STAT 302 Chapter 8Chapter 8: Designing Experiments“To call in the statistician after the experiment is done may be no more than asking him to perform a post-mortem examination: he may be able to say what the experiment diedof.” -Ronald Fisher (1938)Review: In an experiment, the experimenter assigns values of the explanatory variable to each individual. In an observational study, experimenters observe without imposing treatments.- A treatment is a level of the explanatory variable randomly assigned to an individual.Example: Researchers studying Alzheimer’s disease have been searching for genetic causes. In a recent study involving Alzheimer’s patients over 65 year old, researchers found that patients with Alzheimer’s disease had lower levels of SORL1 protein in their blood due to their having a particular variant of the SORL1 gene. They found that having lower levels of SORL1 protein in the blood was statistically significantly associated with an increase in the production of the toxic A-beta peptide involved in the build up of “amyloid plaques” in the brain.- Was this an experiment or observational study?Example: A researcher wanted to know “Are adult women who take an aspirin every other day less likely to get a heart attack than adult women who do not take an aspirin every other day?” To answer this question, a random sample of 20,000 female nurses is selected from the registry of all registered nurses. Each nurse is asked if she takes an aspirin every other day or not. The nurses are then followed for 5 years to determinetheir heart health status. - Was this an experiment or an observational study?- What were the explanatory and response variables?125STAT 302 Chapter 8- What were the treatments?- Experimental units are the individuals or groups assigned to the treatments.Example : How do salinity and temperature affect the growth of shrimp? One researcher randomly assigned two aquariums each to three different salinity levels: low salt, medium salt, and high salt. For each salinity, one tank was assigned to a high temperature and one to a low temperature. Then 100 shrimp were grown in each tank. - What were the explanatory and response variables?- What were the treatments?- What were the experimental units?Key components to experimental designs1. A control group is a treatment group used to compare to the treatment of interest. Often the control group is assigned to take a _____________ medication (a medication that is designed to look like the medication of interest, but that has no actual effect on the body). When placebos are unethical, the group assigned to the standard treatment is the control group.Example : Gastric freezing was once a clever treatment for ulcers. In one study, patients swallowed a deflated balloon attached to tubes. Then a refrigerated liquid was used to fill the balloon with the idea that cooling the ulcer would relieve pain. 70% of patients experienced continued pain relief after experiencing this treatment.What was the treatment in this study? What were the experimental units? How could this study be improved?126STAT 302 Chapter 8A later experiment randomly divided ulcer patients into two groups. One group was treated by gastric freezing as before. The control group received a placebo treatment inwhich the liquid in the balloon was at body temperature rather than freezing. The results: 34% of the 83 patients in the treatment group improved, but so did 38% of the 78 patients in the placebo group. This and other properly designed experiments showed that gastric freezing was no better than a placebo, and its use was abandoned.Control groups are necessary for several reasons:1. Other events besides the treatment may contribute to experimental units’ responses. Patients usually get better, whether or not they take anything.2. On average, placebos work better than no treatment at all. One study1 even found that placebos that cost more relieve pain more frequently.2. Replication repeating the experiment on other experimental units allows us to study variability of the outcome due to measurement error or variability in any variables not being controlled in the experiment. Example: In the shrimp study above, there was no replication of aquariums, so we have no way to measure how many more or less2 pounds of shrimp we might have had if we had a different group of 100 shrimp.3. Subjects are blinded when they don’t know which treatment they were assigned. A study is double-blinded when attending physicians or psychologists making the diagnosis or evaluation are also ignorant of the treatment assigned.The goal of blinding is similar to the reasoning behind using a placebo: we don’t want individuals’ beliefs (whether the patients’ beliefs or the attending physicians’ beliefs) to affect the outcome of the study. Confirmation bias is a type of psychological bias displayed when researchers tend to search for information selectively in favor of their opinion. We wouldn’t want physicians to subconsciously diagnose patients in the placebo group as being sicker when in reality the drug and placebo work equally well.1 http://www.sciencedaily.com/releases/2008/03/080304173339.htm2 For the grammar Nazis in the room, I’m following Wikipedia’s advice about the usage of fewer vs. less found here: http://en.wikipedia.org/wiki/Fewer_vs._less127STAT 302 Chapter 84. Random assignment of treatments to individuals involves using a random number generator, flipping a coin, or drawing names out of a hat to choose whichsubjects get which treatments. Recall that random assignment is necessary for a study to be an experiment. However, we’re not always able to blind an experiment or use a control group. Subjects may be aware of the treatment they’re assigned to, no matter what. It may be unethical to use aplacebo when a known medication works, so sometimes researchers use the standard medication as the control.Which of the following describes randomization of 400 subjects to an antidepressant or placebo?a) Assign the 400 subjects to the two groups as haphazardly as possible.b) Use your own judgment to choose which subjects go into which groups.c) Use a random number generator in a computer to assign the subjects to the groups.d) Line up all 400 subjects according to birthdate, and put every other one in the antidepressant group and the others in the placebo group.Use randomization for assigning subjects to