What Thomas (with help from Richard and Pierre-Simon), Sir Ronald, Claude, and Daniel have to do with Information, Forest Biometrics, and Economics and how Jack got me to think about it. Western Mensurationists Meeting June 24, 2013 Greg Johnson Director, Forest Research, Weyerhaeuser NR Company
25
Embed
Western Mensurationists Meeting June 24, 2013 Greg Johnson Director , Forest Research, Weyerhaeuser NR Company
What Thomas (with help from Richard and Pierre-Simon), Sir Ronald, Claude, and Daniel have to do with Information, Forest Biometrics, and Economics and how Jack got me to think about it . Western Mensurationists Meeting June 24, 2013 Greg Johnson - PowerPoint PPT Presentation
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
What Thomas (with help from Richard and Pierre-Simon), Sir
Ronald, Claude, and Daniel have to do with Information, Forest Biometrics, and Economics and
how Jack got me to think about it.Western Mensurationists Meeting
June 24, 2013
Greg JohnsonDirector, Forest Research, Weyerhaeuser NR Company
My Assignment
How is information used in forest management? What are the key questions from a forest
industry perspective.
My Objective
Is to make you uncomfortable!
The Back-story
• The fields of mensuration and forest biometrics are well-developed and mature.
• Increases in analytical power (both computing and statistical techniques) over the past several decades have been staggering.
• Sampling theory and its application in forest measurements is robust, sophisticated, and efficient.
• The biometricians and economists are even talking to each other now! (just have to get those pesky geneticists to join the party and we’ll be set)
Leading to …
• Better information.• More accurate and precise answers to
business and social questions.• Cost-effective investments in information
technology.
Leading to …
• Better information?• More accurate and precise answers to
business and social questions?• Cost-effective investments in information
technology?
The (Rest of the) Story• We continue to see policy formulated in the absence or in
spite of sound science and data.• We still see abysmal forest and resource inventories (data
and methods) in the public and private sectors.• Planning tools (e.g., growth models) are unvalidated and
frequently produce abiological results.• It is common to see cost cutting initiatives targeting
information-gathering expenses in forestry.• A nearly complete disinvestment in commercial forest
management research in the private sector and a severe retrenchment in the public sector.
Guess What?
Our mixed-effects model accounting autocorrelation and heterogeneous variance
won’t fix it.
Why?
• Because we just don’t get it!
• Despite poor practice, decisions get made, companies make profits, policy is formed.– In each case, there are expected outcomes,
established norms, biological boundaries.– We don’t seem to be running amok.
Our first cast of characters
Presbyterian minister
Chaplain
Attended a school run at a Benedictine priory
Belief versus Outcomes
BP
APABPBAP
Two ways of using this equation:• Bayesian (belief)• Frequentist (proportion of outcomes) US
Everyone Else
Belief versus Outcomes
• From a Bayesian viewpoint, the stronger the belief, the more difficult it is to alter that belief with new evidence (observations).
• Fast versus slow thinking.– System 1: Fast, automatic, frequent, emotional,
stereotypic, subconscious.– System 2: Slow, effortful, infrequent,
logical, calculating, conscious.
Daniel Kahneman
System 1 and 2
The Müller-Lyer Illusion
Belief versus Outcomes• Combined with the Confirmation Bias managers may be
reluctant to spend money on new data when their priors are strong (they’ve seen it before, aren’t looking for differences, and “know” it won’t change).
• Intuitive heuristics: when faced with a difficult question, we often answer an easier one instead, usually without noticing the substitution.
• There is an additional bias against new data that would alter the posterior: “It is the consistency of the available information that matters for a good story, not its completeness. Indeed, you will often find that knowing little makes it easier to fit everything you know into a coherent pattern.” Daniel Kahneman
So what is information anyway?
• We like to distinguish between data and information. Data being the attributes/traits (stuff) you observe, and information being the what gets used.
• “Information” has a more formal interpretation: Given what you know, how surprised can you be in the next outcome?– Entropy quantifies the uncertainty
involved in predicting the value of a random variable.
– Information is predictive.
Claude Shannon (1916-2001)
An Example• So when we go to cruise a stand, we will often have:– An existing inventory estimate.– The time period since the last measurement.– Some expectation of stand development given it’s
prior measurement, age, etc.• When combined, these data are highly predictive– An Information Scientist would say we have low
entropy.– A Bayesian would say we have a strong prior.– A Manager would/should ask: will collecting new
data significantly change my decision/plan/etc?
Here’s our problem
• We don’t have a good answer for the manager.
• Sir Ronald can’t help us – tests of significance and the like are not helpful. We are talking about the same questions and the same data,but weighing outcomes differently. We can test hypotheses, but we are not moving the posterior.
Sir Ronald Aylmer Fisher (1890-1962)
Value of Information
• Value of Information (VOI)*:– VOI = (Vi+ - Vi- ) – Cwhere:
Vi+ = the expected monetary value of the decision with new information
Vi- = the expected monetary value of the decision without new information
C = cost of information acquisition
• Vi+ may not be known a priori – “expected” then becomes a different statistical statement.
*After Kangas 2011 and 2010
VOI: Expected Utility Theory
*Hirshliefer and Riley 1979
Prospect Theory
where:– ω = weighting function
based on outcome probabilities pi.
– ν = value function for outcomes xi.
• Losses hurt more than gains feel good.• Depends on a reference point (not considered in
expected utility theory where higher expected value decisions are simply (linearly) preferred).
in
ii xpU
1
Consequences
• Managers and decision makers often set their reference points differently than we do.
• Potential upsides for new information are discounted and downsides (costs of acquisition with little change in decision outcomes) are over-weighted.
Value of Information
• Some economists treat the uncertainty problem as Bayesians.
• The value of information is tied to beliefs.
Jack Hirshleifer
Hirshleifer and Riley, 1979
What to do … What to do …
• People (managers) preferentially use System 1 (fast) thinking, that is numerically and probabilistically inept.
• Biometricians/mensurationists attempt to inform managers using highly refined analytics useful to System 2 (slow) thinking.
• New information on its own will be less effective at changing decisions when it is aimed at fast thinking situations where there are strong beliefs (priors) and/or complex consequences (in danger of the substitution effect).
Our Challenges
• Slow the decision making process down.• Recognize the inherent Bayesianness of the decision
making process.• Recognize the nonlinear and over-weighted negative
utility of losses.• Work to develop new approaches to couch the value
of information problem in a belief-based decision process.
• Talk to our economist friends more – they discovered this a while ago (went to the dark side?).
“A mistake is not something to be determined after the fact, but in the light of the information