Poster #92 - Examination of dimensionality of scales of the Infant Behavior Questionnaire-Revised (IBQ-R)

• In Event: Poster Session 15

Abstract

The Infant Behavior Questionnaire-Revised (IBQ-R; Gartstein & Rothbart, 2003) is the most popular assessment for infant temperament, which consists of 191 items with 14 scales. The dimensionality of scales, however, has yet to be examined. Dimensionality is important because items in a scale reflect the same underlying construct only when unidimensionality is assured. Although the scale scores of the IBQ-R are computed by averaging all items within each scale assuming unidimesionality, no studies have examined the dimensionality beyond reporting Cronbach’s alpha, which has been argued as not a measure of unidimensionality. In this study, I assessed the dimensionality of the IBQ-R scales using factor analytic methods.
Data sets (K=11) were provided by researchers who published research using the IBQ-R and from two longitudinal studies. The combined dataset included 3152 infants (1521 females) aged 1 to 14 months (M = 7.89, SD = 3.28). Race was available for 1498 infants and 82.3% were white.
I first tested unidimensional models using confirmatory factor analysis (CFA) with weighted least square mean and variance adjusted (WLSMV) estimator and model fit indices of the comparative fit index (CFI). The CFI > .90 was considered acceptable fit (Little, 2013). Next, I assessed dimensionality using exploratory bifactor models. Two bifactor model indices, hierarchical omega (H; Zinbarg et al., 2005) and the explained common variance (ECV; sijtsma, 2009) were examined. H > .80 and ECV > .70 (Reise et al., 2013) were interpreted as support for the unidimensionality, indicating that substantial proportion of total score variance explained by the general factor.
Unidimensional interpretations were supported for only 5 scales among 14 scales. For Vocal Reactivity, High Intensity Pleasure, Smiling and Laughter, Perceptual Sensitivity, and Fear, the unidimensional models fit the data well (CFA > .90) and both H (0.84 < H < .87) and the ECV (0.70 < ECV < 0.77) exceeded the benchmark. For Approach and Cuddliness, the unidimensional models fit the data well (CFA > .90) and H exceeded the benchmark, 0.87 and 0.86, but the ECV did not exceed the benchmark, 0.66 and 0.67. For Activity Level, Sadness, Distress to Limitation, Falling Reactivity, Low Intensity Pleasure, Duration of Orienting, and Soothability, the unidimensional interpretations appeared not appropriate given the poor fit of unidimensional models (0.75 < CFA < 0.88) and that neither H (0.74 < H < .80) or the ECV (0.47 < ECV < 0.58) exceeded the benchmark. For Activity Level, Falling Reactivity, and Soothability, exploratory bifactor models even failed to yield a general factor.
The results first provide information about the dimensionality of the IBQ-R scales beyond the alpha, which warrants more in-depth investigation on each scale as unidimensionality was not substantiated for majority of the scales. It is noteworthy that unidimensional scales are mostly for the higher-order factor of Surgency/Positivity Affectivity. Only one scale, Fear, for Negative Affectivity, and none for Regulatory Capacity factors showed unidimensionality. Therefore, the effects of dimensionality on the higher-order factors structure need to be further investigated.

Author

• Sooyeon Sung, University of Minnesota - Twin Cities
  Presenting Author