The subtraction of accidental tags has a large effect on the
statistical errors in all measured spectra. In addition to reducing
the total number of events via the coincidence requirement, the errors
no longer follow the 
formula after subtraction. Precise
error estimates may be obtained by turning on explicit computation of
errors in the histograms when they are created and then using the
weighted filling technique described in the previous section. It
turns out that the ratio of true/accidental coincidences is on the
order of 1, so the size of a sample can still give an indication of
the expected statistical error. The statistical errors of a subtracted
spectrum with 
events and a signal/background ratio of 1 are roughly
equivalent to those of a sample of 
pure signal events, provided
that the spectrum of signal and background are not too different in
shape. For the actual analysis the correct error treatment must be
applied, but in advance of that some idea of the statistical quality
of the Radphi sample can be obtained with this rule of thumb. In
Table 3 is shown the total count of events of each
multiplicity are shown that remains after the cuts described above
are applied to select fully-contained forward neutral triggers, and
the tagging subtraction has been performed.
| multiplicity | count |
| 2 | 3,613,000 |
| 3 | 2,006,000 |
| 4 | 1,277,000 |
| 5 | 494,000 |
| 6 | 177,000 |
| 7 | 45,000 |