The interest in Physical Human-Robot Interaction (pHRI) has significantly increased over the last two decades thanks to the availability of collaborative robots that guarantee user safety during force exchanges. For this reason, stability concerns have been addressed extensively in the literature while proposing new control schemes for pHRI applications. Because of the nonlinear nature of robots, stability analyses generally leverage passivity concepts. On the other hand, the proposed algorithms generally consider ideal models of robot manipulators. For this reason, the primary objective of this paper is to conduct a detailed analysis of the sources of instability for a class of pHRI control schemes, namely proxy-based constrained admittance controllers, by considering parasitic effects such as transmission elasticity, motor velocity saturation, and actuation delay. Next, a sensitivity analysis supported by experimental results is carried out, in order to identify how the control parameters affect the stability of the overall system. Finally, an adaptation technique for the proxy parameters is proposed with the goal of maximizing transparency in pHRI. The proposed adaptation method is validated through both simulations and experimental tests.