Seasonal decompose of monthly data including NA in r

问题

I need help from you guys to decompose my monthly data which have seasonality, but it does not work because NA values are not removed..There may be another problem. Please look at my data and errors as below.

    ts.monthly<-ts(monthly$rBC.median, frequency=12, start=c(2006, 4))
    ts.monthly
        Jan        Feb        Mar        Apr        May        Jun
    2006                                   5.1656479  6.2847959 19.4833690
    2007  1.4252665  2.9127775  2.8912652  7.5326158  8.6182227 23.2129310
    2008         NA  1.8200842  1.3488755  2.0700927  5.3541366  8.6916708
    2009  1.2531161  1.5075780  2.4955524 10.6724704 10.1367162 16.0362127
    2010  0.8850190  2.4974866  1.8459976  9.2297697  3.8203789  7.1492986
    2011  2.6990434  0.4570701  1.3787403  5.8739804  4.1669501 13.2228535
    2012         NA  2.0670538  1.3758499 11.7306663  4.1248775 12.3604423
                Jul        Aug        Sep        Oct        Nov        Dec
    2006  9.8028986  7.8167810  2.1333807  2.5777504  1.9022561  2.7254065
    2007  4.2121577  8.8604768 12.0017155  4.0978332  1.6053110         NA
    2008  5.7338211  9.7432563  4.6548508  1.3589789  0.9650082  1.2788504
    2009 11.7632775 11.2299683  1.6229679  1.0333217  1.0481580  1.0734208
    2010  3.5996501  4.3245873  4.4586863  1.6403104  2.8622518  1.2564256
    2011  3.0463918  7.1515472  6.5613683  1.3715623  1.9757217  5.4901524
    2012 11.1010563  3.6220968  2.2597341  

   ts.monthly=na.omit(ts.monthly)  
    Error in na.omit.ts(ts.monthly) : time series contains internal NAs
   ts.monthly.com<-decompose(ts.monthly)
    Error in na.omit.ts(x) : time series contains internal NAs
   ts.monthly$seasonal
    Error in ts.monthly$seasonal : $ operator is invalid for atomic vectors

I do not understand why na.omit does not work.. how can I treat this NA??

Finally, after using a function "decompose", I want to take only "trend" without seasonality, and then apply sen's slope estimator to get a slope for linear trend. It will work??

Thanks a lot for your help.

回答1:

Try filling in the missing values using a seasonal Kalman filter by employing na.StructTS from the zoo package first:

library(zoo)
decompose(na.StructTS(ts.monthly))

zoo has many other na. functions as well: na.aggregate, na.approx, na.fill, na.locf, na.spline, na.StructTS, na.trim.

回答2:

The X-13ARIMA-SEATS software, accessible by the R-package seasonal, handles missing values and seasonal decomposition in a single step:

library(seasonal)

# a monthly time series with some missing values
AirPassengersNA <- AirPassengers
AirPassengersNA[c(2, 24)] <- NA

m <- seas(AirPassengersNA, na.action = na.x13)
head(m$data)

        final  seasonal seasonaladj    trend irregular adjustfac
[1,] 122.5860 0.9029705    122.5860 122.6289 0.9996500 0.9136445
[2,] 123.8615 0.9492046    123.8615 123.8656 0.9999671 0.9408045
[3,] 125.0191 1.0701984    125.0191 125.3132 0.9976535 1.0558387
[4,] 127.4633 1.0028864    127.4633 126.6222 1.0066428 1.0120561
[5,] 127.2526 0.9494692    127.2526 126.8592 1.0031006 0.9508650
[6,] 126.0700 1.0771444    126.0700 126.1723 0.9991886 1.0708339

回答3:

I struggled with this for a long time too.

Just use na.locf from the zoo package on your ts object. na.locf returns a ts object, so there are no worries about a changed object type.

Use:

    library(zoo)
    season_ts <- na.locf(season_ts)

where season_ts is your ts object.

回答4:

The imputeTS package is a R package solely dedicated to replacing missing values in time series. You can use a function of the package before performing your decomposition.

The na.seadec() and the na.kalman() are especially good for replacing missing data in seasonal time series. But there are also other advanced methods available (Link to imputeTS Paper).

You would use it like this for this problem:

library(imputeTS)
x <- decompose(na.seadec(yourTimeSeries))

or if you want to use the na.kalman method instead:

library(imputeTS)
x <- decompose(na.kalman(yourTimeSeries))

来源：https://stackoverflow.com/questions/24694558/seasonal-decompose-of-monthly-data-including-na-in-r

标签

time-series