Question in Time-Series -SUMMARY

[Pat Farr <patricia.farra@rogers.com>]

I received prompt, kind, and valuable suggestions from Laura Holt and Henrik Aalborg Nielsen. Many thanks! The suggestions follow. If you are interested in time series, please read to the end. In multivariate time series as considered in this posting, the individuals (subjects) are the variables and measurement occasion gives a serie. I was right saying "I am afraid I am missing theory".

 

Laura Holt wrote:

 

Do you mean a multivariate time series, please?

>x1 <- matrix(rnorm(40),nrow=4,ncol=10)
>x1.ts <- rts(x1,frequency=2,start=1995)
>x1.ts
    Series 1   Series 2   Series 3   Series 4   Series 5   Series 6   Series 7   Series 8   Series 9   Series 10
1995.0  0.4692720  0.7460260 -0.8682816 -0.3630142  0.5585065 -0.3924150 -0.1154636 -1.6124470  1.8504896  0.06445123
1995.5 -0.4837774 -0.5160213 -0.4999353 -0.4086840 -0.4596804 -1.8399909 -2.4181038 -2.4651423 -1.8367840 -0.67806762
1996.0  0.6136543  1.4780755 -1.2902717 -1.4098757 -1.1124667  2.6178822  2.4640857  0.1587029  2.2451479  0.07457770
1996.5 -1.8427487 -0.6714690  0.2860823 -1.0762595 -0.3879411 -0.7800703 -1.0769670 -0.9179550 -0.3266432 -0.57922833
start deltat frequency
  1995    0.5         2
>

You need 10 series, with a frequency of 2 (for the 6 months)

Hope this helps!

Sincerely,
Laura

Your suggestion works! However, it puts me in another difficulty as I
have
more than two variables.

> serie1<-c(11, 9, 11, 22, 21, 19, 19, 15, 13, 7)
> serie2<-c(15, 17, 14, 17, 15, 24, 12, 11, 12, 4)
> serie3<-c(13, 12, 15, 6, 15, 17, 13, 7, 19, 11)
> serie4<-c(18, 15, 15, 9, 13, 13, 12, 10, 17, 12)
> x<-as.matrix(rbind(serie1,serie2,serie3,serie4))
> x.ts <- rts(x,frequency=2,start=2001)
> x.ts
>     Series 1 Series 2 Series 3 Series 4 Series 5 Series 6 Series 7 Series 8 Series 9 Series 10
>serie1       11        9       11       22       21       19       19       15       13         7
>serie2       15       17       14       17      15       24       12       11       12         4
>serie3       13       12       15        6       15       17       13         7       19        11
>serie4       18       15       15        9       13       13       12       10       17        12
  start deltat frequency
   2001    0.5         2

When I consider all the 40 observations on variables together, I have:
> serie<-c(serie1,serie2,serie3,serie4)
> serie
  [1] 11  9 11 22 21 19 19 15 13  7 15 17 14 17 15 24 12 11 12  4 13 12 15& nbsp; 6 15 17 13  7 19 11 18 15 15  9 13 13 12 10
>[39] 17 12
> rts(serie,frequency=2,start=2001)
               1  2  1  2  1  2  1  2  1  2  1  2  1  2  1  2  1  2  1  2  1  2  1 2  1  2  1  2  1  2  1  2  1  2  1  2  1  2
>2001: 11  9 11 22 21 19 19 15 13  7 15 17 14 17 15 24 12 11 12  4 13 12 15  6 15 17 13  7 19 11 18 15 15  9 13 13 12 10
>2020: 17 12
start deltat frequency
2001    0.5         2

showing my observations "ended" in 2020! I would be very
appreciative of your help to extend your suggestion to three variables assuming equal spacing of 6 months (to make it ea sy until I understand what is going on; I would prefer its() to rts()).

Sincerely,
Pat

Laura, very knidly, replied:

 

Ok.  You have 10 variables: seri1 through serie10.

Each series has 4 observations.
What I did was put together some fake dates in the next step:
>obs.times <-
dates(c("01/01/1995","07/01/1998","12/01/1999","01/01/2003"))
>obs.times
[1] 01/01/1995 07/01/1998 12/01/1999 01/01/2003

Here, I just generated a matrix.  What you would want to do is
cbind(seri1,seri2,seri3)
>x <- matrix(rnorm(40),nrow=4,ncol=10)

Now I create an its.  The data is in x, while the observation times are
in obs.times.
>x.its <- its(x, times=obs.times)
>x.its
              Series 1   Series 2   Series 3  Series 4   Series 5   Series 6 Series 7   Series 8   Series 9  Series 10
01/01/1995 -0.96275047  0.3086150 -0.3232615 0.7133833  0.8644014 -1.6 651707 -0.76674859  1.3370470  1.2941335 -0.5012535
07/01/1998  1.00891721 -2.8632682 -0.9608544 0.3475233 -1.2520791  1.6180516  0.72959253  1.0774104  0.7418033  0.1483676
12/01/1999 -1.32420419 -0.7070052 -0.9639293 0.1648563  0.1603369 -0.2901317  0.12260904  0.1657497 -0.4931657 -0.9986250
01/01/2003 -0.03907015  0.1445699  1.9400356 1.0557461  0.7901353 -0.1000546  0.07382149 -1.6310907 -0.2132029  0.2581173
      start            end
01/01/1995 ... 01/01/2003
>

You actually don't have 40 observations in the sense that there are 40 observations in 1 series.

You have 10 variables...maybe they represent people.  X1 is person one, X2 is person two, and so on.

Every six months you return to each person and ask him/her a question.

Does this help?  ; Time series (particularly multivariate time series) can be a bit tricky at first.

Henrik Aalborg Nielsen wrote:

 

Dear Pat,

If I understand you correct you follow 10 individuals for 18 months
and collect data on each individual every six months; resulting in 4
measurements for each individual. I believe this should be handled as
longitudinal data analysis. Venables and Ripley "Modern applied
statistics with S-PLUS" has a section on this (10.3 in the 2nd ed).

Regards,
Henrik

 

 

Enjoy,

Pat