During my research and experiments about using WSJT-X modes through linear transponder satellites, one of the questions I had is by how much do TLEs of different epochs for the same satellite vary. This was glimpsed in part II, where I plotted the "best delay" parameter for TLEs of different age.
The topic of accuracy in TLE computation and propagation is rather complex. A NORAD TLE is the result of an orbit determination after several radar measurements at different epochs, so the elements are in some sense "averaged" over time. Also, the SGP4 propagator is simple and doesn't model many orbit perturbations. However, NORAD TLEs are specially crafted to give improved results when used with SGP4.
Nevertheless, here I present a simple way of studying the rate of change of NORAD TLEs at different epochs. This procedure might not be very meaningful or sophisticate, but still seems to yield some interesting results.