Received date: 2011-10-26
Request revised date: 2012-01-20
Online published: 2012-11-20
Copyright
The problem of aggregation has been always one of the hot and focuses of the region research, and the positioning of aggregation is a prerequisite for continuing in-depth study in aggregation analysis. Aggregation appears a high degree of sensitivity on the scale. When the aggregation was analyzed by using the spatial autocorrelation method, scale choices are often susceptible to subjective judgments of the researchers, and exist the possibility of selection bias. So the spatial weight problem has been controversial. In addition, the aggregation is obviously space-dependent as well as time-dependent since different aggregation takes place within different time and space, which is neglected by the spatial autocorrelation method. Therefore, some scholars have been trying to explore better ways for aggregation analysis. In comparison, the spatial-temporal scan statistic method raised by scholars as Kulldorff shows its superiority. This paper targets at 51 districts and counties within the central and southern regions of Gansu Province, which are densely populated minority areas. Using software Geoda0.9.5 and Clusterseer0.2.3, spatial autocorrelation analysis and space time scan statistical analysis were adopted about 30463 perinatal deaths during 2001 to 2010 according to the report delivered by monitoring points of Ministry of Health. It also gives a detailed comparison between spatial autocorrelation and scanning statistic from the different perspectives of scale choices, scale transforming and space integration, and then tests the analysis results by appealing Sam 4.0. Through theoretical and empirical analysis, it further confirms that spatial-temporal scan statistic method has the significant advantage from the three aspects of scale selection, scale conversion, and spatial-temporal integration,which shows that it is not only effective measure to solve the problem of artificial selection, but also to achieve the scale extrapolation and automatic conversion, and more favorable mix of three-dimensional, dynamic, multi-scale analysis. Therefore, a conclusion is drawn that the space-time scan statistic method is superior than other methods. Specifically, there are three main aspects: 1) Compared to spatial autocorrelation methods and space-time scan statistic method, the adjacency and distance matrix is artificial selection in spatial autocorrelation. The space-time scan statistic method achieves the purpose of scale automatically conversion and avoids the instability. 2) The scale of spatial autocorrelation methods is static, single, and the scale of space-time scan statistic is dynamic, three-dimensional and multi-scale.3) In analyzing the issue of spatial autocorrelation method does not take into account the time factor, while the space-time scan statistic method is not only the full integration of space technology, and achieves good analytical results. By combining the time factor, the range of aggregation can be reduced to a more accurate range. Moreover, space-time scan statistic method can even make a prediction, which can provide a basis for decision-making.
WANG Pei-an , BAI Yong-ping , GUO Jin-xian , LI Jia-yue . [J]. SCIENTIA GEOGRAPHICA SINICA, 2012 , 32(11) : 1410 -1416 . DOI: 10.13249/j.cnki.sgs.2012.011.1410
Table 1 Comparison of spatial autocorrelation results in different adjacency guidelines for cases of perinatal in 2010表1 不同邻接准则空间自相关结果比较(2010年围产儿病例) |
邻接方式 | 邻接准则 | Moran's I值 | 检验P值<5% | 检验次数 |
---|---|---|---|---|
Rook邻接 | 上下左右邻接准则 | 0.3325 | 0.003 | 999 |
Queen邻接 | 上下左右对角线邻接准则 | 0.3317 | 0.002 | 999 |
K-Nearest邻接 | K值最近邻接准则 | 0.3086 | 0.005 | 999 |
Distance邻接 | 5%的门槛距离准则 | 0.1749 | 0.002 | 999 |
Delaunay邻接 | 德劳内三角形构建准则 | 0.25 | 0.003 | 199 |
Gabriel邻接 | 加布里多边形构建准则 | 0.143 | 0.06 | 199 |
Relative邻接 | 相对近邻构建准则 | 0.185 | 0.045 | 199 |
Min Tree邻接 | 最小生成树构建准则 | 0.235 | 0.146 | 199 |
Table 2 Comparison of the results for spatial-temporal scan statistics and pure spatial scan statistics表2 时空扫描统计量和纯空间扫描统计量结果对比 |
时间 | 聚集分类 | 显著聚集区域 | 对数似然率 | 检验P值 | 结论对比 |
---|---|---|---|---|---|
2010年 纯空间扫 描结果 | 一集聚区 | 永靖县、临洮县、东乡县、临夏县、临夏市、和政县、广河县、康乐县、夏河县、碌曲县、卓尼县、临潭县、漳县、岷县、宕昌县、迭部县、舟曲县 | 72.963 | 0.001 | 似然率较低,识别范围过大,比较笼统,实际操作时参考范围不明确 |
二集聚区 | 静宁县、庄浪县、张家川县、清水县 | 38.754 | 0.002 | 似然率较低,范围偏大 | |
2000~2010年 时空扫描 结果 | 一集聚区 | 永靖县、临洮县、东乡县、临夏县、临夏市、和政县、广河县、康乐县、夏河县、碌曲县、卓尼县、临潭县 | 1 045.13 | 0.001 | 似然率显著提高,通过过滤和识别范围变小 |
二集聚区 | 永登县 | 415.773 | 0.003 | 似然率显著提高,通过过滤和识别范围变小 |
Fig. 1 Comparison of results s for spatial autocorrelation and spatial-temporal (spatial) scan statistics图1 空间自相关和时空(空间)扫描结果对比 |
The authors have declared that no competing interests exist.
[1] |
|
[2] |
Sawada M.Global Spatial Autocorrelation Indices—Moran’s I, Geary’s C and the General Cross-Product statistic[Z/OL].
|
[3] |
|
[4] |
|
[5] |
|
[6] |
|
[7] |
|
[8] |
|
[9] |
|
[10] |
|
[11] |
|
[12] |
|
[13] |
|
[14] |
|
[15] |
|
[16] |
|
[17] |
|
[18] |
|
[19] |
|
[20] |
|
[21] |
|
[22] |
|
[23] |
|
[24] |
|
[25] |
|
[26] |
|
[27] |
|
/
〈 |
|
〉 |