dev/AGEINT.R
In timriffe/DemoTools: Standardize, Evaluate, and Adjust Demographic Data
# TR: 2018-08-20 either complementing or replacing previous pop interpolater.
# this one more general.
###############################################################################
#' Interpolate between two population age distributions.
#' @description The interpolation is done by age using a linear, exponential, or power function. This comes from the PAS spreadsheet called \code{AGEINT}. Be aware that this is not cohort-component interpolation.
#' @param popmat numeric. An age-period matrix of data to interpolate over. Age in rows, time in columns.
#' @param datesIn vector of dates. The exact dates in each column. See details for ways to express it.
#' @param datesOut vector of dates. The desired dates to interpolate to. See details for ways to express it.
#' @param method string. The method to use for the interpolation, either \code{"linear"}, \code{"exponential"}, or \code{"power"}. Default \code{"linear"}.
#' @param power numeric power to interpolate by, if \code{method = "power"}. Default 2.
#' @details The age group structure of the output is the same as that of the input. Ideally, \code{datesOut} should be within the range of \code{datesIn}. If not, the left-side and right-side output are held constant outside the range if \code{rule = 2} is passed in, otherwise \code{NA} is returned (see examples). Dates can be given in three ways 1) a \code{Date} class object, 2) an unambiguous character string in the format \code{"YYYY-MM-DD"}, or 3) as a decimal date consisting in the year plus the fraction of the year passed as of the given date.
#'
#' For methods \code{"exponential"} and \code{"power"}, calculations are carried out using linear interpolation through \code{log(popmat)}, or \code{popmat^(1/power)} respectively, then back-transformed. If the data contain 0s, \code{method = "exponential"} will fail, but \code{"power"} would still generally work.
#'
#' @return numeric matrix (age-period) (or vector if \code{length(datesOut) == 1} of the interpolated data for the requested dates.
#' @author Tim Riffe
#' @references
#' \insertRef{PAS}{DemoTools}
#' @export
#' @importFrom stats approx
#'
#' @examples
#' # Example of interpolating over time series of age-structured
#' # data. NOTE: age classes must conform. They don't have to be
#' # uniform, just the same in each year.
#' popmat <- structure(c(2111460L, 1971927L, 1651421L, 1114149L, 921120L,
#' 859806L, 806857L, 847447L, 660666L, 567163L, 493799L, 322936L,
#' 320796L, 163876L, 133531L, 120866L, 2548265L, 2421813L, 2581979L,
#' 2141854L, 1522279L, 1321665L, 1036480L, 1024782L, 935787L, 789521L,
#' 719185L, 481997L, 479943L, 268777L, 202422L, 167962L, 3753848L,
#' 3270658L, 2930638L, 2692898L, 2222672L, 1788443L, 1514610L, 1491751L,
#' 1054937L, 972484L, 796138L, 673137L, 554010L, 352264L, 293308L,
#' 195307L, 3511776L, 3939121L, 4076601L, 3602857L, 2642620L, 2105063L,
#' 1993212L, 1914367L, 1616449L, 1408498L, 994936L, 776537L, 706189L,
#' 507085L, 315767L, 240302L, 3956664L, 3936424L, 3995805L, 4371774L,
#' 4012982L, 3144046L, 2406725L, 2301514L, 2061252L, 1871502L, 1538713L,
#' 1210822L, 896041L, 638954L, 401297L, 375648L), .Dim = c(16L,
#' 5L), .Dimnames = list(c("0", "5", "10", "15", "20", "25", "30",
#' "35", "40", "45", "50", "55", "60", "65", "70", "75"), c("1960",
#' "1976", "1986", "1996", "2006")))
#'
#' # We have irregularly spaced census inputs.
#' \dontrun{
#' matplot(seq(0,75,by=5),popmat,type='l',col=1:5)
#' text(20,popmat["20",],colnames(popmat),col=1:5)
#' }
#' # census dates as decimal: no need for year rounding.
#' # could also specify as "YYYY-MM-DD"
#' # (rounded to 2 digits here- not necessary)
#' datesIn <- c(1960.73, 1976.90, 1986.89, 1996.89, 2006.89)
#' # could ask for single years or anything
#' datesOut <- seq(1960, 2005, by = 5)
#'
#' # NOTE: rule is an argument to approx(), which does the work.
#' # if not passed in, 1960 is returned as NAs (outside the range
#' # of dates given). In this case, rule = 2 assumes constant
#' # equal to nearest neighbor, meaning 1960 data will be returned
#' # as equal to 1960.73 input data. So, function should only be
#' # allowed to extrapolate for short distances.
#' interp(popmat, datesIn, datesOut, "linear", rule = 2)
#' interp(popmat, datesIn, datesOut, "exponential", rule = 2)
#' interp(popmat, datesIn, datesOut, "power", rule = 2)
#'
#' # ----------------------------------------------------------
#' # standard example data, taken from AGEINT PAS spreadsheet.
#' p1 <- c(100958, 466275, 624134, 559559, 446736, 370653, 301862, 249409,
#' 247473, 223014, 172260, 149338, 127242, 105715, 79614, 53660, 31021, 34596)
#' p2 <- c(201916, 932550, 1248268, 1119118, 893472, 741306, 603724, 498818,
#' 494946, 446028, 344520, 298676, 254484, 211430, 159228, 107320, 62042, 69192)
#'
#' # YYYY-MM-DD dates as character
#' d1 <- "1980-04-07"
#' d2 <- "1990-10-06"
#' dout <- "1985-07-01"
#'
#' (p_lin <- interp(cbind(p1,p2),c(d1,d2),dout,method = "linear"))
#' (p_exp <- interp(cbind(p1,p2),c(d1,d2),dout,method = "exponential"))
#' (p_pow <- interp(cbind(p1,p2),c(d1,d2),dout,method = "power"))
#'
#' \dontshow{
#' print(dec.date(d1),digits=18)
#' print(dec.date(d2),digits=18)
#' print(dec.date(dout),digits=18)
#' # these were set to different dates: recreate above dates...
#' expcheck <- c(142637.629438722,658772.565488024,881802.276313982,
#' 790568.050982602,631167.059816327,523673.857092558,426483.092945892,
#' 352375.329549735,349640.068841387,315083.384096823,243376.038026844,
#' 210990.890321914,179772.749510111,149358.515383768,112481.945265698,
#' 75813.0628150498,43827.7491909366,48878.656748965)
#' lincheck <- c(151295.597805253,698759.433295475,935326.835323442,
#' 838554.779337049,669478.29970015,555460.361866426,452370.21082717,
#' 373764.176717154,370862.888286807,334208.645663947,258148.731927464,
#' 223797.836576011,190684.784325522,158424.435131266,119309.492300436,
#' 80414.843580795,46488.0518583646,51845.5446984939)
#'
#' stopifnot(all(abs(expcheck - p_exp)<1e-6))
#' stopifnot(all(abs(lincheck - p_lin)<1e-6))
#' # do controlled spreadsheet test, passing in decimal dates, etc
#' }
#' \dontrun{
#' age <- c(0,1,seq(5,80,by=5))
#' plot(age, p2, type = 'l', lwd = 2, main = "Interpolation options")
#' lines(age, p1, lwd = 2, lty = "8282")
#' lines(age, p_lin,col = "blue")
#' lines(age, p_pow,col = gray(.4))
#' lines(age, p_exp,col = "red")
#' text(30, p1[8], d1, pos = 1, cex = 1)
#' text(20, p2[6], d2, pos = 4, cex = 1)
#' legend("topright", lty = 1, col = c("blue", gray(.4), "red"),
#' legend = c("linear", "power 2", "exponential"),
#' title = paste0("Interpolated 1985-07-01"))
#'
#' }
interp <- function(popmat,
datesIn,
datesOut,
method = c("linear","exponential","power")[1],
power = 2,
...){ # ... args passed to stats::approx . Can give control over extrap assumptions
# a basic check
stopifnot(ncol(popmat) == length(datesIn))
# no sense documenting this wrapper ...
.approxwrap <- function(x, y, xout, ...){
stats::approx(x = x, y = y, xout = xout, ...)$y
}
# -----------------------
# clean method declaration
method <- tolower(method)
if (grepl(method, pattern = "exp")){
pattern <- "exponential"
}
if (grepl(method, pattern = "lin")){
pattern <- "linear"
}
if (grepl(method, pattern = "pow")){
pattern <- "power"
}
# -----------------------
# coerce dates to decimal if necessary
datesIn <- sapply(datesIn, dec.date)
datesOut <- sapply(datesOut, dec.date)
# carry out transform 1
if (method == "exponential"){
if (any(popmat == 0)){
stop("popmat contains 0s, so exponential interpolation won't work.\n
Either handle 0s then do this, or\n
maybe try method = 'power'.")
}
popmat <- log(popmat)
}
if (method == "power"){
popmat <- popmat ^ (1 / power)
}
int <- apply(
popmat,
1,
.approxwrap,
x = datesIn,
xout = datesOut,
...)
dims <- dim(int)
if (!is.null(dims)){
int <- t(int)
rownames(int) <- rownames(popmat)
colnames(int) <- datesOut
} else {
names(int) <- rownames(popmat)
}
# transform back
if (method == "exponential"){
int <- exp(int)
}
if (method == "power"){
int <- int^power
}
int
}
timriffe/DemoTools documentation built on Jan. 28, 2024, 5:13 a.m.
R Package Documentation
Browse R Packages
We want your feedback!
Note that we can't provide technical support on individual packages. You should contact the package authors for that.
# TR: 2018-08-20 either complementing or replacing previous pop interpolater.
# this one more general.
###############################################################################
#' Interpolate between two population age distributions.
#' @description The interpolation is done by age using a linear, exponential, or power function. This comes from the PAS spreadsheet called \code{AGEINT}. Be aware that this is not cohort-component interpolation.
#' @param popmat numeric. An age-period matrix of data to interpolate over. Age in rows, time in columns.
#' @param datesIn vector of dates. The exact dates in each column. See details for ways to express it.
#' @param datesOut vector of dates. The desired dates to interpolate to. See details for ways to express it.
#' @param method string. The method to use for the interpolation, either \code{"linear"}, \code{"exponential"}, or \code{"power"}. Default \code{"linear"}.
#' @param power numeric power to interpolate by, if \code{method = "power"}. Default 2.
#' @details The age group structure of the output is the same as that of the input. Ideally, \code{datesOut} should be within the range of \code{datesIn}. If not, the left-side and right-side output are held constant outside the range if \code{rule = 2} is passed in, otherwise \code{NA} is returned (see examples). Dates can be given in three ways 1) a \code{Date} class object, 2) an unambiguous character string in the format \code{"YYYY-MM-DD"}, or 3) as a decimal date consisting in the year plus the fraction of the year passed as of the given date.
#'
#' For methods \code{"exponential"} and \code{"power"}, calculations are carried out using linear interpolation through \code{log(popmat)}, or \code{popmat^(1/power)} respectively, then back-transformed. If the data contain 0s, \code{method = "exponential"} will fail, but \code{"power"} would still generally work.
#'
#' @return numeric matrix (age-period) (or vector if \code{length(datesOut) == 1} of the interpolated data for the requested dates.
#' @author Tim Riffe
#' @references
#' \insertRef{PAS}{DemoTools}
#' @export
#' @importFrom stats approx
#'
#' @examples
#' # Example of interpolating over time series of age-structured
#' # data. NOTE: age classes must conform. They don't have to be
#' # uniform, just the same in each year.
#' popmat <- structure(c(2111460L, 1971927L, 1651421L, 1114149L, 921120L,
#' 859806L, 806857L, 847447L, 660666L, 567163L, 493799L, 322936L,
#' 320796L, 163876L, 133531L, 120866L, 2548265L, 2421813L, 2581979L,
#' 2141854L, 1522279L, 1321665L, 1036480L, 1024782L, 935787L, 789521L,
#' 719185L, 481997L, 479943L, 268777L, 202422L, 167962L, 3753848L,
#' 3270658L, 2930638L, 2692898L, 2222672L, 1788443L, 1514610L, 1491751L,
#' 1054937L, 972484L, 796138L, 673137L, 554010L, 352264L, 293308L,
#' 195307L, 3511776L, 3939121L, 4076601L, 3602857L, 2642620L, 2105063L,
#' 1993212L, 1914367L, 1616449L, 1408498L, 994936L, 776537L, 706189L,
#' 507085L, 315767L, 240302L, 3956664L, 3936424L, 3995805L, 4371774L,
#' 4012982L, 3144046L, 2406725L, 2301514L, 2061252L, 1871502L, 1538713L,
#' 1210822L, 896041L, 638954L, 401297L, 375648L), .Dim = c(16L,
#' 5L), .Dimnames = list(c("0", "5", "10", "15", "20", "25", "30",
#' "35", "40", "45", "50", "55", "60", "65", "70", "75"), c("1960",
#' "1976", "1986", "1996", "2006")))
#'
#' # We have irregularly spaced census inputs.
#' \dontrun{
#' matplot(seq(0,75,by=5),popmat,type='l',col=1:5)
#' text(20,popmat["20",],colnames(popmat),col=1:5)
#' }
#' # census dates as decimal: no need for year rounding.
#' # could also specify as "YYYY-MM-DD"
#' # (rounded to 2 digits here- not necessary)
#' datesIn <- c(1960.73, 1976.90, 1986.89, 1996.89, 2006.89)
#' # could ask for single years or anything
#' datesOut <- seq(1960, 2005, by = 5)
#'
#' # NOTE: rule is an argument to approx(), which does the work.
#' # if not passed in, 1960 is returned as NAs (outside the range
#' # of dates given). In this case, rule = 2 assumes constant
#' # equal to nearest neighbor, meaning 1960 data will be returned
#' # as equal to 1960.73 input data. So, function should only be
#' # allowed to extrapolate for short distances.
#' interp(popmat, datesIn, datesOut, "linear", rule = 2)
#' interp(popmat, datesIn, datesOut, "exponential", rule = 2)
#' interp(popmat, datesIn, datesOut, "power", rule = 2)
#'
#' # ----------------------------------------------------------
#' # standard example data, taken from AGEINT PAS spreadsheet.
#' p1 <- c(100958, 466275, 624134, 559559, 446736, 370653, 301862, 249409,
#' 247473, 223014, 172260, 149338, 127242, 105715, 79614, 53660, 31021, 34596)
#' p2 <- c(201916, 932550, 1248268, 1119118, 893472, 741306, 603724, 498818,
#' 494946, 446028, 344520, 298676, 254484, 211430, 159228, 107320, 62042, 69192)
#'
#' # YYYY-MM-DD dates as character
#' d1 <- "1980-04-07"
#' d2 <- "1990-10-06"
#' dout <- "1985-07-01"
#'
#' (p_lin <- interp(cbind(p1,p2),c(d1,d2),dout,method = "linear"))
#' (p_exp <- interp(cbind(p1,p2),c(d1,d2),dout,method = "exponential"))
#' (p_pow <- interp(cbind(p1,p2),c(d1,d2),dout,method = "power"))
#'
#' \dontshow{
#' print(dec.date(d1),digits=18)
#' print(dec.date(d2),digits=18)
#' print(dec.date(dout),digits=18)
#' # these were set to different dates: recreate above dates...
#' expcheck <- c(142637.629438722,658772.565488024,881802.276313982,
#' 790568.050982602,631167.059816327,523673.857092558,426483.092945892,
#' 352375.329549735,349640.068841387,315083.384096823,243376.038026844,
#' 210990.890321914,179772.749510111,149358.515383768,112481.945265698,
#' 75813.0628150498,43827.7491909366,48878.656748965)
#' lincheck <- c(151295.597805253,698759.433295475,935326.835323442,
#' 838554.779337049,669478.29970015,555460.361866426,452370.21082717,
#' 373764.176717154,370862.888286807,334208.645663947,258148.731927464,
#' 223797.836576011,190684.784325522,158424.435131266,119309.492300436,
#' 80414.843580795,46488.0518583646,51845.5446984939)
#'
#' stopifnot(all(abs(expcheck - p_exp)<1e-6))
#' stopifnot(all(abs(lincheck - p_lin)<1e-6))
#' # do controlled spreadsheet test, passing in decimal dates, etc
#' }
#' \dontrun{
#' age <- c(0,1,seq(5,80,by=5))
#' plot(age, p2, type = 'l', lwd = 2, main = "Interpolation options")
#' lines(age, p1, lwd = 2, lty = "8282")
#' lines(age, p_lin,col = "blue")
#' lines(age, p_pow,col = gray(.4))
#' lines(age, p_exp,col = "red")
#' text(30, p1[8], d1, pos = 1, cex = 1)
#' text(20, p2[6], d2, pos = 4, cex = 1)
#' legend("topright", lty = 1, col = c("blue", gray(.4), "red"),
#' legend = c("linear", "power 2", "exponential"),
#' title = paste0("Interpolated 1985-07-01"))
#'
#' }
interp <- function(popmat,
datesIn,
datesOut,
method = c("linear","exponential","power")[1],
power = 2,
...){ # ... args passed to stats::approx . Can give control over extrap assumptions
# a basic check
stopifnot(ncol(popmat) == length(datesIn))
# no sense documenting this wrapper ...
.approxwrap <- function(x, y, xout, ...){
stats::approx(x = x, y = y, xout = xout, ...)$y
}
# -----------------------
# clean method declaration
method <- tolower(method)
if (grepl(method, pattern = "exp")){
pattern <- "exponential"
}
if (grepl(method, pattern = "lin")){
pattern <- "linear"
}
if (grepl(method, pattern = "pow")){
pattern <- "power"
}
# -----------------------
# coerce dates to decimal if necessary
datesIn <- sapply(datesIn, dec.date)
datesOut <- sapply(datesOut, dec.date)
# carry out transform 1
if (method == "exponential"){
if (any(popmat == 0)){
stop("popmat contains 0s, so exponential interpolation won't work.\n
Either handle 0s then do this, or\n
maybe try method = 'power'.")
}
popmat <- log(popmat)
}
if (method == "power"){
popmat <- popmat ^ (1 / power)
}
int <- apply(
popmat,
1,
.approxwrap,
x = datesIn,
xout = datesOut,
...)
dims <- dim(int)
if (!is.null(dims)){
int <- t(int)
rownames(int) <- rownames(popmat)
colnames(int) <- datesOut
} else {
names(int) <- rownames(popmat)
}
# transform back
if (method == "exponential"){
int <- exp(int)
}
if (method == "power"){
int <- int^power
}
int
}
R Package Documentation
Browse R Packages
We want your feedback!
Note that we can't provide technical support on individual packages. You should contact the package authors for that.
Embedding an R snippet on your website
Add the following code to your website.
For more information on customizing the embed code, read Embedding Snippets.