This paper studies a principal-agent model of the relationship between o¢ceholders and the electorate, where the o¢ce-holder is initially uninformed about her ability (following Holmström, 1999). If o¢ce-holder e¤ort and ability interact in the “production function” that determines performance in o¢ce, then an o¢ce-holder has an incentive to experiment, i.e. raise e¤ort so that performance becomes a more accurate signal of her ability. Elections reduce the experimentation e¤ect, and the reduction in this e¤ect may more than o¤set the positive “career concerns” e¤ect of elections on e¤ort. Moreover, when this occurs, appointment of o¢cials (random selection from the citizenry and tenure) may Pareto-dominate elections.