Measuring emotional contagion as a multidimensional construct: the development and initial validation of the contagion of affective phenomena scales

ABSTRACT

We offer an alternative conceptualization of the construct of susceptibility to emotional contagion and four related studies where two separate measures were developed and initially validated. The Contagion of Affective Phenomena Scale-General (CAPS-G) is a 5-item scale that measures the general susceptibility to the contagion of affect, and the Contagion of Affective Phenomena Scale – Emotion (CAPS-E) assesses six distinct emotions. Study 1 generated items with experts. Study 2 explored and confirmed construct validity and the factorial structure of both measures using exploratory structural equation modeling. Study 3 established test–retest reliability, concurrent validity, and discriminant validity. Study 4 found predictive validity with a sample of competitive swimmers. In four separate samples, a 21-item and 6-factor first-order structure of CAPS-E provided the best model fit. We provide initial evidence that supports the use of CAPS-E and CAPS-G as reliable and valid measures of the susceptibility to contagion of affective phenomena.

KEYWORDS:

• Affect
• emotional contagion
• measure
• mood convergence
• psychometric

Introduction

The contagion of affective experiences (viz. emotion, mood, or affect) between individuals is a complex interpersonal phenomenon, consisting of unconscious and conscious processes whereby individuals transmit and catch others’ affective experiences. Over the last 20 years, a proliferation of research across multiple domains has explored the contagion of affective phenomena between individuals (Tee, 2015), including the study of individual susceptibility to emotional contagion (e.g., Hatfield et al., 2014), mood contagion (e.g. Bono & Ilies, 2006), mood convergence (e.g., Spoor & Kelly, 2009), affective display transfer (e.g., Ilies et al., 2007), and group-level affective tone (e.g., Tsai et al., 2012). Nevertheless, there exists debate regarding the conceptualization of contagion with multiple operationalizations, an absence of an overarching guiding framework and the lack of consensus regarding the measurement of this concept (Clarkson et al., 2020).

In this paper, we present a new, versatile set of contagion measures, the Contagion of Affective Phenomena Scales (CAPS),¹ to address the substantial and significant shortcomings in the measurement of affective contagion. First, a recent systematic, meta-analytical review identified that scholars often only infer the occurrence of emotional contagion between participants rather than directly measure it (Clarkson et al., 2020). While Doherty’s (1997) Emotional Contagion Scale is a specific self-report measure of susceptibility to emotional contagion, it only measures emotion susceptibility and omits broader affective phenomena (viz. mood, affect). As only 33% of the affective contagion literature investigates emotional contagion (Clarkson et al., 2020), we propose that either the ECS does not adequately align with most researchers’ alternative conceptualizations of contagion, and/or there is a lack of confidence in the useability of the measure. Indeed, the construct validity of the ECS has been questioned (although it is a highly cited measure) and whether susceptibility to emotional contagion is a unidimensional or multidimensional construct (see. Wróbel & Lundqvist, 2014). This reflects a broader academic debate, also noted by Clarkson et al. (2020), concerning the multiple operationalizations of contagion (e.g., as an individual difference variable, and/or process) that is further confounded by researchers’ theoretical subscription to affective experiences (viz. emotion, mood, or affect). As an example of this issue and need for a new complementary measure, if we wanted to empirically test both dispositional mood contagion susceptibility and to anger contagion, the ECS does not provide this versatility. Researchers who subscribe to general susceptibility as unidimensional and the susceptibility to discrete emotions as multidimensional also do not possess a versatile measure. Therefore, and as we assert throughout this introduction, there is a need for a complementary, psychometrically tested versatile instrument that can be used to dynamically measure the susceptibility to contagion of affective phenomena. To address the proposed alternative conceptualization, the introduction contains two main sections: (i) formation of an underlying conceptual framework and (ii) review of empirical research regarding the measurement of contagion.

Overview of the conceptualisation

Emotional contagion can be summarized as a complex, four-dimensional process (Barsade et al., 2018). However, the early conceptualization of “primitive emotional contagion” by Hatfield et al. (1994) is the most adopted theoretical framework by researchers, Kelly and Barsade (2001) and others (discussed below) have argued that labeling emotional contagion as occurring only at an unconscious level ignores the influence of conscious cognitive and emotional social comparison processes. Hatfield et al. (1994) argue individuals continuously and automatically mimic their movements with the facial expressions, vocalizations, postures and movements of others. This mimicry generates afferent feedback and subsequent subjective, momentary emotional experiences, leading to emotional convergence between individuals and what is better known as primitive emotional contagion. This conceptualization neglects the conscious mechanisms (e.g., social comparison and conscious cognitive processes, affective impression management) of emotional contagion and may risk conceptual over-simplification. For example, an individual may realize another person is angry, which may trigger recollections of their own anger, leading to similar felt emotions. Further, according to the Emotions-As-Social-Information model (EASI: Van Kleef, 2009), affective reactions to emotional expressions are influenced by the observer’s information processing and social-relational factors (e.g., how the emotion is expressed, prevailing cultural norms, and the interpersonal relationship) and that these affective reactions impact behavioral responses. Consequently, it may be possible to assess susceptibility to emotional contagion through self-report measures that tap the conscious contagion processes.

It is noteworthy that our distinction of susceptibility to emotional contagion is as an individual difference variable and not as an operationalization of emotional contagion (see Doherty, 1997). Individuals’ susceptibility to emotional contagion can be defined as “the extent to which emotional stimuli elicit an emotional expression characteristic of the eliciting emotion” (Doherty, 1997, p. 134). The strength of the physiological, cognitive, and behavioral reactions when viewing others’ emotions will differ from person to person. If an individual is highly susceptible to emotional contagion, they are likely to mimic others’ emotional expressions, depending on the situation and expressed affective states. Further, noteworthy sex differences in mean susceptibility to contagion scores have been reported (e.g., Doherty et al., 1995; Hatfield et al., 2018), with females considered more susceptible to others’ emotions than males, and therefore, researcher should attend to such considerations in the development and use of contagion instruments.

Social psychology researchers (e.g., Lundqvist & Dimberg, 1995) have questioned the hypothesis that susceptibility to emotional contagion is a unidimensional trait, invariant to the types of emotions involved and instead hypothesized susceptibility to occur at a differential emotion level. Derived from an evolutionary perspective, negative emotional contagion (e.g., anger) has been shown to occur more automatically and be less contextually sensitive than positive emotions (e.g., happiness; Kelly et al., 2016). Moreover, experimental investigations have revealed that different emotions evoke distinct facial reactions when exposed to facial expression stimuli that correspond with discrete emotional experiences (e.g., Lundqvist, 1995). Given that discrete primary emotions are activated and regulated through different processes and mechanisms (Izard, 1993), another person’s emotion may be caught and expressed when most aligned with that emotion. Further to conceptualizing susceptibility to the contagion of discrete emotions, we introduce the need to investigate another distinct component: general susceptibility to the contagion of affect. We posit that the general susceptibility is dispositional and assesses the underlying susceptibility to contagion that is invariant to specific felt emotions. We make this suggestion based on Barsade et al. (2018) assertion that emotional contagion consists of the contagion of discrete emotions and generalized mood. As outlined below, it may therefore be prudent to consider a broader view than moods to general susceptibility to others’ affective experiences.

Emotional contagion measurement

Scholars have used multiple methods to assess the presence of emotional contagion (see Barsade et al., 2018). Our focus for review and study is on self-reported measures of susceptibility to emotional contagion. Self-report measurement has received scarce attention, until recently with the development of The Emotional Contagion Scale (ECS; Doherty, 1997). The 15-item ECS measures individual susceptibility to emotional contagion of three basic negative emotions (anger, fear, and sadness) and two basic positive emotions (love and happiness). It has undergone two revisions; from 38 to 18 items (Doherty et al., 1995), and later from 18 to 15 items in the current version (Doherty, 1997). Doherty (1997) also reported the ECS to have a unidimensional structure using principal component analysis, with better model fit than an obtained two-factor solution (positive and negative affect).

Lundqvist (2006) conducted the first rigorous examination of this scale, testing multiple alternative theoretically and empirically derived models. He found support for a five-factor model (primary factors of happiness, love, anger, fear, and sadness) and a hierarchical two-factor model (first-order factors of positive and negative emotions and five second-order emotion facets) but found the unidimensional solution proposed by Doherty (1997) to have a poor fit to the data. These data might be interpreted as evidence of subtle patterns in susceptibility to emotional contagion not distinguishable at a primary analysis level, while we also acknowledge the potential confounding influence of sample differences between studies. Both models were retained suggesting that susceptibility to emotional contagion can be differentiated by positive and negative emotion states as well as discrete emotions, challenging Doherty’s (1997) proposed unidimensional solution. Indeed, multiple researchers have questioned the generalizability of the ECS as a unidimensional construct (e.g., Bhullar & Bains, 2013; Coco et al., 2014; Lundqvist & Kevrekidis, 2008; Wróbel & Lundqvist, 2014). The multidimensionality of the ECS has also been supported in a Japanese sample by Kimura et al. (2007) who argued that the independence of each emotion should be considered when examining susceptibility to emotional contagion. The unidimensionality of the ECS does not allow for investigations of the susceptibility to the contagion of discrete emotions (e.g., anger) separated from other emotions.

In Clarkson et al. (2020) systematic review of contagion of affective phenomena and leadership, contagion researchers were identified as having predominantly employed either the Positive Affect and Negative Affect Scale (PANAS: Watson et al., 1988) or the Job Affect Scale (JAS: Brief et al., 1988) rather than the ECS. Given the availability of a susceptibility to emotional contagion measure, it is surprising that researchers had employed measures like PANAS and JAS which do not specifically measure the contagion of affective phenomena. This may be due to researchers’ alignment to a conceptualization of unconscious primitive emotional contagion which would assume that individuals cannot assess the frequency of contagion events. Yet, if researchers use Barsade et al.’s (2018) recent theoretical framework of emotional contagion as containing both conscious and unconscious processes, a psychometric assessment of susceptibility to emotional contagion would tap the conscious element and therefore be an appropriate assessment method.

In summary, questions remain regarding the conceptualization of susceptibility to emotional contagion as a unidimensional or multidimensional construct and the construct validity of the ECS. We hypothesize that there are theoretical advantages associated with a complementary multidimensional measure of the susceptibility to contagion that could accompany the existing ECS and address measurement shortcomings. Hence, we sought to systematically develop a psychometrically tested versatile instrument that could be used to dynamically measure the susceptibility to contagion of affective phenomena through a multi-study approach. In doing so, we examine whether susceptibility to emotional contagion is best conceptualized as a unidimensional or multidimensional construct. We present four studies describing the development and initial validation of two Contagion of Affective Phenomena Scales (CAPS).

Study 1: item generation and preliminary item analysis

Before starting this measurement development process, we consulted DeVellis (2003) measurement development guidelines and Holmbeck and Devine’s (2009) criteria and checklist for measure development papers to guide the set of studies and our efforts to develop empirical support for the measures. We followed, and subsequently detailed in this manuscript, eight out of the nine steps within these guidelines, omitting “evaluates diagnostic utility, clinical utility, and cost-effectiveness” (p.695) given the measures were not intentionally developed with a specific clinical population or use.

Stage 1: item generation

Following institutional ethical approval (SFEC 2017–113), 48-item pool was generated which broadly measured the overall susceptibility to contagion of affective phenomena concept, consisting of seven discrete emotions (viz. anger, disgust, fear, anxiety, sadness, excitement, and happiness) and general susceptibility. We consulted discrete emotion theory and the Discrete Emotion Questionnaire (DEQ; Harmon-Jones et al., 2016) to choose discrete emotions. DEQ is a psychometrically valid measure, which has been found to be more sensitive than the PANAS at detecting self-reported emotions (Harmon-Jones et al., 2016). The DEQ is not an exhaustive list of emotional state possibilities and therefore we also included ECS items and the emotional contagion subscale items of the Measure of Empathetic Tendency (MET: Mehrabian & Epstein, 1972) interpreted to demonstrate good reliability (i.e., using squared multiple correlations and item total correlations). Some contagion researchers have questioned the validity of the ECS’ love subscale (e.g., Harmon-Jones et al., 2016; Lundqvist, 2006) due to the high cross-load potential of the factor onto happiness, and thus the love subscale of the ECS was omitted. We included items under the category of general susceptibility in response to research pertaining to the processing of emotional stimuli for which stress researchers report individuals to have a general susceptibility (e.g., Brühl et al., 2011).

Stage 2: face validity by expert preliminary item analysis

Stage 2 sought to gain feedback from leading affective “experts” and refine or eliminate items (Kline, 2003). Expert reviewers have consistently been used as a pretesting method of questionnaire evaluation to identify potential measurement errors (Olson, 2010).

Methods

Participants

Fifteen academic experts from United Kingdom (UK) and Australian institutions with a research interest in affective phenomena and/or measurement development were contacted. Eight experts from five institutions responded and completed the item rating form.

Procedure

Experts were emailed study rationale details and instructions for individually rating the overall validity of the 48-item pool on a 5-point Likert-type scale (1 = poor to 5 = excellent). A definition of each emotion was supplied, and experts were asked to examine each item in terms of whether it was “representative” of the emotion. Space was provided for potential item improvement suggestions or whether the measure required more emotions than were included. For example, one expert identified that certain aspects of the Anger definition (feelings of displeasure or hostility) were not captured by the proposed items. Accordingly, another item “I feel stirred when those around me are angry” was added to the item pool. Experts also provided comment on: “Is the item clearly understood?” and “Is it specific – i.e., the item focused enough and not too general or ambiguous?.” Consistent with social psychology measurement development research (e.g., Cronin & Allen, 2017), items were retained if the mean score was greater than 3.5 out of 5.0. Items with a mean score of 3.0–3.5 were scrutinized and considered for removal if in conjunction with a priori rationale and the experts’ comments.

Results and discussion

The standard deviations for all items fell within a moderate variability range of 0.49 to 1.21, except for “I feel excited if others around me appear to be excited” with SD = 1.60. Upon further inspection, the mean score was 3.75, and in the absence of a priori rationale or tautological justification, we decided to retain the item at this stage. On average, all factors scored 3.5 or higher (Table 1). Frequency analysis revealed 44 of 48 items were rated greater than 3.5 criteria and were retained (22 items had a mean score > 4.0; 22 items had a mean score > 3.5). Four items scored between 3.0 and 3.5 and were further scrutinized. An Anxiety item was inspected: “I feel nervous when someone else is nervy near me” (3.29). Whilst the adjective nervy is common in England to describe an individual’s agitative state, in North America it is also used to describe an individual’s brashly presumptuous manner. Item ambiguity coupled with low expert rating was sufficient for removal. Two Sadness items were inspected: “When I am happy it does not take much to bring me down” (3.29) and “I often cry at sad films” (3.43), in addition to one Excitement item: “When watching films, I get excited when I see exciting things” (3.43). To encourage inclusivity and maintain content coverage, these items were retained at this stage, with removal rationale should they perform poorly at the next analysis stage. Next, measure completion instructions and response format were devised. The directions given to respondents were: “This is a scale that measures a variety of feelings and behaviours in various situations. There are no right or wrong answers, so try very hard to be completely honest in your answers. Results are completely confidential.” While we consulted methodological response scale literature (e.g., Hinkin, 1995), these directions were chosen in line with directions in the ECS. Accordingly, we decided upon the response scale: (1) never true, (2) rarely true, (3) sometimes true, (4) often true, and (5) always true.

Table 1. Frequency analysis of factor overall face validity.

Factor (items)			Ratings
	Mean	SD	Lowest	Highest
Anger (6)	4.04	0.61	3.57	4.86
Disgust (3)	3.71	0.14	3.57	3.86
Fear (4)	4.36	0.18	4.14	4.57
Anxiety (6)	4.06	0.46	3.29	4.43
Sadness (7)	3.97	0.62	3.29	4.71
Excitement (5)	3.64	0.18	3.43	3.86
Happiness (7)	4.06	0.33	3.71	4.57
General Susceptibility (8)	4.18	0.29	3.86	4.71

This study took a pragmatic item generation approach using highly performing items in existing measures rather than from scratch to generate an item pool which were then assessed by experts for face and content validity and application to (adult) populations (Jones et al., 2005). Study 1 results suggested that a final pool of 47 items was representative of the construct and were further analyzed in Study 2 (Table 2).

Table 2. Study 1 items retained after content validity assessment.

Items
(1) I feel stirred when those around me are angry. (A1)
(2) It irritates me to be around angry people. (A2)
(3) I notice myself getting tense when I’m around people who are stressed out. (A3)
(4) I experience feelings of frustration if others around me seem to be frustrated. (A4)
(5) When someone around me is angry, I also feel angry. (A5)
(6) I get provoked by someone else’s anger. (A6)
(7) I am influenced by others’ disgust. (D1)
(8) I feel sickened when other people are repulsed. (D2)
(9) When I see others treated poorly, I share their disgust. (D3)
(10) I feel swayed by others fear. (F1)
(11) I become nervous if others around me seem to be nervous. (F2)
(12) When I see people around me who are scared, I become scared. (F3)
(13) When I see someone who is scared, I share their fear. (F4)
(14) I am disturbed by others’ anxiety. (AX1)
(15) I experience panic when I see others who are scared. (AX2)
(16) It fills me with dread when other people are anxious. (AX3)
(17) My anxiety escalates when those around me are worried. (AX4)
(18) I feel tense when those around me are worrying about something. (AX5)
(19) I am able to remain calm even when those around me worry. (AX6R)
(20) I am moved when I see sadness in those I interact with. (S1)
(21) I cannot continue to feel okay if people around me are depressed. (S2)
(22) I become sad when other people are sad. (S3)
(23) When I am happy it does not take much to bring me down. (S4)
(24) When I see other people down it brings me down too. (S5)
(25) When watching films, I get sad when I see sad things. (S6)
(26) I often cry at sad films. (S7)
(27) I get infected by excitement in others. (EX1)
(28) I often find that I can remain cool in spite of excitement around me. (EX2R)
(29) I feel excited if others around me appear to be excited. (EX3)
(30) I get excited when someone receives exciting news. (EX4)
(31) When watching films, I get excited when I see exciting things. (EX4)
(32) When people around me are buzzing with excitement, I get excited too. (EX5)
(33) Others happiness makes an impression on me. (H1)
(34) Being with a happy person picks me up when I’m feeling down. (H2)
(35) When someone smiles warmly at me, I smile back and feel warm inside. (H3)
(36) Being around happy people fills my mind with happy thoughts. (H4)
(37) I experience feelings of happiness if others around me seem to be happy. (H5)
(38) It makes me delighted when other people are happy. (H6)
(39) When I am in a bad mood, a smile or joke can really cheer me up. (H7)
(40) My emotions are connected to those of the group. (G1)
(41) The people around me have a great influence on my moods. (G2)
(42) Other people’s emotions impact my own emotions. (G3)
(43) The way others feel has no relation to my own feelings. (G4R)
(44) I get influenced by other people’s emotions. (G5)
(45) When I am in groups, I am infected by the mood of the group. (G6)
(46) When watching films, I am moved by emotional scenes. (G7)
(47) The emotions of those I interact with influence me. (G8)

Study 2: Exploring and confirming construct validity and factorial structure

Study 2 explored the psychometric properties of the refined version of CAPS by examining the factorial validity and scale reliability using structural equation modeling (SEM) techniques in two stages, described as Study 2a and Study 2b. The aim of the Study 2a was to eliminate cross-loading items on unintentional factors and initially test the factor structure of the scale. Study 2b replicated the model testing approach with a second independent sample to assess the internal reliability and bolster validity (DeVellis, 2003).

Study 2a: exploring validity and structure

Methods

Participants and procedure

Psychometric measurement researchers have long acknowledged difficulties in determining EFA sample sizes (Pearson & Mundform, 2010). To determine a suitable sample size in all ESEM studies, we followed Comrey and Lee’s (1992) recommendations for sample size adequacy (i.e., 50–very poor, 100–poor, 200–fair, 300–good, 500–very good, and >1,000– excellent). The participants were 455 adults (age range 18–93 years, M = 26.29, SD = 17.93). Accordingly, the Study 2a sample size could be determined as between “good” and “very good.”

Study participation inclusion criteria were adults (aged over 18 years old) who had demonstrable English language proficiency (either by native language confirmation as English or by previous completion of a nationally recognized English proficiency test). This sample included 236 (56%) male and 219 female (48%). Participants identified as 80% White, 7% Black, 4% Asian, and 9% declined to disclose their ethnicity.

Participants were recruited through two strategies: (a) the lead author administered the scale to University students at the end of timetabled lessons (n = 278, f = 67, m = 211), and (b) by responding to a publicized online version of the scale on social media (n = 177, f = 152, m = 25). Participants participated voluntarily, without financial incentive, and completed every item (true for all studies). Any questionnaire packs with missing data were excluded and not reported here.

Data analysis

Data were treated as continuous and analyzed using a combination of CFA and exploratory structural equation modeling (ESEM) through Mplus 8.1 software (Muthén & Muthén, 2018). In this study and following assumptions being met, the analysis involved testing each proposed factor separately to eliminate poor loading items, as well as examining global goodness-of-fit indicators, parameter estimates and residuals. Items that loaded onto their intended factor with a rating >0.32 were retained and conversely items with a cross-loading of >0.32 were excluded (Tabachnick & Fidell, 2013). The models were re-specified if required. Then, inter-factor correlations, residuals, and modification indices were examined, eliminating any ambiguous items, and re-specifying the factors.

Finally, the full model’s fit to the data was tested using ESEM, a relatively new methodology that integrates the advantages of EFA and CFA within a general structural equation model, providing a useful framework for initial validity studies (Asparouhov & Muthén, 2009). While CFA provides a rigorous framework to test factorial invariance and allows a priori content knowledge to guide model specification, it is only appropriate when sufficient a priori measurement theory exists (Myers et al., 2011). Theoretical absence results in a data-driven process, led by modification indices to improve model fit, which can be vulnerable to chance sample characteristics and raises questions as to whether the measurement model is generalizable to other samples. Emotional contagion theory and Emotions-as-Social-Information Model guided the present study; yet, both theories relate specifically to emotions, rather than mood or affect, and the contagion of other affective phenomena (e.g., mood contagion, affective transfer) has no direct a priori guidance. ESEM is advocated when predominantly guided by content knowledge rather than by a priori reasoning (Myers et al., 2011) and was consequently chosen to develop a measurement model for CAPS. Model fit was examined using several goodness-of-fit indices; these were the Comparative Fit Index (CFI), Tucker-Lewis Index (TLI), Standardised Root Mean Square Residual (SRMR), and Root Means Square Error of Approximation (RMSEA). Hu and Bentler’s (1999) recommendations were followed for evaluating model fit (e.g., CFI ≥ .90, TLI ≥ .90, SRMSR < .08, RMSEA ≤ .08). Cronbach’s alphas were also computed for each subscale. Additionally, incremental fit indices indicated the goodness of fit of alternative models. Chen’s (2007) criteria were applied to reflect a lack of measurement invariance, including an increase in RMSEA or SRMR > .015 or a decrease in CFI or TLI of > .01 (Cheung & Rensvold, 2002). Alternative models with differences lower than the stated values were considered irrelevant and instead the most parsimonious model selected. We then conducted configural measurement invariance testing, where the sample was split by sex and model fit was compared between the sex groups (Lee, 2018).

Results: single factor models

Fit statistics and factor loadings for the eight proposed factors were examined. This process revealed that the seven of the eight proposed factors (viz. anger, fear, anxiety, sadness, excitement, happiness, and general susceptibility) demonstrated acceptable factor loadings and fit statistics (Table 3). Single-factor analysis for the three-item disgust factor revealed a poor fit to the data and low Cronbach's alpha score (.51). Subsequently, the disgust factor was removed.

Table 3. Summary of fit statistics for alternative models of CAPS-E.

Model	χ^2	df	CFI	TLI	RMSEA	SRMR
Study 2 (n=455)
First-order one-factor	2153.36	189	.50	.45	.15	.14
First-order six-factor	391.02	174	.95	.93	.05	.05
Higher order two-factor (positive and negative emotions)	973.10	188	.80	.78	.10	.09
Study 2b (n=567)
First-order one-factor	1246.01	188	.57	.52	.13	.12
First-order six-factor	565.73	174	.94	.93	.06	.05
Higher order two-factor (positive and negative emotions)	621.50	188	.82	.80	.09	.08
Study 3 (n = 315)
First-order six-factor	366.75	174	.92	.90	.06	.06
Study 4 (n = 200)
First-order six-factor	453.72	174	.90	.88	.09	.07

During this stage, several problem items were identified, removed and their models successfully re-specified. This was largely data-driven, although attempts were made to provide a conceptual rationale to supplement item removal where possible. For the anger (A) model, A4 and A5 (Table 2) were identified with significantly lower factor loadings (0.48 and 0.44 respectively) as problem items and removed for greater model fit. We hypothesized that A4 (“I experience feelings of frustration if others around me seem to be frustrated”) was conceptually weak as frustration is a negative emotion that is typically goal-directed and may not always lead to anger (Zurawski & Houston, 1983). The removal of A5 was data-driven. The subsequent fit of the four-item anger model was greatly improved with a lower χ²(2) = 1.370, lower values for RMSEA = .00 and SRMR = .001, and a higher value for CFI = 1.00. Likewise, the anxiety (AX) model without A×2 and AX6R (Table 2) showed an improved, re-specified model fit. While the A×2 factor loading was not extremely low (0.58), this item was removed in conjunction with conceptual reasons (feelings of panic are argued to be primarily related to panic and distinct from anxiety, Gray, 1991) and to greatly improve single model fit without its presence. For AX6R, the utility of reverse scored items for measuring anxiety has been questioned (e.g., Rodebaugh et al., 2007), as the reverse of anxiety is not always the opposite. This conceptual rationale was theorized to explain the poor loading (0.22), and the item was removed. Problematic items within the Sadness factor (S2, S4R and S7), the Excitement factor (EX1 and EX2), the Happiness factor (H1, H6 and H7), and the General Susceptibility factor (G1, G4R, G6 and G7) were identified by joint examination of fit statistics and factor loadings, and were removed with improved single-factor model fits. This resulted in six discrete emotion factors; anger (four items), fear (three items), anxiety (five items), sadness (four items), excitement (four items), and happiness (four items), and a factor measuring general susceptibility to contagion of affective phenomena (five items) which proceeded to full model testing.

At this stage, we determined the General factor as conceptually similar but distinct from the susceptibility to contagion of discrete emotions. This was supported by ESEM testing of a two-factor model (general susceptibility and discrete emotions), which revealed a strong interfactor correlation (.74) with a poor fit to the data (χ²(298) = 2429.50, p < .001, CFI = .57, TLI = .53, SRMR = .11, and RMSEA = .13). We theorized that such differences could be interlinked with the discussion surrounding susceptibility to emotional contagion as a unidimensional (as Doherty and colleagues claim) or as a multidimensional construct (as Lundqvist and colleagues claim). The separation of the measure enables the general susceptibility to be viewed as unidimensional and the susceptibility to discrete emotions to be viewed as multidimensional. Thus, further testing was divided into two scales: the multidimensional Contagion of Affective Phenomena – Emotion (CAPS-E) and the unidimensional Contagion of Affective Phenomena – General (CAPS-G).

Results: full model testing – CAPS-E

The retained 24 items of CAPS-E were univariately normally distributed (skewness ranged between −.56 and .51; kurtosis ranged between −.80 and .48), and, however, the data significantly deviated from multivariate normal distribution (Shapiro–Wilk=.995, p = .115) and meant that a maximum likelihood (ML) method was employed as it is robust to non-normality. A satisfactory model fit to the data was achieved with the hypothesized 24-item first-order six-factor model χ²(215) = 475.79, p < .001, with acceptable goodness-of-fit indices (CFI=.94, TLI = .93, SRMR = .05, RMSEA = .05). The significant p value was overlooked in favor of the χ²/df value where values <5.0 are deemed acceptable – a common scholarly benchmark (Myers et al., 2011). The item factor loadings ranged from .40 to .88, while still above traditional .40 cut off, AX5R (.40) from Anxiety, A2 (.53) from Anger, and S3 (.57) from Sadness subscales were removed due to factor loadings distinctly lower than other items in their factor. The re-specified 21-item six-factor model was found to be reliable. Cronbach’s alphas were: anger (.62), fear (.85), anxiety (.82), sadness (.64), excitement (.80), and happiness (.87). There were high correlations between factors with consistent emotional valence (e.g., excitement and happiness = .68) and low correlations between factors with inconsistent emotional valence (e.g., happiness and fear = .21). This model demonstrated acceptable model fit to the data χ²(174) = 391.02, p < .001, and acceptable goodness-of-fit indices (Table 3). CAPS-E consisted of 16 original items, one OECS item (F1) and four ECS items (A2, H1, H2, and H3). CAPS-G comprised four original items and one OECS item (G2).

Plausible alternative models (a one-factor and a higher order two-factor model) were also tested to demonstrate the additional validity of the hypothesized ESEM model. The first model hypothesized that the 21 items of CAPS-E measured a single dimension. The higher order two-factor model categorized susceptibility to contagion into the positive emotion (excitement and happiness) and negative emotion (anger, fear, anxiety and sadness) factors. ESEM results indicated that the CFI and TLI of the first-order six-factor model were above the .93 and .92 respective criteria and the RMSEA and SRMR were below .07 and .05 criteria, indicating that this model had the best fit to the data (Table 3). The fit of the alternative unidimensional one-factor model and higher order two-factor model was not substantially better than the first-order six-factor model (Table 3). Thus, the six-factor solution offered the most parsimonious model fit and proceeded for further testing.

Results: full model testing – CAPS-G

Preliminary analyses demonstrated that normality was violated, and the ML estimation method was employed. Item loadings ranged from .60 to .83, which suggested that the five items tapped the same construct. Furthermore, the 5-item model demonstrated a satisfactory fit to the data χ²(5) = 7.92, p = .16, and acceptable goodness-of-fit indices (Table 4). Cronbach's alpha was .82.

Table 4. Model fit statistics for unidimensional CAPS-G.

Model	χ^2	df	CFI	TLI	RMSEA	SRMR
Study 2a (n=455)	7.92	5	.99	.99	.04	.02
Study 2b (n=567)	52.54	5	.97	.94	.13	.03
Study 3 (n = 315)	7.92	5	1.00	.99	.04	.99
Study 4 (n = 200)	50.71	5	.92	.83	.21	.05

Results: group difference testing

Overall, no substantial sex or recruitment group differences existed between the CAPS-E and CAPS-G model fits. With regard to sex, the six-factor solution model of CAPS-E demonstrated comparably good fit for females (χ²(174) = 307.70, CFI = .93, TLI = .92, SRMR = .06, RMSEA = .06) and males (χ²(174) = 275.84, CFI = .93, TLI = .93, SRMR = .06, RMSEA = .05). Similar findings across most fit indices were evident. Some fluctuation was noted in the χ² (f = 9.34, m = 5.44, df = 5), but both group’s statistics had acceptable χ²/df value below 5.0. The CAPS-G model fit the female (χ²(5) = 307.70, CFI = .99, TLI = .98, SRMR = .06, RMSEA = .02) and male (χ²(5) = 5.44, CFI = .99, TLI = .99, SRMR = .02, RMSEA = .02) groups well with comparable satisfactory fit indices. There was a small difference in χ² values, but both groups χ²/df values were below the common benchmark of 5.0 (Myers et al., 2011).

With regard to the recruitment groups, no meaningful CAPS-E model fit difference was interpreted between the recruitment groups: social media (χ²(174) = 300.45, CFI = .93, TLI = .91, SRMR = .06, RMSEA = .07), and University students (χ²(174) = 274.02, CFI = .95, TLI = .95, SRMR = .05, RMSEA = .05). Concerning CAPS-G, the RMSEA value in the University group was noticeably lower (.00) than the social media group (.08), but this fit difference was interpreted with caution due to the unequal sample sizes and other fit indices being similar, in addition to the RMSEA values still falling within acceptable parameters (≤.08).

Study 2b: confirming reliability and validity

Methods

Participants and procedure

A second independent sample comprised 567 adults with demonstratable English proficiency, with the sample size deemed as between “very good” and “excellent” (Comrey & Lee, 1992). This sample included 226 females (40%) and 341 males (60%) whose age ranged from 18 to 63 years (M=25.70, SD=8.85) and whom were members of local sports teams, recruited by the lead author to complete CAPS after scheduled practice. Participants identified as 61% White, 28% Black, 9% Asian, and 2% declined to disclose their ethnicity.

Measures

CAPS-E. Susceptibility to the contagion of affective phenomena measured using the 21-item CAPS-E that consisted of six discrete emotions (Cronbach's alpha score): anger (3 items, .75), fear (3 items, .86), anxiety (4 items, .87), sadness (3 items, .80), excitement (4 items, .86), and happiness (4 items, .82).

CAPS-G. General susceptibility to the contagion of affect was measured using the 5-item CAPS-G. Cronbach's alpha was .89.

Results: full model testing – CAPS-E

The ML method of estimation (normality violated) was employed. Table 3 shows the goodness-of-fit indices and chi-square statistic calculated for the different models. The first-order six-factor model was again hypothesized to best fit the data and scrutinized against the same alternative models. Results show that the first-order six-factor model (Figure 1) had the best fit to the data (χ²(215) = 565.73, CFI = .94, TLI = .93, SRMR = .05, RMSEA = .06) when compared to the two-factor solution and unidimensional model (Table 3).

Figure 1. The six-factor measurement model confirmed with ESEM in study 3.

Results: full model testing – CAPS-G

The ML method of estimation (normality violated) was employed. ESEM statistics were interpreted to demonstrate some acceptable fit indices (CFI, TLI, SRMR), however, the RMSEA (.13) was above the criterion of < .08 and the χ²/df value (10.51) was above the <5.0 criterion (Table 4). The model fit may have been explained by a lower G1 item loading (.57). However, given the factorial testing conducted in Study 2b that demonstrated a satisfactory fit to the data that was also invariant across groups and the item within acceptable the range, we decided upon caution and did not remove items and re-specify the model.

Discussion

The collective results of Study 2 indicated the existence of two separate but conceptually similar measures: the multidimensional six-factor measure of susceptibility to contagion of discrete emotions, CAPS-E, and the unidimensional measure of general susceptibility to the contagion of affect, CAPS-G. The hypothesized CAPS-E first-order six-factor model, when compared to one-factor or two-factor solutions, showed the best fit to the data. Negligible differences between fits of alternative models in Study 2a were scrutinized in Study 2b to confirm the identified measurement properties of CAPS-E and provide further empirical evidence of reliability and validity. Model testing of CAPS-G indicated that it is a valid single-factor 5-item measure of general susceptibility to the contagion of affect.

Study 3: establishing test-retest reliability, concurrent validity, and discriminant validity

The aims of Study 3 were to: (a) cross-validate the first-order six-factor model supported in Study 2 with a third independent sample; (b) assess test–retest reliability by administering CAPS-E and CAPS-G at two time points (T1 and T2) 6 weeks apart; (c) assess the concurrent validity of CAPS-E and CAPS-G by testing them with established susceptibility to contagion measures that are theorized to correlate and (d) assess the discriminant validity of CAPS-E and CAPS-G by testing them with a variable (i.e., empathy) theorized to be distinct from susceptibility to contagion.

Empathy was selected as a discriminant variable. While many definitions of empathy exist (Hatfield et al., 2011), contagion researchers assert emotional contagion as conceptually distinct from empathy. Examining studies by Decety and Jackson (2004) and Singer and Lamm (2009), it is evident that when reacting to an emotional display, empathy differentiates your own with others’ feelings (top-down process), whereas emotional contagion lacks any differentiation (bottom-up process). Hatfield et al. (2011) described empathy as comprising three components: an ability to share another person’s feelings, a cognitive ability to understand what another person feels, and an intention to respond compassionately to another’s distress. In line with this description, it is probable that there would exist a crossover between empathy and the first component (sharing another’s’ feelings viz. emotional contagion) yet distinct due to the absence of second and third.

Methods

Participants and procedure

The third independent sample comprised 315 adults with demonstratable English proficiency, with sample size deemed as between “good” and “very good” (Comrey & Lee, 1992). This sample included 110 females (35%) and 205 males (65%) whose age ranged from 18 to 34 years (M=19.48, SD=2.31). Participants identified as 74% White, 11% Black, 4% Asian, and 11% declined to disclose their ethnicity.

The first author recruited and administered the scale to university students at the end of timetabled lessons in two semesters, at a different institution from Study 2. Participants participated voluntarily, without financial incentive. Significant recruitment and retention challenges of university students as participants have been noted (Khatamian Far, 2018) and were experienced in this study. All 315 participants completed CAPS-E, CAPS-G and two criterion measures of susceptibility to contagion (i.e., ECS and MET) to assess concurrent validity. A subsample (n = 116) completed a measure of empathy to assess the discriminant validity of CAPS-E and CAPS-G. Due to participant drop-out (classes across semesters and T1 multi-instrument questionnaire pack), test–retest reliability for CAPS-E and CAPS-G was conducted with 27 participants at T2. The sample size was, however, above the minimum recommendations necessary for reliability testing (Bujang & Baharum, 2017). We recommend researchers spread questionnaire items across time-points to distribute load to increase retention.

Prior to analysis, five items from CAPS-E that were shared with the ECS (e.g., A3, H1, H2, and H3), and one item from CAPS-G that was shared with the MET empathetic concern subscale (i.e., G2) were removed from the criterion susceptibility to contagion measures, due to their unidimensional nature compared to CAPS-E’s multidimensionality. Reliability was assessed using Cronbach’s alpha. These concurrent (MET) and discriminant (IRI) measures were chosen in keeping with the procedure followed by Doherty (1997) in his ECS development.

Criterion measures

The Emotional Contagion Scale (ECS). The ECS (Doherty, 1997) is a 15-item unidimensional scale that measures individual differences in the susceptibility to emotional contagion across five basic emotions (happiness, love, fear, anger, and sadness). Responses (e.g., “If someone I’m talking with begins to cry, I get teary-eyed”) are measured using a 5-point Likert-type scale where 1=never true and 5=always true. Cronbach’s alpha was .83.

Measure of Empathetic Tendency (MET-EC). The 7-item emotional contagion subscale of MET (Mehrabian & Epstein, 1972) was administered to assess the concurrent validity of CAPS-E and CAPS-G. Responses (e.g., “I often find I am able to remain cool in spite of excitement around me”) are measured using a 5-point Likert-type scale where 1=strongly disagree and 5=strongly agree. Cronbach’s alpha was .66.

The Interpersonal Reactivity Index (IRI). The 28-item IRI (Davis, 1983) is multidimensional scale that measures individual differences in empathy across four dimensions (perspective taking, fantasy, empathetic concern, and personal distress). The IRI was administered to assess the discriminant validity of CAPS-E and CAPS-G to empathy. Responses (e.g., “I often have tender, concerned feelings for people less fortunate than me”) are measured using a 6-point Likert-type scale ranging from 0 = does not describe me well and 5 = describes me well.

We expected CAPS to correlate with the empathetic concern subscale but overall be conceptually distinct from the overall construct, given that some researchers (e.g., Hatfield et al., 2011) consider contagion as part of an index of empathy. To assess the robustness of the correlation coefficient, bootstrapping was performed with 1,000 resamples. Bias-corrected and accelerated confidence intervals were calculated to account for potential bias in the bootstrap estimates. Given a smaller study subsample completed the IRI, we employed a more stringent assessment to establish discriminant validity, drawing on recent insights from Rönkkö and Cho (2022).

Results

Table 5 shows the means, standard deviations, and correlations of the CAPS and concurrent measures of emotional contagion. ESEM (using the ML estimation method for non-normally distributed data) was performed on CAPS-E and confirmed it had acceptable fit to the data (Table 3). CAPS-G also had a satisfactory fit to the data, and we retained its item structure (Table 4).

Table 5. Means, SD and inter-correlations between CAPS and concurrent measures of contagion.

		1	2	3	4	5	6	7	8	9	10	Mean	SD
1	CAPS-E	(.90)										2.89	0.46
2	Anger	.63**	(.74)									2.72	0.67
3	Fear	.71**	.39**	(.86)								2.37	0.69
4	Anxiety	.76**	.40**	.69**	(.87)							2.34	0.72
5	Sadness	.71**	.37**	.38**	.46**	(.81)						3.00	0.73
6	Excitement	.64**	.23**	.20**	.26**	.32**	(.86)					3.47	0.65
7	Happiness	.62**	.18**	.18**	.22**	.33**	.67**	(.82)				3.47	0.64
8	CAPS-G	.60**	.60**	.43**	.42**	.49**	.30**	.35**	(.89)			2.98	0.67
9	ECS	.68**	.41**	.41**	.42**	.54**	.50**	.49**	.49**	(.83)		2.93	0.56
10	MET – EC	.12*	.14*	.08	.09	.13	.01	.03	.05	.12*	(.56)	3.12	0.33

*p < .05, ** p < .001, reliability scores in brackets.

Test-retest reliability

The T1 mean CAPS-E score was 2.97 (SD = 0.36), and the T2 mean score 6 weeks later was 2.97 (SD = 0.35). Similar results were found in CAPS-G (T1 M = 2.88, SD = 0.57; T2 M = 2.92, SD = 0.54). A paired sample t-test indicated that the mean scores were significantly different for CAPS-G (t(26) = 2.26, p=< .05), but were not significantly different for CAPS-E (t(26) = 1.56, p = .13). On further inspection of CAPS-E scores, the lower 95% confidence interval was only just below zero (−.01 and .08). Hence, we further inspected the reliability between T1 and T2, both CAPS-E (r = .96, p < .001) and CAPS-G (r = .98, p < .001) significantly correlated.

Concurrent validity

CAPS-E significantly correlated with CAPS-G (r = .60), the ECS (r = .68) and the MET-EC (r = .12). CAPS-G did not significantly correlate with the MET-EC (r = .05).

Discriminant validity

Significant positive, weak correlations were observed between empathy and susceptibility to contagion of anxiety (r = .25, p < .01, LLCI = .07, ULCI = .40) and sadness (r = .25, p < .01, LLCI = .09, ULCI = .40). All other correlations were nonsignificant, including empathy and susceptibility to contagion of anger (r = .08, p = .39, LLCI = −1.0, ULCI = .26), fear (r = .25, p < .01, LLCI = .07, ULCI = .40), excitement (r = .11, p = .25, LLCI=−.07, ULCI = .25), and happiness (r = .08, p = .39, LLCI = −1.0, ULCI = .26). General susceptibility to contagion of affect (r=-.03, p = .73, LLCI=−.21, ULCI = .16) was negatively and nonsignificantly correlated with empathy.

Discussion

In Study 3, the validity of the first-order six-factor measurement model for CAPS-E and unidimensional CAPS-G was reaffirmed. The scale statistics were interpreted to demonstrate test–retest reliability, although caution is required given the small sample size and the need for further evaluation of the scales. The study findings indicated concurrent validity of CAPS-E with all other susceptibility to contagion measures. CAPS-G did not significantly correlate with MET-EC, which could be interpreted as indicative of our original assertation that the CAPS-G is conceptually similar but distinct from the susceptibility to contagion of discrete emotions. Indeed, some psychotherapy research indicates emotional states may interact differently with empathy than affect (Arizmendi, 2011). Further examination beyond the present research is needed to determine if the observed pattern in the measurement relationship is reflective of a broader variable-level relationship. In doing so, consideration may be given to the empathy measure used as the reliability score of the MET-EC was low. Nonetheless, interpretations of discriminant validity statistics are suggestive CAPS-E and CAPS-G as distinct from empathy measures. Our findings suggest subtle differences between empathy and individual susceptibility to the contagion of anxiety and sadness than compared with the contagion of other emotions that go beyond the scope of this study but could prove fruitful lines of further scholarly enquiry. Such findings provide initial support for the utility of CAPS-E to assess susceptibility to the contagion of discrete emotions and CAPS-G to assess general susceptibility to the contagion of affect.

Study 4: test of predictive validity

The aim of Study 4 was to test the predictive validity of CAPS-E and CAPS-G. While CAPS could have application in multiple social psychology domains, this investigation focused within our sport psychology expertise and considering an emerging body of work exploring the emotional contagion process in sport (e.g., Cotterill et al., 2020; Van Kleef et al., 2019). We explore the potential role that individual susceptibility to emotional contagion plays with sport performance. While no study has directly examined the link between emotional contagion and performance in sport, there is a wealth of research on the positive association between emotional intelligence and performance (see Laborde et al., 2016). While conceptually distinct from emotional contagion, individuals who can infect others’ emotions and are themselves highly susceptible to the emotions of others could be perceived as emotionally intelligent people (Goleman, 1995). There has also been some indication in other domains that individual susceptibility to emotional contagion can influence performance. In a study of sales organizations’ performance, Verbeke (1997) found a salesperson’s ability to infect others and be sensitive to the emotions of others was an asset to performance. Susceptibility was, however, also a found to be associated with a higher risk of burnout from role stress. Hence, given an existing relationship between emotional intelligence and sporting performance, as well as indications of between emotional contagion and occupational performance, we sought to examine whether susceptibility would predict variance in self-rated satisfaction with athletic performance after controlling for emotional intelligence. In the context of swimming, where athletes taper performance differently for various meets, subjective satisfaction was appropriate for assessing performance.

Method

Participants

Participants were 200 swimmers (M_age = 29.20 years, SD = 10.67), who had on average 12.95 years of competitive swimming experience (SD = 8.70). Sample included 51% females (n = 102) and 49% males (n = 98). Soper’s (2019) a priori minimum sample size calculator for hierarchical multiple regression was employed, entering: anticipated Cohen’s d effect size (0.15), desired statistical power (0.80), the number of predictors (2), and probability level (0.05). The minimum sample size required to be sufficiently powered was 55.

Procedure

English regional swimming competitions were attended, and swimmers competing at these meets in individual competitions were invited to participate in the study prior to competition.

Measures

Susceptibility to emotional contagion. Susceptibility to emotional contagion was measured using the 21-item CAPS-E and 5-item CAPS-G. ESEM analysis (using the ML method of estimation for non-normally distributed data) confirmed CAPS-E’s six-factor solution had acceptable fit to the data (Table 3). CAPS-G fit some model indices but not all (Table 4). Cronbach’s alphas were: CAPS-E=.92 (anger=.83, fear = .86, anxiety = .89, sadness = .80, excitement = .85, and happiness = .88) and CAPS-G=.88.

Emotional intelligence. Emotional intelligence was measured using the 16-item Wong and Law Emotional Intelligence Scale (WLEIS: Wong & Law, 2002). Responses (to e.g., “I am a good observer of others’ emotions”) are measured using a 7-point Likert-type scale where 1=strongly disagree and 7=strongly agree. Cronbach's alpha was .97.

Self-rated subjective performance. A 3-item measure was developed for the current study that assessed subjective athlete satisfaction with individual performance (e.g., “I am satisfied with the degree to which I have reached my performance goals during the season”). As comparing athletes’ performances across sports can be complex, we drew on pragmaticism and discipline traditions (e.g., Levy et al., 2011) to develop a self-report subjective performance measure for the study. The use of subjective indices has been acknowledged for sport performance (Biddle et al., 2001). Cronbach's alpha was .92.

Data analysis

Correlations were computed to better understand the relationship between susceptibility to emotional contagion (and its subscales) and emotional intelligence in swimmers. Then, a two-step hierarchical regression analysis was performed after assumptions were met (1) to assess the role of susceptibility to contagion in predicting satisfaction with performance and (2) to determine whether emotional contagion predicted variance in the satisfaction with performance relationship additional to emotional intelligence.

Results

Table 6 displays the study variable correlations. A significant and weak, negative correlation was observed between CAPS-E and emotional intelligence (r=-.15, p < .05). CAPS-G was similarly weak and negatively correlated with emotional intelligence but did not reach significance (r=-.13, p=.06). Emotional intelligence was significantly negatively correlated with fear (r=-.41, p < .001) and anxiety (r=-.46, p < .001) contagion and significantly positively correlated with excitement (r=.15, p < .05) and happiness (r=.20, p < .001) contagion. Emotional intelligence was not correlated with susceptibility to anger (r=-.13, p=.07) or sadness contagion (r=-.01, p=.91). Correlation analysis also revealed sex differences with significant positive relationships between emotional intelligence and excitement (r=.24, p < .05) and happiness (r=.25, p < .05) contagion for females but not for males (excitement r=.08, p=.43; happiness r=.15, p=.13). All other correlations were consistent in direction and significance between sexes.

Table 6. Summary of variables predicting performance (study 4).

		1	2	3	4	5	6	7	8	9	10	Mean	SD
1	CAPS-E											3.10	0.67
2	Anger	.75**										3.03	0.94
3	Fear	.72**	.61*									2.57	0.94
4	Anxiety	.74**	.50*	.80**								2.56	1.00
5	Sadness	.79**	.50*	.44**	.47**							3.19	0.87
6	Excitement	.70**	.42**	.15*	.20**	.60**						3.49	0.86
7	Happiness	.70**	.36**	.19**	.23**	.51**	.69**					3.62	0.91
8	CAPS-G	.74**	.50**	.52**	.52**	.64**	.53**	.55**				3.10	0.78
9	Emotional intelligence	−.15*	−.13	−.41**	−.46**	−.01	.15*	.20**	−.13			4.70	1.43
10	Performance	−.33**	−.29**	−.41**	−.40**	−.23**	−.06	−.09	−.30**	.39**		3.55	1.07

*p < .05, ** p < .001.

The first regression analysis revealed that the general susceptibility to emotional contagion (as measured by CAPS-G) significantly and negatively predicted subjective performance satisfaction (R² =.09; β=-.30, p = < .01). Susceptibility to the contagion of discrete emotions (as measured by CAPS-E) also significantly and negatively predicted subjective performance satisfaction (R² =.11; β=-.33, p = < .01). Furthermore, our second hierarchical regression analyses revealed that CAPS-G accounted for a significant proportion of variance in subjective performance satisfaction (∆R² =.21; β=-.25, p < .01) over and above accounted for by emotional intelligence (R²=.15; β=.39, p < .01). CAPS-E also accounted for a significant proportion of variance in subjective performance satisfaction (∆R² =.22; β=-.28, p < .01) over and above accounted for by emotional intelligence (R²=.15; β=.39, p < .01). While variations in the beta values were noted, susceptibility to all emotions, that is anger (∆R² =.21), fear (∆R² =.23), anxiety (∆R² =.21), sadness (∆R² =.20), excitement (∆R² =.17), and happiness (∆R² =.18) contagion, accounted for a significant proportion of variance in subjective performance satisfaction over and above accounted for by emotional intelligence. Some sex differences were observed across the different emotions. Susceptibility to anxiety contagion (∆R² =.25; β=-.36, p < .01), excitement contagion (∆R² =.18; β=-.19, p < .05), and happiness contagion (∆R² =.20; β=-.25, p < .01) accounted for a significant proportion of variance in performance satisfaction (over and above emotional intelligence) for male but not for female swimmers (anxiety ∆R² =.18, β=-.19, p=.08; excitement ∆R² =.16, β=-.06, p=.51; happiness ∆R² =.16, β=-.11, p=.26).

Discussion

Study 4 statistics were interpreted to demonstrate the initial predictive validity of CAPS-E and CAPS-G in a sport context. Both measures provided evidence of adequate psychometric and predictive tests, and thus, CAPS allows researchers to examine susceptibility to emotional contagion from both a theoretical and applied perspective, while also advancing scholarly understanding of the conscious elements of this construct. For example, the current measures will enable further exploration of the role that individual susceptibility to emotional contagion potentially plays in group interactions and successful group performance.

In this study, susceptibility to emotional contagion negatively predicted performance satisfaction, suggesting that swimmers were less satisfied with their performance if they were vulnerable to the emotions of others. Our subjective performance satisfaction findings here align with previous research by Verbeke (1997) who concluded that susceptibility to emotional contagion in business can either be an asset or liability. In our study, this might be a swimmer’s ability to infect others with their (discrete) emotions which they perceive to be facilitative for their performance (e.g., anger) or a swimmer’s tendency to catch the emotional displays of their competitors as a liability (e.g., anxiety contagion) (see Robazza & Bortoli, 2007). Given CAPS-E was overall negatively correlated with performance satisfaction, but this did not hold true across all discrete emotions, there is clearly much nuance in this domain that might also be situation-specific (i.e., does the relationship between emotional contagion and performance satisfaction change if you are/are not tapering your performance), which is outside the scope of this study, but we encourage researchers to expand academic discussion in this area.

Our findings can also be interpreted to suggest that different emotions can be differentially contagious dependent on sex, as susceptibility to anxiety, excitement and happiness contagion predicted performance satisfaction over and above emotional intelligence for male swimmers only. This finding demonstrates that the subscales of CAPS-E do function differently (i.e., predicting different outcomes) when sex is considered.

A nuanced relationship between emotional intelligence and susceptibility to emotional contagion was observed in Study 4. The weak but significant correlations imply that the constructs could be linked but unsubstantial. CAPS-E factors performed differently, supporting our underlying conceptual framework of different emotions being differentially contagious. We also observed sex differences where the susceptibility to excitement and happiness contagion was significant and positively related to emotional intelligence in female (but not male) swimmers. This finding extends sex differences in susceptibility to emotional contagion observed by Cotterill et al. (2020), who found women athletes were more susceptible to the emotional influence of their athlete leaders than men.

To extend the present work, researchers might seek to explore susceptibility to emotional contagion in general, including the potential impact of emotional valence (e.g., positive and negative). We acknowledge that the performance measure used in this study was self-report in form and subjective in nature and relates to satisfaction with performance and rather than objective performance, and while we believe this measure to have offered valuable insight which may go beyond the nuanced and uncontrollable outcome performance measures, we do encourage research to also consider incorporating objective performance measures in the future research. Another future research consideration emanates from the complexity of the correlation results observed between emotional intelligence and emotional contagion. We would encourage researchers in performance contexts to undertake further examination of emotional intelligence and emotional contagion subscales and the potential influence of gender, ethical, and cultural identification.

General discussion

Across the four studies, we provide preliminary support for the Contagion of Affective Phenomena Scale – Emotion (CAPS-E) and the Contagion of Affective Phenomena Scale – General (CAPS-G). We address a substantial measurement gap by developing a versatile set of instruments for researchers who subscribe to general susceptibility as unidimensional (e.g., mood contagion and affective transfer scholars) and the susceptibility to discrete emotions as multidimensional that enables investigations of both specific emotional contagion and general susceptibility within the same theoretical and measurement framework.

The two lines of evidence support the validity of CAPS-E and CAPS-G. First, CAPS-E yielded a six-factor structure that was confirmed in four ESEM samples (n = 1540). Moreover, as seen by favorable fit indices, the scale appears to offer a statistically sound measure of susceptibility to contagion of six discrete emotions and supports the multidimensionality of the CAPS construct. The CAPS-E subscales (anger, fear, anxiety, sadness, excitement, and happiness) and CAPS-G were also interpreted to demonstrate high internal consistency and acceptable test–retest reliability. In addition, CAPS-G was confirmed as a distinct (but conceptually similar) measure with statistics indicating construct validity across the four ESEM samples.

The second indicator of the validity of the CAPS-E comes from the subscale inter-correlations as well as with other measures of susceptibility to contagion in a manner consistent with susceptibility to emotional contagion theory. High correlations existed among negative emotions and between positive emotions. Correlations between the ECS and CAPS-E were high and lead us to conclude that the two measures tap the same construct (i.e., susceptibility to emotional contagion). Similarly, correlations between CAPS-E and a measure of empathy were low enough to surmise that the two measures tap distinctly different constructs. Overall, CAPS-E and CAPS-G demonstrated associations with theoretically meaningful criterion measures in expected directions, thereby providing support for its concurrent and discriminant validities.

The findings from this multi-study investigation provide support for CAPS-E as a measure of susceptibility to emotional contagion that taps a multidimensional construct, in line with arguments raised by Lundqvist and colleagues (see Wróbel & Lundqvist, 2014). Given that emotional states differ in three independent and bipolar dimensions (i.e., pleasure – displeasure, arousal – nonarousal, and dominance – submissiveness), it is plausible that individual susceptibility to the contagion of discrete emotions is likely to differ between emotional states. Empirical studies demonstrating negative affective states as more contagious than positive affective states (e.g., Spoor & Kelly, 2009) are consistent with the present study’s finding that there was a stronger correlation between the negative emotions (anger, fear, anxiety, and sadness) than compared to correlations between the positive emotions (excitement and happiness). The transference of negative, threat-related emotions is thought to transfer more automatically than positive, non-threat-related emotions due to the important information value that they possess (Kelly et al., 2016).

Although we have begun assessing the predictive validity of the CAPS measure developed here, further validation work would extend the current set of studies. As hypothesized, susceptibility to the contagion of affective phenomena assessed through the CAPS-E and CAPS-G significantly and negatively acted as a predictor of athlete-reported subjective performance satisfaction. Sport is an emotion-laden context, and the results of this study seem to suggest that the ability to manage your own emotions and the emotions of others (i.e., emotional intelligence) is important for athletes’ satisfaction with performance. Given susceptibility to the contagion of affective phenomena was negatively associated with performance, suggesting that being able to catch the affective experiences of others can lead to reduced performance satisfaction for athletes. For example, if an athlete is the recipient of verbal aggression from an opponent and starts to feel angry (contagion), this may be beneficial for performance if anger is interpreted as helpful (intelligence) but interpreted negatively if deemed unhelpful. There are domain-specific nuances that require more analysis and discussion outside of the scope of this study. It is important to further investigate the predictive validity statistics of CAPS-E and CAPS-G to assess how well both measures predict future behavior. For example, research relating the susceptibility to contagion of affective phenomena with the group-level construct group affective tone is a logical extension of construct validity research. It is expected that individual susceptibility to contagion will predict the development of group affective tone, a state-like construct. Also, deserving of empirical attention are variables such as different leadership approaches, leader effectiveness, team cohesion, and team performance (see Clarkson et al., 2020). The validity of this conceptual framework, and particularly the precise predictions regarding the effects of a two-dimensional emotional contagion construct, needs to be developed further and empirically tested.

There are several applications for CAPS-E and CAPS-G. First, it is important to highlight that both measures can be used independently or in combination with each other as a diagnostic tool in organizational consultancy work, for example, or as a data-gathering tool in research that assesses personality and group dynamics. Second, practitioners may use these measures to guide the design of emotional awareness educational programs. For instance, it may be useful to administer these scales at the start of training to help participants self-assess their susceptibility to the contagion of affective phenomena and contextualize the importance of emotional awareness training to that individual.

Several limitations should be acknowledged. First, it remains unclear whether CAPS-E or CAPS-G have application in ethnically diverse populations as the sample used to confirm validity was predominantly White and all sampled in the UK, an individualistic country. Future research is needed to understand the effects of susceptibility to contagion in other cultures (e.g., collectivistic countries where relationships are hierarchical and highly interdependent). Second, previous studies have shown strong robust sex differences in mean susceptibility to contagion scores, with females considered more susceptible to the emotions of others than men (Hatfield et al., 2018). The sex ratio in the samples used in the present study was, on average, 40:60 (female: male). As such, it is important for future studies to continue to evaluate CAPS-E’s and CAPS-G’s factorial invariance across populations (i.e., sex, ethnic minorities). Nonetheless, our analyses showed that CAPS-E and CAPS-G were consistently valid across sex groups. Further assessment of configural, metric, and scalar invariance testing is also recommended.

Other broader limitations of this series of studies should be acknowledged. First, the removal of items within Study 3 that CAPS shared with the criterion measures could be seen as a limitation of the study. In the absence of guidelines for how to deal with scales with shared variance to avoid false-positive inter-correlations, we performed a subsequent scale analysis that indicated that these scales were reliable and suitable for use as criterion measures. Second, a relatively small sample size (n = 116) of participants completed both CAPS and the IRI, the discriminant measure within Study 3. Although the size was far above the minimum sample recommendations for correlational analysis (n > 25, Bonett & Wright, 2000), an increased sample size could have strengthened the reliability of the conclusions. Third, a small number of self-report measures were used in Study 4 to demonstrate predictive validity. The sensitivity of self-report measures of emotional intelligence, for example, has been questioned by researchers in favor of ability-based measures (e.g., O’Connor & Little, 2003). Nevertheless, the predictive validity of CAPS in predicting performance satisfaction over and above emotional intelligence should not be dismissed and indeed strengthened in light of the issues contended above. Researchers may wish to pursue other outcomes of interest (e.g., leadership effectiveness) for which predictive validity has not yet been demonstrated. All measurements were obtained via self-reports and there was no assessment of social desirability responding. Utilising observer ratings would be an added advantage for future research.

In summary, a reconceptualization of susceptibility to emotional contagion is presented as a two-dimensional construct that acknowledges the under-represented concept of generalized susceptibility in the contagion of affective phenomena. Specifically, four studies were conducted that outline the systematic development of two psychometric measures of the susceptibility to contagion of affective phenomena: discrete emotions (CAPS-E) and general affect (CAPS-G). The multidimensional approach adopted in the present work provides a broader perspective on the dimensions of contagion than has previously been used. The multidimensional nature of susceptibility to emotional contagion has been interpreted to demonstrate construct validity, criterion validity (i.e., concurrent and predictive validity), discriminant validity and test–retest reliability of the measures providing support for the scales as valid and reliable tools

Open scholarship

This article has earned the Center for Open Science badges for Open Data, Open Materials and Preregistered. The data and materials are openly accessible at 10.17605/OSF.IO/GSB6H

Disclosure statement

No potential conflict of interest was reported by the author(s).

Data availability statement

The data that support the findings of this study are publicly available at https://doi.org/10.17605/OSF.IO/U3P8A

Additional information

Funding

The author(s) reported that there is no funding associated with the work featured in this article.

Notes on contributors

Beth G. Clarkson

Beth G. Clarkson works as a Visiting Senior Lecturer in Sport Management at the University of Portsmouth and as a Senior Leadership and Workforce Development Programmes Manager at the Premier League.

Christopher R.D. Wagstaff

Christopher R.D. Wagstaff is a Professor in Applied Psychology at the University of Portsmouth.

Calum A. Arthur

Calum A. Arthur is an Independent Scholar and Organisational Development Consultant to the NHS.

Richard C. Thelwell

Richard C. Thelwell is Interim Executive Dean to the Faculty of Science & Health at the University of Portsmouth.

Notes

1 Measurement availability: CAPS and scoring instructions can be freely accessed at an OSF registration page: https://doi.org/10.17605/OSF.IO/Q3KD9