Much literature has analysed parties’ influence on public education spending. We challenge this literature on theoretical, methodological and empirical grounds. It is standard to regress expenditure on cabinet seat-share weighted party family dummies in time-series cross-section regressions using ‘country-year’ data. But using ‘country-year’ data artificially inflates the number of cases and leads to biased estimates, as governments usually do not change annually. Second, using party families as proxies for party preferences assumes that parties within families hold similar positions while they differ across families. But this is empirically often not the case. Finally, a historical institutionalist perspective suggests that we should not expect party effects anymore in the first place. Empirically, we propose a new design, using direct measures of party preferences in analyses on government-term level. We find that the partisan composition of government did not have any significant effects on education spending from 1995 to 2010 in 21 democracies.