Positive response density estimation from CV interval data affords efficiency gains which must be weighed against the risk of introducing potential bias during questions iteration. This study examines the effect of eliciting a third response on a set of often-used welfare measures derived in a conventional parametric setting. It then compares these with distribution-free nonparametric estimates. A third bound increases censoring probability, introduces welfare estimates sensitivity to inclusion of a theoretically relevant covariate such as wealth which also affects efficiency gains. This might well introduce complications that outweigh the expected efficiency gain. This empirical finding supports and complements previous results obtained via simulation.