ABSTRACT
A randomized trial of fall rates is presented with an intention-to-treat analysis. Many in the intervention arm attended only some of the exercise classes which constituted the main intervention. Can we do an unbiased analysis of the effect of treatment received? To do this we need an instrumental variable, a measurement that is related to the treatment received, but unrelated to the outcome except through the treatment. In a randomized trial, we have such a measurement – the random assignment. An instrumental variable analysis can produce an unbiased estimate of the effect of actually receiving treatment. The reasoning behind instrumental variable analysis is justified using a simple algebraic example. Then two-stage linear regression and instrumental variable linear regression are shown. Next, generalized methods of moments for rates is presented. Limitations of instrumental variable analysis are discussed.
