Missing(ness) in Action : Selectivity Bias in GPS-Based Land Area Measurements

Kilic, Talip ; Zezza, Alberto ; Carletto, Calogero ; Savastano, Sara

World Bank, Washington, DC

关键词: AGRICULTURAL OUTPUT;    AGRICULTURE;    COMMODITY;    COMPUTATION;    CONFIDENCE INTERVALS;   

DOI  :  10.1596/1813-9450-6490
RP-ID  :  WPS6490

学科分类：社会科学、人文和艺术（综合）

来源: World Bank Open Knowledge Repository

【 摘 要 】

Land area is a fundamental component ofagricultural statistics, and of analyses undertaken byagricultural economists. While household surveys indeveloping countries have traditionally relied onfarmers' own, potentially error-prone, land areaassessments, the availability of affordable and reliableGlobal Positioning System (GPS) units has made GPS-basedarea measurement a practical alternative. Nonetheless, in anattempt to reduce costs, keep interview durations withinreasonable limits, and avoid the difficulty of askingrespondents to accompany interviewers to distant plots,survey implementing agencies typically require interviewersto record GPS-based area measurements only for plots withina given radius of dwelling locations. It is, therefore,common for as much as a third of the sample plots not to bemeasured, and research has not shed light on the possibleselection bias in analyses relying on partial data due togaps in GPS-based area measures. This paper explores thepatterns of missingness in GPS-based plot areas, andinvestigates their implications for land productivityestimates and the inverse scale-land productivityrelationship. Using Multiple Imputation (MI) to predictmissing GPS-based plot areas in nationally-representativesurvey data from Uganda and Tanzania, the paper highlightsthe potential of MI in reliably simulating the missing data,and confirms the existence of an inverse scale-landproductivity relationship, which is strengthened by usingthe complete, multiply-imputed dataset. The studydemonstrates the usefulness of judiciously reconstructedGPS-based areas in alleviating concerns over potentialmeasurement error in farmer-reported areas, and with regardsto systematic bias in plot selection for GPS-based area measurement.