Multilevel Time Series Models Research Articles

Mother-infant self- and interactive contingency at four months and infant cognition at one year: A view from microanalysis

Although a considerable literature documents associations between early mother-infant interaction and cognitive outcomes in the first years of life, few studies examine the contributions of contingently coordinated mother-infant interaction to infant cognitive development. This study examined associations between the temporal dynamics of the contingent coordination of mother-infant face-to-face interaction at 4 months and cognitive performance on the Bayley Scales of Infant Development at age one year in a sample of (N = 100) Latina mother-infant pairs. Split-screen videotaped interactions were coded on a one second time base for the communication modalities of infant and mother gaze and facial affect, infant vocal affect, and mother touch. Multi-level time-series models evaluated self- and interactive contingent processes in these modalities and revealed 4-month patterns of interaction associated with higher one-year cognitive performance, not identified in prior studies. Infant and mother self-contingency, the moment-to-moment probability that the individual’s prior behavior predicts the individual’s future behavior, was the most robust measure associated with infant cognitive performance. Self-contingency findings showed that more varying infant behavior was optimal for higher infant cognitive performance, namely, greater modulation of negative affect; more stable maternal behavior was optimal for higher infant cognitive performance, namely, greater likelihood of sustaining positive facial affect. Although interactive contingency findings were sparse, they showed that, when mothers looked away, or dampened their faces to interest or mild negative facial affect, infants with higher 12-month cognitive performance were less likely to show negative vocal affect. We suggest that infant ability to modulate negative affect, and maternal ability to sustain positive affect, may be mutually reinforcing, together creating a dyadic climate that is associated with more optimal infant cognitive development.

Affective dependence, self-regulation, and alcohol problems: Between- and within-person associations.

Positive affect (PA) and negative affect (NA) are associated with goal pursuit in addition to dysregulated behavior. Affective dependence (i.e., the correlation between PA and NA) may be a marker for good self-regulation on the one hand (weaker dependence) and poor self-regulation on the other (stronger dependence). This study sought to elucidate the role of affective dependence as a predictor of goal pursuit and alcohol problems at the within- and between-person levels. Participants were 100 college students aged 18-25 years, who drank alcohol at least moderately, and completed a 21-day ecological momentary assessment study regarding affect, academic goal pursuit, idiographic goal pursuit, alcohol use, and alcohol problems. Multilevel time series models were estimated. Consistent with hypotheses, affective dependence predicted more alcohol problems and decreased academic goal pursuit at the within-person level. Importantly, effects on academic goal pursuit included perceived achievement and progress related to academics, as well as time spent studying, an objective marker for academic engagement. Effects were significant controlling for autoregressive effects, lagged residuals of PA and NA, concurrent alcohol use, day of the week, age, gender, and trait affective dependence. Thus, this study provides robust tests of lagged within-person effects of affective dependence. The effect of affective dependence on idiographic goal pursuit was not significant, contrary to hypothesis. Affective dependence was not significantly associated with alcohol problems or goal pursuit at the between-person level. Results suggest that affective dependence is a common factor explaining problems related to alcohol use and psychological functioning more broadly. (PsycInfo Database Record (c) 2023 APA, all rights reserved).

Temporal associations linking alcohol and cannabis use to cigarette smoking in young adults engaged in a tobacco cessation and relapse monitoring study

Young adulthood remains a developmental period in which cigarette smoking initiation and progression to dependence and regular use is common. Moreover, co-use of alcohol and/or cannabis with tobacco is common in this age group and may have detrimental effects on tobacco use rates and cessation outcomes. Although young adults are interested in quitting smoking, achieving abstinence remains difficult, even with evidence-based treatment strategies. Understanding proximal associations between other substance use (e.g., alcohol and cannabis) and smoking may have important treatment implications. This exploratory analysis investigated the role of alcohol and/or cannabis use in contributing to smoking events on the same day or next day among young adults engaged in a smoking cessation and relapse monitoring study. We used ecological momentary assessment (EMA) data from 43 young adults (ages 18–25; 932 observations) who smoked cigarettes daily and agreed to participate in a 5-week study that included a 2-day smoking quit attempt and provision of tobacco treatment in the form of nicotine replacement therapy, brief cessation counseling, and financial incentives for abstinence (incentives were provided only during the 2-day quit attempt). We tested multilevel time-series models of daily associations between alcohol use, cannabis use, and smoking. Consistent with hypotheses, days on which participants were more likely to drink alcohol predicted increased likelihood of smoking the next day (OR = 2.27, p =.003). This effect was significant after controlling for both the one-day lagged effect of smoking (i.e., autoregression) and the concurrent (i.e., same day) effects of drinking and cannabis use. Although there was a positive concurrent effect of cannabis use on smoking (OR = 12.86, p =.003), the one-day lagged effect of cannabis use and the concurrent effect of drinking was not significant, contrary to hypotheses. Results indicate that alcohol use presents a potential threat to successful smoking cessation that extends to the following day. This suggests a risk-window in which treatment could be supplemented with just-in-time interventions and extending the focus on co-use to include this lagged impact on cessation outcomes.

Associations of maternal postpartum depressive and anxiety symptoms with 4-month infant and mother self- and interactive contingency of gaze, affect, and touch.

Maternal depression and anxiety are associated with infant and mother self- and interactive difficulties. Although maternal depression and anxiety usually co-occur, studies taking this comorbidity into account are few. Despite some literature, we lack a detailed understanding of how maternal depressive and anxiety symptoms may be associated with patterns of mother-infant interaction. We examined associations of maternal postpartum depressive and anxiety symptoms with infant and mother self- and interactive patterns by conducting multi-level time-series models in a sample of 56 Turkish mothers and their 4-month infants. Time-series models assessed the temporal dynamics of interaction via infant and mother self- and interactive contingency. Videotaped face-to-face interaction was coded on a 1s time base for infant and mother gaze and facial affect, infant vocal affect, and mother touch. Results indicated that mothers with high depressive symptoms were vulnerable to infants looking away, reacting with negative touch; their infants remained affectively midrange, metaphorically distancing themselves from mothers' affect. Mothers with high anxiety symptoms were vulnerable to infants becoming facially dampened and mothers reacted with negative facial affect. Altered infant and mother self-contingency patterns were largely opposite for maternal depressive and anxiety symptoms. These patterns describe foundational processes by which maternal postpartum mood is transmitted to the infant and which may affect infant development.

Multilevel time-series models for small area estimation at different frequencies and domain levels

A small area estimation method is developed for repeatedly conducted multipurpose surveys. A multilevel time-series model is proposed that uses direct estimates for the most detailed domains observed at the highest frequency of the repeated survey. A consistent set of estimates at different aggregation levels is then derived by aggregation of the model-based predictions obtained for the most detailed domains observed at the highest frequency. The model borrows strength over time and space via smooth and local level trends at different aggregation levels. The model also borrows information from auxiliary series available from registers with coefficients that can vary over both domains and time. Regional domain random effects are allowed to vary smoothly over space according to a spatial autoregressive process. To account for the diversity of domains and for more volatile time-dependence, nonnormally distributed random effects and trend innovations are used via so-called global-local shrinkage priors. A Bayesian approach is taken, and the model is estimated by MCMC simulation. The method is illustrated with an application to the Dutch Labour Force Survey to produce monthly provincial and quarterly municipal unemployment figures.

Preterm infant contingent communication in the neonatal intensive care unit with mothers versus fathers.

Very few studies have assessed infant capacity for bidirectional, contingent communication at birth, and to our knowledge there are none with preterm infants in the neonatal period. Presence versus absence of such interactive contingency makes a difference for our theories of development. We examined whether preterm infants can contingently coordinate behaviors with mothers and fathers in spontaneous communication in the Neonatal Intensive Care Unit (NICU), and whether mother-infant versus father-infant engagement and contingency differ. Twenty Italian preterm infants (60% girls, born 27-33 weeks, largely middle-class families) lying in a heated cot in the NICU were videotaped at 35 weeks with mothers, and fathers (counterbalanced), in face-to-face communication. Videotapes were coded on a 1-s time-base with parent and infant engagement scales. Multilevel time-series models evaluated self-contingency (auto-correlation) and interactive contingency (lagged cross-correlation). Mothers (vs. fathers) showed higher levels of engagement, interpreted as more arousing. Fathers (vs. mothers) showed more midrange engagement, interpreted as less "demanding" of infant engagement. Infants were more gaze-on-parent's-face and gaze-on-environment with mothers than fathers. Fathers interacted contingently with infants, whereas mothers did not. However, infants interacted contingently with mothers, but not fathers. When infants were in lower engagement levels 1 s prior, fathers stayed in lower engagement levels in the current second, closer to infants than mothers. We suggest that fathers were more coordinated because fathers were more able to join the infant's dampened state. We suggest that infants were more coordinated with mothers because mothers were more socially stimulating, and more familiar. We conclude that preterm infants, shortly after birth, are capable of contingent communication. (PsycInfo Database Record (c) 2022 APA, all rights reserved).

Multilevel Time Series Modelling of Mobility Trends in the Netherlands for Small Domains

Abstract The purpose of the Dutch Travel Survey is to produce reliable estimates on mobility of the Dutch population. In this paper mobility trends are estimated at several aggregation levels, using multilevel time series models. The models account for discontinuities induced by two survey redesigns and outliers due to less reliable outcomes in one particular year. The input for the model is direct annual estimates with their standard errors for the period 1999–2017 for a detailed cross-classification in 504 domains. Appropriate transformations for the direct estimates and generalized variance functions to smooth the standard errors of the direct estimates are proposed. The models are fitted in an hierarchical Bayesian framework using MCMC simulations. From the model outputs smooth trend estimates are computed at the most detailed domain level. Predictions at higher aggregation levels obtained by aggregation of the most detailed domain predictions result in a numerically consistent set of trend estimates for all target variables.

Expanding the Bayesian structural equation, multilevel and mixture models to logit, negative-binomial, and nominal variables

ABSTRACT Recent work on the Polya-Gamma distribution provides a breakthrough for the Bayesian modeling of logit, count, and nominal variables. We describe how the methodology is incorporated in the Mplus modeling framework and illustrate it with several examples: logistic latent growth models, multilevel IRT, multilevel time-series models for count data, multilevel nominal regression, and nominal factor analysis.

Multilevel Hierarchical Bayesian Versus State Space Approach in time Series Small Area Estimation: the Dutch Travel Survey

Summary This study compares state space models (estimated with the Kalman filter with a frequentist approach to hyperparameter estimation) with multilevel time series models (based on the hierarchical Bayesian framework). The application chosen is the Dutch Travel Survey featuring small sample sizes and discontinuities caused by the survey redesigns. Both modelling approaches deliver similar point and variance estimates. Slight differences in model-based variance estimates appear mostly in small-scaled domains and are due to neglecting uncertainty around the hyperparameter estimates in the state space models, and to a lesser extent to skewness in the posterior distributions of the parameters of interest. The results suggest that the reduction in design-based standard errors with the hierarchical Bayesian approach is over 50% at the provincial level, and over 30% at the national level.

Experiencing Racial Humor with Outgroups: A Psychophysiological Examination of Co-Viewing Effects

Research on co-viewing (or group viewing) is scarce. Yet, co-viewing has important implications for the viewers’ entertainment experience and the way viewers respond to and evaluate entertainment—especially those with controversial messages. The present study investigated responses to racial humor content among racial in-group and out-group viewing contexts. Specifically, the study examined the extent to which Blacks and Whites would experience discomfort when viewing racial slurs in comedies with in-group compared to out-group members. Employing real-time psychophysiological data and multilevel time series models, the study found a significant increase in emotional arousal (indicated by SCRs) and distraction (indicated by RSA) among Blacks in the context of out-group viewing compared to in-group viewing, but not for Whites. Implications of findings are discussed.

Multilevel Models for the Analysis of Angle-Specific Torque Curves with Application to Master Athletes.

The aim of this paper was to outline a multilevel modeling approach to fit individual angle-specific torque curves describing concentric knee extension and flexion isokinetic muscular actions in Master athletes. The potential of the analytical approach to examine between individual differences across the angle-specific torque curves was illustrated including between-individuals variation due to gender differences at a higher level. Torques in concentric muscular actions of knee extension and knee extension at 60º·s−1 were considered within a range of motion between 5º and 85º (only torques “truly” isokinetic). Multilevel time series models with autoregressive covariance structures with standard multilevel models were superior fits compared with standard multilevel models for repeated measures to fit angle-specific torque curves. Third and fourth order polynomial models were the best fits to describe angle-specific torque curves of isokinetic knee flexion and extension concentric actions, respectively. The fixed exponents allow interpretations for initial acceleration, the angle at peak torque and the decrement of torque after peak torque. Also, the multilevel models were flexible to illustrate the influence of gender differences on the shape of torque throughout the range of motion and in the shape of the curves. The presented multilevel regression models may afford a general framework to examine angle-specific moment curves by isokinetic dynamometry, and add to the understanding mechanisms of strength development, particularly the force-length relationship, both related to performance and injury prevention.

When the media matter for politics: Partisan moderators of the mass media’s agenda-setting influence on parliament in Belgium

In this study, we investigate which factors moderate the agenda-setting influence of the mass media on the Belgian parliament during the period 1993—2000. Based on elaborate codings of the media, parliamentary questions and interpellations, party manifestos, government agreements and ministerial meetings, we employ a multi-level time-series model. The results indicate that especially party characteristics (party size, incumbent or opposition party, issue ownership) and the government agenda influence the dependency of parliament on media coverage. Furthermore, we find an increase in the extent of media influence through time, suggesting an increasing presence of ‘media logic’ in the behaviour of Belgian MPs. Irrespective of all those contingent factors, the mass media determine the Belgian parliamentary agenda to a considerable degree.

Six-week postpartum maternal self-criticism and dependency and 4-month mother-infant self- and interactive contingencies.

Associations of 6-week postpartum maternal self-criticism and dependency with 4-month mother-infant self- and interactive contingencies during face-to-face play were investigated in 126 dyads. Infant and mother face, gaze, touch, and vocal quality were coded second by second from split-screen videotape. Self- and interactive contingencies were defined as auto- and lagged cross-correlation, respectively, using multilevel time-series models. Statistical significance was defined as p<.05. Regarding self-contingency, (a) more self-critical mothers showed primarily lowered self-contingency, whereas their infants showed both lowered and heightened, and (b) infants of more dependent mothers showed primarily lowered self-contingency, whereas findings were absent in mothers. Regarding interactive contingency, (a) more self-critical mothers showed lowered attention and emotion contingencies but heightened contingent touch coordination with infant touch, and (b) more dependent mothers and their infants showed heightened facial/vocal interactive contingencies. Thus, maternal self-criticism and dependency have different effects on mother-infant communication.

Pulmonary tumour embolism

24 Jacobs B, Young NL, Dick PY, et al. Canadian Acute Respiratory Illness and Flu Scale (CARIFS): development of a valid measure for childhood respiratory infections. J Clin Epidemiol 2000; 53: 793–99. 25 Yang M, Rasbash J, Goldstein H, Barbosa M. MLwiN macros for advanced multilevel modelling, version 2.0. London: Multilevels Models Project, Institute of Education, University of London, 1999. 26 Goldstein H, Healy M, Rasbash J. Multilevel time series models with applications to repeated measures data. Stat Med 1994; 13: 1643–55. 27 Tasche MJA, Uijen JHJM, Bernsen RMD, Jongste JC, van der Wouden JC. Inhaled disodium cromoglycate (DSCG) as maintenance therapy in children with asthma: a systematic review. Thorax 2000; 55: 913–20. 28 Majeed A, Moser K. Ageand sex-specific antibiotic prescribing in general practice in England and Wales in 1996. Br J Gen Pract 1999; 49: 735–36. 29 Stott NCH. Management and outcome of winter upper respiratory tract infections in children aged 0-9 years. BMJ 1979; 1: 29–31. ARTICLES

A multilevel time-series model for the incidence of AIDS cases in Spain

Objective. The main aim of this research is to study the quantitative evolution of the incidence of AIDS in the 19 Spanish Communities. The hypothesis is that incidence follows a multilevel autoregressive model, where each Community shows random variability around a general process. Method. On the basis of official data on the number of existing AIDS cases, an autorregressive multilevel time-series model was developed. Results and Conclusions. Analysis shows that the hypothesis is supported, indicating that overall AIDS incidence in Spain has already reached a maximum and has a tendency to remain stable or to decline in future. Long term expected values have become stable in most Communities; a slight increase is expected only in Extremadura. However, this Community has a relatively sparse population, and its contribution on the overall Spanish incidence is small. Long term expected values are estimated to be around 152.99 new cases per million inhabitants per year. This value is slightly smaller than the maximum incidence, observed in 1994 (179.4 cases).

Discrete Response Multilevel Models for Repeated Measures: An Application to Voting Intentions Data

Repeated measures data can be modelled as a two-levelmodel where occasions (level one units) are grouped byindividuals (level two units). Goldstein et al. (1994)proposed a multilevel time series model when theresponse variable follows a Normal distribution andthe measurements are taken with unequal timeintervals. This paper extends the methodology todiscrete response variables. The models are applied toBritish Election Study data consisting of repeatedmeasures of voting intention.

Discussion of the Paper by Lee and Nelder

Discussion of the Paper by Lee and Nelder

Multilevel time series models with applications to repeated measures data.

The analysis of repeated measures data can be conducted efficiently using a two-level random coefficients model. A standard assumption is that the within-individual (level 1) residuals are uncorrelated. In some cases, especially where measurements are made close together in time, this may not be reasonable and this additional correlation structure should also be modelled. A time series model for such data is proposed which consists of a standard multilevel model for repeated measures data augmented by an autocorrelation model for the level 1 residuals. First- and second-order autoregressive models are considered in detail, together with a seasonal component. Both discrete and continuous time are considered and it is shown how the autocorrelation parameters can themselves be structured in terms of further explanatory variables. The models are fitted to a data set consisting of repeated height measurements on children.

Better Ways to Compare Schools?

The term school effectiveness is found throughout the educational literature, and there is even an annual international conference devoted solely to this topic. The laudable aim of discovering which factors are associated with the success or failure of schools clearly is popular with policy makers and administrators as well as researchers. Much of the activity in this area is concerned with achievement testing, and some of the most interesting work is based on the use of multilevel models. These models are interesting because they attempt to remain faithful to the hierarchical structure of educational data where, for example, pupils are nested within schools and schools within boards or authorities. Applications of these models have been successful in studying the way in which group differences (e.g., between ethnic minorities) vary from school to school; the way it is possible to measure value added, or progress, during schooling by using suitable measures of intake achievement; and the way achievement changes over time using repeated measurements on students. Other work has looked at how schools change over time using repeated measurements of cohorts of students. The application of these models to multivariate data has shown how multiple matrix designs can be modeled efficiently and how efficient estimates can be obtained for generalizability models. A detailed account of these models with applications can be found in Raudenbush and Bryk (1986) and Goldstein (1987). Some of the current development work is concerned with the formulation of multilevel loglinear models, multilevel time series models, multilevel structural equation models, and random cross classifications, and publications in these areas are beginning to appear. All of this is exciting because it provides us with a general tool for the valid analysis of educational data and promises to yield new insights. Indeed, it promises this for all observational data in the human sciences and especially in areas such as complex sample surveys, where direct modeling of the data structure is now possible. In education, the intelligent application of these models promises a deeper understanding of the processes of schooling and the determinants of achievement. With suitably large samples of schools, it should be possible to study how both student and school characteristics play