Skip to contents

Exponentially Weighted Moving Average (EWMA)

Source: R/child_clean.R
ewma.Rd

ewma calculates the exponentially weighted moving average (EWMA) for a set of numeric observations over time.

Usage

ewma(agedays, z, ewma.exp, ewma.adjacent = TRUE, window = 15, cache_env = NULL)

Arguments

agedays

Vector of age in days for each z score (potentially transformed to adjust weighting).

z

Input vector of numeric z-score data.

ewma.exp

Exponent used for age-gap weighting. May be a single scalar (recycled across all observations) or a per-observation vector matched to agedays / z; the algorithm's main internal callers pass a per-observation vector that varies the exponent by the widest neighbor age gap.

ewma.adjacent

Specify whether EWMA values excluding adjacent measurements should be calculated. Defaults to TRUE.

window

Maximum number of observations on each side for EWMA weighting. Default 15. Set to Inf to disable windowing.

cache_env

Optional environment for caching EWMA intermediate results. Used internally for performance. Defaults to NULL.

Value

A named list of numeric vectors, each the same length as agedays.

  • ewma.all - EWMA at each observation's age, excluding only the observation itself.

  • ewma.before - also excludes the immediate prior observation (sorted by agedays); equals ewma.all for the first observation and when n <= 2.

  • ewma.after - also excludes the immediate subsequent observation; equals ewma.all for the last observation and when n <= 2.

When ewma.adjacent = FALSE, only ewma.all is returned (still wrapped in a one-element named list).

Examples

# Run on 1 subject, 1 type of parameter
df_stats <- as.data.frame(syngrowth)
df_stats <- df_stats[df_stats$subjid == df_stats$subjid[1] &
                       df_stats$param == "HEIGHTCM", ]

# Get the uncentered z-scores
measurement_to_z <- read_anthro(cdc.only = TRUE)
sd <- measurement_to_z(df_stats$param,
                       df_stats$agedays,
                       df_stats$sex,
                       df_stats$measurement,
                       TRUE)

# Calculate exponentially weighted moving average
e_df <- ewma(df_stats$agedays, sd, ewma.exp = -1.5)