School of Continuous Improvement researchers re-estimate the matching step behind a state's published level-crossing occurrence rates, and find the crossings the register reports as quietest are the ones whose reports it could not resolve.
A re-linkage of 6,142 driver occurrence reports against a 2,388-crossing register finds more than a quarter resolving to no crossing at all, and the failures concentrated where the signage carries only a number
Four. That is how many of the 120 crossings a state’s current rail safety education campaign was directed at are the kind identified on the ground by a kilometrage plate rather than a road name. There are 1,043 such crossings on the network. Quiet Where the Sign Has No Name, published today by the School of Continuous Improvement, sets out why so few of them are on the list.
The published occurrence rate for each crossing is assembled by reading the free-text location field on a train driver’s report and finding the matching entry in the asset register. Dr Thandiwe Solberg, Dr Sten Okwuosa and Dr Bram Ntuli treated that step as the record-linkage problem it is, and re-estimated it across five financial years of reports. At the operating point that reproduces the published counts, 71.4 per cent of reports resolve to a unique crossing. Reports naming a road and a locality resolve at 94.6 per cent; reports giving only a kilometrage resolve at 34.7. Checked against a hand-resolved sample, the occurrence rate at the quietest class of crossing rises from 0.06 to 0.33 per crossing-year.
Work of this kind depends on people being willing to let us look at the joins between one record and the next. We are grateful for that, and we think the questions it raises are ones worth sitting with together.
— Dr Thandiwe Solberg, Senior Lecturer and Convenor of Demo Quarter
The release extends a line of work the School has been assembling on the seams where one administrative record is fastened to another. Dr Sten Okwuosa, who convenes the Impact Pathway Atlas, noted that “the register is entirely consistent with the reports it can read”, and that the University’s contribution is to widen what can be read.
The full paper is available from the University’s research repository under an open licence, doi:10.5555/slop.ba3177.