Exploiting Zero-Inflated Consumption Data using Propensity Score Matching and the Infrequency of Purchase Model, with Application to Climate Change Policy

We apply propensity score matching (PSM) to the estimation of household motor fuel purchase quantities, to tackle problems caused by infrequency of purchase. The results are compared to an alternative, regression-based, imputation strategy using the infrequency of purchase model (IPM). Using data from the UK’s National Travel Survey (NTS) we observe that estimated mean obtained from the PSM imputation is closer to the estimated mean from the consumption diary, than that obtained from fitted values from the IPM. The NTS also contains an interview question on household mileage which can be used to assess the results of imputation. We find that the order statistics of the imputed distribution are more plausible for the PSM estimates than those obtained using the IPM, judging by the sample distribution of household mileage. We argue that there are some applications for which the PSM method is likely to be superior, including estimates of distributional effects of policies. On the other hand, the IPM is more suitable for analysing conditional effects and associations of consumption with covariates. We illustrate our arguments using a simple microsimulation exercise on CO2 emissions reduction policies, an area where methods for coping with zero-inflated data seem currently to be under-used.