Michael J. Rovine, Ph.D. - Associate Director

Professor of Human Development

119-C Henderson Building
The Pennsylvania State University
University Park PA 16802

(814) 865-7094


Bio Statement

My current interests are in areas related to statistical modeling. In the area of structural equations modeling, I am looking at ways to estimate a number of different multilevel models as SEM. One model of particular interest is a multilevel autoregressive model that could have important implications for those collecting relatively intensive time series data. I am also working on a general model, the nonstationary autoregressive moving average model that can be used to describe essentially any latent variable model. This general model has important implications for model comparison and testing.

Another interest of mine relates to the history of statistics, in particular the contributions of the philosopher C.S. Peirce to the development of statistical methodology. My interest in this area was sparked by my attempts to develop variations of the correlation coefficient that could be used to describe effect sizes in uncontrolled studies. Looking to see whether similar work had been done in the past, I discovered an interesting history of correlation and regression that predated the better known work of Pearson and Galton.

My main focus, however, is on a more idiographic approach to the description of developmental phenomena. Along with Erik Loken (HDFS), David Nembhard (Engineering), Cynthia Stifter (HDFS) and Peter Molenaar (HDFS), I am beginning an NSF funded study to develop and apply time series models to developmental data. We will be working with a number of different models including multilevel ARMA, state space, and control models. A brief abstract for the study follows.

Research interests

  • Structural equations modeling of longitudinal data
  • Time Series approaches to modeling individual level developmental phenomena
  • Structural equations modeling approaches to linear mixed models

Research Projects

Application of Time Series Modeling Techniques for Optimal Control of Asthmatic Symptoms at the Person-Specific Level.

Michael J. Rovine (PI)
Peter C.M. Molenaar (Co-PI)
Carol Gold, Project Director
Graduate students: Siwei Liu, Tamara Goode, Lawrence Lo

This project will apply time series and control modeling techniques to determine continuous optimal medication therapy for patients with asthma, with the proximate goal to keep each patient free of disease symptoms and with the ultimate goal of maintaining healthy lung functioning. The application of these techniques will allow the investigators to identify and optimally accommodate individual differences in how drugs operate within persons over time, as well as differences due to variations in levels of stress, exercise, and exposure to environmental triggers such as smoke (first-hand and second-hand) and pollen (as measured by local meteorologists). These methods will enable the health care provider to make adjustments on an individual basis to medication dosage and/or type of medication, as well as patient-specific recommendations regarding avoidance of behaviors and contexts that trigger symptoms, so that the best outcome is achieved for each individual patient.

Asthma requires daily monitoring of lung function, which can be affected by variations in medication usage, in addition to the behaviors and triggers described above. Problems associated with inappropriate dosage and type of medication are especially important. Despite our recognition of the variability in patients' responses to various types of medications and dosage levels, little has been put into place that would monitor a patient's responses to medication in a continuing way so that proper adjustments can be made for the duration of the use of that drug therapy.

Our approach will optimize the effectiveness of ongoing medical treatment by using adaptive control techniques. These are time series modeling techniques that are used by engineers to ensure the best possible outcomes in dynamic processes. Such models can be accommodated to the needs of medicine and social sciences to increase the positive results of interventions. The basic idea is that first a criterion for determining a good outcome is established (e.g., minimum asthma symptoms with minimum medication dose). Repeated measurements are taken to monitor the outcome. As soon as the outcome deviates from the criterion, the level or content of the intervention is modified to counteract this deviation. At the next measurement occasion, the direction of the outcome is checked for improvement and, if necessary, subsequent modifications of the intervention are made. Unlike many other intervention strategies, the degree or content of the intervention can be adjusted on an individual basis.

Multilevel ARMA and Dynamic Models for the Longitudinal Study of Human Interactions.

Michael J. Rovine (PI)
Peter C.M. Molenaar (Co-PI)
Cynthia Stifter (Co-PI)
Graduate students: Siwei Liu, Katie Gates, Katerina O. Sinclair

Time series methods are used to model phenomena as varied as patterns in the weather, fluctuations in the stock market, changes in populations, quality of industrial products, patterns of sleep, physiological characteristics such as heart rate, blood pressure, and brain wave activity, and the flight of the space shuttle. The use of these methods, which are so common in the areas of engineering and econometrics and which seem so naturally suited for application in the study of human interactions have been surprisingly overlooked in the developmental sciences. In this project we adapted, extended, and where necessary, developed new methods that are particularly well-suited to developmental research and the study of human interactions. We implemented these models in ways that will make them easily accessible to developmental researchers studying human interaction. We demonstrated the utility of these methods by analyzing data from the Infant and Child Temperament Study related to infant's self regulation of emotion, and parent-infant interaction related to the parents ability to soothe a distressed child.

The main focus for this project has been the development and implementation of the Hidden Markov Model. We have been working both on implementing and improving the modeling software and in describing an optimal model for the Stifter mother-infant interaction data. Along with this, we are currently developing a menu-driven front end to the DEPMIX program that will lead users through a set of decisions that will allow them for properly implement the model. The Stifter data set follows and assesses infant who receive an inoculation. The mother and infant are videotaped and the data are coded to indicate (among other variables) the child's level of distress and the mothers attempted soothing behaviors. These data are collected at two- and six-month visits. We have just submitted a paper with based on separate common loading and transition matrices for the two month and six month data. The six month model tended to show more differentiation than the two-month model Individual differences were represented by the posterior probabilities (state transition sequence) for each dyad at each occasion of measurement. This approach allowed us to show that the characteristic pattern of soothing strategies changed from two to six months. We were able to show that different soothing patterns at each age met with success.

These results were recently submitted as:

Stifter, C., Rovine, M., Sinclair, K.(2010).
Modeling dyadic processes using Hidden Markov Models: A time series approach to mother-infant interactions during 2 and 6 month infant inoculations. Manuscript submitted to Developmental Psychology.

A second paper describes the method and compares this approach to other approaches that appear in the literature, including lagged contingency table approaches such as Yule's Q and methods based on Bakeman's sequential analytic strategies.

A chapter on the approach comparing it some other analytic approaches was accepted for publication as:

Rovine, M.J., Sinclair, K.O., Stifter, C.A.(2009).

Modeling mother-infant Interactions using hidden Markov models. In K. Newell & P.C.M. Molenaar (Eds.), Individual pathways of change in learning and development. Washington, DC: APA Press.

A new revision of the software, DEPMIX, developed by our collaborator, Ingmar Visser (University of Amsterdam) was recently released. This new version incorporates some additional requirements developed for this project. These include more flexible modeling by allowing all parameters in the model to be conditioned on a set of covariates and more flexibility in the input allowed by the program. Our website includes annotated versions of the program and the output. We are in the process of developing a menu-driven front end that will take the user through a set of steps required to properly execute a model. In addition, we are greatly expanding the diagnostics that are currently available as part of the DEPMIX routine.

In addition to the Stifter data, we are working along with other researchers to implement hidden Markov models. These include data designed to link sleep patterns with eating behavior collected by Birch and Marini through the Center for Childhood Obesity Research, data collected by Teti as part of Project Siesta that relates to infant sleep patterns, and additional data collected by Warren on mother-infant interactions. We are also developing other collaboration related to the modeling of emotional regulation and other psychological constructs through the Developmental Systems Group.

In addition to the above approach for describing patterns in both mothers' and infants' behaviors, we continued looking at Association Rule Mining (ARM) approaches to analyzing the same data. As previously stated, the goal of these analyses is to discover ( mine for) association/correlation among set of data items showing patterns in the mother's behavior that are more or less successful in determining patterns of change in the child. The work has become the focus of the dissertation of Kelly Yip under the direction of David Nembhard. She is considering some of the problems with ARM approaches including adding temporal ordering and increasing the flexibility of these models by building time and sequence dependence into the ARM approach.

Dissertations related to these projects:

Nissa Towe-Goodman - Interparental conflict and regulation in early childhood: person-oriented approaches.

Stephanie Anzman-Frasca - A Multi-method Investigation of Infant Behaviors and Weight Status.

Select Publications

Rovine, M., & Lo, L. (2011, in press). Issues and perspectives in person specific time series models. In B. Laursen, T. Little, & N. Card (Eds.), Handbook of Developmental Research Methods. New York, NY: Guilford.

Rovine, M., & Liu, S. (2011, in press). Structural equations modeling approaches to longitudinal data. In R. Jones, J. Newsom, & S. Hofer (Eds.), Best practices for data analysis of longitudinal studies of aging. New York, NY: Psychology Press.

Rovine, M. J., & Anderson, D. R. (2011). Peirce's coefficient of the science of the method: an early form of the correlation coefficient. In D. R. Anderson & C. Hausman (Eds.), A conversation on Peirce (pp 246-274). New York: Fordham University Press.

Rovine, M.J., Sinclair, K.O., Stifter, C.A. (2010). Modeling mother-infant Interactions using hidden Markov models. In K. Newell & P.C.M. Molenaar (Eds.), Individual pathways of change in learning and development, (pp 51-67). Washington, D.C.: APA Press.

Rovine, M. J., & Walls, T. A. (2006). A multilevel autoregressive model to describe interindividual differences in the stability of a process. In J. L. Schafer & T. A. Walls (Eds.), Models for intensive longitudinal data (pp. 124-147). New York: Oxford.

Rovine, M. J., & Molenaar, P. C. M. (2005). Relating factor models for longitudinal data to quasi-simplex and NARMA models. Multivariate Behavioral Research, 40(1), 83-115.

Rovine, M. J., & Anderson, D. R. (2004). Peirce and Bowditch: An American contribution to correlation and regression. American Statistician, 59(3), 232-236.

Rovine, M. J., & Molenaar, P. C. M. (2003). Estimating analysis of variance models as structural equation models. In B. Pugesek, A. Tomer, & A. von Eye (Eds.), Structural equation modeling: Applications in ecological and evolutionary biology research (235-280). New York: Cambridge.

Rovine, M. J., & Molenaar, P. C. M. (2000). A structural modeling approach to the random coefficients model. Multivariate Behavioral Research, 35(1), 51-58.

Rovine, M. J., & von Eye, A. (1997). a 14th way to look at a correlation coefficient: Correlation as the proportion of matches. American Statistician, 51, 42-46.

Professional Experience

2004-present: Director, Health and Human Development Methodology Consulting Center

2004-2005: Visiting Professor, Applied Psychology and Human Development, University of Pennsylvania

1997-1998: Visiting Professor, Department of Developmental Psychology, University of Amsterdam

1991-Present: Associate Professor of Human Development, Department of Human Development and Family Studies, College of Health and Human Development, The Pennsylvania State University

1992-2004: Associate Director, Center for Developmental and Health Research Methodology, College of Health and Human Development, The Pennsylvania State University

1993-1994: Acting Director, Center for Developmental and Health Research Methodology, College of Health and Human Development, The Pennsylvania University

1990-1993: Research Associate, Center for Developmental and Health Research Methodology, College of Health and Human Development, The Pennsylvania State University

1984-1991: Assistant Professor of Human Development, Department of Human Development and Family Studies, College of Health and Human Development, The Pennsylvania State University

1982-1983: Post-doctoral Research Associate, Infant and Family Development Project, The Pennsylvania State University

1981: Instructor, Department of Educational Psychology, College of Education, The Pennsylvania State University--Altoona Campus


Ph.D., 1982, Ed. Psychology, The Pennsylvania State University
M.S., 1979, Ed. Psychology, The Pennsylvania State University
B.S., 1971, Mathematics, University of Pennsylvania