The Human Factor in Software Engineering

Recently, I again used my students as guinea pigs for a risk estimation experiment. The electrotechnics students at University of Applied Sciences at Bingen estimated the probabilities of some frequent risk events like the crash of a laptop. We applied the Delphi method which includes two estimation rounds. Between both rounds, the results of the first round are made transparent and discussed.

These are the experiment´s main results:
- The students were better than estimators in previous experiments where the same risk events were used. They overestimated the probabilities, but only by 114% (so, they approximately estimated the probabilities twice as high as they are).
- It was again found that having experienced a certain risk event before makes the estimator overestimate the probability even more.
- Correlations with two out of the Big Five personality characteristics were also investigated. Those estimators with a low value in neuroticism and a high value in diligence were better estimators than the others.
- As the estimates were too high in the first round, and those with low estimates raised their values in the second round and vice versa, the estimates converged, but overall worsened. The general and systematic bias to overestimate risk event probabilities can not overcome by applying the Delphi method.