Matching cases with controls

[PGStats]

The 5 and 7 denominators were chosen arbitrarily to correspond to the allowed mismatch range for both variables. You could use 

euclid((c.age-s.age)/3, (c.edu-s.edu)/2) as distance

instead to scale both mismatches from -1 to +1 or any other scaling that reflects the relative importance of matching age and edu.

Following the same logic, you could add dobw as a matching variable in the distance calculation with

euclid((c.age-s.age)/3, (c.edu-s.edu)/2, (c.dobw-s.dobw)/14) as distance

with the extra join condition

c.dobw between s.dobw-14 and s.dobw+14 and

The format given to distances is only used for display purposes, it has no impact on calculations.

[wajmsu]

Dear PGStats:

Referring to Proc Optnet, I am trying to use it for my original dataset with libref 9not work library) it does not work. Log says 'Work.Links.Data' not found. Please guide me.

Thanks

[PGStats]

The way to refer to a dataset is MYLIBREF.MYDATASET . If you do not specify MYLIBREF then the temporary library WORK is used. So if your LINK dataset is not in the WORK library you must prefix its name with your libref.

[wajmsu]

As I mentioend earlier, I've Links dataset in my permanent library.

While using Proc optnet data_links=links, the program takes it, probably, as coming from Work library which is actually not.

How can I use my libref which is "pe." here? Where should I write it?

Thanks

[PGStats]

Proc optnet data_links=pe.links

[wajmsu]

Thanks.

[wajmsu]

Thanks.

Going back to the creation of data Links, the controlNode becomes character and sampleNode becomes numeric variables. So, it does not work for the 'proc optnet' statement due to different in the type. How can I bring the controlNode (which is 4 digit label) to sampleNode (lenght 8, infromats 10.2 and format 10). I added 'format controlNode 4.0.' the log statement showed 'converted from character to numeric but in actuality it did not.

Here is the orginal code, (without format for controlNode, which does not work):

---

data pe.links;

set pe.matches;

where not missing(controlId);

controlNode = controlId;

do i = 1 to min(99, &controlsPerSample);

    sampleNode = sampleId + i/100;

    output;

    end;

format sampleNode 10.2;

keep sampleNode controlNode distance;

run;

/* Perform best linear assignment algorithm */

proc optnet data_links=pe.links graph_direction=directed;

data_links_var

from=sampleNode to=controlNode weight=distance;

linear_assignment out=pe.bestMatches;

run;

---

Thanks

[wajmsu]

Hi PG,

For my matching strategy, the final step is creating problem. Please see the log below and guide me how to fix it.

proc sql;
217  create table pe.allMatches as
218  select a.group, a.type, a.ctrlId, b.*
219  from pe.stackedMatches as a inner join pe.sample as b on a.id=b.id
220  union all corresponding
221  select a.group, a.type, a.ctrlId, b.*
222  from pe.stackedMatches as a inner join pe.control as b on a.id=b.id
223  order by group, type, ctrlId;
NOTE: Data file PE.SAMPLE.DATA is in a format that is native to another host, or the file encoding
      does not match the session encoding. Cross Environment Data Access will be used, which might
      require additional CPU resources and might reduce performance.
NOTE: Data file PE.CONTROL.DATA is in a format that is native to another host, or the file encoding
      does not match the session encoding. Cross Environment Data Access will be used, which might
      require additional CPU resources and might reduce performance.
ERROR: Expression using equals (=) has components that are of different data types.
ERROR: Expression using equals (=) has components that are of different data types.
224  quit

[PGStats]

Looks like your Ids are of type character in one of your datasets. You could use a.Id = input(b.Id, best.) as your join condition if b.Id is character.

[wajmsu]

Thanks PG; it did work.

[LinusH]

What I meant was that you should attach (or paste) the desired output.
What about the SQL idea?