When selecting the binning region for a spectrum, I stretch the image to make the full top and bottom edges of the spectrum visible. I then set the edge of the spectrum bin to include all of the signal, even the weak parts, while trying not to include any unnecessary background.
Yes this can be particularly important if there is any chromatism or atmospheric dispersion which will change the width of the spectrum along the wavelength axis (or any residual tilt) so the fraction sampled varies. (ISIS has an optimal binning option which seems to cope with this while minimising the sky contribution. It is rather effective on faint targets, though I am not exactly sure how it does it). Similarly the sky background zones need to be set far enough away not to include any of the spectrum. Stretching also helps spot any weak spectra which might be in the selected sky background zones, a bigger problem with slitless setups.
Cheers
Robin
