Crop Hail Evaluation
(Smith et al. 1987). This study defined a control area comprised of upwind eastern Montana counties adjacent to North Dakota. Hail insurance data were analyzed for the seeded areas (target) and unseeded areas (control) showing a significant reduction of losses in the target with strong statistical significance. The significance level is an indicator of the degree of confidence which may be accorded the result of a statistical test of two populations (seeded vs unseeded, for example). The lower the numerical significance, the greater the confidence. For example, a significance level of 0.08 indicates a 92% chance that the two populations are really different, i.e, that the observed effect is real.
Claims recorded for a lengthy pre-project period from this control area were compared to those for the modern-day NDCMP target to first establish the historical relationship. Crop-hail losses accrued in both areas during the NDCMP were compared to establish the magnitude of the change. The loss ratio is the ratio of the amount of claims paid to the amount of insurance sold. Thus, a high loss ratio indicates a high hail loss. When the loss ratios for the target were plotted versus those for the upwind (west) control area for each year, a significant difference emerged between the historical years and those of the NDCMP. Sixty-one years are compared; ten for the NDCMP years (circles), and 51 years for the "historical" period which preceded the NDCMP. In Fig. 1, a single (solid) regression line is plotted for all years, seeded and unseeded. When separate lines are plotted for the seeded and historical years however, a striking difference is observed. The difference between the slopes of the Historical and NDCMP regressions can be interpreted as the change induced by seeding. In this evaluation, the NDCMP target averaged 43.5% less crop hail damage when compared to the upwind control area. The statistical confidence level for this test was .002, indicating that "...the target area loss ratio values during these [NDCMP] years were considerably lower than would have been predicted by the historical and control area regression lines". The authors of the report conclude with the observation that, "Therefore, the sponsors of the NDCMP would seem to be fully justified in continuing their support of seeding operations."
Since the completion of the Smith et al. report, an additional three years' data from the NDCMP (1986-88) were added to the original data base. In an identical analysis, the expanded 13-year NDCMP data base showed a 45% decrease in losses in the target area (Smith, et al., 1997). The findings were published in the Journal of Applied Meteorology in 1997.