社会经济统计数据热点探测的MAUP效应

MAUP Effects on the Detection of Spatial Hot Spots in Socio-economic Statistical Data

为探讨不同尺度下社会经济统计数据热点的变化规律及其影响因子,本文基于2000年全国县级农业统计数据和2008年北京市第二次经济普查数据,按照一定的聚合规则得到不同尺度的数据,计算不同尺度下的局部空间自相关指标G统计值并对其进行显著性检验得到热点分布,分析不同聚合尺度下热点的变化规律。然后运用Logistic回归分析探测了影响聚合前后热点变化的因素,并根据探测结果建立了预测聚合前后热点变化的Logistic模型。分析结果表明,基于G统计探测的热点分布具有明显的空间尺度效应,聚合水平越高、空间尺度越大,热点数目越少。Logistic回归分析的显著性分析表明,热点包含的面状单元数目和热点的平均G统计值是影响热点探测尺度效应的主要因素。热点包含的面状单元越多,热点的平均G统计值越大,热点探测结果受尺度效应的影响越小。研究建立的热点变化预测模型,可以在细尺度热点分布状况已知时,根据热点包含的面状单元数目和热点的平均G统计值来预测聚合后热点的变化。对模型精度的交叉验证结果表明,模型对全国县级农业统计数据热点变化预测精度可达到93.8%,对北京市第二次经济普查数据热点变化预测精度达到94.2%。两套数据试验得到的结论一致,说明热点探测的尺度效应变化规律和所选变量以及研究区域的大小无关。

The study of spatial distribution of population and economic situations is important for government policy making. County-level agriculture statistical data in 2000 and Beijing＇s second economic census data in 2008 were collected in order to explore the hot spots＇ scale effects. First, China＇s county-level agriculture statistical data and Beijing＇s second economic census data were aggregated to different scales based on certain aggregation rules. Second, hot spots detection was implemented based on G value at each scale respectively. Third, the changes of hot spots at different scales were analyzed. Fourth, factors affecting the changes were identified by employing Logistic Regression Model and a prediction model was built. Results show that, space hot spots explored by G value have significant MAUP effects. The higher the aggregation level, the greater the spatial scale, the less the number of hot spots. The number of units in a hot spot on the confidence level of 99.9% has a significant effect on the changes of hot spots. The mean G value of a hot spot on the confidence level of 98% has a significant effect on the changes of hot spots. Hot spots will become less susceptible to MAUP when they have more units and a larger G value. When the hot spot distribution is already known in the fine scale, changes of a hot spot can be predicted based on the model we built, which depends on unit number that the hot spot contains and mean G value of the hot spot. The prediction accuracy of China＇s county-level agriculture statistical data can reach 93.8% and that of Beijing＇s second economic census data can reach 94.2%. The consistent conclusion of the two datasets shows that scale effects on the detection of spatial hot spots have nothing to do with variables and study areas.