Additional Material

David C. Howell

The material on this page is mainly textual and represents some of the stuff that I couldn't fit in either book, but wanted to write up. This material is all in draft form, and it may be revised from time to time. Please remember where you found it, and always give credit if you use it. Again, I would appreciate suggestions for further additions and/or clarification.

Missing Data

This is a discussion of alternative ways to handle missing data, whether those data come from a multigroup experiment or are continuous variables used in a regression problem.

I have written a chapter on missing data for the Handbook of Social Science Methodology, edited by Outhwaite and Turner and published by Sage. A preprint of that chapter is available by writing to me at David.Howell@uvm.edu I also recommend that you look at the following entry if your missing data occur in repeated measures designs.

Missing Data with Logisitic Regression

The purpose of this page is more about handling missing data than about logistic regression, but I think that it is helpful for both. The issues about missing data are very similar whether you are planning on a basic multiple regression or whether you have a dichotomous dependent variable and are using logistic regression.

Mixed Models for Repeated Measures

  • Analysis of Repeated Measures as a Mixed Model--Part 1
  • Analysis of Repeated Measures as a Mixed Model--Part 2
  • The use of mixed models offers another way of looking at repeated measures designs, and will return the same results as standard repeated measures ANOVA, but only when the design is balanced and there is no missing data. When there are missing data, the mixed model approach has a great deal to offer. You can obtain Part II at Mixed Models for Repeated Measures Designs-Part II

    Confounding Effects with Unequal Sample Sizes

    This is a discussion that begins to show how sample sizes can affect the interpretation of a study when you have unequal cell frequencies. Near the end of the article is an e-mail message that I sent to someone else, illustrating how what appears to be one effect can actually come out to be a different effect.

    Power when we're hoping for a difference greater than 0.00.

    This is a discussion under construction of what it means to talk about power when we wouldn't be satisfied just to prove that one mean is trivially greater than another.

    Intraclass Correlation Coefficients

    If you are correlating variables (such as scores from twins or gay partners) where there is no ordering within a pair (e.g. either twin could be considered twinA or twinB), you want an intraclass correlation coefficient.

    Multiple Comparisons for Repeated Measures Designs

    This is a discussion of ways to run multiple comparisons when you have a repeated measure. It addresses the often-asked question "How do I do a Tukey test on my repeated measure?".

    Type I, II, and III Sums of Squares

    I was asked for a demonstration of how you would compute the different types of sums of squares using the general linear model. Here that is.

    Contingency Tables

    This is a discussion of the traditional Pearson's chi-square, Fisher's Exact Test, and randomization tests of 2 x 2 contingency tables. It demonstrates that when marginal totals are not considered as fixed, the standard chi-square test is to be preferred to Fisher's Exact Test.

    Power Calculations

    This is a response to a question about how I used the normal distribution to create the power tables at the end of both books and in Table 15.1 in the Fundamentals book..

    Contingency Tables come in Different Flavors

    This is a discussion different ways of testing hypotheses about contingency tables. The basic idea is chi-square, but there is more to it than that.

    Testing Dependent Proportions

    This is a discussion of the designs in which you have two different groups, one receiving an intervention, and you test both before and after an intervention, if given, occurred.

    Permutation Tests on Factorial Anova

    There are a number of ways to create permutation tests with factorial analyses of variance, and this document covers a number of them. Code is also given for programming in R.

    Graphical Demonstrations Great for Class

    There are a number of very good graphical demonstrations of statistical issues available in the R package. This document shows how to download R and run those demonstrations even if you don't know R.

    Single Case Methodology

    I have had very little to say about single case studies in what I have written, but an excellent web site on this topic is available at the above link. John Crawford has published extensively in this field, and his web page is excellent. I have published several papers with John, but that really means that he does 95% of the work and I tell him how good the paper is.

    Return to Home Page

    David C. Howell
    Last revised: 02/23/200910/30/2008