To my experience, this question should be considered regarding the size or characteristic length scale of the analyzed microstructure. For coarse grain microstructure and/or statistical analysis, we are generally working in the range of 15/20kv. For "advanced" steel microstructure, for exemple a mixture of ferrite and austenite with grain size below 1µm, we are more working in the range of 10-15kV. We observed in fact a real improvement with low kV for the analysis of fine structure.

Was searching a little bit and found an article here on researchgate which may be usefull

https://www.researchgate.net/publication/233598275_Towards_ultrahigh_resolution_EBSD_by_low_accelerating_voltage

Article Towards ultrahigh resolution EBSD by low accelerating voltage

@David: do you use also higher kV, e.g. 30kV? If you reduce the electron energy you certainly reduce the probe size but you also increase possible effects preparation artifacts. How do you know that the results you are derive are characteristic for your bulk matrial?

We are only using 30kV for TKD applications.

It's clear that by using low kV you are more sensitive to surface preparation and artifacts. But technicians are well experienced here and really master the complicated art of sample preparation! We have well established samples preparation protocols with comparative experiments on the effects of sample preparation (OPS vs. OPU, polishing time and pressure...). 

We never know if what we are measuring is representative from the bulk or from the surface. But we always have detailed metallographic analysis (with some times serial sectioning, but not always as it is time consuming)  of each sample measured by EBSD, which can help at knowing if what we are measuring is representative or not. But to my opinion, it is important to have in mind that whatever the characterization method, we never expect to have an "absolute" measurement of the microstructure, but relative data to be compared according to the same reference protocol.

Here, experience of the person preparing the samples and running the experiments (who should be the same person I think!) is a key element ! 

For coarse grain material and step width in the scale on microns I also use sometimes higher acceleration voltages then 20kV. To my opinion, especially for phases with low diffraction signal/background ratio (which is only predictable by dynamical pattern simulation, i.e. I don't talk about the signal blurring caused by crystal defects!) it may help to get better patterns. However, in case of tiny grains the reduction of acceleration voltage and the use of a FE-SEM are the main option to get satisfying patterns.

I normally use 20 kV and 8 nA, it is a sort of good setup for most geological materials, even those with low diffraction signal (for some of them I have to use a 2x2 binning to get decent indexation. If the grain sizes are too small (like below 1 µm mentioned by Dave) I change to 15 kV and works quite well, I actually never tried 10 kV :-) !