Early work, but your previous comment got me interested in determining whether a significant correlation exists between change in TS% and change in Usage between seasons (between the 2013-14 & 2014-15 seasons [n = 384], and between the 2014-15 & 2015-16 seasons[n = 333]). TS% and Usage, obviously, had high and significant correlations between seasons with themselves (very similar results in two separate models), but I found that change in TS% did not significantly correlate with change in Usage between seasons (p > 0.20).
Now, this may be due to the fact that players on the tails of the distributions don't see major shifts in their usage/TS% further away from the median. Quality of teammates likely factors in a bit, too. Nonetheless, it's an interesting start. My knowledge in basketball is not as strong as it is with baseball and hockey, so it's possible I missed some important confounding variables to account for in the model.