A patch of errors
Errorbars are one of the most common representations of standard deviations, standard errors and confidence intervals. Another representation is an errorpatch, as observed in the figure above. An example of such a graphical representation can be found in figure 2 of this research article. How to create such an errorpatch in Matlab is described here.
To plot data, we first need the data. If you have your own data to plot, you can skip this section. Otherwise, please create some random data in Matlab:
%% Create fake data X = -90:10:90; % stimulus X Nrep = 50; % 50 repetitions X = repmat(X,1,Nrep); % apply repeats Y = 50*sind(1.2*X); % let's take a sinusoidal relationship between stimulus and response Y = Y+(abs(X)+20)/3.*randn(size(Y)); % add some Gaussian noise/variation (standard deviation depends on stimulus size)
The data will appear somewhat as in figure 1, that can be created in Matlab by:
figure(1) h = plot(X,Y,'k.'); axis square box off xlabel('X'); ylabel('Y'); axis([-100 100 -100 100]); set(gca,'XTick',-90:30:90,'YTick',-90:30:90);
Next, you should determine the mean 'response' for every unique 'stimulus', and the standard deviation. This can be achieved by:
%% Determine mean and std response for each stimulus uX = unique(X); % unique values of X nX = numel(uX); % number of unique X mu = NaN(size(uX)); % initialization mean vector sd = mu; % initialization std vector se = mu; % also determine the standard error for ii = 1:nX sel = X==uX(ii); mu(ii) = mean(Y(sel)); sd(ii) = std(Y(sel)); n = sum(sel); % number of responses se(ii) = sd(ii)./sqrt(n); % standard error end ci = 1.96*se; % 95% confidence interval
This mean data and the standard deviation can be easily represented with Matlab's errorbar function:
figure(2) h = errorbar(uX,mu,sd,'ko-'); set(h,'MarkerFaceColor','w','LineWidth',2); axis square box off xlabel('X'); ylabel('Y'); set(gca,'XTick',-90:30:90,'YTick',-90:30:90); axis([-100 100 -100 100]);
How can we visualize the data as an errorpatch? With Matlab's patch. The only tricky part is to make a data matrix that describes the data:
Making good use of Matlab's patch function and its figure and axis handles, we obtain figure 3.
figure(3) hpatch = patch(x,y,'k'); % the errorpatch set(hpatch,'EdgeColor','none'); % remove the edge set(hpatch,'FaceColor',[.7 .7 .7]); % and change the hold on; hline = plot(uX,mu,'k-'); % central line set(hline,'LineWidth',2); % make nice thick line set(hline,'Color','k'); % change color axis square box off xlabel('X'); ylabel('Y'); set(gca,'XTick',-90:30:90,'YTick',-90:30:90);
The errorpatches have been colored light gray, and the edge colors have been removed. The mean is also plotted as a thick black line.
Instead of changing the 'facecolor' of the patch to make it appear lighter, you can also make the patch transparent by setting the patch's alpha.
hpatch = patch(x,y,'k'); % the errorpatch set(hpatch,'EdgeColor','none'); % remove the edge alpha(hpatch,0.4); % another option would be to use transparency
This creates a problem if you want to save this figure in a vector-format such as eps. By default, Matlab converts vector figures with transparancy / alpha to bitmaps. This can be circumvented by supplying the -painter option when saving the figure. However, Matlab then removes the transparency. In Adobe Illustrator, you can manually reinstate the transparency.
print('-depsc2','-painter','errorpatch'); % save eps-file with -painter option with filename errorpatch