- Mark as New
- Bookmark
- Subscribe
- Mute
- RSS Feed
- Permalink
- Report Inappropriate Content
I’m doing segmented regression. I understand the theory behind it, but I can’t figure out which is the correct SAS model/output. I found 2 different models in 2 books and they give slightly different results. Any help is appreciated.
Example data:
Month Mar Apr May Jun Jul (bp) Aug Sep Oct Nov
rate 55 51 48 49 150 156 165 172 180
X 1 2 3 4 5 6 7 8 9
X2 -4 -3 -2 -1 0 1 2 3 4
X3 -4 -3 -2 -1 0 0 0 0 0
X4 0 0 0 0 0 1 2 3 4
Which of these three models is correct?
Model 1: Proc reg; model rate = X X4; run; (from Biostatistics for Animal Science)
Model 2: Proc reg; model rate = X2 X4;run;
Model 3: Proc reg; model rate = X3 X4;run; (from Draper and Smith, Applied reg analysis text book)
Model 2 output is the exact same at model 1 except the intercept parameters are more meaningful.
Model 3 is the same output as Model 2 except the slope after the break point (bp) is larger, and I don’t know why.
Accepted Solutions
- Mark as New
- Bookmark
- Subscribe
- Mute
- RSS Feed
- Permalink
- Report Inappropriate Content
I would certainly expect models 1 and 2 to give the same output, as X2 is a centered version of X1.
The slope in Model 3 after the break is larger as it is the deviation from a plateau starting in July, as opposed to an additional increase over the linear trend modeled by X2.
Steve Denham
- Mark as New
- Bookmark
- Subscribe
- Mute
- RSS Feed
- Permalink
- Report Inappropriate Content
I would certainly expect models 1 and 2 to give the same output, as X2 is a centered version of X1.
The slope in Model 3 after the break is larger as it is the deviation from a plateau starting in July, as opposed to an additional increase over the linear trend modeled by X2.
Steve Denham
- Mark as New
- Bookmark
- Subscribe
- Mute
- RSS Feed
- Permalink
- Report Inappropriate Content
Thanks, Steve.
Yes, I just figured out the slope after the break point reported by model 3 is the sum of the slope pre break point + the change in slope post break point. I was not realizing models 1 and 2 were showing the change in slope rather than the new slope.