- Statistics Without Statistics: Did They Use the Right Test? (continued)
Regression (continued)
For regression you need continuous variables. The dependent variable (y) is the variable to "explain".
The slope (m) describes the change when testing (y) against an independent variable (x). The y-intercept
(b) is also used to determine the "baseline" value of (y) when no influence of the independent
variable exists. (i.e. the value of y when x=0.) Several independent variables can be used in the model
(y = m1x1 + m2x2 +m3x3…). Authors can generate an equation to represent the value of a continuous
variable. High blood pressure (y) may be estimated by cholesterol (x1), weight (x2) and age (x3). Although
complicated at times pay attention to the final result. Many times authors will put a spin on the original
hypothesis stating they found all these other great results, but never answer the question.
- Ten Ways to Cheat on Statistical Tests When Writing Up Results
A study which shows a positive result is about 10 times as likely to be published as one which shows a
negative result. Authors and researchers therefore have an incentive to find "good news" in
their work. This is not necessarily nefarious. However, it is our job to watch out for these strategies or
accidental omissions. Here are 10 strategies that Trisha Greenhalgh suggests we be on the lookout for.
- The authors threw all their data into a computer and looked for enough results so that something was positive by accident.
- Baseline differences between the groups were ignored if they favored the authors' bias.
- Data, which were not normally distributed, were tested with as normal data.
- Patients who dropped out were ignored.
