[14] On the basis of dimensional analysis and observations we have derived a linear relationship between the rate of volumetric margin retreat *V* = *R* · *h* (i.e., expressed in m^{2}/yr) and the mean annual wave power density, _{i}, which, if the cliff height, *h*, at a given site is constant in time, also implies a linear relationship between linear margin retreat *R* (i.e., expressed in m/yr) and _{i}. The most commonly accepted view, on the contrary, assumes a power law relation between *R* and _{i} [*Schwimmer*, 2001; *Kamphuis*, 1987]. We suggest that such a power law assumption is the result of a subjective interpretation of the existing data, which can equally well be interpreted by the theoretically justified linear relationship (6) between retreat and mean incident power density (note also that the power law exponent proposed by *Schwimmer* [2001], *b* = 1.1, is quite close to unity). Figure 3c summarizes the data available in the previous literature, which, in lack of cliff height information, have been plotted in the retreat versus power density plane. The graphical interpretation of the different datasets yields quite different slopes, which can be ascribed to differences in sediment properties (e.g., glacial till for *Gelinas and Quigley* [1973] and marsh sediment for *Schwimmer* [2001]), but also to the possibly different procedures used to estimate/measure incident wave energy, not fully described in some of the previous literature. The data are compared with power law and linear fits, the latter stemming from equation (6) when (the unknown) depth is assumed to remain constant during the time of the observations at each study site. Indeed, Figure 3c does not allow to conclude in favor of either the linear or the power law model on statistical grounds alone. Under such circumstances the linear model, supported by the results in Figure 3b and by theoretical arguments, appears to be preferable. We also note that the dimensional analysis approach proposed here has the significant advantage of allowing us to pool together all the available observations, irrespective of the specific cliff height characterizing each single study site. This allows the derivation of more robust statistics and a greater generality of the conclusions that may be drawn from a statistical analysis applied collectively to diverse study sites. The specific value of the proportionality constant linking power density and retreat in equation (6) is clearly site-dependent (e.g., it contains the effective cohesion term, *c*, which must be a function of sediment and vegetation properties) and requires the analysis of site-specific data. However, the robustness of the common trend emerging in Figure 3b, in spite of the several possibly neglected spatial heterogeneities (e.g., in wind forcing, depth, material, vegetation cover), the theoretical support afforded by dimensional analysis, and the compatibility with previous literature results, provide a remarkable support on the wide validity of the proposed proportionality between *R* and _{i}.