Lisa A. Bragg

Date of Award


Level of Access Assigned by Author

Campus-Only Thesis

Degree Name

Master of Science (MS)


Resource Economics and Policy


Timothy J. Dalton

Second Committee Member

Kathleen P. Bell

Third Committee Member

Hsiang-Tai Cheng


Firm efficiency analysis has evolved from mathematically estimating an industry's mean level of efficiency and deriving producer-specific inefficiency scores to focusing on isolating "pure" inefficiency. Failing to isolate inefficiency attributable only to producer behavior, as opposed to other systematic factors, can affect parameter estimates and calculated efficiency scores. Misspecification bias may create significant statistical implications for frontier modeling. Scores may be biased and if so, the tendency is to under-estimate. Correlates may be misidentified due to this bias. Empirical studies have historically defined the production frontier and investigated correlates of inefficiency with a two-stage modeling process. The first-stage defines the structure of the frontier from which efficiency scores are derived. The second stage identifies systematic variation across these scores. This approach has been criticized since systematic variation found in the secondstage is indicative of omitted variable bias introduced during the first-stage. Incorporating exogenous farm-level characteristics into a one-stage frontier modeling process is hypothesized to better measure efficiency. Also, spatial information is incorporated into frontier estimation to control for the influence of a farm's relative landscape position on production decisions. Von Thiinen-influenced distance-to-market measures and regionalized dummy variables have been used to capture spatial dependence in agriculture. These spatial measures are not highly refined and ignore knowledge sharing and social network externalities. If information-based externalities hold significant influence and are ignored, parameter estimates and efficiency scores are biased reflections of industry conditions. Using maximum-likelihood econometric techniques, short-run efficiency scores are estimated for Maine dairy production using farm-level data collected in 2001. All three types of efficiency are investigated: technical, allocative, and economic. Technology is represented by the Translog functional form. Efficiency scores obtained from the one- and two-stage frontiers are tested for the influence of spatial autocorrelation. Evidence of spatial clustering led to the inclusion of geo-specific variables into both the one-stage frontier and secondstage Tobit regression. This approach eliminated spatial clustering. Further comparison of the one- and two-stage modeling results finds that a larger portion of the variation of efficiency scores is accounted for by the onestage model. On average, observed production practices are found to be more efficient. Pair-wise means testing and cumulative frequency distributions support this finding. Efficiency levels are not improved for all producers. In some cases, the difference between scores is large enough to negatively affect the ranking of a particular producer. Thus, these findings suggest that rankings are sensitive to the modeling method.