The New Statistics: Estimation for better research
The bookCumming, G. (2012). Understanding The New Statistics: Effect Sizes, Confidence Intervals, and Meta-Analysis. New York: Routledge
- Explains estimation, with many examples.
- Designed for any discipline that uses statistical significance testing.
- For advanced undergraduate and graduate students, and researchers.
- Comes with free ESCI software.
- May be the first evidence-based statistics textbook.
- Assumes only prior completion of any intro statistics course.
- See the dance of the confidence intervals, and many other intriguing things.
The main message of the book is summarised in two short magazine articles, in The Conversation and InPsych, and on ABC Radio, and you can also download the book's Preface, Contents, and a sample chapter. There is further information about the book available from the publisher and a Users’ Guide [DOC 75KB] explaining how to get the best out of the book.
The top journal Psychological Science now “embraces the new statistics.” See the new policy. It published my invited tutorial article to help authors adopt the new statistics where possible. Here is a brief summary of the advantages of making the change.
My main current research is in the area of statistical cognition, which is the study of how people understand - or misunderstand - statistical concepts, and various different ways to present the results of statistical analyses. I advocate the evidence-based practice of statistics, meaning that our selection of a statistical technique should be supported by cognitive evidence that people understand it well.
I am especially interested in replication, which is the topic of Chapter 5 in the book. One of many reasons that CIs are better than p values is that CIs generally give quite good information about what is likely to happen on replication of an experiment, whereas a p value gives almost no information about replication. The dance of the p values illustrates how p values vary enormously with replication, thus indicating how terribly uninformative they are.
You can contact me at E: firstname.lastname@example.org
See the reviews on Amazon.
In this book Cumming does exactly what's needed - he clearly explains significance-testing and confidence-interval techniques, and in the process leaves no doubt as to which should be preferred. I hope that this book will be read by anyone who plans to do research of any kind. Geoffrey Loftus, University of Washington, Seattle, USA
It is about time that the ‘new statistics’ get the place they deserve in statistics education, to further true understanding instead of following statistical rituals. Geoff Cumming’s excellent book is a milestone towards reaching that goal. Peter Sedlmeier, Chemnitz University of Technology, Germany
Geoff Cumming is a preeminent author in statistics reform and his new book is an important work in this area. Written in a clear, accessible way, Cumming covers essential topics in data analysis, including effect size estimation, confidence intervals, and meta-analysis. Readers in the social and behavioral sciences will learn new ways to look at their own results and avoid common traps of traditional significance testing. Rex Kline, Concordia University, Canada
Cumming makes strong arguments for confidence intervals and meta-analysis as better ways of representing and thinking about research. His graphical, interactive approach makes the ‘new statistics’ highly accessible and remarkably impactful on our statistical understanding. Robin K. Henson, University of North Texas, USA
See also a review by Eoin O'Connell in Significance: Statistics Making Sense
ESCI (Exploratory Software for Confidence Intervals)
- Pronounced 'ESS-key'
- Developed to accompany the book.
- Runs under Microsoft Excel.
- Easy and free to download.
ESCI Readme [DOC 80KB] has the latest information about ESCI, and hints for getting going. [updated 24 July 2012]
To download and use ESCI
- Choose one of the three versions below
- Save the ESCI .zip file to your hard disk (N.B. Don’t open the .zip file straight from this webpage)
- Open the saved .zip file, extract (unzip) all files and save the contents in a single folder on your hard drive
- If you encounter a problem, it may be a known Microsoft problem with Windows XP and .zip files: You need to have the ‘Confirm open after download’ option selected. See Microsoft Support Article ID 308090
- Open an ESCI module—a regular Excel workbook—in Excel.
- Make sure macros are enabled.
- Adjust zoom factor if necessary, so the display fits your screen.
- Enjoy the mean heap, dance of the p values, and other neat simulations.
ESCI for Windows (Excel 2007 or 2010) [ZIP 7MB] (updated 24 July 2012)
ESCI modules are .xlsm macro-enabled files.
ESCI for Windows (Excel 2003) [ZIP 3.9MB] (updated 24 July 2012)
ESCI modules are .xls files.
This version runs fastest and has best interactivity. Choose this if you have access to Excel 2003.
ESCI for Macintosh (Excel 2011) [ZIP 5.7MB] (updated 24 July 2012)
ESCI modules are .xlsm macro-enabled files.
ESCI modules released during 2001 to 2010 were designed to accompany various journal articles. To a large extent, but not entirely, those earlier modules are now superseded by ESCI that accompanies the book. The earlier modules and information about the journal articles are still available.
View the original dance of the p values, as discussed in Chapter 5.
There are now 10 additional videos that present basic aspects of the new statistics. The list below gives the YouTube links. (Alternatively, go to YouTube and search for ‘Geoff Cumming.’) No. 9 in the list below is a new version of the dance of the p values.
The same 10 videos are also available via iTunes U.
- Intro Statistics 1 Frequency Distributions
- Intro Statistics 2 Descriptive Statistics
- Intro Statistics 3 The Normal Distribution
- Intro Statistics 4 Sampling
- Intro Statistics 5 Standard Error
- Intro Statistics 6 Estimation
- Intro Statistics 7 Confidence Intervals
- Intro Statistics 8 NHST to Estimation
- Intro Statistics 9 Dance of the p Values
- Intro Statistics 10 Meta-Analysis
About 2.5 years after publication the first error found in the text was reported: In Equations 4.1 and 4.2 on p. 94, the left parenthesis should come just after the big sigma, not before it. My apologies!