When you think about the first problem, you soon realize that it is hard to say exactly what a “natural” conversation is. But even if we can’t describe it, we can try to find examples of it. I experimented with several ways of making people talk that could be recognized as _un_natural: word games, list-checking, shape description and recognition, and a system of rewards for spurts of talk. None of these schemes seemed to generate the real conversational interplay we wanted. Finally, I noticed that my coworkers often got involved in vigorous conversations on political and social issues at the lunch table. So I circulated a questionnaire to help me pick pairs of people who might enter into lively discussions on one or more topics. I arranged seven conversations of this sort, and they became the basic material of my study. Only one of these lacked sufficient spirit to yield good data, and six out of seven is a pretty good percentage when you try to study human behavior in such a free situation. Note that the conversers expressed ideas that came from _within_ them at the time—not from any external materials—and that they usually felt rather strongly about what they were saying to the other fellow, who disagreed and therefore needed some convincing. Of course, these conversations do not represent the whole range of possible conversations; they are only a small sample of one type. This doesn’t limit the truth of the particular result we got, but it does limit how far we may generalize from these results. What Should We Observe? You might think we could solve this problem simply by asking the conversers their opinions. We found out long ago, however, that the opinions you get are affected by a lot of things: how you ask the question, the attitude of the respondent, and his unrelated experiences outside the experiment. So we usually try a more subtle approach. In this case, my basic observation was of what psychologists call _escape behavior_. The conversers were told that they would start talking over a normal circuit, and that delay would be introduced at some point. (The delay was inserted in such a way that an abrupt change could not be noticed.) All the conversers had pushbuttons for signaling the experimenter. If they thought they noticed a delay, they were told that it would be removed if they pushed the button. Thus they could always escape from this possibly unpleasant condition.