The error bars scale with appropriate axis scales, and negative error bars or error bars equal; to zero do not plot at all on a logarithmic axis —
x = 1:10;
y = rand(size(x));
err = randn(size(x))*0.1;
figure
errorbar(x, y, err)
grid
figure
errorbar(x, y, err)
grid
set(gca, 'YScale','log')
(This also neatly illustrates the problem of doing a regression on logarithmically-scaled variables, because the actual errors behave the same way. They are multiplicative, not additive, violating the assumptions of least-squares parameter estimation.)
.
