In: Statistics and Probability
A study by Lefèvre et al. (2010, PLoSONE 5: e9546) investigated whether alcohol consumption affected peoples' attractiveness to the mosquito, Anopheles gambiae, which is the primary vector for malaria in Africa. A total of 43 human participants were tested, and the attractiveness of each participant was tested twice. The first test collected a baseline attractiveness for each participant. The participants then drank either a liter of beer (n=25 participants) or a liter of water (n=18), and after 15 minutes, the test was repeated. The type of drink was randomly determined for each subject.
To estimate a person's attractiveness to A. gambiae, a vial of 50 mosquitos was released into the Y-tube apparatus shown in Figure 1. Participants were seated in a small tent, which was connected by an air hose to one of two mosquito traps. This allowed the mosquitos to detect and respond to olfactory cues from the participants, without exposing the participants to the mosquitos. The proportion of mosquitos that flew toward the participant was recorded as the response variable.
You can find the Lefèvre data on the website in the file lefevre_2010_mosquitos_and_beer.csv. Each row in in the data represents a single participant. The first column is an id code. The second column is a factor that indicates the subject’s treatment-level for drink (beer vs. water). The third column gives the proportion of mosquitos that flew in the participants' direction in the baseline measurement, and the fourth column gives the proportion after drinking. Column five is the change from baseline after drinking (i.e., afterDrink – beforeDrink).
I
In the following questions, we will use two different strategies to analyze the data. In Question 2, we will conduct two completely separate tests that compare attractiveness before drinking a beverage to attractiveness after drinking a beverage, and then compare the result that we get with water to the result with beer (if it helps, imagine two different research groups test the same question, but one uses water and one uses beer; afterward, they meet up and compare notes). In the second strategy (Question 4), we will directly analyze the difference in average change from baseline for beer versus water. One of these strategies is a much better approach to use in real life. Which one is better, and why?
Here two different methods are being used to test the proportion of mosquitos attracted with beer and water.
In the first approach, two completely separate tests that compare attractiveness before drinking a beverage to attractiveness after drinking a beverage, and then compare the result that we get with water to the result with beer. This involves at least two different tests and might lead to multiplicity.
In the second appraoch, we will directly analyze the difference in average change from baseline for beer versus water.
The second appraoch is better in that the participant acts as his/her own control. When we use the first appraoch, the overall proportion before and after the alcohol/water is summarised and might lead to more variation into the data. In teh second appraoch, these variations are eliminated to a larger extent because the difference after drink is taken from before for the same participant. That is if a participant inherently has an attraction even without alcohol or water or any other factor which we deem that affects attraction gets nullified since we take the difference from baseline for the same pariticipant and the differences so obtained will have only the variation due to alcohol or water. Also the first test inflates the because of number of tests.