影響台灣總人口數的因素

Abstract

在此報告中，探討各項反應變數對解釋變數(台灣總人口數)具有較大的解釋能力，先對原始資料做Excel迴歸分析後，去除相關係數不符合範圍內者，查看常態圖是否為45度、殘差圖是否以0作為界線，上下均分。 對原始資料做迴歸分析，本組仍保留其分析結果，原先的七個變數利用SAS檢定，但初步結果發現解釋變數中的自然增加人數為出生人數減去死亡人數、總增加人數為自然增加人數加上社會增加人數，因此將自然增加人數和總增加人數從變數中剔除，再利用以下三種分析方法：第一為向前選取法(Forward Selection)，第二為向後消去法(backward method)，第三為逐步迴歸法 (Stepwise)。發現「出生人口數」解釋變數不具有線性迴歸關係，進而減少一個變數，才可增加其餘變數之解釋能力。 將遺漏值加以刪減，漸進推出正確公式，交叉比較變數之結果，得出最具代表性之線性迴歸線。再以殘差分析檢定資料是否為常態，變異數是否為常數及迴歸模型是否為最佳的線性迴歸模型，如此才能加以將變數數據化，其影響程度一覽無遺。 最後所得到的結論為：在上述三種方法的篩選後，留下具可信的變數：死亡人數、社會增加人數、車禍死亡人數、出國人數，此四種變數為本組所選變數下影響總人口數較大之因素。

In this report, explore each response variable on the independent variables(total population of Taiwan) has a greater explanatory ability, doing the regression analysis with the original data by Excel, then remove the correlation coefficient those who does not comply with range, check whether the regression curve was 45 degrees、whether residual plots with 0 as the boundary and evenly distributed. Doing the regression analysis on the original data, we still retain its analysis and analyze the original seven variables with SAS. However, preliminary results showed that the natural increase in the number of explanatory variables is equal to the number of births minus deaths、the total increase in the number equal to the natural increase in the number plus the increase in the number of social. Therefore, the natural increase in the number and total increase in the number were removed from the variables, then use the following three analysis methods-the first method is “Forward Selection”, the second method is “Backward method”, the third method is “Stepwise”. We found that "the number of births." This variable does not have a linear regression relationship, so it is removed in order to increase explanatory ability of the remaining variable. We delete the missing values in the data, then using the three test methods to select the best variable, it obtained to the most representative of the linear regression line. Using the residual analysis to test whether residual sum is zero and whether the model is the best linear regression model, so in order to obtain the best variable. The final conclusion- after the filtering of the three methods, leaving credible variables“the number of deaths”、“ the increase in the number of social”、“the number of car accident deaths”、“the number of people abroad”. The following of four variables are factors that we selected affect the total population of relatively large.