To identify clicker questions likely to provoke rich and engaging student discussions, we recorded the percentage of the class voting on each option on every clicker question used throughout each semester, encompassing nineteen sections of introductory statistics, taught by three instructors. Working from the hypothesis that seeing student votes broadly distributed across several different options is an indication of a good discussion question, we present the four questions with the most broadly distributed votes and discuss their use. Next, we present and consider the educational value of the questions with the lowest percentage of students voting correctly. We find that both broadly distributed votes and a low percentage of students voting correctly can be good indicators of effective discussion questions. Further, we find that questions devoid of numbers and based on general cases have fewer students voting correctly and have a significantly broader distribution of votes, thus indicating that this characteristic may be an important indicator of good discussion questions.