10.1 Hypothesis Testing Overview of Process
Process for Hypothesis Testing
- Start with a research question: Are job applicants with criminal records less likely to receive call backs for interviews than applicants without criminal records?
- Develop a theory of how the world works
- E.g., “Job prospects are, in part, a function of someone’s criminal record.”
- Construct “null” and “alternative” hypotheses
- E.g., \(H_o\): “Applicants with a criminal record will receive call backs at similar rates as those without a criminal record” (I.e., no difference)
- E.g., \(H_A\): “Applicants with a criminal record will be less likely to receive call backs than those without a criminal record” or “Applicants with a criminal record will receive call backs at different rates than those without a criminal record” (I.e., some nonzero difference)
Example of Implied Framework
Health Savings Experiment (Dupas and Robinson 2013): Researchers conducted a field experiment in rural Kenya in which they randomly varied access to four innovative saving technologies and observed the impact on asset accumulation. Participants who were given a box locked with a padlock and key saved about 150 Kenyan Shillings more after 12 months relative to those who were simply encouraged to save money for health. Did the lock box work?
- What are the implied null and alternative hypotheses?
Process for Hypothesis Testing
- Carry out a test of the hypothesis, such as a difference-in-means.
- Applicants with a criminal record receive 12.5 percentage points fewer call backs than those without a criminal record
- Calculate the uncertainty around this estimate.
- Decide whether you can reject or fail to reject the hypothesis of no difference