The following table indicates suitable statistical methods for a certain aim of an analysis (see first column below) given the type of available data (categorical or numerical).

Aim | Data | ||
---|---|---|---|

Categorical, binomial (two possible outcomes) | Numerical (from normal distribution, or non-normal as long as n > 20*) | Numerical (from non-normal distribution with outliers and n < 20*) | |

Describe/summarize a group | Proportions, contingency tables | Mean, standard deviation | Median, interquartile range |

Compare one group to a hypothesized value** | One-proportion, chi-square or Binomial test | One-sample t-test | Wilcoxon test |

Compare two unpaired groups | Two-proportion or Fisher's test (chi-square for large samples) | Unpaired (two-sample) t-test | Mann-Whitney test |

Compare two paired (matched) groups | McNemar's test | Paired t-test | Wilcoxon test |

Compare three or more unmatched groups | Chi-square test | One-way ANOVA | Kruskal-Wallis test |

Compare three or more matched groups | Cochrane Q | Repeated-measures ANOVA | Friedman test |

Quantify association between two variables | Contingency coefficients | Pearson correlation | Spearman correlation |

Predict value from another measured variable | Simple logistic regression | Regression (linear or nonlinear) | Nonparametric regression |

Predict value from several measured or binomial variables | Multiple logistic regression | Multiple regression (linear or nonlinear) |

* To check whether data (can be assumed to) come from a normal distribution, do a normality test (e.g., Anderson-Darling). If your data strongly deviate from normality, consider transforming them, e.g., taking the (natural) logarithm, the square root, or inverse; transformed data may be more normal.

** A hypothesized value can be a standard/reference value from previous literature, a legal limit, a recommended value, etc.