A recent study [Dan S. P. Smith and Bruce M. Law, Phys. Rev. E 54, 2727 (1996)] presented measurements of the ellipsometric coefficient at the Brewster angle \ensuremath{\rho}-bar on the liquid-vapor surface of four different binary liquid mixtures in the vicinity of their liquid-liquid critical point and analyzed the data analytically for large reduced temperatures t. In the current report we analyze this (\ensuremath{\rho}-bar,t) data numerically over the entire range of t. Theoretical universal surface scaling functions ${\mathrm{P}}_{\ifmmode\pm\else\textpm\fi{}}$(x) from a Monte Carlo (MC) simulation [M. Smock, H. W. Diehl, and D. P. Landau, Ber. Bunsenges. Phys. Chem. 98, 486 (1994)] and a renormalization-group (RG) calculation [H. W. Diehl and M. Smock, Phys. Rev. B 47, 5841 (1993); 48, 6470(E) (1993)] are used in the numerical integration of Maxwell's equations to provide theoretical (\ensuremath{\rho}-bar,t) curves that can be compared directly with the experimental data. While both the MC and RG curves are in qualitative agreement with the experimental data, the agreement is generally found to be better for the MC curves. However, systematic discrepancies are found in the quantitative comparison between the MC and experimental (\ensuremath{\rho}-bar,t) curves, and it is determined that these discrepancies are too large to be due to experimental error. Finally, it is demonstrated that \ensuremath{\rho}-bar can be rescaled to produce an approximately universal ellipsometric curve as a function of the single variable ${\ensuremath{\xi}}_{\ifmmode\pm\else\textpm\fi{}}$/\ensuremath{\lambda}, where \ensuremath{\xi} is the correlation length and \ensuremath{\lambda} is the wavelength of light. The position of the maximum of this curve in the one-phase region, (${\ensuremath{\xi}}_{+}$/\ensuremath{\lambda}${)}_{\mathrm{peak}}$, is approximately a universal number. It is determined that (${\ensuremath{\xi}}_{+}$/\ensuremath{\lambda}${)}_{\mathrm{peak}}$ is dependent primarily on the ratio ${\mathrm{c}}_{+}$/${\mathrm{P}}_{\mathrm{\ensuremath{\infty}},+}$, where ${\mathrm{P}}_{+}$(x)\ensuremath{\cong}${\mathrm{c}}_{+}$${\mathrm{x}}^{\mathrm{\ensuremath{-}}\mathrm{\ensuremath{\mathrm{B}}}\mathrm{/}\ensuremath{\nu}}$ for x\ensuremath{\ll}1 and ${\mathrm{P}}_{+}$(x)\ensuremath{\cong}${\mathrm{P}}_{\mathrm{\ensuremath{\infty}},+}$${\mathrm{e}}^{\mathrm{\ensuremath{-}}\mathrm{x}}$ for x\ensuremath{\gg}:1. This enables the experimental estimate of ${\mathrm{c}}_{+}$/${\mathrm{P}}_{\mathrm{\ensuremath{\infty}},+}$=0.90\ifmmode\pm\else\textpm\fi{}0.24, which is significantly large compared to the MC and RG values of 0.577 and 0.442, respectively.