Uncertainty in experimental data and models for punching shear

Abstract A database of about sixteen hundred monotonic or cyclic tests of shear‐critical slab‐column connections (concentric or eccentric interior, edge or corner ones, with shear reinforcement in the slab or not) is used to investigate how much of the test‐to‐model scatter may be due to the data. Test results are compared to predictions of the independent models of punching shear resistance in fib MC2020 and second‐generation Eurocode 2, both modified in this work to give zero mean natural log of test‐to‐model‐ratios and be free of lack‐of‐fit with respect to key independent variables. A combination of the two modified models is proposed to express the central tendency of test results. The analysis suggests that each testing campaign has its own bias, which inflates scatter with respect to models and induces correlation between tests of the same campaign. Removing the estimated effects of this bias reduces the coefficient of variation of the ratio of experimental punching shear resistance to the predictions of fib MC2020 or second‐generation Eurocode 2 to 12%–16% or 10%–14%, respectively. Criteria to identify sources of data incompatible with the central tendency of test results are proposed and applied.