The effectiveness of Peer Instruction is often associated to the importance of in-class discussions between peers. Typically, a greater number of students have correct answers after peer discussions. However, other cognitive and metacognitive processes such as reflection or time-on-task may also explain this increase because students answering conceptual questions reflect more and spend more time thinking about their understanding. An identical sequence of conceptual questions was given to three groups of students. All groups were polled twice on each question. Between polls, students were asked either to discuss their choice with a peer, or to reflect for a minute (no discussion), or were given a distraction task (sequence of cartoons: no discussion and no reflection). Increases in the rates of correct answers between the first and the second poll were found across all conditions. The `Distract' condition had a small but positive increase (3.4%). The `Reflect' condition had a greater increase (9.7%) while the `Discuss' condition had the greatest (21.0%). All conditions showed gains, possibly because of `testing effects', though peer-discussions clearly yield greatest increases. Our findings show that learning gains through peer discussions cannot be explained only by additional time-on-task or self-reflection.