The basic shape you're looking for is a single-pole response.
The basic setup will work in reflection (s11, resistor wired across RF jack from signal to GND), or transmission (s11, s12*) with a series (jack 1---resistor---jack 2) or shunt (jack 1---jack 2, resistor from trace to GND) arrangement. Which one is best, depends on value: small values (under Zo) are better in shunt, large values in series.
*s22 = s11 and s21 = s12 by symmetry, but you can always check all four just in case, or, say, to reduce noise by combining curves.
Then, fit a curve such that the component under test, in the jig arrangement given, is modeled as an R+L, or even (R+L || C) (or (R || C) + L, but you may not have data to disambiguate the two possibilities, for example, full transmission line effects might take over by the time series/parallel LC comes into question, and then a much more developed model is required; this will be in the GHz, corresponding to lead length).
You'll find that carbon comp are more or less ideal, with some capacitance at high frequencies, and inductance given by lead+body length (major downside: excess noise, and value tends to drift up over time, and under voltage bias); carbon/metal film (normally spiral cut) aren't much worse (despite what you may think of the spiral), and, wirewound fall apart in the MHz. Wirewound are available into the megs, where C will also dominate at such frequencies; these will most strongly call into question the L/C placement in the model, but more than that, C is distributed along the winding so you may end up with a more detailed model anyway, even without transmission line effects.
Tim