|
|
 |
Listserv Message Center
 |
Teaching Stat: A Little Computation Is Good? |
 |
|
|
|
| Posted by: | Rebecca Warner |
| Title/Position: | Professor |
| School/Organization: | University of New Hampshire |
| Sent to listserv of: | SPSP |
| Date posted: | July 28th, 2010 |
|
|
|
|
 |
 |
|
I'm following this discussion with great interest. I agree that students do not gain much understanding from some kinds of by hand computation, but I do think there is a role for limited computation exercises, with a focus on what the numbers mean and how values are related to design decisions.
For example, when we first talk about deviations from the mean, I take a simple variable such as height, provide a population mean and standard deviation, and have each class member compute his or her own deviation from the mean. I talk about my own deviation from the mean for height for women (I am shorter than the average woman, but not to the point where I would be regarded as unusually short) and ask them to think about the information from their own deviation from the mean on this and other variables. Making it personally relevant in this way seems to bring it to life for them. Then we discuss the problem of how to summarize information about differences among people represented by deviations from the mean, and why the obvious first choice (just add them up) does not work.
When we discuss the independent samples t test, the focus is not on computation of all the sums of squared deviations. At that point, we talk about the important summary information we have for each group (mean, N, and standard deviation) and how we use that information to set up a t ratio. Also, we discuss how design decisions (e.g. selection of dosage levels in a simple experiment, control over extraneous variables, increases in sample size) are likely to influence the size the size of the t ratio.
As noted by others, I think it's very helpful to students to acknowledge and discuss the many ways in which actual research practice departs from ideal procedures described in introductory research methods.
Based upon considerations similar to those in this discussion, I've written a book that spans a wide range of topics (starting with sampling, significance testing, and bivariate tests, up through multiple regression, ANCOVA, MANOVA, and factor analysis) in which I use this approach all the way through (that is, focus on how design decisions influence numerical outcomes, and on interpretation more than on computation). The book is: Warner (2007) Applied statistics: From bivariate through multivariate techniques, Sage. SPSS is used for analyses and each chapter focuses on one example throughout, starting with design issues and ending with a Results section. The publisher provides examination copies upon request from instructors; supplements available to instructors include all data sets, fully worked SAS examples, answers to comprehension questions, and PowerPoint slides for all figures and tables. Sage produced the book nicely and priced it very reasonably (about $80.00 through on line booksellers). I use it in both advanced undergraduate and beginning graduate courses, but based on this discussion, I wonder if a subset of the chapters might also work well in an introductory level undergraduate course.
Link for Complimentary Copy Requests:
http://www.sagepub.com/booksProdDesc.nav?prodId=Book225779&currTree=Courses&level1=Course12&level2=Course67&
Please forgive the self promotion,
Rebecca
|
|
 |
|
|
|
|
|
|
Forum Home Page
