This article takes as given the accuracy of opinion polls. It offers instead a suggestion towards improving forecasts of elections where traditional assumptions may not be valid. What is offered is a consistent, and ex-ante correction, to polling data. This correction is based on an intuitively simple and age old assumption - people lie. Not all the time, and not even most of the time. Indeed, they lie very rarely. But when they do, they can cause serious errors. And given the high price of going wrong, it pays pollsters to heed, and calculate, a "lying index". If the basic premise of the lying index is accepted, the question arises : can it be measured, and measured successfully ? Success involves several dimensions, but the test should pass two in particular: it should broadly reflect lying when there is lying, and not reflect lying when there is no lying. In other words, there should be no "bias"in the estimate of lying.