.
G154 Stone Building Tuesdays 2:00-4:15 PM The Multivariate Analysis of Categorical Data Susan Carol Losh Department of Educational Psychology and Learning Systems Florida State University |
LINKS TO INFORMATION ABOUT EXERCISES AND FEEDBACK GENERALLY GO HERE AT THE TOP. |
WATCH FOR ANNOUNCEMENTS ABOUT :
ANY SCHEDULE CHANGES (UPDATED) PAPER INFORMATION GENERIC FEEDBACK ETC. |
OVERVIEW |
MATERIALS |
|
INFO |
TOPICS |
MY OFFICE: 3204 Stone Building
850-644-8778 OFFICE HOURS: Any exceptions to be announced 1:00-2:00 P.M. Tuesday 2:00-3:15 Wednesday & by appointment slosh@fsu.edu |
FSU OLD Stone Building INSTRUCTOR: Professor Susan Carol Losh
|
PLEASE INFORM ME IMMEDIATELY
IF YOU REQUIRE ANY ASSISTANCE WITH DISABILITIES!
|
The Multivariate Analysis of Categorical Data addresses models with categorical dependent variables. Analyses with these variables do not lend themselves to some of the more traditional statistics used in education, the behavioral or the social sciences. Part of our course centers around causality and testing causal models. Along the way we will encounter maximum likelihood estimators, Likelihood-Ratio Chi-Squares, and diverse fit estimates. This material requires familiarity with one course past the basic introductory statistics class (e.g., multiple regression, ANOVA, the General Linear Model or structural equation models).
By the end of the semester, you should be able to set up, test and interpret:
--loglinear
and logit models
--logistic
regression equations for both binary and multinomial dependent variables
Simply reading materials does not confer peace of mind, let alone a healthy respect for everything that can go wrong. Until you try basic applications, these techniques are difficult to understand or evaluate. Once you work an exercise, the materials become much clearer. You will complete four short exercises designed to familiarize you with terminology, basic concepts, and computer techniques. Finally, you will analyze data and write a short paper, as well as give a short presentaton based on your paper. A later handout will set the fundamental parameters for the analysis paper. Several datasets are available to you or you may use your own data.
It is insufficient to simply report results,
results
are interpreted. Did your original substantive hypotheses receive
any support? Were they unequivocally rejected? Did they make sense in your
causal system? We will address causal issues several times throughout the
semester.
|
|
MAIN TEXT: Alan Agresti, An Introduction to Categorical Data Analysis. SECOND EDITION! (Wiley, 2007)
I will introduce several ways to save some
money on the Agresti book. Please wait until the first day of class.
|
Other readings from: Nigel Gilbert, Analyzing Tabular Data: Loglinear and Logistic Models (1993; ISBN = 1857280903) and
Online Course Guides
The
Agresti book is at Bill's and the FSU Bookstore. Used copies ARE available.
It is quite possible that you can find
used copies online at Amazon and other sites for a good deal.
I will make copies of the Gilbert chapters
for you. Sometimes you can find VERY cheap copies of Gilbert at online
sites, such as Amazon.
ALL MY COURSE LECTURES will be placed on the Internet and linked in with each course topic.
Course guides will be keyed to the readings.
See the top of each Guide as it is posted.
The lecture urls have the general form
of:
Please type in course urls EXACTLY. There is no "www" in these urls.
Some material in the guides will be covered during class. However, we will also use class time for instruction related to each exercise, demonstrations, presentations and feedback.
|
Here is information about assignments, due dates, and course weights.
There will be four short equally weighted exercises. While each exercise will focus on the immediately prior units, please be advised that this material is cumulative in nature.
Exercises have several purposes:
To familiarize you with terminology, basic operations, and associated computer programs.
To help practice your basic analysis and results interpretation skills. For example, logistic regression coefficients are often used and equally often misinterpreted.
To alert you to common problems that occur with different kinds of analyses and ways to solve these problems.
All four exercises put together will count a total of 40% toward your final grade.
Details on each assignment are posted to our course WEB site prior to the due date.
As exercises and exercise feedback sites are created and posted, watch the space at the top of the Guides for information and links.
An analytic paper will count 40% toward your final grade.
A presentation based on your analytic paper will count 20% toward your final grade.
In the paper, you will analyze data using course material and interpret your findings. The general format of the paper will resemble a journal article.
You may use your own data, data from
your major professor, or one of the many data sets I have available. There
are many databases online.
I use plus and minus grading, throughout and for final grades.
If I think you are having trouble with
the material, I will alert you immediately. I expect you will seek help
as quickly as possible. If you receive such an alert, please take it very
seriously.
|
|
|
1: Terminology and purpose | February 7 | 10 percent |
2. Using a hierarchical loglinear program | March 7 | 10 percent |
3. Exercise on general loglinear models | March 28 | 10 percent |
4. Loglinear to logit transformations (includes program exercise) and logistic regression | April 11 | 10 percent |
PRESENTATION ON PAPER TOPIC & ANALYSIS | April 18 & 25 | 20 percent |
COURSE PAPER | May 3 by NOON | 40 percent |
EXERCISE DUE DATES TURNING IN EXERCISES |
We are on a tight schedule so exercises must reach me Because of our schedule, I try to return assignments quickly. If you are late, I just might hand them back before you turn yours in. |
PLEASE DO NOT SEND ME ANY EMAIL ATTACHMENTS. THEY WILL NOT BE OPENED! OUR BLACKBOARD COURSE DISCUSSION BOARD takes Word and .pdf files. ONE MORE NOTE ON EMAIL: Widespread viruses spread through email use subject lines such as "hi" "hello" "hi there" "thanks" or "my test" or no subject heading at all. If not a virus, some of these subject lines are used to camophage advertisements for products I neither use nor want (I received over 75 of these over Break!) PLEASE USE SOME OTHER SUBJECT LINE. INFORMATIVE SUBJECT LINES INCLUDE:
|
I use plus and minus grading for final grades.
|
Our course is WEB assisted through Blackboard at FSU. You MUST be registered for edf6937-05 to access our Blackboard site. To access our course through Blackboard, here is what to do. Go online to:
(You will be forwarded to the new, more complicated url. . The above works and is easy to remember.) Enter your FSU username (USERNAME ONLY!) and password to log in. For example, I would enter "slosh" ONLY and omit the "@fsu.edu" part. Then click on:
Seminar in Advanced Research Problems
If you DON'T have an FSU account, you need to get one NOW. Go to the Computing Services website and follow the links to register online for your FSU account and email. (Of course you can set up your FSU email account to forward to the email account of your choice.)
You need an FSU account to log into BlackBoard.
Our course can also be accessed directly through the Internet and FSU's mailer system. Go to:
(That's THIS WEB address.)
You can link to nearly all the course sites
from this central location.
|
I will use WEB-assist for several course features; this is sometimes called a "flipped course":
|
There may be some variations from this syllabus. Please check back weekly and watch Blackboard for any announcements.
|
|
|
January 10-17 | Introduction: Issues in Modeling
Causal issues in experimental, nonexperimental and observational data and their implications for models |
Navigating our course web sites
What
are the fundamental analytic problems?
|
January 17-24 | Review General Linear Model; issues of
basic terminology
Introduce terminology (Odds-Ratios; MLEs; iteration; Chi-square as goodness of fit); The basic two-way (2 X 2) cross-tabulation table recast |
What
are important aspects of the GLM?
What are the building blocks and terms of multivariate analyses of non-numeric data? |
January 24-31
|
Basic loglinear models.
A probabilistic model for table cell counts. Poisson and multinomial distributions. Begin with two way table, extend initially to three way (2 dichotomous, 1 not, then to multinomial). Frequency equations; transformed to log-linear equations. |
What
do loglinear models "look like"?
What is a general cell frequency model? Show me an example! |
February 7 | EXERCISE 1 DUE | Basic terminology exercise |
February 7-14 | N-way tables
Model construction and model testing in the loglinear model |
Extending
the model to "n dimensional" cross-tabulation tables
How do I test my loglinear model? Which parameters can I drop? |
February 14-21 | Introduction to programs | |
March 7 | EXERCISE 2 DUE | Program 1 and writing a loglinear equation |
February 28-March 7 | The logit model.
Transformation from the loglinear to the logit model. Transform equations to logit model. |
What
is a logit model?
What are the advantages to a logit model? What's the relationship between a loglinear model and a logit model? |
March 13-17 | No class--Spring Break | |
March 21 | Issues in testing logit models and demonstrations. | |
March 28 | EXERCISE 3 DUE | Nonhierarchical Loglinear models |
March 28 - April 4 | The special case
of logistic regression. Dichotomous and polytotomous dependent variables.
Poisson, binomial and multinomial regression More programs: Multinomial logistic regression |
What's
the relationship between the GCF and logit models and logistic regression?
How do I handle dependent variables with more than two categories? |
April 11 | EXERCISE 4 DUE | Loglinear to
logit transformation
Running a logit model program |
April 18 | First
draft analytic paper (allows me to get
it back to you in time to do revisions)
Presentations |
Please submit paper draft through turnitin
onBlackboard
PRESENTATION INFO |
April 4-25 | Various extensions and special cases:
Model Fitting Ordinal response variables versus nominal models; Quasi independence; Structural versus sampling zeros Presentations |
When
Chi-square just isn't enough
Many many measures of fit Are there advantages to an ordinal dependent variable? What about cells with no cases? What are some other extensions? |
May 3 by noon | Analytic paper due FINAL DUE DATE!! submit through turnitin | Paper Guide |
A LECTURE (AND ASSOCIATED MATERIALS)
WILL BE LINKED WITH EACH TOPIC AS THE SEMESTER PROGRESSES.
|
READINGS AND ASSIGNMENTS |
|
This page was created with
Netscape Composer.
There may be some minor
changes as the semester progresses.
Your patience is appreciated.
Susan Carol Losh
December 31 2016