NICER PI bins greatly oversample the detector energy resolution (FWHM ~80 eV at 1 keV and ~200 eV at 10 keV, dark conditions). Using PI data without worrying about the energy resolution misrepresents the number of independent data points and will reduce the signal-to-noise of your data set. This means a poorer model will have a misleadingly better reduced chisq, and that constraints on subtle features like broad reflection lines or narrow absorption features will be somewhat handicapped. As the instrument calibration improves towards the statistical limit, this will become increasingly important in getting the most out of our data.
For those interested, Kaastra & Bleeker (2016) present an analysis on the optimal binning for detector data. For NICER data with counts per resolution element ranging from ~1-1e6, optimal binning would oversample by ~1-3 times the FWHM (see their Figure 2). Unfortunately, the ftool most used in grouping data, grppha, won't allow a user to supply two simultaneous constraints (e.g., group min 25 AND group to resolution).
If anyone would like, you can download the IDL tool and NICER binning table (aimed to produce approximately 3× oversampling) that will let you bin to resolution while also achieving a minimum number of counts per bin. Note that for sufficiently bright data, the resolution table will be fine for grppha to work with directly (provided if you don't have to worry about separately reaching a count threshold).
1) 1st run grppha: % grppha nicer_channels_to_group.dat 2) then run js_groupspec in IDL ; will achieve a minimum of 25 counts per bin and match the resolution criterion at the same time. IDL> js_groupspec,"my_favorite_data.pha",out="my_grouped_output.grp",min=25