Roger Bakeman pictureRoger Bakeman
Professor Emeritus
Georgia State University
Atlanta, Georgia, USA
Email: bakeman@gsu.edu.  Click here for GSU profile.

Welcome to my website. It lets you download memos I have written (please feel free to share) and gives you information about (and lets you download at no cost) computer programs that Vicenç Quera and I have written. These program run on Windows computers, or on Apple computers running a Windows simulator.


Putting Effect Sizes First: Magnitudes for Standard Statistics
Cohen (1988) provided guidelines for small, medium, and large effects. Using his guidelines can make results sections more readable. This memo explains some standard effect size statistics (e.g., standardized mean difference for between- and within-subjects t tests,  partial and generalized eta squared for analyses of variance) and, when necessary, how to compute them. This matters for designs with repeated measures because SPSS computes only partial eta squared when generalized eta squared is appropriate.

Cohen on Small, Medium, and Large Effects Sizes
A deeper dive into Cohen (1988) on effect sizes. He was the first to say that the values he provided are a last resort, that different areas of research should develop their own standards for what is a meaningful effect size, but that absent such consensus (rarely seen), his arbitrary guidelines are useful.

Why You Should Use Microsoft Word Styles
Applying Microsoft Word Styles to a document offers several advantages including standardization, flexibility, use of the Navigation Pane, and automatic Tables of Contents. This memo explain why you should avoid hard formatting and use Styles instead. (Send docx file to your Downloads folder.)

How the 7th Edition of the APA Publication Manual Differs from Previous Editions
There are few major changes from the 6th Edition.  Generally, I find the 7th edition both more detailed and more precise, while making many matters simpler and more consistent, computer aware, and flexible. This memo highlights the differences.

Outliers, Transformations, Nonparametric Statistics, and Recodes
When, if ever, to modify data before analysis? Many of us have learned “rules” that don’t always serve us well. This memo examines some of these rules and makes recommendations, including: always examine distributions before analysis, regard data transformations with skepticism—especially the rank order transformation underlying parametric statistics, and consider binary and ordinal recodes when they make data analytic and conceptual sense. 

Why You Should Not Use the Bonferroni Correction
Requests for the Bonferroni correction to “correct” for multiple tests continue to pop up, among other places, in reviews of articles submitted for publication. But the Bonferroni adjustment defies common sense and increases type II errors (the chance of false negatives). The memo includes arguments and references you can use to push back against misguided reviewers.

How to Create Box-and-Whisker Plots in Excel
Box-and-whisker plots (Tukey, 1977) provide a clear picture of how variable values are distributed. The main impediment to their use is the lack of an easy-to-use and widely available computer program to produce them. This memo explains how to produce publishable box-and-whisker plots with Excel.

What Type Sum of Squares Should You Use: I, II, or III?
For SPSS’s General Linear Model, the default is Type III sums of squares, but is that always the best choice?  Are there times when you should specify Type I or Type II instead?  This memo explains how to decide.

How to Transfer a Correlation Table from SPSS to Microsoft Word
This memo shows how to create correlation tables in a Microsoft Word document beginning with SPSS output in a way that is easy, accurate, and relatively free of error. SPSS output is first transferred to Excel (which, among other things, avoids the double-round error).

How are Percentile Scores  Computed (Not as Easy as You think)
Probably most of us think we know how compute percentile scores—until we actually try. Most statistics admit to a single computational formula; not so percentile scores. There is not one method but many. This memo explains the various possibilities.


GSEQ iconThe Generalized Sequential Querier (GSEQ) is a computer program for analyzing sequential observational data. It computes a variety of simple and contingency table statistics. Simple statistics include frequencies, rates, durations, and proportions (percentages).  Vicençe Quera (University of Barcelona, Spain) and I have been developing this program since the 1990s.

ILOG iconThe Interactive Log-linear (ILOG) program performs, as the name suggests, log-linear analyses interactively. Log-linear analysis is a useful but under-used technique for investigators who categorize some entity (persons, dyads, various kinds of events such as bids for attention or turns of talk in conversations, etc.) on several dimensions using nominal scales and then tally the results in a multi-dimensional contingency table.

The KappaAcc program computes various kappa statistics and estimates observer percentage accuracy.  When observers code or rate events, Kappa is a statistic that gauges inter-observer agreement corrected for chance. Because no one value of kappa can be regarded as universally acceptable, estimated accuracy provides a way to judge the adequacy of the computed value.

The BootCI program compute percentile bootstrap confidence intervals (CIs).  For means and correlations, estimation formulas for the standard errors required to compute CIs are easily found in standard statistics texts, but standard errors for the increases in variance accounted for when using hierarchic regression (ΔR2, also called the semipartial correlation coefficient squared) are more problematic.  The major merit of BootCI is that it computes these using bootstrap methods.