傳統上國人寄望養兒防老，但面臨少子化及壽命延長，家庭已無法獨力負擔照顧老年人的責任，必須仰賴個人（老年人自己）、國家及政府分擔人口老化造成的需求，這也是政府在過去二十年來積極投入更多資源，制訂與老年人有關的社會保險、福利及政策的原因。像是1995年開辦的全民健康保險提升了全民健康，其中老年人受惠尤多；2005年的勞工退休金條例、2008年的國民年金保險等，則是因應我國國民壽命延長的社會保險制度。對於未來費用的需求估算，需要依賴可靠的死亡率預測，但大多數預測並沒有將死亡率改善列入考量，勢必低估長壽風險的衝擊，影響個人的財務規劃、增加國家負債。 有鑑於此，本文研究常用的死亡率模型，評估哪些適合用於描述高齡死亡率的變化，且能用於計算年金商品的定價。本文考量的模型大致分成兩類：關係模型(Relational Models)及隨機模型(Stochastic Models)，第一類包括常用於高齡的Gompertz、Coale-Kisker模型，以及Discount Sequence模型，第二類則有Lee-Carter及CBD等模型。模型比較的方式以長期預測和短期預測，選用交叉驗證的方式驗證死亡率模型的預測結果與觀察值之間的差異。研究結果顯示Discount Sequence、Lee-Carter、CBD隨機模型較能準確描述台灣、日本與美國等三個國家的死亡率特性；但這三個模型在年金險保費並沒有很明顯的訂價差異。另外，若用於短期預測、長期預測比較，又以Discount Sequence的預測結果優於Lee-Carter模型的預測。 / Traditionally in Asia, families played the main role in caring their own elderly (i.e., parents and grand-parents), but the declining fertility rates and longer life expectancy make it difficult for the families to take care of the elderly alone. The elderly themselves and the government need to share the burden caused by the aging population. In fact, most Taiwan’s major social policies in the past 20 years are targeting the elderly, such as National Health Insurance, Labor Pension Act and National Pension Insurance. Their planning and financial solvency rely on reliable mortality models and their projections for the elderly population. However, many mortality models do not take into account the mortality improvements and thus underestimate the cost. In this study, we look for elderly mortality models which can reflect the mortality improvements in recent years and use them to price the annuity products. Two types of mortality models are of interest: relational models and stochastic models. The first group includes the Gompertz model, Coale-Kisker model and Discount Sequence; the other group includes the Lee-Carter and CBD models. We utilize these mortality models to project future mortality rates in Taiwan, Japan and U.S., along with the block bootstrap and ARIMA for projection. The model comparison is based on cross-validation, and both short-term and long-term projections are considered. The results show that the Discount Sequence, Lee-Carter model and CBD model have the best model fits for mortality rates and, for the short-term and long-term forecasts, the Discount Sequence is better than the Lee-Carter model.

壽命延長及生育率下降使得人口老化日益明顯，成為全球多數國家在21世紀必須面對的議題，由於各區域人口老化的速度不同，必須根據各地特性而調整因應對策。其中研究死亡率變化為面對人口老化的必備課題，尤其是高齡族群的死亡率，這也是近年高齡死亡模型廣受重視的主因之一。因為樣本數與變異數成反比，人口較少的區域或是高齡人口，死亡率的觀察值通常會有較大震盪，為了降低震盪多半會經過修勻，以取得較為穩定的死亡率推估值（王信忠等人，2012）。此外，Li and Lee (2005)的Coherent Lee-Carter模型也是另一種可行方法，透過參考大區域的資訊降低小區域的估計誤差。 本文探討結合上述修勻、死亡率模型的可能，希冀能綜合兩者的優點，提高小區域死亡率推估的精確性。因為Coherent Lee-Carter模型的想法類似增加小區域的人數（加入大區域的人數），本文探討人口數與Lee-Carter模型參數估計值的關係，再以修勻調整大小區域的差異，透過電腦模擬及資料分析，驗證本文提出方法是否有效。其中，仿造王信忠等人的作法，假設小區域與大區域死亡率間的七種可能情境，以平均絕對百分誤差(Mean Absolute Percentage Error)為衡量標準，找出調整修勻、相關模型的方法。另外，本文也以臺灣縣市為研究區域，驗證本文方法的估計結果。研究發現適當地使用修勻方法，可降低小區域的死亡率估計值，其效果優於Coherent Lee-Carter模型。

在生態學、生物學、和比較文學的研究中，物種個數通常是評估生物多樣性的重要指標，單一群落物種數的估計已有非常豐富的相關研究。較為知名者包括Good (1953)提出未出現物種的機率，作為估計物種數的參考，往後Good的想法被大量延伸，推演出不少新的估計方法，像是Burnham and Overton (1978)的Jackknife估計法，Chao and Lee (1992)利用涵蓋機率的估計。相對而言，兩群落共有物種數的研究較少，現有研究中較為知名的有Chao et al. (2000)的估計式。 本研究延伸Good想法，探討Jackknife估計法在兩群落的應用，以出現一次的共有物種（一階Jackknife估計），推估未出現共有物種機率，並且仿造Burnham and Overton的想法，建立共有物種數的估計值及變異數。本文除了以電腦模擬，也使用實例（包括：金庸武俠小說、台灣野生水鳥、巴拿馬螃蟹和巴洛科羅拉多森林）檢驗本文的Jackknife估計法，利用涵蓋機率角度發現抽出某特定比例樣本時，估計值涵蓋母體共有物種數之機率值達到九成以上，且也與Chao提出的估計值比較。 / The number of species is frequently used to measure the biodiversity of a population in ecology, biology, and comparative literature. There are quite a lot of studies related to estimating the number of species. Among these studies, Good (1953) proposed a famous estimate (Turing’s estimate) for the probability of unseen species. Subsequently, many methods have been proposed for estimating the number of species based on Good’s idea. For example, the Jackknife estimator by Burnham and Overton (1978) and sample coverage probability by Chao and Lee (1992) are two famous estimates for the number of species. In contrast, there are not many studies for the number of shared species in two communities, and Chao et al. (2000) is probably the only one. This article extends Good’s idea and the Jackknife method to estimate the number of shared species in two communities. Similar to Burnham and Overton, we establish the estimate and its estimated variance, based on the number of species appearing exactly once. We also use computer simulation and real data sets (Jin-Yong martial arts novels, Taiwan wild birds, Panama crustacean, and Barro Colorado Island forest) to evaluate the proposed method. We found that the coverage probability for confidence interval covering the true number of shared species is more than 90%. In addition, we compare the proposed method with Chao’s method.

一個國家對全國人口有充分瞭解，方能依據國情制定適合的政策，地方發展更是如此，更須洞悉各地的人口結構，以善用有限的資源。台灣近年人口老化日益明顯，各縣市的老化速度及人口問題也不盡相同，若可獲得各地區未來的人口相關數值 (亦即人口推估)，當能減輕未來人口老化對台灣造成的衝擊。本文以縣市層級的人口推估，也就是小區域人口推估為研究目標，探討需注意的事項，尋找適合台灣地區的小區域推估方法。 本文整理小區域人口推估方法，並使用人口要素變動合成法 (Cohort Component Method)，以雲嘉兩縣、臺北市、澎湖縣為範例，測試縣市層級的人口推估。人口推估與生育、死亡、遷移三者的假設有密切關係，我們以死亡率為目標，比較不同模型的優劣，考慮的模型包括 Lee-Carter 模型、區塊拔靴法 (Block Bootstrap)、篩網拔靴法 (Sieve Bootstrap) 以及泛函資料分析 (Functional Data Analysis) 中的主成份分析 (Principle Component Analysis)，以估計誤差為衡量方法優劣的標準。分析發現篩網拔靴法、區塊拔靴法、Lee-Carter 模型三者的結果較佳，因此在小區域推估中使用較簡便的區塊拔靴法。研究發現對小區域的人口推估而言，遷移假設扮演非常重要的角色，此與全國規模的人口推估結果截然不同。研究過程亦發現人口三要素對人口推估有明顯的影響，若假設三要素間互相獨立 (也就是傳統推估時的假設)，推估結果的預測區間遠小於三要素不獨立。 / The government can make policy according to the population change in this country, while the local government can develop their district by using their limited resources well after realizing the populaton structure. The population ageing is becoming more serious and being more different among every counties in Taiwan day by day. If we can get the relative numbers of population in the future (population projection), we can decrease the attack of population ageing for Taiwan. The aim of this paper is to find an appropriate method and some notations of small area population projection in Taiwan. The paper includes the summary of methods of small area population projection and the results by using cohort component method on three areas in Taiwan, YunLin & ChiaYi, Taipein City and PengHu. Population projection is highly related with birth, death and migration, hence we test the mortality rate by using several methods, Lee-Carter, block bootstrap, sieve bootstrap and principal component analysis of functional data analysis are included. We found that the result of sieve bootstrap, block bootstrap and Lee-Carter are much better than the others, therefore, we take block bootstrap which is much simpler than the other two to analysis the effect of birth, death and migration in population projection. The sutdy found that, in small area population projecton, migration plays an important role, which is totally different from the whole country population projection.

由於台灣地區的生育率變化較大，之前的研究發現其他各國的生育率模型不見得適用，亟需建立可反映我國國情的生育率模型。本文引用台灣、日本、荷蘭、美國（亞洲、歐洲、美洲）等經濟發達國家的出生資料，配適包括Gamma、Lee-Carter、主成份分析、年齡組個別估計法、擴散模型等較為常用的模型，比較這些國家配適結果的異同。分析發現，如果要預測總生育率，台灣、日本、美國都是以年齡組生育率個別配適擴散模型的結果最佳，荷蘭則是年齡組個別估計法；在年齡組生育率的預測方面，台灣、日本、荷蘭都是以年齡組生育率個別配適擴散模型最好，美國則是以年齡組個別估計最好。此外再從相對穩定性或相對效率的角度來評判，一樣是以年齡組生育率個別配適擴散模型或年齡組個別估計的總生育率預測結果最佳。最後還觀察到台灣地區和各縣市的有偶婦女比例和生育率呈正向關係，平均生育年齡和生育率呈反向關係，各縣市在有偶婦女比例、生育率、平均生育年齡的變化並不一致，各年齡組有偶婦女比例和生育率的改變也不盡相同。

在觀察性研究或非隨機試驗研究中，欲探討因果效應時，研究者需要重新對觀察性研究進行設計，設計目的在於重新建立一個隨機指派受試者的機制，使其得以近似一個隨機試驗研究，這樣的研究一般稱為「類隨機試驗研究」(quasi-randomized-experiments)。 傾向分數分析即為一種設計觀察性研究的方法，在不牽涉到反應變數結果之下進行設計。本文於一個病例對照研究(case-control study)中使用傾向分數進行配對接著再進一步估計處理效果，傾向分數配對是可降低觀察性研究中的選擇性偏誤的方法，透過配對可減少實驗組與對照組間的系統性差異，使研究群體在所觀察到的控制變數分配達到相似，進而得到處理效果(treatment effect)的不偏估計，為近年廣受流行病學、經濟學以及社會學領域使用的方法之一。傾向分數本身為一個條件機率，定義為研究受試者在其所觀察到的控制變數之下，接受某處理或被指派至某特定群體的機率，估計傾向分數最常見的方法為羅吉斯迴歸。 此外，自1970年代起，配對方法(matching method)開始被使用來選取合適的實驗組與對照組並進行兩群體的比較，其中「確切配對」屬於最常使用的配對方法，過去文獻中經常可見各種配對方法的結合使用，因此，本文電腦模擬研究部份，欲比較四種情境之下「傾向分數配對」與「確切配對」結合使用的效果，分別以偏誤降低比例、信賴區間覆蓋率、均方誤衡量兩種配對方法結合使用的適合情境。結果顯示若對「與處理指派中度相關的變數」且「與反應變數高度相關的變數」，其效果最為明顯。根據結果，我們總結認為「確切配對與傾向分數配對合併使用」確實會有較好的表現，但表現的好壞也取決於確切配對的變數。實證研究部份，探討家庭結構對青少年偏差行為之影響，欲了解來自非完整家庭之青少年是否較來自完整家庭之青少年更有容易出現偏差行為。 / In observational or nonrandomized studies, treatments are not randomly assigned so that baseline differences between treated and control groups are typically observed. Without properly executed, the differences would bias the treatment effect estimates. There has been a long history of using matching to eliminate confounder bias, and inferences are made based on the matched observations. The theoretical basis for matching has been developed since 1970, and among those matching methods commonly in use, the exact matching is probably the most popular one. On the other hand, introduced by Rosenbuam and Rubin in 1983, propensity scores, the conditional probability of being exposed or treated given the observed covariates, has been a welcome alternative used to adjust for baseline differences between study groups of late. Instead of matching a treated with an untreated subject by their covariates, subjects in both treated and control groups are matched by their propensity scores. In this study, we explore the benefits of using propensity score matching together with the exact matching for adjusting for baseline differences through Monte Carlo simulations. An empirical study is also be provided for illustration.

人口政策的制定需要人口推估作基礎。近年世界各國人口推估逐漸從專家意見推估走向機率推估，常見的機率推估分成三大類，隨機推估、模擬情境、推估誤差三種，本文所使用的人口推估方法為隨機推估法結合生育率之模擬情境方法，在人口變動要素組合法 (Cohort Component Method) 之下輔以電腦模擬的區塊拔靴法 (Block Bootstrap)，針對台灣地區與台灣北、中、南、東四地區進行人口推估。另外，本文試圖在隨機模型人口推估中加入遷移人口之考量，以期針對遷移人口在數量與其影響上都能有較深入的了解，比較區塊拔靴法與經建會推估之差異後發現遷移之考量確實會影響人口推估之結果。 / 針對與全區相符的小區域人口推估，本文亦提出可使得推估一致的方法，但其缺點為限制了生育、死亡人口要素之變動性。此推估在總數上與隨機推估方法差異不大，但在人口結構上則有明顯的差別，此差別可能是來自於死亡率在四區間差異造成。 / Population projection is important to policy making, and only with accurate population projection can the government achieve suitable policy planning and improve the welfare of the society. The most popular and well-known population projection method is the Cohort Component method, proposed since 1930’s. The trends of future fertility, mortality and migration are required, in order to apply the cohort component method. Currently in Taiwan, these trends are determined according to experts’ opinions (or scenario projection) and three future scenarios are assumed: high, median and low scenarios. One of the drawbacks in applying experts’ opinions is that the projection results of these three scenarios do not have the meaning in probability. / To modify the expert’ opinions and let the projection results carry the meaning in probability, many demographic researchers have developed stochastic projection methods. The proposed stochastic methods can be categorized into three groups: stochastic forecast, random scenario and ex post methods. In this study, we introduce these stochastic methods and evaluate the possibility of applying the methods in projecting the population in Taiwan. / In this study we use block bootstrap, a computer simulation and stochastic forecast method, to determine the trends of future fertility, mortality and migration in Taiwan, and combine it with the cohort component method for population projection in Taiwan. We compare the projection results with those from the Council for Economic Planning and Development (a scenario projection). We found that the block bootstrap is a possible alternative to the scenario projection in population projection, and the numbers of migration is small but have a non-ignorable influence on the future population. However, we also found that the block bootstrap alone might not be appropriate for population projection in small areas.

在二十世紀九十年代，學者提出地理加權迴歸（Geographically Weighted Regression;簡稱GWR)。GWR是一個企圖解決空間非穩定性的方法。此方法最大的特性，是模型中的迴歸係數可以依空間的不同而改變，這也意味著不同的地理位置可以有不同的迴歸係數。在係數的估計上，每個觀察值都擁有一個固定環寬，而估計值可以由環寬範圍內的觀察值取得。然而，若變數之間的特性不同，固定環寬的設定可能會產生不可靠的估計值。 為了解決這個問題，本文章提出CGWR(Conditional-based GWR)的方法嘗試修正估計值，允許各迴歸變數有不同的環寬。在估計的程序中，CGWR運用疊代法與交叉驗證法得出最終的估計值。本文驗證了CGWR的收斂性，也同時透過電腦模擬比較GWR, CGWR與local linear法(Wang and Mei, 2008)的表現。研究發現，當迴歸係數之間存有正相關時，CGWR比其他兩個方法來的優異。最後，本文使用CGWR分析台灣高齡老人失能資料，驗證CGWR的效果。 / Geographically weighted regression (GWR), first proposed in the 1990s, is a modelling technique used to deal with spatial non-stationarity. The main characteristic of GWR is that it allows regression coefficients to vary across space, and so the values of the parameters can vary depending on locations. The parameters for each location can be estimated by observations within a fixed range (or bandwidth). However, if the parameters differ considerably, the fixed bandwidth may produce unreliable or even unstable estimates. To deal with the estimation of greatly varying parameter values, we propose Conditional-based GWR (CGWR), where a different bandwidth is selected for each independent variable. The bandwidths for the independent variables are derived via an iteration algorithm using cross-validation. In addition to showing the convergence of the algorithm, we also use computer simulation to compare the proposed method with the basic GWR and a local linear method (Wang and Mei, 2008). We found that the CGWR outperforms the other two methods if the parameters are positively correlated. In addition, we use elderly disability data from Taiwan to demonstrate the proposed method.