Statistical Process Control with Tidyverse-Native Workflows.

shewhartr

shewhartr is a tidyverse-native toolkit for Statistical Process Control (SPC). It implements the classical Shewhart chart family — variables (I-MR, Xbar-R, Xbar-S) and attributes (p, np, c, u) — alongside a flagship regression-based control chart for processes with trend, where stationarity is too strong an assumption to make.

The package is built around a small set of design choices:

• Tidy by default. Every constructor takes data first, supports tidy-eval column references, and returns an S3 object that integrates with broom via tidy(), glance() and augment().
• Diagnostics in the box. Average Run Length (ARL) by Monte Carlo simulation, all eight Nelson runs rules, Box-Cox guidance, and a Tukey-style residual panel are first-class citizens.
• Phase I / Phase II. Explicit calibrate() and monitor() functions make the distinction between estimation and prospective monitoring impossible to confuse, following Woodall (2000).
• Multilingual plots. Pass locale = "pt" (or "es", "fr") and chart titles, axis labels, and legends are translated.
• Statistically honest counts. c and u charts accept limits = "poisson" for exact Poisson quantile limits, instead of the normal approximation that breaks down at small means.

Installation

# Development version
remotes::install_github("castlaboratory/shewhartr")

A 30-second tour

library(shewhartr)
library(ggplot2)

# Classical I-MR chart on a 100-observation series with a small drift
fit <- shewhart_i_mr(bottle_fill, value = ml, index = observation)

print(fit)
autoplot(fit)

When to use which chart

Multi-phase regression chart

The flagship chart for trended processes splits the series into phases when a runs rule fires, fits a local model in each, and flags points that depart from the local trend. The example below uses the COVID-19 mortality series for Recife (cvd_recife) with the original analysis settings from Ferraz et al. (2020):

fit <- shewhart_regression(
  cvd_recife,
  value       = new_deaths,
  index       = .t,
  model       = "loglog",
  phase_rule  = "we_seven_same",
  rules       = c("nelson_1_beyond_3s", "we_seven_same"),
  lower_bound = 0
)

length(fit$fits)              # number of phases detected
nrow(fit$violations)          # individual flagged observations
autoplot(fit)

Each shaded band is one phase, the solid line is the local regression centre, the dashed lines are the phase’s 3-sigma limits, and the firebrick points are the days flagged by the rule set as departing from the local trend.

Phase I vs Phase II

A Shewhart chart serves two different purposes that are easy to conflate. Phase I is retrospective: take historical data, identify out-of-control points, eliminate assignable causes, and arrive at trustworthy estimates of the process mean and variability. Phase II is prospective: take those estimated limits and apply them to new data, signalling alarms when something departs from the established baseline. The package draws this line in code:

# Phase I: estimate limits from a clean baseline
calib <- calibrate(historical_data, value = y,
                   chart = "i_mr", trim_outliers = TRUE)

# Phase II: apply the limits to new data
alarms <- monitor(new_observations, calib)
alarms$violations

Architecture

Every chart constructor — variables (shewhart_i_mr, shewhart_xbar_r, shewhart_xbar_s), attributes (shewhart_p, shewhart_np, shewhart_c, shewhart_u), regression (shewhart_regression), memory-based (shewhart_ewma, shewhart_cusum), and multivariate (shewhart_hotelling) — returns a shewhart_chart S3 object with a uniform layout. The same object then feeds into:

• Inspect: print(), summary(), shewhart_diagnostics(), shewhart_capability(), shewhart_arl().
• Plot: autoplot() (ggplot2) and as_plotly() (interactive, plotly in Suggests).
• Tidy: broom::tidy(), glance(), augment().
• Phase II: calibrate() produces a Phase I chart whose limits are frozen by monitor() for prospective monitoring.

Documentation

The website hosts:

• A full reference for every exported function.
• Topical articles: variables charts, attributes charts, regression charts, Phase I/II, ARL by simulation, residual diagnostics, Box-Cox guidance.
• A case study on epidemiological monitoring with the regression chart.

Citation

If you use shewhartr in academic work, please cite:

Leite, A., Vasconcelos, H., Ospina, R., & Ferraz, C. (2025). shewhartr: Statistical Process Control with Tidyverse-Native Workflows. R package version 1.0.0. https://castlaboratory.github.io/shewhartr/

References

• Shewhart, W. A. (1931). Economic Control of Quality of Manufactured Product. D. Van Nostrand.
• Montgomery, D. C. (2019). Introduction to Statistical Quality Control (8th ed.). Wiley.
• Nelson, L. S. (1984). The Shewhart Control Chart — Tests for Special Causes. Journal of Quality Technology, 16(4), 237-239.
• Woodall, W. H. (2000). Controversies and Contradictions in Statistical Process Control. Journal of Quality Technology, 32(4), 341-350.
• Box, G. E. P., & Cox, D. R. (1964). An Analysis of Transformations. Journal of the Royal Statistical Society B, 26(2), 211-252.
• Tukey, J. W. (1977). Exploratory Data Analysis. Addison-Wesley.
• Wheeler, D. J., & Chambers, D. S. (1992). Understanding Statistical Process Control (2nd ed.). SPC Press.
• Perla, R. J., Provost, S. M., Parry, G. J., Little, K., & Provost, L. (2020). Understanding variation in reported COVID-19 deaths with a novel Shewhart chart application. International Journal for Quality in Health Care, 32(S1), 49-55.
• Ferraz, C., Petenate, A. J., Wanderley, A. L., Ospina, R., Torres, J., & Moreira, A. P. (2020). COVID-19: Monitoramento por gráficos de Shewhart. Revista Brasileira de Estatística.