It is common for political scientists to investigate the degree to which partisanship affects government policies --- especially expenditure. This paper advances the view that such partisanship effects are likely to be conditioned by strategic calculations about the probability of future policy changes. Defining the partisanship effect as the difference between the policy that would be implemented by left-wing and right-wing governments, a formal model is developed that relates the size of this effect to two parameters: the electoral bias faced by left-wing parties and the degree to which today''s policies are likely to be rolled back by a future government. In some cases when they face a negative electoral bias, left-wing parties spend `inefficiently'' highly, which itself exacerbates their electoral difficulties. The model yields an unanticipated non-linear hypothesis that finds support when tested with welfare expenditure data across `developed democracies''.
