how to compile contious visit dates

[mkeintz]

The "then start_date=date" is correct.

The "if first.client and first.date" condition would be wrong, because:

1. There is no BY statement to even generate the first.client and first.date dummy variables
2. Even if there were, the condition ("first.client and first.date") would select EVERY observation from a daily file.
   1. But a "first.client" alone would work, although it's overkill.

What I meant was

  "if _N_=1 then start_date=date;"

placed in front of the do group.  The automatic variable _N_ counts the current iteration number in the DATA step.  In this sample _N_ is tracking the number of MERGE actions  (once for each incoming observation).  So _N_=1 is the first observation.

. It selects the first record for a givne client/date - which means every record in your daily file.

Now "if first.client" would work, but is a sligh

[Astounding]

As a rule, I find it easier to handle this type of problem in two steps instead of one.  First, find the groupings:

data temp;

set have;

by client;

if first.client or dif(date) ne 1 then group + 1;

run;

Then find the endpoints:

proc summary data=temp;

   by client group;

   var date;

   output out=want (keep=client start_date end_date) min=start_date max=end_date;

run;

************ EDITED:

sorry, didn't notice a very similar solution.  However, notice that this condition is not sufficient:

date-lagdate>1

If one client started earlier than the previous client, the value for that expression could be negative.