Wednesday, February 1, 2012

Delphi Methods (ii)



1. Finish Hands-on Delphi Method if needed

Recall the process


. . . each student
will extract the
wisdom of the group
by using
the Delphi Method


. . . (I will help as needed)

. . . Each person will also
      have a chance
      to be the recorder

. . . What are the steps?


Let's finish using the
    Delphi Method
    now . . .

. . . each person gets 
      a turn to lead it

. . . I'll help at the board

. . . each person will also 
      get to be a recorder

. . . recorder gives the 
      results to the leader

. . . leader keeps results 
      for a HW problem 




Before we start
   

Recall from last time:

(Source: JC Glenn and TJ Gordon, eds. Futures Research Methodology V3.0, chap. 4, p. 10)

But what if the # responses is not divisible by 4?

   . . . What should we do?

   . . . Conceptually:
         If there were millions of data points
         What would be the range of the middle half?
         Think about it!
         If time, could be an in-class exercise






Source: http://www.is.njit.edu/pubs/delphibook/index.html


. . . Entire books
      have been written on
      the Delphi Method



2. Let's reflect a bit . . .
We're doing it for practice

. . . how would doing it
      "professionally"
      differ?


. . . and what are the
      "so whats?" of
      these differences?


Consider:

"How social influence can undermine 

        the wisdom of crowd effect

(2011)


. . . Groups can give better estimates that individuals

. . . Expert estimates can be inferior to group medians (Galton, 1907)

But on the other hand:

. . . Crowd-sourced wisdom can go wrong

. . . Group dynamics can reduce group wisdom

. . . Examples?

. . . They did 5 rounds of re-estimates 

. . . . . . (we're doing 2)

. . . . . . Knowing others' estimates reduces range of estimates

. . . . . . . . . This can reduce quality of crowd estimates

. . . "Social influence effect":

. . . . . . reduces range without increasing accuracy

. . . . . . groups can converge without improving

. . . . . . how to minimize this problem?

. . .  "Range reduction effect":

. . . . . . true value moves to the edge of the range

. . . . . . sounds like the social influence effect but worse

. . . . . . how to minimize this problem?

. . . "Confidence effect":

. . . . . . increases confidence in, but not accuracy of, estimates

. . . . . . how to minimize this problem?

So how to do the Delphi Method so as to actually get group wisdom?



3. Ambiguity in the question
- another important issue



. . .we've noticed
     this problem

     let's look at a
     well-known
     example
     Often, public parks
     have rules

    The rules try to
    make them fun
    for everyone


. . . The Greenway Park
      Community Board
      met and proposed
      this rule:










. . . What does this
      really say . . . 
      - and how can we fix it??



No comments:

Post a Comment