You might want to check the ANOBS attribute first, to find out whether SAS knows the number of observations. If all you want to know is whether the dataset has 0 obs or non-zero obs, you could use the ANY attribute instead of NLOBS (I haven't checked to see whether deleted records show up).

Also, in version 7, the NLOBSF attribute is available; it takes OBS=, FIRSTOBS=, and WHERE= into account when counting observations.

-- JackHamilton@FirstHealth.com Development Manager, METRICS First Health, West Sacramento, California USA

>>> GORRELP1 <GORRELP1@WESTAT.COM> 03Jun1999 5:37 AM >>> David,

An efficient way to get the number of obs in a SAS dataset is as follows:

%let id_num = %sysfunc(open(SOME_DSN)); %let numobs = %sysfunc(attrn(&id_num,nlobs)); %let clz_dsn = %sysfunc(close(&id_num));

Now you have a macrovariable NUMOBS you can reference in conditionals or whereever you need. Also, NLOBS contrasts with NOBS in referring only to observations not marked for deletion. See the Online Help for ATTRN.

Paul Gorrell gorrelp1@westat.com

____________________Reply Separator____________________ Subject: Re: missing obs data sets question Author: David Johnson <dkvjcns@IBM.NET> Date: 06/03/1999 12:22 AM

Jacks caution is quite timely. We have had similar difficulties recently when reading the dictionary.

I also discovered that reading the TABLES dictionary was taking some 15 seconds in our application, which was an unconscionable delay in an interactive user session. Here is a comparison I ran off site:

230 Proc SQL STimer; NOTE: The SQL Statement used 0.0 seconds.

231 Create Table TEMPCONT as 232 Select LIBNAME, MEMNAME, NOBS from 233 DICTIONARY.TABLES 234 Where Substr( MEMNAME,1,4) Eq 'LINE'; ERROR: Libname USAGE is not assigned. NOTE: Table WORK.TEMPCONT created, with 25 rows and 3 columns.

NOTE: The SQL Statement used 3.41 seconds.

235 Quit; NOTE: The PROCEDURE SQL used 0.05 seconds.

Granted, the lack of the USAGE library may affect the outcome, but nowhere near as much as 24 librefs for the application and most of these assigned on a LAN through UNCs. As if this wasn't slow enough, it is even slower on NT than on Windows 95.

The following is my preferred method, longer hand, easier for newer programmers less experienced in SQL to understand and considerably faster.

236 237 Proc Datasets Lib = WORK NoList; 238 Contents Data = _ALL_ NoPrint Out=WORK.TEMPCONT; 239 Quit;

NOTE: The data set WORK.TEMPCONT has 53 observations and 35 variables. NOTE: The PROCEDURE DATASETS used 0.38 seconds.

240 241 Data TEMPCONT; 242 Set TEMPCONT(Keep = LIBNAME MEMNAME NOBS 243 Where = ( MEMNAME Eq: 'LINE') ); 244 By LIBNAME MEMNAME; 245 If First.MEMNAME; 246 Run;

NOTE: The data set WORK.TEMPCONT has 25 observations and 3 variables. NOTE: The DATA statement used 0.11 seconds.

Now you have a dataset containing one entry per member with its count. One caution, the use of 'OUT=WORK.TEMPCONT' as a fully specified destination is deliberate. Using a single level name does not necessarily write WORK.TEMPCONT, but in fact writes TEMPCONT into the library specified on the Datasets LIB= parameter. A small trap there, especially if you are reading a protected drive.

regards

David

----- Original Message ----- From: Jack Hamilton <JackHamilton@FIRSTHEALTH.COM> To: <SAS-L@AKH-WIEN.AC.AT> Sent: 02 June 1999 20:59 Subject: Re: missing obs data sets question

> Yes, but keep in mind that there are circumstances under which NOBS will not return the actual number of observations in a dataset. It doesn't happen often, but it does happen. The main things to watch out for are views and datasets with deleted observations. > > > > -- > JackHamilton@FirstHealth.com > Development Manager, METRICS > First Health, West Sacramento, California USA > > > >>> "Newhouse, Russell" <Russell.Newhouse@WL.COM> 02Jun1999 12:42 PM >>> > One possible method could be to look at the DICTIONARY.TABLES table that SAS > creates. You can use a SQL query to look at the value of NOBS (number of > observations) for the appropriate MEMNAME (data set name). > > Russell Newhouse > Parke-Davis Pharmaceutical Research > 2800 Plymouth Road > Ann Arbor, MI 48188 > (734) 622-5006 > Russell.Newhouse@wl.com > > -----Original Message----- > From: Doug Conrad [mailto:DConrad@HER-CHER.ORG] > Sent: Wednesday, June 02, 1999 3:53 PM > To: SAS-L@UGA.CC.UGA.EDU > Subject: missing obs data sets question > > > I have a range of small data sets sequentially named line1--line25. > These are generally 1 line datasets that go into a proc report. > My problem: sometime zero obs will be returned in the datasets > thereby dropping them as a line item in the proc report. I want all lines > that have zero/missing to just produce a "0.00". > What I would like to do is go through a macro loop and check if each > dataset is empty, and if it is, then give it one observation that will > then produce a corresponding line in proc report. > > %syserror doesn't look right. I've set the program up so that the files > 'exist' but am not quite sure how to macro code this. > > Any ideas? > Thanks > doug conrad > dconrad@her-cher.org >