傳統的統計技術不外乎就是以迴歸方法探討房屋價格與特徵間的關係,過去有相當多的文獻資料在探討房屋價格的組成,以特徵價格法(Hedonic Price Method)估計房屋特徵價格,但是此方法有一缺點,其為將所有的區位條件以「同質區」的方式處理,尤其針對外部環境,僅是得到房屋特徵與房價之間的「單一數據」關係,而非特徵屬性在空間上因區位差異與每一筆房價的關係,這種無法藉由空間上的落實來了解各屬性區位對於房價的影響,在空間統計領域稱之為「全域式」(Global)觀點的分析方法,而這種將空間中房屋特徵與房價間的關係假設為是為「穩定(Statinary)」的,但其實兩者的關係是不穩定的,因為每筆房價所座落的建物,都是位於空間上,而空間每天都會產生不同的變化,可能隨著周邊設施有不同的變動,或者因為某些因素而吸引相當多人來此購屋而產生聚集等等,且每一座房屋具有不一樣的內部特徵,卻以同一個係數解釋之,並不客觀,因此為了修正因為全域式分析而被忽略的訊息,也發展出地方性(Local)的分析觀點。地方性的觀點加入空間區位的「座標」條件,而本研究所使用的地理加權迴歸(Geographically Weighted Regression)為空間分析中的一種方法,其為地理學領域所發展出來,因此一定具備有空間的概念,透過頻寬的設置,加權在房屋樣本周邊一定比例的樣本。 從文獻回顧中得到比較特徵價格法與地理加權迴歸兩種方法多得到後者有較高的影響力,且可以找到某些在全域型裡被忽略掉的空間變數與房屋特徵價格所呈現的空間不穩定性,藉此方法來分析台北市的房屋特徵價格之空間分佈型態,並可以GIS的技術透過圖面表達這些房屋特徵的空間變化,。 透過分析及空間型態的展示發現地理加權迴歸的解釋能力高於特徵價格法,且在傳統迴歸分析中得到幾個具有顯著性的變數,在地理加權迴歸當中轉變為無明顯針對哪些變數呈現不顯著而說明,因為每個變數針對不同房屋樣本座落位置會有各個不同變數的顯著性,以及房屋特徵價格的確是呈現空間不穩定性的情形,並可以從GIS所呈現出來的圖上看出整個空間的變化情形。
Traditional statistics technology is nothing more than to probe into the relation among the price of the house attribute by regression method. There are a lot of paper that was probed into the price of the house in the past, and no matter the inside, outside or social economic attributes of the house, mostly using hedonic price method to estimate the hedonic price of the house, but this method has a shortcoming, that is all position conditions by way of “Homogeneity district “, especially directs against the external environment condition, only get ' the single datum ' relation between the housing attributes and price, is not a spatial relation, this is called Global analysis, and that assumes there are a stationary between price and attributes, actually that is non-stationary because of the housing price representing the building value, but it is located in the space. The space changes everyday because of many reason, in order to calibrate some neglected information like that, it is developed in geography, and called “Local” analysis. Geographically Weighted Regression is one of the local analysis method through the coordinate and bandwidth, and observations in proximity will be weighted. From the past paper to find out comparison of geographically weighted regression and hedonic price method, most of these can get better explanation power in the former, also find out some variations are spatial variations, and housing attribute price is non-stationary in the space, so I will use Taipei housing market to be an approach of Geographically Weighted Regression. And I get the explanation power of geographically weighted regression is higher than hedonic price method, find out some variations are spatial variations, and housing attribute price is non-stationary in the space using GIS to present the spatial pattern.