Psychology 290

Special Topics Study Course

Advanced Meta-analysis

Spring Semester, 2014

Syllabus for Spring, 2014

Instructor:

Jack L. Vevea (jvevea@ucmerced.edu)
Social Science and Management Bulding, Room 306-A
Office hours: Wednesdays, 9:00-11:00 A.M. Telephone: (209) 658-1706

Text:

There is no required text for the class; articles to be read before some class meetings will be posted on the web site. Look for links in the syllabus.

Meeting times:

We will meet Mondays and Wednesdays from 7:30 to 8:45 P.M. in room 279 of the Classroom Building.

Course description:

Psychology 290 will introduce the student to some of the mathematical tools often used by quantitative psychologists. The context in which these tools will be developed is advanced meta-analysis.

Course learning goals:

In the class, you will:

learn about strategies for proof using simplifying assumptions;

learn about the estimation principle of maximum likelihood;

gain exposure to simulation as a method for investigating the properties of statistical models;

apply maximum likelihood methods to conventional meta-analytic models;

use maximum likelihood to extend these to more complex models (e.g., analyses in which the magnitude of the variance component depends on study characteristics);

learn about weighted distribution theory and its application to weight-function selection models;

explore the development of models for meta-analysis of growth curves;

learn about techniques for dealing with missing data;

learn the rudiments of Bayesian analysis, using some of the same meta-analytic models that are developed earlier in the course.

Course learning outcomes:

By the end of the class, you will be able to:

implement maximum-likelihood models using software for numerical optimization;

perform and interpret simple computer simulations;

implement some advanced meta-analytic methods, including:

models for variance components;

weight-function selection models;

perform and interpret Bayesian meta-analysis.

Each of these learning goals will be assessed with a homework assignment in which the student implements and interprets the method with an instructor-provided meta-analytic data set.

In addition, students will apply at least two of these methods to their own data.

Prerequisites:

Graduate status in a field that typically employs meta-analytic methods; prior exposure to the basic ideas of meta-analysis, or consent of the instructor.

Evaluation:

Grading will be based on a combination of written homework, a final project and presentation, and class participation. Homework will count for 55% of your final grade, the project and presentation will count for 35%, and participation will count for 10%.

These components make up the final grade in the following manner. First, each component (homework, project, participation) gets a grade point value: A+ = 4.3, A = 4.0, A- = 3.7, B+ = 3.3, B = 3.0, B- = 2.7, and so on. The weighted average of the grade points from the three components determines your final grade. The following table shows the mapping of grade point averages to letter grades:

Grade Point Range	Letter Grade
GPA > 4.25	A+
3.75 < GPA < 4.25	A
3.50 < GPA < 3.75	A-
3.25 < GPA < 3.50	B+
2.75 < GPA < 3.25	B
2.50 < GPA < 2.75	B-
2.25 < GPA < 2.50	C+
1.75 < GPA < 2.25	C
1.50 < GPA < 1.75	C-
0.75 < GPA < 1.50	D
GPA < 0.75	F

Academic Integrity

Students should be familiar with University policies on academic honesty. Here is a link to the most recent policy on acedemic honesty.

For this class, my expectations are:

Cooperative work on the computational aspects of homework assignments is strongly encouraged, but you are expected to work independently on discussion and interpretation. The words you submit in your written assignments should be entirely your own.

You are expected to work entirely independently on your project.

UC Merced Guiding Principles

Where does this fit into my general graduate eduation at UC?

Here are the UC Merced Guiding Principles for General Education. Although they are thought of as particularly relevant to undergraduate training, they still provide a reasonable framework for thinking about graduate education. This class is particularly relevant to the first (Scientific Literacy), second (Decision Making), and eighth (Developing Personal Potential).

Course Outline

Week One

January 22: class introduction.

Week Two

January 27: Vote counting; proof with simplifying assumptions.

Readings are Hedges, Laine, & Greenwald (1994) and Hanushek's reply. Hedges, Laine, & Greewald's response to Hanushek's reply may also be of interest.

January 29: Discussion of readings.

Week Three

February 3: Introduction to maximum likelihood.

Reading: read sections 1 and 2 of the web document found at http://statgen.iop.kcl.ac.uk/bgim/mle/sslike_1.html.

February 5: Maximum likelihood, continued. Properties of maximum likelihood estimates. MLE for fixed-effects meta-analysis.

Reading: read sections 3 and 4 of the web document found at http://statgen.iop.kcl.ac.uk/bgim/mle/sslike_1.html.

Homework One is available (due 2/12/2014).

Week Four

February 10: Random-effects meta-analytic models using maximum likelihood.

Read this article by Wolfgang Viechtbauer for a fairly sound introdution to the logic of random- and mixed-effects models.

February 12: Using optimizers for maximum likelihood estimation. Mixed-effects models.

Read this article by Hedges & Vevea for a discussion of conditionally random inference.

Week Five

Reading:

This article by Thompson & Sharp introduces restricted maximum likelihood in the context of meta-analysis.

No class meeting on February 17.

February 19: Restricted maximum likelihood. Using simulation to assess statistical methods.

Homework Two is available (due 2/26/2014).

Week Six

February 24: No class meeting.

February 26: Using maximum likelihood to estimate models for variance components.

Reading: A draft form of a paper by Vevea, Citkowicz, & Betts will be made available here. Please note the prohibition on citations. This is a work in progress, but it will give you an idea of what we are talking about this week.

Week Seven

March 3: Models for variance components (continued). Reading: No new reading.

Homework Three is available (due 3/12/2014).

March 5: Catching up (as we will probably be running behind by this point).

Week Eight

March 10: Wrapping up models for variance components: the prisoner's dilemma example.

Reading: this paper will help clarify the variables used in the example.

Homework Four is available (due 3/19/2014).

March 12: Reviewing results from the simulation homework. Weighted distributions. Weight-function models for funnel plot asymmetry.

Reading: here is an early simple example of an application of weight-function models for publication bias.

Week Nine

March 17:

Weight-function models with covariates. Reading: here is the paper describing the weight-function model with covariates.

March 19: Weight-function models with covariates (cont.).

Spring Break; no class meetings March 24 or 26.

Week Ten

March 31: Funnel plot asymmetry (continued). Working with the weight-function model for publication bias.

Homework Five is available (due 4/9/2014).

April 2: Funnel plot asymmetry (continued). Sensitivity analysis using a priori weight functions.

Reading: An article by Vevea and Woods is available here.

Week Eleven

Reading: articles at the following two links provide a useful introduction to Bayesian thinking. First, here is a pretty good Wikipedia introduction to Bayesian inference. Second, here is a Scholarpedia article on Bayesian methods.

April 7: Introduction to Bayesian statistics.

Homework Six is available (due 4/14/2014).

April 9: Markov chain Monte Carlo methods.

Week Twelve

Homework Seven is available (due 4/28/2014).

Reading: TBA.

April 21: Applying Bayesian methods to meta-analysis (fixed-effects models).

April 23: Bayesian analysis with random-effects models.

We will refer to this chapter (which is freely available on SAS's web site) in today's class.

Week Thirteen

Reading: TBA.

April 28: Bayesian methods with more complex meta-analytic models.

April 30: Bayesian methods with models for variance components.

Homework Eight is available (due 5/7/2014).

Weeks Fourteen and Fifteen

No new reading.

Class presentations.