科学软件网是一个以引进国研软件,提供软件服务的营业网站,网站由北京天演融智软件有限公司创办,旨在为国内高校、科研院所和以研发为主的企业事业单位提供的科研软件及相关软件服务。截止目前,科学软件网已获得数百家国际软件公司正式授权,代理销售科研软件达一千余种,软件涵盖领域包括经管,仿真,地球地理,生物化学,工程科学,排版及网络管理等。同时,还提供培训、课程(包含34款软件,66门课程)、实验室解决方案和项目咨询等服务。
. Introduction
In practice, many multivariate data sets contain missing values. These missing values may result from nonresponses in a survey, absenteeism of participants in a longitudinal study, etc. The traditional way of dealing
with these missing data values is to use list wise deletion to generate a data set that only contains the complete
data cases. However, list wise deletion may result in a very small data set. It is a well-known fact that most
multivariate statistical methods require a large sample size, especially if the number of observed variables is
large. Consequently, alternative statistical methods for dealing with data with missing values are of interest.
Multiple Imputation (MI) and Full Information Maximum Likelihood (FIML) estimation are two popular
statistical methods for dealing with data with missing values. Both these methods are available in LISREL
(Jöreskog & Sörbom 2003). The Multiple Imputation module of LISREL implements the Expected
Maximization (EM) algorithm and the Markov Chain Monte Carlo (MCMC) method for imputing missing
values in multivariate data sets. Technical details of these methods are available in Schafer (1997) and Du
Toit & Du Toit (2001). Supplementary notes on these methods are also provided by Du Toit & Mels (2002).
In this note, the Multiple Imputation and FIML methods for data with missing values of LISREL are illustrated
by fitting a measurement model to a multivariate data set consisting of the scores of a sample of girls on six
psychological tests. This data set is described in the next section. The measurement model is described in
section 3. Thereafter, the method of Multiple Imputation is used to fit the measurement model to the data set
for girls. In section 5, the measurement model is fitted to the girls’ data by means of the FIML method
Introduction
In practice, the variables of interest are often latent (unobservable) variables, such as intelligence, job
satisfaction, organizational commitment, socio-economic status, ambition, alienation, verbal ability, etc.
These latent variables are modeled by specifying a measurement model and a structural model. The
measurement model specifies the relationships between the observed indicators and the latent variables
while the structural model specifies the relationships amongst the latent variables. However, it is also
possible and often desired to include observed variables as part of the structural model.
LISREL (Jöreskog & Sörbom 2006) implements the Maximum Likelihood (), Robust Maximum
Likelihood (RML), Generalized Least Squares (GLS), Un-weighted Least Squares (ULS), Weighted Least
Squares (WLS), Diagonally Weighted Least Squares (DWLS) and Full Information Maximum Likelihood
(FIML) methods to fit structural equation models to data. More information on these methods is provided
in Jöreskog & Sörbom (1999) and Du Toit & Du Toit (2001).
In this note, the method of LISREL is used to fit a structural equation model to the values of a sample
of school children on 10 observed variables. The data set is described in the next section. The structural
equation model is described in section 3. In section 4, the structural equation model is fitted to the data by
means of the method. The LISREL output file is reviewed in section 5
LISREL不仅能处理结构方程模型,还能用于其他统计应用中:
LISREL:结构方程模型
PRELIS:数据处理与基本统计分析
MULTILEV:分层线性和非线性建模
SURVEYGLIM:广义线性模型
MAPGLIM:多级数据的广义线性建模
处理数据类型包括
分类和连续变量的完整和不完整的复杂调查数据
分类变量和连续变量的完全不完全随机样本数据
科学软件网的客户涵盖产品涵盖教育、、交通、通信、金融、保险、电力等行业,并且为诸如北京大学、*大学、中国大学、中科院、农科院、社科院、环科院、国家、交通部、南方电网、国家电网、许继、南瑞等国内大型企事业单位、部委和科研机构长期提供相关产品。我们的品质,值得您信赖。