|
On Tue, 24 Jan 2006 02:51:28 -0800, GuidoT <cymraeg_erict@HOTMAIL.COM>
wrote:
>Hi s,
>
>Sorry, it was my confusion about what the WEEKV format does. WEEKV is
>a SAS 9 format. It gives a 1 to 1 match for input date and output week
>string (Week and day number). I'm running SAS 8.2 and I had
>(incorrectly) assumed that WEEKV was a private format that resulted in
>a week number - So 7 days go to one output string. With this
>assumption, the SQL without the CALCULATED gives a summary by week for
>each of the input records. WIth CALCULATED (or 1) it produces one
>record for each week.
>
>That is, without calculated it performs the summary statistics grouped
>by the formatted value of the group, but outputs the statistics for
>every record selected.
>
>Using GROUP BY CALCULATED DATUM_C (or GROUP BY 1) it produces the stats
>for the group and only outputs one record for each group.
>
>++ Guido
Alright, thanks for your reply.
Being practically braindead about sql myself,
I'm not going to jump into discussions on how it works! :-)
On the other hand, I'm not convinced your explanation
holds water.. (i.e. does it really matter to sql who supplied the
format?) Just consider this:
data test;
infile datalines;
input datum date9. user $;
*format datum weekv5.;
datalines;
01JAN2006 4452
05JAN2006 4457
07JAN2006 4457
11JAN2006 4392
12JAN2006 4889
14JAN2006 4400
15JAN2006 4490
17JAN2006 4901
;
run;
proc format;
value userdef
16802='05W52'
16806,16808='06W01'
16812,16813,16815,16816='06W02'
16818='06W03'
;
proc sql;
select put(datum, weekv5.) as datum_c, count(user) as n
from test
group by datum_c;
quit;
proc sql;
select put(datum, userdef.) as datum_c, count(user) as n
from test
group by datum_c;
quit;
But like I said, I have no idea really...
If I got you wrong, which is quite likely, please see
if you can examplify your point on the above data.
That way I'll probably get it..
Thanks,
s
| 
