I met with Kristina, the statistician, this week to clear up some of my questions. Her verdict: I didn't do everything wrong. So that's always good news. On the other hand, I still have my work cut out for me with some of these individual symptom graphs. See, I can't just take the difference between the correlation coefficients and call it a day. I also need to decide whether the difference between study and control is statistically significant. Kristina recommended a Generalized Linear Model (GLM), which is used to analyze linear regressions like mine. Apparently, this model is commonly utilized by medical researchers who are studying extremely rare diseases, and thus are restricted to a very small population of subjects. This method will account for error resulting from my ridiculously tiny sample size, as well as the fact that there are multiple variables involved (symptom severity vs. number of visits, study vs. control, etc.). Typically, this model is created using an actual statistics program such as SPSS. I have Excel. I'll figure it out.
Obviously, I want to analyze my data in the most accurate way possible. Regardless of my attempts to combat sample size error, the fact still remains: this study is underpowered. We knew this ahead of time. However, we're not looking for anything revolutionary. In reality, the only novel aspect of this study is the fact that no other prior research has been done on this specific population (domestic violence victims with TBI who undergo speech therapy). Our goal is to change that. We want to encourage other doctors to conduct further research on this population despite all of the other variables that make this research difficult. Most of all, we want to prove that the treatments provided here at BNI are even working. Hopefully, that's what this GLM will demonstrate.
No comments:
Post a Comment