How Many Discoveries Have Been Lost by Ignoring Modern Statistical Methods? Rand R. Wilcox.

Slides:



Advertisements
Similar presentations
Nonparametric Statistics Timothy C. Bates
Advertisements

8. Heteroskedasticity We have already seen that homoskedasticity exists when the error term’s variance, conditional on all x variables, is constant: Homoskedasticity.
Confidence Interval and Hypothesis Testing for:
Copyright © 2007 Pearson Education, Inc. Publishing as Pearson Addison-Wesley Chapter 19 Confidence Intervals for Proportions.
Confidence Intervals for Proportions
MARE 250 Dr. Jason Turner Hypothesis Testing II To ASSUME is to make an… Four assumptions for t-test hypothesis testing: 1. Random Samples 2. Independent.
MARE 250 Dr. Jason Turner Hypothesis Testing II. To ASSUME is to make an… Four assumptions for t-test hypothesis testing:
PSY 1950 Confidence and Power December, Requisite Quote “The picturing of data allows us to be sensitive not only to the multiple hypotheses that.
Hypothesis Tests for Means The context “Statistical significance” Hypothesis tests and confidence intervals The steps Hypothesis Test statistic Distribution.
Copyright © 2010 Pearson Education, Inc. Chapter 19 Confidence Intervals for Proportions.
Chapter 11: Inference for Distributions
Educational Research by John W. Creswell. Copyright © 2002 by Pearson Education. All rights reserved. Slide 1 Chapter 8 Analyzing and Interpreting Quantitative.
Getting Started with Hypothesis Testing The Single Sample.
Review for Exam 2 Some important themes from Chapters 6-9 Chap. 6. Significance Tests Chap. 7: Comparing Two Groups Chap. 8: Contingency Tables (Categorical.
Overview of Robust Methods Analysis Jinxia Ma November 7, 2013.
Linear Regression 2 Sociology 5811 Lecture 21 Copyright © 2005 by Evan Schofer Do not copy or distribute without permission.
Chapter 19: Confidence Intervals for Proportions
Bootstrapping applied to t-tests
The t-test Inferences about Population Means when population SD is unknown.
Chapter 4. Exercise 1 The 95% CI represents a range of values that contains a population parameter with a 0.95 probability. This range is determined by.
Review I volunteer in my son’s 2nd grade class on library day. Each kid gets to check out one book. Here are the types of books they picked this week:
Marshall University School of Medicine Department of Biochemistry and Microbiology BMS 617 Lecture 6 – Multiple comparisons, non-normality, outliers Marshall.
Statistical Significance R.Raveendran. Heart rate (bpm) Mean ± SEM n In men ± In women ± The difference between means.
CHAPTER 16: Inference in Practice. Chapter 16 Concepts 2  Conditions for Inference in Practice  Cautions About Confidence Intervals  Cautions About.
Jan 17,  Hypothesis, Null hypothesis Research question Null is the hypothesis of “no relationship”  Normal Distribution Bell curve Standard normal.
Basic Business Statistics, 11e © 2009 Prentice-Hall, Inc. Chap 8-1 Confidence Interval Estimation.
Bootstrapping (And other statistical trickery). Reminder Of What We Do In Statistics Null Hypothesis Statistical Test Logic – Assume that the “no effect”
The Robust Approach Dealing with real data. Review With regular analyses we have certain assumptions that are made, or requirements that have to be met.
Copyright © 2012 Pearson Education. Chapter 23 Nonparametric Methods.
B AD 6243: Applied Univariate Statistics Correlation Professor Laku Chidambaram Price College of Business University of Oklahoma.
1 rules of engagement no computer or no power → no lesson no SPSS → no lesson no homework done → no lesson GE 5 Tutorial 5.
10.2 Tests of Significance Use confidence intervals when the goal is to estimate the population parameter If the goal is to.
CHAPTER 17: Tests of Significance: The Basics
Basics of Data Cleaning
The Robust Approach Dealing with real data. Estimating Population Parameters Four properties are considered desirable in a population estimator:  Sufficiency.
Copyright © 2010 Pearson Education, Inc. Chapter 19 Confidence Intervals for Proportions.
Sampling distributions rule of thumb…. Some important points about sample distributions… If we obtain a sample that meets the rules of thumb, then…
Correlation Review and Extension. Questions to be asked… Is there a linear relationship between x and y? What is the strength of this relationship? Pearson.
Robust Estimators.
Correlation. Correlation Analysis Correlations tell us to the degree that two variables are similar or associated with each other. It is a measure of.
Stats Lunch: Day 3 The Basis of Hypothesis Testing w/ Parametric Statistics.
Copyright © 2009 Pearson Education, Inc. Chapter 19 Confidence Intervals for Proportions.
Chapter 19 Confidence intervals for proportions
Tuesday, April 8 n Inferential statistics – Part 2 n Hypothesis testing n Statistical significance n continued….
Copyright © 2010, 2007, 2004 Pearson Education, Inc. Chapter 19 Confidence Intervals for Proportions.
Thinking Robustly. Sampling Distribution In order to examine the properties of a statistic we often want to take repeated samples from some population.
HL Psychology Internal Assessment
Copyright © 2010, 2007, 2004 Pearson Education, Inc. Chapter 19 Confidence Intervals for Proportions.
Marshall University School of Medicine Department of Biochemistry and Microbiology BMS 617 Lecture 7 – Non-normality and outliers.
Power Point Slides by Ronald J. Shope in collaboration with John W. Creswell Chapter 7 Analyzing and Interpreting Quantitative Data.
1 Copyright © 2010, 2007, 2004 Pearson Education, Inc. All Rights Reserved. Example: In a recent poll, 70% of 1501 randomly selected adults said they believed.
Copyright © 2013, 2009, and 2007, Pearson Education, Inc. Chapter 12 Analyzing the Association Between Quantitative Variables: Regression Analysis Section.
Non-parametric Approaches The Bootstrap. Non-parametric? Non-parametric or distribution-free tests have more lax and/or different assumptions Properties:
Copyright © 2010 Pearson Education, Inc. Slide
Statistics 19 Confidence Intervals for Proportions.
Confidence Intervals for Proportions
Data analysis Research methods.
Non-Parametric Tests 12/1.
LBSRE1021 Data Interpretation Lecture 9
Non-Parametric Tests 12/1.
Confidence Intervals for Proportions
Confidence Intervals for Proportions
Non-Parametric Tests 12/6.
Fundamentals of regression analysis
Non-Parametric Tests.
Quantitative Methods PSY302 Quiz Chapter 9 Statistical Significance
Inferential statistics,
Chapter 25: Paired Samples and Blocks
Review for Exam 2 Some important themes from Chapters 6-9
Section 10.1: Confidence Intervals
Presentation transcript:

How Many Discoveries Have Been Lost by Ignoring Modern Statistical Methods? Rand R. Wilcox

The theme Despite what we learn, standard methods are NOT robust to violations of normality Despite what we learn, standard methods are NOT robust to violations of normality –Heteroscedasticity –Skewness –Outliers => Reduce chances of detecting true differences & obtaining accurate confidence intervals

Alternatives to the Mean: Need an estimator that performs as well as the mean under normal conditions AND is robust to departures from normality Need an estimator that performs as well as the mean under normal conditions AND is robust to departures from normality –4 options: 10%t trimmed mean 10%t trimmed mean 20% trimmed mean 20% trimmed mean Μ m – Mean estimator by some chap called Huber. Μ m – Mean estimator by some chap called Huber. Ө.5 – Median estimator by some chaps called Harrell & David Ө.5 – Median estimator by some chaps called Harrell & David

Dealing with Outliers: Sample mean & sample SD are inflated by outliers => masks them Sample mean & sample SD are inflated by outliers => masks them Trimming is not simply “throwing” data away and applying standard methods Trimming is not simply “throwing” data away and applying standard methods –This is a bad idea! If you take out extreme values and then continue => use of the wrong SE.

Skewness:

How much trimming & what to choose? Rule of thumb = 20% Rule of thumb = 20% Trimmed means tend to perform better that M estimators in more situations; M estimators are better with correlation & regression Trimmed means tend to perform better that M estimators in more situations; M estimators are better with correlation & regression

Why can’t we just test normality and then decide? Because conventional tests are insensitive..... Because conventional tests are insensitive..... Only way to determine if modern methods are useful is to use them Only way to determine if modern methods are useful is to use them Modern methods can be extended to more complex designs as well; including multivariate analyses Modern methods can be extended to more complex designs as well; including multivariate analyses

Correlation: Pearson’s r is not resistant to outliers; modern methods/alternatives can help e.g. Kendall’s Tau & Spearman’s rho Pearson’s r is not resistant to outliers; modern methods/alternatives can help e.g. Kendall’s Tau & Spearman’s rho Percentage Bend correlation: Percentage Bend correlation: –Population value of assoc is zero under independence (unusual apparently) –Good control over type I error in broad range of situations –Allows flexible choice re: how many outliers can be handled

Regression: OLS = poor choice for researchers; SE can be more than 100 times larger than some modern methods! OLS = poor choice for researchers; SE can be more than 100 times larger than some modern methods! Recommends a bootstrap method in conjunction with a robust estimator e.g. S- PLUS function regci Recommends a bootstrap method in conjunction with a robust estimator e.g. S- PLUS function regci Critics argue that robust regressoin estimators fail to check for curvature of the line – this can be fixed by using a “smoother” Critics argue that robust regressoin estimators fail to check for curvature of the line – this can be fixed by using a “smoother”

An example.....

An example (2):

Conclusions: Use of trimmed means and funky modern methods is recommended Use of trimmed means and funky modern methods is recommended Education in psychology should reflect modern advances in stats Education in psychology should reflect modern advances in stats Not all problems are solved, but you could be missing something really important due to the vulnerability of standard methods to minor departures from normality. Not all problems are solved, but you could be missing something really important due to the vulnerability of standard methods to minor departures from normality.