We used inverse probability weights to correct for disproportionate sampling of particular physician subgroups when presenting data on overall physician attitudes. Because 2 major physician groups [surgeons and those who had graduated from medical school since 1990] were oversampled in our study, we divided respondents into 4 separate weighting groups: 1) nonsurgeons who graduated before 1990, 2) nonsurgeons who graduated during or after 1990, 3) surgeons who graduated before 1990, and 4) surgeons who graduated during or after 1990. We used data from the AMA Physician Masterfile to estimate the total number of U.S. physicians in each of these 4 groups when calculating our sampling weights. The probabilities used to determine these weights were 1 in 590, 1 in 513, 1 in 223, and 1 in 187, respectively. For statistical comparisons, we condensed “strongly support” and “generally support” responses into a single “support” category and “strongly oppose” and “generally oppose” responses into a single “oppose” category. We used logistic regression to assess relationships between demographic and professional characteristics and attitudes toward national health insurance. Multivariate models were adjusted for covariates chosen a priori, including sex; year of graduation; training status; specialty distinction; practice setting; practice location; and self-reported proportions of patients on Medicare, on Medicaid, and without insurance. In adjusted analyses, we dichotomized proportions of patients receiving Medicare, patients receiving Medicaid, and uninsured patients into categories of “low” and “high” using the approximate median values for each of those insurance options. We excluded neutral respondents from adjusted analyses to determine major predictors of support or opposition. Calculations were performed using the Stata statistical package, version 7.0 (Stata Corp., College Station, Texas).