Addressing Non-response Bias in Urban Forest Inventories: An Estimation Approach

Abstract

When conducting urban forest inventories, complete or partial inaccessibility of sample plots results in non-response for a portion of the selected sample. Unfortunately, the non-response is rarely random and thus a potential bias may be imparted in the sample and associated data analyses. In this study, results from an alternative estimation method that employs response homogeneity groups (RHGs) appeared to be more robust to non-random non-response when compared to those of a standard estimation method. Across the six cities studied, the total non-response rates varied from 8.0 to 20.4%. Percent differences between the two methods in estimated number of trees ranged from −0.7 to 12.6%; whereas 1.4 to 14.8% differences were found for tree biomass density. While these differences only approximate the amount of non-response bias present under standard estimation methods, there is a clear indication that misleading results may be obtained if non-response bias is not adequately addressed. By implementing methods that mitigate potential non-response bias, urban forest inventory practitioners would increase the reliability of information used by city planners to make effective management and policy decisions.