臺大學術期刊資料庫NTU Journal Database

年輪變動比用於小區域人口推估的探討

A Study of Applying Cohort Component Ratio to Small Population Projection

余清祥(Jack C. Yue);王信忠(Hsin-Chung Wang);陳譽騰(Yu-Teng Chen)

國立政治大學統計學系教授(Professor, Department of Statistics, National Chengchi University);真理大學統計資訊與精算系教授(Professor, Department of Statistical Information and Actuarial Science, Aletheia University);國立政治大學統計學系碩士(Master, Department of Statistics, National Chengchi University)

類型
研究論文
卷期
63期
出版
2021年12月
頁碼
99–133
編號
NTUEJ00001977
全文限校內本刊全文只能在臺大校內網路閱讀。

你目前的連線不在校內範圍,所以無法開啟全文。書目與摘要不受限制;回到校內網路後重新整理本頁,全文就會直接顯示。

摘要

人口推估常用於制訂國家發展方針的依據,根據歷史數據及政策方向預測全國人口數及年齡結構,作為政府研擬政策及分配資源的依據。年輪組成法為我國官方的人口推估方法,需要詳細生育、死亡、遷移等資料,很難直接套用至縣市層級及以下之人口推估。小區域推估可採Hamilton-Perry法(簡稱HP法)的年輪變動比(cohort change ratio, CCR),本文以臺灣各級行政區域為研究區域,驗證HP法是否能用於推估臺灣縣市、鄉鎮市區層級的人口及結構。本文根據1975-2019年臺灣全國、縣市、鄉鎮市區的人口紀錄,透過區塊拔靴法與歷年平均法估計CCR,並運用回測法得出之平均絕對百分比誤差(mean absolute percentage error)作為評估依據。研究發現HP法可用於小區域人口,15年之內短期推估與年齡組成法相當,但推估誤差未必隨著人口數減少而增加。另外,推估時建議採用單齡推估(五齡組誤差較大),基底年數與地區特性有關,推估年數建議不超過15年。

Abstract

Population policy is essential to national development and population projection is often used to provide insightful suggestions for planning government policies and allocating public resources. The cohort component method is currently used in projecting the national level population in Taiwan, but this method requires detailed population data, such as the records of births, deaths, and migration. It is difficult to acquire these data in county and township levels and we need to seek an alternative method for the subnational population projection. In this study, we evaluate whether the cohort change ratio (CCR), proposed by Hamilton-Perry method, is suitable for the subnational population projection via backcasting the historical data in Taiwan (1975-2019). In specific, we are interested in comparing the projecting accuracy of CCR and cohort component methods, and we found that the CCR method can be used for short-term projections (e.g., 15 years or less) for county and township levels. Also, our projection errors are smaller by using the single-age data (comparing to 5-age group data), and there are little differences in using block bootstrap or weighted average to predict the future CCR.

關鍵詞

程式與 AI 助手可以改用資料介面(JSON、MCP),查詢比讀頁面快。給程式與 AI 代理的存取說明