Skip to contents

Trend analysis

Source: vignettes/trend-analysis.Rmd
trend-analysis.Rmd

run_grouped_stl_trend() detects trends in grouped monthly time series - for example incidence and TPR by district. For each group and indicator it runs an STL decomposition (seasonal-trend decomposition using loess) to separate seasonality from the underlying trend, then tests that trend with the Mann-Kendall non-parametric test and estimates its magnitude with Sen’s slope.

This is a natural follow-on to the routine-data workflow: feed it the monthly incidence/TPR series you produced there.

Usage 

library(sntmethods)

trends <- run_grouped_stl_trend(
  data      = monthly_district_data,
  group_col = c("adm1", "adm2"),
  date_col  = "date",
  indicators = list(
    list(col = "n2_incidence", type = "incidence"),
    list(col = "tpr",          type = "positivity")
  )
)
  • group_col - one or more columns that define each series (e.g. admin units).
  • date_col - the monthly date column.
  • indicators - a list of indicators to analyse, each naming the value column and its type (used for sensible defaults and labelling).

What you get back

For each group × indicator, the result reports the decomposed components plus the trend test:

  • the Mann-Kendall statistic and p-value (is there a monotonic trend?),
  • Sen’s slope (how steep, robust to outliers),
  • and the direction/significance classification you can map or tabulate.

Because the output is tidy and keyed by your grouping columns, it joins straight back onto admin geometries for trend maps, or into a table of which districts are improving or worsening.

See also