MPH505 WK 5 Discussion

Erika Lindwall
MPH505 Biostatistics
Week 5 Discussion on Hypothesis Testing
The purpose of the article was to demonstrate the limitations of hypothesis testing and
presenting only a P value to show statistical significance (Gardner & Altman, 1986). One
important limitation the authors point out is that showing statistical significance with a P value
can be misleading to its actual medical importance or lack there of. Instead of using hypothesis
testing, the authors assert that confidence intervals should be applied to gain a range of possible
values in which the actual value of interest may lie. They conclude that by using confidence
intervals as the standard, the field will move away from focusing on statistical significance and
instead focus on what is really helpful in a practical sense.
I would agree with the statement that in medical practice, it is more important to know
how much a certain illness or certain drug effects the people in the sample versus how
"statistically significant" the results were. This is because, as the authors state on page 746,
minute differences in results between subjects in a large sample study will show statistical
significance where there really is no clinical significance. Conversely, in a study with a small
sample, clinically significant results may appear to be statistically insignificant with the reported
P value. So, to summarize, statistical significance may not have much meaning in a practical
sense.
When the authors state that soley presenting the P value encourages lazy thinking, I think
what they mean is when readers see a P value that shows significant or insignificant results, the
reader will jump to the conclusion that it is medically significant when that might not be the case.
I would agree with this statement. When one sees results that denote significance, it is easier to
assume clinical significance instead of searching for more information about what that statistical
significance really means.
After reading this article, I think I do have some reservations about hypothesis testing.
However, I think that the information gained from hypothesis testing can still be valuable when
coupled with confidence intervals. It also seems that hypothesis testing provides a good starting
point for further study as it provides with some precision whether or not control and
experimental groups of are truly "different" (Hubbard & Lindsay, 2008).
References:
Gardner, M. J., & Altman, D. G. (1986). Confidence intervals rather than P values: estimation
rather than hypothesis testing. British Medical Journal (Clinical Research Edition), 292(6522),
746.
Hubbard, R., & Lindsay, R. (2008). Why P Values Are Not a Useful Measure of Evidence in
Statistical Significance Testing. Theory & Psychology, 18(1), 69-88.