In order to build four polynomial models we calculated a correlation matrix and selected the best independent variables.
Using HMM models, we calculated τ for different tunnels and compared the results with the τ of real termites.
To summarize predictions across the three final models, we calculated median values for each pixel, and assessed uncertainty in the model outputs as the range of the medians of 10 replicate analyses for each model.
To examine agreement with models, we calculated intraclass correlation coefficients (ICC) and kappa statistics.
In order to compare the magnitude of group differences as estimated by the four models, we calculated effect sizes between the group means of proficiency.
To validate our models, we calculated validation parameters for our correction model according to the principles of the Organization for Economic Cooperation and Development (OECD) [41].
For each modelling scenario (i.e., NFI- and LiDAR-calibrated AGB models), we calculated a mean AGB value from 100 Monte Carlo realizations.
In order to compare the scaling models, we calculated the log-likelihood, the Bayesian information criterion (BIC; Schwarz 1978) and Akaike's information criterion (AIC; Akaike 1974) for each of the four models.
To develop protein specific prediction models, we calculated around 18,000 molecular descriptors which include geometric, electrostatic, structural, constitutional, path and graph fingerprints etc. utilizing the open source PaDEL software [28].
Using unconditional logistic regression models, we calculated the risk of hemorrhagic stroke associated with migraine, adjusting for age, sex, calendar year, alcohol, body mass index, hypertension, previous cerebrovascular disease, oral contraceptive use, and health services utilization.
Based on the best models we calculated predictive distributions to evaluate goodness-of-fit.
