HLM 7

Anthony S. Bryk - University of Chicago
Stephen W. Raudenbush - Michigan State University
Richard T. Congdon, Jr - Harvard University

New in HLM 7

• Four-level nested models
• Four-way cross-classified and nested mixtures
• Hierarchical models with dependent random effects

HLM, Hierarchical Linear Modeling, allows for the analysis of models with two to four levels of nesting (i.e., multilevel analysis). Such nested models may be used to analyze growth and change within individuals; to study responses of persons in organizations such as schools, businesses, community, and religious groups; and to conduct meta-analysis of research results.

The HLM program handles both hierarchical linear and nonlinear models. All models can easily be formulated within this Windows version, as HLM leads the user step-by-step through the specification of the model at the respective levels. After specifying the variables at each level, the relevant equations for each level are immediately constructed and displayed on screen.

All models can be formulated in Windows. As you specify variables at each level, the relevant equations for each level are immediately constructed in a graphics box. These are saved and can be easily modified for subsequent analysis. Data are also easily read into HLM using Windows.

HLM interfaces with widely available statistical packages: HLM 4 can read data from a variety of statistical packages, Including SPSS, SAS, SYSTAT, and STATA, to construct the input file. HLM outputs residual files that can immediately be read into these packages. Thus, all of the familiar exploratory analysis methods, data transformations, and graphical capabilities of these packages are readily available.

Three kinds of parameters may be estimated: empirical Bayes estimates of randomly varying lowest level(s) coefficients; generalized least squares estimates of the level-2 coefficients, or maximum-likelihood estimates of the level-3 coefficients; and maximum-likelihood estimates of the variance-covariance components.

HLM allows estimation of Bernoulli and binomial models for binary data with logit link function and Poisson models for count data with constant or variable exposure with log link function. Estimation is available for two- and three-level models with and without over-dispersion. Data may be at the person level or grouped by covariate set.

HLM provides estimation of population-average models using Generalized Estimating Equations (GEE) with and without robust standard errors as described by Zeger, Liang, and Albert (1988).

To maximize the likelihood, HLM uses acceleration via the Fisher scoring algorithm. The Fisher/EM combination produces a fast and reliable convergence for both 2- and 3-level programs.

Full maximum likelihood for 2- and 3-level hierarchical linear models and full penalized quasi-likelihood estimates for hierarchical generalized linear models are accompanied by standard errors for variance-covariance components.

HLM prints standard errors and t-tests for each of the fixed effects, as well as chi-square tests of homogeneity of each random effect. A residual file is produced which can be used for checking model assumptions. HLM contains options to perform multivariate hypothesis tests for the fixed effects and for the variance-covariance components, to estimate models without a level-1 intercept, and to constrain the fixed effects to any positive number.

Replicated analyses for multiply imputed data sets such as the National Assessment of Educational Progress, the National Adult Literacy Survey, and the International Adult Literacy Survey are available for the two-level model.

Student Version of HLM 6

The student edition of HLM 6 is available as a single, self-extracting executable, "HLM6Student.exe".

• Download this file from the link below to a temporary folder (for example C:\tempfiles).
• Run "HLM6Student.exe" by double-clicking on it.
• Default installation will be to a new folder "C:\Program Files\HLM6S". You may change the name and location of this folder. After successful installation the downloaded file "HLM6Student.exe" may be deleted.

The student edition contains the following:

• All the examples distributed with the full HLM 6 version. These examples may be run with the student edition.
• An on-line helpfile, as provided for the full version too. The helpfile includes most of the new HLM 6 manual and a complete tutorial showing the use of HLM.

The student edition can run all the analyses the full version can in terms of models selected, statistical options and output. Restrictions are, however, placed on the data used and the size of the model selected. The following restrictions apply in this edition:

• The STAT/Transfer utility used for the importation of data is not included. The student edition will only accept ASCII, SYSTAT, SPSS for Windows or SAS transport data files. Note: SPSS data files created with SPSS 16 or earlier can be used with the student edition.
• For a level-3 model, the maximum number of observations that may be used at levels 1, 2 and 3 are approximately 7500, 1700 and 60 respectively. Note that the restriction applies to observations in the case of the level-2 file, for example, and not to actual number of level-2 units to be included in the analysis.
• For a level-2 model, the maximum number of observations at the two levels are 7200 at level-1 and 350 at level-2 of the hierarchy.
• No more than 5 effects may be included in any HLM equation at any level of the model, and the grand total of effects can not be 25 or higher.

For more information on how to set up these models and how to interpret the output, please see the on-line Help file that comes with the program.

• Download the HLM 6.06 Student version (15 MB)