Step 7: Identify which attributes need special attention in the design, and determine how they should be attended to in the sampling process to avoid bias.

The superintendent anticipates differences in outcomes due to experience but is not interested in obtaining an estimate of the interaction of experience and treatment effectiveness. Rather she wants an estimate of treatment effectiveness that can be extrapolated to the universe of teachers in the district without the necessity of weighting experimental results to adjust for teaching experience differences between the teachers participating in the experiment and all teachers in the district. Because you suspect that novice teachers are more likely than experienced teachers to volunteer for your study, you are concerned about drawing a biased sample. Consequently, you decide to stratify your sample of teachers on level of experience. You draw two random samples of teachers, one novice, one experienced, according to their proportions in the population.