When faced with a series of decisions, we tend to view each decision one at a time (narrow view), instead of taking a broad view. System 2 is lazy, and rather than analyze the combinations, it is perfectly fine letting System 1 handle each decision separately. Unfortunately, this leads to irrational behaviour in the aggregate: series with better total outcomes are ignored in favour of less optimal outcomes made due to individual decisions (which are affected by loss aversion, for example).
Kahneman also discusses Mental Accounting, which causes a great deal of irrational behaviour including the disposition effect (selling winners and holding losers) and the effect of sunk costs (which keep people in poor jobs and unhappy marriages).
Interestingly, we also give different answers to questions depending on which questions are asked at the same time. For example, when given two near-identical legal cases, subjects are likely to dole out similar penalties. But if the two cases are given separately to subjects, the penalties vary tremendously based on variables (including location) that should have no bearing on the result! This was part of a series of experiments that finally got economists to talk about the idea that the perfectly rational human that underlay all models is not so perfectly rational after all in very predictable ways.