What are predictor units in proc reg?

noetsi

Fortran must die
I ran proc reg code [without changing defaults] for an interval DV. One of the predictors was cost in dollars and cents. The slope then is the change in the Y for a one unit change in X. But I am not sure what one unit is in SAS when you have a predictor measured in dollars and cents.

I am guessing it is one dollar so the results reflect the change in the DV for a one dollar change in cost.

hlsmith

Less is more. Stay pure. Stay poor.
Did the outputted estimates make sense? Do you really need cents or can you round the observe?

noetsi

Fortran must die
I don't really know what makes sense is this case - no theory and no one ran this before that I know of. The parameter was about 7 cents so I am guessing that this means 7 cents per dollar - but I was hoping SAS had some automated feature that showed what a "unit" was.

hlsmith

Less is more. Stay pure. Stay poor.
When it comes to a decimal I am unsure. Well you can always set the unit and see if you get the same result. Most regression procedure allows for the Unit statement. I believe it is as easy as

Code:
PROC REG;
MODEL y = x;
UNIT = 0.01;
RUN;

noetsi

Fortran must die
How would you change the unit if you have multiple predictors and you only want to change one?

I spent a good deal of time looking for this in SAS documentation today without finding it.

Dason

It's not quite clear to me what you're saying. Are you saying your predictor took values like 5.67 which represents five dollars and sixty seven cents? Betas always reflect a change in 1 unit. So what's a change in one unit for this? It's a difference of a dollar. Why would SAS think anything else?

How would you change the unit if you have multiple predictors and you only want to change one?
Just change the predictor directly in the data

hlsmith

Less is more. Stay pure. Stay poor.
I believe the following changes variable X5.

Code:
[COLOR=blue][FONT=Courier New]UNIT [/FONT][/COLOR][FONT=Calibri] X1 = 5;[/FONT]
Perhaps, place different values in for 5. If you are still confused, you can probably back up and just run a simple regression and see what makes sense.

P.S., just saw Dason's post and agree that SAS is probably using unit=1 as default. I would imagine that this is a way to make SAS tell you this, but put 1 in the statement above and confirm nothing changes.

noetsi

Fortran must die
It's not quite clear to me what you're saying. Are you saying your predictor took values like 5.67 which represents five dollars and sixty seven cents? Betas always reflect a change in 1 unit. So what's a change in one unit for this? It's a difference of a dollar. Why would SAS think anything else?

Just change the predictor directly in the data
I actually was not sure what SAS interpreted a unit to be. If I understand your comment, its the value immediately to the left of the decimal place

Amazing what I never think of