| Abstract: |
Survey questions that elicit point predictions regarding uncertain events face
an important challenge as human forecasters use various statistics to
summarise their subjective expectations. In this paper, we take up the
challenge and study whether alternative formulations of the questions used to
elicit point predictions can be successful in driving forecasters towards
reporting a particular central tendency (median or mean) of their subjective
expectations distribution. We set up a laboratory experiment in which the
participants act as forecasters and are asked to predict the next realisation
of iid random draws coming from an objectively known distribution. We elicit
the subjects' point predictions in four treatments, in which we ask for either
(1) a "guess" of the next draw, as is standard in survey measures, (2) a
"guess" as close as possible to the next 6 draws, and (3) the mean, or (4) the
median of the next six draws. We then compare the predictions reported in the
different treatments and their proximity to the three main central tendencies
(mean, median, mode) of the objectively known distributions. We also
investigate the cognitive process that affects the production of point
predictions. We find that the majority of predictions in the two guess
treatments, (1) and (2), are close to the mode. In treatment (2) ("one guess
for the next six draws"), the forecasters report the mean and the median more
often in comparison to (1) ("guess for the next draw"), but the mode remains
the central tendency around which most of the predictions are located. In
treatments (3) and (4), we find that forecasters adjust the point they report
in the direction of a particular central tendency when specifically asked to
report the mean or the median. Adjustments are more precise for forecasters
with higher measures of numeracy and for those who have more experience.
However, numeracy has no explanatory power when the forecasters are asked to
report a "guess for the next draw" in treatment (1) which suggests that
forecasters have different ways to summarise a distribution. |