crossStatR

crossStatR is an R package of statistics for experimental science. It supports biological assays, electrochemical measurements, optical and spectroscopic signals, and physics experiments. The package combines established methods (descriptive statistics, t-tests, regression, ANOVA, power analysis) with experimental diagnostics (region-level signal-difference testing, peak-evidence assessment, calibration-transfer risk, kinetic discrimination, experimental fragility, drift-attributed uncertainty, analytical detection-reliability curves, and stage-wise uncertainty-origin attribution).

This is a development version (0.99.2). The API may change before 1.0.0.

Local simulation diagnostics found poor peak-test calibration and under-coverage of transfer and drift intervals in the tested scenarios. These methods remain exploratory. Results and reproducible scripts are documented in the validation report.

Installation

From R-universe:

install.packages(
  "crossStatR",
  repos = c(
    rotsl = "https://rotsl.r-universe.dev",
    CRAN = "https://cloud.r-project.org"
  )
)

Alternatively, from GitHub using remotes:

install.packages("remotes")
remotes::install_github("rotsl/crossStatR")

Then verify it loads:

library(crossStatR)
packageVersion("crossStatR")
#> [1] '0.99.2'

Function catalogue

The package exports 32 user-facing functions, grouped into three categories.

Category Function What it does
A: Core statistics sci_describe Descriptive statistics, with explicit biological vs. technical replicate handling
sci_compare Two-group comparisons (Welch, Student, paired, permutation, Wilcoxon) with effect sizes
sci_anova ANOVA, including one-way, two-way, and mixed-effects models
sci_regression Linear, weighted, and robust regression with diagnostic output
sci_correlation Correlation analysis with CIs (Pearson, Spearman, Kendall)
sci_normality Normality assessment with multiple tests and graphical diagnostics
sci_outliers Outlier diagnostics (IQR, MAD, Grubbs, Dixon)
sci_power Power and sample-size calculation for common designs
sci_effect_size Standardized effect sizes (Cohen's d, Hedges' g) with CIs
sci_bootstrap Bootstrap resampling with percentile, BCa, and normal CIs
B: Domain methods adaptive_baseline Asymmetric least squares or lowess baseline estimation for spectra
adaptive_smoothing Edge-preserving smoothing for noisy signals
batch_effect_quantify Quantify batch-to-batch variance, with mixed-effects decomposition
calibration_compare Compare two calibration curves via ANCOVA
curve_similarity Similarity between two curves sampled on possibly different grids
drift_correct_stat Drift correction with uncertainty quantification
kinetic_model_compare Compare candidate kinetic models via likelihood-ratio tests
noise_characterize Summarise residual variation and autocorrelation
parameter_uncertainty Propagate parameter uncertainty to derived quantities
replicate_decompose Variance-component decomposition for nested and crossed replicates
robust_lod Limits of detection and quantification (Currie-style)
spectral_change_test Test for differences across complete spectra
timepoint_optimize Optimise sampling timepoints for a planned experiment
weighted_peak_fit Weighted multi-peak fitting (Gaussian/Lorentzian)
C: Advanced diagnostics signal_difference_regions Identify contiguous regions where two groups of curves differ
peak_evidence Frequentist evidence (LR + permutation FPP) that a candidate feature is a true peak
calibration_transfer_risk Risk of transferring a calibration between instruments without recalibration
kinetic_discrimination Probability that a planned experiment can discriminate between kinetic models
experiment_fragility Sensitivity of a result to defensible analysis decisions
drift_uncertainty Decompose measurement variance into drift and noise components
detection_reliability Concentration-dependent PD/FPR/FNR curves with bootstrap bands
uncertainty_origin Stage-wise attribution of measurement uncertainty (Shapley)

Every function returns an S3 object of class crossstatr_<name> with print, summary, tidy, and as.data.frame methods.

Quick start

The two common starting functions are sci_describe (for summarizing a sample) and sci_compare (for comparing two groups).

library(crossStatR)
set.seed(1)

# Biological assay: control vs. treatment, 8 biological replicates per group
ctrl <- rnorm(8, mean = 10, sd = 1.2)
trt  <- rnorm(8, mean = 11.5, sd = 1.2)

sci_describe(ctrl)
#> <crossStatR::sci_describe>
#> Groups: overall  (n = 8)
#>     mean     sd     se    median       ci_low    ci_high   cv
#>   10.21    1.18   0.42    10.30      9.23       11.19     0.12

sci_compare(ctrl, trt, test = "welch")
#> <crossStatR::sci_compare> Welch two-sample t-test
#>   t = -2.93, df = 13.4, p = 0.011
#>   mean difference = -1.42 [95% CI: -2.49, -0.36]
#>   Cohen's d = -1.04 [95% CI: -1.95, -0.10]

Examples by domain

Biological: replicate-aware comparison

set.seed(1)
df <- data.frame(
  y     = c(rnorm(12, 1.0), rnorm(12, 1.4)),
  grp   = factor(rep(c("ctrl", "trt"), each = 12)),
  bio   = factor(rep(1:12, times = 2)),
  tech  = factor(rep(1:3, times = 8))
)
# Summarise at the biological-replicate level first; do not treat
# technical replicates as independent observations.
d <- sci_describe(df$y, group = df$grp, bio_rep = df$bio, tech_rep = df$tech)
print(d)
c <- sci_compare(y ~ grp, data = df, test = "welch")
print(c)

Electrochemical: drift-attributed variance

set.seed(2)
t  <- seq(0, 10, length.out = 60)
y  <- 2 + 1.5 * t + 0.3 * (t - mean(t)) + rnorm(length(t), sd = 0.2)
dr <- drift_uncertainty(y, time = t, drift_model = "linear")
print(dr)

Optical/spectroscopic: region-level differences

set.seed(3)
wl <- seq(400, 800, length.out = 80)
g1 <- matrix(replicate(15, sin((wl - 400) * pi / 200) * 0.2 +
                          rnorm(80, sd = 0.15)), nrow = 15, byrow = TRUE)
g2 <- matrix(replicate(15, {
  base <- sin((wl - 400) * pi / 200) * 0.2 + dnorm(wl, 530, 18) * 0.8
  base + rnorm(80, sd = 0.15)
}), nrow = 15, byrow = TRUE)
sdr <- signal_difference_regions(wl, g1, g2, n_perm = 200)
print(sdr)

Physics: uncertainty origin attribution

# Four uncertainty sources, two correlated. Stage variances have
# already been estimated upstream (e.g. via replicate_decompose()
# and drift_uncertainty()).
stages <- c(sampling = 0.05, instrument = 0.10,
            calibration = 0.02, drift = 0.03)
uo <- uncertainty_origin(
  stages,
  correlations = list(sampling__instrument = 0.3,
                      calibration__drift = 0.2)
)
print(uo)

Reproducibility

Randomness enters the package in three places: permutation tests, the bootstrap, and Monte Carlo simulation. Set the seed explicitly whenever the analyst wants a reproducible number.

set.seed(42)
res <- sci_compare(x, y, test = "perm", seed = 42)

Seed handling varies by function. Calls that use a function-level seed may change the global random-number state; reproducible analyses should set and record seeds explicitly.

Beyond the seed, recording the analysis decisions that produced the reported result is at least as important. The documentation recommends running experiment_fragility() on the final result to see how sensitive the conclusion is to defensible choices: which replicates were included, how outliers were handled, which model was fit, and how much measurement noise would have to be added before the decision flips. If a single decision flips the sign of the effect, the claim is fragile and should be reported as such.

af <- function(opts) {
  d <- opts$data
  if (!is.null(opts$rep_lo)) d <- d[d$rep != opts$rep_lo, ]
  diff(tapply(d$y, d$grp, mean))
}
fr <- experiment_fragility(
  analysis_fn = af, data = df, replicates = "bio",
  decision_threshold = 0, n_boot = 200, seed = 1
)
print(fr)

Citation

The development version can be cited as follows:

Rohan R (2026). crossStatR: Scientific Statistics for Biological, Electrochemical, Optical and Physics Experiments. R package version 0.99.2. https://github.com/rotsl/crossStatR

In R:

citation("crossStatR")

License

MIT. See LICENSE.md and LICENSE. Copyright (c) 2026 Rohan R.

Author

Rohan R.

Issues, bug reports, and pull requests are welcome at https://github.com/rotsl/crossStatR.

The test data is taken from Soedarmadji and Gregoire (2021), CaltechDATA: https://doi.org/10.25989/198p-1x22. scripts/validate_caltech.R reproduces the local checks and summaries.

Method descriptions and source references are available in the methods guide.