Table 2 shows the set of variables used, the final cut values, and the rejection power (R) of each variable.
| Variable # | Cut variable | Data | MC(Sig) | R |
| 1 | ldirc(2)>120 | 0.12 | 0.86 | 7.17 |
| 2 | jkrchi(3)< 4.1 | 0.35 | 0.86 | 2.46 |
| 3 | zenith(2)-zenith(1)<13 | 0.38 | 0.92 | 2.42 |
| 4 | ndirb(2)>4 | 0.71 | 0.95 | 1.34 |
| 5 | ldirb(2)>50 | 0.81 | 0.97 | 1.20 |
| 6 | jkrchi(2)< 7.7 | 0.82 | 0.96 | 1.17 |
| 7 | ndirc(2)>9 | 0.85 | 0.98 | 1.15 |
Variable one (ldirc(2)) is the projected length of direct hits in a 15 to 75 ns window and variable five (ldirb(2)) is the same except with a stricter time window of 15 to 25 ns. Variable two (jkrchi(3)) is the likelihood parameter of the phit/nohit method of energy reconstruction. Variable three (zenith(2)-zenith(1)) is the difference in zenith angle between the maximum likelihood reconstructed fit and the first guess fit in the zenith direction. Variable four (ndirb(2)) is the number of direct hits in a -15 to 25 ns time window and the same with the seventh variable (ndirc(2)) except now in a -15 to 75 ns time window. Lastly, the sixth variable is the likelihood parameter of the maximum likelihood reconstruction. The numbers in the parenthesis indicate level of reconstruction. The line fit is (1). It is the first guess for the full minimization fit, (2). Level three (3) is the energy fit.
The variables in table 2 are sorted according to rejection power. The first three variables, ldirc(2), jkrchi(3), and zenith(2)-zenith(1) produce the greatest impact on signal to noise. The detail comparisons and discussion of these variables are shown in the next section (section 3.2.2).
An important component in this analysis procedure is the comparison of passing efficiencies between experimental data and simulated background (table 3). This is necessary to insure the accuracy of simulation. Table 3 shows significant discrepancies in the passing rates for several of the variables. The poorest agreement is observed for the likelihood parameter of the energy fit (jkrchi(3)), where the background MC predicts 1/2 the passing efficiency. This is surprising since the passing efficiencies agreed after passing the Swedish filter (table 11). This discrepancy is not understood yet. The fact the Monte Carlo passing efficiencies are less than the data, may imply that the true signal passing efficiencies are higher than what the signal Monte Carlo predicts. Table 3 shows clearly the necessity to better match Monte Carlo with experimental data. Only variable three (zenith(2)-zenith(1)) may imply more signal is being cut out than what Monte Carlo predicts.
| Variable # | Cut variable | Data | MC(Bg) |
| 1 | ldirc(2)>120 | 0.094 | 0.078 |
| 2 | jkrchi(3)< 4.1 | 0.091 | 0.045 |
| 3 | zenith(2)-zenith(1)<13 | 0.51 | 0.64 |
| 4 | ndirb(2)>4 | 0.29 | 0.21 |
| 5 | ldirb(2)>50 | 0.36 | 0.36 |
| 6 | jkrchi(2)< 7.7 | 0.28 | 0.17 |
| 7 | ndirc(2)>9 | 0.18 | 0.084 |