The key idea of democracy is that voters reward politicians for policies that benefit them, and punish them for policies that harm them. In reality, however, voters rarely know which actors (politicians or bureaucrats) are responsible for important policy changes, particularly in complex policy areas like healthcare. This creates a risk that politicians who wish to shrink the size of government can restrict access to or quality of government-subsidized healthcare, shift citizens' blame to the bureaucracy, and benefit from a backlash against government incompetence and inefficiency. In this way, anti-government politicians can gain support among the same voters their restrictive policies harm. We use data on receipt of Medicaid (government-subsidized healthcare) among 35 million US adults from 2008-2022. Using modern strategies for causal inference (a staggered rollout difference-in-difference design), we show that Republican governors implement restrictive Medicaid access measures and reduce Medicaid receipt by about 20%. We will train a machine learning algorithm to estimate the heterogeneous causal effects across different demographic groups, and will then use nationally representative data on public opinions about the size of government to test whether the same demographic groups who lose Medicaid access also shift their views towards favoring smaller government.