Decreasing values of operations in compcost increase gedscore? - sas

Question

I'm afraid I'm running across the following:
Method 1:
proc sql;
create table as
...
compged(a.plan_id, b.plan_id,&maxscore.,'iL') as gedscore
from view_a a, view_b b
where a.state = b.state and calculated gedscore < &maxscore.
order by calculated gedscore;
This works, it's all fine and dandy, but I would like to adjust my results slightly with compcost. So I adopt Method 2:
proc sql;
create view tempview as select
...
from view_a a, view_b b
where a.state = b.state;
quit;
data modified_gedscore
set tempview;
if _N_ = 1 then call compcost('delete=',10,'truncate=',10);
gedscore = compged(el_plan, clms_plan,&maxscore.,'iL');
if gedscore < &maxscore.;
run;
There's a bit more to it, but I've tried to isolate the relevant bits. I have tried to decrease the cost of the operations delete and truncate (it makes sense given the data I'm working with and what I'm trying to accomplish). My expected result would be due to delete and truncate operations having a lower cost, more observations would have a gedscore < &maxscore. However, I'm afraid I am seeing the following: the call compcost is actually dramatically decreasing the number of observations I see. Do I have a basic misunderstanding as to how call compcost works? If the above is incorrect, how would I adjust compged to have deletion of characters be more likely to fall under the maxscore threshold?
Edit: Also, I understand that the different structuring of the two methods would raise the possibility of something other than call compcost causing the unexpected results, but if I simply comment out the call compcost line I get results equivalent to that in Method 1. So, nope.
Edit2: sample data. First observation is equivalent (0). Second yields higher gedscore under method 2 than method 1, even though the compcost of delete and truncate has been lowered, with no other changes.
data sample_data;
input state1 $ plan1 $ plan2 $;
datalines;
ID DENTAL DENTAL
GA GBHC GBCH
;
Edit3: I think I may have found the problem. It appears that the default compged costs (here) are different from the default compcost costs (here). When compcost is called, all operations not specified are set to the compcost defaults, which are usually higher. If anybody feels like confirming, feel free.
Thanks for your help

Answer 1

The issue is that COMPGED is not using the SWAP cost, but instead only using DELETE and INSERT (the latter of which costs 100). That's because of how CALL COMPCOST works; for some reason (that makes little sense to me), CALL COMPCOST's default values are not equal to COMPGED's default values - and it inserts a default value into every other operation that you do not specify.
In order to make this work, it looks like you'll have to specify a value for everything that you want it to use, in particular, APPEND, BLANK, PUNCTUATION, SINGLE, SWAP, and TRUNCATE (the latter of which you do specify already). From the doc, as of 9.2, the defaults were 50,10,30,20,20,10 for COMPGED for those.
In your example:
data sample_data;
input state1 $ plan1 $ plan2 $;
call compcost('del=',10,'truncate=',10,'swap=',20);
compged_1 = compged(plan1,plan2,'il');
put compged_1=;
datalines;
ID DENTAL DENTAL
GA GBHC GBCH
;
run;
Now returns 20 instead of 110.