| |

**SUMMARY: The Most Important
Points**

- Think about differences between group means in terms of standard deviations, not standard errors of the mean. Mean ± SD or Mean ± SEM?
- Learn exactly what a trivial, small, moderate, large, very large, and almost perfect effect is, for a correlation, a frequency difference, and the effect-size statistic. Magnitudes for Effect Statistics
- Present as few numbers as possible: no more than two significant digits for effect statistics and standard deviations. How Many Digits?
- Understand how validity impacts on cross-sectional studies and reliability impacts on longitudinal studies. Cross·Sectional Designs, Longitudinal Designs
- Test enough subjects to allow you to publish any result. Sample·Size Estimation, Based on Confidence Limits
- Try sample size "on the fly" in your next project, but base it on width of the confidence interval, not statistical significance. On The Fly
- Explore a stats program by making up data with an effect, then analyzing them. Simulation
- Before you do any modeling (statistical tests), look at your data to see what's going on. Effect Statistics
- Keep your eye on standard deviations or scatter of points, to decide whether log or rank transformation is needed before you fit a model. Residuals: Bad, Log Transformation, Rank Transformation
- If you have repeated measures with missing data, get a statistician to help you model covariances. Modeling Covariances
- Know the difference between statistically significant and substantial. Confidence Limits, Magnitudes for Effect Statistics
- Show confidence intervals instead or, or as well as, p values. P Values
- Stop asking "is there an effect?" Start asking "how big is the effect?" Hypothesis Testing
- Stop thinking about testing. Start thinking about estimating. Hypothesis Testing

Go to: Next · Previous · Contents · Search · Home

Last updated 25 April 97