Leading With Data· a Bicycle Guide

Partial book profile

Hierarchical Linear Models Raudenbush Bryk

This book provides a comprehensive theoretical and practical guide to hierarchical linear models (HLM), a class of statistical methods for analyzing data with nested or multilevel structures, such as students within schools or repeated measures within individuals.

The full book profile is not available yet.

Get the book →

What it’s about

Researchers in the social, behavioral, and medical sciences frequently encounter data with a hierarchical structure—students nested in classrooms, employees within firms, patients within clinics, or repeated observations over time on individuals. Traditional statistical methods like Ordinary Least Squares (OLS) regression are ill-equipped to handle such data, often leading to biased standard errors and a failure to capitalize on the rich, multilevel nature of the phenomena under study. 'Hierarchical Linear Models' by Raudenbush and Bryk offers a complete and accessible solution, presenting a powerful statistical framework that explicitly models these nested structures. The book guides readers from the fundamental logic of multilevel modeling, through practical applications in organizational research, individual growth studies, and meta-analysis, to advanced topics like generalized models for non-normal outcomes, latent variables, and Bayesian inference. By learning to properly partition variance, model cross-level interactions, and improve estimation of unit-specific effects, readers will be empowered to ask more sophisticated questions and draw more valid and nuanced conclusions from their complex data.

The through-line

Who it’s for
A social, behavioral, or educational researcher, or a data analyst, who works with complex data where individuals are nested within groups (like students in schools) or have repeated measurements over time.
The problem
Traditional statistical methods (like OLS regression or ANOVA) are not designed for hierarchical data and produce flawed results, such as incorrect standard errors and misleading coefficient estimates, creating a 'unit of analysis' problem. The researcher feels frustrated, confused, and uncertain about the validity of their conclusions, worrying that they are either missing crucial insights hidden in the data's structure or drawing conclusions that are statistically indefensible.
The plan
  1. Grasp the fundamental logic of HLM by seeing how it extends familiar statistical models like ANOVA and regression.
  2. Learn the core principles of multilevel estimation and hypothesis testing.
  3. Apply the two-level model to key research areas: organizational effects, individual growth, and meta-analysis.
  4. Master techniques for assessing model assumptions and making critical modeling decisions, such as how to center predictors.
  5. Extend your capabilities to more complex scenarios using three-level models, generalized models for discrete outcomes, models for latent variables, cross-classified data, and Bayesian inference.
The payoff
The researcher can confidently analyze complex, multilevel data, producing valid and defensible results. · They can formulate and test sophisticated hypotheses about how contexts influence individuals and how relationships vary across those contexts. · Their research becomes more powerful, nuanced, and conceptually aligned with the multilevel nature of the phenomena they study.

See our guide

Related profiles we’ve built

Additional reading

  • Bayes estimates for the linear model · Lindley, D. V., & Smith, A. F. M. (1972)

    This paper is cited as providing the seminal contribution to the Bayesian estimation of linear models, introducing the term 'hierarchical linear models' and elaborating the general framework for nested data.

  • Maximum likelihood from incomplete data via the EM algorithm · Dempster, A. P., Laird, N. M., & Rubin, D. B. (1977)

    The book identifies this paper as providing the 'needed breakthrough' for estimating covariance components in hierarchical models, which was a major barrier to their practical use.

  • Random-effects models for longitudinal data · Laird, N. M., & Ware, H. (1982)

    Cited as a key early application of the EM algorithm approach to the study of growth, demonstrating how HLM could resolve long-standing problems in longitudinal data analysis.

  • Contextual analysis through the multilevel linear model · Mason, W. M., Wong, G. M., & Entwistle, B. (1983)

    Cited as a parallel key application (along with Laird & Ware) that applied the EM algorithm to cross-sectional data with a multilevel structure, demonstrating the broad applicability of HLM.

  • Multilevel statistical models · Goldstein, H. (1995)

    This is the major alternative textbook on multilevel modeling, developed in parallel. It is frequently cited, particularly regarding iteratively reweighted generalized least squares (IGLS) estimation and cross-classified models.

  • Random coefficient models · Longford, N. (1993)

    Another key textbook in the field, often cited for its development of the Fisher scoring algorithm for estimation and for work on generalized linear models.

Featured in these guides

Co-cited books