Noise considerations for experiments that aim to statistically estimate the 21 cm signal from high redshift neutral hydrogen during the Epoch of Reionisation (EoR) using interferometric data are typically computed assuming a tracked observation, where the telescope pointing centre and instrument phase centre are the same over the observation. Current low frequency interferometers use aperture arrays of fixed dipoles, which are steered electronically on the sky, and have different properties to mechanically-steered single apertures, such as reduced sensitivity away from zenith, and discrete pointing positions on the sky. These properties encourage the use of two additional observing modes: (1) zenith drift, where the pointing centre remains fixed at the zenith, and the phase centre tracks the sky, and (2) drift + shift, a hybrid mode where the telescope uses discrete pointing centres, and the sky drifts during each fixed pointing. These three observing modes view the sky differently, and therefore yield different uncertainties in the power spectrum according to the balance of radiometric noise and cosmic variance. The coherence of measurements made by the instrument in these modes dictates the optimal reduction in thermal noise by combination of coherent modes, and the reduction in cosmic variance by combination of incoherent modes (views of different patches of the sky). Along with calibration and instrument stability considerations, the balance between these noise components provides one measure for the utility of these three modes for measuring a statistical signature of the EoR signal. We provide a general framework for estimating the uncertainty in the power spectrum for a given observing mode, telescope beam shape, and interferometer antenna distribution. We then apply this framework to the Murchison Widefield Array (MWA) using an analysis of the two-dimensional (2D) and one-dimensional (1D) power spectra for 900 hours of observing. We demonstrate that zenith drift scans can yield marginally lower uncertainty in the signal power compared with tracked scans for the MWA EoR experiment, and that moderately higher signal-to-noise ratio (S/N) estimates of the amplitude (3%) and slope (1%) of the 1D power spectrum are accessible, translating directly into a reduction in the required observing time to reach the same estimation precision. We find that the additional sensitivity of pointing at zenith, and the reduction in cosmic variance available with a zenith drift scan, makes this an attractive observing mode for current and future arrays.