Some experiments, particularly those associated with weather modification, are best formulated in terms of the"null" hypothesis, the hypothesis that the treatment has no effect. For example, in a cloud-seeding experiment the null hypothesis may be that release of seeding material has no effect on the precipitation. To test this hypothesis, we determine if the precipitation falling from seeded clouds differs significantly from that falling from unseeded clouds. If so, one "rejects the null hypothesis" -- i.e., concludes that precipitation differs in seeded vs unseeded clouds. (Note that this still does not imply that seeding caused the difference; the difference may arise because the seeded clouds were naturally more vigorous, or because of some other unrecognized cause.) In this case, a type-I error occurs if we reject the null hypothesis, and conclude there is a significant difference, when in fact there is none.
The experiment needs to be designed so that the expected result provides a definitive test. For example, if the result of a seeding experiment is that the null hypothesis is accepted (i.e., there was no significant difference between seeded and unseeded cases), this may only reflect an inadequate sample size or too small an effect to detect with the selected experimental design. Numerical experiments and statistical simulations can help avoid poor experimental designs. For example, one might want to test how well current formulations of collision efficiencies predict the rates of coalescence of water droplets leading to rain. Numerical experiments ahead of the field experiment can help determine the accuracy needed in measurement of such relevant factors as updraft speed, liquid water content, droplet size distribution, radar reflectivity, cloud condensation nucleus population, etc. These experiments can help avoid the all-too-common inconclusive experiment, one producing the primary conclusion that a more careful experiment is needed.