Ghost Infrastructure: Historical Industrial Geography and the Persistence of Path-Dependent Accessibility in Bochum, Germany

SAKSHI D. MASKE 

The complete 69,393 pedestrian street node network from Bochum was used for a Welch's t-test, which revealed as a surprise that street nodes with low accessibility are actually significantly farther from historical industrial sites than street nodes with high accessibility (t=42.887, p<0.00001, Cohen's d=0.589 - medium-to-large effect), which is the opposite of what the original hypothesis expected. This is done by determining whether the industrial geography of Bochum in the 19th and 20th centuries in relation to the location of coal mines and the housing colonials for miners (Zechensiedlungen) can still be regarded as a prediction of the "15-minute city" access to everyday areas by walking or cycling, more than 50 years after the last coal mine was shut down. Each of 13 historical coal mines and 4 worker colonies have been digitized from existing archival sources, and accessibility for present-day mines was simulated by computing true network distance isochones over the entire street network. This inverse link remained significant in a logistic regression model when stratified by proximity to the inner core of the city which has been previously established as an important predictor of level of accessibility in the wider 15-minute city literature: with the city-center distance as the only independent variable modeled, the city-center coefficient remained statistically significant (coefficient=-0.0005, p<0.001); and the correlation between the two distance measures was very low (r=0.063). Combined with this, a spatial-clustering analysis of Local Moran's I (a complementary statistical method) confirms the same findings: that 97.1% of low-accessibility nodes can be identified as part of statistically significant spatial cold-spot clusters, indicating that low accessibility is not random but represents real, contiguous geographical areas that are measurably further from historical industrial infrastructure.

This finding was also tested a second time, in a second city, and the honest result is a mix: some parts confirmed, some did not, and both parts are reported here in full rather than only the confirming half. The same raw reversed relationship, however, still present in direction and significance (t=24.731, p<0.00001), was present in the 72,027 nodes of pedestrian network (95.5% of the low-accessibility nodes of the city in significant cold-spot clusters, compared to 97.1% of the nodes in Bochum). Cohen's d was lower (0.338 versus 0.589) and the Local Moran's I spatial clustering result was replicated almost exactly. However, this was not found to replicate — a logistic regression that jointly controls for both distance-to-historical-site and distance-to-city-center shows the coefficient of the historical-site effect changing sign (from negative in Bochum to positive in Essen), meaning the historical-site effect is highly correlated with, and non-independent of, the city-center effect in Essen. The "path dependency of centrality" effect in itself seems to generalize, but the finding that it doesn't happen if a city is not represented in the city center itself is, according to current evidence, a Bochum-specific and not general result. This is corroborated by a separate walking-threshold sensitivity analysis, which was performed prior to the Essen replication; the results, both before and after that replication, showed the same reversed correlation at an even stricter 10-minute (750m) and even more permissive 20-minute (1500m) walking-threshold (Cohen's d=0.413, p<0.00001 and Cohen's d=0.661, p<0.00001 respectively).