Unit 3a: Introducing the Multilevel Regression Model © Andrew Ho, Harvard Graduate School of EducationUnit 3a – Slide 1

Slides:



Advertisements
Similar presentations
Which Test? Which Test? Explorin g Data Explorin g Data Planning a Study Planning a Study Anticipat.
Advertisements

AP Statistics Course Review.
Unit 4a: Basic Logistic (Binomial Logit) Regression Analysis © Andrew Ho, Harvard Graduate School of EducationUnit 4a – Slide 1
Hierarchical Linear Modeling: An Introduction & Applications in Organizational Research Michael C. Rodriguez.
Unit 6a: Motivating Principal Components Analysis © Andrew Ho, Harvard Graduate School of EducationUnit 6a– Slide 1
Forecasting Using the Simple Linear Regression Model and Correlation
ADVANCED STATISTICS FOR MEDICAL STUDIES Mwarumba Mwavita, Ph.D. School of Educational Studies Research Evaluation Measurement and Statistics (REMS) Oklahoma.
Irwin/McGraw-Hill © Andrew F. Siegel, 1997 and l Chapter 12 l Multiple Regression: Predicting One Factor from Several Others.
Conclusion to Bivariate Linear Regression Economics 224 – Notes for November 19, 2008.
6-1 Introduction To Empirical Models 6-1 Introduction To Empirical Models.
© 2008 McGraw-Hill Higher Education The Statistical Imagination Chapter 12: Analysis of Variance: Differences among Means of Three or More Groups.
Copyright (c) 2004 Brooks/Cole, a division of Thomson Learning, Inc. Chapter 13 Nonlinear and Multiple Regression.
© Willett, Harvard University Graduate School of Education, 5/21/2015S052/I.3(b) – Slide 1 More details can be found in the “Course Objectives and Content”
LINEAR REGRESSION: Evaluating Regression Models Overview Assumptions for Linear Regression Evaluating a Regression Model.
LINEAR REGRESSION: Evaluating Regression Models. Overview Assumptions for Linear Regression Evaluating a Regression Model.
Regression and Correlation
Clustered or Multilevel Data
Treatment Effects: What works for Whom? Spyros Konstantopoulos Michigan State University.
Topics: Regression Simple Linear Regression: one dependent variable and one independent variable Multiple Regression: one dependent variable and two or.
Topic 3: Regression.
© 2000 Prentice-Hall, Inc. Chap Forecasting Using the Simple Linear Regression Model and Correlation.
Analysis of Variance & Multivariate Analysis of Variance
Analysis of Individual Variables Descriptive – –Measures of Central Tendency Mean – Average score of distribution (1 st moment) Median – Middle score (50.
Today Concepts underlying inferential statistics
Correlation and Regression Analysis
Unit 5c: Adding Predictors to the Discrete Time Hazard Model © Andrew Ho, Harvard Graduate School of EducationUnit 5c– Slide 1
Structural Equation Modeling Intro to SEM Psy 524 Ainsworth.
Chapter 12 Section 1 Inference for Linear Regression.
S052/Shopping Presentation – Slide #1 © Willett, Harvard University Graduate School of Education S052: Applied Data Analysis Shopping Presentation: A.
Relationships Among Variables
Basic Analysis of Variance and the General Linear Model Psy 420 Andrew Ainsworth.
Unit 5c: Adding Predictors to the Discrete Time Hazard Model © Andrew Ho, Harvard Graduate School of EducationUnit 5c– Slide 1
Unit 4c: Taxonomies of Logistic Regression Models © Andrew Ho, Harvard Graduate School of EducationUnit 4c – Slide 1
Unit 3b: From Fixed to Random Intercepts © Andrew Ho, Harvard Graduate School of EducationUnit 3b – Slide 1
Analysis of Clustered and Longitudinal Data
Multiple Regression Farrokh Alemi, Ph.D. Kashif Haqqi M.D.
Review Guess the correlation. A.-2.0 B.-0.9 C.-0.1 D.0.1 E.0.9.
Unit 2b: Dealing “Rationally” with Nonlinear Relationships © Andrew Ho, Harvard Graduate School of EducationUnit 2b – Slide 1
Unit 4c: Taxonomies of Logistic Regression Models © Andrew Ho, Harvard Graduate School of EducationUnit 4c – Slide 1
Unit 4b: Fitting the Logistic Model to Data © Andrew Ho, Harvard Graduate School of EducationUnit 4b – Slide 1
© Willett, Harvard University Graduate School of Education, 8/27/2015S052/I.3(c) – Slide 1 More details can be found in the “Course Objectives and Content”
PS 225 Lecture 15 Analysis of Variance ANOVA Tables.
© 2005 The McGraw-Hill Companies, Inc., All Rights Reserved. Chapter 12 Describing Data.
Unit 6b: Principal Components Analysis © Andrew Ho, Harvard Graduate School of EducationUnit 6b – Slide 1
From GLM to HLM Working with Continuous Outcomes EPSY 5245 Michael C. Rodriguez.
Andrew Ho Harvard Graduate School of Education Tuesday, January 22, 2013 S-052 Shopping – Applied Data Analysis.
Unit 5b: The Logistic Regression Approach to Life Table Analysis © Andrew Ho, Harvard Graduate School of EducationUnit 5b– Slide 1
© 1998, Geoff Kuenning General 2 k Factorial Designs Used to explain the effects of k factors, each with two alternatives or levels 2 2 factorial designs.
Hierarchical Linear Modeling (HLM): A Conceptual Introduction Jessaca Spybrook Educational Leadership, Research, and Technology.
Error Component Models Methods of Economic Investigation Lecture 8 1.
User Study Evaluation Human-Computer Interaction.
Unit 1c: Detecting Influential Data Points and Assessing Their Impact © Andrew Ho, Harvard Graduate School of EducationUnit 1c – Slide 1
S052/Shopping Presentation – Slide #1 © Willett, Harvard University Graduate School of Education S052: Applied Data Analysis What Would You Like To Know.
Maths Study Centre CB Open 11am – 5pm Semester Weekdays
© Willett, Harvard University Graduate School of Education, 12/16/2015S052/I.1(d) – Slide 1 More details can be found in the “Course Objectives and Content”
Psychology 202a Advanced Psychological Statistics November 12, 2015.
I271B QUANTITATIVE METHODS Regression and Diagnostics.
Robust Regression. Regression Methods  We are going to look at three approaches to robust regression:  Regression with robust standard errors  Regression.
Why do we analyze data?  It is important to analyze data because you need to determine the extent to which the hypothesized relationship does or does.
Unit 2a: Dealing “Empirically” with Nonlinear Relationships © Andrew Ho, Harvard Graduate School of EducationUnit 2a – Slide 1
Anticipating Patterns Statistical Inference
Applied Biostatistics: Lecture 2
What we’ll cover today Transformations Inferential statistics
POSC 202A: Lecture Lecture: Substantive Significance, Relationship between Variables 1.
Understanding Research Results: Description and Correlation
From GLM to HLM Working with Continuous Outcomes
When You See (This), You Think (That)
Introductory Statistics
Presentation transcript:

Unit 3a: Introducing the Multilevel Regression Model © Andrew Ho, Harvard Graduate School of EducationUnit 3a – Slide 1

Revisiting the assumption of independent population residuals Identifying and visualizing a multilevel structure Contrasting “Total,” “Between,” and “Within” Regression models © Andrew Ho, Harvard Graduate School of Education Unit 3a– Slide 2 Multiple Regression Analysis (MRA) Multiple Regression Analysis (MRA) Do your residuals meet the required assumptions? Test for residual normality Use influence statistics to detect atypical datapoints If your residuals are not independent, replace OLS by GLS regression analysis Use Individual growth modeling Specify a Multi-level Model If time is a predictor, you need discrete- time survival analysis… If your outcome is categorical, you need to use… Binomial logistic regression analysis (dichotomous outcome) Multinomial logistic regression analysis (polytomous outcome) If you have more predictors than you can deal with, Create taxonomies of fitted models and compare them. Form composites of the indicators of any common construct. Conduct a Principal Components Analysis Use Cluster Analysis Use non-linear regression analysis. Transform the outcome or predictor If your outcome vs. predictor relationship is non-linear, Use Factor Analysis: EFA or CFA? Course Roadmap: Unit 3a Today’s Topic Area

© Andrew Ho, Harvard Graduate School of Education Unit 3a – Slide 3 If the population residuals are correlated across observations, then OLS-estimated standard errors will be too small. So, t-statistics will be inflated, and null hypotheses will be rejected more frequently than is proper (Increased Type I Error). … the errors must be independent from observation to observation.” “In the population, … Residual Independence Assumption How Does Failure of the Assumption Affect OLS Regression Analysis? Once you have addressed linearity and measurement error conditions, then you should consider the following assumptions about population residuals (a.k.a. errors)… in this rough order of diminishing priority: In a regression model, all unobserved effects end up in the residuals, and so the residuals of students in the same school may lose their required independence. Then, OLS regression analysis will provide incorrect standard errors and inference. In a regression model, all unobserved effects end up in the residuals, and so the residuals of students in the same school may lose their required independence. Then, OLS regression analysis will provide incorrect standard errors and inference. Students within the same school share many common unobserved experiences that may impact their values of the outcome, in a similar way. Students within the same school share many common unobserved experiences that may impact their values of the outcome, in a similar way.

© Andrew Ho, Harvard Graduate School of EducationUnit 3a – Slide 4 RQ: What is the relationship between SES and math achievement? We can download datasets from online sources directly!  This is an oft-analyzed dataset from the High School and Beyond survey, that tracks young adults through schooling and their careers.High School and Beyond  These data are a subsample of 7185 students from 160 schools assessed on their mathematics achievement at a single timepoint in  We also have their socioeconomic status estimated from a composite scale that incorporates parental education, parental occupation, and parental income.  This is an oft-analyzed dataset from the High School and Beyond survey, that tracks young adults through schooling and their careers.High School and Beyond  These data are a subsample of 7185 students from 160 schools assessed on their mathematics achievement at a single timepoint in  We also have their socioeconomic status estimated from a composite scale that incorporates parental education, parental occupation, and parental income. I’m including some generally useful formatting that allows you to reorder variables and capitalize all variable names.

© Andrew Ho, Harvard Graduate School of EducationUnit 3a – Slide 5 Understanding a Multilevel Structure A random sort allows us to sample 10 random observations from our dataset. A “School ID” code that tells us that all of these students are from the same school. We see that we’ve sampled 10 students from different schools. They differ on math achievement and SES, but we know regression inferences will be flawed if we don’t take school membership into account. We see that we’ve sampled 10 students from different schools. They differ on math achievement and SES, but we know regression inferences will be flawed if we don’t take school membership into account.

© Andrew Ho, Harvard Graduate School of EducationUnit 3a – Slide 6 Identifying Your Grouping Variable with xtset This allows you to identify a single observation from each group. Seems silly, for now. This command allows you to identify your “grouping variable” for subsequent commands. A grouping variable can be a classroom, school, district, state, or hospital, as long as there are multiple observations within this group and multiple groups. A grouping variable is also often a participant or patient on whom multiple measures are gathered. This command allows you to identify your “grouping variable” for subsequent commands. A grouping variable can be a classroom, school, district, state, or hospital, as long as there are multiple observations within this group and multiple groups. A grouping variable is also often a participant or patient on whom multiple measures are gathered. This is one way to count the number of schools that you have. How many 1s? So we have 160 schools. Still silly, but you’ll see…

© Andrew Ho, Harvard Graduate School of EducationUnit 3a – Slide 7 The Super-Helpful egen Command with its by Option The egen command allows you to create new variables by your grouping variable. Want counts, means, standard deviations, or alternative identifiers by school? Look to egen. Instead of wild integer IDs with gaps between them, this allows you to number your schools 1…N. This should always be a knee-jerk reaction with multilevel data. What is the distribution of group sizes, that is, what is the distribution of the number of observations per group? When there is variance here, we call this “unbalanced.” We often (but do not always) have “balanced” designs in measurements of individuals, where a participant ID is the grouping variable, and we measure all participants the same number of times. This should always be a knee-jerk reaction with multilevel data. What is the distribution of group sizes, that is, what is the distribution of the number of observations per group? When there is variance here, we call this “unbalanced.” We often (but do not always) have “balanced” designs in measurements of individuals, where a participant ID is the grouping variable, and we measure all participants the same number of times.

Exploratory Data Analysis for Multilevel Structures This is a naïve perspective that does not take the grouping variable into account. It’s not a bad place to start as long as you understand that grouping lurks underneath. This is a naïve perspective that does not take the grouping variable into account. It’s not a bad place to start as long as you understand that grouping lurks underneath. © Andrew Ho, Harvard Graduate School of EducationUnit 3a – Slide 8 Tabulate and graph the means and variances for each group, or a random sample of groups. Try to distinguish “within-group” variation, the standard deviations and spreads of each school distribution, from “between-group” variation, the variation in school means from different schools. Here, we’re looking for differences in central tendency across groups, or “between group variation,” vs. the average spread within each group, or “within- group variation. We can also visualize hetero- scedasticity.

Between-Group Variation on the Outcome Variable We can easily calculate group means using the egen command. © Andrew Ho, Harvard Graduate School of EducationUnit 3a – Slide 9 We show the distribution of school average mathematics achievement. Compare the variance of this distribution to the unconditional, total distribution of the MATHACH variable (the relationship should not be surprising).

Within-Group Centering or “De-Meaning” to Visualize Within-Group Variation © Andrew Ho, Harvard Graduate School of EducationUnit 3a – Slide 10 We center all of the within-school distributions on the grand mean. A bit off because box plots show medians, not means

Decomposing and distinguishing variance with the xtsum command. © Andrew Ho, Harvard Graduate School of EducationUnit 3a – Slide 11 Unconditional standard deviation Standard deviation of school means: “Between” Standard deviation of de- meaned scores: “Within” Number of observations and groups, respectively. Average group size.

© Andrew Ho, Harvard Graduate School of EducationUnit 3a – Slide 12 Not-so-good, old-fashioned regression, completely ignoring school membership and possible correlations among residuals

© Andrew Ho, Harvard Graduate School of EducationUnit 3a – Slide 13

© Andrew Ho, Harvard Graduate School of EducationUnit 3a – Slide 14

Contrasting “Between,” “Total,” and “Within” Regression Lines © Andrew Ho, Harvard Graduate School of EducationUnit 3a – Slide 15 Between: Means on Means, ignores within-group variation. Total: Points on Points, ignores group membership. Within: Demeaned Points on Demeaned Points. By demeaning, we have taken group membership into account! Between: Means on Means, ignores within-group variation. Total: Points on Points, ignores group membership. Within: Demeaned Points on Demeaned Points. By demeaning, we have taken group membership into account!