The impact of probe sample bias on the accuracy of commercial floating car data speeds

Floating car data (FCD) reliability studies regularly report differences between FCD and benchmark speeds, called speed bias. Comparatively few studies consider factors causing speed bias. Various researchers have suggested, but not proven, that the sample of probes reporting FCD may result in FCD accuracy discrepancies between regions and sources. This paper defines a new FCD accuracy contributor - sample bias - and offers the first statistically motivated investigation of the impact that the FCD sample has on FCD accuracy, measured by speed bias. Evidence for sample bias is considered by observing how speed bias is impacted by region and FCD source. The magnitude of speed bias on freeways differed significantly between six cities in South Africa (p < .001) and between two commercial FCD sources (p = .046), from which it is inferred that sample bias does impact FCD speed bias. These findings inform the application of FCD for comparative studies and overall accuracy.