Despite much recent interest in the flipped classroom, quantitative studies are slowly emerging, particularly in the sciences. We report a year-long parallel controlled study of the flipped classroom in a second-term general chemistry course. The flipped course was piloted in the off-semester course in Fall 2014, and the availability of the flipped section in Spring 2015 was broadly advertised prior to registration. Students self-selected into the control and flipped sections, which were taught in parallel by the same instructor; initial populations were 206 in the control section, 117 in the flipped. As a pretest, we used the ACS first-term general chemistry exam (form 2005), given as the final exam across all sections of the first-term course. Analysis of pretest scores, student percentile rankings in the first-term course, and population demographics indicated very similar populations in the two sections. The course designs required comparable student effort, and five common exams were administered, including as a final the ACS second-term general chemistry exam (form 2010). Exam items were validated using classical test theory and Rasch analysis. We find that exam performance in the two sections is statistically different only for the bottom third, as measured by pretest score or percentile rank; here improvement was seen in the flipped class across all five exams. Following this trend was a significant (56%) decrease in DFW percentage (Ds, Fs, withdrawals) in the flipped courses as compared with the control. While both courses incorporated online homework/assessments, the correlation of this indicator with exam performance was stronger in the flipped section, particularly among the bottom demographic. We reflect on the origin and implication of these trends, using data also from student evaluations.