The absolute value of the difference between the replicated sample size calculation and the authors' target sample size was greater than 10% for 47 articles (30%) and greater than 50% for 10 (6%).
While there were frequently significant differences between replicates, the replicates typically followed the same trend and only interacted with the treatment group when indicated (Figs. 2b and 5b).
Systematic variation of only two factors was sufficient to mimic the range of differences between the replicate experiments, resulting in almost perfect reproducibility [34].
For each variable, the cluster effect was expressed as the mean of the relative differences between the replicates of the strains within a given cluster and the control without bacteria.
(5) ϵ t = 〈 | x t − μ t | 〉 = ∫ | x t − μ t | p (x ) d N x (6) p (x | { μ t, ϵ t } ) = ∏ t = 1 m 1 2 ϵ t exp (− | x ˜ t − μ t | ϵ t ) Equation (5) defines the normalizing constant ϵ t as the expected value of the moduli of the differences between the replicate observations at time t and the values predicted by the kinetic model (μ t).
Hierarchical clustering (HCL) analyses were initially performed to obtain an overview of the differences between the technical replicates, resampled biological replicates (sub-biopsies/biopsies), and biological replicates (patients).
This clustering was performed to obtain an overview of the differences between the technical replicates (that is, replicates of the same amplified RNA aliquot with reversed fluorophores, denoted A and B), the resampled biopsies (denoted 1 or 2) and the biological replicates in the form of different biopsy types (denoted close to cartilage or not close to cartilage) from the patients.
No qualitative difference between the replicates of each extract was found so the four spectra (two replicate FT-IR measurements of two extractions per solvent) were averaged.
The ANOVA F-values of SHAP tasks and the FP and IOF functionality scores do not exceed Fcritical = 3.28, indicating there is no difference between the replicates and demonstrating, thus, good reliability [ 1, 28].
The possible differences between the three replicate experiments were modeled as a random block effect.
