Using imprecise estimates for weights Alan Jessop Durham Business School.

Slides:



Advertisements
Similar presentations
STATISTICS HYPOTHESES TEST (III) Nonparametric Goodness-of-fit (GOF) tests Professor Ke-Sheng Cheng Department of Bioenvironmental Systems Engineering.
Advertisements

Statistical vs Clinical or Practical Significance
Overview of Lecture Partitioning Evaluating the Null Hypothesis ANOVA
Quantitative Methods Lecture 3
STATISTICAL INFERENCE ABOUT MEANS AND PROPORTIONS WITH TWO POPULATIONS
Chapter 7 Sampling and Sampling Distributions
HYPOTHESIS TESTING. Purpose The purpose of hypothesis testing is to help the researcher or administrator in reaching a decision concerning a population.
Chapter 7 Hypothesis Testing
Estimation of Means and Proportions
“Students” t-test.
Module 16: One-sample t-tests and Confidence Intervals
Inferential Statistics
Statistical Inferences Based on Two Samples
Chapter 18: The Chi-Square Statistic
Chapter 8 Estimation Understandable Statistics Ninth Edition
CHAPTER 15: Tests of Significance: The Basics Lecture PowerPoint Slides The Basic Practice of Statistics 6 th Edition Moore / Notz / Fligner.
CHAPTER 14: Confidence Intervals: The Basics
Chapter 16 Inferential Statistics
Lecture (11,12) Parameter Estimation of PDF and Fitting a Distribution Function.
Objectives 10.1 Simple linear regression
CHAPTER 25: One-Way Analysis of Variance Comparing Several Means
CHAPTER 25: One-Way Analysis of Variance: Comparing Several Means ESSENTIAL STATISTICS Second Edition David S. Moore, William I. Notz, and Michael A. Fligner.
CHAPTER 21 Inferential Statistical Analysis. Understanding probability The idea of probability is central to inferential statistics. It means the chance.
Sampling Mathsfest Why Sample? Jan8, 2003 Air Midwest Flight 5481 from Douglas International Airport in North Carolina stalled after take off, crashed.
Confidence Interval and Hypothesis Testing for:
Bayesian inference Gil McVean, Department of Statistics Monday 17 th November 2008.
Maximum Likelihood. Likelihood The likelihood is the probability of the data given the model.
1 Statistical Inference H Plan: –Discuss statistical methods in simulations –Define concepts and terminology –Traditional approaches: u Hypothesis testing.
Chapter 7 Sampling and Sampling Distributions
IEEM 3201 Two-Sample Estimation: Paired Observation, Difference.
Inference.ppt - © Aki Taanila1 Sampling Probability sample Non probability sample Statistical inference Sampling error.
Using ranking and DCE data to value health states on the QALY scale using conventional and Bayesian methods Theresa Cain.
Chapter 9 Hypothesis Testing.
Are the results valid? Was the validity of the included studies appraised?
Estimation and Hypothesis Testing Faculty of Information Technology King Mongkut’s University of Technology North Bangkok 1.
Copyright © Cengage Learning. All rights reserved. 13 Linear Correlation and Regression Analysis.
COLLECTING QUANTITATIVE DATA: Sampling and Data collection
McGraw-Hill/Irwin Copyright © 2007 by The McGraw-Hill Companies, Inc. All rights reserved. Statistical Inferences Based on Two Samples Chapter 9.
Confidence Intervals Confidence Interval for a Mean
Andrew Thomson on Generalised Estimating Equations (and simulation studies)
Assume we have two experimental conditions (j=1,2) We measure expression of all genes n times under both experimental conditions (n two- channel.
First approach - repeating a simple analysis for each gene separately - 30k times Assume we have two experimental conditions (j=1,2) We measure.
VI. Evaluate Model Fit Basic questions that modelers must address are: How well does the model fit the data? Do changes to a model, such as reparameterization,
Agresti/Franklin Statistics, 1 of 122 Chapter 8 Statistical inference: Significance Tests About Hypotheses Learn …. To use an inferential method called.
Academic Research Academic Research Dr Kishor Bhanushali M
Marshall University School of Medicine Department of Biochemistry and Microbiology BMS 617 Lecture 13: One-way ANOVA Marshall University Genomics Core.
Various Topics of Interest to the Inquiring Orthopedist Richard Gerkin, MD, MS BGSMC GME Research.
Education 793 Class Notes Inference and Hypothesis Testing Using the Normal Distribution 8 October 2003.
SAMPLING DISTRIBUTION OF MEANS & PROPORTIONS. SAMPLING AND SAMPLING VARIATION Sample Knowledge of students No. of red blood cells in a person Length of.
SAMPLING DISTRIBUTION OF MEANS & PROPORTIONS. SAMPLING AND SAMPLING VARIATION Sample Knowledge of students No. of red blood cells in a person Length of.
Hypothesis Testing. Statistical Inference – dealing with parameter and model uncertainty  Confidence Intervals (credible intervals)  Hypothesis Tests.
Copyright © 2009 Pearson Education, Inc t LEARNING GOAL Understand when it is appropriate to use the Student t distribution rather than the normal.
1 Life Cycle Assessment A product-oriented method for sustainability analysis UNEP LCA Training Kit Module k – Uncertainty in LCA.
Lecture Slides Elementary Statistics Twelfth Edition
HYPOTHESIS TESTING.
Tutorial 11: Hypothesis Testing
Hypothesis Testing Hypothesis testing is an inferential process
Chapter 9 Hypothesis Testing.
Tests of Significance The reasoning of significance tests
Hypothesis Testing and Confidence Intervals (Part 1): Using the Standard Normal Lecture 8 Justin Kern October 10 and 12, 2017.
HLM with Educational Large-Scale Assessment Data: Restrictions on Inferences due to Limited Sample Sizes Sabine Meinck International Association.
Chapter 8 Hypothesis Testing with Two Samples.
CPSC 531: System Modeling and Simulation
Ten things about Inference
12 Inferential Analysis.
CHAPTER 11 CHI-SQUARE TESTS
STA 291 Spring 2008 Lecture 18 Dustin Lueker.
Carey Williamson Department of Computer Science University of Calgary
STA 291 Spring 2008 Lecture 13 Dustin Lueker.
Statistical Inference for the Mean: t-test
Presentation transcript:

Using imprecise estimates for weights Alan Jessop Durham Business School

Motivation In a weighted value function model weights are inferred from judgements. Judgements are imprecise and so, therefore, weight estimates must be imprecise. Probabilistic weight estimates enable the usual inferential methods, such as confidence intervals, to be used to decide whether weights or alternatives may justifiably be differentiated.

Testing & sensitivity Single parameter. Results easily shown and understood. But partial. Multi parameter. Examine pairs (or more) to get a feel for interaction. Global (eg. Monte Carlo). Comprehensive but results may be hard to show simply. Using some familiar methods uncertainty can be inferred from judgements and the effects of global imprecision can be shown. An analytical approach rather than a simulation.

Sources of imprecision Statements made by the judge are inexact. This is imprecise articulation: variance = σ a ² The same judgements may be made in different circumstances, using different methods or at different times, for instance. This is circumstantial imprecision: variance = σ c ²

articulation single pointthree-point degrees of freedom implicit in questions 0none available a2a2 >0 c2c2  c 2 +  a 2 Sources of imprecision Redundancy e.g. ask at different times reciprocal matrix No redundancy e.g. simple rating related?

3 point estimate:  a 2 μ = aM + (1-a)(L+H)/2 σ a = b(H-L) Previous studies for PERT analyses. Generalise as a = 1.800x c b = c Beta distribution But because  w = 1 variances will be inconsistent. Solution: fit a Dirichlet distribution.

Dirichlet f(W) = k  i w i u i -1 ; 0<w i <1,  i w i = 1, u i  0,  i wherek =  (  i u i ) /  i  (u i ) which has Beta marginal distributions with mean  i = u i / v variance  i ² = u i (v- u i ) / v²(v+1) =  i (1-  i ) / (v+1) and covariance  ij = -u i u j / v²(v+1) = -  i  j / (v+1) ; i≠j wherev =  i u i Relative values of parameters u i determine means. Absolute values determine variance via their sum, v.

Dirichletset parametersweight values Dirichlet judgements: marginal characteristics mean e i and variance s i 2 consistent variances  i ² Usually used by specifying parameters (eg in Monte Carlo simulation) But can also be used to ensure compatibility: Put  i = e i Then get least squares best fit to minimise S =  (  i ² - s i 2 ) 2  S/  v = 0 → v+1 =  [e i (1- e i )] 2 /  e i (1- e i )s i 2 so  i ² = e i (1-e i ) / (v+1) ( NOTE: only have to know mean values and v ) Sum over available estimates s i ² so can tolerate missing values

Experiment: FT full-time MBA ranking 7 variables used in experiment

Experiment: 3 point estimate →  a 2 3 point judgementscaled from which, mean and standard deviation Dirichlet consistent missing value tolerated

Summarising discrimination between programmes y =  i w i x i var(y) =  i  j σ ij x i x j = [  i w i (1-w i )x i ² - 2  i  j w i w j x i x j ] / (v+1) j>i For two alternatives replace x values with differences (x a -x b )

Northern Europe: UK, France, Belgium, Netherlands, Ireland

Summarising discrimination between programmes Summaries (v+1) = Proportion of all pairwise differences significantly different at p = 0.1: discrimination = 81%

Lines show indistinguishable pairs; p = 10%

Lines show indistinguishable pairs; p = 1%

Lines show indistinguishable pairs; p = 25%

Weights. discrimination = 71%

Weights; p = 10%

Different circumstances →  c 2. Reciprocal matrix.

 a 2 and  c 2. Reciprocal matrix. Give each judgement in a reciprocal matrix as a 3-point evaluation. Then treat each column as a separate 3-point evaluation and find Dirichlet compatible  a 2 as before. For each weight the mean of these variances is the value of  a 2 as in aggregating expert judgements (Clemen & Winkler, 2007). The mean of the column means is the weight and the variance of the means is  c 2.

Results from 10 MBA students standard deviations σ = [  a 2 +  c 2 ] ½

Are the two sources of uncertainty related? mean r = 0.70 taken together r = 0.33 consistently σ c > σ a

Student G is representative

Scores. (v+1) = discrimination = 30%

Lines show indistinguishable pairs; p = 10%

Lines show indistinguishable pairs; p = 50% decide that 1 & 5 can be distinguished

A possible form of interaction Assume that new discrimination is due to increased precision rather than difference in scores: statistical significance rather than material significance. So, change precision by changing (v+1) and leave weights unaltered. z is directly proportional to √(v+1). In this case (v+1) = 8.54 → z 1,5 = 0.55 p = 50% → z* = 0.67 (v+1) new = (z*/z 1,5 )² × (v+1) = (0.67 / 0.55)² × 8.54 = and so...

Weights. discrimination = 14%

Group aggregation Do this for all ten assessors.

Scores. (v+1) = discrimination = 16%

Lines show indistinguishable pairs; p = 10%

Lines show indistinguishable pairs; p = 50%

Weights. discrimination = 0%

Tentative conclusions Even though results are imprecise there may still exist enough discrimination to be useful, as in forming a short list. May give ordering of clusters. Makes explicit what may justifiably be discriminated. Choosing confidence levels and significance values is, as ever, sensible but arbitrary. Explore different values. Once a short list is identified, further analysis needed, probably using some form of what-if interaction to see the effect of greater precision. Variation between circumstances seems to be consistently greater than self- assessed uncertainty. Does this matter? Do we want to justify one decision now or address circumstantial (temporal?) variation?

end