Question 4: Which of these would be suitable here?
The samples are fairly large, both greater than 50, so we could use the large sample z method. The distributions are slightly skew, with two rather extreme values for the males, and the spread of the distribution looks slightly wider for the males. The two sample t test is very robust to small departures from assumptions, so it would make very little difference which method we used.
As many programs do not provide a separate large sample test, we may have to use the command for the two sample t method with unequal variances is any case. This give the mean difference, males minus females, = 18.6 mm, standard error = 3.4 mm, 95% confidence interval = 11.9 to 25.4 mm, t = 5.45, P<0.0001. The t method with equal variances gives mean difference, males minus females, = 18.6 mm, standard error = 3.3 mm, 95% confidence interval = 12.2 to 25.1 mm, t = 5.73, P<0.0001.
There does not seem much point in giving the confidence interval to three significant figures and we can round both confidence intervals to 12 to 25 mm.
Back to question.
Back to IAPT index.
To Martin Bland's home page.
This page maintained by Martin Bland.
Last updated: 27 February, 2009.
Back to top.