Search

In this study, we reanalyzed the classic Hawai'i Family Study of Cognition (HFSC) data using contemporary multilevel modeling techniques. We used the HFSC baseline data ("N" = 6,579) and reexamined the factorial structure of 16 cognitive variables using confirmatory (restricted) measurement models in an explicit sequence. These models were initially fitted using multilevel confirmatory factor analysis techniques and the invariant six-factor models with two higher order factors fit fairly well (e[subscript a < 0.08) to the total, between- and within-family data. More crucially, a model requiring metric factorial invariance proved to be a reasonable fit to the between and within matrices, and allowed the ratio of the between-family variation to total family variation (eta-squared) to be calculated separately for each common factor (?[superscript 2]: Gc = 0.27, Gf = 0.22, Gm = 0.15, Gs = 0.04, Gv = 0.30, and SP = 0.16). Higher order factors were fitted using multilevel structural equation modeling techniques and these suggested a reasonable two-factor solution with unequal family impacts. These results suggest that (a) A "G only model" does not fit the data very well, and there are many sources of individual differences in cognitive abilities; (b) the sources of the individual differences in cognition can be measured the same way between and within families; and (c) even after the unique test components are removed, "cognitive differences are larger within families than between families". We consider other general multivariate family models, and we raise questions about family influences.

In the behavioral and social sciences, structural equation models (SEMs) have become widely accepted as a modeling tool for the relation between latent and observed variables. SEMs can be seen as a unification of several multivariate analysis techniques. SEM Trees combine the strengths of SEMs and the decision tree paradigm by building tree structures that separate a data set recursively into subsets with significantly different parameter estimates in a SEM. SEM Trees provide means for finding covariates and covariate interactions that predict differences in structural parameters in observed as well as in latent space and facilitate theory-guided exploration of empirical data. We describe the methodology, discuss theoretical and practical implications, and demonstrate applications to a factor model and a linear growth curve model. (Contains 6 figures.)

Background: Educational policy makers and test critics often assert that standardized test scores are strongly influenced by factors beyond individual differences in academic achievement such as family income and wealth. Unfortunately, few empirical studies consider the simultaneous and related influences of family income, parental education, and high school achievement on college admissions test scores. Focus Of Study: This research was animated by the nagging question of the association of family income with SAT performance. For example, is the relationship between family income and SAT performance non-linear? Does the relationship differ markedly by race? More importantly, how strong are the effects of poverty on SAT performance? Research Design: This study is a secondary analysis of a large national sample of Black and White college-bound high school students who took the SAT in 2003 (N = 781,437). Data Collection And Analysis: Employing data from the College Board's Student Descriptive Questionnaire, this study used structural equation modeling (SEM) to estimate the effects of family income on SAT scores for Black and White examinees accounting for the simultaneous effects of parental education and high school achievement. Findings/Results: Results suggest the effects of family income on SAT scores, though relatively modest in contrasts to high school achievement, are substantial, non-linear, and nearly twice as large for Black students. Moreover, the unstandardized direct effect of high school achievement on SAT performance is not enough to address the substantial effects of poverty for Black students. Conclusions/Recommendations: The findings are discussed with respect to social inequality and educational opportunity in college admissions.

This is an application of contemporary multilevel regression modeling to the prediction of academic performances of 1st-year college students. At a first level of analysis, the data come from N greater than 16,000 students who were college freshman in 1994-1995 and who were also participants in high-level college athletics. At a second level of analysis, the student data were related to the different characteristics of the C = 267 colleges in Division I of the NCAA. The analyses presented here initially focus on the prediction of freshman GPA from a variety of high school academic variables. The models used are standard multilevel regression models, but we examine nonlinear prediction within these multilevel models, and additional outcome variables are considered. The multilevel results show that (a) high school grades are the best available predictors of freshman college grades, (b) the ACT and SAT test scores are the next best predictors available, (c) the number of high school core units taken does not add to this prediction but does predict credits attained, (d) college graduation rate has a second-level effect of a small negative outcome on the average grades, and (e) nonlinear models indicate stronger effects for students at higher levels of the academic variables. These results show that standard multilevel models are practically useful for standard validation studies. Some difficulties were found with more advanced uses and interpretations of these techniques, and these problems lead to suggestions for further research. (Contains 6 tables and 6 figures.)

In this paper, we model gender differences in cognitive ability in China using a new sample of middle-aged and older Chinese respondents. Modeled after the American Health and Retirement Study (HRS), the CHARLS Pilot survey respondents are 45 years and older in two quite distinct provinces--Zhejiang, a high-growth industrialized province on the East Coast, and Gansu, a largely agricultural and poor province in the West--in a sense new and old China. Our cognition measures proxy for two different dimensions of adult cognition--episodic memory and intact mental status. On both measures, Chinese women score much lower than do Chinese men, a gender difference that grows among older Chinese cohorts. We relate both these cognition scores to schooling, urban residence, family and community levels of economic resources, and height. We find that cognition is more closely related to mean community resources than to family resources, especially for women, suggesting that in traditional poor Chinese communities there are strong economic incentives to favor boys at the expense of girls. We also find that these gender differences in cognitive ability have been steadily decreasing across birth cohorts as the economy of China grew rapidly. Among cohorts of young adults in China, there is no longer any gender disparity in cognitive ability. This parallels the situation in the United States where cognition scores of adult women actually exceed those of adult men. (Contains 1 figure, 9 tables, and 7 footnotes.)

This study used statistical simulation to calculate differential statistical power in dynamic structural equation models with groups (as in McArdle & Prindle, 2008). Patterns of between-group differences were simulated to provide insight into how model parameters influence power approximations. Chi-square and root mean square error of approximation (RMSEA) power approximation procedures were used to compare the effects of parameter manipulations and how researchers should interpret findings. The chi-square power of perfect fit calls for at least 270 individuals to detect moderate differences, whereas the RMSEA procedure of close fit seems to require as many as 1,450 participants. It is shown that parameters that provide input into the change score that the transfer leads to affect power versus indirect pathways. A discussion of differences in approximation values and future research directions follows. (Contains 3 tables and 4 figures.)

In recent years the use of the latent curve model (LCM) among researchers in social sciences has increased noticeably, probably thanks to contemporary software developments and the availability of specialized literature. Extensions of the LCM, like the the latent change score model (LCSM), have also increased in popularity. At the same time, the R statistical language and environment, which is open source and runs on several operating systems, is becoming a leading software for applied statistics. We show how to estimate both the LCM and LCSM with the sem, lavaan, and OpenMx packages of the R software. We also illustrate how to read in, summarize, and plot data prior to analyses. Examples are provided on data previously illustrated by Ferrer, Hamagami, and McArdle (2004). The data and all scripts used here are available on the first author's Web site. (Contains 5 figures.)

Latent difference score models (e.g., McArdle & Hamagami, 2001) are extended to include effects from prior changes to subsequent changes. This extension of latent difference scores allows for testing hypotheses where recent changes, as opposed to recent levels, are a primary predictor of subsequent changes. These models are applied to bivariate longitudinal data collected as part of the Baltimore Longitudinal Study of Aging on memory performance, measured by the California Verbal Learning Test, and lateral ventricle size, measured by structural MRIs. Results indicate that recent increases in the lateral ventricle size were a leading indicator of subsequent declines in memory performance from age 60 to 90. (Contains 4 tables, 2 footnotes, and 8 figures.)

The authors use multiple-sample longitudinal data from different test batteries to examine propositions about changes in constructs over the life span. The data come from 3 classic studies on intellectual abilities in which, in combination, 441 persons were repeatedly measured as many as 16 times over 70 years. They measured cognitive constructs of vocabulary and memory using 8 age-appropriate intelligence test batteries and explore possible linkage of these scales using item response theory (IRT). They simultaneously estimated the parameters of both IRT and latent curve models based on a joint model likelihood approach (i.e., NLMIXED and WINBUGS). They included group differences in the model to examine potential interindividual differences in levels and change. The resulting longitudinal invariant Rasch test analyses lead to a few new methodological suggestions for dealing with repeated constructs based on changing measurements in developmental studies. (Contains 5 tables and 4 figures.)

Previous analyses have identified a genetic contribution to the correlation between declines with age in processing speed and higher cognitive abilities. The goal of the current analysis was to apply the biometric dual change score model to consider the possibility of temporal dynamics underlying the genetic covariance between aging trajectories for processing speed and cognitive abilities. Longitudinal twin data from the Swedish Adoption/Twin Study of Aging, including up to 5 measurement occasions covering a 16-year period, were available from 806 participants ranging in age from 50 to 88 years at the 1st measurement wave. Factors were generated to tap 4 cognitive domains: verbal ability, spatial ability, memory, and processing speed. Model-fitting indicated that genetic variance for processing speed was a leading indicator of variation in age changes for spatial and memory ability, providing additional support for processing speed theories of cognitive aging. (Contains 6 figures, 7 tables, and 1 footnote.)

Depression and posttraumatic stress disorder (PTSD) are highly comorbid conditions that may arise following exposure to psychological trauma. This study examined their temporal sequencing and mutual influence using bivariate latent difference score structural equation modeling. Longitudinal data from 182 emergency room patients revealed level of depression symptom severity to be positively associated with changes in PTSD intrusion, avoidance, and hyperarousal over 3 time intervals, beginning shortly after the traumatic event. Higher scores on depression anticipated increases (or worsening) in PTSD symptom severity. The pattern of influence from PTSD symptom severity to change in depression symptom severity simply followed the general trend toward health and well-being. Results are discussed in terms of the dynamic interplay and associated mechanisms of posttrauma depression and PTSD symptom severity. (Contains 2 figures, 5 tables, and 4 footnotes.)

Bayesian methods are becoming very popular despite some practical difficulties in implementation. To assist in the practical application of Bayesian methods, we show how to implement Bayesian analysis with WinBUGS as part of a standard set of SAS routines. This implementation procedure is first illustrated by fitting a multiple regression model and then a linear growth curve model. A third example is also provided to demonstrate how to iteratively run WinBUGS inside SAS for Monte Carlo simulation studies. The SAS codes used in this study are easily extended to accommodate many other models with only slight modification. This interface can be of practical benefit in many aspects of Bayesian methods because it allows the SAS users to benefit from the implementation of Bayesian estimation and it also allows the WinBUGS user to benefit from the data processing routines available in SAS. (Contains 3 figures and 3 tables.)

Score limitation at the top of a scale is commonly termed "ceiling effect." Ceiling effects can lead to serious artifactual parameter estimates in most data analysis. This study examines the consequences of ceiling effects in longitudinal data analysis and investigates several methods of dealing with ceiling effects through Monte Carlo simulations and empirical data analyses. Data were simulated based on a latent growth curve model with T = 5 occasions. The proportion of the ceiling data [10%-40%] was manipulated by using different thresholds, and estimated parameters were examined for R = 500 replications. The results showed that ceiling effects led to incorrect model selection and biased parameter estimation (shape of the curve and magnitude of the changes) when regular growth curve models were applied. The Tobit growth curve model, instead, performed very well in dealing with ceiling effects in longitudinal data analysis. The Tobit growth curve model was then applied in an empirical cognitive aging study and the results were discussed. (Contains 2 figures and 7 tables.)

The main purpose of this research is to evaluate the performance of a Bayesian approach for estimating unknown change points using Monte Carlo simulations. The univariate and bivariate unknown change point mixed models were presented and the basic idea of the Bayesian approach for estimating the models was discussed. The performance of Bayesian estimation was evaluated using simulation studies of longitudinal data with different sample sizes, varying change point values, different levels of Level-1 variances, and univariate versus bivariate outcomes. The numerical results compared the performance of the Bayesian methods with the first-order Taylor expansion method and the adaptive Gaussian quadrature method implemented in SAS PROC NLMIXED. These simulation results showed that the first-order Taylor expansion method and the adaptive Gaussian quadrature method were sensitive to the initial values, making the results somewhat unreliable. In contrast, these simulation results showed that Bayesian estimation was not sensitive to the initial values and the fixed-effects and Level-1 variance parameters can be accurately estimated in all of the conditions. One concern was that the estimates of the Level-2 covariance parameters were found to be biased when the Level-1 variance was large in the bivariate model. However, and in general, the new Bayesian approach to the estimation of turning points in longitudinal data proved to be quite robust and practically useful. (Contains 7 tables, 2 figures, and 2 notes.)

The authors applied linear dynamic models to longitudinal data to examine the dynamics of reading and cognition from 1st to 12th grade. They used longitudinal data (N=445) from the Connecticut Longitudinal Study (S. E. Shaywitz, B. A. Shaywitz, J. M. Fletcher, & M. D. Escobar, 1990) to map the dynamic interrelations of various scales of the Wechsler Intelligence Scales for Children--Revised (i.e., Full, Performance, and Verbal) and specific markers of the Woodcock-Johnson Psycho-Educational Battery--Revised reading cluster (i.e., Letter-Word ID, Decoding, and Comprehension). The results of these analyses indicate that (a) there is a positive dynamic relation between reading and cognition across the selected age range; (b) this dynamic relation is symbiotic, with positive influences in both directions; (c) the influence from cognition to reading is stronger when considering the Wechsler Intelligence Scales for Children Performance--Revised Performance scale and weaker with the Verbal scale; (d) when examining the different Reading subtests, the influences from cognition are more apparent for Letter-Word ID and Comprehension and are less perceptible for Decoding; and (e) the dynamics of reading and cognition appear to be of stronger magnitude during 1st to 3rd grade, less strong during 4th to 8th grade, and weaker from 9th to 12th grade.

Emerging adulthood is a period in which profound role changes take place across a number of life domains including finance, romance, and residence. On the basis of dynamic systems theory, change in one domain should be related to change in another domain, because the concept of development according to this approach is a relational one. To evaluate this hypothesis dynamic systems analysis was applied to data from narrative interviews of 200 respondents covering the years between 17 and 27 to examine how change in one domain affects change in another domain. In each dyad, the fit of the model significantly deteriorated when the coupling between domains was removed providing support for the assumption of interdependency. On average, assuming greater responsibility in one domain was associated with assuming greater responsibility in the other domain. However, imbalances were also observed in which role assumption in one domain far exceeded role assumption in another domain. These imbalances can have detrimental effects and indicate the utility of a balanced approach to development. The findings underscore the importance of studying the relational unit between domains, which is critical to understanding development over time within domains.

Objectives: The prevalence rates of disorders among a community-based sample of Hawaiian youths were determined and compared to previously published epidemiological studies. Method: Using a two-phase design, 7,317 adolescents were surveyed (60% participation rate), from which 619 were selected in a modified random sample during the 1992-1993 to 1995-1996 school years: 590 selected randomly and 29 at risk (i.e., Center for Epidemiologic Studies-Depression score of [greater than or equal to] 35 and suicidal risk) from grades 9-12. The Diagnostic Interview Schedule for Children-Version 2.3, was used to determine DSM-III-R diagnoses. Prevalence rates, weighted for ethnicity, Center for Epidemiologic Studies-Depression scores, and suicide attempts, were calculated for any diagnosis and various disorders. Meta-analyses compared the Hawai'i sample to four community-based studies (randomly selected youths from community populations) and two high-risk studies (homeless, low-income, or high unemployment communities). Results: Hawaiian females had the highest rate for any diagnosis (37.7%; 95% confidence interval [CI] 28.4%-48.0%) and non-Hawaiian males had the lowest rate (19.6%; 95% CI 14.8%-25.5%). Hawaiian males (26.8%; 95% CI 18.2%-37.5%) and non-Hawaiian females (27.9%; 95% CI 22.2%-34.4%) had intermediate and comparable rates. Overall, Hawaiians had significantly higher rates (32.7%; 95% CI 26.1%-40.1%) than non-Hawaiians (23.7%; 95% CI 19.9%-28.0%) when controlling for gender, and girls had significantly higher rates (30.8%; 95% CI 25.8%-36.3%) than boys (21.1%; 95% CI 16.8%-26.1%) when controlling for ethnicity. These findings were primarily the result of the significant differences in rates regarding anxiety disorders. Meta-analyses showed the Hawaiian youth rate for any diagnosis was comparable to high-risk studies and nearly three times higher than the community studies. Conclusions: Hawaiian youths, especially females, are at high risk. Research on the sociocultural factors that underpin both the genesis of and protection from psychopathology is imperative for Hawaiian and non-Hawaiian mixed-ethnicity youths. (Contains 4 tables.)

Though many cognitive abilities exhibit marked decline over the adult years, individual differences in rates of change have been observed. In the current study, biometrical latent growth models were used to examine sources of variability for ability level (intercept) and change (linear and quadratic effects) for verbal, fluid, memory, and perceptual speed abilities in the Swedish Adoption/Twin Study of Aging. Genetic influences were more important for ability level at age 65 and quadratic change than for linear slope at age 65. Expected variance components indicated decreasing genetic and increasing nonshared environmental variation over age. Exceptions included one verbal and two memory measures that showed increasing genetic and nonshared environmental variance. The present findings provide support for theories of the increasing influence of the environment with age on cognitive abilities.

This study examined the dynamics of cognitive abilities and academic achievement from childhood to early adulthood. Predictions about time-dependent "coupling" relations between cognition and achievement based on R. B. Cattell's (1971, 1987) investment hypothesis were evaluated using linear dynamic models applied to longitudinal data (N=672). Contrary to Cattell's hypothesis, a first set of findings indicated that fluid and crystallized abilities, as defined by the Woodcock-Johnson Psycho-Educational Battery-Revised (WJ-R; R. W. Woodcock & M. B. Johnson, 1989-1990), were not dynamically coupled with each other over time. A second set of findings provided support for the original predictions and indicated that fluid ability was a leading indicator of changes in achievement measures (i.e., quantitative ability and general academic knowledge). The findings of this study suggest that the dynamics of cognitive abilities and academic achievement follow a more complex pattern than that specified by Cattell's investment hypothesis.

This article offers different examples of how to fit latent growth curve (LGC) models to longitudinal data using a variety of different software programs (i.e., LISREL, Mx, Mplus, AMOS, SAS). The article shows how the same model can be fitted using both structural equation modeling and multilevel software, with nearly identical results, even in the case of models of latent growth fitted to incomplete data. The general purpose of this article is to provide a demonstration that integrates programming features from different software. The most immediate goal is to help researchers implement these LGC models as a useful way to test hypotheses of growth.

Using Bowen and Bok's data from 23 selective colleges, we fit multilevel logit models to test two hypotheses with implications for affirmative action and group differences in attainment of science, math, or engineering (SME) degrees. Hypothesis 1, that differences in precollege academic preparation will explain later SME graduation disparities, was fully supported with respect to the outcome gap between Whites and underrepresented minorities, partially supported for that between Asians and underrepresented minorities, and between men and women. Hypothesis 2, that college selectivity, after accounting for student characteristics, will be positively associated with SME persistence, was not supported. We demonstrate that the significance of the selectivity effect is overestimated when unilevel models are used. Admission officials are advised to carefully consider the relative academic preparedness of science-interested students, and such students choosing among colleges are advised to compare their academic qualifications to those of successful science students at each institution.

Describes some applications of latent growth curve models in the context of structural equation modeling using data from P. Trickett and F. Putnam (1993) on the physical height of abused (n=77) and nonabused (n=75) adolescent girls. Presents power calculations for the ability of the different models to discern the growth of the abuse sample from the control sample. (SLD)

Analyzes differences in self-reported Center for Epidemiologic Studies Depression inventory results among ethnic Hawaiian and non-Hawaiian high school students, using different forms of latent variable structural equation models. Finds a high degree of invariance between students on depression. Discusses issues about common features and relationships among standard analysis of variance approaches and structural equation methodology. (Author/JPB)

This study examined dyadic interrelations between episodic memory and depressive symptom trajectories of change in old and advanced old age. The authors applied dynamic models to 10-year incomplete longitudinal data of initially 1,599 married couples from the study of Asset and Health Dynamics Among the Oldest Old ([M.sub.age] = 75 years at Time 1). The authors found domain-specific lead-lag associations (time lags of 2 years) among wives and husbands as well as between spouses. For memory, better performance among husbands protected against subsequent memory decline among wives, with no evidence of a directed effect in the other direction. For depressive symptoms, wives' scores predicted subsequent depression increase and memory decline among husbands. Possible individual covariates (age, education, functional limitations) and spousal covariates (length of marriage, number of children, and whether the couple remained intact over the study period) did not account for differential lead-lag associations. The findings of antecedent-consequent relations between wives and husbands are consistent with life-span notions that individual development both influences and is influenced by contextual factors such as close social relationships. Keywords: dyad, spouses, successful aging, dynamic modeling, growth curve modeling DOI: 10.1037/a0016346

Previous analyses have identified a genetic contribution to the correlation between declines with age in processing speed and higher cognitive abilities. The goal of the current analysis was to apply the biometric dual change score model to consider the possibility of temporal dynamics underlying the genetic covariance between aging trajectories for processing speed and cognitive abilities. Longitudinal twin data from the Swedish Adoption/Twin Study of Aging, including up to 5 measurement occasions covering a 16-year period, were available from 806 participants ranging in age from 50 to 88 years at the 1st measurement wave. Factors were generated to tap 4 cognitive domains: verbal ability, spatial ability, memory, and processing speed. Model-fitting indicated that genetic variance for processing speed was a leading indicator of variation in age changes for spatial and memory ability, providing additional support for processing speed theories of cognitive aging. Keywords: cognitive aging, longitudinal twin study, behavioral genetics, temporal dynamics, speed-cognition relationship