|
Latent Variable Scores From SEM Analysis
|
| Presenter(s):
|
| Karen Larwin, University of Akron, drklarwin@yahoo.com
|
| Abstract:
Latent variable scores are traditionally calculated from the averaging of scores from items associated with an individual construct. These values are then used in further analyses, such as multiple regression analyses. Unfortunately, this traditional approach to computing latent variable scores is problematic in that the resulting composite scores do not take into consideration the strength of the relationship between the each item and the associated construct. In Joreskog (2000) a new approach to latent scores development was proposed in which latent variable scores can be generated from the SEM model provide more accurate values with which to examine the effect of exogenous variables, such as participants' gender, age, or major area on factors. The present application of latent variable scores provides a new level of flexibility to accurately address research questions not previously be answered with the SEM model.
|
|
Internal Mental Distress Among Adolescents Entering Substance Abuse Treatment: Examining Measurement Equivalence Across Racial/Ethnic and Gender Groups
|
| Presenter(s):
|
| Mesfin S Mulatu, Centers for Disease Control and Prevention, MMulatu@cdc.gov
|
| Dionne Godette, University of Georgia, dgodette@uga.edu
|
| Kimberly Leonard, The MayaTech Corporation, kjleonard@mayatech.com
|
| Suzanne M Randolph, The MayaTech Corporation, srandolph@mayatech.com
|
| Abstract:
Objective: It is uncertain whether racial/ethnic or gender disparities in co-occurring mental health outcomes reflect true differences in prevalence or simply measurement bias. This study examines cross-racial/ethnic and cross-gender measurement equivalence of the Internal Mental Distress Scale (IMDS) of the Global Appraisal of Individual Needs (GAIN) - a widely used assessment tool in treatment settings.
Sample and Methods: We randomly selected equal number of male and female White (n=600), Black (n=600) and Hispanic (n=600) adolescents from a large pool of admissions to federally-funded substance abuse treatment programs throughout the U.S. Using the Mplus software, we tested increasingly strict models of measurement equivalence of IMDS'5-factor structure (43 dichotomous items) with a series of multi-group confirmatory factor analyses (CFA).
Results: Multi-group CFA showed that the most restrictive model proposing equal factor loadings, thresholds, and residuals fit the data very well in both racial/ethnic (CFI=.988; TLI=.992; RMSEA=.037) and gender (CFI=.987; TLI=.992; RMSEA=.033) group comparisons. Inspection of item-level estimates indicated potential for further improvement of cross-group equivalence.
Conclusions: GAIN's IMDS appears to measure internal mental distress fairly equally across race/ethnic and gender groups. Group differences on IMDS factor scores are less likely due to measurement bias; thus, its use among diverse populations is supported.
|
|
Evaluation of Translated Instruments: Cross Cultural Factorial Invariance of Multiple Group Confirmatory Factor Analysis and Multiple Indicators, Multiple Causes (MIMIC) Models
|
| Presenter(s):
|
| Fatma Ayyad, Western Michigan University, fattmah@hotmail.com
|
| Brooks Applegate, Western Michigan University, brooks.applegate@wmich.edu
|
| Abstract:
If factorial invariance is established across translated forms of research instruments then it is clear that the meaning of the construct crosses cultures. However, if invariance is not established it is not clear if the construct fails to replicate in the translated instrument or the actual translation in faulty.
This study disentangles this dilemma by determining if cultural/language variance can be decomposed from a more general form of construct variance. Specifically, in translated instruments, is there both construct-pure variance and variance related to language and culture? If so can these two sources of variances be separately estimated?
This paper presents a model of selected instruments translated from English into Arabic. Forward and blind-back translation strategies were conducted by bilingual English-Arabic speaking professionals to achieve conceptual equivalence between the original and translated instruments. Empirical Evaluation of the Scales was conducted using data collected from four samples of two different language populations.
|
|
Recapturing Time in Evaluation of Causal Relations: Illustration of Latent Longitudinal and Nonrecursive SEM Models for Simultaneous Data
|
| Presenter(s):
|
| Emil Coman, Institute for Community Research, comanus@netscape.net
|
| Abstract:
We compare two ways of estimating cross-time relations between causes and effects from simultaneously sampled data, the nonrecursive SEM models and the Gollob & Reichardt's latent longitudinal SEM (LLM) model. Both attempt to distinguish direction of causation from cross-sectional measurements. We evaluated them against cross-lagged SEM models of panel data, the Add Health 3 Wave survey. The cross-lagged analyses of stress and alcohol use in men showed a consistent pattern of stress[1] to alcohol[2] then to stress[3]. We tested nonrecursive SEM loop models for both the W2 and W3 stress-alcohol relations. We tested LLM for the W2 and then W3 (pretending no prior time data was available), with prior latent (unobserved) measures of stress and alcohol. LLMs provided indication that the latent alcohol[1] variable causes the measured stress[2], while the latent stress[W1]->alcohol[W2] was not significant, and no paths lead to observed stress[3] or alcohol[3]. LLM seems to fare better here.
|
| | | |