What the Association Can—and Cannot—Prove

The central quantitative claim about U.S. kindergarten MMR coverage comes from a comparison of jurisdictions in the 2024–2025 school year. This chapter focuses on what that comparison estimates, how to read its magnitude and uncertainty, and where the evidence stops. The practical value is a disciplined interpretation: a strong inverse relationship between exemption prevalence and MMR coverage can identify a policy-relevant pattern, but it does not prove that changing exemptions alone will change every community’s coverage by a fixed amount.

Read the coefficient as a jurisdiction-level comparison

The study analyzed 39 jurisdictions with 2024–2025 data using a simple ordinary-least-squares, or OLS, regression. OLS fits a line through paired observations—in this case, each jurisdiction’s reported exemption prevalence and kindergarten MMR coverage. The study estimated that each 1 percentage-point increase in exemption prevalence was associated with an approximately 1.08 percentage-point decrease in MMR coverage. The reported 95% confidence interval ran from −1.34 to −0.83, the standard error was 0.13, and the p-value was less than 0.001.

The units are easy to lose in a headline. A percentage point is an absolute difference between percentages. The result does not say that a child becomes 1.08% less likely to be vaccinated for every additional exemption in a state. It says that, among the observed jurisdictions, places with higher exemption prevalence tended to have lower reported kindergarten MMR coverage, and the fitted difference was about 1.08 coverage points for each one-point difference in exemption prevalence.

That distinction becomes clearer with a hypothetical comparison. If Jurisdiction A reported an exemption prevalence two percentage points higher than Jurisdiction B, the fitted relationship would place A at roughly 2.16 percentage points lower in MMR coverage than B, all else in the model being equal. “All else” is doing important work here: the model contains no additional covariates, and the comparison is between jurisdiction-level summaries. The calculation is a way to read the slope, not a forecast for an individual child, school, or policy change.

The confidence interval describes uncertainty around the estimated slope under the model and data used. It does not give a range of outcomes for every state or a probability that a future policy will land inside those endpoints. The interval does exclude zero, which is consistent with a strong observed inverse association in this sample. Statistical strength answers whether the pattern is difficult to attribute to random variation under the model; it does not answer which mechanism produced the pattern.

The study also reports that exemption prevalence accounted for approximately two-thirds of the observed variation in jurisdiction-level MMR coverage, with R² around 0.66. In plain language, the fitted relationship tracks a substantial share of the differences in the observed coverage values. It does not mean that exemptions caused two-thirds of the national decline. It does not mean that two-thirds of each child’s vaccination status is explained by an exemption.