Context

The description of statistical methodology in scientific articles, particularly in preclinical research, is almost never exhaustive. Too often, it is vague, incomplete, or plainly negligent. This is not a minor oversight. It fuels confusion, weakens reproducibility, and normalizes sloppy statistical habits that the field can no longer afford. So when you come across the phrase “a t-test, a non-parametric test, an ANOVA, a regression model was used,” it should not surprise you. It is a perfect example of how low our standards for methodological transparency have sunk and why that must change.

An ANOVA was Used

The phrase “an ANOVA was used” is problematic because it is vague, incomplete, and uninformative. It tells the reader almost nothing about what was actually done. Unfortunately, even in 2025, this phrase is still prevalent.

⚠️“ANOVA” is NOT a Single Test

Several ANOVA Models Exist

  • One-way, two-way, or multi-factor ANOVA
  • Randomized block design ANOVA
  • Repeated-measures ANOVA
  • Mixed-design ANOVA
  • Nested ANOVA
  • Split-Plot ANOVA
  • Analysis of Covariance (ANCOVA)
  • Multivariate ANOVA (MANOVA)
  • Etc.

Each addresses different experimental designs and assumptions. Without specifying which was used, the statement is ambiguous.

⚠️ Such a Description Omits Key Details about Design & Assumptions

ANOVA possesses underlying assumptions that must be verified in order to make correct statistical inference.

  • Normal distribution of residuals
  • Homogeneity of variances
  • Independent observations

If these assumptions were violated, the results may be invalid — yet the phrase “an ANOVA was used” does not tell the reader whether these checks were done or what alternatives were considered.

⚠️ Such a Description Hides the Structure of the Analysis

Additional information is required.

  • What were the studies factors and levels?
  • How are factors related: crossed, nested.
  • Was there a blocking variable or random effect?
  • Was it a fixed-effect or mixed-effect model?
  • Single or multiple error terms?
  • Were post-hoc comparisons performed, and if so, how were they corrected for multiple testing?
  • And so on.

Without that, replication and interpretation are impossible.

⚠️ Such a Description Lacks Clarity about the Dependent Variable(s)

ANOVA can be applied to a wide range of outcomes — continuous, transformed, or even aggregated data. Simply saying “ANOVA” doesn’t specify which variable was analyzed, nor how data were preprocessed.

⚠️ Such a Description Fails Transparency & Reproducibility Standards

Modern reporting guidelines (e.g., ARRIVE, SAMPL, APA, CONSORT) require:

  • The type of ANOVA used
  • The model structure
  • The assumptions tested
  • The post-hoc procedures
  • The software and version

Without these, the description doesn’t meet minimal reporting standards.


A Better Way to Report it

Clearly Describe the Analysis

Before reporting results, specify:

  • Design type: one-way, two-way, repeated measures, mixed model, etc.
  • Dependent variable(s) and independent factors (with number of levels).
  • Any transformations, covariates, or adjustments (e.g., log-transformed outcome, controlling for body weight).

Example: “A two-way ANOVA was conducted to assess the effects of treatment (3 levels) and sex (2 levels) on tumor volume.”

Report Test Statistics Completely

ElementDescriptionExample
F-valueThe ANOVA test statisticF(2, 27) = 6.42
Degrees of freedomFor factor and residualF(df_effect, df_error)
p-valueExact p-value if possiblep = 0.005
Effect sizeEstimated mean value (contrast, difference, etc.)mean difference 2.1 & 95% CI [0.8, 3.3],
Assumption checksNote violations or correctionsVisual exploration of the data and Levene’s test indicated equal variances

Example:
“The effect of treatment was significant, F(2, 27) = 6.42, p = 0.005, η² = 0.32.”

Check & Mention Underlying ANOVA Assumptions

At minimum, confirm whether key assumptions were tested:

  • Normality of residuals (e.g., Shapiro–Wilk test or Q–Q plot or other approach) justify your choice
  • Homogeneity of variances (e.g., Levene’s or Bartlett’s test or other approach)
  • Sphericity (for repeated measures; note Greenhouse–Geisser correction if violated, provide details about covariance structure used and why)

Example:
“Assumptions of normality and homogeneity were met. For repeated measures, Greenhouse–Geisser correction was applied when sphericity was violated.”

Post Hoc or Simple Effects Tests (if Applicable)

If the ANOVA is significant:

  • Specify the post hoc test used (Tukey, Bonferroni, Holm, etc.).
  • Report pairwise mean differences, confidence intervals, and adjusted p-values.

Example:
“Post hoc Tukey tests revealed that Treatment A significantly increased response compared to Control (mean difference = 2.1, 95% CI [0.8, 3.3], p = 0.004).”

Provide Context for Interpretation

  • Summarize practical meaning or direction of effects.
  • Avoid over interpreting p-values.
  • If nonsignificant, provide confidence intervals and discuss precision.

Recommended

  • Include a visual summary (e.g., group mean ± 95% CI plots).
  • Share exact model code (R, SAS, etc.) or ANOVA table in supplementary material.
  • Report data transformations explicitly if applied.

Example of full Reporting Paragraph

“A one-way ANOVA was conducted to examine the effect of three treatments on tumor volume.

Independent biological replicates were used to assess the effect of treatments on tumor volume (mice). 10 independent mice were randomized to the treatment groups.

There was a significant effect of treatment, F(2, 27) = 6.42, p = 0.005.

Post hoc Tukey comparisons indicated that Treatment B resulted in significantly higher tumor volumes than Control (mean difference = 4.5 mm³, 95% CI [1.3, 7.6], p = 0.003), whereas Treatment A did not differ significantly from Control (mean difference = 2.2 mm³, 95% CI [−0.5, 4.9], p = 0.11).

The group means (95% CI) were 10.4 mm³ [8.7, 12.1] for Control, 12.6 mm³ [10.9, 14.3] for Treatment A, and 14.9 mm³ [13.2, 16.6] for Treatment B.

Assumptions of normality was assessed using a Q-Q plot and homogeneity of variance using plots and Levene’s test. Both assumptions were met.”