Automation and Container Throughput: Heterogeneous Evidence from a Staggered Difference-in-Differences Design

document divergence between engineering targets and operating parameters at commissioned terminals.This gap matters for tender evaluation, infrastructure lending, and decarbonization policy.Despite the scale of these investments, the empirical foundation for the productivity claim remains thin.Kon et al. [2] synthesize three decades of automated-containerterminal research.The methodological basis is dominated by discrete-event simulation, queuing models, and crosssectional data envelopment analysis (DEA).Reported simulation productivity gains cluster in the 25-50% range under favorable parameterization [3], measured at the operational unit rather than the terminal or port level.These estimates are internally valid for a stylized regime; they do not account for software-integration friction, labor-relations adjustment, weather disruption, and variable demand.Crosssectional DEA confounds the decision to automate with the AbstractFully automated container terminals promised step changes in throughput, yet the empirical record rests on engineering simulations and crosssectional efficiency scores that do not identify the causal effect of adoption.We provide an exploratory causal assessment of port-level throughput after automation by applying the Callaway-Sant'Anna staggered difference-in-differences estimator to a port-year panel of 56 container ports from 2001 to 2024, covering seven adoption cohorts.Synthetic control, changes-in-changes, and a size-matched placebo provide complementary checks.The pooled average treatment effect on log throughput is +0.112 [95% uniform confidence band (-0.015, 0.240)].One atypical cohort (Abu Dhabi 2020) is the single largest influence on the pooled estimate; its weight in the aggregate is limited by a short post-adoption window.Confidence bands for five of the remaining six cohorts include zero, and the Abu Dhabi estimate is fragile with respect to identification under the donor-pool and Oster-bound diagnostics.The size-matched placebo yields an ATT of +0.075 (-0.114, 0.263); therefore, the treated-port estimate is not statistically distinguishable from placebo variation.A dose-response specification at two dual-terminal port-clusters (Rotterdam and San Pedro Bay) is consistent with saturation rather than additive gains.The evidence reframes the port-level throughput response to automation as contextconditional rather than uniform, with implications for infrastructure lending and terminal investment appraisal.