Other's On-Line Resources (Chapter 10)

 Comparing 2 Correlated Means With a t-Test Description: In using this interactive on-line resource, you'll be able to see how a t-test functions to compare the means of 2 correlated samples. You'll also see a wonderful pictorial representation of the "matching" (or what some would refer to as "pairing") within the data. What to Do: Click on the colored title of this on-line resource: "Comparing 2 Correlated Means With a t-Test." After clicking on "Begin," click "Simulate" in the next screen you see. Look inside the gray box and find the 2 sample means, the t-test's df, the calculated value, and the critical value. (Is df+1 = the number of pairs of scores?) Look in the upper right-hand corner to see whether your t-test comparison produced a statistically significant result. Repeat Steps 2-4 several times. Then click "Simulate 5000." (Do approximately two-thirds of your replications turn out significant?) Sky Huck's Puzzle Question: If you click "Reset" and then change "n" from 8 to 100, how likely is it that the t-test will yield a significant difference between the two sample means? After making your guess, click "Simulate 5000" to see. Comparing 2 Independent Means With a t-Test Description: In using this interactive on-line resource, you'll be able to see how a t-test functions to compare the means of 2 independent samples. You'll also see the absence of any "matching" (or what some would refer to as "pairing") within the data. What to Do: Click on the colored title of this on-line resource: "Comparing 2 Independent Means With a t-Test." After clicking on "Begin," change (in the next screen) the number for "Population rho" from 0.50 to 0.00. Then click "Simulate." Look inside the gray box and find the 2 sample means, the t-test's df, the calculated value, and the critical value. (Is df+2 = the total number of scores in both samples combined?) Look in the upper right-hand corner to see whether your t-test comparison produced a statistically significant result. Repeat Steps 2-4 several times. Then click "Simulate 5000." Sky Huck's Puzzle Question: If you click "Reset" and then change "n" from 8 to 100, how likely is it that the t-test will yield a significant difference between the two sample means? After making your guess, click "Simulate 5000" to see. Estimation With a Mean Description: This interactive on-line resource allows you to build a 95% confidence interval around a sample mean, with you being in control of the numerical values of the sample's M, SD, and n. What to Do: Click on the colored title of this on-line resource: "Estimation With a Mean." In the 3 boxes, set the sample mean = 100, the standard deviation = 15, and n = 25. Click on the gray bar that says "Calculate the Confidence Interval." On the next screen that pops up, look at the CI that's been built for your situation. Click on your browser's "Back" button, and then repeat Steps 2-4, varying the values of M, SD, and n. Sky Huck's Puzzle Question: With M, SD, and n initially set equal to 100, 15, and 5, respectively, how much do you need to increase n (while holding M and SD constant) in order to cut the CI in half? Regression Toward the Mean Description: You'll sometimes be confronted with a "research discovery" that comes from a study in which a single group of people was first pretested, then given some form of treatment, and finally posttested, with a correlated t-test used to compare the group's pretest and posttest means. This clearly-written on-line resource, with its wonderful graphics, explains why such a t-test comparison can, in certain circumstances, fail to assess correctly the treatment's worth. What to Do: Click on the colored title of this on-line resource: "Regression Toward the Mean." Carefully read the text material and examine the pictures. Sky Huck's Puzzle Question: Suppose a group of 200 of people is tested, yielding a mean of 80 and a SD of 8. Next suppose that those folks (n = 10) who score the worst (M =60, SD = 2) are identified, given some special training for 2 weeks to improve their skills, and then administered a posttest after the training. (A correlated t-test is used to compare the small group's pretest and posttest means to see whether the training worked.) Finally, assume that (1) the test used to in this study is known to have a test-retest reliability of .60 when used with folks like those who went through the training, and (2) that people generally improve their test performance by 5 points when retested, simply because they understand the format better, are familiar with the time limit, etc. Knowing all this, how well do you think the subgroup that went through training would do on the posttest, on average, EVEN IF THE TRAINING PROGRAM HAD ABSOLUTELY NO BENEFIT WHATSOEVER?