Choose Your Own Data-Analytic Adventure

bob.8
So what should I write in the IRB proposal'

First, since you randomized the intervention and control groups, they are equal (in expectation). Therefore, if you observe a statistically significant difference in anxiety, you can reasonably infer that the difference was caused by your child life intervention.

If you are going to polychotomize MYPAS scores (into, say, "not," "somewhat," "very," and "extremely anxious"), you need reasonable cut scores, preferably specified now.

There's nothing in the literature that suggest multiple cuts. What's reasonable'

Perhaps use quartiles, thus breaking the distribution of anxiety into 4 equal-sized categories' Think about it. You may decide you want 3 or 5 categories. Maybe when we look at the actual data, natural cuts may suggest themselves. For example, if the distribution is trimodal or something like that.


Conduct a 2x(') contingency table analysis to explore the relationship between the intervention and anxiety (however you decide to polychotomize it).


Using the '2 statistic, test the null hypothesis that there is no relationship in the population between the intervention and anxiety. If you reject the null, conclude that there is a relationship in the population, that intervention patients are less likely to be anxious (using residuals to dig deeper), and that the intervention caused this effect.


Use differences in percentages (from your sample) to convey the effect size.

What should I do about the section for statistical power and sample size'

Before we can answer that, you really have to hammer out how many categories your outcome has. Then we can can do the delicate work of thinking about the effect size. To determine statistical power, we need to input effect size, sample size and alpha level. To determine sample size, we need to input effect size, statistical power and alpha level.

You decide to:
Submit this data analytic strategy to the Institutional Review Board.
or
Reconsider how you will characterize the outcome.