proccontents@GMAIL.COM wrote back:
>On 1/8/07, David L Cassell <firstname.lastname@example.org> wrote:
>>peter.crawford@BLUEYONDER.CO.UK sagely replied:
>> >there is (or may be) an effect, created by excluding empty data
>> >sets from a set statement.
>> >The order of variables in the output dataset (and their internal
>> >lengths) is determined in the "order" they are discovered by the
>> >compiler, among the contributing data sets of the data step. Thus
>> >empty data sets may disturb variable order, data lengths and
>> >crucially, data type.
>> >Rather than consider observations, I would be looking to ensure
>> >that all variable contributions have consistent data types, and
>> >that the longest lengths get predefined to ensure no loss of
>> >But , that isn't the point the original poster was addressing,
>> >so perhaps I should keep my side-issues to myself ;-)
>> >Happy New Year
>>I was thinking about some of those side-issues myself, so I
>>think you are right to bring them up.
>>I have not seen any sign that the Original Poster was concerned
>>about such issues, but that does not mean that they are non-existent.
>>David L. Cassell
>I do not have any knowledge of the internal lengths but the way I did is
>took proc contents of all the datasets then using nobs variables I found
>if the dataset is empty (nobs >0 ) then captured the memname's in a macro
>using proc sql and then set it. So I was thinking is that round about way
>is there better way. I am sorry for the late reply .
But are the data similar across data sets? I mean, do the data
sets have the same variables, which all have the same types and
lengths and formats?
If so, then you probably do not have to knock yourself out to
append these together.
David L. Cassell
3115 NW Norwood Pl.
Corvallis OR 97330
From photos to predictions, The MSN Entertainment Guide to Golden Globes has
it all. http://tv.msn.com/tv/globes2007/?icid=nctagline1