Hi
I'm just stepping through the decompose() function, in "stats". Does
this contained line of code not work if you have a time series ending
"unevenly" (i.e., middle of the year), or am I missing something?
season <- na.omit(c(as.numeric(window(season, start(x) +
c(1, 0), end(x))), as.numeric(window(season, start(x),
start(x) + c(0, f)))))
The line seems to cut the first year of incomplete data, pare off NAs,
and stick the incomplete first year at the end. It makes sense if I
have an data series that is a multiple of the frequency, but I am
perplexed otherwise.
Here's something to try. I calculate the seasonal adjustment figures
in two ways here. One is trying to replicate decompose(), and another
way which doesn't rely on the above line.
# I get identical results for:
x <- ts(c(6.3, 6.6, 7.5, 5.4, 6.7, 5.6, 6.9, 7.3, 5.3, 5.3, 4.8, 5.5,
5.4, 4.5, 7.7, 7.1, 6.1, 5.6, 7, 8.8, 7.8, 6, 5.8, 5.3, 6.2,
6.2, 6.3, 8.7, 6.8, 6.7, 9.3, 10.1, 8.2, 5.8, 5.2, 5.8), start=c(1987,
1), frequency=12)
# But different results for:
x <- ts(c(6.3, 6.6, 7.5, 5.4, 6.7, 5.6, 6.9, 7.3, 5.3, 5.3, 4.8, 5.5,
5.4, 4.5, 7.7, 7.1, 6.1, 5.6, 7, 8.8, 7.8, 6, 5.8, 5.3, 6.2,
6.2, 6.3, 8.7, 6.8, 6.7, 9.3, 10.1, 8.2, 5.8, 5.2, 5.8, 5.8,
7.5, 6.7, 6.4, 7.8, 6.5, 9.1, 10.7, 8.8, 5.5, 4.3, 4.3, 5, 6.2,
6.6, 5.3, 8.4), start=c(1987, 1), frequency=12)
# Here's what I run:
# This part takes the centered moving averages: trend
l <- length(x)
f <- frequency(x)
filter <- c(0.5, rep(1, f - 1), 0.5)/f
trend <- filter(x, filter)
# Assuming multiplicative
season <- x/trend
# This part confuses me!
season <- na.omit(c(as.numeric(window(season, start(x) +
c(1, 0), end(x))), as.numeric(window(season, start(x),
start(x) + c(0, f)))))
# Calculating the mean of seasonal factor for each month
periods <- l%/%f
index <- c(0, cumsum(rep(f, periods - 2)))
figure <- numeric(f)
for (i in 1L:f) figure[i] <- mean(season[index + i])
# Neutrality adjustment
figure <- figure/mean(figure)
# Here's an alternate method of calculating the monthly mean
figure_alt <- numeric(f)
for(i in 1L:f) figure_alt[i] <- mean((x/trend)[cycle(x)==i],na.rm=T)
figure_alt <- figure_alt/mean(figure_alt)
# For comparison
decompose(x,"multiplicative")$seasonal[1:12]
figure
figure_alt
Have I missed something fundamental, or is this a problem?
Peter
______________________________________________
R-help@r-project.org mailing list
https://stat.ethz.ch/mailman/listinfo/r-help
PLEASE do read the posting guide http://www.R-project.org/posting-guide.html
and provide commented, minimal, self-contained, reproducible code.
