The Changepoint Approach to SPC Douglas M. Hawkins, Peihua Qiu University of Minnesota Chang-Wook Kang Hanyang University.

Slides:



Advertisements
Similar presentations
Chapter 7 Hypothesis Testing
Advertisements

Chapter 4 Inference About Process Quality
Chapter 16 Inferential Statistics
Hypothesis Testing. To define a statistical Test we 1.Choose a statistic (called the test statistic) 2.Divide the range of possible values for the test.
CHAPTER 21 Inferential Statistical Analysis. Understanding probability The idea of probability is central to inferential statistics. It means the chance.
Copyright (c) 2009 John Wiley & Sons, Inc.
EPIDEMIOLOGY AND BIOSTATISTICS DEPT Esimating Population Value with Hypothesis Testing.
G. Cowan Lectures on Statistical Data Analysis 1 Statistical Data Analysis: Lecture 8 1Probability, Bayes’ theorem, random variables, pdfs 2Functions of.
Tests Jean-Yves Le Boudec. Contents 1.The Neyman Pearson framework 2.Likelihood Ratio Tests 3.ANOVA 4.Asymptotic Results 5.Other Tests 1.
Horng-Chyi HorngStatistics II41 Inference on the Mean of a Population - Variance Known H 0 :  =  0 H 0 :  =  0 H 1 :    0, where  0 is a specified.
Inferences About Process Quality
BCOR 1020 Business Statistics
5-3 Inference on the Means of Two Populations, Variances Unknown
Analysis of Variance. ANOVA Probably the most popular analysis in psychology Why? Ease of implementation Allows for analysis of several groups at once.
Hypothesis Testing:.
Overview Definition Hypothesis
Review of Statistical Inference Prepared by Vera Tabakova, East Carolina University ECON 4550 Econometrics Memorial University of Newfoundland.
+ DO NOW What conditions do you need to check before constructing a confidence interval for the population proportion? (hint: there are three)
Estimating a Population Mean
1 Least squares procedure Inference for least squares lines Simple Linear Regression.
June 18, 2008Stat Lecture 11 - Confidence Intervals 1 Introduction to Inference Sampling Distributions, Confidence Intervals and Hypothesis Testing.
Section 8.3 Estimating a Population Mean. Section 8.3 Estimating a Population Mean After this section, you should be able to… CONSTRUCT and INTERPRET.
CHAPTER 18: Inference about a Population Mean
Introduction to Statistical Quality Control, 4th Edition
Regression Part II One-factor ANOVA Another dummy variable coding scheme Contrasts Multiple comparisons Interactions.
Business Statistics for Managerial Decision Comparing two Population Means.
Introduction to inference Use and abuse of tests; power and decision IPS chapters 6.3 and 6.4 © 2006 W.H. Freeman and Company.
Exam Exam starts two weeks from today. Amusing Statistics Use what you know about normal distributions to evaluate this finding: The study, published.
Maximum Likelihood Estimator of Proportion Let {s 1,s 2,…,s n } be a set of independent outcomes from a Bernoulli experiment with unknown probability.
10.2 Tests of Significance Use confidence intervals when the goal is to estimate the population parameter If the goal is to.
Inference We want to know how often students in a medium-size college go to the mall in a given year. We interview an SRS of n = 10. If we interviewed.
+ The Practice of Statistics, 4 th edition – For AP* STARNES, YATES, MOORE Chapter 8: Estimating with Confidence Section 8.3 Estimating a Population Mean.
1 Lecture 16: Point Estimation Concepts and Methods Devore, Ch
What Does the Likelihood Principle Say About Statistical Process Control? Gemai Chen, University of Calgary Canada July 10, 2006.
KNR 445 Statistics t-tests Slide 1 Introduction to Hypothesis Testing The z-test.
1 The Monitoring of Linear Profiles Keun Pyo Kim Mahmoud A. Mahmoud William H. Woodall Virginia Tech Blacksburg, VA (Send request for paper,
AP Statistics Section 11.1 B More on Significance Tests.
Sampling and estimation Petter Mostad
+ The Practice of Statistics, 4 th edition – For AP* STARNES, YATES, MOORE Unit 5: Estimating with Confidence Section 11.1 Estimating a Population Mean.
Statistical Inference Statistical inference is concerned with the use of sample data to make inferences about unknown population parameters. For example,
Chapter 13 Sampling distributions
Dr. Dipayan Das Assistant Professor Dept. of Textile Technology Indian Institute of Technology Delhi Phone:
1 SMU EMIS 7364 NTU TO-570-N Control Charts Basic Concepts and Mathematical Basis Updated: 3/2/04 Statistical Quality Control Dr. Jerrell T. Stracener,
Review of Statistical Inference Prepared by Vera Tabakova, East Carolina University.
Week 21 Order Statistics The order statistics of a set of random variables X 1, X 2,…, X n are the same random variables arranged in increasing order.
Hypothesis Testing. Suppose we believe the average systolic blood pressure of healthy adults is normally distributed with mean μ = 120 and variance σ.
Uncertainty and confidence Although the sample mean,, is a unique number for any particular sample, if you pick a different sample you will probably get.
+ Chapter 8 Estimating with Confidence 8.1Confidence Intervals: The Basics 8.2Estimating a Population Proportion 8.3Estimating a Population Mean.
The inference and accuracy We learned how to estimate the probability that the percentage of some subjects in the sample would be in a given interval by.
Statistical Inference for the Mean Objectives: (Chapter 8&9, DeCoursey) -To understand the terms variance and standard error of a sample mean, Null Hypothesis,
Week 21 Statistical Model A statistical model for some data is a set of distributions, one of which corresponds to the true unknown distribution that produced.
Chapter 8: Estimating with Confidence
SPC Born in the ’20’s Walter A. Shewhart
Chapter 8: Estimating with Confidence
Hypothesis Testing: Hypotheses
Discrete Event Simulation - 4
ENGM 620: Quality Management
Chapter 8: Estimating with Confidence
Special Control Charts II
Chapter 8: Estimating with Confidence
Chapter 8: Estimating with Confidence
Chapter 8: Estimating with Confidence
Inference on the Mean of a Population -Variance Known
Chapter 8: Estimating with Confidence
Chapter 8: Estimating with Confidence
2/5/ Estimating a Population Mean.
Chapter 8: Estimating with Confidence
Chapter 8: Estimating with Confidence
Chapter 8: Estimating with Confidence
Presentation transcript:

The Changepoint Approach to SPC Douglas M. Hawkins, Peihua Qiu University of Minnesota Chang-Wook Kang Hanyang University

Changepoint approach to SPC2 Background to SPC Have stream of process readings X 1, X 2,…X n,…. Need to decide whether all follow common statistical model, versus Isolated (transient) special causes (affect individual readings) or Persistent special causes that remain until detected and fixed.

Changepoint approach to SPC3 The simplest statistical model In control the X n are iid N( 2 ) Isolated special causes change mean and/or variance then revert. Persistent special cause shifts the mean and/or variance. For example, step change in mean to

Changepoint approach to SPC4 Standard SPC methods Shewhart Xbar and R/S chart used for isolated special causes. Persistent causes need memory – cumulative sum (cusum) or exponentially weighted moving average (EWMA) chart. For now we concentrate on latter.

Changepoint approach to SPC5 Designing a chart An upward cusum is defined by where K is reference value or allowance. The chart signals a change if where H is the decision interval.

Changepoint approach to SPC6 The things you need to know Cusum is the optimal way to detect step shift if K is halfway between in-control and out-of-control means. So you must know and You decide H by setting acceptable in- control average run length (ARL). To do this, you also need to know

Changepoint approach to SPC7 Who told you the Greek stuff? Very rarely, you do actually know it. More commonly, –do a Phase I study to estimate and –carefully check data for control (can use fixed- sample-size methods for this) –pick a big enough to matter, small enough not to be easy to see.

Changepoint approach to SPC8 An estimate is not a parameter But sample estimates are not population parameters. So you have a target ARL, but your actual ARL will be a random variable. For sensitive methods like cusum with small K, EWMA with small, resulting uncertainty in your ARL can be large.

Changepoint approach to SPC9 What cusum optimality? On top of this, cusum is optimal only for shift it is tuned for. Get a much different shift, you lose performance. Similarly for EWMA.

Changepoint approach to SPC10 The changepoint-in-mean model For this model –X i ~ N( 2 ) for i <= ~ N( for i > None of the Greeks is known a priori. Suppose we are at observation number n.

Changepoint approach to SPC11 Likelihood approach Write If we knew changepoint was (say) k then MLEs for would be 2 MLE would be S k,n = (V 0,k + V k,n )/(n-2) (after the usual bias adjustment

Changepoint approach to SPC12 …. continued Two-sample t for H 0 : =0 (no change) is Finally, estimate as k maximizing |T k,n | And diagnose step change if T max,n > h n

Changepoint approach to SPC13 Phase II use Changepoint formulation for fixed-sample (Phase I setting) is classical. For Phase II SPC use n is not constant. Modify the procedure to: –If T max,n, < h n, diagnose in control, continue –If T max,n, > h n, conclude out of control. Use the MLEs to diagnose time of change and pre- and post-change means.

Changepoint approach to SPC14 Getting the control limits We need sequence of control limits h n. Fixed-sample theory not much help. A conceptual objective: Pick the h n so that Pr[T max,n > h n | no signal before time n] =. With such a sequence, in-control RL would be geometric (like Shewhart), and with –In-control ARL = 1/

Changepoint approach to SPC15 How to get the h n Big simulation: 16 million data sets. Estimated h n for several values. All on web site

Changepoint approach to SPC16 So why have a Phase I? Dont need in-control parameter estimates, and so dont need Phase I data gathering, Can get up and running in Phase II. As time goes by in control, ever-growing data base gives ever-better estimates (unlike conventional Phase I/II dichotomy)

Changepoint approach to SPC17 ….continued But most folk would dry run at least some readings before turning on testing. For lack of obvious best choice, suggest starting testing at n=10 (but Web tables give cutoffs for starts of n=3 through 21) For example, for =0.005:

Changepoint approach to SPC18

Changepoint approach to SPC19 The cutoff The cutoffs seem to tend to around 3.2 This corresponds roughly to the two-sided point of a N(0,1) This Bonferroni multiplier of 5 is what you pay for the multiple testing.

Changepoint approach to SPC20 Do we need the Shewhart? Changepoint formulation with compares latest X with mean of all previous data; this includes Shewhart I chart as one of its tests. Asymptotic cutoff of 3.2 is close to European standard. and tests the newest mean against grand mean of all previous data; this includes Shewhart Xbar chart for rational groups of any and all sizes.

Changepoint approach to SPC21 How does method perform? Compared to what? Methods that fix IC ARL with unknown parameters scarce. Self-starting cusum doesnt need IC parameter values. Also seamless from Phase I to Phase II. Does however need size of shift for tuning purposes.

Changepoint approach to SPC22 A method comparison Three cusums, k=0.25, 0.5, 1 (tuned for shifts of 0.5, 1, 2 sds) Two in-control ARLs – 100, 500 Shift occurring early (observation 10) or later (observation 100) a: ARL 100, early; b: ARL 100, later c: ARL 500, early; d: ARL 500, later

Changepoint approach to SPC23

Changepoint approach to SPC24 Results Changepoint is sometimes best. Mostly is second best (no surprise, given cusums theoretical optimality). Where not best, it is a close second best and has by far most robustly good performance.

Changepoint approach to SPC25 Example – triglyceride data Data set kindly supplied by Dr. Dan Schultz, Rogasin Institute, New York. Assay triglyceride standard every week. Use as a QC check on unknowns. Triglyceride reading should be constant (doesnt much matter what its value is). Heres one year of data (given as I chart):

Changepoint approach to SPC26 Outlier? Upward shift at end?

Changepoint approach to SPC27 First clear exceedance is at week 40

Changepoint approach to SPC28 What are estimates of the changepoint?

Changepoint approach to SPC29 and of the before- and after-change means

Changepoint approach to SPC30 Focus Dont interpret estimate of changepoint or of separate means in non-significant bit. First signal is 5 weeks after apparent shift. Pre-change mean estimate is 117 mg/dL Post-change mean estimate is124 mg/dL Right from first signal, all three estimates highly stable.

Changepoint approach to SPC31 Conclusions Conventional Shewhart, cusum, EWMA calibrated assuming known parameters. Random errors of estimation in parameters become systematic distortions in run distribution of any particular chart making IC and OOC ARLs random. Ugly tradeoff between Phase I sample size and control over IC RL distribution.

Changepoint approach to SPC32 … The unknown-parameter changepoint formulation lets you fix in-control run length distribution exactly, with or without sizeable Phase I sample. Furthermore, interval alternative means performance competitive regardless of size of the shift.

Changepoint approach to SPC33 References Hawkins, D. M., Qiu, P., and Kang, C.-W. (2003) The Changepoint Model for Statistical Process Control to appear in Journal of Quality Technology. Pollak, M. and Siegmund, D., (1991), 'Sequential Detection of a Change in a Normal Mean When the Initial Value Is Unknown', Annals of Statistics, 19, Siegmund, D, (1985), Sequential analysis : tests and confidence intervals, Springer-Verlag, New York. Siegmund, D. and Venkatraman, E. S., (1995), 'Using the Generalized Likelihood Ratio Statistic for Sequential Detection of a Change- point', Annals of Statistics, 23,