Tufte's better known for his TheVisualDisplayOfQuantitativeInformation, but here one gets a chance to see him use graphics to make a point, instead of a meta-point about graphics. His thesis is that there is a strong correlation between the timing of government expenditures and election years, and furthermore, in the context of the temptation to buy votes, this correlation likely indicates causation. In particular, the timing of increases in veterans' benefits is particularly damning for presidential elections.
Of interest is also his observation that politicians' goals and public policy are at times at odds. Because voters remember recent changes, it is better to reduce unemployment near the next election, rather than reducing it at the beginning of one's term. This presupposes that politicians control enough government expenditures to influence the economy, which is currently true in the US.