Rural-urban land conversion probability was calibrated through a spatial logistic regression.
The dependent variable is binary (whether the rural land use developed into urban land use or
remains in the current state) and the predictors are the seven factors discussed in the
preceding section. First of all, the predictors were prepared in ArcInfo. Then, spatial
sampling was implemented to reduce spatial dependence. This study employs a sampling
scheme integrating systematic and random sampling. A systematic sampling with a 20th order
lag (20 pixels interval in east-west and north-south directions) is implemented for the
population. After that, another 10% from sample 0 was randomly selected to gain unbiased
parameter estimation. Finally, a binary logistic regression was processed using MiniTab.
The rural-urban land conversion models of 1984-1992 and 1992-1997 were analyzed
respectively and the regression results presented in table 2. Two models, both significant at
0.000, show some regularity in land use conversion. The logistic regression model was
estimated using maximum likelihood algorithm. One efficient way to assess the goodness-offit
of logistic regression is to cross tabulate prediction with observation and to calculate the
percentage correctly predicted (PCP). The overall percentages of correctness were 80.7% and
74.0% respectively for 1984-1992 and 1992-1997. Hence, the models can be regarded as
effective descriptions of the rural-urban land conversions, as only a limited number of
explanatory variables are used and land use conversions are usually distributed in a
complicated way.