1.3 Experimental Design
What is the goal of every statistical Study? Collect data Use data to make a decision If the process to collect data is flawed, so is the conclusion.
Designing a Statistical Study Identify the variable of interest and the population. Develop a detailed plan for collecting data. If using a sample, make sure it represents the population. Collect data. Describe the data. Interpret the data and make decisions about the population. Identify any possible errors.
Data Collection Do an observational study Observes and measures characteristics of a sample but does not change existing conditions. Perform an experiment Treatment is applied to part of the group, results are noted Control group is used. No treatment is applied A Placebo may be used.
Data Collection(cont.) Use a simulation Use of a mathematical or physical model to reproduce the conditions of the situation or process studied. Allows study of situations that are difficult or dangerous to reproduce. Use a survey Ask questions, get answers.
Experimental Design Key Elements Control Why do we need to control the experiment? Confounding variable – when you can’t tell the difference between effects of different factors on a variable. Placebo effect – subject react favorably to placebo. No real treatment given. Blinding – subject does not know if they have been given a treatment or placebo. Double-blind – neither the researcher or the subject knows who gets the actual treatment until study is over.
Key Elements(cont) Randomization Process of randomly assigning subjects to different treatment groups. Randomized block design – subjects are classified by similar characteristics then randomly assigned within that group. (pg 20) Matched pair design – subjects paired by similarity. One gets treatment, one gets placebo.
Key Elements(cont) Replication Repetition of the experiment using a large number of subjects. Rule of large numbers An experiment that is performed a large number of times approaches theoretical results.
Sampling WHY? Population vs. sample Census or sample Error – may not be avoidable. Needs to be minimized. Biased – not representative of the population.
Sampling Techniques Random – every member has an equal chance of being selected. Use tables or random number generators. (A7 in book) Stratified - when it is important to have members from each segment of the population. Random sample within the smaller strata.
Sampling Techniques (Cont) Cluster – population falls into naturally occurring subgroups. One or more of the subgroups is used. Systematic – population is ordered and then every ___ is chosen. Convenience – because they are there. Often leads to biased results.