By using cavity ring-down spectroscopy (CRDS), we have obtained visible overtone absorption spectra of HCN which display a large collisional line-mixing effect in the proximity of the R branch band heads, for J similar to 18. We consider in detail the 106<--000 (1=CN, 0=bend, 6=CH) parallel transition. The R branch profile was modeled using the modified-exponential-gap (MEG) and energy-corrected-sudden approximation (ECS) population transfer rate laws. We used the rates previously determined by Pine and Looney (PL) by fitting the self broadening coefficients measured for the Q branches of Pi-Sigma infrared perpendicular stretch-bend combination bands of HCN [J. Chem. Phys. 96, 1704 (1992)]. Contrary to what is found by these authors, in the present case the MEG law reproduces the R branch line-mixing satisfactorily, while the ECS model fails. This reflects an increasing propensity at higher J for collisional transitions with smaller Delta J. Using the MEG law, we found we need to include, as had PL in their fits to the infrared Q branches, an empirical dephasing scale factor F similar to 0.6 for the coherence transfer rates to obtain a satisfactory simulation of the R band head. PL suggested that dephasing in the Q branch spectra are due to cross relaxation across l-type doublet levels of the Pi state, but no such mechanism would be available in the present case. However, we have found that by using a 50/50 linear combination of the ECS and MEG rate laws, it is possible to fit our data even with F=1, which would imply no dephasing of coherence. We take this as a demonstration that the dephasing factor F cannot be reliably extracted from line-mixing studies alone but instead requires some independent source of information on the relative value for state to state inelastic collision rates.