1
Course in multilevel analysis for
grouped and longitudinal data
by
Joop Hox
Utrecht University
Social research often concerns relationships between individuals and the social contexts to which they belong. Individuals and their social contexts can be conceptualized as a hierarchical structure, with individuals nested within groups. Classical examples are educational research, with pupils nested within schools, and cross-national research, with individuals nested within their national units. Such systems can be observed at two levels, and as a result we have data with group level variables and individual level variables. To analyze such hierarchical structures, we need multilevel modeling, which allows us to study the relationships between variables observed at different levels in the hierarchical structure.
Multilevel modeling can also be used to analyze data from longitudinal research, by viewing measurement occasions as being nested within respondents. This has several advantages compared to more classical approaches to longitudinal data. In addition, multilevel models have been generalized to cover situations where data do not have a simple multilevel structure, such as cross-classified data or multiple-membership models.
This short course is intended as a basic and nontechnical introduction to multilevel analysis. It starts with a description of some examples, and shows why multilevel models are necessary if the data have a hierarchical structure. It then covers the basic theory of two- and three-level models. Next it explains how multilevel models can be applied to analyzing longitudinal data, and why and when this may be an attractive analysis approach, as compared to more classical analysis methods such as multivariate analysis of variance (Manova). Further topics are multilevel logistic models to analyze data where the outcome variable is dichotomous or a proportion, and multilevel multivariate modeling to analyze where there are multiple outcome variables.
The course includes three computer labs, where multigroup and longitudinal data are analyzed. The computer labs in the course use the SPSS Mixed procedure, which is available in SPSS starting with version 11.5. The course assumes reasonable familiarity with analysis of variance and multiple regression analysis, but prior knowledge of multilevel modeling is not assumed.
The course is based on: J.J. Hox (2002). Multilevel Analysis. Techniques and Applications. Mahwah, NJ: Lawrence Erlbaum Associates.
Time Schedule Summer School:
Thursday 28.09 Friday 29.09 Saturday 30.09 9.30 Lecture: Examples of multilevel data Lecture: Analyzing longitudinal data
12.30 Lunch Break Lunch Break 14.00 Lecture:
Two-and three level models
Lecture: Analyzing categorical data
3
Resumen
Joop J. Hox
Department of Methodology & Statistics, Utrecht University P.O.B. 80140
NL-3508 TC Utrecht, the Netherlands + 31 30 2539236
e-mail: [email protected] / homepage http://www.fss.uu.nl/ms/jh
Joop Hox is Professor of Social Science Methodology at the department of Methodology and Statistics of the Faculty of Social Sciences at Utrecht University. As Methodology chair, he is responsible for the research, development and teaching carried out at the faculty in the field of social science methods & techniques.
His research interests focus on two lines of work: data quality in social surveys and multilevel modeling. In survey methodology, he has written articles on nonresponse problems and the effects of data collection mode and interviewers on various aspects of data quality. In multilevel modeling, he has written numerous articles, book chapters, and an introductory handbook, with a newly written monograph currently in press. The two lines of research reinforce each other, for instance in using multilevel methods to model complex survey data. He has acted as coordinator for a large international study of interviewer effects on nonresponse, which combines interviewer and respondent-level data from 32 surveys carried out in nine different countries.
He graduated from the Psychological Methods department at the Faculty of Psychology of the University of Amsterdam (doctoral degree 1977; Ph.D. 1986). He was a Fulbright scholar at the Program on Social Statistics at UCLA in 1990. He worked as associate professor at the Faculty of Education at the University of Amsterdam. At present he is full professor at Utrecht University. In addition to undergraduate and graduate courses, he teaches at the international University College Utrecht.
Joop Hox has taught postgraduate courses on missing data problems at the Max Planck Institute (Berlin, 2001) and at the Zentral Archiv (Köln, 1999). He also taught postgraduate courses on multilevel modeling at the ISR Summer School (Ann Arbor, USA, 2000, 2001), the Swiss Institute for Alcohol and Drug Research (Lausanne, 1998), the Max Planck Institute (Berlin, 1997), LLWB
longitudinal modeling workshop (Potsdam, 1997), and the Instituto Europeano (Florence, 1995). He has been invited speaker, a.o., at the NISS Survey Quality workshop (Washington, DC, 2005), at the 1997 Conference of the German Gesellschaft für Klassification (Potsdam, 1997), at the Conference of the Society of Multivariate Analysis in the Social and Behavioral Sciences (Leuven, BE, 1996), and at the American Statistical Association Conferences on Survey Measurement and Process Quality (Bristol, UK, 1995) and on Measurement Errors in Surveys (Tucson, USA, 1990). He is a long-standing and active member of the International Workshop on Household Survey Nonresponse.
Throughout his career at Utrecht University, and before that at the University of Amsterdam, Joop Hox has maintained an active research program in both survey methodology and multilevel modeling. He is a member of the editorial board of Structural Equation Modeling and Using Statistics. He has acted as reviewer for national and international journals in the fields of survey methodology and statistics, and has been guest editor for special issues. He has written numerous publications, organized workshops and conferences, and edited several books. His recent publications focus on survey nonresponse, interviewer effects, survey data quality, missing data problems, and multilevel analysis of regression and structural equation models.
A list publications since 2000 is given below. A complete list of publications, some of which available online, can be found on his website at http://www.fss.uu.nl/ms/jh.
Publications, J.J. Hox (2000-2005)
Books
J.J. Hox (2002). Multilevel Analysis. Techniques and Applications. Mahwah, NJ: Lawrence Erlbaum Associates.
J. Blasius, J. Hox , E. de Leeuw & P. Schmidt (eds.) (2002). Social Science Methodology in the New Millennium. Proceedings of the Fifth International Conference on Logic and Methodology. Second expanded edition. Opladen, RG: Leske + Budrich Verlag (CD-ROM).
5
Articles
De Leeuw, E.D. & Hox, J.J. (2004). I am not selling anything: 29 experiments in telephone introductions. International Journal of Public Opinion Research, 16, 4, 464-473.
Hox, J.J. & Lensvelt-Mulders, G. (2004). Randomized Response Analysis in Mplus. Structural Equation Modeling, 11, 4, 615-620.
Maas, C.J.M. & Hox, J.J. & G.J.L.M. Lensvelt-Mulders (2004). Longitudinal meta-analysis. Quality & Quantity, 38, 381-389.
Maas, C.J.M. & Hox, J.J. (2004). The influence of violations of assumptions on multilevel parameter estimates and their standard errors. Computational
Statistics & Data Analysis, 46, 427-440.
Maas, C.J.M. & Hox, J.J. (2004). Robustness issues in multilevel regression analysis. Statistica Neerlandica, 58, 127-137.
Stoel, Reinoud D., van den Wittenboer, Godfried & Hox, Joop (2004). Including time-invariant covariates in the latent growth curve model. Structural Equation Modeling, 11, No. 2: 155-167.
Borgers, Natacha; Hox, Joop; Sikkel, Dirk (2004). Response effects in surveys on children and adolescents: the effect of number of response options, negative wording, and neutral mid-point. Quality & Quantity, 38, 17-34.
Wassenberg-Severijnen, J.E., Custers, J.W.H., Hox, J.J., Vermeer, A. &
Helders, P.J.M. (2003). Reliability of the Dutch Pediatric Evaluation of Disability Inventory (PEDI). Clinical Rehabilitation, 17, 457-462.
E.D. de Leeuw, J.J. Hox & M. Huisman (2003). Prevention and treatment of item nonresponse. Journal of Official Statistics, 19, 153-176.
N. Borgers, J.J. Hox & D. Sikkel (2003). Response quality in survey research with children and adolescents: the effect of labeled response options and vague quantifiers. International Journal of Public Opinion Research, 15, 1, 83-94.
E.D. de Leeuw & J.J. Hox (2003). The use of meta-analysis in cross-national studies. Pp. 329-345 in: J.A. Harkness, F.J.R. van de Vijver & P.Ph. Mohler (Eds.) (2003). Cross-Cultural Survey Methods. New York: Wiley.
J.J. Hox & E.D. de Leeuw (2003). Multilevel models for meta-analysis. Pp. 90-111 in: S.P. Reise & N. Duan (Eds.) (2003). Multilevel Modeling. Methodological Advances, Issues, and Applications. Mahwah, NJ: Lawrence Erlbaum
J.J. Hox, E.D. de Leeuw, G. Snijkers (2003). Nonresponse in surveys: determining the research agenda for the future. Bulletin de Méthodologie Sociologique, N.79, 47-60.
J.J. Hox & C.J.M. Maas (2002). Sample sizes for multilevel modeling. In: J. Blasius, J. Hox , E. de Leeuw & P. Schmidt (eds.) (2002). Social Science Methodology in the New Millennium. Proceedings of the Fifth International Conference on Logic and Methodology. Second expanded edition. Opladen, RG: Leske + Budrich Verlag (CD-ROM).
J.J. Hox & E.D. de Leeuw (2002). The influence of interviewers' attitude and behavior on household survey nonresponse: an international comnparison. Pp. 103-120 in R.M. Groves, D.A. Dillman, J.L. Eltinge & R.J.A. Little (Eds.) Survey Nonresponse. New York: Wiley.
J.J. Hox (2002). Review of: S. James Press & Judith M. Tanur "The subjectivity of scientists and the Bayesian approach" New York: Wiley 2001. Journal of Official Statistics, 18, 305-307.
C.A. Hartman, J.J. Hox, G.J. Mellenbergh, M.H. Boyle, D.R. Offord, Y. Racine, J. McNamee, K.D. Gadow, J. Sprafkin, K.L. Kelly, E.E. Nolan, R. Tannock, R. Schachar, H. Schut, I. Postma, R. Drost, J.A. Sergeant. (2001). DSM-IV internal construct validity: when a taxonomy meets data. Journal of Child Psychology and Psychiatry, 42, 6, 817-836.
N. Borgers & J.J. Hox (2001). Item nonresponse in questionnaire research with children. Journal of Official Statistics, 17, 321-335.
N. Borgers, B. van den Bergh & J.J. Hox (2001).Testing measurement and structural equivalence in different age groups of children. Kwantitatieve Methoden, 67, 65-80.
J.J. Hox & C.J.M. Maas (2001). The Accuracy of Multilevel Structural Equation Modeling With Pseudobalanced Groups and Small Samples. Structural
Equation Modeling, 8, 2, 157-174.
S. Kef, J.J. Hox & H.T. Habekothé (2000). Social networks of visually impaired and blind adolescents. Structure and effect on well-being. Social Networks, 22, 73-91.
N. Borgers, E. de Leeuw & J. Hox. (2000). Children as respondents in survey research: Cognitive development and response quality. Bulletin de
Méthodologie Sociologique, 66, 60-75.
P.G.M. van der Heijden, G. van Gils, J. Bouts & J.J. Hox (2000). A comparison of randomized response, computer assisted self interview, and face-to-face direct questioning: Eliciting sensitive information in the context of welfare and unemployment benefit. Sociological Methods & Research, 28, 505-537.
7 G. van den Wittenboer, J.J. Hox & E.D. de Leeuw (2000). Latent class analysis of respondent scalability. Quality & Quantity, 34, 177-191.
J.J. Hox (2000). Multilevel analysis of grouped and longitudinal data. In: T.D. Little, K.U. Schnabel & J. Baumert (Eds.). Modeling longitudinal and multiple-group data. Practical issues, applied approaches, and specific examples. Hillsdale, NJ: Lawrence Erlbaum.
