Both would capture a residual effect but 1 - R^2 is perhaps easier to interpret (i.e., as the proportion of variability in the outcome or dependent variable that remains unexplained in the path model). The square root of (1 - R^2) does not have a straightforward interpretation as far as I can tell.