Document Sample
mpa Powered By Docstoc
					  De-biasing Anchoring and
Adjustment: The Use of Multiple
            Bridgett Milner
                Ed Hirt
          MPA Presentation
              Thanks to:
  Amy Johnson, Kelly Koch, Erin Steury,
          and Kristin Hendrix
     Anchoring and Adjustment!
                     (a brief review)

• The anchoring and adjustment heuristic
  involves the significant influence of an
  arbitrary reference point (an anchor) on
  estimates and judgments (Tversky and
  Kahneman, 1974).
o Is Mt. Everest taller or shorter than 10 miles?
  How tall is Mt. Everest?
     • 5.5 miles
        More on Anchoring and
• Anchoring and adjustment allows for judgments
  to be made under uncertainty conditions,
  through allowing the individual (most often) a
  reasonable starting point from which to adjust
  (Tversky and Kahneman, 1974).
• The influence of this anchor is such that
  judgments are insufficiently adjusted away from
  this reference point (the anchor) toward the true
• The use of this strategy, or heuristic,
  typically causes a bias in estimates
  reflected in this insufficient adjustment,
  such that judgment is biased toward the
  initial values presented (the anchor). Bias
  becomes more extreme as anchor points
  become more extreme relative to the true
   Standard Anchoring Procedure
• Most often demonstrated through the standard
  anchoring paradigm in which participants are
  given an anchor to which they make two
  responses. These responses are:
  1. a “comparative” judgment
    • asked to compare the anchor to their estimate or target
      response (most often asked to judge whether the actual,
      true value is above or below the supplied anchor)
  2. an “absolute” judgment
    • asked to give their best estimate of the target response
      (or the “true” value).
  Example of Standard Anchoring
• Is the percentage of African Nations in the
  UN greater or less than ____? (e.g.,65%
  or 10%)
  – Respond
• What is the number of African Nations in
  the UN?
  – Respond (mean estimate of high anchor =
    45%, mean estimate of low anchor= 25%)
• Regardless of type of anchoring task used,
  findings are quite robust. Anchoring occurs in
  many paradigms and with many types of
  estimates including those on issues individuals
  care deeply about such as nuclear war
• Anchoring occurs even when the anchor
  provided is irrelevant to the study question
  (such as an assigned participant number or a
  completely unrelated statistic) or so extreme as
  to be completely implausible as a response.
            Really Robust
• Anchoring has been demonstrated in
  information rich real world settings and
  with experts in the anchoring estimation
  task (Northcraft and Neale, 1987).
   Avoiding the Bias—Background for
               this Study
• Use of the anchoring and adjustment heuristic in
  decision making seems fairly unavoidable.
  – laboratory experiments find that participants are often
    completely unaware that an anchor influenced their
    response (Tversky and Kahneman, 1974; Northcraft and
    Neale, 1987, many others).
  – when told about anchoring and adjustment and warned to
    avoid the use of this heuristic, effects remain (Plous, 1993).
  – attempts at accuracy such as monetary incentives for
    correct judgments have no effect (Wilson, Houston, Eitling, &
    Brekke, 1996).
  – To date, within traditional anchoring studies, the only time in
    which bias was not found was when participants had not
    paid attention to the supplied anchor (Wilson et al,
    1996)…well sort of (we’ll get to that)
         Mechanism: SAM – More
         Background for this Study
• Mussweiler and Strack’s Selective Accessibility
  Model is used to account for anchoring findings.
  – When given an anchor, initially set value as equal to
    anchor. Test this using a positive test strategy
    which selectively scans ones memory for
    supporting evidence. This scan causes confirming
    evidence to become more available. This available
    information is used to make estimate.
  – More robust than semantic priming due to self-
    generated nature of prime.
   A new attempt to de-bias (at least
         in part): This study!
• Perhaps can de-bias anchoring and adjustment
  in the same way one de-biases the explanation
  bias in social judgment---through the “consider
  the opposite technique” (Lord, Lepper, &
  Preston, 1984; Mussweiler, Strack, & Pfeiffer,
  2000) or “consider any plausible alternative”
  (Hirt & Markman, 1995).
  – De-biases through priming stimuli other than the
    ones previously assessed (Hirt & Markman, 1995)---
    participants use ‘simulation heuristic’ to engage in
    “multiple simulation runs of potential outcomes”.
       Consider the Opposite
   (or the Red Sox not winning against the Yankees)

• Within social judgment (and anchoring)
  considering the opposite outcome (or
  anchor) de-biases social judgments.
  – When first asked to think about “why might the
    Red Sox win against the Yankees” –
    judgments become biased and people think
    “yeah, the Red Sox are going to win.”
  – If asked to then think about “why the Red Sox
    might not win against the Yankees”--
    judgments now become unbiased.
          Consider an Alternative
               (or Carrie winning American Idol)

• Interestingly, work by Hirt & Markman (1995) finds that
  you needn’t just consider the opposite to produce de-
  biasing effects though. They found that considering any
  alternative outcome (even when that outcome was in
  the same direction as the original one) produced de-
  – So in their paper, when participants explained a team winning
    by both a close-margin and a landslide, de-biasing occurred!
    Also when explain 2 teams winning de-biasing occurs!
  – So think about Bo winning American Idol---explanation bias
    leads to individuals thinking this is more likely---now think
    about Carrie winning---de-biased.
        This study’s hypotheses
1. Traditional anchoring effects will occur in
   standard traditional anchoring questions.
2. When participants “consider the opposite,” de-
   biasing will occur.
3. When participants “consider an alternative,” de-
   biasing will occur
• 2 conditions (1 control; 1 experimental)
  – Control condition asked trivia questions (given no
  – Experimental condition –standard anchoring
    paradigm--(5 different questionnaires all with same
    15 trivia questions---each participant responds to:
     •   3 high anchor questions
     •   3 low anchor questions
     •   3 2-high anchor questions
     •   3 2-low anchor questions
     •   3 1-high 1-low anchor questions
           Additional Measures
• Prior to completing the anchoring questions
  participants completed a short questionnaire
  asking about individual difference information.
• Participants responded to
  – demographic information (age, gender, and race)
  – A self-report of “trivia-knowledge” on a 1-7 scale
  – The NFC scale
1.    When did the Titanic sink?
        1912 Mean Cntrl=1910 anch: 1850, 1867, 1947, 1954
2.    How long is a blue whale at birth (in feet)?
        25      19.45          2,     4,     44,     60
3.    How many hours does it take for a snake to digest a frog (on
        50       21.2          1,     2,     34,      42
4.    How many eggs does the average American consume each
        286     236.4         24,     30,    400,     500
5.    In what year was the planet Neptune discovered?
        1846   1829.6        1601,   1756,   1972,   1974

• Overall, differences between the 5
  experimental conditions were found
  F(4,800)=390.44, p<.001.
• Post hoc t-tests follow on the upcoming
  slides as do t-tests comparing
  experimental conditions with control
     Traditional Anchoring Effects
Traditional anchoring effects were found
  – T-test comparing (z-scores of) high anchoring
    conditions and control conditions
     • t (293)= -6.380, p<.001
        – High anchor mean = .6002
        – Control mean =    -.0889
  – Low conditional and control
     • t (297) = 5.878, p<.001
        – Low anchor mean = -.5853
        – Control mean =    -.0889
            Consider the Opposite
Considering the Opposite was seemingly an effective de-
 biasing technique.
  – Control versus High and Low anchor given btw Ss t-test
     • t (299) = -.834, p=.405
         – Control mean =            -.0889
         – High and Low anchor mean = .0046
  – High and Low anchor given leads to significantly less high
    estimates than high anchor condition and significantly higher
    estimates than low anchor condition.
     • t (217) = 11.725, p<.001 (high vs both w/in Ss t-test)
     • t (221) = -12.043, p<.001 (low vs both w/in Ss t-test)
         – High mean = .5287
         – Both mean = .0008
         – Low mean = -.5551
      Consider an Alternative
• Considering 2 anchors in the same
  direction did not de-bias overall but
  instead led to even more severe biasing
  than single anchors .

    • t (216) = -4.011, p<.001 (high vs 2-high w/in Ss t-
    • t (221) = 2.449, p<.05 (low vs 2-low w/in Ss t-test)
       –   High mean = .5276
       –   2-High mean = .6687
       –   Low mean = -.5556
       –   2-Low mean = -.6472
  More on Consider an Alternative
• Recent follow up research by Hirt, Kardes, &
  Markman (2004) finds that this consider an
  alternative de-biasing strategy is moderated by
  the individual difference “need for structure”
  such that:
  – individuals low in need for structure are more
   affected by this de-biasing technique than those
   who are high.
• Within their research they find in fact that for
  low need for structure individuals, the
  alternative considered needn’t even be in the
  same domain for effective de-biasing to occur.
             Need for Structure
• Need for structure is characterized by the desire to
  maintain with some extent of permanence a specific
  solution that has been seized upon.
• This personality characteristic is measured by facets
  1, 2, and 4 of the Need for Closure scale (which can
  be broken into 5 different facets [3 being decisiveness
  and 5 being closed-mindedness]) (Neuberg, Judice, &
  West, 1997).
• Past research has found that this individual difference
  (NFS) leads to stronger anchoring effects (Kruglanski).
       Need for Structure Results
• Significant differences between high and low
  NFS participants were found across the 4
  experimental conditions in which bias was
  found (low, 2 low, high, & 2 high anchors). This
  difference was always that individuals low in
  NFS were significantly less biased than
  individuals high in NFS.
• Low: F(1, 257)=6.41, p<.05         high=-.61 low=-.48
• High: F(1,253)=6.85, p<.01         high= .63 low=.48
• 2 Low: F(1, 254)=109.23, p<.001 high=-.78 low=-.41
• 2 High: F(1, 254)=109.23, p,.001 high= .94 low=.36
                 More on NFS
• For the high and low anchor conditions
  (consider the opposite), all participants were
  de-biased, showing no differences from the
  control condition.
• For conditions in which 2 anchors were
  presented in the same direction (consider an
  alternative) :
  – For low NFS participants, de-biasing effects were
    found such that when 2 anchors were presented
    participants showed less bias than in the traditional
    single anchor conditions
     • Low: t(114)=1.17, p=.09 1=-.48 2=-.41
     • High: t(111)=2.83, p<.01 1= .49 2= .36
– For high NFS participants, presenting 2
  anchors in a single direction worked in the
  opposite direction such that, rather than de-
  biasing estimates, estimates became more
   • Low: t(131)=3.18, p<.01   1=-.61 2=-.78
   • High: t(135)=4.64, p<.001 1= .63 2= .94
   Discussion---Hypotheses revisited
1. Traditional anchoring effects will occur in
   standard traditional anchoring questions.
2. When participants “consider the opposite,” de-
   biasing will occur.
3. When participants “consider an alternative,”
   de-biasing will occur.
  Discussion---Hypotheses revisited
 Traditional anchoring effects will
  occur in standard traditional
  anchoring questions.
   Single high anchor led to significantly
    higher estimates than control and
    single low anchor led to significantly
    lower estimates than control.
2. When participants “consider the opposite,” de-
   biasing will occur.
3. When participants “consider an alternative,” de-
   biasing will occur.
      Discussion---Hypotheses revisited
    Traditional anchoring effects will occur in standard traditional anchoring

 When participants “consider the opposite,” de-
  biasing will occur.
      Consider the opposite condition estimates did not
       significantly differ from control estimates.
3.   When participants “consider an alternative,” de-biasing will occur
     Discussion---Hypotheses revisited
    Traditional anchoring effects will occur in standard traditional anchoring questions.
    When participants “consider the opposite,” de-biasing will occur.

 When participants “consider an alternative,” de-
  biasing will occur.
     Consider an alternative ONLY worked as a de-biasing
      strategy for participants low in need for structure. For
      individuals high in need for structure, considering two
      anchors in the same direction actually led to more extreme
      anchoring effects.
     Individuals high in NFS (consistent with Kruglanksi’s
      findings) made significantly more biased estimates in the
      direction of the anchor (or anchors) than individuals low in
      need for structure
• De-biasing of anchoring (like the explanation
  bias) using a consider an alternative strategy is
  moderated by the individual difference NFS.
  – For individuals low in NFS, considering an
    alternative is effective as a de-biasing technique.
  – For individuals high in NFS, de-biasing is much
    more difficult. For these individuals considering an
    alternative does not work. To de-bias individuals
    high in NFS one must really be extreme in their de-
    biasing strategy and provide an opposite anchor
    rather than any other anchor.

Shared By: