We present a new distributed approach that establishes reputation-based trust among sensor nodes in order to identify malfunctioning and malicious sensor nodes and minimize their impact on applications. Our method adapts well to the special characteristics of wireless sensor networks, the most important being their resource limitations. Our methodology computes statistical trust and a confidence interval around the trust based on direct and indirect experiences of sensor node behavior. By considering the trust confidence interval, we are able to study the tradeoff between the tightness of the trust confidence interval with the resources used in collecting experiences. Furthermore, our approach allows dynamic scaling of redundancy levels based on the trust relationship between the nodes of a wireless sensor network. Using extensive simulations we demonstrate the benefits of our approach over an approach that uses static redundancy levels in terms of reduced energy consumption and longer life of the network. We also find that high confidence trust can be computed on each node with a relatively small memory overhead and used to determine the level of redundancy operations among nodes in the system.