You must take into consideration the order in which observations are
returned to the calling step when you use the SPDE engine, since the
engine often does not read a data set sequentially.  In addition, the
order in which the observations are returned by the engine may not be
consistent over multiple submissions of a specific PROC or DATA step.

Incorrect results or errors may occur if the data is not read
sequentially and the step expects the data to be returned in physical
order.  Order-dependent processes include, but are not limited to,
SAS/ETS procedures, PROC GPLOT with spline or join interpolation, the
DIF and LAG functions in the DATA step, and Durbin-Watson statistics
in PROC REG or PROC GLM.

The SPDE engine provides some options to ensure that observations are
returned in consistent physical order.  The chart below describes the
appropriate settings to use:

  When:                           Use:

  BY present, no index            BYSORT=YES LIBNAME or data set option
                                  BYNOEQUALS=NO data set option

  BY present, index exists        BYSORT=YES LIBNAME or data set option
                                  BYNOEQUALS=NO data set option
                                  IDXWHERE=YES data set option

  WHERE present, no index         THREADNUM=1 data set option

  WHERE present, index exists     THREADNUM=1 data set option
                                  BYNOEQUALS=NO data set option
                                  IDXWHERE=YES data set option

  BY and WHERE present, no index  BYSORT=YES LIBNAME or data set option
                                  BYNOEQUALS=NO data set option

  BY and WHERE present,           BYSORT=YES LIBNAME or data set option
           index exists           BYNOEQUALS=NO data set option

If neither a BY statement nor a WHERE clause is specified, the engine
performs a sequential read, thus you can use the default option
values.

Operating System and Release Information