The rise of online polling has provided near real-time insights into voting intentions and likely electoral outcomes, making polls highly visible to political actors, the media, and the public. Research shows that polls can influence attitudes and voting behaviour, raising concerns about bias, especially when estimates deviate from election results or other pollsters.

In Spain, the Spanish Centre for Sociological Research (CIS), a public institute regularly releasing pre-election estimates, has faced criticism for allegedly overestimating the vote share of ruling parties since 2018, when a new director with ties to the socialist party PSOE took over under the PSOE-Sumar government. We analyse CIS pre-election estimates from 2012 to 2024, a period marked by alternance in government and CIS's directorship, covering 69 regional and national elections, to assess time trends in CIS’s estimate accuracy. We specifically focus on estimates that exceed the margin of error. Our Bayesian change point analyses and regression discontinuity models show that, particularly since 2018, CIS has consistently overestimated the vote share of PSOE and its allies, though this does not involve underestimating the conservative PP, or its allies. A synthetic control analysis supports the causal nature of this relationship: using estimates from other Spanish pollsters, we construct a synthetic control to benchmark CIS’s estimation, revealing that no such systematic misestimation of specific parties occurs in this counterfactual scenario.

This highlights how political control over polling institutions may affect the accuracy of election forecasts, raising questions about accountability and public trust in state-sponsored polling.