Question 8: Compare the square root transformed numbers of attacks on the two drug treatments using the paired t test. How does the result compare with those from the sign test and from the t test on the raw data?
There are two ways to do this.
If we were to proceed as we have so far, we would choose option 1. We create two new variables, placeboroot = SQRT(placebo), and pronethroot = SQRT(proneth), using the same method as in Question 2. We then do the paired t test as in Question 1 and get the following output:
|Paired Samples Statistics|
|Mean||N||Std. Deviation||Std. Error Mean|
|Paired Samples Correlations|
|Pair 1||placeboroot & pronethroot||12||.988||.000|
|Paired Samples Test|
|Paired Differences||t||df||Sig. (2-tailed)|
|Mean||Std. Deviation||Std. Error Mean||95% Confidence Interval of the Difference|
|Pair 1||placeboroot - pronethroot||1.14219||.90408||.26099||.56776||1.71661||4.376||11||.001|
If we were to use the the second approach, the one sample t test using the differences, we proceed as follows. Click Analyze, Compare Means, One-Sample T Test. Choose variable 'Difference, square root number of attacks'. Click OK.
We should get the following output:
|N||Mean||Std. Deviation||Std. Error Mean|
|Difference, square root number of attacks||12||1.1422 .26099||.90408||.26099|
|Test Value = 0|
|t||df||Sig. (2-tailed)||Mean Difference||95% Confidence Interval of the Difference|
|Difference, square root number of attacks||4.376||11||.001||1.14219||.5678||1.7166|
The mean difference, its standard error, the confidence interval, and the P value are the same either way.
Using the square root transformed data, the P value is 0.001, smaller than but quite similar to the 0.0063 for the sign test. Both are significant. Compare this to the test for the untransformed data, which gave P = 0.107. By comparison, the square root transformed data give a much smaller P value and a different conclusion.
If the assumptions are met and the test is valid, the paired t test is more powerful than the sign test and has a better chance of finding a difference when there really is a difference in the population.
When the assumptions are not met, the t test may be less powerful than less demanding alternative methods.
Back to Exercise: Normal plot for angina and pronethalol.
To Applied Biostatistics index.
To Martin Bland's M.Sc. index.
To Martin Bland's home page.
This page maintained by Martin Bland.
Last updated: 30 October, 2006.
Back to top.