Choosing the Best Statistical Test

Introduction

Decisions on how to evaluate data are crucial first steps in planning a study. Such decisions become part of the wider consideration of how to design the study. Statistics can be a challenging subject for most investigators. There are several statistical tests to choose from, and choosing the right one for a particular set of data can be an overwhelming task, particularly if such decisions are rendered after the data is collected. Even if an investigator doesn’t directly perform his own analysis, he or she should have a working knowledge of the subject to communicate intelligently and effectively with the statistician or epidemiologist. In the end, the investigator is responsible for the choice of statistical methods used but few receive adequate training in this area. Some of us who review articles lack the knowledge to determine if the best statistical handling was used in a particular study. This review focuses on compiling a "Best of the Web" on sites that aid in choosing the best statistical test for a given set of data. However, it is always recommended that the investigator begin with a solid foundation of statistics.

Last Update: July 2008

Search

A Google® search for "correct statistical test" returned 4.18 million hits. Adding the term "choosing" to "correct statistical test" returned 1.84 million hits. A valuable tool for finding more sites was using links on searched sites. Searching through the first 100 sites as well as multiple links through these sites produced several useful websites. The most helpful sources were often from university departments, more specifically computing and academic technology service departments. The sites varied from descriptive text to summary tables to interactive step-by-step guides to choosing the correct statistical test based on the data characteristics.

Best Websites

University of California, Los Angeles, Academic Technology Services

This UCLA website is run by the Statistical Consulting Group from UCLA. Their online sources include online classes and seminars and a comprehensive website on a variety of statistical issues and techniques. This site includes a page called "What Statistical Analysis Should I Use?" which includes a table of statistical tests to choose from based on the data characteristics. This table was adapted from a more detailed table by James D. Leeper, PhD, from the University of Alabama Health Sciences. The table also includes links to programming examples for common statistical software (i.e., SAS, STATA, and SPSS).

1. Authority: This well-recognized UCLA Statistical Consulting Group consists of four full-time staff with Ph.Ds with expertise in research methodology and applied statistics. Rating: 5

2. Currency: The statistical methodology presented is well established, and as such, reflects current knowledge. The website welcomes comments to improve and update the content. Rating: 5

3. Accuracy: The material seems accurate and their views agree with other recognized figures in the area. Rating: 5

4. Navigation & Readability: The layout is clear and the links are well identified. The site is searchable and easy to understand. Rating: 5

5. Utility: This website is free for all researchers worldwide. It is helpful to the investigator who is a beginning or intermediate programmer and has a "Frequently Asked Questions" section that includes many common questions regarding programming. Rating: 5

Summary: This site is an excellent overall resource for statistical help, including programming examples. It is often included as a useful link in other academic computing center websites.

British Medical Journal

This site includes the ninth revision of a statistical textbook, Statistics at Square One. Section 13 covers study design and choosing a statistical test. It contains a summary of what to consider when choosing a test and includes a similar table to the UCLA site.

1. Authority: Original text was written by T.D.V. Swinscow, and this edition was revised by M. J. Campbell from the University of Southampton. The British Medical Journal is well respected. Rating: 5

2. Currency: The statistical methodology presented is well established, but this online text was published in 1997 and not updated. Rating: 4

3. Accuracy: The material seems accurate and their views agree with other recognized figures in the area. Rating: 5

4. Navigation & Readability: The layout is clear and the links are well identified. The site is mostly text with few illustrations. Rating: 4

5. Utility: This website is free. Since it is an online text, it is not interactive and somewhat dense. Rating: 3

Summary: This site is a good reference, but others are more user friendly and easier to read. However, this may be a good resource to those not yet familiar with the nature of the variables in their dataset.

University of Texas Medical Branch, School of Allied Health Sciences

This site is a section of a web book, The Whole Art of Deduction, by Dr. Rodger Marion. The contents include sections about developing the research question to drawing conclusions, with a section on matching statistics with the research design.

1. Authority: This site is maintained by Dr. Rodger Marion, who is a professor in Humanities and Basic Sciences at the UTMB School of Allied Health Sciences. Dr. Marion’s Ph.D. is in psychology and communications and he has been involved in developing multimedia, instructional software that is used internationally at many colleges and universities. Rating: 5

2. Currency: The statistical methodology presented is well established, and as such, reflects current knowledge. The website welcomes comments to improve and update the content. Rating: 5

3. Accuracy: The material seems accurate and their views agree with other recognized figures in the area. Rating: 5

4. Navigation & Readability: This is a one-page layout without much to navigate. It is mostly text and somewhat busy. Rating: 4

5. Utility: The site is free and contains information found elsewhere. Rating: 4

Summary: This site contains a long list of statistical tests with short descriptors of when to use each one. It is not a site for quick answers as to which test should be used.

Thomson Wadsworth, Thomson Higher Learning

Thomson Wadsworth is part of Thomson Higher Learning, a company that provides customized learning solutions for colleges, universities, professors, students, reference centers, government agencies, corporations and professionals worldwide. This site includes educational workshops on statistics with a tutorial on choosing the correct statistical test. The tutorial focuses on the process of making the decision and important factors to consider before the decision is made. There are practice exercises at the end to test one’s knowledge.

1. Authority: The individual or group who established this workshop is not clearly stated. However, Thomson Higher Learning is well known. Rating: 3

2. Currency: The statistical methodology presented is well established, and as such, reflects current knowledge. Rating: 4

3. Accuracy: The material seems accurate and their views agree with other recognized figures in the area. Rating: 5

4. Navigation & Readability: The layout is clear and easy to navigate. Rating: 4

5. Utility: The site was accessible without having to register. Rating: 4

Summary: This site is a good overview for a beginning investigator before making a decision on which statistical method to use.

Web Center for Social Research Methods

This site is one of several that have a step-by-step question-answer section to lead you to the correct statistical test. Although there is a complete hypertext textbook on applied social research methods on the site, one needs to really know their data prior to using their interactive section on selecting statistics.

1. Authority: This site is run by Dr. William Trochim, a Professor in the Department of Policy Analysis and Management at Cornell University. He is well-published, has authored several books, and helped develop Concept Systems Incorporated, a software and consulting service company. Rating: 5

2. Currency: The statistical methodology presented is well established, and as such, reflects current knowledge. Rating: 5

3. Accuracy: The material seems accurate and their views agree with other recognized figures in the area. Rating: 5

4. Navigation & Readability: The layout is clear and easy to navigate, but the section on selecting statistics has no help links along the way. Rating: 3

5. Utility: The site was free and a good overall resource, but less so for the topic of choosing the correct statistical test. Rating: 3

Summary: This is a very good overall resource for statistics, but there is no website help available for the question and answer portion of the site.

Other Important Sites

• Interactive Statistics Pages, John C. Pezzullo’s Home Page
  http://statpages.org

  This site is a wonderful resource to find other links by topic of interest. There are links to online statistical software packages for analysis of data, including sections on power analysis. John Pezzullo was an Associate Professor in the Department of Pharmacology and Biostatisitcs at Georgetown University, in Washington, DC.

• Van Eck Computer Consulting, The Decision Tree for Statistics
  http://www.microsiris.com/Statistical%20Decision%20Tree

  This is a step-by-step interactive site developed by the makers of MicrOsiris, a statistical software package. Most answers include MicrOsiris commands corresponding with the answer. There is a table that also includes similar commands for SPSS and SAS.

• University of Minnesota, Computer Science personal site
  http://www-users.cs.umn.edu/~ludford/Table_of_Content.htm

  This site is maintained by Pamela Ludford, a doctorate student in computer sciences from the University of Minnesota, who self admittedly is not a statistician. This site was written in Spring 2003 with no evidence of an update. This site contains a list of statistical tests, with each test having a link to a short descriptor. This site is good for quick reminders.

• Canterbury Christ Church University
  http://www.whichtest.info

  This site is a quick step-by-step interactive site that leads to the correct test based on the choices of the investigator. Similar to other sites, the investigator needs to be familiar with the data, but there are help sections that contain examples to help you choose.

Disclaimer

The author has no personal or financial interest in any of the websites discussed above. The author has no personal or financial interest in any of the websites discussed above.