Running hedonic house price regression

[Sean_OConnor]

Hi All,

I'm trying to replicate some work a colleague has done but using some different techniques.

proc reg data=RegA&i. outest=StatsA&i. RSQUARE noprint;
	model log_price = &Variables. qrt02-qrt04;
	output out=RegB&i. (keep=log_price
		&Variables. qrt02-qrt04 mtch lev cd fv res r) 
		residual=res rstudent=r h=lev cookd=cd predicted=fv;
run;

Above is an example of some code for a hedonic regression for house prices.

I would like to build on this, but one of my issues with proc reg  is it forces me to code any non numeric variables into categories.

I would like to use either proc robust reg or proc glm for my model instead as I can specify a class statement. One of my issues is that I need to calculate cooks distance in order to take out high leverage points and rerun the regression. I can't seem to find this option in either package.

Would anyone be able to provide some insight how I could output the same kind of results for the code above with either proc robust reg or proc glm.

Many thanks,

Sean  

[Reeza]

What if you use PROC GLMMOD to create the dummy variables for you automatically and then use PROC REG?

Here's how you can use PROC GLMMOD to create those dummy variables for you and then you can use the data in PROC REG:

https://communities.sas.com/t5/SAS-Communities-Library/How-to-create-dummy-variables-Categorical-Var...

---

@Sean_OConnor wrote:

Hi All,

I'm trying to replicate some work a colleague has done but using some different techniques.

proc reg data=RegA&i. outest=StatsA&i. RSQUARE noprint;
	model log_price = &Variables. qrt02-qrt04;
	output out=RegB&i. (keep=log_price
		&Variables. qrt02-qrt04 mtch lev cd fv res r) 
		residual=res rstudent=r h=lev cookd=cd predicted=fv;
run;

Above is an example of some code for a hedonic regression for house prices.

I would like to build on this, but one of my issues with proc reg  is it forces me to code any non numeric variables into categories.

I would like to use either proc robust reg or proc glm for my model instead as I can specify a class statement. One of my issues is that I need to calculate cooks distance in order to take out high leverage points and rerun the regression. I can't seem to find this option in either package.

Would anyone be able to provide some insight how I could output the same kind of results for the code above with either proc robust reg or proc glm.

Many thanks,

Sean  

---

[PaigeMiller]

Another option is PROC PLS, which does have a CLASS statement, and produces an output statistic called TSQUARE, which is similar in its idea to Cook's D, in that TSQUARE will identify points that are extreme in the x-directions. A difference between TSQUARE and Cook's D is that if an x-variable is not a good predictor, it is weighted very lightly in TSQUARE, and the variables that are a good predictor are weighted heavily; whereas I think Cook's D weights all the x-variables equally.