Comparative performance of logistic regression and survival analysis for detecting spatial predictors of land-use change

Although survival analysis is known to outperform logistic regression, theoretically and according to evidence from other disciplines, little is known about how true this is in situations where the goal is detecting spatial predictors of land change. Furthermore, with the increasing availability of longitudinal land-change data, evidence is needed on the relative performance of these two different methods in situations with differing levels of data abundance. To fill this gap, we generated a pseudo land-change data set using an agent-based model of residential development in a virtual landscape. This agent-based model simulated the decisions of homebuyers in choosing residential locations based on the values of several spatial variables. Pseudo land-change maps, generated by the agent-based model with different weights on these spatial variables, were exposed to statistical analysis under the logistic and survival approaches. We evaluated how well the two approaches could reveal the spatial variables that were used in the agent-based model and compared the performance of the two methods when land-change data were collected under different sampling frequencies. Our results suggest that survival analysis outperforms logistic regression in detecting the variables that were included in agent decisions, largely because it takes into account time-dependent variables. Also, this research suggests that various properties of land-change processes (like amount of developed area and access of agents to information) affect the relative performance of these statistical approaches aimed at uncovering land-change predictor variables.