TABLE 1:
Matching types of data appropriately with commonly used statistical tests (Crawley, 2013; Whitlock and Schluter, 2014).
| Response type | Treatment type | Statistical tests | Typical null hypothesis | Assumptions | Functions in Excel and R | Examples |
|---|---|---|---|---|---|---|
| Continuous numerical (e.g., reaction rate) | Binary (e.g., wild type vs. mutant) | t test | Means equal for two treatments | Randomized treatment; normally distributed response; equal variance of response for two treatment groups | Excel: TTEST(controlrate, genotype, tails = 2, type = 2) R: t.test(rate∼genotype, var.equal = TRUE) | Zhuravlev et al., 2017; Figure 1C |
| Continuous numerical (e.g., reaction rate) | Categorial (e.g. wild type vs mutant 1 vs mutant 2) | ANOVA followed by Tukey-Kramer post-hoc test | Means equal across treatments | Randomized treatment; normally distributed response; equal variance of response between treatment groups | Excel: Analysis ToolPak add-in for single factor ANOVA required if treatment has more than two levels R: aov(rate∼genotype) TukeyHSD() | Plooster et al., 2017; Figure 1D |
| Continuous numerical (e.g., reaction rate) | Continuous numerical (e.g., drug concentrations) | Linear regression followed by t test on regression coefficients or correlation test | Coefficients equal to zero; correlation coefficient equal to zero | Randomized treatment; linear relationship between treatment and response; treatment and response bivariate normally distributed; normally distributed residuals | Excel: LINEST(rate, drug, const = TRUE, stats = TRUE) R: lm(rate∼drug) summary() | Spencer et al., 2017; Figure 4 |
| Continuous numerical (e.g., reaction rate) | More than one treatment: Continuous numerical (e.g., drug concentrations) plus categorical (e.g., wild type vs. mutant 1 vs. mutant 2) | Analysis of covariance (ANCOVA) | Coefficients equal to zero | Randomized treatment; numerical treatment and response bivariate normally distributed; equal variance of response between treatment groups; linear relationship between numerical treatment and response; normally distributed residuals | Excel: LINEST(rate, treatments, const = TRUE, stats = TRUE) R: lm(rate∼drug + genotype) summary() | Nowotarski et al., 2014; Figures 3 and 4 |
| Categorical (e.g., cell cycle stage) | Categorical (e.g., wild type vs. mutant 1 genotype vs. mutant 2 genotype) | Chi-square or G contingency test; binomial test; Fisher’s exact test | Proportions between response categories are equal between treatments | Randomized treatment; all expected counts are one or more and no more than 20% of expected counts less than five; binomial: response or treatment sample sizes fixed; Fisher’s: response and treatment sample sizes fixed | Excel: CHISQ.TEST(actual_range,expected_range) R: chisq.test(stage, genotype) | Bartolini et al., 2016; Figures 2F and 5B |
| Binary categorical (e.g., alive or dead) | Continuous numerical (e.g., drug concentrations) | Logistic regression; generalized linear model | Slope or intercept equal to zero | Randomized treatment; linear relationship between treatment and log odds of one response | Excel: Not available R: glm(alive∼drug, family = binomial) summary() | Atay and Skotheim, 2014; Figure 2 |
| Binary categorical (e.g., alive or dead) | More than one treatment: Categorical (e.g., wild type vs. mutant 1 vs. mutant 2) plus categorical (e.g., day 1 vs. day 2 vs. day 3) | Logistic regression; generalized linear model | Slope or intercept equal to zero | Randomized treatment | Excel: Not available R: glm(alive∼genotype + day, family = binomial) summary() | Kumfer et al., 2010; Table 1 |