推薦HLM軟件操作簡介

1、hlm6 軟件操作中國人民大學(xué)統(tǒng)計(jì)學(xué)院陳堰平 分層線性模型 (hierarchical linear model，簡稱 hlm，又稱多層線性模型，multilevel linear model)，hlm6.0 是分層線性模型軟件，包含線性和非線性部分，可以讀取大部份統(tǒng)計(jì)軟件的數(shù)據(jù)如 spss, sas, systat及stata等等。hlm常用于社會(huì)科學(xué)和行為科學(xué)，因?yàn)樗Ｓ星短捉Y(jié)構(gòu)(nested structure)的數(shù)據(jù)，因此需用次模型(sub-model)或分層線性模型(hierarchical model)，hlm就是設(shè)計(jì)來專門解決此類問題的，hlm提供的模型包括2-level mod

2、els、3-level models、hierarchical generalized linear models (hglm)和hierarchical multivariate linear models (hmlm)等。 hlm6 的五個(gè)模塊the hlm program encompasses 5 modules that may be used to fit different types of models: the hlm2 module is used to fit two-level linear and non-linear (hglm) models. it offers

3、 the widest array of special features, output, and hypothesis testing options. the hlm3 module is used to fit three-level linear and non-linear (hglm) models. the range of features are similar to that of hlm2. the hmlm module allows estimation of multivariate normal models from incomplete data. with

4、in the framework of hmlm, it is possible to estimate models having an unrestricted covariance structure, that is a full covariance matrix.a model with homogenous level-1 variance and random intercepts and/or slopes at level-2.a model with heterogeneous variances at level 1 (a different variance for

5、each occasion) and random intercepts and/or slopes at level 2.a model that includes a log-linear structure for the level-1 variance and random intercepts and/or slopes at level 2.a model with first-order auto-regressive level-1 random errors and random intercepts and/or slopes at level 2. hmlm2 allo

6、ws for study of multivariate outcomes for persons who are, in turn, nested within higher-level units and offers similar modeling features as hmlm. the hcm2 module is used for two-level cross-classified random effects models, where lower-level units are cross-classified by two higher-level units. 原始數(shù)

7、據(jù)的格式：spss、ascii、sas、systat 學(xué)生版的限制：對(duì)于兩層模型，層1最多7200條觀測(cè)，層2最多350條觀測(cè)對(duì)于三層模型，層1、2、3最多只能有7500、1700、60條觀測(cè)每層不能超過5個(gè)效應(yīng)（effects），效應(yīng)的總數(shù)不能超過25個(gè) 下載地址：ftp:/ 兩層模型 三層模型 分層廣義線性模型（hglm） 分層多元線性模型（hmlm） hlm6的作圖功能一、兩層模型 對(duì)于ascii格式的數(shù)據(jù)，id變量必須是字符型，對(duì)于其它格式，可以是字符或數(shù)值型 層一數(shù)據(jù)必須以層二單元id分組，并在導(dǎo)入hlm前排序 如果id是數(shù)值型，范圍必須在 到 之間。如果是浮點(diǎn)數(shù)，則小數(shù)部分被忽略

8、字符型的id不能超過12位。 給定層的id必須一樣寬度關(guān)于id變量（p17） hierarchical linear models 第四章實(shí)例講解實(shí)例講解層層2數(shù)據(jù)數(shù)據(jù)層層1數(shù)據(jù)數(shù)據(jù)指定層1變量指定層2變量 保存mdm模板 生成mdm文件 查看mdm的統(tǒng)計(jì)量mdm的描述統(tǒng)計(jì)量選擇層選擇層1的結(jié)果變量的結(jié)果變量無條件模型無條件模型 bold 黑體：按總平均數(shù)對(duì)中 bold italic 黑斜體：按層-2平均數(shù)（組平均數(shù)）對(duì)中對(duì)中方式對(duì)中方式填加層填加層1解釋變量解釋變量 in the level-2 model, both the intercept and ses slope are to b

9、e modeled as dependent on the schools mean social class (meanses) and school sector (sector). 填加層填加層2的解釋變量的解釋變量混合模型混合模型using level subscripts指定層指定層1系數(shù)為隨機(jī)的或非隨機(jī)的系數(shù)為隨機(jī)的或非隨機(jī)的結(jié)果分析（與課本結(jié)果分析（與課本82頁表頁表4.5比較）比較）基本設(shè)置基本設(shè)置層層1的殘差數(shù)據(jù)的殘差數(shù)據(jù)層層2的殘差數(shù)據(jù)的殘差數(shù)據(jù) 手冊(cè)p3646 結(jié)合spss的描述統(tǒng)計(jì)和統(tǒng)計(jì)圖功能層1殘差的分析層2殘差的分析殘差分析殘差分析 層1模型在生成mdm文件時(shí)逐條刪

10、除在分析時(shí)逐條刪除analysis of multiply-imputed data（section 9.2） 層2模型，因假設(shè)是完全的，如果有缺失 impute a value刪除 methods described in section 9.2確實(shí)數(shù)據(jù)的處理（確實(shí)數(shù)據(jù)的處理（p4648）其他設(shè)置其他設(shè)置收斂的控制（收斂的控制（p49）estimation control to put constraints on fixed effects層層1異方差建模異方差建模假設(shè)檢驗(yàn)假設(shè)檢驗(yàn) models without a level-1 intercept coefficients having a random effect with no corresponding fixed effect二、三層模型數(shù)據(jù)是課本第八章數(shù)據(jù)是課本第八章data input requires a level-1 file (in our