Does Knowledge of Constitutional Principles Increase Support for Civil Liberties? Results from a Randomized Field Experiment

1,215 students assigned (627 treatment in 30 courses; 588 control in 29 courses); 25 teachers in 10 high schools; analysed N = 1,045 (2007), 591 (2008), 508 (2009)

US high-school civics students in 10 high schools, including many students described as less academically inclined and unlikely to attend college. Followed for two years post-treatment.

Cluster-randomized field experiment with 19 randomization strata (groups of courses taught by one teacher or cluster of teachers); classrooms, not students, were assigned, and analysis uses GLS with classroom random effects and strata fixed effects. The treatment was 21 lessons - 'The Bill of Rights for Real Life' - developed for the study by the Bill of Rights Institute, delivered as a supplement to the regular civics course; teachers took on average one and a half months to complete it, and the curriculum itself emphasises classroom DISCUSSION and active participation, i.e. the very pedagogy the observational literature credits with producing tolerance. The control is an ACTIVE alternative: conventional civics instruction. Attrition was balanced (2007: 87.4% control vs 84.7% treated completed; by 2009 only ~53% in both arms, because schools left the study). A randomization check found slight imbalance (6.3% of 20,000 simulated assignments were more extreme), which the authors show is immaterial - results are unchanged with and without covariates. NOTE ON THE TEST'S DIRECTION: the authors deliberately report the REDUCED FORM (assignment to attitudes) rather than an IV estimate, which they explain is the CONSERVATIVE choice - it credits to the knowledge channel any effect the curriculum might have had on attitudes directly through discussion. The null therefore cannot be rescued by saying the mediation model was wrong.

The curriculum taught the content, taught nothing else, and changed no attitudes - and the knowledge gain then faded. Civil Liberties Knowledge Index: +1.220 correct answers out of 15 (SE 0.155, p<0.001, Cohen's d = 0.49) in 2007; +0.762 out of 18 (SE 0.226) in 2008; +0.221 out of 16 (SE 0.269), NOT significant, in 2009. Calibrating across waves, the control group mean was flat while the treatment group showed an initial surge and then decayed to the control mean - textbook fadeout on a domain-skill outcome within two years. General Political Knowledge Index: essentially zero in every wave (0.023, SE 0.076; 0.051, SE 0.259; -0.037, SE 0.202), so the gain was entirely domain-specific with no transfer and no crowding-out. Support for Civil Liberties Index: -0.071 (SE 0.051) in 2007, -0.073 (SE 0.131) in 2008, -0.080 (SE 0.119) in 2009 - null, and if anything slightly negative, in every wave including the waves where knowledge was significantly higher. Item-by-item regressions on each tolerance question produced no significant effect either. An open-ended applied-reasoning essay showed a small significant gain in 2007 (+0.182 constitutional principles identified, SE 0.068) that was gone by 2008 (-0.007, SE 0.128). The authors: 'it is possible to increase awareness and understanding of civil liberties without producing an increase in support for those civil liberties', and this 'calls into question the longstanding argument that beliefs and attitudes are causally linked in this domain'. They stress the experiment had ample power to detect even small effects.

Low. Classroom-level randomization within teacher-defined strata balances heritable ability and family political socialisation across arms. The attitude null is a null under a design that detected a d=0.49 knowledge effect in the same students on the same questionnaire, which forecloses the 'instrument-too-blunt' explanation. The authors also note the support-for-civil-liberties index, though noisy (alpha 0.31-0.45), correlates with knowledge cross-sectionally exactly as the observational literature reports - so it reproduces the correlation and still shows no causal effect.

No direct replication exists. The FINDING - knowledge moves, attitudes do not - is independently echoed by Campbell and Niemi 2016 (state civics exams raise knowledge but not attitudes or participation), by Persson et al. 2020 (deliberative teaching moves nothing), and by Jung and Gopalan 2024 (civics-test mandates do not move turnout). Campbell's 2019 review flags this specific result as the one most in need of replication. Note the study also independently undercuts Dee 2004: Green et al. explicitly observe that Dee's instrumental-variables tolerance result 'may be an artifact of unmeasured factors that are related to both labor laws and levels of tolerance', and their experiment finds no tolerance effect.