Progress in Private Forest Landowner Estimation

Abstract

T wo recent issues of the Journal of Forestry contained a discussion of methods for estimating private forest landowner (PFL) population parameters. Initially, Metcalf et al. (2012) raised questions about the application of unequal probability estimation equations to samples of PFLs drawn proportional to property size (i.e., PPS). In response, Dickinson and Butler (2013) presented corrections to National Woodland Owner Survey (NWOS) methods that clarified sampling protocols and revised equations and procedures used to generate estimates of PFL parameters. Dickinson and Butler (2013) also raised questions about the validity of the analysis and conclusions reached by Metcalf et al. (2012). This response addresses those questions, amends conclusions presented in Metcalf et al. (2012), and expands the discussion of implications for managers and researchers concerned with private forest ownership and management. It was never our intent to critique and/or debate methodological issues. Our research was funded by the Pennsylvania Department of Conservation and Natural Resources, Bureau of Forestry with the specific goal of informing outreach and extension efforts promoting stewardship of the Commonwealth’s vast private forest (approximately 11.5 million acres, 70% of the state’s forestland). This project did not focus on the exploration of the functionality of estimation equations. Our methodological approach was fashioned after the NWOS methods presented in US Department of Agriculture (USDA) Forest Service publications (e.g., Birch 1996, Butler et al. 2005), but coupled to a substantial increase in sampling intensity. After collecting data and conducting different analyses, we were left with curiously incongruent results (e.g., estimates of PFL population size in the millions rather than the hundreds of thousands reported in the NWOS). This finding led us to initiate a systematic review of both the NWOS methods and our own to explore possible sources for bias and/or error. The review uncovered problems with the NWOS estimation equations and their application. We presented these findings in Metcalf et al. (2012), in which we specifically addressed the confusion of inclusion probabilities with selection probabilities and the erroneous exclusion of duplicate PFL selections when using the Hansen-Horvitz estimator (HHE). The implications of these errors on our data and the resources needed to conduct the review itself severely constrained our research. The conclusions tied to estimation approaches helped clarify our research, led to changes in NWOS estimation procedures as presented by Dickinson and Butler (2013), and allowed other researchers to have significantly more confidence when designing and interpreting results from PFL studies. Dickinson and Butler (2013) dismissed the issues presented in Metcalf et al. (2012) as trivial because of an unestablished theoretical basis (p. 319), case study methodology (p. 320), limited sampling intensity of NWOS (p. 323), and unaddressed issues (i.e., nonprivate forest sampling points and variance estimation, p. 323, 324). We disagree. Consumers of NWOS results need to be assured that they are unbiased; researchers using USDA Forest Service methods also need to be assured that their resulting estimates are unbiased. This was not the case before the publication of Metcalf et al. (2102). Indeed, Dickinson and Butler (2013) acknowledged the following two errors in past NWOS methods: confusion of selection probability and inclusion probability calculations; and unwarranted omission of duplicate selections of PFLs in the sampling process. Such errors, when applied to sampling designs identical to NWOS, produce biased estimates (increasingly so as sample size increases), thus potentially affecting past NWOS panels and any PFL study replicating NWOS methods before this discussion. Further, because the source of bias is largely affected by the exclusion of duplicate selections of PFLs, estimates for areas of interest (e.g., regions and states) characterized by larger private forest parcels (i.e., the western United States) may be disproportionately biased. Dickinson and Butler (2013) suggested a review of past NWOS data to explore possible changes in results due to updated methods; we strongly encourage this analysis of legacy NWOS data because it will help us understand the full implication of these errors on past results and reveal true trends in private forest ownership characteristics. Allocations of federal formula funds are, in part, determined by estimates of PFL population sizes by state and/or region. Parcelization patterns vary disproportionately across states: some states are highly parcelized, whereas others are less so. Thus, the likelihood for duplicate selections of PFLs in past NWOS sampling varies from state to state. It follows that the bias in estimates of PFL population parameters (including total population size) may be more pronounced in some states than others; this bias results in skewed resource allocations. Understanding the nature of such errors would not be possible without analysis of past NWOS data using revised methods. This type of legacy analysis to remove past bias is essential for a longitu-