Caldera collapse events can be sudden and violent in the case of large explosive volcanic eruptions, or incremental in the case of long-lived eruptions. Faults nucleating during collapse are associated with seismic activity, yet the kinematic behavior of newly formed faults is poorly constrained. We conducted a series of novel sandbox experiments using piezoelectric sensors to monitor stress perturbations during a caldera collapse. We found excellent spatial and temporal correlations among (1) fault nucleation, inferred from the stress sensor data, (2) the appearance of faults on the surface, and (3) final fault structure, obtained via cross sections. We estimated fault propagation rates for early inner faults and found that these rates increase with increasing magma evacuation rates. We applied our experimental results to seismic data from natural caldera-forming episodes in order to estimate rates of fault propagation for these systems. Our experiments are consistent with en masse caldera collapse events, such as at Mount Katmai (Alaska, USA) in A.D. 1912 and Mount Pinatubo (Philippines) in 1991.