11.2 Simple Filters

A reasonable first approach to data is to figure out which variables have the highest predictive power on your dataset.

Outcome Categorical

  • Predictor categorical
    • 2 levels odds-ratio
    • 3+ levels Contingency table with χ2 test
  • Predictor Continuous
    • 2 levels categorical t-test
    • 3+ levels categorical ANOVA F-stat

Outcome Continuous

  • Predictor categorical
    • 2 levels categorical t-test
    • 3+ levels categorical ANOVA F-stat
  • Predictor Continuous
    • Pairwise or Rank correlation (linear)
    • non-linear
      • MIC or A statistic
      • GAM