3.3.1. Non-Parametric Analysis

For RQ2, given the small sample size, 35 accidents with maintenance contributions, testing was limited to utilizing Fisher's exact test [26]. This type of testing with small sample sizes has previously been utilized in other post-accident analyses [27]; as with that work, Fisher's exact tests were undertaken in MATLAB to determine the *p*-values. Specifically, in the testing of the various characteristics or features coded, the observed (O) data were the 35 accidents with maintenance contributions, while the expected (E) are those of the complete set of 1277 official accidents. The statistical hypotheses to be tested are given as

$$\begin{array}{cc} \mathrm{H}\_{0}\mathrm{:} & \mathrm{P}\_{\mathrm{O,n}} = \mathrm{P}\_{\mathrm{E,n}}\\ \mathrm{H}\_{\mathrm{A}}\mathrm{:} & \mathrm{P}\_{\mathrm{O,n}} \neq \mathrm{P}\_{\mathrm{E,n}} \end{array}$$

where P is the proportion of the n'th category. That is, n separate Fisher's exact tests were conducted.
