Why do we need significance testing in epidemiology? What do we indicate with a confidence interval? Let's try to use a practical example to illustrate how both concepts contribute to our better understanding of "the truth" behind investigations, for example studying causes of outbreaks within a population.
In most field epidemiology investigations we perform measurements (e.g. the occurrence of disease) and we compare the results between different parts of the study population (e.g. those exposed to a certain food item and those not exposed). If a food item that we investigate caused the outbreak, then we should expect that the risk of becoming ill is significantly higher among those exposed to this food item, than among those unexposed.
So what exactly do we mean with "significantly higher"? Let's look at the (fictitious) example of a Botulism outbreak, where the investigators found:
"The risk of illness was higher among diners who ate home preserved green olives (RR=3.6)."
Suppose this is all we know about the investigation: does this tell us that home preserved green olives have caused disease? Important questions to answer are: can this association be due to chance? How confident can we be in the result?
In order to answer these questions, let's take a closer look at the idea of statistical inference. That is a necessary background to understand significance and the probability that our measurement is different from what we expect. After that it will be easier to express our level of confidence in our results.
After reading this chapter, you will be better able to:
This chapter is based on the lecture "Significance testing and Confidence Intervals" of the EPIET Introductory Course.
1. Rothman KJ, Greenland S, Lash TL, editors. Modern Epidemiology. 3rd ed. Philadelphia: Lippincott Williams & Wilkins; 2008.
2. Goodman SN, Royall R, Evidence and Scientific Research, AJPH 78, 1568, 1988
3. Goodman SN, Towards Evidence Based Medical Statistics. !: the P Value Fallacy, Ann Intern Med. 130, 995, 1999
4. Poole C, Low P values or Narrow Confidence Intervals: which are more durable? Epidemiology 12, 291, 2001