ABSTRACT An experimental database of over one thousand shear‐critical reinforced concrete beam‐column joints from about 220 testing campaigns is analyzed with three unbiased models to estimate possible campaign bias. Two of the models are improvements of earlier mechanical ones; the third is a new empirical model. Test campaign bias produces and relates to strong correlation of test‐to‐prediction ratios of the three models and between tests of the same campaign. The outcomes of all three models are combined to give the maximum likelihood prediction for each test, which reflects the central tendency of the data. About one‐fifth of the test campaigns in the database, accounting for 12% of the tests, violate at least one of three criteria of compatibility with the so‐established central tendency. Exclusion of these campaigns reduces scatter of model predictions with respect to test results, but leaves in place correlations between predictions of different models and between tests of the same campaign. If, instead, the best estimate of the perceived campaign bias is removed from all test results, correlations are essentially eliminated and scatter is reduced as much as with removal of test campaigns considered as incompatible with the central tendency. Scatter is further reduced if both approaches are applied together. The reduced scatter points to a lower bound of 15%–20% for the standard error (or the coefficient of variation) of the three models, that is, about half the gross standard error of reported test results incorporating campaign bias. An upper bound of 20%–25% is estimated for the contribution of campaign bias to gross standard error.
Michael N. Fardis (Tue,) studied this question.