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
- /
- Analytics
- /
- Stat Procs
- /
- How to include error of input variables in the reg...

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

02-09-2017 07:56 PM

I am using Proc SGPLOT to create a regression plot that charts 'occurances of condition x' (yaxis) against year of admission (xaxis) and I know how to include the 95% confidence intervals for that regression line (like this https://support.sas.com/kb/40504).

However my values for 'occurances of condition x' are estimates and each have an associated standard error. Eg for 2004, occurances could be 34.9 with stddev of 4.3 but 2005 could be 55 with a stddev of 0.3. Is there anyway to incorporate this error into my regression analysis. If not, what is the reason it cannot be done (ie: is it because SAS doesn't have the capability or is it something flaw in my understand of statistics).

Thanks

Jon

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

02-09-2017 08:05 PM

Regression models aren't designed to handle the information AFAIK. There maybe more complex techniques that can, but one quick workaround is to do a simulation to determine ranges of intervals.

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

02-09-2017 08:28 PM

Any idea how to do that?

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

02-09-2017 11:31 PM

Look at the documentation of the WEIGHT statement for proc reg. This is most likely the way to go. Define* w = 1 / stdErr**2*; and use *w* in the weight statement.

PG