Presentation on theme: "Part 2 CHAPTER 6.1. DESIGNING AN EXPERIMENT The first goal in designing an experiment is to ensure that it will show us the effect of the explanatory."— Presentation transcript:
DESIGNING AN EXPERIMENT The first goal in designing an experiment is to ensure that it will show us the effect of the explanatory variables on the responses variables. Confounding often prevents experiments from doing this. The remedy is to compare two or more treatments.
EXAMPLE Sickle cell anemia is an inherited disorder of the red blood cells that in the United States affect mostly African Americans. It can cause severe pain and many complications. The National Institutes of Health carried out a clinical trial of the drug hydroxyurea for treatment of sickle cell anemia. The subjects were 299 patients who had had at least three episodes of pain from sickle cell anemia in the previous year.
CONTINUED Simply giving hydroyxurea to all 299 subjects would confound the effect of the medication with the placebo effect and other lurking variables such as the effect of knowing that you are the subject of an experiments. Instead, half of the subjects received hydroyxurea, and the other half received a placebo that looked and tasted the same. All the subjects were treated exactly the same (same schedule of medical checkups for example), except for the content of the medicine they took. Lurking variables, therefore, existed in both group equally and should not cause any difference in the results. The two groups must be similar in all respects before starting the medication. The best way to avoid bias is to choose who takes which medication randomly. Use a SRS.
WHAT DOES THAT LOOK LIKE Random Assignment Group 1 152 Patients Group 2 147 Patients Treatment 1 Hydroxyurea Treatment 2 Placebo Compare Pain Episodes
RESULTS The experiment was stopped ahead of schedule because the evidence was so compelling. The treatment group had much fewer pain episodes than the placebo group.
RANDOM ASSIGNMENT Random Assignment: One group for each treatment (relatively the same size), control group, response variable, then compare results The randomized comparative experiment is one of the most important ideas in statistics. It is designed to allow us to draw cause and effect conclusions. Logic of Experimental Design: Random assignment produces groups of subjects that should be similar in all respects before we apply treatments. A proper comparative design ensures that influences other than the experimental treatments operate equally on all groups. Therefore, different in the response variable must be due to the effects of the treatment.
EXAMPLE: STOPPING DRUNK DRIVERS Even though most people know how dangerous drunk driving is, some drivers are citied multiple times for driving while intoxicated. Is there anything that can be done to change their behavior? A subject was done to answer just that….300 people convicted of drunk driving three times in one year. The treatments, imposed after the third conviction, area fine plus a suspended jail sentence plus one of the following: (1) No treatment, (2), attend alcoholism clinic, or (3) participate in AA One response variable is whether or not they get arrest in the next year. Create chart to show the treatment groups.
TREATMENT GROUPS Random Assignment Group 1 80 Drunk Driver Group 2 98 Drunk Driver Group 3 122 Drunk Driver Treatment 1 Control Treatment 2 Alcoholism Clinic Treatment 2 Alcoholism Anonymous Compare Arrests the following year
CONTINUED Is this a randomized comparative experiment? Why or why not? How could we have made it more random? Is that realistic in this situation?
PRINCIPLES OF EXPERIMENTAL DESIGN 1.Control the effects of lurking variables on the response. Use a comparative design and ensure that the only systematic difference between the groups is the treatment administered. 2.Randomize- use impersonal chance to assign subjects to treatments. 3.Use enough subjects in each group to reduce chance variation in the results. We should insist on a different in the responses so large that it is unlikely to happen just because of chance variation.
STATISTICAL SIGNIFICANCE Statistical Significance: an observed effect so large that it would rarely occur by chance Statistically Significant results from randomized comparative experiments are the best available evidence that changing the explanatory variable really causes change in the response.