Study Design

1. Statistics is used to answer questions in a rational and defensible manner. The reliability of the answer is also estimated. At the beginning of the semester we called this "Decision Making". Some examples are:
    
   1. Do warning labels on cigarettes deter smoking?
   2. Should women in their 40's get yearly mammograms?
   3. Do gun control laws prevent violent crimes or gun accidents?
   4. Does driving while using a cellular phone increase the risk of an accident?
   5. Do sentences in criminal cases depend on the race of the accused?
   6. Does taking aspirin regularly reduce the chance of a heart attack?
       
2. Observational studies
    
   1. Compare 2 or more groups. Researcher does not decide who will be in each group. The groups are referred to as "treatments"
   2. "treatment" is an aspect of the subject
   3. Examples:
       
      1. smoking and cancer
          

         	smoking/ non-smoking are the treatments
         	treatment in not assigned to the subjects

      2. smoking and divorce rate
          

         	Do smokers and non-smokers differ in other ways than smoking alone?
         	Could some of these differing traits explain the discrepancies in divorce rates?

   4. observational studies are often difficult to interpret due to confounding factors. (read this)
   5. how would you determine if the discrepancy in median income of men and women is due to sex discrimination? (National Committee on Pay Equity)
       

      	Could the discrepancy between male and females be due to job selection?
      	It would be more useful to compare salaries of men and women in similar jobs and with comparable abilities. For instance you might want to compare incomes of male and female bank tellers with over 5 years experience who live in major cities (since people in cities generally are paid more than rural workers)
      	When we limit our analysis in this way we are controlling for confounding factors. In the above example we are controlling for job selection, job experience and location.

   6. Read about how to control for selection of major in a study of gender bias in college admissions.
   7. Example: Lead exposure linked to bad teeth in children
   8. Example: Alcohol protects from ischemic stroke
   9. "Correlation is not Causation"
        
3. Designed experiments
    
   1. Two or more groups are compared. The experimenter assigns subjects to the groups.
   2. "treatment" is assigned (randomly) to the subject
   3. Often compares a treatment to a control group using placeboes.
   4. randomized controlled double-blind study
   5. Example: Lovastatin study
   6. Example: Flu vaccine study
   7. Example: Physician's health study
   8. Example: Doctor Bias May Affect Heart Care
   9. Are confounding factors an issue in a designed experiment? Compare Table 1 from the Ischemic Stroke study with Table 1 from the Lovastatin study.
   10. double blind studies
   11. There are two different levels where randomization can be employed here:
        
       1. selection of participants in the study (recruitment), this is typically non-random selection
       2. assignment of subjects to treatment, this should always be random (read about the Salk polio trials)
           
   12. What should you do about subjects who don't comply with the study design? (read)

This page was last modified on October 17, 1999.