topic Re: Creating Dummy Variable in Statistical Procedures

Creating Dummy Variable

Mon, 22 Jun 2015 16:02:03 GMT

I have a categorical variable, I will call it City. Each City has a number between 1-1500, though only 75 actual numbers are used between this range, with no pattern. I want to create a dummy variable City_1 - City_75 for the 75 cities actually in the dataset.

Example data and dummy coded variable:

Current: Desired:

City = 1 City_1 = 1

City = 12 City_2 = 1

City = 44 City_3 = 1

City = 163 City_4 = 1

City = 199 City_5 = 1

So implicitly if City_1 = 1, then City_2,...,City_5 would all = 0 for that observation (person).

The dummy code should be a numeric value.

Thanks in advance, and please ask questions if I did not provide enough information.

Re: Creating Dummy Variable

Tom — Mon, 22 Jun 2015 16:43:21 GMT

Look at PROC GLMMOD.

Do you really need the dummy variables? Most SAS proc will automatically create them for you if you use a CLASS statement.

Re: Creating Dummy Variable

Ksharp — Tue, 23 Jun 2015 13:44:11 GMT

IML 's function DESIGN( ) . OR data step's proc transpose.

Re: Creating Dummy Variable

Wed, 24 Jun 2015 13:14:30 GMT

Thanks for the input. I ended up not recoding the variable with all of the groups (not in sequential integer order). I just recoded another variable using the following code:

DATA New;

SET Old;

ARRAY dummys {*} 3. Cat_Var_1 - Cat_Var_4;

DO i=1 TO 4;

dummys(i)=0;

END;

dummys(Cat_Var) = 1;

DROP i;

RUN;

Re: Creating Dummy Variable

RW9 — Wed, 24 Jun 2015 13:52:35 GMT

Sorry, I am still not seeing why you need the variables. In your first example, if you create 1500 empty variables, and then fill only 75 you are both wasting storage space and making further processing more difficult. Also, by group processing and class processing is far quicker than multiple calls with different variables. Just create a categorising variable and use by group processing:

your_variables city [The grouping var]

... CITY5

... CITY205

... CITY1000

Then you only have rows where needed and can do:

proc xyz;

group by city.

Re: Creating Dummy Variable

arame — Fri, 26 Jun 2015 10:54:23 GMT

I think you should recode your variable like this :

if City = 44 then City = 3

else if City = 163 City = 4 etc.

after that you can use the proc transreg :

proc transreg data=a(keep = city) noprint design ;

model class (city /zero=none) ;

output out=a_dummicode(drop=_name_ _type_ intercept) ;

run ;