分類變量數(shù)據(jù)分析1

1、分類變量數(shù)據(jù)分析（Categorical Data Analysis）應(yīng)用與實(shí)踐 2012-02-151一、課程簡(jiǎn)介課程緣起課程內(nèi)容及目標(biāo)課程講授方法二、課程導(dǎo)論基本概念2主要內(nèi)容：1）二分變量Logit模型（Binary Logit model）及二分變量模型的應(yīng)用實(shí)例；2）列聯(lián)表與Loglinear模型；3）定序變量Logit模型（Ordinal Logit model）及 定序變量Logit模型的應(yīng)用實(shí)例討論；4）多項(xiàng)分類變量Logit模型（Multinomial logit model）及多項(xiàng)分類變量Logit模型的應(yīng)用實(shí)例等。3本課程注重培養(yǎng)學(xué)生的實(shí)踐能力和實(shí)際應(yīng)用。要求學(xué)生具備基本

2、統(tǒng)計(jì)知識(shí)和方法并能夠使用統(tǒng)計(jì)軟件。通過(guò)本課程的學(xué)習(xí)，掌握社會(huì)科學(xué)中最常用的分類數(shù)據(jù)分析方法，并可以較熟練地進(jìn)行實(shí)際應(yīng)用，提高學(xué)生定量研究方法的分析技能。 4授課方式：本課程以研究生為對(duì)象。講課形式以教師和學(xué)生共講、討論為主。要求學(xué)生在課前認(rèn)真準(zhǔn)備，課上積極發(fā)言和參與討論。成績(jī)?cè)u(píng)定：課程成績(jī)將由學(xué)生出勤、課堂表現(xiàn)（發(fā)言、討論）和期末成績(jī)而確定。5參考書(shū)：1、Scott Long, J. Scott. 1996. The Analysis of Categorical and Limited Dependent Variables. New York: SAGE2、Daniel A. Powers

3、 & Yu Xie 2008. Statistical Methods for Categorical Data Analysis ACADMIC PRESS（中文版）3、王濟(jì)川, 郭志剛 Logistic回歸模型 方法與應(yīng)用 高教出版社，6二、課程導(dǎo)論2.1 Categorical VariableWe define categorical variables as those variables that can be measured using only a limited number of values or categories.Vs Continuous Variable特征屬

4、性、行為選擇及結(jié)果、態(tài)度意愿等72.2 Categorical Dependent VariablesDependent Variable (response, outcome, endogenous)Independent Variable (explanatory, predetermined, exogenous)因變量的測(cè)量類型決定了研究者該選何種統(tǒng)計(jì)分析模型或方法。82.3 Types of MeasurementQuantitative & Qualitative92.4 Basics of RegressionThree interpretations of Regression

5、Causation Observed=True Mechanism + DisturbancePrediction Observed=Predicted + ErrorDescription Observed=Summary + Residual102.5 準(zhǔn)確（accuracy）與簡(jiǎn)約（parsimony）There is always a tension between accuracy and parsimony.Occams Razor: Principle of parsimony.The principle states that one should not make more

6、assumptions than the minimum needed. 112.6 Types of RegressionCaseD.VInD.VM. of A.Exam.1ContinuousContinuousCorrelation, LRF.Inc.on IQ2ContinuousCategoricalANOVA,LRGender on Inc.3CategoricalContinuousLogit/Probit RF.Inc on Edu. 4CategoricalCategoricalLoglinear , Logit/Probit RRace on Edu.12例1 教育水平與生

7、育意愿 | des edu | 0 1 2 3 | Total-+-+- 1 | 24 257 960 56 | 1,297 2 | 58 987 2,897 73 | 4,015 3 | 52 1,565 3,144 40 | 4,801 4 | 6 306 555 7 | 874 -+-+- Total | 140 3,115 7,556 176 | 10,987 13a. LR Model1. regress des edu fweight=freq Source | SS df MS Number of obs = 10987-+- F( 1, 10985) = 115.66 Mode

8、l | 30.2967799 1 30.2967799 Prob F = 0.0000 Residual | 2877.59218 10985 .261956502 R-squared = 0.0104-+- Adj R-squared = 0.0103 Total | 2907.88896 10986 .264690421 Root MSE = .51182- des | Coef. Std. Err. t P|t| 95% Conf. Interval-+- edu | -.0653997 .0060812 -10.75 0.000 -.0773201 -.0534794 _cons |

9、1.869079 .0158408 117.99 0.000 1.838028 1.90013-14b. LR Model2 Source | SS df MS Number of obs = 10987-+- F( 3, 10983) = 41.15 Model | 32.3179575 3 10.7726525 Prob F = 0.0000 Residual | 2875.571 10983 .261820177 R-squared = 0.0111-+- Adj R-squared = 0.0108 Total | 2907.88896 10986 .264690421 Root MS

10、E = .51168- des | Coef. Std. Err. t P|t| 95% Conf. Interval-+- _Iedu_2 | -.0645565 .0163425 -3.95 0.000 -.0965907 -.0325223 _Iedu_3 | -.1473228 .0160125 -9.20 0.000 -.1787102 -.1159354 _Iedu_4 | -.1638537 .0223927 -7.32 0.000 -.2077474 -.1199601 _cons | 1.808019 .014208 127.25 0.000 1.780168 1.83586

11、9-15c. Logit Model1 edu | 0 1 | Total-+-+- 1 | 281 1,016 | 1,297 -+-+- 2 | 1,045 2,970 | 4,015 -+-+- 3 | 1,617 3,184 | 4,801 -+-+- 4 | 312 562 | 874 -+-+- Total | 3,255 7,732 | 10,987 16Log likelihood = -6620.592749 AIC = 1.205532BIC = -88968.39586-des1 | Coef. Std. Err. z P|z| 95% Conf. Interval-+- edu | -.2780525 .0265706 -10.46 0.000 -.3301299 -.2259751 _cons| 1.564346 .0709542 22.05 0.000 1.425279 1.703414 -17d. Logit Model2Log likelihood = -6616.715701 AIC = 1.205191BIC = -88957.54102-des1 | Coef. Std. Err. z P|z| 95% Conf. Interval-+- edu_2 | -.2407289 .0763977 -3.15 0.002 -.3904