n = 1000000;
observe = 2;

% Naive way
% 1
actual = randn(n, 1) * 4;
offset = randn(n, 1) * 3;
measured = actual + offset;
% 2
antecedent = observe - .1 < measured & measured < observe + .1;
% 3
% hist(actual(antecedent))
% mean(actual(antecedent))

% Less naive way
% 1
actual = randn(n, 1) * 4;
offset = observe - actual;
weight = normpdf(offset, 0, 3);
% 3
% sum(actual .* weight) / sum(weight)
% total = sparse(floor(actual)-min(floor(actual))+1, ones(size(actual)), weight);
% plot(min(floor(actual)):max(floor(actual)), total)

% Even less naive way -- towards Kalman filtering
% (http://www.cs.unc.edu/~welch/kalman/maybeck.html)
actual = -20:.1:20;
offset = observe - actual;
weight = normpdf(actual, 0, 4) .* normpdf(offset, 0, 3);