Unlike SPSS, where value labels are an attribute of the variable, in SAS a variable may or may not have a permanent association with a set of labels for its values. A format may be assigned to a variable in a DATA step or in SQL or in PROC DATASETS, or as needed in a PROC. As you noted, this merely assigns the name of the associated format as a property of the variable. The user defined formats are stored separately - not in the dataset itself. To associate the values from the format with the variables that have assigned formats one could join DICTIONARY.COLUMNS with a cntlout dataset on the format name. In the cntlout dataset the format name is stored without the period or field width(s). In DICTIONARY.COLUMNS it is stored with the period and any field width information. The following snippet would find the assigned user formats for the dataset &dataset in the library &lib

proc sql; create table myColumns as select *, prxchange('s/\d*\.\d*//',1,format) as fmtName from dictionary.columns where libname=upcase("&lib") and memname=upcase("&dataset") order by name;

create table userFormats as select myFormats.* from myformats,myColumns where myColumns.fmtName=myFormats.fmtname and myFormats.type in ("N","C");

I expect it is common practice, though, for SAS programmers to not permanently associate any user format with a variable, especially in the case where there are several forms of labels to be associated with the values of a variable. Some output, like axis labels for a chart, might need short forms of labels while other output, like row headings for a table, might use a longer form. Multiple formats may also be used to group values in alternative ways. Another inducement to not create permanent associations is the fact that opening a dataset that has user formats assigned when the formats are not accessible produces an error unless the NOFMTERR option is in effect.

All of this means that archiving all of the associations between user formats and SAS variables will require metadata about the metadata, such as a table listing the associations between variables and formats.

Larry Hoyle Associate Scientist Institute for Policy & Social Research University of Kansas Blake Hall 1541 Lilac Lane Lawrence, KS 66044-3177

http://www.ipsr.ku.edu

>> -----Original Message----- Larry,

Thanks for your response.

With libname XML engine I was not aware that XMLTYPE is actually referencing also a tag set. So the only difference between XMLTYPE and TAGSET seems to be this: some default tag sets are available with XMLTYPE, additional and user-defined tag sets can be defined by TAGSET.

proc format cntlout=myformats; Yes, this seems to be the appropriate approach to have a basis for exporting user-defined formats (my description with "output dataset from PROC FORMAT" was heading to this).

An approach by PROC SQL and the DICTIONARY tables seems to provide only the names of the user-defined formats, but not the formats themselves (like "category labels"). I don't see a possibility to get the details of the user-defined formats by PROC SQL.

Achim