Factor Mixture Modeling Research Articles

Factor Mixture Modeling of the Insomnia Severity Index among Psychology Clinic Outpatients

Insomnia symptoms are common among individuals with psychiatric disorders, and associated with increased symptom severity. However, the Insomnia Severity Index (ISI) has rarely been psychometrically evaluated in a psychiatric sample. Furthermore, the latent structure of the ISI has not been evaluated using factor mixture modeling, which improves upon prior analytic techniques by integrating factor structure and including exogenous variables. Thus, the current study used factor mixture modeling to test the latent structure of the ISI among two samples of psychology clinic outpatients (Ns = 366 and 331). The ISI was best represented with a three-class structure representing “no or minimal,” “moderate,” and “severe” insomnia symptoms, with significantly escalating symptoms of depression, anxiety, and posttraumatic stress across these classes. Clinical cut-scores for the “severe” class were also derived using receiver operating curve analyses. Implications for theoretical conceptualization of insomnia among those with psychiatric disorders and clinical implications are discussed.

Offender Subtypes Based on Psychopathic Traits

Abstract. Psychopathy is a primary risk factor of re-offending in sexual offenders. Conceptually, both variable-centered (e.g., factor analysis) and clustering methods (e.g., latent profile analysis) have been used in previous research. Variable-centered and clustering methods were merged in a simultaneous modeling strategy for two purposes: First, to test assumptions on the emergence of psychopathic versus sociopathic (antisocial) sub-groups. And second to compare the predictive validity of clusters with that afforded by a dimensional cut-score. Using mixture modeling, two types of models were estimated: Latent class factor-analytic (LCFA) and factor-mixture models (FMM). The four-factor model of psychopathy as assessed with the Psychopathy Checklist-Revised (PCL-R) was estimated for up to 12 latent classes in a sample of adult male sexual offenders from Austria ( N = 1,266). Solutions with five (LCFA) and two latent classes (FMM) provided a good and parsimonious fit for the data. The two-latent-class FMM solution yielded higher predictive validity than a cut-score but only for general offense recidivism. Theoretically, this solution goes against etiological models that distinguish psychopathic from sociopathic (antisocial) individuals. Official data on offense recidivism (at a fixed 7-year-interval post-release) corroborate the importance of psychopathic offender subtypes. The rates of recidivism varied considerably between the subgroups.

Distinct Risk Profiles in Social Anxiety Disorder.

Using a factor mixture model (FMM) approach, this study examined if SAD could be subtyped by distinct risk profiles, and whether these subtypes predicted different manifestations of the disorder. We derived risk profiles from neurotic temperament (NT), positive temperament (PT), and autonomic arousability (AA), which are hypothesized to be important in the maintenance of anxiety disorders such as SAD. In our sample of 758 SAD outpatients, a two-class FMM solution fit the data best. Class 1 was characterized by very low PT whereas PT in Class 2 was substantially higher. The two classes differed to a lesser extent on NT, but were virtually equivalent on AA. Class 1 had significantly more males, individuals with depressive disorders, generalized SAD, and higher SAD severity. Class 2 had more individuals with performance subtype SAD. These findings provide initial support for distinct risk profiles within SAD that may be predictive of its clinical expression.

Identifying unstable and empty phenotypes of borderline personality through factor mixture modeling in a large nonclinical sample.

Borderline personality disorder (BPD) is serious, prevalent, and symptomatically heterogeneous. Identifying distinct phenotypes of BPD features promises useful diagnostic and treatment implications. Although a series of subtyping studies exist, only two have examined BPD symptom configurations while taking into account BPD severity. We used factor mixture modeling to identify discrete subtypes of BPD features, simultaneously considering symptom severity, in the largest nonclinical young adult sample to date. Undergraduates (N = 20,010; 63.86% women; Mage = 18.75, SD = 1.73) completed the McLean Screening Instrument for BPD, which was condensed to measure the 9 Diagnostic and Statistical Manual of Mental Disorders BPD criteria dichotomously. We used a model comparison approach to determine the optimal latent factor and class structure of BPD symptoms and validated classes via BPD-relevant constructs. The sample consisted of three subtypes: Asymptomatic (70%), Unstable (19%), and Empty (11%). The Unstable and Empty classes displayed elevated BPD symptomatology along a single continuum of BPD severity. Individuals in the Empty class displayed the highest levels of emptiness and dissociation, emotional distress, and attachment avoidance, whereas individuals in the Unstable class displayed a high frequency of reckless and self-damaging behaviors. Our results suggest the importance of a hybrid dimensional/categorical conceptualization of BPD as displayed in a nonclinical sample. Unstable and Empty classes may be associated with different treatment targets for subthreshold BPD presentations. The findings are discussed in terms of their clinical implications regarding diagnosis, treatment, and theoretical conceptualization of BPD. (PsycINFO Database Record (c) 2020 APA, all rights reserved).

Supplemental Material for Factor Mixture Modeling of Anxiety Sensitivity: Support for the Three-Class Solution in a Serbian Sample

Supplemental Material for Factor Mixture Modeling of Anxiety Sensitivity: Support for the Three-Class Solution in a Serbian Sample

Caring for the elderly: A person-centered segmentation approach for exploring the association between health care needs, mental health care use, and costs in Germany.

BackgroundPerson-centered care demands the evaluation of needs and preferences of the patients. In this study, we conducted a segmentation analysis of a large sample of older people based on their bio-psycho-social-needs and functioning. The aim of this study was to clarify differences in health care use and costs of the elderly in Germany.MethodsData was derived from the 8-year follow-up of the ESTHER study—a German epidemiological study of the elderly population. Trained medical doctors visited n = 3124 participants aged 57 to 84 years in their home. Bio-psycho-social health care needs were assessed using the INTERMED for the Elderly (IM-E) interview. Further information was measured using questionnaires or assessment scales (Barthel index, Patients Health Questionnaire (PHQ) etc.). The segmentation analysis applied a factor mixture model (FMM) that combined both a confirmatory factor analysis and a latent class analysis.ResultsIn total, n = 3017 persons were included in the study. Results of the latent class analysis indicated that a five-cluster-model best fit the data. The largest cluster (48%) can be described as healthy, one cluster (13.9%) shows minor physical complaints and higher social support, while the third cluster (24.3%) includes persons with only a few physical and psychological difficulties (“minor physical and psychological complaints”). One of the profiles (10.5%) showed high and complex bio-psycho-social health care needs (“complex needs”) while another profile (2.5%) can be labelled as “frail”. Mean values of all psychosomatic variables—including the variable health care costs—gradually increased over the five clusters. Use of mental health care was comparatively low in the more burdened clusters. In the profiles “minor physical and psychological complaints” and “complex needs”, only half of the persons suffering from a mental disorder were treated by a mental health professional; in the frail cluster, only a third of those with a depression or anxiety disorder received mental health care.ConclusionsThe segmentation of the older people of this study sample led to five different clusters that vary profoundly regarding their bio-psycho-social needs. Results indicate that elderly persons with complex bio-psycho-social needs do not receive appropriate mental health care.

Advancing Precision Medicine for Alcohol Use Disorder: Replication and Extension of Reward Drinking as a Predictor of Naltrexone Response.

BackgroundPrecision medicine aims to identify those patients who will benefit the most from specific treatments. Recent work found large effects of naltrexone among “reward drinkers,” defined as individuals who drink primarily for the rewarding effects of alcohol. This study sought to replicate and extend these recent findings by examining whether the desire to drink mediated the effect of naltrexone among reward drinkers.MethodsWe conducted a secondary analysis of a 12‐week randomized clinical trial of daily or targeted naltrexone among problem drinkers (n = 163), with a focus on 86 individuals (n = 45 naltrexone and n = 41 placebo) who received daily medication. Interactive voice response technology was used to collect daily reports of drinking and desire to drink. Factor mixture models were used to derive reward and relief phenotypes. Moderation analyses were used to evaluate naltrexone effects, with phenotype as a moderator variable. Multilevel mediation tested average desire to drink as a mediator.ResultsResults indicated 4 phenotypes: low reward/low relief; low reward/high relief; high reward/low relief; and high reward/high relief. There was an interaction between the high reward/low relief subgroup (n = 10) and daily naltrexone versus placebo on drinks per drinking day (DPDD; p = 0.03), percent heavy drinking days (p = 0.004), and daily drinking (p = 0.02). As compared to placebo, individuals in the high reward/low relief phenotype who received daily naltrexone had significantly fewer DPDD (Cohen's d = 2.05) and had a lower proportion of heavy drinking days (Cohen's d = 1.75). As hypothesized, reductions in average desire to drink mediated the effect of naltrexone on average daily drinking among the high reward/low relief drinkers (moderated mediation effect: p = 0.029).ConclusionsThis theory‐driven study replicates the empirical finding that naltrexone is particularly efficacious among high reward/low relief drinkers. Our study brings the field a step closer to the potential of using a precision medicine approach to treating alcohol use disorder.

Differential Patterns of the Division of Parenthood in Chinese Family: Association With Coparenting Behavior.

The current study explored the division of parenthood in Chinese families with adolescents by identifying the parental involvement patterns in the data obtained from 786 pairs of parents. Division-of-parenthood patterns were created via factor mixture modeling using self-reported three dimensions of father and mother involvement. Three differential division-of-parenthood patterns were identified: (a) parent-cooperation pattern, where moderate and equivalent involvement existed between mothers and fathers; (b) mother-dominated pattern, where mother involvement was particularly greater than father involvement; and (c) father-dominated pattern, where father involvement was particularly greater than mother involvement. Families were more likely to be in the mother- or the father-dominated pattern as their levels of positive coparenting behaviors increased. By contrast, as the levels of paternal conflict behavior increased, families were likely to be in the mother-dominated pattern. This study highlighted parents’ individual parenting role, the diverse division-of-parenthood patterns in the family, and the important role of coparenting behavior.

The latent structure of depression symptoms and suicidal thoughts in Brazilian youths

BackgroundMany studies have investigated the existence of discrete groups comprising the clinical concept of depression. Nevertheless, identifying true latent depression groups might require the inclusion of indicators of severe manifestations of depression, such as suicidal thoughts, in the analysis. Another issue is that relatively few studies have addressed the latent structure of depression in children and adolescents from Latin American cultures. MethodsIn the present study, we combined latent profile and taxometric analysis to investigate the latent structure underlying depression symptoms (negative emotionality and low positive emotionality indicators) and suicidal thoughts in four aggregated Brazilian youth samples (total N = 2587; mean age = 12.86 years; SD = 2.60; 50.8% females). ResultsLatent profile analysis indicated five classes that clearly represented distinct levels on a continuum of depression. However, taxometric results were ambiguous with regard to highlighting depression as a purely dimensional or categorical latent entity (mean CCFI = 0.497). LimitationsThe use of few indicators from a single instrument, the potential heterogeneity in the clinical group, and the non-random nature of the samples included in the study. ConclusionThe mixed findings support the existence of a dimension of depression, as well as latent classes of individuals. Factor mixture models are discussed as a strategy for further exploring the nature of depression among young people.

The complexity of trauma exposure and response: Profiling PTSD and CPTSD among a refugee sample.

This study investigated the latent dimensional and categorical structure of ICD-11 complex posttraumatic stress disorder (CPTSD) within a refugee sample. A subsample that identified as refugee (n = 308) was selected from the National Epidemiological Survey on Alcohol and Related Conditions. Factor mixture modeling (FMM) was employed to establish the dimensional structure of CPTSD symptomology and the categorical distribution of these dimensions. It was then evaluated whether trauma history could differentiate between the distribution of trauma response profiles. A correlated 6-factor model with 5 latent classes was the best fitting model. Two classes were characterized by symptom profiles that were consistent with ICD-11 CPTSD and PTSD formulations. The remaining classes were characterized by nonspecific variation across dimensions. CPTSD class membership was predicted by traumas that were predominantly interpersonal in nature (serious neglect, physical assault, and sexual assault), whereas PTSD class membership was predicted by situational traumatic experiences (unarmed civilian in a conflict environment and a serious accident). A distinct dose-response effect was evident between cumulative traumatic exposure and CPTSD class membership. FMM class profiles distinguished between PTSD and CPTSD symptom formulations. Moreover, class membership was determined by specific trauma-exposure histories. (PsycINFO Database Record (c) 2019 APA, all rights reserved).

Are dietary patterns differently associated with differentiated levels of mental health problems? Results from a large cross-sectional study among Iranian manufacturing employees

ObjectivesThe present study aimed to classify participants based on mental health problems profile and to evaluate its relationship with dietary patterns among Iranian manufacturing employees.DesignObservational study with a cross-sectional design.SettingThis...

A quantitative study of body dysmorphic disorder: Latent structure and correlates

Body dysmorphic disorder (BDD) is a body image disturbance of perceived ugliness that is underdiagnosed and often untreated. Although the nature of BDD is assumed to be synonymous with its conceptualization as a discrete DSM category, recent research suggests the possibility of a more complex latent structure. To clarify BDD’s latent structure, data from a large (N = 1,385), unselected sample was analyzed using an integrated, quantitative approach. From multiple measures of BDD, three symptom indicators were extracted that had the following appearance-related content: (1) perception—negative self-evaluation, (2) avoidance—fear of negative social evaluation, and (3) behavior—activities to modify flaws. Factor mixture modeling of the three indicators supported a hybrid categorical-dimensional latent structure with three distinct classes that had varying levels of the three extracted BDD symptoms. The three classes (subgroups of individuals) were a large non-symptomatic class (i.e., Non-BDD) and two smaller classes (i.e., Elevated Avoidance BDD, Elevated BDD). Analyses of conceptually overlapping personality and symptom measures showed social anxiety symptoms were highest in the Elevated Avoidance class and lowest in the Non-BDD class. Depression and obsessive-compulsive disorder symptoms were higher in the two BDD symptomatic classes than in the Non-BDD class, whereas negative affect was highest in the Elevated Avoidance class and lowest in the Non-BDD class. These findings suggest that modifications to current conceptualizations BDD could lead to improvements in BDD identification.

Factor mixture analysis of paranoia in young people.

Paranoid thoughts are relatively common in the general population and can increase the risk of developing mental health conditions. In this study, we investigate the latent structure of paranoia in a sample of young people. Cross-sectional survey; 243 undergraduate students (males: 44.9%) aged 24.3years (SD 3.5). The participants completed the Green et al. Paranoid Thought Scales GPTS, a 32-item scale assessing ideas of social reference and persecution; the 12-item General Health Questionnaire (GHQ-12), and the 74-item Schizotypal Personality Questionnaire (SPQ). Confirmatory factor analysis (CFA) was used to confirm the two-factor structure of the GPTS. Factor mixture modeling analysis (FMMA) was applied to map the best combination of factors and latent classes of paranoia. The GPTS showed excellent internal reliability and test-retest stability. Convergent validity was good, with stronger links with measures of ideas of reference and of suspiciousness than with other measures of psychosis-proneness. CFA showed excellent fit for the two-factor solution. FMMA retrieved a three-class solution with 176 subjects (72.5%) assigned to a baseline class, 54 (22.2%) to a "suspicious and mistrustful" class, and 13 (5.3%) to a "paranoid thinking" class. Compared to the baseline class, the other two classes had a higher risk of psychological distress and psychosis-proneness. The latent structure of paranoid thinking in young people appears dimensional. Although caution is advised when generalizing from studies on college students, screening for paranoid ideation in young people who complain about psychological distress might prove useful to prevent the development of severe and potentially debilitating conditions.

Negative self-evaluation and the genesis of internal threat: beyond a continuum of suicidal thought and behaviour.

Death by suicide is often preceded by attempted suicide, suicidal ideation and non-suicidal self-injury. These extreme thoughts and behaviours have been considered in terms of a continuum of suicidality. Little known research, however, has considered a suicide continuum that extends beyond these extreme thoughts and behaviours and incorporates a much wider array of phenomena that may vary in severity and may constitute a broader negative self-evaluation (NSE) continuum. Harvesting key indicators of NSE from a British epidemiological survey (N = 8580), the current study used exploratory factor analysis, confirmatory factor analysis and factor mixture modelling to (i) identify the dimensional structure of NSE in the general population and (ii) profile the distribution of the resultant NSE dimensions. Multinomial logistic regression was then used to differentiate between classes using an array of risk variables, psychopathology outcome variables and a suicide attempt indicator. A 4-factor model that reflected graded levels of NSE was identified; (F1) Low self-worth & subordination (F2) depression, (F3) suicidal thoughts, (F4) self-harm (SH). Seven classes suggested a clear pattern of NSE severity. Classes characterised by higher levels across the dimensions exhibited greater risk and poorer outcomes. The greatest risk for suicide attempt was associated with a class characterised by engagement in SH behaviour. Low self-worth, subordination and depression, while representative of distinct groups in the population are also highly prevalent in those who entertain suicidal thoughts and engage in SH behaviour. The findings promote further investigation into the genesis and evolution of suicidality and internal threat.

The Performance of Model Fit Indices for Class Enumeration in Multilevel Factor Mixture Models

Factor mixture models combine the common factor model and latent class analysis. Given that multilevel data structures are very common in educational and social research, the multilevel factor mixture model (ML FMM) is appropriate for analyzing nested measurement data when population heterogeneity is unobserved. This simulation study aims to investigate the performance of model fit indices with multilevel factor mixture models under various conditions. In data simulation, the five-items and one-factor model with between- and within-cluster was chosen. Two subgroups with the factor mean difference were simulated so two-class was the correct number of classes. To investigate the performance of information criterions, the following conditions were manipulated in this study: class separation, the intraclass correlation ( ICC ), sample size. For each of the generated dataset, one correct model and three mis-specified models were analyzed to fit the data. The results showed that class separation was an important factor on detecting the correct number of classes in multilevel factor mixture models. The proportion correct increases as the class separation gets larger. Although no single criterion is always best, AIC yield a more accurate model selection than aBIC and BIC overall. Only when class separation is large, aBIC is more trustworthy for model selection. The results of this study can provide the information for educational researchers interested in analyzing multilevel data when the heterogeneity of the population is unknown.

Investigating Sources of Heterogeneity with Three-Step Multilevel Factor Mixture Modeling: Beyond Testing Measurement Invariance in Cross-National Studies

We propose the three-step multilevel factor mixture modeling (ML FMM) to test measurement invariance (MI) across many groups and furthermore to model predictors of latent class membership that possibly induce measurement noninvariance. This Monte Carlo simulation study found that information criteria such as Bayesian Information Criterion tended to select a more complex model when sample size was very large. Thus, the adequacy of three-step ML FMM regarding the correct MI detection was demonstrated with an empirically derived information criterion for large data. However, the number of latent classes was overestimated when intraclass correlation was large. For the test of covariate effects, Type I error was well controlled and power was generally adequate when a correct model was identified at Step 1. Using background variables selected from Trends in International Mathematics and Science Study 2011, the application of three-step ML FMM to a cross-national MI test is demonstrated.

Testing the Assumption of Population Homogeneity in the Measurement of Dispositional Optimism: Factor Mixture Modeling Analysis

The assumption of population homogeneity in the measurement of dispositional optimism was examined. U.S. and Canadian respondents (N = 591) completed an optimism scale. Possible population heterogeneity was analyzed using factor mixture modeling. Two major results emerged. First, population homogeneity was not supported: A large class of participants had trouble giving consistent answers to optimism items, optimism and pessimism items, or, especially, pessimism items. Second, after the removal of this problematic class, the correlation between optimism and pessimism was found to be .94, a magnitude large enough to support the assumption of unidimensionality. Although psychometric problems with the measurement of optimism have not been identified previously, this study suggests that the typical measurement of dispositional optimism requires substantial revision. The findings showcase the importance of factor mixture modeling in evaluating the psychometric properties of a measurement scale.

Investigating Response Heterogeneity in the Context of Positively and Negatively Worded Items by Using Factor Mixture Modeling

Factor mixture modeling was used to investigate potential response incongruity between positively and negatively worded items. Survey respondents ( N = 591) answered questions about job satisfaction and dissatisfaction. Results revealed two classes of respondent: a majority class, who generally do not have problems answering positively and negatively worded items, and a minority class, who have serious trouble with negatively worded items. With the exclusion of the minority class, job satisfaction and dissatisfaction were found to be essentially unidimensional rather than bidimensional as previous research had suggested. These results challenge previous findings regarding the bidimensionality of job satisfaction and cast doubt on the widespread research practice of assuming population homogeneity in survey responses. A detailed flowchart illustrating the analytic procedure and an Mplus syntax program are provided so that researchers can conduct similar investigations on constructs of interest.

Profiles of Psychological Resilience in College Students With Disabilities

The present article reports on research conducted to identify profiles of psychological resilience using factor mixture models. We also examine gender as a predictor of resilience profile membership and career optimism, academic satisfaction, and psychological well-being as outcomes of profile membership. Based on resilience data from university students with disabilities, factor mixture modeling revealed three distinct profiles of resilience (viz., “vulnerable,” “spirituality-dominant,” and “engaged-resilient”). Results also revealed that females were almost 4 times as likely to be in the spirituality-dominant profile than the vulnerable profile. Finally, distal outcome analyses revealed that career optimism, academic satisfaction, and well-being were higher in the engaged-resilient profile than the other profiles. Notably, spirituality-dominant and vulnerable individuals possessed about the same levels of career optimism, satisfaction, and well-being. The findings have important implications for the theory and assessment of resilience, suggesting the tenability of a person-centered assessment of psychological resilience.

Reliability, Population Classification and Weighting in Multidimensional Poverty Measurement: A Monte Carlo Study

In poverty measurement, differential weighting aims to take into account the unequal importance of the diverse dimensions and aspects of poverty and to add valuable information that improves the classification of the poor and the not-poor. This practice, however, is in contention with both classical test theory and modern measurement theories, which state that high reliability is a necessary condition for consistent population classification, while differential weighting is not so. The literature needs a clear numerical illustration of the relationship between high/low reliability and good/poor population classification to dissolve this tension and assist applied researchers in the assessment of multidimensional poverty indexes, using different reliability statistics. This paper uses a Monte Carlo study based on factor mixture models to draw up a series of uni-and multidimensional poverty measures with different reliabilities and predefined groups. The article shows that low reliability results in a high proportion of the poor group erroneously classified as part of the not poor group. Therefore, reliability inspections should be a systematic practice in poverty measurement. The article provides guidelines for interpreting the effects of unreliability upon adequate population classification and suggest that the classification error of current unreliable multidimensional indexes is above 10%.