Bug in bootstrap.func

R versions of bootstrap.func downloaded before January 2003 had a potentially serious bug. The bug occurs with small data sets when the bootstrap resample does not include all the years in the original data frame. It causes the entries in the bootstrap matrix to be shifted along one place for every year that is missing. For the rest of the matrix, index values are therefore wrongly mapped to year.

• Corrected function: bootstrap.func

The corrected version was placed in the tar file R.all.tar and linked from the main page on January 2003. For analyses conducted using the previous version, check to see whether the bug was present as follows.

• Look at the output from sp.plot. If the confidence intervals for year 1 have width 0 (that is, lower limit = upper limit = 1), then the bug has probably not arisen, or has had little effect. The confidence limits for correct output can be seen here.

• Check the matrices sp.bootind.nrep.df. The first column should have entry 1.0 in every place. For example, type:

  > cb.bootind.119.10[ , 1]
  

      [1] 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1
    [38] 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1
    [75] 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1
  [112] 1 1 1 1 1 1 1 1
  

  If the entries are not all 1, the bug has appeared.

• When bootstrapping, the bug is signalled by the following warning message:

  Warning message:
  Replacement length not a multiple of the elements to replace in matrix(...)

Apologies for any inconvenience this bug may have caused.

Rachel Fewster      28th January 2003