Monte Carlo study of several methods for estimating linear demand for outdoor recreation from censored and truncated data

Graduation date: 1993Objective of this study was to compare accuracy of\ud several travel cost methods, by employing Monte Carlo\ud simulations in the single site framework. Estimates from\ud five methods were compared to the "true" travel cost\ud coefficient and "true" consumer surplus. Each method was\ud ranked, using the root mean squared error (RMSE) criterion.\ud Each method was applied to one of two kinds of travel\ud cost data: censored data that includes both users and nonusers\ud of recreation, and truncated data from on-site surveys\ud consisting of only users. For censored data, the Tobit\ud (TOBIT) method and OLS (OLSALL) were used, and for truncated\ud data, the zone average travel cost (ZATC) method, the\ud truncated normal (TRUNCN) estimator, and OLS (OLSUSR) were\ud employed.\ud Two types of consumer surplus were compared:\ud (1) Predicted (traditional) consumer surplus (TRCS) that\ud uses predicted values of the dependent variable, and (2) "actual" or Gum-Martin consumer surplus (GMCS) that uses\ud observed values.\ud Hellerstein's (1992) method for computing TOBIT TRCS\ud was used in this study, which supposedly gives better\ud estimates of TOBIT TRCS. This improvement, however, did not\ud always give a better RMSE ranking than ZATC or OLSALL TRCS.\ud The smaller the proportion of non-users in the data,\ud the better were the TOBIT estimates. When the non-users\ud portion was larger, ZATC and OLSALL sometimes gave better\ud estimates of consumer surplus than TOBIT.\ud An unexpected identity of linear demand estimated by\ud OLSALL and the classic ZATC model was discovered. An\ud original proof of this identity is given in Appendix 1.\ud Because of this identity, both OLSALL and ZATC gave\ud identical TRCS estimates. However, GMCS estimates differed\ud because of averaging involved with ZATC.\ud For censored data, TOBIT gave the best linear demand\ud and GMCS estimates, but for truncated (user-only) data, best\ud estimates of consumer surplus were from the supposedly\ud obsolete ZATC model, fitted by Bowes-Loomis weighted least\ud squares.\ud Another new finding was that for data suitable for\ud ZATC, missing zero observations can be synthesized, thus\ud permitting use of TOBIT. However, thorough testing of such a\ud scheme was beyond the scope of this thesis