Data Intelligence

Methods Accagrity actually calculates

This is the current deterministic catalogue. Choose a method because it suits your research question and design—not because a menu makes it look available. The platform validates inputs and records diagnostics, but it does not turn association into causation or replace domain expertise.

Describe, compare, and associate

Descriptive and grouped descriptive statistics, frequency tables, mean confidence intervals, and covariance matrices.
Pearson, Spearman, and Kendall tau-b correlations; Pearson, Spearman, and Kendall matrices include their declared confidence-interval approximation where defined.
One-sample, independent, and paired t-tests; one-way ANOVA; Mann–Whitney, Wilcoxon signed-rank, Kruskal–Wallis, and governed pairwise group comparisons.
Chi-square independence and Fisher exact tests.
One-way repeated-measures ANOVA for a balanced complete subject-by-condition design, and one-way MANOVA using Pillai trace with a deterministic permutation p-value.

Regression, counts, survival, and time

Simple and multiple linear regression.
Inferential binary logistic regression with numeric or categorical predictors, model or HC0 robust standard errors, odds ratios and intervals, and a fail-closed Firth bias-reduced fallback for unsafe maximum-likelihood fits.
Poisson and negative-binomial count regression for non-negative integer outcomes. Poisson output warns when its dispersion diagnostic suggests a negative-binomial model may be more appropriate.
A Gaussian random-intercept linear mixed model with numeric fixed predictors and one clustering variable. It does not claim random slopes, crossed effects, or generalized mixed models.
Kaplan–Meier survival curves with Greenwood log-log confidence intervals, plus a two-group log-rank test when a comparison group is supplied.
Linear time trend and simple-exponential-smoothing forecasts. These are not ARIMA, seasonal, causal, or intervention models.

Study planning

Required sample size, achieved power, and power curves for one- and two-sample means, paired means, correlation, one-way ANOVA, and two proportions.
Declared-assumption planning extensions for clustered and two-level multilevel means, complex-survey means, repeated measures, multivariable logistic event adequacy, and log-rank survival. Each extension states when it is an approximation or adequacy rule rather than a full protocol simulation.
Attrition, allocation ratio, intraclass correlation, design effect, event proportion, and measurement assumptions are recorded with the result.

Machine learning and visual analysis

Linear and regularised regression, binary logistic classification, decision trees, k-means clustering, and principal-component analysis (PCA).
Visual outputs include distributions, grouped and stacked bars, heatmaps, scatterplots, error bars, ROC and precision–recall curves, cluster summaries, PCA scatter only when two components exist, and Kaplan–Meier curves.

Important boundaries

Accagrity does not currently provide Cox regression, competing-risks survival analysis, exact logistic regression, random-slope or generalized mixed models, ARIMA/seasonal time series, causal inference, or automatic method selection. If your design needs one of these, consult a qualified statistician and use a validated specialist workflow.