PM78778: CROSS-CORRELATION FUNCTION (CCF) MISALIGNS CROSS-CORRELATION TABLE IF EXCESSIVE NUMBER OF LAGS REQUESTED

APAR status

• Closed as program error.

Error description

• You have used the CCF (Cross-correlation function) procedure in
  SPSS Statistics to find the cross-correlation between a pair of
  variables in a small data set. You have requested a value for
  the maximum number of lags to display (/MXCROSS) that is equal
  to the number of time points for the 2 variables (which we'll
  call K),
  i.e., the number of records in the file. You have also run
  sequence plots for the 2 variables in one graph and it appears
  that the maximum cross-correlation will be at lag 0. However,
  the Cross-correlation table in the CCF output indicates that the
  largest Cross-correlation is at lag - 2. Also, the
  Cross-correlations for the last 4 positive lags are all 0.0. If
  you rerun the CCF with a number of maximum lags that is much
  smaller than the number of time points, the CCF table seems
  properly aligned, with the largest cross-correlation (the value
  that was at lag -2 for /MXCROSS=k) at the position suggested by
  the Sequence plots and without the Cross-correlation values of 0
  near the bottom of the table (i.e. at the largest positive
  lags).
  
  You have encountered a bug in the CCF procedure that occurs when
  you request cross-correlations for more lags
  than can be computed for the data. You have K valid cases on the
  variables involved and requested K lags. There are no cases that
  are K time points apart, so there are no x,y pairs with a lag of
  -K or +K.
  There is only one x,y pair with a lag of +(K-1) and one x,y pair
  with a lag of -(K-1), but a correlation
  cannot be computed with 1 case (as the standard deviation of
  each of x and y equals 0, so the correlation is undefined).
  The CCF at lags -K,-(K-1),(K-1) and K should be system missing
  and/or you should get a warning that those correlations cannot
  be calculated. You should not get a misaligned Cross-Correlation
  table.
  
  A program defect has been filed for the misalignment of the
  Cross-Correlation table when the number of lags requested is
  greater than the data will support. You can avoid the defect by
  requesting a maximum number of lags that is much smaller than
  the number of rows in the sequence. Keep in mind that
  cross-correlations for lags that are even close to K (but
  smaller than (K-2)) will be based on a very small number of
  cases.
  

Local fix

Problem summary

• ****************************************************************
  * USERS AFFECTED:                                              *
  * User working with cross-correlation function                 *
  ****************************************************************
  * PROBLEM DESCRIPTION:                                         *
  * You have used the CCF (Cross-correlation function) procedure *
  * in                                                           *
  * SPSS Statistics to find the cross-correlation between a pair *
  * of                                                           *
  * variables in a small data set. You have requested a value    *
  * for                                                          *
  * the maximum number of lags to display (/MXCROSS) that is     *
  * equal                                                        *
  * to the number of time points for the 2 variables (which      *
  * we'll                                                        *
  * call K),                                                     *
  * i.e., the number of records in the file. You have also run   *
  * sequence plots for the 2 variables in one graph and it       *
  * appears                                                      *
  * that the maximum cross-correlation will be at lag 0.         *
  * However,                                                     *
  * the Cross-correlation table in the CCF output indicates that *
  * the                                                          *
  * largest Cross-correlation is at lag - 2. Also, the           *
  * Cross-correlations for the last 4 positive lags are all 0.0. *
  * If                                                           *
  * you rerun the CCF with a number of maximum lags that is much *
  * smaller than the number of time points, the CCF table seems  *
  * properly aligned, with the largest cross-correlation (the    *
  * value                                                        *
  * that was at lag -2 for /MXCROSS=k) at the position suggested *
  * by                                                           *
  * the Sequence plots and without the Cross-correlation values  *
  * of 0                                                         *
  * near the bottom of the table (i.e. at the largest positive   *
  * lags).                                                       *
  *                                                              *
  * You have encountered a bug in the CCF procedure that occurs  *
  * when                                                         *
  * you request cross-correlations for more lags                 *
  * than can be computed for the data. You have K valid cases on *
  * the                                                          *
  * variables involved and requested K lags. There are no cases  *
  * that                                                         *
  * are K time points apart, so there are no x,y pairs with a    *
  * lag of                                                       *
  * -K or +K.                                                    *
  * There is only one x,y pair with a lag of +(K-1) and one x,y  *
  * pair                                                         *
  * with a lag of -(K-1), but a correlation                      *
  * cannot be computed with 1 case (as the standard deviation of *
  * each of x and y equals 0, so the correlation is undefined).  *
  * The CCF at lags -K,-(K-1),(K-1) and K should be system       *
  * missing                                                      *
  * and/or you should get a warning that those correlations      *
  * cannot                                                       *
  * be calculated. You should not get a misaligned               *
  * Cross-Correlation                                            *
  * table.                                                       *
  ****************************************************************
  * RECOMMENDATION:                                              *
  * This issue has been addressed in Statistics 21 Fix Pack 1.   *
  * Please upgrade to this version.                              *
  ****************************************************************
  

Problem conclusion

• This issue has been addressed in Statistics 21 Fix Pack 1.
  Please upgrade to this version.
  

Temporary fix

Comments

APAR Information

• APAR is sysrouted FROM one or more of the following:

• APAR is sysrouted TO one or more of the following:

Fix information

• Fixed component name

  SPSS STATISTICS

• Fixed component ID

  5725A54ST

Applicable component levels

• RL00 PSN

     UP

• RL00 PSY

     UP