On Mon, 12 Oct 2009 02:45:28 -0700, Martin Bourgeois wrote:
>Nothing hard to understand about it; you dummy code a dichotomous 
>variable and correlate it with a continuous one. On test items, you can 
>dummy code a question as right/wrong and correlate it with test scores 
>to see if people who get it right tend to do better on the test.

To add to what Martin has written:

(1)  The point-biserial coefficient is the Pearson r calculated on
a "truly" dichotomous variable (e.g., an item is right or wrong,
gender/sex coded 0=female, 1=male, illness status, 0=not ill,
1=ill, life status, 0=dead, 1=alive, etc.) and an interval or ratio
scale continuous variable (e.g., total score on a test, height,
weight, number of years of smoking, degree of depression, etc.)
The specific equation for the point-biserial is simplify hand 
calculation, simplifications that arise from dealing with dichotomies
in form of zero and one.  The Wikipedia entry provides additional
background as well as the point-biserial's relationship to tests
which appear similar but are not based on the Pearson r (e.g.,
the biserial coefficient and the rank-biserial); see:
http://en.wikipedia.org/wiki/Point-biserial_correlation_coefficient 

(2)  The point-biserial coefficient is now often covered in 
introductory statistics textbooks.  For example, Gravetter &
Wallnau include it and present it as a way of obtaining an
effect size measure from the two-sample t-test.  One version
of the formula is the following:

point biserial r**2 = t**2/(t**2 + df-TOTAL) 
where **2 means raised to the power of 2.

Taking square roots of both sides provides the point-biserial r.
I like Glass & Hopkins (3rd ed) coverage of correlations and
suggest it for additional background:

Gene V. Glass and Kenneth D. Hopkins (1995). Statistical 
Methods in Education and Psychology (3rd edition ed.). 
Allyn & Bacon. ISBN 0205142125

This book is available on books.google.com but only in a
"snippet view" (I believe that it continues to be a popular book
that is used in a variety of undergraduate statistics courses); see:
http://books.google.com/books?lr=&num=100&id=SFmdAAAAMAAJ&dq=%22Glass+%26+Hopkins%22&q=point-biserial#search_anchor
or
http://tinyurl.com/yg2ftct 

-Mike Palij
New York University
[email protected]



________________________________
On Monday, October 12, 2009 4:45 AM, Michael S wrote:
>
>I never did understand the point-biserial statistic.I came across 
>it at Mizzou when I was in charge of grading lots of scantrons.
>It had something to do with test items but I never figured out why.

---
To make changes to your subscription contact:

Bill Southerly ([email protected])

Reply via email to