Foreground subtraction in global redshifted 21 cm measurements is limited by frequency-dependent (chromatic) structure in antenna beam patterns. Chromatic beams couple angular structures in Galactic foreground emission to spectral structures that may not be removed by smooth functional forms. We report results for simulations based on two dipole antennas used by the Experiment to Detect the Global EoR Signature (EDGES). The residual levels in simulated foreground-subtracted spectra are found to differ substantially between the two antennas, suggesting that antenna design must be carefully considered. Residuals are also highly dependent on the right ascension and declination of the antenna pointing, with RMS values differing by as much as a factor of 20 across pointings. For EDGES and other ground-based experiments with zenith pointing antennas, right ascension and declination correspond directly to the local sidereal time and the latitude of the deployment site, hence chromatic beam effects should be taken into account when selecting sites. We introduce the "blade" dipole antenna and show, via simulations, that it has better chromatic performance than the "fourpoint" antenna previously used for EDGES. The blade antenna yields 1-5 mK residuals across the entire sky after a 5-term polynomial is removed from simulated spectra, whereas the fourpoint antenna typically requires a 6-term polynomial for comparable residuals. For both antennas, the signal-to-noise ratio of recovered 21 cm input signals peaks for a 5-term polynomial foreground fit given realistic thermal noise levels.