Investigating the Psychometric Properties of the Emotionality Activity and Sociability Temperament Survey among Young Preschoolers in Norway

The early years of life are characterized by rapid brain growth and the important establishment of nurturing relationships with parents and professional caregivers (Shonkoff et al., 2012). This period is crucial for children’s early development, with lasting effects throughout life. A true understanding of young children’s development requires consideration of individual differences, as these differences can significantly impact their interaction with important caregivers and developmental trajectories (Berger, 2016). With the increasing attendance of toddlers in early childhood education and care (ECEC)(Statistics Norway, 2023), teacher-child interactions in ECEC play a prominent role for children’s development. Individual differences in child behavior in the form of temperamental characteristics is an important child-related factor that may play a crucial role in shaping children’s interactions with their teachers (Susa-Erdogan et al., 2022). Caspi and Moffitt (1993) emphasize that a child’s temperament can alter how they respond to environmental stimuli, which environmental stimuli they approach, and how others interact with them. For instance, a highly active and emotionally reactive child may face challenges in a structured classroom environment in ECEC (compared to e.g. a home setting), whereas a more adaptable and calm child may excel in the same setting (Baardstu et al., 2020; Phillips et al., 2011). In studies of children’s development in association with their environments, such as ECEC, it is increasingly recognized to include children’s temperamental differences (Phillips et al., 2011; Pluess & Belsky, 2009; Wilhelmsen et al., 2021). Therefore, it is important to measure children’s temperament in a simple and valid manner, and to understand the validity across different contexts and observers in children’s early life.

The Emotionality Activity and Sociability (EAS) Temperament Survey, developed by Buss and Plomin (1984a, b), is one of the most widely used measures within the field of developmental psychology. It is also one of the few measures of temperament designed specifically to assess young children, originally targeting ages 1–9 years (Walker et al., 2017). Originally, it was meant to measure emotionality, activity, sociability, and impulsivity. However, impulsivity was removed for lack of stability, and sociability was redefined as shyness while an experimental sociability subscale was included (Buss & Plomin, 1984a, b). The EAS Temperament Survey in its current form evaluates four temperament dimensions: Emotionality, Activity, Sociability, and Shyness.

The emotionality dimension assesses the intensity of emotional reactions. Individuals scoring high on emotionality tend to experience emotions strongly and may exhibit greater mood variability. Conversely, those scoring low are more even-tempered and less prone to emotional fluctuations. The activity dimension measures the level of physical movement and energy expenditure. High scorers are typically more energetic and may seek out active pursuits, while low scorers are more relaxed and may prefer quieter, less physically demanding activities. The third dimension, sociability, evaluates the tendency to seek out and enjoy social interactions. High scorers are typically outgoing, enjoy being around others, and seek social stimulation. Low scorers are more reserved, preferring solitude or smaller social gatherings. Finally, the fourth dimension, shyness, indicates discomfort or anxiety in social situations, particularly when meeting new people or being at the center of attention. High scorers are more likely to avoid unfamiliar social interactions or feel anxious in such situations, while low scorers are more confident and at ease in social settings.

Despite being widely used, a recent review of the EAS concluded that the mixed results regarding the four-factor structure of the parent-rated EAS warrant further examinations of its psychometric properties across different ages and gender (Walker et al., 2017). Although the review included only a few studies on the factor structure of the EAS, and excluded all translated versions, the authors summarized that the four-factor structure of the EAS has received varying levels of support with concerns particularly regarding the reliability of the sociability and shyness factors. Using parent ratings of six to twelve-year-old Dutch children, Boer and Westenberg (1994) failed to replicate the four-factor structure of EAS, because the sociability and shyness subscales were not distinct. Similar results were found for parent ratings of three-year-old children in the UK (Stingaris et al., 2010), where a three-factor model was supported combining the sociability and shyness factors while two items (no.5 and no.18) had weak factor loadings. However, subsequent analyses of the same UK sample following the children until six years of age, supported the four-factor structure for three, five, and six-year-old children, although one item (no.18) did not load on the expected sociability scale for the younger children (Bould et al., 2013). These results support the findings from a previous evaluation by Mathiesen and Tambs (1999) of the factor structure and reliability of the EAS and its stability using parent report across three age levels (18, 30, and 50 months) in a Norwegian preschool sample. They concluded that the four-factor structure of the EAS was appropriate and largely stable from 18 to 50 months of age, and also noted that the internal consistency of each subscale improved with age, in particular for the sociability subscale which demonstrated poor internal consistency at 18 months (Cronbach’s alpha from 0.48 at 18 months to 0.60 at 50 months).

Using both parent and teacher reports of the EAS among children aged six to twelve years in France, Gasman et al. (2002) reported suboptimal fit indices of the four factors of the EAS (e.g., CFIs = 0.72–0.73), while each of the four factors had sufficient reliability among both parents (Cronbach’s alpha = 0.69–0.77) and teachers (Cronbach’s alpha = 0.79–0.84). Importantly, they found a strong negative correlation between sociability and shyness (r = −.91 and − 0.98). The authors argue that cultural and contextual differences in the interpretation of children’s temperament encourage further examinations of children’s temperament scales based on different observers (e.g., ECEC and schoolteachers) and in different cultures.

In addition to potential contextual and observer differences of understanding the EAS, one should also recognize differences in children’s age. Buss and Plomin (1984a, b) suggested that the temperament factors would be more distinct with increasing age, as infant children show more diffused behavior. Whereas emotionality and activity are usually found to be distinct across all ages, the age difference in sociability and shyness is supported in the longitudinal studies of the EAS (Mathiessen & Tambs, 1999; Bould et al., 2013). However, other studies with elementary school children aged six to twelve years still do not find support for the four-factor model either and find overlapping items for sociability and shyness (Boer & Westernberg, 1994; Gasman et al., 2002). Aligned with theory, that differences in temperament in infant years play a role for children’s interactions and environmental stimuli (Belsky & Pluess, 2009;Caspi & Moffit 1993), and research acknowledging individual temperamental differences in response to ECEC quality (Baardstu et al., 2022; Phillips et al., 2011; Pluess & Belsky, 2009; Wilhelmsen et al., 2021), we need evaluations of the EAS factor structure for toddlers in ECEC.

Studies examining children’s development depending on individual differences in temperament, often include or control for potential sex differences as well. While Gasman et al. (2002) did not find sex differences in either parent- nor teacher-rated temperament scores, Mathiesen and Tambs (1999) found that boys and girls had somewhat different mean levels on the different temperament scales across age, but not in a consistent pattern. Bould et al. (2013) concluded that girls had higher scores on emotionality, sociability and shyness, while boys had higher activity, and Baardstu et al. (2020) found that parents rated girls to exhibit more shyness than boys at 18 months of age, while boys had a peak in shyness in middle childhood. However, none of these studies directly examined measurement invariance (MI) of the EAS temperament survey across child sex i.e., whether the different temperament scores can be interpreted and measured by the items in the same way for boys and girls. Without MI, it is difficult to evaluate whether such differences in temperament across sex are due to true differences in levels of temperament or just differences in the measurement properties of the EAS. To allow accurate comparisons of sex differences in temperaments and ensure that the factors are understood in consistent ways across these groups of individuals, we need to address the MI of the EAS across boys and girls.

In sum, there is a lack of psychometric assessment of instruments applied in the Nordic countries in general, especially for the youngest children (Peltonen et al., 2023). As highlighted by Peltonen et al. (2023) in their review, it is important to use validated and culturally appropriate assessment tools to ensure accurate and meaningful results when assessing children’s development. As a result, the current study aims to test the factor structure and psychometric properties of the EAS Temperament Survey using ECEC professionals ratings of children under three years old. Although Mathiessen and Tambs (1999) concluded that the EAS could be used for children from 18 months of age, Gasman et al. (2002) argued that one should acknowledge the contextual differences in children’s environment and evaluate measures by different raters, as they did by examining teachers’ ratings in elementary school. Despite examining sex differences in temperament, neither study evaluated MI between boys and girls. With the increased use of ECEC for children under the age of three years, and continued attention to sex differences in early development, we therefore extend both Mathiessen’s and Gusman’s studies by examining the factor structure of the EAS with teacher ratings in ECEC for children under three years old, and evaluating the MI across children’s sex.

The descriptives and reliability analyses can be found in Table 2. Regarding the factors reliability in terms of internal consistency, active (ω = 0.79 [0.77, 0.81]), emotional (ω = 0.86 [0.84, 0.87]), and shyness (ω = 0.75 [0.72, 0.77]) exhibited satisfactory internal consistency, while social (ω = 0.64 [0.61, 0.68]) exhibited ω estimate below the recommended threshold of ≥ 0.70.

The preliminary CFA of the EAS indicated a poor model fit of the data for the full sample (RMSEA = 0.094 [0.091, 0.098], CFI = 0.77, TLI = 0.74, SRMR = 0.084) with similar estimates when grouped by children’s sex (Table 3). The standardized factor loadings for the full sample (Fig. 1) ranged between − 0.04 (item 18, Social “When alone, child feels isolated”) and − 0.83 (item 12, “Is very sociable”). The lowest correlation between the factors was found for Active and Emotional (-0.10), while the highest was found between Shyness and Social (-0.98) (Table 4). Thus, an empirical approach was applied to see if we could gain better data fit by inspecting the chi square modification fit indices. Three options were tested: (1) removing item pairs stepwise based on their chi square, (2) relaxing covariates stepwise based on their chi square, (3) and testing a higher order model. None of the approaches yielded satisfactory model fit. As this approach was purely explorative and empirical driven in nature, further testing was abandoned. Also, as the model fit for boys and girls individually indicated poor fit, investigation of the MI was also abandoned.

The aim of the current study was to investigate the psychometric properties of the EAS and its applicability in an ECEC context by examining its factor structure and measurement invariance (MI) across sex. The EAS factors generally exhibited satisfactory internal consistency with the exception of the factor Sociability, which was slightly below the recommended threshold of ≥ 0.70. However, the data exhibited poor model fit for the original factor structure of the EAS and post hoc explorative analysis did not improve the model fit to satisfactory levels. Several authors propose that post hoc analysis should be carried out with great caution and only performed if it is theoretically and practically plausible, as modifications may lead to under-identified or overfitted models and capitalize on chance variance in the sample (Flora & Flake, 2017; Jackson et al., 2009; Tarka, 2017). Model modifications are usually made by omitting or freeing parameters and should be considered purely explorative and empirically driven. Consequently, model modifications may improve the model fit at the expense of theory and replicability (Hermida, 2015). In the current study, neither omitting or freeing parameters improved the model fit to a satisfactory level. Due to the poor model fit, the investigation of the MI was abandoned, as invariance testing of unreliable factors would be meaningless.

Despite poor model fit, the factor structure of the EAS with teacher reports of children under three years old do not deviate a lot from previous examinations of factor structure with parents and teachers. Specifically, studies which included children under three years old found poor fit and reliability for the sociability subscale at this age (Stingaris et al., 2010; Bould et al., 2013; Mathiessen & Tambs 1999). As we found poorest fit for sociability and shyness, the current results are consistent with the argument that sociability and shyness might become more distinct with age (Buss & Plomin, 1984a, b).

However, age differences cannot explain the poor fit alone. Previous research suggests that the overlapping items and high correlations between shyness and sociability are present when assessing older children as well (Gusman et al., 2002; Boer & Westernberg, 1994). In our analyses, the sociability and shyness subscales correlated strongly (r = −.98), which corresponds with the results of parent and teacher ratings of six- to twelve-year-old children in France (Gusman et al., 2002). The very high correlation between Shyness and Sociability indicates issues with multicollinearity, suggesting redundancy, as these two factors largely measure a common factor. Additionally, the presence of multicollinearity has shown to impact psychometric properties negatively, such as unreliable factor structures and inconsistent factor loadings (Kyriazos & Poga, 2023). Theoretically, shyness and sociability differ in that shyness is more closely aligned with fear or insecurity with new people or social situations, while sociability is not (Buss & Plomin, 1984a, b), and early shyness seems linked to higher neuroticism in later adolescence (Baardstu et al.,2020). Examining the factor loadings on the shyness subscale in the current analyses, we found the highest loadings for item 12 “the child is very sociable” and item 8 “the child easily makes friends”. These two items were less associated with fear or with strangers than the other shyness items, and the stronger factor loading on these items from the shyness subscale may inflate the correlation between the sociability and shyness subscales.

Additionally, these two items were reverse coded. Reverse coded items are commonly used to control for acquiescence and social desirability effects, but may have negative effects of an instrument’s psychometric properties (e.g., internal consistency and model fit), as it may introduce multidimensionality to factors which are supposed to be unidimensional (Vigil-Colet et al., 2020). Mathiesen and Tambs (1999) also mentioned that these two items had low face validity as they also loaded on the sociability subscale, and they suggested a rewording of items to appropriately reflect the sociability and shyness constructs. Such modifications could potentially emphasize the distinction between shyness and sociability based on social fear.

An item with low factor loading (-0.04) on the sociability scale was item 18 “the child dislikes being alone”. This item could potentially convey more fear or insecurity, compared to item 16 “prefers to be alone” which convey a choice. In the UK sample of three-to-six-year-olds rated by mothers, item 18 (here phrased “when alone, child feels isolated”) also had low factor loadings on sociability (0.26 to 0.33) (Bould et al., 2013). It could be that this distinction would be even more pronounced in the ECEC context compared to the home context. Although Gusman et al., (2002) found higher factor loadings for this item with ratings by primary school teachers (0.65) than by parents (0.26), the ECEC context (non-parental care) might be less familiar to a three-year-old than the school context is to a six-year-old, which could explain the low factor loading in the current sample. Furthermore, it should be acknowledged that the EAS was originally designed for parent-child interactions, with parents responding based on their close, personal knowledge of their child. However, the dynamics between early childhood educators and children differ significantly due to the group-based context of ECEC. Educators typically manage groups of 14–19 children, which shapes their professional, less personal relationships. This professional context might influence how educators interpret EAS items. Some nuances, like individual temperamental traits, may be harder to detect in group settings. Daily interactions with many children provide educators with a broad comparative framework, contrasting parents’ focus on one or a few children. Additionally, educators’ responses may emphasize behaviors that disrupt group dynamics, affecting ratings on items like item 1 “The child is easily shy.” This trait might stand out more in a classroom than at home. Cultural norms in preschools also value adaptability and group harmony, potentially altering how traits like sociability are assessed. Thus, educators may focus on behaviors that impact group dynamics or learning environments, interpreting traits through a lens of classroom management. This underscores the importance, as Gasman et al. (2002) argued, of recognizing the contextual differences in children’s environments and assessing measures using various raters also in the future. Moreover, it further suggests that future modifications of the EAS items could compare parent and teacher ratings and perhaps retain those that are less context specific in order to measure these factors as traits of the child.

Finally, the poor model fit may also be explained by sub-optimal methods used when developing the EAS. The EAS was developed with varimax rotation and by extracting factors with eigenvalues larger than one (Buss & Plomin, 1984a, b), methods in later years criticized for its inappropriateness and inaccuracy (Flora & Flake, 2017). For instance, varimax rotation is an orthogonal type of rotation, which forces the factors to be uncorrelated. However, the consensus among methodologist is that oblique rotations, where factors are allowed to correlate, should be preferred in almost every situation as it is unrealistic to expect that factors are perfectly uncorrelated (Flora & Flake, 2017). As seen in Fig. 1, the factors in the current study are indeed correlated, indicating that orthogonal rotation of factor loadings in the explorative stage would produce worse model fit than an oblique rotation would.

The EAS generally demonstrates satisfactory internal consistency; however, the original four-factor structure was not supported in this study. The findings align closely with those of studies that have examined parent-reported data, often focusing on older children, and suggest that the EAS is not a reliable instrument in its current form. Future research should explore the psychometric properties of the EAS across diverse contexts, using samples with varying characteristics and different types of informants. Additionally, as noted in previous studies (e.g., Walker et al., 2017), consideration should be given to potential modifications to the scale. At present, the issues appear to stem from the distinctions or overlaps between the constructs of shyness and sociability, while emotionality and activity seem to perform more consistently. Moving forward we suggest to address these issues by modifications of items in the Sociability and Shyness factors, to make the factors more distinct based on social fear, and also make item wordings less context specific.