Common use of Dependent Variable Selection Clause in Contracts

Dependent Variable Selection. The Fusion LIDAR processing program CloudMetrics, (▇▇▇▇▇▇▇▇▇, 2016), was used to calculate 100 different metrics for each of the 113 plots from the LIDAR data. The models that most successfully predict heights needed to be identified. These models will use some of these LIDAR metrics and not others. Additionally, some LIDAR metrics are highly collinear, and models should avoid using multiple collinear metrics. Our experience has shown that Intensity, L Moment, and Count metrics could be removed from consideration for modeling heights, because they have not been useful in previous height modeling work. The remaining metrics tend to form three categories: height, cover, and distribution of heights. To avoid collinearity issues, models were limited to using only one metric from each category. To this end, only models with three or fewer predictor variables were considered. An automated process was developed in the statistical software R to compare all possible models with one, two, or three predictor variables. This process considered all distinct metric combinations, with and without interaction between predictor variables, and tested each predictor variable without transformation and with log and square root transformations. Each model was ranked by its R2 value (these were back-transformed if the dependent variable was transformed). This allowed for a relatively simple, straight forward and repeatable, but effective method of model selection.