turn on suggestions

Auto-suggest helps you quickly narrow down your search results by suggesting possible matches as you type.

Showing results for

Find a Community

- Home
- /
- SAS Programming
- /
- General Programming
- /
- should I model the shifted mean based on the origi...

Topic Options

- Subscribe to RSS Feed
- Mark Topic as New
- Mark Topic as Read
- Float this Topic for Current User
- Bookmark
- Subscribe
- Printer Friendly Page

- Mark as New
- Bookmark
- Subscribe
- Subscribe to RSS Feed
- Highlight
- Email to a Friend
- Report Inappropriate Content

10-03-2013 12:01 PM

I have a series, which obviously increased the mean at one point. I was thinking to create a variable x=1 after the change point and x=0 before the change point.

however, the series also has a seasonality. I did span of 12 differencing to remove the seasonality. but the shift disappeared after the differencing. instead it shows some big or low value around the change point.

if I test x for the original series , it is very significant. but after differencing, x is non-significant any more. I guess the differencing just removed the effect of x...what should I do ?

please help!

Xiu Yan

- Mark as New
- Bookmark
- Subscribe
- Subscribe to RSS Feed
- Highlight
- Email to a Friend
- Report Inappropriate Content

Posted in reply to xiuyanzhao

10-03-2013 05:49 PM

You should span of 12 difference x as well. - PG

PG

- Mark as New
- Bookmark
- Subscribe
- Subscribe to RSS Feed
- Highlight
- Email to a Friend
- Report Inappropriate Content

Posted in reply to PGStats

10-03-2013 05:55 PM

how to span of difference X? it is only has 0 and 1, not a series? does that make sense? in addition how to do it in terms of syntax? Could you pls elaborate it and guide me?

Thanks!

Xiu Yan

- Mark as New
- Bookmark
- Subscribe
- Subscribe to RSS Feed
- Highlight
- Email to a Friend
- Report Inappropriate Content

Posted in reply to xiuyanzhao

10-03-2013 09:35 PM

Let's say your series is Y(t), t=1...96 and, as you described, X(t) = 0, t=1...63 and X(t) = 1, t=64...96. Then the deseasonalized series will be Y'(t) = Y(t)-Y(t-12) and X'(t) = X(t)-X(t-12), t=13...96. The X'(t) series will be zero everywhere except for t=64...75 where it will be = 1.

If X turns out to be significant, it will signal that observations up to one year after the change of level were higher than observations for the same month, the year before.

Hope this helps.

PG

PG

- Mark as New
- Bookmark
- Subscribe
- Subscribe to RSS Feed
- Highlight
- Email to a Friend
- Report Inappropriate Content

Posted in reply to PGStats

10-04-2013 10:05 AM

Hi PG,

Thanks! do you mean if the coef. of x is still not sig. after using differencing the x on span of 12, we can conclude that there is no change on that point at 5% level.

we then can just remove the x from the model? or we can leave it in the model.

Thanks!

Xiu Yan

proc **arima**;

var=copd_ohip(**12**) crosscorr=(x(12)) ;

p=(**1**)(**4**) input=(x) method=ml ;

run;quit;

- Mark as New
- Bookmark
- Subscribe
- Subscribe to RSS Feed
- Highlight
- Email to a Friend
- Report Inappropriate Content

Posted in reply to xiuyanzhao

10-04-2013 11:36 AM

You cannot conclude that there is no changepoint. You can only say that you don't have the evidence to detect it statistically. It can also mean that some other aspect of your time series model is inadequate, including the assumed timing of the changepoint.

PG

PG

- Mark as New
- Bookmark
- Subscribe
- Subscribe to RSS Feed
- Highlight
- Email to a Friend
- Report Inappropriate Content

Posted in reply to PGStats

10-04-2013 02:20 PM

Hi PG,

Thanks! That is very helpful! I asked the researcher, she felt this is important from clinical point of view. I would like to take it into account for the forecast, however, my forecast plot does not seem to do this.

Do you have an idea how to incorporate this mean increases in the forecast? could you please guide me?

Xiu Yan

- Mark as New
- Bookmark
- Subscribe
- Subscribe to RSS Feed
- Highlight
- Email to a Friend
- Report Inappropriate Content

Posted in reply to xiuyanzhao

10-04-2013 03:05 PM

Hi Xiu Yan,

If your forecasts are more than a year away from the changepoint, they will not depend on the series before the changepoint, they will depend almost entirely on recent values and the corresponding values from the previous year. What's wrong with your forecasts? - PG

*Prediction is very difficult, especially about the future*. - Niels Bohr (1885 - 1962)

PG

- Mark as New
- Bookmark
- Subscribe
- Subscribe to RSS Feed
- Highlight
- Email to a Friend
- Report Inappropriate Content

Posted in reply to PGStats

10-07-2013 10:38 AM

- Mark as New
- Bookmark
- Subscribe
- Subscribe to RSS Feed
- Highlight
- Email to a Friend
- Report Inappropriate Content

Posted in reply to xiuyanzhao

10-07-2013 10:48 AM

Hi PG,

Thanks! I got it. I thought the forecasts should have similar

behaviour. I mean two means-one mean increases after point.

You are right. The incorporation of that change will depending on the value

after the point since we usually predict more than one year.

I rechecked my data, and re-set the changing point. now it is sig. in my

model. I guess your guidance have helped me solve this problem.

please see my forecast plot. Does this looks OK?

Thanks!

Xiu Yan

- Mark as New
- Bookmark
- Subscribe
- Subscribe to RSS Feed
- Highlight
- Email to a Friend
- Report Inappropriate Content

Posted in reply to xiuyanzhao

10-07-2013 11:10 AM

Hi Xiu Yan,

Yes! That looks pretty good! Congratulations. Glad I could help.

PG

PG

- Mark as New
- Bookmark
- Subscribe
- Subscribe to RSS Feed
- Highlight
- Email to a Friend
- Report Inappropriate Content

Posted in reply to PGStats

10-07-2013 11:35 AM

Hi PG,

One more thought, will my forecast be questionable due to the fact that my forecast is more rely on the recent value ?

I incorporate the mean increases behavior, but I lost something. .. right?

how about an example, that the series keep flat before the point and downward after the point. The forecast would be decreasing then...is that appropriate?

Xiu Yan