RE: Spontaneous Group Accuracy
Mon, 1 Feb 1999 08:46:02 -0800

There is a well known prediction/estimation technique called Delphi forecasting, which dates back to the 1960s. It is designed to get an improved group estimate. Here is a description, from:

: First, no participant is told the identity of the other members of the
: group, which is easily accomplished if, as is common, the forecasts are
: obtained by means of questionnaires or individual interviews. When the
: Delphi is conducted in a workshop setting-one of the more productive
: ways to proceed in many cases-this rule cannot be honored, of course.
: Second, no single opinion, forecast, or other key input is attributed
: to the individual who provided it or to anvone else. Delphi
: questionnaires, interviews, and computer conferences all easily
: provide this protection. In the workshop setting, it is more difficult
: to ensure, but it can usually be obtained by using secret ballots or
: various electronic machines that permit anonymous voting with immediate
: display of the distribution of answers from the group as a whole.
: Third, the results from the initial round of forecasting must be collated
: and summarized by an intermediary (the experimenter), who feeds these
: data back to all participants and invites each to rethink his or her
: original answers in light of the responses from the group as a whole. If,
: for example, the participants have individually estimated an event's
: probability by some future year, the intermediary might compute the
: mean or median response, the interquartile range or upper and lower
: envelopes of the estimates, the standard deviation, and so forth, and
: pass these data back to the panelists for their consideration in making
: a new estimate. If the panelists provided qualitative information as
: well-for example, reasons for estimating the probabilities as they did
: or judgments as to the consequences of the event if it were actually to
: occur-the role of the intermediary would be to edit these statements,
: eliminate the redundant ones, and arrange them in some reasonable order
: before returning them for the group's consideration.
: Fourth, the process of eliciting judgments and estimates (deriving
: the group response, feeding it back, and asking for re-estimates in
: light of the results obtained so far) should be continued until either
: of two things happens: The consensus within the group is close enough
: for practical purposes, or the reasons why such a consensus cannot be
: achieved have been documented.
: In sum, the defining characteristics of Delphi are anonymity of the
: estimates, controlled feedback, and iteration. The promise of Delphi
: was that if these characteristics were preserved, consensus within the
: panel would sharpen and the opinions or forecasts derived by the process
: would be closer to the "true" answer than forecasts derived by other
: judgmental approaches.