Cascaded Inference David Lagnado Evidence Project.

Slides:



Advertisements
Similar presentations
BEHAVIORAL RESEARCH IN MANAGERIAL ACCOUNTING RANJANI KRISHNAN HARVARD BUSINESS SCHOOL & MICHIGAN STATE UNIVERSITY 2008.
Advertisements

Overview of Inferential Statistics
Naïve Bayes. Bayesian Reasoning Bayesian reasoning provides a probabilistic approach to inference. It is based on the assumption that the quantities of.
Bayesian Network and Influence Diagram A Guide to Construction And Analysis.
The Logic of Intelligence Pei Wang Department of Computer and Information Sciences Temple University.
Mean, Proportion, CLT Bootstrap
PROBABILITY. Uncertainty  Let action A t = leave for airport t minutes before flight from Logan Airport  Will A t get me there on time ? Problems :
Chapter 3 Flashcards. obligation of an individual to other individuals based on a social or legal contract to justify his or her actions; the processes.
The influence of hierarchy on probability judgment David A. Lagnado David R. Shanks University College London.
Probabilistic Models of Cognition Conceptual Foundations Chater, Tenenbaum, & Yuille TICS, 10(7), (2006)
Modeling Human Reasoning About Meta-Information Presented By: Scott Langevin Jingsong Wang.
Uncertain Reasoning CPSC 315 – Programming Studio Spring 2009 Project 2, Lecture 6.
From: Probabilistic Methods for Bioinformatics - With an Introduction to Bayesian Networks By: Rich Neapolitan.
Cognitive Psychology Chapter 7. Cognitive Psychology: Overview  Cognitive psychology is the study of perception, learning, memory, and thought  The.
Solved the Maze? Start at phil’s house. At first, you can only make right turns through the maze. Each time you cross the red zigzag sign (under Carl’s.
Introduction to Decision Analysis
Knowing Semantic memory.
Non-Experimental designs: Developmental designs & Small-N designs
Non-Experimental designs: Developmental designs & Small-N designs
Prelude to the Research Validity Lecture A RH: is a guess about the relationships between behaviors In order to test our RH: we have to decide on a research.
What is Cognitive Science? … is the interdisciplinary study of mind and intelligence, embracing philosophy, psychology, artificial intelligence, neuroscience,
Meta-analysis & psychotherapy outcome research
How Science Works Glossary AS Level. Accuracy An accurate measurement is one which is close to the true value.
Social Cognition: Thinking About People
PSY 2012 General Psychology Chapter 8: Thinking and Intelligence Samuel R. Mathews, Ph.D. Associate Professor The Department of Psychology The University.
For Better Accuracy Eick: Ensemble Learning
Quiz 4: Mean: 7.0/8.0 (= 88%) Median: 7.5/8.0 (= 94%)
DO NOW:  What is cognition (it’s okay to guess)?  Prepare your spring break extra credit to turn in (if you have it).
S519: Evaluation of Information Systems Week 14: April 7, 2008.
Soft Computing Lecture 17 Introduction to probabilistic reasoning. Bayesian nets. Markov models.
© 2009 McGraw-Hill Higher Education. All rights reserved. C H A P T E R 9 Complex Cognitive Processes.
Design Experimental Control. Experimental control allows causal inference (IV caused observed change in DV) Experiment has internal validity when it fulfills.
Bayesian Learning By Porchelvi Vijayakumar. Cognitive Science Current Problem: How do children learn and how do they get it right?
1/20 Remco Chang (Computer Science) Paul Han (Tufts Medical / Maine Medical) Holly Taylor (Psychology) Improving Health Risk Communication: Designing Visualizations.
Scientific Inquiry & Skills
One-sample In the previous cases we had one sample and were comparing its mean to a hypothesized population mean However in many situations we will use.
Chapter 2 Research in Abnormal Psychology. Slide 2 Research in Abnormal Psychology  Clinical researchers face certain challenges that make their investigations.
Introduction Osborn. Daubert is a benchmark!!!: Daubert (1993)- Judges are the “gatekeepers” of scientific evidence. Must determine if the science is.
Bayesian Networks for Data Mining David Heckerman Microsoft Research (Data Mining and Knowledge Discovery 1, (1997))
For ABA Importance of Individual Subjects Enables applied behavior analysts to discover and refine effective interventions for socially significant behaviors.
Ensembles. Ensemble Methods l Construct a set of classifiers from training data l Predict class label of previously unseen records by aggregating predictions.
Methodological Problems in Cognitive Psychology David Danks Institute for Human & Machine Cognition January 10, 2003.
Revisiting Sampling Concepts. Population A population is all the possible members of a category Examples: the heights of every male or every female the.
Perception, Cognition, and Emotion in Negotiation
Decision Making Chapter 7. Definition of Decision Making Characteristics of decision making: a. Selecting a choice from a number of options b. Some information.
Cognitive Computer Vision Kingsley Sage and Hilary Buxton Prepared under ECVision Specific Action 8-3
METU Informatics Institute Min720 Pattern Classification with Bio-Medical Applications Lecture notes 9 Bayesian Belief Networks.
Issues concerning the interpretation of statistical significance tests.
Chapter 4 Decision Support System & Artificial Intelligence.
Slide 1 UCL JDI Centre for the Forensic Sciences 21 March 2012 Norman Fenton Queen Mary University of London and Agena Ltd Bayes and.
Thinking  Cognition  mental activities associated with thinking, knowing, remembering, and communicating  Cognitive Psychology  study of mental activities.
Human and Optimal Exploration and Exploitation in Bandit Problems Department of Cognitive Sciences, University of California. A Bayesian analysis of human.
1 DECISION MAKING Suppose your patient (from the Brazilian rainforest) has tested positive for a rare but serious disease. Treatment exists but is risky.
RESEARCH METHODS IN INDUSTRIAL PSYCHOLOGY & ORGANIZATION Pertemuan Matakuliah: D Sosiologi dan Psikologi Industri Tahun: Sep-2009.
Cognition and Language. Cognition: thinking, gaining knowledge, and dealing with knowledge. I. Categorization A. Categorization: in general, we categorize.
Concepts And Generic Knowledge
Helpful hints for planning your Wednesday investigation.
#1 Make sense of problems and persevere in solving them How would you describe the problem in your own words? How would you describe what you are trying.
Probability judgement. AO1 Probability judgement ‘Probability’ refers to the likelihood of an event occurring, such as the likelihood that a horse will.
Decision Making ET 305, Spring 2016
The Psychology of Inductive Inference Psychology 355: Cognitive Psychology Instructor: John Miyamoto 5/26/2016: Lecture 09-4 Note: This Powerpoint presentation.
Irwin/McGraw-Hill © Andrew F. Siegel, 1997 and l Chapter 7 l Hypothesis Tests 7.1 Developing Null and Alternative Hypotheses 7.2 Type I & Type.
Uncertain Judgements: Eliciting experts’ probabilities Anthony O’Hagan et al 2006 Review by Samu Mäntyniemi.
Meta-analysis Overview
Hypothesis Tests l Chapter 7 l 7.1 Developing Null and Alternative
Approaches to social research Lerum
Cognition: Thinking and Language
Cognitive Processes: Thinking and Problem Solving
chance Learning impeded by two processes: Bias , Chance
David Kellen, Henrik Singmann, Sharon Chen, and Samuel Winiger
Presentation transcript:

Cascaded Inference David Lagnado Evidence Project

Plan of seminar Some ideas about cascaded inference Previous work on hierarchical inference

Cascaded inference A chain of probabilistic inferences  ‘Inference upon inference’ Pervasive in many domains  Legal  Medical  Criminal  Advertising  Everyday

Cascaded inference Horse race example WEATHER (Sun, rain, drizzle, frost…) TRACK CONDITION (Heavy, soft, good, firm…) WINNER (‘Waltzing Along’, ‘Persian Weaver’, ‘Ride the Storm’…)

Cascaded inference Legal example TESTIMONY Witness testifies that accused was at crime scene EVENT Accused was at crime scene CLAIM Accused is guilty of crime

Cascaded inference Used car example Salesman’s report Maintenance history Car reliability

Cascaded inference How do people do it? Very little recent research  Most done in  Using abstract problems  No comprehensive normative model Current psychological models also seem inadequate

Normative model for cascaded inference? Modified Bayes theorem (Dodson), Jeffrey’s rule, Chain rule Weighted sum across all possible paths ABC Given evidence A: P(C|A) = P(C|B).P(B|A) + P(C|~B).P(~B|A) A ~B C B ~C E.g., A = heavy rain; B = muddy track; C = ‘Ride the Storm’ wins P(C|B) P(C|~B) P(B|A) P(~B|A) P(~C|~B) P(~C|B)

Simplifying heuristics Various studies claim that people use simplifying strategies (as-if, best guess)  Gettys et al., 1973; Schum et al., 1973; Steiger & Gettys, 1972 Treat inference from first stage as-if it is true Focus on most probable path, ignore alternative less probable paths Such strategies lead to overestimation of final probability judgments

As-if or Best guess model A ~B C B ~C Select the most probable outcome at first stage of inference (from A to B) E.g., A = heavy rain; B = muddy track; C = ‘Ride the Storm’ wins

As-if or Best guess model Select the most probable outcome at first stage of inference (from A to B) Assume this is true for the second stage inference (from B to C) Judged probability of C given A =.7 This overestimates the correct value =.52 A ~B C B ~C.7 E.g., A = heavy rain; B = muddy track; C = ‘Ride the Storm’ wins

Current study Looking at both abstract and applied settings  Some studies claim that overestimation only occurs in abstract settings Vary presentation of probability information to see if this improves inferences  Frequency tables  Pie charts  Network diagrams?

Shortcomings with previous research Neglects structural assumptions  Causal models  Conditional independencies etc These determine appropriate normative model, and likely to influence people’s reasoning Current psychological models of probabilistic reasoning do not take these factors into account  Belief activation (neural networks)  Belief updating (rule-based)  Heuristic models

Influence of causal model How are cascaded inferences affected by assumptions about causal structure? Different causal models can imply different conditional independencies ABC A and C are (unconditionally) dependent BUT A and C are conditionally independent given B ABC A and C are (unconditionally) independent BUT A and C are conditionally dependent given B

Causal models and cascaded inference MeaslesSpotsItchy Evidence = Measles Infer that spots are probable Infer that itchiness is probable MeaslesSpotsChicken pox Evidence = Measles Infer that spots are probable Do not infer that Chicken pox is probable Evidence does not alter P(Chicken pox) Causal structure influences permissible inferences

Causal models and cascaded inference Schum – ‘conditional non-independence’ ABC E.g., A = Weather, B = Track condition, C = Winner Some horses may run better/worse in the rain, independent of the track condition

Moral Normative models for cascaded inference depend on causal and structural assumptions

Future research Construct cognitive model of cascaded inference that allows for causal and structural assumptions How well does this conform to normative models? Various studies suggest that people are poor at probability estimation/computation, but are quite good at qualitative causal reasoning Decision aids  To support inference when there are many variables, complex computations etc.  To correct for any systematic biases  Better presentation of probability information  E.g., graphical representations …

Interdisciplinary aid What are the appropriate normative models (statistics)? What are the plausible computational algorithms? Are there economical heuristic procedures? What are the cognitive mechanisms that people use (psychology, neuroscience)? Are there naturally occurring inference problems of this kind (epidemiology, forensic, legal, history)?

Some earlier studies on hierarchical inference Cascaded inference Hierarchically structured information Learning and judgment

Learning and judgment with category hierarchies Hierarchical structure  Pervasive feature of how we represent the world  Organizes knowledge and structures inference FLU Type B Type AType C

Inference using a hierarchy One powerful feature of a category hierarchy is that given information about categories at one level, you can make inferences about categories at another level. This allows you to exclude alternatives, or reduce the number you need to consider TabloidBroadsheet TimesGuardian Mirror Sun

Probabilistic inference using a hierarchy In many real-world situations we must base our initial category judgments on imperfect cues, degraded stimuli, or statistical data. What effect do such probabilistic contexts have on the hierarchical inferences that we are licensed to make? TabloidBroadsheet TimesGuardian Mirror Sun

Commitment heuristic Reason ‘as-if’ probable info is true (but reduce overall confidence) Commitment heuristic - When people select the most probable category at the superordinate level, they assume that it contains the most probable subordinate category (and vice-versa) This leads to the neglect of subordinates from the less probable superordinate TabloidBroadsheet TimesGuardian Mirror Sun

How effective is such a heuristic? Depends on the structure of the environment In certain environments it is advantageous  increases inferential power by focus on appropriate subcategories  reduces computational demands by avoiding complex Bayesian calculations But in some environments it leads to anomalous judgments and choices

Non-aligned hierarchy The most frequently read type of paper is a Tabloid, but the most frequently read paper is the Guardian (a Broadsheet). Non-aligned hierarchy: the most probable superordinate category does not contain the most probable subordinate category. Tabloid 60Broadsheet 40 Times 5 Guardian 35 Mirror 30 Sun 30

Cascaded inference Inference from hierarchy to a related category Different levels can support opposing predictions Tabloid 60Broadsheet 40 Times 5 Guardian 35 Mirror 30 Sun 30 Party A Party B

Experiments 1 & 2 What is effect of manipulating level of representation on subsequent probability judgment? Learning phase - participants exposed to a non- aligned hierarchical environment in which they learn to predict voting behavior from newspaper readership. 100 trials ‘reading/voting profiles’

Screen during learning phase Broadsheet Sun Tabloid MirrorGuardianTimes ○ Party A ○ Party B

Screen during learning phase Broadsheet Sun Tabloid MirrorGuardianTimes ○ Party A ○ Party B Reading profile for J. K.

Screen during learning phase Broadsheet Sun Tabloid MirrorGuardianTimes ○ Party A ○ Party B Reading profile for J. K. Outcome feedback

Structure of environment Tabloid 60Broadsheet 40 Times 5 Guardian 35 Mirror 30 Sun 30 Party A Party B 50 NB Overall each party equally frequent

Judgment phase Which paper is X most likely to read? X is selected at random What is the probability that X votes for Party B rather than A? Which kind of paper is X most likely to read? Baseline General Specific

Predictions Which paper is X most likely to read? X is selected at random What is the probability that X votes for Party B rather than A? Which kind of paper is X most likely to read? Baseline General Specific General choice Tabloid -> Party A Specific choice Guardian -> Party B Based on commitment heuristic

Results Which paper is X most likely to read? X is selected at random What is the probability that X votes for Party B rather than A? Which kind of paper is X most likely to read? Baseline General Specific General choice Tabloid -> Party A Specific choice Guardian -> Party B 47% 28% 75% Mean probability rating

Summary People allow their initial categorization to shift their inferences, even though all judgments are based on the same statistical data. Simplifying heuristic that assumes that environment is aligned Empowers inference when hierarchical structure is aligned, otherwise can lead to error Suggests tendency to reason as if a probable conclusion is true

Applied to Medical choices In medical settings treatment options (and survival rates) often grouped to facilitate understanding and communication Can this lead to errors? SURGERY DRUGS Type2 Type1 Type2 Type1

Medical choices Suppose success rates are non-aligned Will grouping affect people’s treatment choices? SURGERY 60% DRUGS 40% Type2 10% Type1 70% Type2 60% Type1 60%

Medical choices experiment Participants learn about success rates trial-by-trial SURGERY 60% DRUGS 40% Type2 10% Type1 70% Type2 60% Type1 60%

Medical choices experiment Ungrouped control – 75% chose most effective treatment Drug4 10% Drug3 70% Drug2 60% Drug1 60%

Medical choices experiment Grouped condition – 25% chose most effective treatment (first asked about best superordinate treatment) SURGERY 60% DRUGS 40% Type2 10% Type1 70% Type2 60% Type1 60%

Summary Grouping with a non-aligned hierarchy can lead to poorer choices How generalizable is this?  Does it depend on memory processes?  Will it apply to summary presentations? Consider situations where people must use large databases with various levels of hierarchy (e.g., NHS statistics)

Useful biases? Systematic biases reveal something about judgment and learning processes Need not be indictment of human reasoning Heuristic strategies might be well adapted to the inferential tasks that we commonly confront

Challenge for future Do people use simplifying heuristics in cascaded and hierarchical inference? If so, how can we work with these heuristics to improve human judgment?