The goal of Shewhart is to provide a streamlined workflow for detecting and modeling process phase changes using Shewhart control charts. It features tools for fitting common transformations (log, log-log, Gompertz), identifying phases via the “7 consecutive points” rule, and computing control limits on both transformed and original scales. The package leverages the tidyverse for intuitive data manipulation and is designed to be easily integrated into broader data analysis pipelines.
You can install the development version of Shewhart from GitHub with:
# install.packages("pak")
pak::pak("castlaboratory/Shewhart")This package provides functions for creating and analyzing Shewhart control charts (using either ggplot2 or plotly) and detecting potential phase changes via the “7 consecutive points” rule. Key functions include:
shewhart(): Creates the control chart plot (either ggplot2 or plotly).shewhart_7points(): Detects phase changes by checking for 7 consecutive points above or below the central line.shewhart_model(): Builds a data frame (or tibble) with annotated phase information and control limits.shewhart_fit(): Internally fits the specified model (e.g., “log”, “loglog”, or “gompertz”).
For demonstration, we’ll use a dataset (cvd_recife) included in the
package:
cvd_recife <- readr::read_rds(
system.file("extdata",
package = "Shewhart",
file = "recife_2020_covid19.rds")
) %>%
slice_head(n = 50)
cvd_recife %>%
glimpse()
#> Rows: 50
#> Columns: 2
#> $ data <date> 2020-03-28, 2020-03-29, 2020-03-30, 2020-03-31, 2020-04-0…
#> $ obitosNovos <int> 4, 0, 0, 0, 1, 1, 0, 4, 4, 2, 2, 4, 5, 5, 2, 3, 9, 3, 13, …In this dataset, data is a date column, and obitosNovos are new
daily deaths (integer counts).
You can create a ggplot2-based Shewhart chart simply by setting
type = "ggplot" (the default). The shewhart() function will detect
whether the dataset already has phase information—if not, it internally
calls shewhart_7points() and shewhart_model().
shewhart(
data = cvd_recife,
index_col = data,
values_col = obitosNovos,
locale = "en_US"
)
You can specify a different model (e.g., “loglog”):
shewhart(
data = cvd_recife,
index_col = data,
values_col = obitosNovos,
model = "loglog",
locale = "en_US"
)
In the next example, we first add a new factor column wd to the
cvd_recife dataset by extracting the day of the week from the data
column. By specifying dummy_col = wd in the shewhart() call, we
allow the underlying Shewhart model (here using model = "log") to
account for potential day-of-week effects. In other words, each weekday
becomes a “dummy” factor that might influence the estimated control
limits, helping reveal whether there are systematic shifts or patterns
associated with particular days. We also set locale = “en_US” so that
date labels and text are formatted in English.
shewhart(data = cvd_recife %>% mutate(wd = factor(wday(data))),
index_col = data,
values_col = obitosNovos,
model = "log",
dummy_col = wd,
locale = "en_US")
shewhart(data = cvd_recife %>% mutate(wd = factor(wday(data))),
index_col = data,
values_col = obitosNovos,
model = "loglog",
dummy_col = wd,
locale = "en_US")
To generate an interactive chart with plotly, set
type = "plotly":
shewhart(
data = cvd_recife,
index_col = data,
values_col = obitosNovos,
locale = "en_US",
model = "log",
type = "plotly"
)Hover over points or ribbons to see tooltips, zoom in/out, and more.
You can explicitly detect phase changes using the 7-point rule via
shewhart_7points():
phase_dates <- shewhart_7points(
data = cvd_recife,
index_col = data,
values_col = obitosNovos
)
print(phase_dates)
#> [1] "2020-04-07" "2020-05-06"You can also build the phase model data frame yourself by calling
shewhart_model(), passing the detected phase changes:
shwt_model <- shewhart_model(
data = cvd_recife,
index_col = data,
values_col = obitosNovos,
phase_changes = phase_dates
)
shwt_model %>% head()
#> # A tibble: 6 × 18
#> phase data obitosNovos change model flag N fit tidied fitted
#> <int> <date> <int> <lgl> <chr> <int> <int> <list> <list> <dbl>
#> 1 0 2020-03-28 4 FALSE log 0 1 <lm> <tibble> 0.330
#> 2 0 2020-03-29 0 FALSE log 0 2 <lm> <tibble> 0.419
#> 3 0 2020-03-30 0 FALSE log 0 3 <lm> <tibble> 0.508
#> 4 0 2020-03-31 0 FALSE log 0 4 <lm> <tibble> 0.598
#> 5 0 2020-04-01 1 FALSE log 0 5 <lm> <tibble> 0.687
#> 6 0 2020-04-02 1 FALSE log 0 6 <lm> <tibble> 0.776
#> # ℹ 8 more variables: residuals <dbl>, CONL_1 <dbl>, UCL <dbl>, LCL <dbl>,
#> # CL <dbl>, UL_EXP <dbl>, LL_EXP <dbl>, phase_string <chr>This tibble now has columns like phase, CL, UL_EXP, LL_EXP,
etc., which can be used directly in your custom plots or analyses.
If you already have the phase_changes or the annotated data frame from
shewhart_model, you can pass them to shewhart() to avoid
recalculating:
# Using phase_changes explicitly:
shewhart(
data = cvd_recife,
index_col = data,
values_col = obitosNovos,
model = "loglog",
locale = "en_US",
phase_changes = phase_dates
)
# Or using the 'shwt_model' directly:
shewhart(
data = shwt_model,
index_col = data,
values_col = obitosNovos,
model = "loglog",
locale = "en_US"
)
Both approaches yield the same final plot but skip re-running the 7-point detection or the modeling if the data is already annotated.
This package uses:
- lubridate (>= 1.8.0) for date handling and wday extraction
- tidyverse (>= 1.3.0) for data manipulation (dplyr, tidyr, etc.)
- tibbletime (>= 0.1.6) or slider for rolling sum calculations
- tidymodels (>= 1.0.0) for modeling convenience
- pals (>= 1.7) and scales (>= 1.2.1) for color palettes and scaling
- plotly (>= 4.1) for interactive plots
Ensure these are installed and loaded (as needed) before using
Shewhart.
Visit https://castlab.org for updates and other packages/research from the CASTLab team. If you have suggestions or encounter issues, please open an Issue on GitHub or contact us at leite@castlab.org.
Happy charting!