NotesFAQContact Us
Search Tips
ERIC Number: ED531719
Record Type: Non-Journal
Publication Date: 2011-May-23
Pages: 54
Abstractor: ERIC
Reference Count: 25
Analyzing Multilevel Data: An Empirical Comparison of Parameter Estimates of Hierarchical Linear Modeling and Ordinary Least Squares Regression
Rocconi, Louis M.
Association for Institutional Research (NJ1), Paper presented at the 2011 Association for Institutional Research Annual Forum (Toronto, ON, May 23, 2011)
Hierarchical linear models (HLM) solve the problems associated with the unit of analysis problem such as misestimated standard errors, heterogeneity of regression and aggregation bias by modeling all levels of interest simultaneously. Hierarchical linear modeling resolves the problem of misestimated standard errors by incorporating a unique random effect for each institution into the statistical model; moreover, the variability in these random effects is taken into account in estimating the standard errors. Until the advent of HLM, heterogeneity of regression had often been viewed as a methodological nuisance. However, the cause of heterogeneity of regression is often of substantive interest. HLMs enable a researcher to estimate a separate set of regression coefficients for each higher level organizational unit and then model variation among the higher level units in their sets of coefficients as multivariate outcomes to be explained by higher level factors. HLMs solve the problem of aggregation bias by modeling each level of the hierarchy with its own model. Today, many higher education scholars are rushing to use this new, sophisticated analytic procedure. This rush seems to be based on the assumption that HLM might yield substantively different findings than those from studies based on ordinary least squares (OLS) regression analyses. With this in mind, the current study investigates the different conclusions that may be drawn depending upon the type of analysis chosen. This paper focuses on the three types of analyses discussed above. The first analysis will be an OLS regression with the student as the unit of analysis, the second analysis will be an OLS regression with the student level variables aggregated to the institutional level with the institution as the unit of analysis, and the third analysis will be a three-level hierarchical linear model with student characteristics modeled at Level 1, characteristics about the major modeled at Level 2 and characteristics of the institution modeled at Level 3. Appended are: (1) Items comprising the variables used in the analyses and the construction of scales; and (2) List of Majors and Biglan (1973a, 1973b) classification. (Contains 8 tables.)
Association for Institutional Research. 1435 East Piedmont Drive Suite 211, Tallahassee, FL 32308. Tel: 850-385-4155; Fax: 850-383-5180; e-mail:; Web site:
Publication Type: Reports - Evaluative; Speeches/Meeting Papers
Education Level: Higher Education; Postsecondary Education
Audience: N/A
Language: English
Sponsor: N/A
Authoring Institution: Association for Institutional Research
Identifiers - Assessments and Surveys: National Survey of Student Engagement