Dissections, Resections, and Reflections of Personality Traits in Australian Children: Adaptations and Validation of a Translated Big Five Children’s Personality Scale
thesis
posted on 2017-02-08, 23:54 authored by Dianne WattThe lexical
hypothesis has allowed the development of comprehensive personality trait
modelling with measures derived from language, culminating in the
well-researched Big Five and the five-factor model (FFM). Research using
parental descriptors has seen this extended to the construction of developmentally
appropriate scales for children, not least the 144-item Dutch-English language
translated Hierarchical Inventory of Personality in Children (HiPIC) developed
through principal component analysis of Flemish parental sentences describing
child behaviour.
Translation introduces potential error variance through questions of semantic and cultural equivalence. Principal component analysis use in scale construction adds unidentified error variance to the scale’s ability to present the true reflection of the intended constructs. Unidentified response variance in scale construction further confounds this reflection. Furthermore, the effect of response burden or the cognitive effort required to respond to items can multiply such impact. Finally, personality trait research in children has principally been considered from a variable-centred perspective: augmentation with a person-centred, typological approach can substantially facilitate understanding and clinical utility.
This thesis aimed to identify and remove the potential error variances embedded within an a priori five-factor solution of the HiPIC to determine whether the resultant scale versions maintained construct and predictive validity. A sequence of increasingly rigorous, metaphorical dissections or psychometric techniques removed or resected the error from the HiPIC. Initially, semantic invariance was identified and removed through a mixed method approach incorporating cognitive interviewing of ten HiPIC end-users integrated with iterative single-component principal component analyses of the internal structure of facets, to develop an adapted 124 item HiPIC-A (N = 399). In subsequent studies, guided by Classical test theory, single congeneric modelling, and a blended variable, structural equation modelling (SEM) approach, unidentified error variance and the response biases of acquiescence and social desirability were resected from the HiPIC. Three measures of potential social desirability were estimated for comparison, including factor score imputation of a bifactor model. A shortened version of the HiPIC, was presented as a means of reducing the multiplier effect of error variance. Latent profile analyses of the models allowed an understanding of the HiPIC models at the person-centred level. Predictive validity of the resultant versions was explored using a broadband psychosocial scale.
Initial findings show that the initial HiPIC-A structure was confirmed through procrustes rotation (Tucker’s φ = .9972). The short version of the HiPIC (34 items) also demonstrated excellent congruence (φ = .98) advancing the methods used to explore the HiPIC. Discriminant validity is reduced in two HiPIC factors with resection of unidentified error variance however, this loss is restored when acquiescence and social desirability were removed at the item level in a bifactor model of the HiPIC. This suggests that the parental response biases play a role in confounding constructs measured by the HiPIC. The general factor of this bifactor model correlated (average Pearson’s r = .82) with two measures of social desirability, lending support for this general factor of personality as a measure of social desirability.
The person-centred typologies demonstrated theoretically predictable relationships with the external psychosocial measure, including a Vulnerable child typology, at risk of both internalising and externalising symptomatology. The blended variable, bifactor model of the HiPIC-A presented four types of children, although the research shows a loss of ability of the imputed factor scored Emotional Stability to discriminate between types. This surprising result is discussed and one explanation proffered is the impact the true factor score presented in the bifactor, blended variable model, has on predictive validity at the factor level, relative to an enforced simple structure of sum scoring facets and factors.
Taken together, the studies presented and evaluated several error-resected models of the HiPIC as a measure of the FFM of personality in children. Methods for embedding and shortening a translated scale in an alternative culture were progressed. Modelling social desirability as a general factor confirmed a blended variable FFM version of the HiPIC, simultaneously questioning the existence of the general factor of personality. Demonstrated differential impact of social desirability across the HiPIC factors and facets has meaning for clinical interpretation of this scale indicating at the least, a development of an impression scale within the current HiPIC would benefit interpretation of this scale. Further research is also considered.
Translation introduces potential error variance through questions of semantic and cultural equivalence. Principal component analysis use in scale construction adds unidentified error variance to the scale’s ability to present the true reflection of the intended constructs. Unidentified response variance in scale construction further confounds this reflection. Furthermore, the effect of response burden or the cognitive effort required to respond to items can multiply such impact. Finally, personality trait research in children has principally been considered from a variable-centred perspective: augmentation with a person-centred, typological approach can substantially facilitate understanding and clinical utility.
This thesis aimed to identify and remove the potential error variances embedded within an a priori five-factor solution of the HiPIC to determine whether the resultant scale versions maintained construct and predictive validity. A sequence of increasingly rigorous, metaphorical dissections or psychometric techniques removed or resected the error from the HiPIC. Initially, semantic invariance was identified and removed through a mixed method approach incorporating cognitive interviewing of ten HiPIC end-users integrated with iterative single-component principal component analyses of the internal structure of facets, to develop an adapted 124 item HiPIC-A (N = 399). In subsequent studies, guided by Classical test theory, single congeneric modelling, and a blended variable, structural equation modelling (SEM) approach, unidentified error variance and the response biases of acquiescence and social desirability were resected from the HiPIC. Three measures of potential social desirability were estimated for comparison, including factor score imputation of a bifactor model. A shortened version of the HiPIC, was presented as a means of reducing the multiplier effect of error variance. Latent profile analyses of the models allowed an understanding of the HiPIC models at the person-centred level. Predictive validity of the resultant versions was explored using a broadband psychosocial scale.
Initial findings show that the initial HiPIC-A structure was confirmed through procrustes rotation (Tucker’s φ = .9972). The short version of the HiPIC (34 items) also demonstrated excellent congruence (φ = .98) advancing the methods used to explore the HiPIC. Discriminant validity is reduced in two HiPIC factors with resection of unidentified error variance however, this loss is restored when acquiescence and social desirability were removed at the item level in a bifactor model of the HiPIC. This suggests that the parental response biases play a role in confounding constructs measured by the HiPIC. The general factor of this bifactor model correlated (average Pearson’s r = .82) with two measures of social desirability, lending support for this general factor of personality as a measure of social desirability.
The person-centred typologies demonstrated theoretically predictable relationships with the external psychosocial measure, including a Vulnerable child typology, at risk of both internalising and externalising symptomatology. The blended variable, bifactor model of the HiPIC-A presented four types of children, although the research shows a loss of ability of the imputed factor scored Emotional Stability to discriminate between types. This surprising result is discussed and one explanation proffered is the impact the true factor score presented in the bifactor, blended variable model, has on predictive validity at the factor level, relative to an enforced simple structure of sum scoring facets and factors.
Taken together, the studies presented and evaluated several error-resected models of the HiPIC as a measure of the FFM of personality in children. Methods for embedding and shortening a translated scale in an alternative culture were progressed. Modelling social desirability as a general factor confirmed a blended variable FFM version of the HiPIC, simultaneously questioning the existence of the general factor of personality. Demonstrated differential impact of social desirability across the HiPIC factors and facets has meaning for clinical interpretation of this scale indicating at the least, a development of an impression scale within the current HiPIC would benefit interpretation of this scale. Further research is also considered.
