Table 3.
Accuracy of kit for predicting whether mean XRF soil As is above a 20 mg/kg, 30 mg/kg, and 40 mg/kg threshold for the regression method (regression on an average of 12 kit samples) compared with the Bayesian method (probabilities calculated from 12 kit samples) calibrated and tested on plots from the aman 2015 through aman 2017 growing seasons. For this comparison, we use the probability threshold for the Bayesian method that gives the same number of false negatives as given by the linear regression method.
XRF | XRF | ||||||
---|---|---|---|---|---|---|---|
≤20 mg/kg | >20 mg/kg | ≤20 mg/kg | >20 mg/kg | ||||
Kit (12-sample average) | >0.09 mg/L | 20 (19%) | 141 (85%) | Kit (Probability threshold) | >19% | 19 (18%) | 141 (85%) |
≤0.09 mg/L | 84 (81%) | 25 (15%) | ≤19% | 85 (82%) | 25 (15%) | ||
≤30 mg/kg | >30 mg/kg | <30 mg/kg | ≥30 mg/kg | ||||
>0.13 mg/L | 29 (16%) | 74 (87%) | >15% | 29 (16%) | 74 (87%) | ||
≤0.13 mg/L | 156 (84%) | 11 (13%) | ≤15% | 156 (84%) | 11 (13%) | ||
≤40 mg/kg | >40 mg/kg | <40 mg/kg | ≥40 mg/kg | ||||
>0.17 mg/L | 23 (10%) | 39 (78%) | >82% | 22 (10%) | 39 (78%) | ||
≤0.17 mg/L | 198 (90%) | 10 (22%) | ≤82% | 199 (90%) | 10 (22%) |