This paper examines the role of multiple measures of performance in a principal-agent model incorporating both moral hazard and adverse selection. The outcome of interest to the principal depends stochastically on the agent's unobservable ability and effort, while the principal implements a contract contingent on two noisy measures of the outcome. There are three main findings. First, the weights assigned the performance measures are reduced in the presence of adverse selection because the

more »

... mational rent paid to the agent lowers the return to the principal of hiring the agent, but it does not affect how informative one signal is relative to another. Second, the weights assigned the signals are decreasing in the sensitivity of performance to ability. Third, a signal is assigned more weight if and only if it is more precise and sensitive to the agent's effort; thus, the Banker and Datar (1989) result is robust to the introduction of adverse selection. An empirical test of the model is provided in the context of the CEO pay-for-performance sensitivity and the investment opportunities set (IOS) of the firms they manage. If high IOS firms are more ability-intensive, the model predicts the weights on the performance measures are decreasing in IOS. We examine a sample of 12,221 firm-year observations for 1,411 firms spanning the period 1992-2006 obtained from ExecuComp, CRSP, and Compustat. In agreement with the model, we find that CEO compensation is less sensitive to accounting and stock returns in high IOS firms.