The modifiable temporal unit problem in competitive accessibility measures and implications for regression inference

Transport agencies increasingly rely on accessibility metrics to evaluate transport investments and assess spatial equity. Many accessibility measures require the analyst to define a travel time threshold that determines which opportunities are considered within reach. While the sensitivity of cumulative accessibility measures to threshold choice has been explored at the measurement level, much less is known about how this issue affects competitive accessibility metrics, which explicitly account for competition over limited opportunities, or whether such sensitivity carries consequences for inference. This study examines how small variations in travel time thresholds influence competitive accessibility estimates and whether such sensitivity propagates into regression analyses that use accessibility as an explanatory variable. Using transit-based job accessibility in Taichung City, Taiwan, we compute accessibility minute-by-minute across thresholds ranging from 20 to 90 min and compare traditional single-cutoff specifications with an interval-based approach that summarizes accessibility across a range of travel times. Results show that single-cutoff specifications generate substantial fluctuations in accessibility and inequality estimates. These variations also translate into instability in regression coefficients and inferential conclusions. In contrast, interval-based specifications produce smoother accessibility profiles and more stable regression outcomes. These findings suggest that travel time thresholds are not neutral modelling choices. Using interval-based specifications can provide a more robust basis for integrating competition into accessibility analyses and reduce the risk of selective inference in transport planning and policy evaluation.