Electoral competition is a democratic mechanism to guarantee high governmental performance. In reality, however, it often leads to policy failure due to Government Capture and Government Accountability. An understanding of both phenomena has to be based on voter theory and nowadays the probabilistic voter model is the workhorse model applied in voter studies. In this paper we first proceeded to derive a theoretical model to estimate voter behavior including three voting motives: non-policy oriented, policy oriented and retrospective oriented. Then, we derived government performance indicators to estimate Capture and Accountability based on marginal effects and relative importance of the three components. Subsequently, we tested our theory estimating a probabilistic voter model for Ghana using own election survey data. In particular, we calculated different mixed logit model specifications and, to allow heterogeneity, we followed the latent class approach. Using the results of the estimations, we were able to calculate marginal effects and relative importance of each voting motive and we found that the non-policy component is the most important whereas the retrospective component is the less relevant. Finally, the government performance indicators were estimated and they suggest that, although the political weights are unequally distributed in Ghana, the government is partially accountable towards the voter and elections provide an effective mechanism to promote democracy.