Take for example those graphs from Martorell-Riera et al. 2014:
I am sure you can agree that the selection of grey shades and multiple error bars cover the actual curve of measured values. Also the axis looks quite cluttered with attempt of showing the frequency of the measurement, a case completely lost on panel C.
So I advise to simplify the graph and replace individual error bars with common interval (values are kept as in the original!) in the lighter saturation of the color:
This is a much cleaner view on the measurement with the SD values turned into a "cloud". You can try to argue that this means the single measurements are gone, but was it really possible to read them from the previous ones?
As the last remark I want to briefly come back to the previous post about axis tricks and redo graphs from Rassmusen et al. 2016
They also show several measurements over time, so they can be redone in the same manner as graphs above (keeping original colors and alternatives - both color blind friendly):
Now let's have a look at the context of the entire panel:
Traces look quite similar, but you will probably notice that the scales on every single panel are different! If they were re-scaled the graphs would look the following:
Which seems like a more honest representation of acquired data and should be a good practice of all the data representation.
