Abstract
The Timber Products Output (TPO) survey is used to determine industrial uses of roundwood, reporting on volumes of roundwood received and residues generated by the primary forest industry by tree species and counties of harvest. This knowledge aids stakeholders in making informed decisions about available forest resources and/or harvest intensity. The widespread use of TPO estimates makes it important to understand the scale and scope of missing data in the survey. This first attempt analyzed respondent-level and question-level nonresponse (RLNR and QLNR, respectively) for Northern and Southern TPO regions, comparing response to mill profile questions (those providing general business information) to those related to mill activity (that related to wood processing information). The RLNR differed between regions, by question grouping, survey mode, and mill volume. The QLNR results for selected mill activity questions indicate that the Southern region generally has lower nonresponse than the Northern region. Parametric analysis of RLNR indicated survey mode was significant for both question groups in the Northern region whereas mill type was significant for mill activity questions in the Southern region. The QLNR parametric analysis indicated self-administered surveys in the Northern region were associated with higher nonresponse, and surveys completed by sawmills in the Southern region were associated with lower nonresponse. Study Implications: Analysis of survey item nonresponse in the national Timber Products Output survey provides analytic background needed to assess the accuracy and completeness of the survey data. Population estimates from the data are used to monitor roundwood production and make informed decisions about forest resources. Responses varied across specific questions, and regional differences appeared to be related to survey mode. Current survey design could be improved to allow for analysis of all survey questions and assessment of the quality of responses. For nonresponse adjustment methods to be comparable across regions, factors causing observed regional difference should first be addressed.