate.txt
+35
View File
@@ -0,0 +1,35 @@
==============================================================================
LOG-DAY MODEL + COHEN'S f + POWER -- unmerge_d6_13 [metric: success RATE]
==============================================================================
model: behavior ~ stim + log(day) + stim:log(day) + (1|rat) (behavior = success RATE)
log(day) uses 1-indexed training day (our day 0 = paper "Day 1")
observed groups: stim n=3, control n=2 nrep=120, alpha=0.05
--- fitted on real data ---
stim x log(day) interaction: F(1,30)=1.838 p(resid)=0.1853 p(Satt)=0.1853 (df=30)
honest per-animal random slope (log-day): F(1,3.6)=0.01 p=0.9233
Cohen's f (interaction, partial eta^2=0.016) = 0.129 (small-medium; f: .10 small, .25 medium, .40 large)
--- power simulation (log-day ground truth: stim:logday=+0.1165, ratSD=0.04832, resSD=0.05061) ---
true stim:log(day) = +0.1165 (100% of observed)
N/group | per-animal power | LME power
------------------------------------------
3 | 0.28 | 0.46 <- observed
5 | 0.58 | 0.70
8 | 0.78 | 0.83
12 | 0.90 | 0.89
16 | 0.99 | 1.00
24 | 1.00 | 1.00
true stim:log(day) = +0.05823 (50% of observed)
N/group | per-animal power | LME power
------------------------------------------
3 | 0.08 | 0.17 <- observed
5 | 0.18 | 0.27
8 | 0.23 | 0.32
12 | 0.34 | 0.38
16 | 0.47 | 0.51
24 | 0.79 | 0.82
Read per-animal as the honest power; LME matches the paper's power code (optimistic).
analysis/matlab/variations/unmerge_d6_13/logpowersim.m
+38 -39
View File
@@ -1,50 +1,52 @@
% Variation log-day analysis + Cohen's f + power simulation. % Variation log-day analysis + Cohen's f + power simulation, for BOTH metrics:
% metric = count : behavior = # successes -> logpower_result.txt
% metric = rate : behavior = success / attempts -> logpower_result_rate.txt
% %
% The paper's power code models behavior against LOG training day, not raw day: % The paper's power code models behavior against LOG training day:
% behavior ~ stim + log(day) + stim:log(day) + (1|rat). % behavior ~ stim + log(day) + stim:log(day) + (1|rat).
% Their day is 1-indexed (1..10); our data.csv day is 0-indexed (day 0 = paper % Their day is 1-indexed; our data.csv day is 0-indexed, so log(day + 1)
% "Day 1"), so log(day + 1) reproduces their transform exactly. % reproduces their transform (our day 0 = paper "Day 1"). For each metric this
% % refits that model, reports the interaction (residual / Satterthwaite / honest
% This script (a) refits that log-day model on data.csv, (b) reports the % per-animal random-slope DF) and Cohen's f (partial-eta^2 effect size), then
% interaction (residual DF, Satterthwaite DF, and the honest per-animal % runs the Monte-Carlo power sim (per-animal cluster-honest + LME power).
% random-slope test) and Cohen's f -- the partial-eta^2 effect size of the % Run: matlab -batch "logpowersim"
% interaction, var(fitted_full) - var(fitted_no_interaction) over var(behavior)
% -- and (c) runs the Monte-Carlo power simulation on the log-day model,
% scoring per-animal (cluster-honest) and LME power across N.
% Writes logpower_result.txt. Run: matlab -batch "logpowersim"
% (Copy of analysis/matlab/variation_logpower.m; see make_variation_logpower.m.) % (Copy of analysis/matlab/variation_logpower.m; see make_variation_logpower.m.)
here = fileparts(mfilename('fullpath')); here = fileparts(mfilename('fullpath'));
if isempty(here); here = pwd; end if isempty(here); here = pwd; end
vname = regexprep(here, '.*[/\\]', ''); vname = regexprep(here, '.*[/\\]', '');
D = readtable(fullfile(here, 'data.csv'), 'TextType', 'string'); D = readtable(fullfile(here, 'data.csv'), 'TextType', 'string');
FORMULA = 'behavior ~ stim + day + stim:day + (1|rat)'; % 'day' column = log(day+1)
NS = [3 5 8 12 16 24];
EFFMULS = [1 0.5];
NREP = 120;
warnState = warning('off', 'all'); localLogPower(D, 'count', here, vname);
rng(1); localLogPower(D, 'rate', here, vname);
logday = log(D.day + 1); % 0-indexed day -> their log(1-indexed day) % ---------------------------------------------------------------- per metric
tbl0 = table(D.success, logday, double(D.stim), categorical(D.subject), ... function localLogPower(D, metric, here, vname)
NS = [3 5 8 12 16 24]; EFFMULS = [1 0.5]; NREP = 120;
FORMULA = 'behavior ~ stim + day + stim:day + (1|rat)'; % 'day' = log(day+1)
if strcmp(metric, 'rate')
D = D(D.total > 0, :); beh = D.success ./ D.total; mlabel = 'success RATE'; suffix = '_rate';
else
beh = D.success; mlabel = '# successes (count)'; suffix = '';
end
warnState = warning('off', 'all'); rng(1);
logday = log(D.day + 1);
tbl0 = table(beh, logday, double(D.stim), categorical(D.subject), ...
'VariableNames', {'behavior', 'day', 'stim', 'rat'}); 'VariableNames', {'behavior', 'day', 'stim', 'rat'});
nStim = numel(unique(D.subject(D.stim == 1))); nStim = numel(unique(D.subject(D.stim == 1)));
nCtrl = numel(unique(D.subject(D.stim == 0))); nCtrl = numel(unique(D.subject(D.stim == 0)));
bar = repmat('=', 1, 78); bar = repmat('=', 1, 78);
s = sprintf('%s\nLOG-DAY MODEL + COHEN''S f + POWER -- %s\n%s\n', bar, vname, bar); s = sprintf('%s\nLOG-DAY MODEL + COHEN''S f + POWER -- %s [metric: %s]\n%s\n', bar, vname, mlabel, bar);
s = [s sprintf('model: behavior ~ stim + log(day) + stim:log(day) + (1|rat) (success COUNT)\n')]; s = [s sprintf('model: behavior ~ stim + log(day) + stim:log(day) + (1|rat) (behavior = %s)\n', mlabel)];
s = [s sprintf('log(day) uses 1-indexed training day (our day 0 = paper "Day 1")\n')]; s = [s sprintf('log(day) uses 1-indexed training day (our day 0 = paper "Day 1")\n')];
s = [s sprintf('observed groups: stim n=%d, control n=%d nrep=%d, alpha=0.05\n', nStim, nCtrl, NREP)]; s = [s sprintf('observed groups: stim n=%d, control n=%d nrep=%d, alpha=0.05\n', nStim, nCtrl, NREP)];
if nStim < 2 || nCtrl < 2 || numel(unique(tbl0.day)) < 2 if nStim < 2 || nCtrl < 2 || numel(unique(tbl0.day)) < 2
s = [s sprintf('\nInsufficient data for this analysis (need >=2 animals/group and >=2 days).\n')]; s = [s sprintf('\nInsufficient data for this analysis.\n')];
localFinish(s, here); warning(warnState); return localFinish(s, here, suffix); warning(warnState); return
end end
% ---- fitted model on the real data ----
full = fitlme(tbl0, FORMULA); full = fitlme(tbl0, FORMULA);
An = anova(full); Asatt = anova(full, 'DFMethod', 'satterthwaite'); An = anova(full); Asatt = anova(full, 'DFMethod', 'satterthwaite');
ii = strcmp(An.Term, 'day:stim'); is = strcmp(Asatt.Term, 'day:stim'); ii = strcmp(An.Term, 'day:stim'); is = strcmp(Asatt.Term, 'day:stim');
@@ -52,7 +54,6 @@ reduced = fitlme(tbl0, 'behavior ~ stim + day + (1|rat)');
eta2part = max((var(fitted(full)) - var(fitted(reduced))) / var(tbl0.behavior), 0); eta2part = max((var(fitted(full)) - var(fitted(reduced))) / var(tbl0.behavior), 0);
cohenf = sqrt(eta2part / (1 - eta2part)); cohenf = sqrt(eta2part / (1 - eta2part));
% honest per-animal random-slope interaction
rsP = NaN; rsDf = NaN; rsF = NaN; rsOk = false; rsP = NaN; rsDf = NaN; rsF = NaN; rsOk = false;
try try
mr = fitlme(tbl0, 'behavior ~ stim + day + stim:day + (day|rat)'); mr = fitlme(tbl0, 'behavior ~ stim + day + stim:day + (day|rat)');
@@ -76,19 +77,18 @@ end
s = [s sprintf('Cohen''s f (interaction, partial eta^2=%.3f) = %.3f (%s; f: .10 small, .25 medium, .40 large)\n', ... s = [s sprintf('Cohen''s f (interaction, partial eta^2=%.3f) = %.3f (%s; f: .10 small, .25 medium, .40 large)\n', ...
eta2part, cohenf, mag)]; eta2part, cohenf, mag)];
% ---- power simulation on the log-day ground truth ----
cn = full.CoefficientNames; be = full.fixedEffects; cn = full.CoefficientNames; be = full.fixedEffects;
b0 = be(strcmp(cn, '(Intercept)')); bStim = be(strcmp(cn, 'stim')); b0 = be(strcmp(cn, '(Intercept)')); bStim = be(strcmp(cn, 'stim'));
bDay = be(strcmp(cn, 'day')); bInt = be(strcmp(cn, 'day:stim')); bDay = be(strcmp(cn, 'day')); bInt = be(strcmp(cn, 'day:stim'));
psi = covarianceParameters(full); sRat = sqrt(psi{1}); sRes = sqrt(full.MSE); psi = covarianceParameters(full); sRat = sqrt(psi{1}); sRes = sqrt(full.MSE);
days = unique(tbl0.day); % the log(day) grid days = unique(tbl0.day);
s = [s sprintf('\n--- power simulation (log-day ground truth: stim:logday=%+.2f, ratSD=%.2f, resSD=%.2f) ---\n', ... s = [s sprintf('\n--- power simulation (log-day ground truth: stim:logday=%+.4g, ratSD=%.4g, resSD=%.4g) ---\n', ...
bInt, sRat, sRes)]; bInt, sRat, sRes)];
for eMul = EFFMULS for eMul = EFFMULS
bI = bInt * eMul; bI = bInt * eMul;
s = [s sprintf('\n true stim:log(day) = %+.2f (%.0f%% of observed)\n', bI, eMul * 100)]; %#ok<AGROW> s = [s sprintf('\n true stim:log(day) = %+.4g (%.0f%% of observed)\n', bI, eMul * 100)];
s = [s sprintf(' %-8s | per-animal power | LME power\n %s\n', 'N/group', repmat('-', 1, 42))]; %#ok<AGROW> s = [s sprintf(' %-8s | per-animal power | LME power\n %s\n', 'N/group', repmat('-', 1, 42))];
for N = NS for N = NS
sigPA = 0; sigL = 0; sigPA = 0; sigL = 0;
for r = 1:NREP for r = 1:NREP
@@ -102,19 +102,18 @@ for eMul = EFFMULS
end end
star = ''; star = '';
if N == nStim || N == nCtrl; star = ' <- observed'; end if N == nStim || N == nCtrl; star = ' <- observed'; end
s = [s sprintf(' %-8d | %5.2f | %5.2f%s\n', N, sigPA / NREP, sigL / NREP, star)]; %#ok<AGROW> s = [s sprintf(' %-8d | %5.2f | %5.2f%s\n', N, sigPA / NREP, sigL / NREP, star)];
end end
end end
s = [s sprintf(['\nRead the per-animal column as the honest power; the LME column matches the\n' ... s = [s sprintf('\nRead per-animal as the honest power; LME matches the paper''s power code (optimistic).\n')];
'paper''s power code (anova interaction p, observation-level DF) and is optimistic.\n'])]; localFinish(s, here, suffix);
localFinish(s, here);
warning(warnState); warning(warnState);
end
% ---------------------------------------------------------------- helpers % ---------------------------------------------------------------- helpers
function localFinish(s, here) function localFinish(s, here, suffix)
fprintf('%s', s); fprintf('%s', s);
fid = fopen(fullfile(here, 'logpower_result.txt'), 'w'); fid = fopen(fullfile(here, ['logpower_result' suffix '.txt']), 'w');
fprintf(fid, '%s', s); fclose(fid); fprintf(fid, '%s', s); fclose(fid);
end end
analysis/matlab/variations/unmerge_d6_13/plotcurve.m
+29 -19
View File
@@ -1,11 +1,12 @@
% Variation learning-curve plot, in the style of the paper: % Variation learning-curve plots, in the style of the paper:
% "Lines indicate mean (and SEM) across animals in the anodal (red) and % "Lines indicate mean (and SEM) across animals in the anodal (red) and
% control (blue) groups." % control (blue) groups."
% Plots mean +/- SEM successful reaches per training day for the treatment / % Produces TWO figures from this folder's data.csv:
% anodal group (stim = 1, red) and the control group (stim = 0, blue), reading % learning_curve.png # successes (count) per training day
% this folder's data.csv and saving learning_curve.png. The per-group N is read % learning_curve_rate.png success rate (success/attempts) per training day
% from the data (each variation pools different groups), so the legend shows the % anodal / treatment = stim 1 (red); control = stim 0 (blue). Per-group N is
% actual counts. Training day is 1-indexed (our day 0 = the paper's "Day 1"). % read from the data. Training day is 1-indexed (our day 0 = paper "Day 1") and
% the x-axis tick labels are drawn vertically.
% Run: matlab -batch "plotcurve" % Run: matlab -batch "plotcurve"
% (Copy of analysis/matlab/variation_plot.m; see make_variation_plot.m.) % (Copy of analysis/matlab/variation_plot.m; see make_variation_plot.m.)
@@ -14,15 +15,20 @@ if isempty(here); here = pwd; end
vname = regexprep(here, '.*[/\\]', ''); vname = regexprep(here, '.*[/\\]', '');
D = readtable(fullfile(here, 'data.csv'), 'TextType', 'string'); D = readtable(fullfile(here, 'data.csv'), 'TextType', 'string');
days = unique(D.day); % 0-indexed days = unique(D.day);
xd = days + 1; % plot as 1-indexed training day (paper axis) xd = days + 1; % plot as 1-indexed training day (paper axis)
red = [0.85 0.10 0.10]; red = [0.85 0.10 0.10];
blue = [0.10 0.30 0.85]; blue = [0.10 0.30 0.85];
[Ma, Sa, na] = localCurve(D, 1, days); % anodal / treatment (stim = 1) localPlot(D, days, xd, 'count', '# successes', ...
[Mc, Sc, nc] = localCurve(D, 0, days); % control (stim = 0) fullfile(here, 'learning_curve.png'), vname, red, blue);
localPlot(D, days, xd, 'rate', 'success rate', ...
fullfile(here, 'learning_curve_rate.png'), vname, red, blue);
% ------------------------------------------------------------------ helpers
function localPlot(D, days, xd, metric, ylab, outFile, vname, red, blue)
[Ma, Sa, na] = localCurve(D, 1, days, metric); % anodal / treatment
[Mc, Sc, nc] = localCurve(D, 0, days, metric); % control
fig = figure('Visible', 'off', 'Color', 'w', 'Position', [100 100 560 460]); fig = figure('Visible', 'off', 'Color', 'w', 'Position', [100 100 560 460]);
hold on hold on
e1 = errorbar(xd, Ma, Sa, '-o', 'Color', red, 'MarkerFaceColor', red, 'LineWidth', 2); e1 = errorbar(xd, Ma, Sa, '-o', 'Color', red, 'MarkerFaceColor', red, 'LineWidth', 2);
@@ -30,26 +36,30 @@ e2 = errorbar(xd, Mc, Sc, '-o', 'Color', blue, 'MarkerFaceColor', blue, 'LineWid
hold off hold off
legend([e1 e2], {sprintf('anodal, N = %d', na), sprintf('control, N = %d', nc)}, ... legend([e1 e2], {sprintf('anodal, N = %d', na), sprintf('control, N = %d', nc)}, ...
'Location', 'northwest', 'Box', 'off'); 'Location', 'northwest', 'Box', 'off');
xlabel('training day'); xlabel('training day'); ylabel(ylab);
ylabel('# successes');
title(vname, 'Interpreter', 'none'); title(vname, 'Interpreter', 'none');
set(gca, 'XTick', xd, 'FontName', 'Arial', 'FontSize', 13, 'LineWidth', 1.5, 'Box', 'off'); set(gca, 'XTick', xd, 'FontName', 'Arial', 'FontSize', 13, 'LineWidth', 1.5, 'Box', 'off');
xtickangle(90); % vertical x-axis tick labels
outFile = fullfile(here, 'learning_curve.png');
exportgraphics(fig, outFile, 'Resolution', 150); exportgraphics(fig, outFile, 'Resolution', 150);
close(fig); close(fig);
fprintf('%s: wrote learning_curve.png (anodal N=%d, control N=%d)\n', vname, na, nc); fprintf('%s: wrote %s (anodal N=%d, control N=%d)\n', vname, outFile, na, nc);
end
% ------------------------------------------------------------------ helper function [M, S, n] = localCurve(D, stimVal, days, metric)
function [M, S, n] = localCurve(D, stimVal, days) %LOCALCURVE Per-day mean and SEM across the animals in a group, for a metric.
%LOCALCURVE Per-day mean and SEM of successes across the animals in a group.
subs = unique(D.subject(D.stim == stimVal)); subs = unique(D.subject(D.stim == stimVal));
n = numel(subs); n = numel(subs);
X = nan(numel(days), n); X = nan(numel(days), n);
for j = 1:n for j = 1:n
for i = 1:numel(days) for i = 1:numel(days)
r = D.subject == subs(j) & D.day == days(i); r = D.subject == subs(j) & D.day == days(i);
if any(r); X(i, j) = mean(D.success(r)); end if ~any(r); continue; end
if strcmp(metric, 'rate')
tot = sum(D.total(r));
if tot > 0; X(i, j) = sum(D.success(r)) / tot; end
else
X(i, j) = mean(D.success(r));
end
end end
end end
M = mean(X, 2, 'omitnan'); M = mean(X, 2, 'omitnan');
analysis/matlab/variations/unmerge_d6_13/power_result.txt
+6 -7
View File
@@ -1,11 +1,11 @@
============================================================================== ==============================================================================
POWER SIMULATION -- unmerge_d6_13 POWER SIMULATION -- unmerge_d6_13 [metric: # successes (count)]
============================================================================== ==============================================================================
model: behavior ~ stim + day + stim:day + (1|rat) (success COUNT; day within-window) model: behavior ~ stim + day + stim:day + (1|rat) (behavior = # successes (count); day within-window)
observed groups: stim n=3, control n=2 nrep=120, alpha=0.05 observed groups: stim n=3, control n=2 nrep=120, alpha=0.05
ground truth: stim:day=+2.27/day, rat SD=7.12, residual SD=7.73, days=8 ground truth: stim:day=+2.267/day, rat SD=7.117, residual SD=7.733, days=8
true stim:day interaction = +2.27 (100% of observed) true stim:day interaction = +2.267 (100% of observed)
N/group | per-animal power | LME power N/group | per-animal power | LME power
------------------------------------------ ------------------------------------------
3 | 0.46 | 0.62 <- observed 3 | 0.46 | 0.62 <- observed
@@ -15,7 +15,7 @@ ground truth: stim:day=+2.27/day, rat SD=7.12, residual SD=7.73, days=8
16 | 1.00 | 1.00 16 | 1.00 | 1.00
24 | 1.00 | 1.00 24 | 1.00 | 1.00
true stim:day interaction = +1.13 (50% of observed) true stim:day interaction = +1.133 (50% of observed)
N/group | per-animal power | LME power N/group | per-animal power | LME power
------------------------------------------ ------------------------------------------
3 | 0.12 | 0.20 <- observed 3 | 0.12 | 0.20 <- observed
@@ -25,5 +25,4 @@ ground truth: stim:day=+2.27/day, rat SD=7.12, residual SD=7.73, days=8
16 | 0.72 | 0.74 16 | 0.72 | 0.74
24 | 0.91 | 0.93 24 | 0.91 | 0.93
Read the per-animal column as the honest power. At the observed N this study Read the per-animal column as the honest power; LME is optimistic (obs-level DF).
is typically underpowered; per-animal power reaches ~0.8 only at larger N.
analysis/matlab/variations/unmerge_d6_13/power_result_rate.txt
+28
View File
@@ -0,0 +1,28 @@
==============================================================================
POWER SIMULATION -- unmerge_d6_13 [metric: success RATE]
==============================================================================
model: behavior ~ stim + day + stim:day + (1|rat) (behavior = success RATE; day within-window)
observed groups: stim n=3, control n=2 nrep=120, alpha=0.05
ground truth: stim:day=+0.01216/day, rat SD=0.04815, residual SD=0.05028, days=8
true stim:day interaction = +0.01216 (100% of observed)
N/group | per-animal power | LME power
------------------------------------------
3 | 0.32 | 0.47 <- observed
5 | 0.63 | 0.76
8 | 0.84 | 0.88
12 | 0.91 | 0.96
16 | 1.00 | 1.00
24 | 1.00 | 1.00
true stim:day interaction = +0.006078 (50% of observed)
N/group | per-animal power | LME power
------------------------------------------
3 | 0.07 | 0.17 <- observed
5 | 0.23 | 0.28
8 | 0.27 | 0.35
12 | 0.39 | 0.40
16 | 0.48 | 0.58
24 | 0.83 | 0.84
Read the per-animal column as the honest power; LME is optimistic (obs-level DF).
analysis/matlab/variations/unmerge_d6_13/powersim.m
+35 -34
View File
@@ -1,45 +1,49 @@
% Variation power simulation -- Monte-Carlo power for the paper's stim x day % Variation power simulation -- Monte-Carlo power for the paper's stim x day
% interaction, using THIS folder's data as the ground truth. % interaction, using THIS folder's data as the ground truth, for BOTH metrics:
% metric = count : behavior = # successes -> power_result.txt
% metric = rate : behavior = success / attempts -> power_result_rate.txt
% %
% Ground truth: fitlme(behavior ~ stim + day + stim:day + (1|rat)) on data.csv % Ground truth: fitlme(behavior ~ stim + day + stim:day + (1|rat)) on data.csv
% (success COUNT; day within-window). Its fixed effects, per-rat intercept SD, % (day within-window). Its fixed effects, per-rat intercept SD, and residual SD
% and residual SD generate NREP synthetic datasets at each rats-per-group N and % generate NREP synthetic datasets at each rats-per-group N and each true-effect
% each true-effect multiplier (1 = observed slope, 0.5 = half). Each dataset is % multiplier (1 = observed, 0.5 = half). Each is scored at alpha=0.05 by:
% scored at alpha = 0.05 two ways: % per-animal : Welch t on per-rat behavior~day slopes (cluster-honest power)
% per-animal : Welch t on per-rat behavior~day slopes (cluster-honest -- the
% honest power, matching the random-slope / per-animal inference)
% LME : the fitlme stim:day p (observation-level DF -- optimistic) % LME : the fitlme stim:day p (observation-level DF -- optimistic)
% Writes power_result.txt beside this script. Run: matlab -batch "powersim" % Writes power_result[_rate].txt. Run: matlab -batch "powersim"
% (Copy of analysis/matlab/variation_power.m; see make_variation_power.m.) % (Copy of analysis/matlab/variation_power.m; see make_variation_power.m.)
here = fileparts(mfilename('fullpath')); here = fileparts(mfilename('fullpath'));
if isempty(here); here = pwd; end if isempty(here); here = pwd; end
vname = regexprep(here, '.*[/\\]', ''); vname = regexprep(here, '.*[/\\]', '');
D = readtable(fullfile(here, 'data.csv'), 'TextType', 'string'); D = readtable(fullfile(here, 'data.csv'), 'TextType', 'string');
localPower(D, 'count', here, vname);
localPower(D, 'rate', here, vname);
% ---------------------------------------------------------------- per metric
function localPower(D, metric, here, vname)
NS = [3 5 8 12 16 24]; EFFMULS = [1 0.5]; NREP = 120;
FORMULA = 'behavior ~ stim + day + stim:day + (1|rat)'; FORMULA = 'behavior ~ stim + day + stim:day + (1|rat)';
NS = [3 5 8 12 16 24]; if strcmp(metric, 'rate')
EFFMULS = [1 0.5]; D = D(D.total > 0, :); beh = D.success ./ D.total; mlabel = 'success RATE'; suffix = '_rate';
NREP = 120; else
beh = D.success; mlabel = '# successes (count)'; suffix = '';
warnState = warning('off', 'all'); end
rng(1); warnState = warning('off', 'all'); rng(1);
day0 = min(D.day); day0 = min(D.day);
tbl0 = table(D.success, D.day - day0, double(D.stim), categorical(D.subject), ... tbl0 = table(beh, D.day - day0, double(D.stim), categorical(D.subject), ...
'VariableNames', {'behavior', 'day', 'stim', 'rat'}); 'VariableNames', {'behavior', 'day', 'stim', 'rat'});
nStim = numel(unique(D.subject(D.stim == 1))); nStim = numel(unique(D.subject(D.stim == 1)));
nCtrl = numel(unique(D.subject(D.stim == 0))); nCtrl = numel(unique(D.subject(D.stim == 0)));
bar = repmat('=', 1, 78); bar = repmat('=', 1, 78);
s = sprintf('%s\nPOWER SIMULATION -- %s\n%s\n', bar, vname, bar); s = sprintf('%s\nPOWER SIMULATION -- %s [metric: %s]\n%s\n', bar, vname, mlabel, bar);
s = [s sprintf('model: %s (success COUNT; day within-window)\n', FORMULA)]; s = [s sprintf('model: %s (behavior = %s; day within-window)\n', FORMULA, mlabel)];
s = [s sprintf('observed groups: stim n=%d, control n=%d nrep=%d, alpha=0.05\n', nStim, nCtrl, NREP)]; s = [s sprintf('observed groups: stim n=%d, control n=%d nrep=%d, alpha=0.05\n', nStim, nCtrl, NREP)];
if nStim < 2 || nCtrl < 2 || numel(unique(tbl0.day)) < 2 if nStim < 2 || nCtrl < 2 || numel(unique(tbl0.day)) < 2
s = [s sprintf('\nInsufficient data for a power simulation (need >=2 animals/group and >=2 days).\n')]; s = [s sprintf('\nInsufficient data for a power simulation.\n')];
localFinish(s, here); warning(warnState); return localFinish(s, here, suffix); warning(warnState); return
end end
lme = fitlme(tbl0, FORMULA); lme = fitlme(tbl0, FORMULA);
@@ -49,14 +53,13 @@ bDay = be(strcmp(cn, 'day')); bInt = be(strcmp(cn, 'day:stim'));
psi = covarianceParameters(lme); sRat = sqrt(psi{1}); sRes = sqrt(lme.MSE); psi = covarianceParameters(lme); sRat = sqrt(psi{1}); sRes = sqrt(lme.MSE);
days = (0:max(tbl0.day))'; days = (0:max(tbl0.day))';
s = [s sprintf('ground truth: stim:day=%+.2f/day, rat SD=%.2f, residual SD=%.2f, days=%d\n', ... s = [s sprintf('ground truth: stim:day=%+.4g/day, rat SD=%.4g, residual SD=%.4g, days=%d\n', ...
bInt, sRat, sRes, numel(days))]; bInt, sRat, sRes, numel(days))];
for eMul = EFFMULS for eMul = EFFMULS
bI = bInt * eMul; bI = bInt * eMul;
s = [s sprintf('\n true stim:day interaction = %+.2f (%.0f%% of observed)\n', bI, eMul * 100)]; %#ok<AGROW> s = [s sprintf('\n true stim:day interaction = %+.4g (%.0f%% of observed)\n', bI, eMul * 100)];
s = [s sprintf(' %-8s | per-animal power | LME power\n', 'N/group')]; %#ok<AGROW> s = [s sprintf(' %-8s | per-animal power | LME power\n %s\n', 'N/group', repmat('-', 1, 42))];
s = [s sprintf(' %s\n', repmat('-', 1, 42))]; %#ok<AGROW>
for N = NS for N = NS
sigPA = 0; sigL = 0; sigPA = 0; sigL = 0;
for r = 1:NREP for r = 1:NREP
@@ -70,20 +73,18 @@ for eMul = EFFMULS
end end
star = ''; star = '';
if N == nStim || N == nCtrl; star = ' <- observed'; end if N == nStim || N == nCtrl; star = ' <- observed'; end
s = [s sprintf(' %-8d | %5.2f | %5.2f%s\n', N, sigPA / NREP, sigL / NREP, star)]; %#ok<AGROW> s = [s sprintf(' %-8d | %5.2f | %5.2f%s\n', N, sigPA / NREP, sigL / NREP, star)];
end end
end end
s = [s sprintf('\nRead the per-animal column as the honest power; LME is optimistic (obs-level DF).\n')];
s = [s sprintf(['\nRead the per-animal column as the honest power. At the observed N this study\n' ... localFinish(s, here, suffix);
'is typically underpowered; per-animal power reaches ~0.8 only at larger N.\n'])];
localFinish(s, here);
warning(warnState); warning(warnState);
end
% ---------------------------------------------------------------- helpers % ---------------------------------------------------------------- helpers
function localFinish(s, here) function localFinish(s, here, suffix)
fprintf('%s', s); fprintf('%s', s);
fid = fopen(fullfile(here, 'power_result.txt'), 'w'); fid = fopen(fullfile(here, ['power_result' suffix '.txt']), 'w');
fprintf(fid, '%s', s); fclose(fid); fprintf(fid, '%s', s); fclose(fid);
end end
analysis/matlab/variations/unmerge_d6_13/result.txt
+4 -4
View File
@@ -1,5 +1,5 @@
============================================================================== ==============================================================================
VARIATION: unmerge_d6_13 VARIATION: unmerge_d6_13 [metric: success COUNT]
============================================================================== ==============================================================================
model: behavior ~ stim + day + stim:day + (1|rat) (behavior = success COUNT) model: behavior ~ stim + day + stim:day + (1|rat) (behavior = success COUNT)
day = training day within window (0 = first analyzed day) day = training day within window (0 = first analyzed day)
@@ -60,9 +60,9 @@ effect t(df) / F(df1) p (resid) Satterthwaite: p (df)
stim x day (interaction) t(30)= 1.71 F(1)= 2.933 p=0.0971 p=0.09701 (df=30) stim x day (interaction) t(30)= 1.71 F(1)= 2.933 p=0.0971 p=0.09701 (df=30)
day (learning) t(30)= 1.61 F(1)= 2.598 p=0.1175 p=0.1174 (df=30) day (learning) t(30)= 1.61 F(1)= 2.598 p=0.1175 p=0.1174 (df=30)
stim (main, window start) t(30)= 1.69 F(1)= 2.856 p=0.1014 p=0.1269 (df=9) stim (main, window start) t(30)= 1.69 F(1)= 2.856 p=0.1014 p=0.1269 (df=9)
interaction 95% CI: [-0.44, +4.97] interaction 95% CI: [-0.4363, +4.97]
HONEST LME (per-animal random slope, day|rat): interaction F(1,13.9)=2.48, p=0.1382 HONEST LME (per-animal random slope, day|rat): interaction F(1,13.9)=2.48, p=0.1382
(Satterthwaite DF ~= residual on this random-intercept model; the random-slope (Satterthwaite DF ~= residual on this random-intercept model; the random-slope
model above is the honest learning-rate test -- DF collapses toward the animal count.) model above is the honest learning-rate test -- DF collapses toward the animal count.)
INTERPRETATION: stim x day interaction n.s. -- slopes parallel (no differential learning rate) (p=0.0971, slope diff=+2.27) INTERPRETATION: stim x day interaction n.s. -- slopes parallel (no differential learning rate) (p=0.0971, slope diff=+2.267)
Paper (N=24): interaction t(227)=2.68, F(1)=7.12, p=0.008. Paper (N=24, count): interaction t(227)=2.68, F(1)=7.12, p=0.008.
analysis/matlab/variations/unmerge_d6_13/result_rate.txt
+68
View File
@@ -0,0 +1,68 @@
==============================================================================
VARIATION: unmerge_d6_13 [metric: success RATE (success/attempts)]
==============================================================================
model: behavior ~ stim + day + stim:day + (1|rat) (behavior = success RATE (success/attempts))
day = training day within window (0 = first analyzed day)
treatment (stim=1): Electrode-Box-B2
control (stim=0): Electrode-Box-A2
N = 5 rats, 34 sessions raw day coverage: treat 6..13, control 6..13
(equal day coverage over this window)
==============================================================================
FULL MODEL SUMMARY -- fitlme
==============================================================================
Linear mixed-effects model fit by ML
Model information:
Number of observations 34
Fixed effects coefficients 4
Random effects coefficients 5
Covariance parameters 2
Formula:
behavior ~ 1 + day*stim + (1 | rat)
Model fit statistics:
AIC BIC LogLikelihood Deviance
-85.035 -75.877 48.518 -97.035
Fixed effects coefficients (95% CIs):
Name Estimate SE tStat DF pValue
{'(Intercept)'} 0.53562 0.0416 12.875 30 9.415e-14
{'day' } 0.0032833 0.0069042 0.47554 30 0.63784
{'stim' } 0.10269 0.053584 1.9164 30 0.06488
{'day:stim' } 0.012156 0.0086128 1.4114 30 0.16841
Lower Upper
0.45066 0.62058
-0.010817 0.017384
-0.0067429 0.21212
-0.0054334 0.029746
Random effects covariance parameters (95% CIs):
Group: rat (5 Levels)
Name1 Name2 Type Estimate
{'(Intercept)'} {'(Intercept)'} {'std'} 0.048154
Lower Upper
0.023654 0.098032
Group: Error
Name Estimate Lower Upper
{'Res Std'} 0.050284 0.038906 0.064989
effect t(df) / F(df1) p (resid) Satterthwaite: p (df)
----------------------------------------------------------------------------
stim x day (interaction) t(30)= 1.41 F(1)= 1.992 p=0.1684 p=0.1684 (df=30)
day (learning) t(30)= 0.48 F(1)= 0.226 p=0.6378 p=0.6378 (df=30)
stim (main, window start) t(30)= 1.92 F(1)= 3.673 p=0.06488 p=0.09052 (df=8)
interaction 95% CI: [-0.005433, +0.02975]
HONEST LME (per-animal random slope, day|rat): interaction F(1,2.3)=0.03, p=0.8749
(Satterthwaite DF ~= residual on this random-intercept model; the random-slope
model above is the honest learning-rate test -- DF collapses toward the animal count.)
INTERPRETATION: stim x day interaction n.s. -- slopes parallel (no differential learning rate) (p=0.1684, slope diff=+0.01216)
Paper (N=24, count): interaction t(227)=2.68, F(1)=7.12, p=0.008.