Epidemiology/Stats Flashcards
For which study designs does one use risk ratio/rate ratio (relative risk)?
Calculated for study designs that collect data on incidence: cohort and RCT
For which study design does one use risk odds ratio?
Calculated for study designs that use prevalent cases: cross-sectional and case-control studies
What is type I (alpha) error?
True null hypothesis incorrectly rejected (false positive)
What is type II (beta) error?
Failing to reject false null hypothesis
Power equation
1 – beta
ROC axes?
Sensitivity on Y axis and 1-specificity on X axis
ROC AUC?
AUC = area under curve (accuracy of test)
Left shift of ROC has what result?
Increased accuracy
Parametric vs non-parametric tests in large samples
The nonparametric tests nearly as powerful as the parametric test from Gaussian populations
Parametric vs non-parametric tests in small Gaussian samples
Non-parametric tests have much less power
Figure out what this means…
Usually report p values and not CIs because this requires additional assumptions
Cannot be readily extended to regression models
Consider transforming data to create a Gaussian distribution rather than use nonparametric and the decision of whether to use parametric vs nonparametric most important with small data sets
Parametric and non-parametric tests for 2 continuous variables
T test and Mann-whitney U (Wilcoxon Rank Sum) respectively
Parametric and non-parametric tests for 2 paired continuous variables
Paired t-test and Wilcoxon Signed Rank respectively
Parametric and non-parametric tests for 3 continuous variables
ANOVA and Kruskal-Wallis respectively
Parametric and non-parametric tests for 3 continuous variables repeat measurements
Repeated measures ANOVA and Friedman respectively