Examining emotional intelligence in older adults with chronic pain: A factor analysis approach

Abstract

Objective

The current study explored whether the three-factor structure of an emotional intelligence measure (attention to emotions, clarity in understanding emotions, and emotion regulation) developed in a sample of college students would replicate in a sample of older adults with chronic pain.

Method

Confirmatory and exploratory factor analyses were conducted to examine the factor structure of the 30-item Trait Meta-Mood Scale among 340 older adults with knee osteoarthritis.

Results

Confirmatory factor analyses indicated that the original three-factor model of emotional intelligence did not fit well with the data for older adults. Exploratory factor analyses revealed a four-factor model of emotional intelligence: (1) Confusion, (2) Acceptance, (3) Rejection, and (4) Insight. Correlations between the original and new subscales were explored.

Conclusion

While the newly derived emotional intelligence scales resembled the original conceptualization of emotional intelligence proposed by Salovey et al. (1995), the current study highlights the differences in emotional intelligence likely representative of older adults with chronic pain.

Introduction

Emotional intelligence is a form of social intelligence that encompasses the ability to accurately evaluate emotions in oneself and others and to use that information to direct subsequent thoughts, feelings and behaviors (Mayer & Salovey, 1990). Mayer and colleagues (1989, 1995, 1998) conceptualized emotional intelligence as paying attention to emotions, experiencing emotions clearly, and regulating emotions to enhance positive mood states and decrease negative ones. Mayer & Salovey (1993) posited that individuals who pay attention to their emotions and have clarity about what they are feeling are likely to be able to regulate their moods more quickly and effectively. Building upon this, Mayer and Stevens (1994) differentiated emotion evaluation from emotion regulation, acknowledging that despite their relatedness, emotion regulation is a distinct ability.

A large body of research suggests that emotion regulation varies across the lifespan (Carstensen, 1992, 1995, 1998). Socioemotional selectivity theory (SST) posits that cognitive appraisal of one’s time left to live influences social goals and motivations, affecting the emphasis that is placed on experiencing and regulating emotions in the present (Carstensen, 1992). For younger adults, goals tend to be future-oriented and rooted in the desire to learn and acquire knowledge. In later life, goals tend to be present-oriented and focused on enhancing current feelings, emotions, and meaningful relationships. Thus, attention to and active regulation of emotions may function differently among older versus younger adults. In particular, emotion regulation may act as a protective factor against the negative effects of chronic health conditions prevalent in late life (Löckenhoff & Carstensen, 2004).

In the Strength and Vulnerability Integration (SAVI) model, Charles (2010) agrees that utilizing emotion regulation strategies represents a strength for older adults; however, she includes an important caveat: older adults who experience sustained distress over an extended period of time are less likely to demonstrate this strength. Rather, prolonged distress lessens or eliminates the ability to regulate one’s emotions and return to baseline emotional state. According to Charles (2010), sustained physiological arousal leads to decreased physiological tolerance and recovery, which influences the capacity to regulate other systems effectively. This suggests that older adults with chronic pain may not possess the strengths in emotion regulation that are often found in later life, leaving them physiologically and emotionally vulnerable to negative experiences.

Osteoarthritis is a chronic health condition that primarily affects older adults. It most commonly affects the weight-bearing joints that get the heaviest use across the lifespan, and it is a leading cause of chronic pain among older adults (Lawrence et al., 2008). Myriad studies have demonstrated the relationship between chronic pain and depression among older adults (Casten, Parmelee, Kleban, Lawton, & Katz, 1995; McCarthy, Bigal, Katz, Derby, & Lipton, 2009; Parmelee, Tighe, & Dautovich, 2015). In fact, epidemiological studies suggest that living with chronic pain puts older adults at two to four times greater risk of developing depression (Panagiotis et al., 2017). Despite this increased risk, not all older adults living with chronic pain experience depression. Given the potential for pain to adversely affect mood, emotional intelligence may play an important role in facilitating positive adjustment to chronic pain.

Broadly, research supports the notion that emotional intelligence may buffer against the negative impact of chronic pain (Wright & Schutte, 2014). For example, in a study of older women with chronic pain, clarity in understanding one’s emotions attenuated the negative impact of pain on positive affect (Zautra, Smith, Affleck, & Tennen, 2001). Across two different studies in chronic pain populations, emotion regulation was found to significantly reduce pain intensity (Connelly et al., 2007, Paquet, Kergoat, & Dubé, 2005). By clearly understanding and regulating their emotions, older adults with chronic pain were able to lessen the impact of pain on psychological and physical well-being. These findings support previous research (Carstensen, Fung & Charles, 2003; Fung & Carstensen, 2004) on the importance of emotional regulation among older adults and extend it by demonstrating its value for coping with chronic pain.

Salovey, Mayer, Goldman, Turvey, and Palfai (1995) developed the Trait Meta Mood Scale (TMMS) to assess emotional intelligence across three domains: (1) Attention (to emotions) (2) Clarity (of emotional experiences), and (3) Mood Repair (emotion regulation). Unlike the State Meta Mood Scale (Mayer & Gaschke, 1988), which was designed to evaluate how a person experiences emotion in the present, the TMMS aimed to capture more stable attitudes and thoughts about experiencing emotions. The TMMS, like many measures of emotional intelligence, was derived using samples of college students. Given the lifespan changes in emotions and emotion regulation emphasized in SST and SAVI, is the TMMS a valid measure of emotional intelligence in a developmentally different sample of older adults with chronic pain? Will it look the same in older adults as it did in college students?

Currently, only one study has addressed this question. Delhom, Gutierrez, Lucas-Molina, and Melendez (2017) examined the psychometric properties of the Spanish version of the TMMS-24 (Fernandez-Berrocal, Extremera, & Ramos, 2004) in a sample of 215 older adults without cognitive impairment. A confirmatory factor analysis of the original three-factor structure (i.e., Attention, Clarity, Mood Regulation/Repair) revealed acceptable statistical and descriptive model fit, and Cronbach’s alpha corroborated goodness of fit for all three scales.

The current study builds upon the research by Delhom and colleagues (2017) by analyzing the structure of the TMMS-30 (Salovey & Mayer, 1995) in a sample of older adults with osteoarthritis. The high prevalence of chronic pain in older adults and the large body of research linking it to negative mood states underscore the need to examine how emotional intelligence functions in this at-risk population. The aims of the current study were, therefore, twofold. The first aim was to examine the fit of observed data for older adults with knee osteoarthritis to the original three-factor structure of emotional intelligence derived in college students. The second aim was to examine in exploratory fashion the structure of emotional intelligence in the current sample of older adults with osteoarthritis pain.

Method

Participants

Three hundred forty older adults with confirmed diagnoses of osteoarthritis of the knee participated in the study. Participants were recruited from two sites: Tuscaloosa, Alabama and Stony Brook, New York. Participants’ ages ranged from 48 years to 97 years; the average age was 63.79 years. Of the 340 participants, 55.3% (n = 188) identified as Caucasian (non-Hispanic white) and 44.7% (n = 152) as African-American. See Table 1 for a full description of demographic characteristics

Table 1.

Participant Demographic Information (N = 340).

	M (SD)
Age (in years)	64.63 (9.32)
Sex, n (%)
Female	264 (77.6)
Male	76 (22.4)
Location, n (%)
Alabama	186 (54.7)
New York	154 (45.3)
Ethnicity, n (%)
Caucasian	188 (55.3)
African-American	152 (44.7)
Education, n (%)
Less than high school	34 (10.1)
High school degree	75 (22.2)
Some college	99 (29.1)
College degree	59 (17.5)
Graduate degree	71 (21.0)
Family income per year, n (%)
Under $10,000	31 (15.3)
$10,000–30,000	53 (26.1)
$30,000–50,000	31 (15.3)
$50,000–70,000	23 (11.3)
Over $70,000	65 (32.0)
Marital Status, n (%)
Married	144 (42.4)
Living with partner	8 (2.4)
Divorced/separated	91 (26.8)
Widow/widower	63 (18.5)
Never married	34 (10.0)

Procedure

Data were drawn from an ongoing longitudinal, multi-site study examining everyday quality of life in older adults with knee osteoarthritis. Institutional Review Board approval was obtained from participating universities prior to conducting the study. Participants were recruited from rheumatology clinics, general geriatric outpatient clinics, community organizations, and public service announcements, with sources and procedures varying slightly at the two sites.

To be included in the study, participants must have received a confirmed diagnosis of knee osteoarthritis and been over age 45. Participants were excluded if they had significant cognitive impairment, a life-threatening physical illness, another painful rheumatologic disorder (e.g., rheumatoid arthritis, lupus, fibromyalgia), or difficulty understanding and completing an interview in English. Eligibility was confirmed in a telephone screening interview. Qualified participants completed a set of self-report measures and an in-person interview.

Measures

The TMMS (Salovey & Mayer, 1989) is a 30-item measure of emotional intelligence that contains three subscales representing distinct areas of mood and emotion regulation: Attention [to feelings], Clarity [in discrimination of feelings], and Mood Repair/Regulation. Attention is composed of 13 items (e.g., “I pay a lot of attention to how I feel”). Clarity is composed of 11 items (e.g., “I am usually very clear about how I feel”). Mood Repair consists of 6 items (e.g., “Although I am sometimes sad, I have a mostly optimistic outlook”). All items are rated on a 5-point Likert scale ranging from strongly disagree (1) to strongly agree (5). Half of the items (n = 15) are worded negatively. Cronbach’s α for the subscales in the current sample was .72, .77, and .60, respectively, suggesting that the Mood Repair scale did not cohere well among participants in this study.

Statistical Analyses

First, confirmatory factor analyses of the 30 items from the TMMS were conducted using Mplus version 8 (Muthen & Muthen, 1998) to determine whether the three-factor model of emotional intelligence established with young adults fit our sample of older adults. To determine model fit, (1) the chi square test of normality, (2) the root mean squared error of approximation (RMSEA), and (3) the comparative fit index (CFI) were assessed.

Second, exploratory factor analyses of the 30 TMMS items were conducted using SPSS version 24 (IBM, 2016) to find the optimal model of emotional intelligence for this sample. Factors were extracted using principal axis factoring (PAF) with direct oblimin rotation (Δ 0.4) and Kaiser normalization. The variance accounted for by the solution and by each individual factor, along with interpretability of the factors, were all evaluated to determine the plausibility of the factor structure.

Exploratory factor analyses were conducted iteratively until all remaining items had a primary factor loading greater than |.3| and did not load onto more than one factor. A minimum cutoff of |.15| between the two strongest factor loadings was used to determine whether items cross-loaded onto multiple factors. Once the optimal number of factors was derived, Cronbach’s alpha reliabilities and alpha-item-if-deleted analyses were conducted for each mood subscale.

Results

Confirmatory Factor Analysis

The initial confirmatory solution indicated that the three-factor model of EI (Attention, Clarity, Mood Repair) did not fit well statistically, χ² [401] = 1763.647, p < .001, or descriptively, CFI = .530, RMSEA = .100, SRMR = .115. Modification indices (MI) recommended loading items onto multiple factors and correlating the residual variances of many of the items. Despite making modifications (MI > 40), the final model was not a good fit, χ² [394] = 1398.647, p < .001, CFI = .654, RMSEA = .087, SRMR = 1.07. Given the failure of the confirmatory model to achieve adequate fit, exploratory factor analyses were conducted.

Exploratory Factor Analyses

The initial exploratory solution suggested that a four-factor model was the best fit for the data. The variance accounted for by the solution was 36.96%, and the four factors individually accounted for 16.49%, 9.05%, 7.42%, and 4.0% of total variance, respectively. Subsequent analyses explored three-factor, four-factor, and five-factor solutions; however, four factors resulted in the best statistical and substantive fit.

The initial four-factor analysis yielded eight items that either did not load strongly onto a single factor (greater than |.3|) or loaded onto multiple factors (less than |.15| difference between the two highest loadings). Of the items removed, four were originally on the Mood Repair subscale, three were originally on the Attention scale, and one was from the Clarity subscale. Put another way, 66% of the Mood Repair items were removed, 23% of the Attention items were removed, and 9% of the Clarity items were removed.

In the final four-factor analysis, the Kaiser-Meyer-Olkin (KMO) measure of sampling adequacy indicated that the remaining 22 items shared 80.7% of common variance, and Bartlett’s Test of Sphericity (χ² (231) = 1846.855, p < .001) suggested that the correlation matrix between the items was factorable. The communalities, which explain the unique variance accounted for by each item, ranged from .14 to .64, with the majority of items falling into the .35 to .45 range. The variance accounted for by the solution was 38.44%, and the four factors individually accounted for 18.86%, 8.93%, 8.88%, and 3.73% of the variance, respectively. See Table 2 for a complete list of the 22 items and their factor loadings.

Table 2.

Four Factor EFA Rotated Pattern Matrix

Confusion (α = .813)	C	A	R	I
I can never tell how I feel. (C)	.680	−.096	.002	.012
When upset, I realize the “good things in life” are illusions. (R)	.646	.088	.072	.113
I can’t make sense of my feelings. (C)	.621	−.050	.025	−.170
Sometimes I can’t tell what my feelings are. (C)	.619	−.056	−.175	.090
I am usually confused about how I feel. (C)	.526	.091	−.026	−.118
My beliefs & opinions change depending on how I feel. (C)	.433	.237	.169	−.154
Although sometimes happy, have mostly pessimistic outlook (R)	.428	.144	.151	−.209
Acceptance (α = .656)
I pay a lot of attention to how I feel. (A)	−.105	.683	−.093	−.099
Feelings give direction to life. (A)	.019	.548	−.072	.111
I often think about my feelings (A)	.204	.521	−.153	.009
I believe in acting from the heart (A)	−.066	.423	.090	.007
The best way to handle feelings is experience to the fullest. (A)	.089	.390	.145	.221
Rejection (α = .70)
One should never be guided by emotions. (A)	−.116	−.033	.624	−.112
Feelings are a weakness that humans have. (A)	.137	.067	.597	−.259
People would be better if they felt less and thought more. (A)	.021	−.030	.523	.113
I don’t think it’s worth paying attention to your emotions (A)	.273	−.190	.505	.222
I never give into my emotions. (C)	−.110	.069	.477	−.046
Insight (α = .724)
I usually know my feelings about a matter. (C)	.041	−.041	−.060	.816
I almost always know exactly how I am feeling. (C)	.016	−.049	.148	.710
I am often aware of my feelings on a matter. (C)	−.093	.116	−.023	.599
I am rarely confused about what my feelings are. (C)	−.184	.138	.172	.568
I am usually very clear about my feelings. (C)	.053	.034	−.134	.358

Extraction Method: Principal Axis Factoring. Rotation Method: Oblimin with Kaiser Normalization.

Note: (A) = Attention, (C) = Clarity, (R) = Mood Repair from original TMMS-30

Factor 1: Confusion about Feelings

The seven items that loaded onto Factor 1 were all negatively worded items representative of being confused or having difficulty understanding one’s feelings. The two items that loaded most strongly on this factor were “I can never tell how I feel” and “When I am upset, I think the good things in life are illusions.” This factor had a strong α of .812. Note that five of the seven Confusion items were on the original Clarity subscale, and two were on the Mood Repair subscale. Additionally, the latter two items were the only Mood Repair items that were not removed due to weak primary factor loadings and/or cross-loading.

Factor 2: Acceptance of Feelings

The five items comprising Factor 2 were all positively worded and representative of accepting or embracing what one is feeling. The two most representative items are “I pay a lot of attention to how I feel” and “Feelings give direction to life.” This factor had a marginal α of .663, suggesting some lack of cohesion among component items. Item statistics revealed, however, that removal of any one item would weaken the scale’s reliability; hence, the scale was retained as is.

Factor 3: Rejection of Feelings

The five items of Factor 3 were all negatively worded and representative of rejecting or disregarding what one is feeling. The two strongest items include “One should never be guided by emotions” and “Feelings are a weakness that humans have.” This factor had a satisfactory α of .703. Four of the items in this factor were originally on the Attention subscale, and one was originally part of the Clarity subscale.

Factor 4: Insight about Feelings

The five items comprising Factor 4 represented being insightful or certain about what one is feeling. The two most representative items include “I usually know my feelings about a matter” and “I almost always know exactly how I am feeling.” This factor had an acceptable α of .719. All four insight items were originally on the Clarity subscale.

Correlations Among EI Subscales

The correlations among the original and newly derived subscales of emotional intelligence appear in Table 3. The overlap between the items on the original and newly derived subscales resulted in significant correlations (p < .01) among many of the factors. Overall, the strength of the correlations reflected the number of shared items, and the direction conveyed the valence of the item wording. Confusion was negatively associated with Clarity (r = −.804), Mood Repair (r = −.593) and Attention (r = −.202). Acceptance was positively associated Clarity (r = .606). Rejection was negatively correlated with Attention (r = −.664) and Clarity (r = −.329). Insight was positively associated with Clarity (r = .792, p < .01), Repair (r = .406) and Attention (r = .163).

Table 3.

Correlations between new and original EI factors using TMMS items

	Confusion	Acceptance	Rejection	Insight	Attention	Clarity	Repair
Confusion	1.000
Acceptance	.231**	1.000
Rejection	.326**	.073	1.000
Insight	−.343**	.153**	−.087	1.000
Attention	−.202**	.606**	−.664**	.163**	1.00
Clarity	−.804**	−.044	−.329**	.792**	.257**	1.00
Repair	−.593**	.087	−.095	.406**	.20**	.592**	1.00

Note:

^*indicates p < .05

^**indicates p < .01

The negative association between Confusion and Insight (r = −.343, p < .01) underscored substantive differences between the EI abilities. The significant positive correlations between Confusion and Rejection (r = .326, p < .01) and Confusion and Acceptance (r = .231, p < .01) highlighted the variability in responding to feelings that are difficult to understand. The significant positive correlations between Acceptance and Confusion (r = .231, p < .01) and Acceptance and Insight (r = .153, p < .01) suggested that emotions may be embraced regardless of how well they are understood.

Discussion

Mayer and colleagues (1994, 1998) conceptualized emotional intelligence as involving distinct emotion evaluation (Attention, Clarity) and regulation (Mood Repair) processes. Similarly, Carstensen and colleagues (1991, 1993, 2004) characterized emotion regulation as varying uniquely across the lifespan as perspectives on time left to live influence motivation and goals. In the current study, confirmatory and exploratory factor analyses provided support for these two lines of research by showing (1) that the structure of emotional intelligence developed in a sample of college students was not a strong fit in a sample of older adults with chronic pain, and (2) that emotion evaluation and emotion regulation processes appear to differ for younger and older adults.

Corroborating the weak alpha for the Mood Repair scale in our sample, confirmatory factor analysis revealed that the three-factor model of emotional intelligence was not replicable among older osteoarthritis sufferers. Rather, exploratory analysis yielded four underlying factors (Confusion, Acceptance, Rejection, and Insight) in our sample. Our obtained Confusion and Insight subscales drew primarily from the original Clarity subscale; two Mood Repair items completed the Confusion scale. Acceptance derived solely from the original Attention subscale; Rejection also drew heavily from the original Attention scale, with one additional Clarity item.

Our four factors reflect Mayer and Steven’s (1994) processes of emotion evaluation more closely than processes of emotion regulation. In fact, the obtained factor structure did not include a unique mood regulation scale; rather, two Mood Repair items split onto the Confusion factor, and the rest failed to cohere or to load meaningfully on other scales. Although this effect was predictable in light of the low internal consistency of the original Mood Repair scale in our sample, it was somewhat surprising given the demonstrated utility of emotion regulation in late life. Our findings suggest that, in contrast, it may be the ability clearly to monitor and define one’s affective states that dominates emotional processes in later life. Of course, it remains to be seen whether our obtained factors are associated with physical and/or mental health, as the biopsychosocial nature of pain would suggest.

The SAVI model (Charles, 2010) provides an interesting and compelling framework to interpret our findings in the contexts of aging and living with chronic pain. Charles (2010) summarized research indicating that age-related changes in cognition and physiology make it more difficult for older adults to recover from physiological and psychological distress. The SAVI model suggests that prolonged stress can attenuate the age-related strength of emotion regulation. Older adults with chronic pain may be especially likely to have difficulty regulating their emotions in response to negative experiences (Carstensen, 1992) given their additional long-term stressor of living with chronic pain. This may explain why the emotion regulation scale failed to replicate in our sample.

Our obtained factors also resemble the four dimensions of wisdom described by Levenson and Aldwin (2013). Our Insight scale parallels their Self-knowledge dimension, defined as self-awareness and understanding of oneself; our Rejection factor similarly crosswalks with Levenson and Aldwin’s (2) nonattachment, which involves being able to detach from negative feelings and behaviors; The wisdom dimension of Integration, or learning to act in ways that are consistent with aspects of oneself, is consistent with our obtained Acceptance factor and Self-transcendence is the product of the three mood-relevant processes. Given these similarities, it is possible that older adults’ increased life experience and consequent wisdom may help explain the age-related differences in emotional intelligence supported by this study.

On an empirical level, it is possible that the wording of the items in the Mood Repair scale influenced the outcomes of the factor analyses. The two negatively worded Mood Repair items that remained in the EFA, “When I am upset, I realize that ‘good things in life’ are illusions” and “Although I am sometimes happy, I have a mostly pessimistic outlook,” are both negatively worded and part of our obtained Confusion subscale. The four positively worded Mood Repair items all failed to load on any of the four obtained factors. The fact that the two Mood Repair items that loaded onto the Confusion subscale were negatively worded while the four that were removed were positively worded supports the first possibility: that the wording of items caused older adults to react differently to them than do younger persons.

Including negatively worded items in psychological measures has been criticized. First, negatively worded items have a tendency to create confusion and error (Zhang & Savalei, 2016). Second, negative wording may influence the covariance structure and hence bias the results (Savalei & Falk, 2014). Our two all-positively worded (Acceptance, Insight) and two all-negatively worded (Confusion, Rejection) subscales produced by our EFA exemplify the potential implications of “directional” wording in scale development in the older adult population.

More generally, the four subscales derived via EFA seem to reflect two overarching domains: Confusion/Insight, and Acceptance/Rejection. Interestingly, however, constraining the solution to two factors did not reveal the same statistical and substantive factors, and combining the items from parallel factors resulted in weak alpha coefficients. Again, it is possible that switching from positively worded items to negatively worded items may have introduced method bias into the covariance matrix. This is a significant point to consider and deserves further attention in scale development for this population.

It is also possible that the Mood Repair items did not tap into that construct as older adults experience it, possibly due to developmental differences in motives and processes (Carstensen, 1991, 1993, 1996). The TMMS Mood Repair items, which seem to capture the ability to “accentuate the positives” over negative emotions, may not reflect the full range of strategies older adults likely use to optimize mood. For example, SST emphasizes the importance of meaningful social relationships but the TMMS contains no items that address either origin of negative moods in social interactions or their regulation by seeking social support. This is an interesting area for further exploration.

Examining the findings another way, the two overarching factors reflect an emotion evaluation component (i.e., Confusion, Insight) and an emotional response component (i.e., Acceptance, Rejection). It is possible that older adults regulate their emotions through their initial evaluation and response to them, rather than engaging in cognitive reappraisal of the situation. Research suggests that older adults may intentionally avoid emotion regulation strategies involving cognitive reappraisal (Charles, 2010), choosing instead to shift attention away from negative events in favor of positive ones (Carstensen et al., 1995; Livingstone & Isaacowitz, 2018). Given the additional resources necessary to regulate chronic pain on top of age-related declines, this may be more pronounced in the current sample.

In summary, our findings indicated that the construct of emotional intelligence looks different in older adults than in college students, providing support for SST and the evolution of socioemotional processes across the lifespan. Additionally, findings suggested that older adults with chronic pain place less emphasis on emotion regulation strategies than is typical in late life, providing support for the SAVI model.

Clinical Implications

In chronic pain populations, the ability to understand and regulate emotions is associated with decreases in pain intensity (Connolly et al., 2007; Paquet et al., 2005) and its negative impact on well-being (Zautra et al., 2001). The potential for prolonged physiological and cognitive stressors to attenuate the age-related strength of emotion-regulation for older adults (Charles, 2010) has important clinical implications. Our findings highlight the need to develop clinical interventions for older adults with chronic pain that utilize their age-related strengths in emotional intelligence to enhance their capacity and ability to regulate their emotions.

Future research is needed to replicate the current findings in other samples of older adults. It is recommended that future research on EI in older adults examine variation in the structure and process of emotional intelligence within the older adult population, specifically as it relates to chronic pain and age-related physiological decline. Additionally, to fully capture the construct of emotion regulation described in SST, it may be useful to develop a measure of socioemotional intelligence that assesses emotion regulation with items including positive reappraisal and focusing on the positive in addition to other age-relevant strategies such as seeking social support. Finally, our findings underscore the importance of examining the psychometric properties of measures validated in college students and/or with negatively worded items when conducting research with samples of older adults.