Many of the most commonly used generation planning models have been formulated in a way that neglects the chronological sequence of demand and the mixed-integer nature of generating units. The generator schedules assumed by these models are inaccurate and become increasingly divorced from real schedules with increasing variability. This paper seeks to characterize and quantify the limitations of these models over a broad set of input parameters. For an illustrative set of test systems, wind capacities and generator types, annual system costs are determined for all combinations of generating units using a unit-commitment model, which captures the chronological behavior of units and a dispatch model which does not. It is seen that the relative performance of the dispatch model is highly system specific but generally degrades with increasing variability. The difference in cost estimates between the models is decomposed into start costs, starts avoidance and average cost estimation error. The impact on least-cost portfolios is shown and finally sensitivities are performed with the addition of hydro and nuclear power to assess their impact.