study3 upload

Matlab scripts of study 3 of my PhD thesis. Refer to main and main_f for no feedback and feedback experiments.

add_responseinstr.m

@@ -0,0 +1,4 @@
+%add response istructions
+Screen('TextSize', Sc.window, 13);Screen('TextFont', Sc.window, 'Myriad Pro');
+DrawFormattedText(Sc.window, instr{1}, 'center', (Sc.size(2)).*cfg.bar.positiony+80, 0);
+DrawFormattedText(Sc.window, instr{2}, 'center', (Sc.size(2)).*cfg.bar.positiony+110, 0);

add_responseinstr_.m

@@ -0,0 +1,9 @@
+% This script was executed in study3 under the name of add_responseinstr.m
+% I changed the name in order to run study3f so that the main script could
+% run the add_responseintr.m script in /myfunctions
+% 17.04.2015
+
+%add response istructions
+Screen('TextSize', Sc.window, 13);Screen('TextFont', Sc.window, 'Myriad Pro');
+DrawFormattedText(Sc.window, instr{1}, 'center', (Sc.size(2)).*cfg.bar.positiony+80, 0);
+DrawFormattedText(Sc.window, instr{2}, 'center', (Sc.size(2)).*cfg.bar.positiony+110, 0);

agreement_functions.m

@@ -0,0 +1,38 @@
+clear all
+close all 
+
+addpath(genpath('C:\Users\npescetelli\Documents\Paris\toolboxes'));
+
+%% plot confidence distribution
+clf
+x1=[0:1:1000];
+z = zscore(x1);
+C = normpdf(z,0,1);
+
+plot(z,C,'LineWidth',3)
+xlim([-10 10])
+ylim([0 1])
+
+set(gcf, 'Position', get(0,'Screensize')); % Maximize figure.
+
+%% plot conditional probability
+hold on
+Y1=sigmoid(z,1,2)*.5+.5; % agreeing in confidence
+Y2=sigmoid(z,1,-2)*.5+.5; % agreeing in uncertainty
+Y3=sigmoid(z,1,2)*.5+.5;% accurate agreeing
+Y4=sigmoid(z,1,-2)*.5+.5;% accurate agreeing
+
+h1 = line([0 0],[1 0],'LineWidth',2,'LineStyle','--','color','b');
+
+h2 = plot(z,Y1,'LineWidth',3,'color','r')
+h3 = plot(z,Y2,'LineWidth',3,'color','b')
+h4 = plot(z,Y3,'LineWidth',3,'color','r','LineStyle','--')
+h5 = plot(z,Y4,'LineWidth',3,'color','b','LineStyle','--')
+
+set(gca,'FontSize',24)
+xlim([-10 10]),xlabel('confidence')
+ylim([0 1]),ylabel('p(agree)')
+
+legend([h2 h3 h4 h5], 'high agreement in confidence','high agreement in uncertainty',...
+    'low agreement in confidence','low agreement in uncertainty','Location','SouthWest')
+

agreementf.m

@@ -0,0 +1,110 @@
+function [agree, step] = agreementf(rawx,type,distribution,funct)
+% use : [agree step] = agreementf(raw-x,obstype,distribution,function type)
+%
+% agree: logical 
+% step: -1-left tail, 0-central mass, 1-right tail of the distribution
+% raw-x : current confidence judgment
+% obstype: 1-agree in confidence, 2-agree in uncertainty, 0- baseline 70%
+% distribution: vector containing (correct) confidence judgments so far 
+% function: one among 'semilinear' 'sigmoid' 'stepwise'
+x =abs(rawx);
+b = quantile(distribution,[.3 .7]); % boundaries (30% and 70% of the mass)
+
+switch funct
+    case 'sigmoid'
+        % define sigmoid function parameters
+        sat = .5; % lower saturation at 50%
+        phase = 1;
+        slope1 = 2; % obstype 1
+        slope2 = -2; % obstype 2
+
+        step   = NaN;
+        switch type
+            case 1 % agreement under confidence
+                pagree= sigmoid(x,phase,slope1)*.5 + sat; % agreeing in confidence
+            case 2 % agreement under uncertainty
+                pagree = sigmoid(x,phase,slope2)*.5 + sat; % agreeing in uncertainty
+            case 0 % baseline type
+                pagree = .7;
+        end
+
+    case 'semilinear'
+        switch type
+            case 1 % agreement under confidence
+                if x > b(1) && x <= b(2) % around 60% of the mass is between -.52 and +.52 sigmas
+                    pagree = .7;
+                    step   = 0;
+                elseif x <= b(1)
+                    minv = min(distribution); % minimum value of the distribution corresponds to 60%
+                    beta = -.1 / (b(2) + minv);
+                    intercept = .7 + beta * b(2);
+                    pagree = intercept + beta * x;
+                    step   = -1;
+                elseif x > b(2)
+                    maxv = max(distribution);
+                    beta = .1 / (b(1) + maxv);
+                    intercept = .7 - beta * b(2);
+                    pagree = intercept + beta * x;
+                    step   = 1;
+                end
+            case 2 % agreement under uncertainty
+                if x > b(1) && x <= b(2)
+                    pagree = .7;
+                    step   = 0;
+                elseif x <= b(1)
+                    minv = min(distribution); % minimum value of the distribution corresponds to 60%
+                    beta = .1 / (b(2) + minv);
+                    intercept = .8 - beta * (minv);
+                    pagree = intercept + beta * x;
+                    step   = -1;
+                elseif x > b(2)
+                    maxv = max(distribution);
+                    beta = -.1 / (b(1) + maxv);
+                    intercept = .7 - beta * b(2);
+                    pagree = intercept + beta * x;
+                    step   = 1;
+                end
+            case 0 % baseline type
+                pagree = .7;
+                step   = 0;
+        end
+    case 'stepwise'
+        switch type
+            case 1 % agreement under confidence 
+                if x > b(1) && x <= b(2)
+                    pagree = .7;
+                    step   = 0;
+                elseif x <= b(1)
+                    pagree = .6;
+                    step   = -1;
+                elseif x > b(2)
+                    pagree = .8;
+                    step   = 1;
+                end
+            case 2 % agreement under uncertainty
+                if x > b(1) && x <= b(2)
+                    pagree = .7;
+                    step   = 0;
+                elseif x <= b(1)
+                    pagree = .8;
+                    step   = -1;
+                elseif x > b(2)
+                    pagree = .6;
+                    step   = 1;
+                end
+            case 0 % baseline type
+                if x > b(1) && x <= b(2)
+                    pagree = .7;
+                    step   = 0;
+                elseif x <= b(1)
+                    pagree = .7;
+                    step   = -1;
+                elseif x > b(2)
+                    pagree = .7;
+                    step   = 1;
+                end
+        end        
+end
+
+agree = rand<pagree;
+

build_trials.m

@@ -0,0 +1,130 @@
+trials = [];
+trialid = 0;
+index=1;
+[block_trials0, block_trials1] =deal([]); %initialize as empty vectors
+
+
+% practice trials
+for b = 1: cfg.nblocksprac
+    block_trials0 = []; %clear vector
+    for t = 1:cfg.ntrialsprac
+        trialid                                 = trialid+1;
+        block_trials0(end+1).trialid            = trialid;
+        block_trials0(end).estim_obsacc         = [];
+        block_trials0(end).block                = b;
+        block_trials0(end).feedback             = true; 
+        block_trials0(end).cj1                  = [];
+        block_trials0(end).obsacc               = NaN;
+        block_trials0(end).qanswers             = [];
+        if b > 1
+            block_trials0(end).obstype          = 0; % practice
+            block_trials0(end).pic              = cfg.observer.pic(4);     % practice picture
+            block_trials0(end).voice            = cfg.observer.voice(4);   % practice voice
+        else
+            block_trials0(end).obstype          = NaN;
+            block_trials0(end).pic              = NaN;
+            block_trials0(end).voice            = NaN;
+        end
+    end
+    block_trials0   = block_trials0(randperm(length(block_trials0)));      % randomize trials within block
+    trials          = cat(2,trials,block_trials0);                         % concatenate practice block
+end
+
+% experimental trials
+for b = b+1 : cfg.nblocks+cfg.nblocksprac
+    block_trials1 = [];
+    for o = 1:cfg.nobs
+        for t = 1:((cfg.ntrials - cfg.nullt) ./ 3) % 10 presentations of observers/block
+            trialid                                 = trialid+1;
+            block_trials1(end+1).trialid            = trialid;
+            block_trials1(end).estim_obsacc         = [];
+            block_trials1(end).block                = b;
+            block_trials1(end).feedback             = false;
+            block_trials1(end).cj1                  = [];
+            block_trials1(end).obsacc               = NaN; % conditional on subjects confidence
+            block_trials1(end).obstype              = o-1;
+            block_trials1(end).pic                  = cfg.observer.pic(o); 
+            block_trials1(end).voice                = cfg.observer.voice(o); 
+            block_trials1(end).qanswers             = [];
+        end
+    end
+    for t = 1:cfg.nullt % 5 null trials per block
+        trialid                                 = trialid+1;
+        block_trials1(end+1).trialid            = trialid;
+        block_trials1(end).estim_obsacc         = [];
+        block_trials1(end).block                = b;
+        block_trials1(end).feedback             = false;
+        block_trials1(end).cj1                  = [];
+        block_trials1(end).obsacc               = NaN; 
+        block_trials1(end).obstype              = NaN;
+        block_trials1(end).pic                  = NaN;
+        block_trials1(end).voice                = NaN;
+        block_trials1(end).qanswers             = [];
+    end
+    block_trials1    = block_trials1(randperm(length(block_trials1)));         % randomize trials within a block
+    trials    = cat(2,trials,block_trials1);                                   % concatenate to main trial vector
+end
+clear  block_trial*;
+
+wl0 = repmat([1 2],1,(cfg.ntrialsprac/2) * cfg.nblocksprac);
+
+% this method for where larger computation has been implemented only for
+% study3f. Check previous versions of the study!
+wl1 = repmat([1 2],1,(cfg.ntrials./2) * cfg.nblocks);
+%% WARNING! Massive problem detected. 
+% This way of assigning wherelarger is imbalanced!
+% wl1 = repmat([1 2],1,(cfg.ntrials./2 +cfg.nullt) * cfg.nblocks);
+% Has this method been implemented for study3???
+%
+%% 
+wl0 = wl0(randperm(length(wl0)));
+wl1 = wl1(randperm(length(wl1)));
+wl  = cat(2,wl0,wl1);
+for t = 1 : length(trials)
+    trials(t).wherelarger = wl(t);
+    trials(t).wheredots = zeros(2,400);
+end
+clear wl
+
+%-- extra information at each trial
+for t = 1:length(trials)
+    %-- add breaks and instruction points
+    if t == 1
+        trials(t).break         = false;
+        trials(t).instr         = true;
+        trials(t).questionnaire = false;
+    elseif ...
+            trials(t-1).block ~= trials(t).block 
+        trials(t).break = true;
+        trials(t).feedback = true;
+        if trials(t).block <= cfg.nblocksprac+1
+            trials(t).instr = true;
+        else trials(t).instr = false;
+        end
+        if ismember(trials(t).block,[3:2:cfg.nblocks+cfg.nblocksprac]), % questionnaires every 2 blocks
+            trials(t).questionnaire = true; 
+        else trials(t).questionnaire = false; 
+        end
+    else
+        trials(t).break         = false;
+        trials(t).instr         = false;
+        trials(t).questionnaire = false;
+    end
+end
+
+%-- check fields: none should be empty
+fields = fieldnames(trials);
+for f= 1:length(fields)
+    if eval(['length({trials.' fields{f} '})']) ~= length(trials)
+        disp(['Empty fields in ' fields{f} '!']);
+    end
+end
+clear f
+if 0
+    % check the design manually
+    % after randomization
+    img_dsg(trials,{'block', 'feedback', 'obstype', 'instr' 'wherelarger'})
+    figure(gcf+1);
+    [z index] = sort([trials.trialid]);clear z;
+    img_dsg(trials(index),{'block', 'feedback', 'obstype','instr' 'wherelarger'})
+end

check_agreementf

@@ -0,0 +1,12 @@
+for c=1:1000
+    % generate confidence judgement
+    conf(c)=25+5*randn;
+    agree1(c) = agreementf(1,conf,'sigmoid');
+    agree2(c) = agreementf(2,conf,'sigmoid');
+end
+
+mean([agree1(conf>40); agree2(conf>40)],2)
+
+figure(1),hold on
+X1 = normpdf([-2:.01:2],0,1);
+plot([-2:.01:2],X1)

check_agreementf.m

@@ -0,0 +1,43 @@
+%%
+addpath(genpath('/home/niccolo/Dropbox/Oxford/myfunctions'));
+addpath(genpath('/media/niccolo/Yupi/Paris/toolboxes/JR_toolbox'));
+
+
+%%
+clear all
+figure(1)
+% generate confidence judgement
+conf=25+5*randn(50,1);
+
+for c=1:5000
+    cj(c) = 25+5*randn;
+    agree1(c) = agreementf(cj(c),1,conf,'stepwise');
+    agree2(c) = agreementf(cj(c),2,conf,'stepwise');
+    agree0(c) = agreementf(cj(c),0,conf,'stepwise');
+end
+
+hist(conf)
+
+%%
+clear X*
+figure(2),clf
+bins = [0 22 28 50];
+for i =1:length(bins)-1 
+    hold on
+    clear toi
+    toi = cj>bins(i) & cj<bins(i+1);
+    X1(i) = mean(agree1(toi));
+    X2(i) = mean(agree2(toi));
+    X0(i) = mean(agree0(toi));
+    plot(i,X1(i),'rd')
+    plot(i,X2(i),'s')
+    plot(i,X0(i),'gs')
+end
+plot(X1,'r')
+plot(X2,'b')
+plot(X0,'g')
+
+% glmfit([0:5:45]',X1','binomial','link','logit')
+
+
+