topic Re: Create dummy variable in New SAS User

Create dummy variable

Tobiasp94 — Mon, 25 Feb 2019 21:56:57 GMT

I need help changing my year variable into a dummy variable, so i can run this macro.
The variable year variable is currently a numeric variable (2010-2017).

Thank you

Re: Create dummy variable

Astounding — Mon, 25 Feb 2019 22:03:04 GMT

Most likely, you will end up using YEAR as is. Analytical procedures will normally let you use this statement:

class year;

That will create dummy variables as needed.

If you really have to have a dummy variable, one dummy variable is not enough. You have 8 values for YEAR. You will need 7 dummy variables (rarely 8, depending on your intentions). Does that fit your plan?

Re: Create dummy variable

Reeza — Mon, 25 Feb 2019 22:03:36 GMT

Here's one way, data step methods:


*declare an array;
array _years(2012:2017) year2012-year2017;

*set all values to 0 initially;
do i=lbound(_years) to hbound(_years);
    _years(i) = 0;
end;

*assign year to 1;
_years(year) = 1;

and here's another, more automated:

https://communities.sas.com/t5/SAS-Communities-Library/How-to-create-dummy-variables-Categorical-Variables/ta-p/308484

PS please post your code as text not as an image. An image means if we wanted to modify or include your code we have to type it out.

@Tobiasp94 wrote:

I need help changing my year variable into a dummy variable, so i can run this macro.
The variable year variable is currently a numeric variable (2010-2017).

Thank you

Re: Create dummy variable

Tobiasp94 — Mon, 25 Feb 2019 22:07:43 GMT

How do I use the class statement in the macro?
If i have to create a dumme wouldn´t it be
2010=0
2011=1
........
2017= 8 ?

Re: Create dummy variable

Astounding — Mon, 25 Feb 2019 22:10:37 GMT

Assuming you successfully did something along these lines:

2010=0

2011=1

2012=2

...

2017=7

You would be no better off than if you had done nothing. The original variable YEAR with values 2010-2017 could have been used any place these new values could be used.

If you want advice about using a set of values within a macro, you might need to post the macro.

Re: Create dummy variable

Tobiasp94 — Mon, 25 Feb 2019 22:20:37 GMT

I can se that this creates a dummies.
But i can´t get it to work in my model.
The idea is that year will take a value of time=1 time=2 time=3 etc.

Thank you for your quick response

%let yvar=COPD;

%let xvars=Healthexpenditure Share_18_66 Share67_ Healthcosts Healthcosts_cofinancing Costelderlyadults Incomefromtaxes Costsports Socioeconomicindex Vocationaleducation Furthereducation Westerns Nonwesterns Dailysmokers Heavysmokers Physicalactivity Unhealthydiet Alcohol Familycontact;

%let wset=wnykom94;

%let datset=final;

%let region=Kkode;

%let time=year;

libname wcat '/folders/myfolders/sasuser.v94';

data set1; set work.&datset;
data sety; set set1; by year;if first.year;
yearch=put(year,4.);

proc iml;
print 'SUR model, assuming equal coefficients across time periods',,;
use set1;
yname={&yvar};
read all var yname into y;
xname={&xvars};
incxname={'Constant'}||xname;
read all var xname into x0;
use sety;
read all var {'year'} into year;
read all var {'yearch'} into yearch;

Re: Create dummy variable

Astounding — Mon, 25 Feb 2019 22:29:44 GMT

Not familiar with IML, so here is an educated guess. If you create a new variable in this way:

Time = year - 2010;

Couldn't you just use Time instead of Year within IML?

Re: Create dummy variable

Tobiasp94 — Mon, 25 Feb 2019 22:49:00 GMT

Im really not that experienced in SAS, so i have difficulties creating a new variable and implementing it into my macro.

Re: Create dummy variable

PaigeMiller — Mon, 25 Feb 2019 22:49:27 GMT

You are doing some type of modeling in IML.

Unless it is a modeling technique not available elsewhere in SAS, don't do it in IML. Save yourself a lot of trouble, use the modeling PROCs that SAS has built and debugged, most of which use the CLASS statement so you don't have to create your own dummy variables.

Re: Create dummy variable

Astounding — Mon, 25 Feb 2019 23:02:04 GMT

In that case you are probably best off removing all the macro language. Any place you see an ampersand, get rid of it. For example, replace &yvar with an actual variable name. That way, you stand a fighting chance of understanding your final program.

Re: Create dummy variable

Tobiasp94 — Mon, 25 Feb 2019 23:05:09 GMT

How would you suggest me to use the class statement. I have created a balanced dataset before i created the macro. I could easily use the class statement in to create a new data sheet before i run the macro with new values?
I would like if you could give me the specific kode since im not use to work in SAS.

Re: Create dummy variable

Tobiasp94 — Mon, 25 Feb 2019 23:07:31 GMT

Im sorry i can´t do that. I have an assignment due to tomorrow and there is no way i can get it done without the macro.
What if a implemented the class year statement in my dataset before i implemente the dataset in my macro?
What would the specific code be then?

Re: Create dummy variable

Reeza — Mon, 25 Feb 2019 23:10:58 GMT

Have you tried using YEAR just as is? Has that caused an issue? You haven't posted any code so far that seems to require dummy variables, so it's hard to understand what you need here. Or why.

Re: Create dummy variable

Tobiasp94 — Mon, 25 Feb 2019 23:15:54 GMT

Hi Reeza.
I have tried with the variable year since i have data for 2010-2017 and the value is numeric i had som terrible results:
My goal is to create a SUR estimation model assuming equal coefficient across time periods.

%let yvar=COPD;
%let xvars=Healthexpenditure Share_18_66 Share67_ Healthcosts Healthcosts_cofinancing Costelderlyadults Incomefromtaxes Costsports Socioeconomicindex Vocationaleducation Furthereducation Westerns Nonwesterns Dailysmokers Heavysmokers Physicalactivity Unhealthydiet Alcohol Familycontact;
%let wset=wnykom94;
%let datset=final;
%let region=Kkode;
%let time=year;

libname wcat '/folders/myfolders/sasuser.v94';

data set1; set work.&datset;
data sety; set set1; by year;if first.year;
yearch=put(year,4.);

proc iml;
print 'SUR model, assuming equal coefficients across time periods',,;
use set1;
yname={&yvar};
read all var yname into y;
xname={&xvars};
incxname={'Constant'}||xname;
read all var xname into x0;
use sety;
read all var {'year'} into year;
read all var {'yearch'} into yearch;
use wcat.&wset;
read all var _NUM_ into w;
n=ncol(w);
do i=1 to n;
w[i,]=w[i,]/sum(w[i,]);
end;
n=nrow(w);
nt=nrow(y);
t=nt/n;
xc=j(nt,1,1)||x0;
k=ncol(xc); tk=t*k;
***********OLS estimation assuming common coefficients**************;

bols=inv(t(xc)*xc)*t(xc)*y;
e=y-xc*bols;

***SUR estimate assuming time-specific coefficients***************;
b=bols;
goon=1;
iter=0;maxiter=1000;
do while (goon=1);
sigma=j(t,t,0);
do i=1 to t;
do j=i to t;
s=t(e[(i-1)*n+1:i*n])*e[(j-1)*n+1:j*n]/n;
sigma[i,j]=s; sigma[j,i]=s;
end;
end;
sigmai=inv(sigma);
kc=ncol(xc);
del1=j(kc,kc,0); del2=j(kc,1,0);
do i=1 to t;
do j=1 to t;
del1=del1+sigmai[i,j]*t(xc[(i-1)*n+1:i*n,])*xc[(j-1)*n+1:j*n,];
del2=del2+sigmai[i,j]*t(xc[(i-1)*n+1:i*n,])*y[(j-1)*n+1:j*n,];
end;
end;
del1=inv(del1);
V=del1;
b1=del1*del2;
e=y-xc*b1;
dif=max(abs(b-b1));iter=iter+1;
b=b1;
if (dif<0.0001 | iter>maxiter) then goon=0;
end;
print iter dif;
vsur=v; bsur=b;
ypred=xc*bsur;
predset=y||ypred;
create predset from predset[colname={'y' 'ypred'}];append from predset;

stdbsur=sqrt(vecdiag(Vsur));
tsur=bsur/stdbsur;
esur=y-xc*bsur;
psur=2*(1-probnorm(abs(tsur)));
resultsur= bsur||stdbsur||psur;
resultsurname={'Estimate' 'std' 'p-value'};
resultsur=round(resultsur*1000)/1000;
print 'SUR estimation results',,yname,,
resultsur[rowname=incxname colname=resultsurname];

Re: Create dummy variable

Reeza — Mon, 25 Feb 2019 23:18:59 GMT

What does SUR stand for again? OLS is linear, which means my original answer was likely correct.

Re: Create dummy variable

Tobiasp94 — Mon, 25 Feb 2019 23:20:34 GMT

Seemingly unrelated regression.

Re: Create dummy variable

Tobiasp94 — Mon, 25 Feb 2019 23:39:53 GMT

It seems like this code has changed my entire dataset. How do i convert it back to numeric. You suggested that the variable should stay numeric through the regression

Re: Create dummy variable

Reeza — Tue, 26 Feb 2019 03:04:41 GMT

What code and/or where? You never mentioned IML initially so my code isn't IML. Just remove that section and you need to create it using IML functions. Unfortunately I'm not familiar with IML.

Re: Create dummy variable

PaigeMiller — Tue, 26 Feb 2019 13:11:23 GMT

@Tobiasp94 wrote:
Seemingly unrelated regression.

You don't have to program this yourself in PROC IML in SAS.

Here's a hint: type SAS seemingly unrelated regression into any search engine. It's already been programmed into SAS, so you don't have to write IML code to do this.

Apparently, you will need dummy variables since there is no CLASS statement for this modeling procedure. Dummy variables are easily created by PROC GLMMOD, and by several other procedures in SAS. Examples are at the links.