Empirical Study
Preparing the Petition | UW Human Subjects
Review | Sample Petition
Data Management | Data
Analysis | The Final Report
Preparing an III Petition for an Empirical Study
The III petition should be brief, generally 1-2 typed pages, but should provide enough information to give the committee a good idea of what you plan to do. The intent of the committee is to spare you the grief and heartache that come from discovering at the end of your third year that your III project is too ambitious, is giving you uninterpretable data, or is in some other way ill-conceived.
A good petition (and a good study) begins with a simple, clear purpose. This purpose should be reflected in each of the components of the study described below. The purpose will dictate which subjects to choose, what study design to use, what variables to measure, and what analyses to perform.
The III committee is diverse in terms of medical and technical background. Therefore, write your petition for a general audience. Avoid using jargon, and define any technical terms you can't avoid. If the information can be better presented in non-narrative form (graph, bulleted list, flow diagram, etc.), by all means do so.
Below are some guidelines for what to include. Because each study is different, not all of the items below will be pertinent to every study.
Background Provide a brief introduction to the problem you are investigating. This might include:
- Why is the problem important?
- What is already known about the problem and what remains unknown?
- What distinguishes your study from others tackling the same problem?
Hypothesis
A hypothesis is a testable assertion about the relationship between variables in your study. The study hypothesis is different from the null hypothesis which is a statistical construct only.
Methods
Subjects
- What are the inclusion and exclusion criteria for subject selection (experimental and control subjects)?
- How will subjects be identified/recruited?
- How many subjects will be used?
Study design. This doesn't have to be a lexical nightmare (i.e., a retrospective placebo-controlled double-blind cohort study). Just be sure to indicate what the basic structure of the study is. Again, this should flow directly from your purpose.
- Are 2 (or more groups) of subjects being compared?
- Are you looking to see whether 2 (or more) variables in a single group of subjects are correlated?
- Are you comparing pre-intervention to post-intervention values?
- Are you trying to predict an outcome variable from 1 (or more) other variables?
Variables
Generally, variables fall into one of 3 categories:
- Independent variables. These are the variables that are putative causal factors. In many cases these play out in the study as the basis for subject selection. For example, in a study looking at the role of body weight on survival after MI, the study could be conducted by comparing a group of obese patients to a group of normal weight patients.
- Dependent variables. These are the outcomes that are thought to be caused or influenced by the independent variable. In a study of the effect of a dietary regimen in diabetes, body weight could be a dependent variable.
- Control (confounding) variables. These are variables which, if not controlled, could exert influence on the dependent variable and so cloak or accentuate the apparent role of the independent variable. For example, imagine a study investigating the effect of hormone replacement therapy on breast cancer risk. Factors such as age at first pregnancy, family history, and body weight all contribute to breast cancer risk. Unless these factors are controlled in some way, the effect of hormone replacement therapy cannot accurately be assessed. In a prospective study, random assignment to treatment groups can effectively control confounding variables. In a retrospective study, other control means must be used.
Notice that body weight played a different role in each of the studies described above. Because of this, it is crucial that you describe the variables you will use in terms of the role they play in your study. This can be done simply and elegantly via a bulleted list of each type of variable.
Procedures
- How will data collection take place?
- If the study is a spin-off of a larger study being conducted by your sponsor, which aspects of the study you are responsible for?
Analysis
- How will you use the measures you collect to test your hypothesis?
The statistical procedures you choose will depend on:
- Your purpose
- Your study design
- The scale of measurement of the variables
- For help, try Selecting Statistics (http://socr.stat.ucla.edu/Applets.dir/ChoiceOfTest.html)
Common problems
The majority of III petitions require some tweaking before they are approved.
- The study was not adequately described. You must present sufficient detail about what you want to do and how you will do it.
- There is no hypothesis. There is a difference between general data gathering and data gathering targeted toward answering a specific question or testing a specific hypothesis. In some cases, a descriptive or hypothesis-generating study can be appropriate, but only when there is not sufficient prior knowledge to formulate a hypothesis.
- There is not sufficient power. If you are looking for a subtle effect or need to control more that 1 or 2 confounding variables by means of multivariate statistics, you will need lots of subjects.
- The purpose is not clear/design doesn't conform to stated purpose. The subjects sampled, the outcome variables, the study design and the analysis plan should all reflect the study's purpose. Are you collecting data that's not directly relevant to your research question? Is the study design described as a comparative study but the statistical analysis only mentions correlations? Is it clear just what you will do with all the data you collect?
- There is a logical flaw in the study. A common example is when important confounding variables are not controlled in the study or when cause and effect relationships will be inferred from data that is only correlational.
- The data to be collected is unlikely to be valid. Survey questions of a sensitive nature, for example, or questions where the respondent is asked to admit to poor professional or personal conduct are unlikely to yield valid results.
- The data you intend to use isn't really accessible. Find out what is and is not recorded in patient charts. Physicians' intentions or reasons for taking action (or not taking action), for example, cannot be found in charts.
- Student's role was not clear. The student should make an intellectual contribution to the study, not serve as the sponsor's technician.
- Project is too ambitious/complex. Ask yourself whether your study is feasible. Can you really get 250 subjects to respond to your survey? Are you looking at multiple outcomes or asking multiple research questions? Are there multiple phases to your project? I have never seen a petition go back for revision because the study was too simple.
Click here for Sample petition
Preparing the Petition | UW
Human Subjects Review | Sample Petition
Data Management | Data
Analysis | The Final Report