about the equivalent of two standard deviations on the income distribution. Even compared to the widely
cited predictors of political knowledge, the experimental effects look big. Together, they amount to two
thirds of the difference between respondents who follow politics “only now and then” or “hardly at all”
(the omitted categories in table 4) and the most interested respondents (“most of the time”). Similarly, the
difference between a college graduate and a respondent with only a high school education is about the
same as the effect of time and pay.
Why Do Time and Money Make a Difference? And for Whom?
We know that our experimental treatments led some respondents to take more time in completing the
survey. Perhaps some of them used that time to ask someone else or look up some answers. If extra time
alone explains the experimental effects, they should disappear when we control for the lengths of the
interview (or, with our imperfect measures of response time and heterogeneity in the relationship between
interview time and performance, such effects should at least decline substantially.) They do not. Adding
logged interview time and its square lower the two main effects by about a fifth. Using a set of dummies
for different lengths of the interview has an even smaller effect. Perhaps these effects are negligible only
because we do not know how much time respondents spent on the knowledge questions specifically.
More likely, the main effect of monetary incentives and even additional time to complete the questions
was in motivating respondents to take the questions seriously, think about them, and try harder than usual
to answer them correctly.
To understand what explains the effect of our experimental treatments, we examine if and how they
changed the role of other predictors of political knowledge. Respondents in the “24 hour” conditions can
perform well on the knowledge test even if they do not know many answers when they first read the
questions (or cannot retrieve the information from memory fast enough to maintain the normal pace of a
survey interview.) Accordingly, access to information resources and skills in finding the correct answers
6
The education variables are coded so that “College Degree” is the additional effect of having at least a college
degree over having at least a high school degree. The total effect of at least a college degree compared to someone
without even a high school degree would be the sum of the two coefficients (.60 + .98 = 1.58).
17
Convention