- 期刊
小區域人口推估研究：臺北市、雲嘉兩縣、澎湖縣的實證分析
A Study of Small Area Population Projection in Taiwan
摘要
一個國家的建立及發展與該國人口有關，依各地區人口成長趨勢推動相關政策，方能制定適合當地特性、符合經濟效益的政策。臺灣近年人口老化日益明顯，各縣市老化速度及人口結構不盡相同，若能獲得各地區未來人口相關資訊（亦即人口推估），再根據地區特性規劃需求，可減輕未來人口老化對臺灣造成的衝擊。本文即以縣市層級的人口推估，也就是小區域人口推估為目標，希冀提供臺灣各地未來政策規劃的參考。 本文考量的小區域人口推估，使用人口要素變動合成法(cohort component method)，加入生育、死亡、遷移三個因素，以臺北市、雲嘉兩縣、澎湖縣三個人口數不同的地區為範例，測試縣市層級的人口推估。對於生育、死亡、遷移三要素的未來數值推估，介紹不同模型，包括區塊拔靴法(block bootstrap)、篩網拔靴法(sieve bootstrap)、Lee-Carter模型及函數主成份分析(functional principal component analysis)。以估計誤差為衡量標準，比較年齡別死亡率，發現篩網拔靴法、區塊拔靴法、Lee-Carter模型三者的結果較佳；再將區塊拔靴法用於小區域推估，發現遷移對小區域人口扮演重要的角色，而且三個地區的人口老化速度也不一致，此與全國規模的人口推估結果截然不同。研究也發現人口三要素間是否相關也有影響，在傳統的人口推估通常假設三要素互相獨立，這種假設得出的預測區間遠小於三要素不獨立的結果。
並列摘要
The government can engage in better policy planning if the future population and its structure can be known in advance. This is especially true for local governments (e.g. at the county or township level) due to limited resources. Like many countries, Taiwan has been experiencing rapid population aging, and the allocation of resources has become very important. The aim of this paper is to find an appropriate method for projecting populations of small areas in Taiwan. First, we use 3 areas (county level) of Taiwan: Taipei City, Yunlin & Chiayi, and PengHu, to explore whether the cohort component method can be used for small area population projection. Also, to decide the future values of birth, death, and migration, we experiment with several probabilistic methods, including the Lee-Carter model, block bootstrap, sieve bootstrap, and functional principal component analysis. The block bootstrap is chosen for its simplicity and good accuracy. Combining the block bootstrap with the cohort component, we found that migration plays an important role in small area projection, while birth is more important in country level projection.