Previous work by the author used Markov Latent Class Analysis (MLCA) to make aggregate estimates of the underreporting of household expenditure by category (e.g. clothes, furniture, and electricity) by exploiting the four interview, rotating panel design of the Consumer Expenditure Interview Survey (CE). This analysis collapsed a few years of the CE into a pooled single panel. Estimates from this analysis were shown to be consistent with both internal and some external indicators of measurement error. However, there is no “gold standard“ for expenditure reports and MLCA models require strong assumptions for identifiability making evaluation of model estimates difficult. The current analysis uses MLCA on data from 15 years of the CE, in order to examine the reliability of measurement error estimates over time. Both sequential and overlapping panels are used in order to assess the sensitivity of MLCA estimates to major survey revisions. Finally, these estimates are compared to survey nonresponse and external benchmarks where available.