A longitudinal examination of social support as a mediator of the personality-health relationship in a national sample of African Americans

Abstract

The present study investigates whether social support mediates the relationship between personality traits and health among African Americans over a five-year period, filling a gap in the literature on longitudinal tests of the personality-health association. Data were collected from a national probability sample of African American adults (N = 200). Personality was assessed at Time 1 (T1), social support was assessed 2.5 years later (T2), and physical functioning was examined 5 years (T3) after T1. Telephone surveys included measures of the Five Factor Model personality traits (T1), social support (T2), and physical functioning (T3). Results suggested that relationships between the T1 personality traits and T3 physical functioning were not mediated by T2 social support. Secondary analyses found that among all T1 personality traits, higher openness and lower neuroticism uniquely predicted higher T2 social support. Further, among T1 personality traits, higher conscientiousness uniquely predicted better T3 physical functioning. This information may be useful to healthcare providers and community members in developing prevention and intervention strategies for African Americans.

The U.S. Department of Health and Human Services’ Healthy People 2020 identified the elimination of health disparities as a primary goal of the nation’s public health agenda. Compared to European Americans, African Americans have higher rates of mortality from heart disease, cerebrovascular disease, kidney disease, and some cancers (National Center for Health Statistics, 2007). Understanding the combination of individual and social factors is important for developing interventions to alter health behaviors (Lee et al., 2018). Interventions for both improving health-enhancing behaviors and decreasing health-compromising behaviors of African Americans can be developed by leveraging based on knowledge about personality traits and social support.

It is critical to discern the processes that link personality traits to health outcomes. Understanding personality allows for predicting what individuals value and how they will respond to various situations. In addition, personality traits influence individuals’ (a) selection of some situations and avoidance of others, (b) evocation of reactions from others, and (c) reactions to their network’s social support (Larsen & Buss, 2018; Mroczek, 2020; Neyer & Lehnart, 2006; Pierce et al., 1997). The influential nature of personality traits has implications for how one processes and acts on information related to social support as well as health. For example, extraverted individuals may be more be likely to search out and notice opportunities to interact with others, with these behaviors eliciting improved social support. Conscientiousness individuals may be more likely to plan, organize, and maintain interpersonal relationships, as well as a healthy diet and exercise program.

Social support is a particularly important variable in the African American experience. Families are arguably the most important health-maintenance system (Bowles & Kington, 1998). Support from families can help African Americans cope with illnesses (Ha et al., 2011) and family is the major means of support for African American senior citizens (Bowles & Kington, 1998). Similarly, African American communities, both social and religious, provide important health-related support and social capital (Holt et al., 2018; Holt et al., 2013; Steers et al., 2019) but more than cross-sectional studies are needed to understand these relationships.

“Longitudinal studies provide the best windows into personality-health relationships” (Kern & Friedman, 2017, p. 403). While cross-sectional designs yield important insights, examining the personality-health connection over time provides a more complete picture of the processes underlying the that connection. Incorporating the variable of time allows researchers to explore whether individuals’ current personality traits influence future information processing related to social support as well as whether traits influence future health outcomes. While longitudinal studies do not allow for causality to be inferred, they do provide information about possible causal processes. Therefore, examining personality, social support, and health outcomes using a longitudinal design is essential to understanding the relationships between these variables. Although research on the personality-health connection is ample, we know little about the importance of social support as an underlying process and no studies have examined these issues using a longitudinal design with an African American sample.

Theoretical Frameworks

The current research is consistent with several theoretical frameworks, all of which emphasize the interplay of individual and social factors. The biopsychosocial model (Engel, 1980; Lee et al., 2018) suggests that health outcomes result from a combination of biological factors, psychological factors (such as personality traits), and social factors (such as social support from one’s network). Social cognitive theory and the related notion of reciprocal determinism suggest that personal factors (e.g., personality traits), the environment (e.g., social support), and behavior (e.g., physical functioning) are constantly affecting one another (Bandura, 1999). More specifically, Bandura posits that this mutual influence occurs in the social environment and emphasizes internal and external social reinforcements, past experiences, and expectations, which are all important to the development of personality. The current study is also consistent with the risk and protective factors model (Fitzpatrick & LaGory, 2000; Hawkins et al., 2002), which suggests that protective factors may mediate the negative consequences of exposure to risk factors and thereby improve outcomes. Protective factors include both individual (e.g., personality traits) and community (e.g., social support) resources.

Similar to the present study’s approach, other researchers (Neyer & Lehnart, 2006; Sanderson, 2013; Swickert, 2009) have suggested mediational models of personality-health, such that persons with certain personality traits or individual differences may be better (or worse) at eliciting social support, making beneficial connections with others, and participating in community activities, all of which, in turn, may influence health. For example, social support has been identified as a mediator between the personality trait of optimism and mental health in a sample of African American college students (Mosher et al., 2006). Below is a brief review of research of each path.

Personality Traits and Health

A considerable amount of prior research (e.g., Cohen et al., 2012; Smith, 2006; Strickhouser et al., 2017) has documented a relationship between personality traits and health. However, only two studies (Aiken-Morgan et al., 2014; Clark et al., 2019) have examined these relationships among African Americans and neither utilized a longitudinal design. In the first cross-sectional study of African American adults, higher neuroticism was related to poorer self-rated general health and cardiovascular health, higher extraversion was related to better self-rated general and cardiovascular health, while higher agreeableness and conscientiousness were related to better self-reported general health (Aiken-Morgan et al., 2014). In the second cross-sectional study, that utilized the same African American sample as the current study, higher openness to experience, higher conscientiousness, higher extraversion, and lower neuroticism were related to better self-reported physical functioning (Clark et al., 2019). Unlike the Aiken-Morgan et al. (2014) study, agreeableness was not associated with physical functioning in the Clark et al. (2019) study.

Other studies that have not examined African American samples have corroborated a relationship between personality traits and health. These traits include openness to experience (Allen et al., 2017; Lockenhoff et al., 2011), conscientiousness (Allen et al., 2017; Chapman et al., 2007; Lokenhoff et al., 2011; Magee et al., 2013), extraversion (Chapman et al., 2007; Ibrahim et al., 2015; Lokenhoff et al., 2011; Magee et al., 2013), agreeableness (Chapman et al., 2007; Allen et al., 2017), and neuroticism (Chapman et al., 2007); Lokenhoff et al., 2011; Magee et al., 2013), although samples (e.g., adults, patients, informal caregivers) and health outcomes (e.g., self-reported health, physical activity, quality of life) varied across studies. Finally, several reviews have further confirmed the relationship of the Five Factor Model personality traits to health (Cohen et al., 2012; Smith, 2006; Strickhouser et al., 2017). In sum, there is an abundance of evidence that personality traits are related to health, but little of this evidence has examined this relationship in African American population. There is a similar gap in the literature on the relationship between personality traits and social support.

Personality Traits and Social Support

The personality-social support connection is relatively well documented, but none of this research has focused on an African American population using a longitudinal design. A cross-sectional study examining an African American sample found that higher openness to experience, higher conscientiousness, higher extraversion, higher agreeableness, and lower neuroticism were related to higher perceived social support (Clark et al., 2019). Even though this study utilized the same sample as the current study, it did not examine these variables longitudinally. Individuals with certain personality traits (e.g., higher extraversion, lower neuroticism) have been found to have larger social networks and/or were better able to elicit social support from others (Friedman et al., 1993; Larsen & Buss, 2018; Smith, 1992; Wiebe & Fortenberry, 2006). Some research suggests the importance of conscientiousness, social competence, emotional stability (lower neuroticism), and/or agreeableness in eliciting social support (Jensen-Campbell & Malcolm, 2007; Kern & Friedman, 2017; Lodi-Smith & Roberts, 2007). In a college student sample, when all of the Five Factor personality traits were considered together, only higher extraversion and lower neuroticism uniquely predicted higher total social support, while higher levels of openness to experience was marginally significantly predictive of higher social support (Swickert et al., 2010). Therefore, while there is ample evidence for the relationship between personality traits and social support, it is unknown if these personality traits predict health outcomes over time among African Americans (Brown, 2008).

Social Support and Health

A large body of research confirms a relationship between social support and physical health (e.g., Holden et al., 2015; Ibrahim et al., 2015; Matire & Frank, 2014; Moak & Agrawal, 2009; Sriram et al., 2018), reflecting a variety of dimensions of support and indicators of physical health. However, no studies have examined this relationship longitudinally in an African American population. In a study of patients with kidney disease, higher affectionate social support was associated with higher physical health-related quality of life (Ibrahim et al., 2015). A 12-year longitudinal study of young women found that social support was significantly associated with both current and subsequent general health in early adulthood (Holden et al., 2015). A review of research on the relationship between social support and coronary heart disease concluded that lack of social support is a risk factor in both healthy persons as well as in patients with coronary heart disease (Lett et al., 2005). In sum, there is ample evidence for the relationship between social support and health, but few studies have been conducted with an African American sample and none has examined these relationships longitudinally.

Social Support as a Mediator of the Personality-Health Relationship

There is some evidence that social support mediates the relationship between personality and health, but it is not known if these relationships persist over time in African Americans. Some studies have found evidence that social support mediated the personality-health connection (Clark et al., 2019; Connell & D’Augelli, 1990; Shen et al., 2004; Staniute et al., 2015), but only one of these examined included an African American sample. Clark et al. (2019) found that relationships between the Five Factor Model personality traits and physical functioning were at least partially mediated by social support in a cross-sectional analysis of the African American dataset used in the current study. However, mediation was not examined longitudinally.

The Present Study

The purpose of the present study is to examine social support as an explanation for the relationship between personality traits and physical functioning in an African American adult sample using the Five Factor Model of Personality (e.g., Costa & McCrae, 2004) in a five-year longitudinal project. The study examines personality traits at Time 1 (T1), social support at Time 2 (T2; 2.5 years after T1), and physical functioning at Time 3 (T3; 5 years after T1). Despite continuing calls for more investigation of mediating pathways between personality traits and health, only a few studies have examined the potential mediating role of social support in explaining the personality-health connection (e.g., Neyer & Lehmart, 2006). None of these studies tested social support as a mediator of the personality-health association among African Americans utilizing a longitudinal design.

We focus on African Americans because they carry a disproportionate burden of chronic disease and other poor health outcomes (Williams, 2012). Given the history of racial discrimination and stigma, social support may be particularly influential in these relationships for African Americans (Chandler, 2010). Prior research on collectivism and extended family suggests that social support among African Americans has been essential to the advancement of the African American community (Richardson, 2009). Persistent ethnic health disparities are embedded in the sociocultural context of African Americans’ lives (Harvey & Afful, 2011) and necessitate deepening scholars’ understanding of the personality-health association among African Americans. We focus on the Five Factor Model of personality because prior research suggests that all personality traits tend to be organized around these five factors and many personality-health studies have utilized the Five Factor Model (Costa & McCrae, 2004; Smith, 2006).

This study advances knowledge by using a longitudinal design, which may add insights about possible causal processes. This study also may provide useful information for developing interventions to help individuals modify their personality traits over time (e.g., increase conscientiousness) and better leverage resources from their social network to ultimately improve their health. Finally, this study may encourage healthcare providers to consider the patients in their entirety, and not just the medical and biological aspects of their health.

Consistent with prior research, we hypothesize that persons with higher openness to experience, conscientiousness, and extraversion, and lower neuroticism (but not agreeableness) at Time 1 will have more success in promoting supportive close relationships at Time 2, because they are likely to have larger social networks, be better skilled at utilizing their networks, and less likely to have conflicts with those around them. Supportive relationships, in turn, will be related to higher levels of self-reported physical functioning at Time 3. Therefore, we hypothesize that social support will at least partially account for the relationship between each of the five personality traits and physical functioning (Clark et al., 2019).

We also performed two secondary analyses in order to clarify our mediation findings as well as to extend prior studies on the relationships between personality, social support, and health. First, similar to Swickert et al. (2010), we examined which T1 personality traits uniquely predicted T2 social support. Second, consistent with previous studies on personality, support, and health (e.g., Sriram et al., 2018; Strickhouser et al., 2017), we examined which variables among the T1 personality traits and T2 social support uniquely predicted T3 physical functioning. In summary, while some prior research suggests that support is a mediator of the personality-health connection, no studies have used a longitudinal design with an African American sample.

Method

Participants

Archival data were obtained from the Religion and Health in African Americans study (RHIAA; described below). The analytic sample for the current study (N = 200) included 115 women and 85 men who completed all three waves (T1, T2, T3) of the study. At T1 (in 2010), participants’ average age was 59.30 years (SD = 11.99), with an age range from 26 to 89 years and participants indicated a median income in the $30,000 to $40,000 category. Many (41.0%) were married or living with a partner, had a high school education (31.5%) or 1 to 3 years of college (37.5%), and worked full-time (27.5%) or were retired (34.5%). Geographically, participants were from the Southern (57.5%), Midwestern (19.5%), Northeastern (21.5%), and Western (1%) United States based on the US Census regions. The distribution of regional percentages was similar to those reported for African Americans in the 2010 US Census (Rastogi et al., 2011). Table 1 contains the demographic information for the sample.

Table 1.

Demographic Characteristics of the Analytic Sample at Time 1 (N = 200)

Variable	Categories	N (%)
Gender	Male	85 (42.5%)
	Female	115 (57.5%)
Age	M = 59.30	SD = 11.99
Income	Less than $5,000	9 (5.5%)
	$5,001-$10,000	14 (8.5%)
	$10,001-$20,000	26 (15.8%)
	$20,001-$30,000	24 (14.5%)
	$30,001-$40,000	18 (10.9%)
	$40,001-$50,000	19 (11.5%)
	$50,001-$60,000	10 (6.1%)
	More than $60,000	45 (27.3%)
Relationship Status	Never been married	20 (10.0%)
	Currently single	39 (19.5%)
	Currently married or living with partner	82 (41.0%)
	Separated or divorced	31 (15.5%)
	Windowed	28 (14.0%)
Education	Elementary	4 (2.0%)
	Some high school	11 (5.5%)
	High school graduate	63 (31.5%)
	Some college or technical school	47 (23.5%)
	College graduate	75 (37.5%)
Employment Status	Full-time employed	55 (27.6%)
	Part-time employed	23 (11.6%)
	Not currently employed	30 (15.1%)
	Retired	69 (34.7%)
	Receiving disability	22 (11.1%)
Region of US	Southern	115 (57.8%)
	Midwestern	39 (19.6%)
	Northeastern	43 (21.6%)
	Western	2 (1%)

Hayes (2018) notes the difficulty of conducting power analyses with mediational models because researchers often do not have the needed information. However, one survey of prior mediational models that tested indirect effects found that the median sample size was 142.5 and that 53% of studies had sample sizes of 200 or less (Fritz & MacKinnon, 2007), so our sample size is consistent with most of the prior research.

Telephone Survey Procedure

The Religion and Health in African Americans (RHIAA) project is a national probability based-survey designed to test a theoretical model of the religion-health connection (Holt et al., 2014). A professional sampling firm generated a call list of households from all 50 United States using probability-based methods. The firm obtained home phone numbers from publicly available data such as motor vehicle records. Professional interviewers dialed telephone numbers from a purchased list of U.S. households. The interviewer identified an adult who lived at the household and introduced the project. If the contact expressed interest, a brief eligibility screener was administered to determine whether the person self-identified as African American and was 21 or older. Interested and eligible contacts were read an informed consent script and provided verbal consent. Participants who agreed to be re-contacted were called 2.5 years after Time 1 (T1) and interviewers administered the Time 2 (T2) survey. The final Time 3 (T3) survey was administered 5 years after T1. Participants were mailed a $25 gift card upon completion of each of the three surveys. For the most part, the same questionnaires were presented at each of the three waves, although the personality measures were only presented at T1. Questionnaires were also presented in the same order each of the three waves. It took participants about 45 minutes to complete the measures at each of the three times. The study was approved by the Institutional Review Board.

Response Rates

The response rate at T1 was calculated as the proportion of complete interviews to the total number of eligible individuals. Only 13 individuals who were screened and eligible refused to participate, resulting in an upper bound response rate of 98% (803/816). The overall response rate was 27%, 803 accepted/(803 accepted + 2,195 refused). Another 379 individuals were not eligible for various reasons: 31 were younger than 21 years, 159 refused to provide an age for use in eligibility screening, and 189 were not African American. A total of 3,390 calls were made (summing each of these dispositions). A brief refusal survey was conducted to compare responders to non-responders (N = 73) using t and ꭓ² tests. Compared to responders, non-responders were older (nonresponders M = 65.52 years old, SD = 15.28 vs. responders M = 56.01 years old, SD = 15.00), more likely to be men (62.0% for nonresponders vs. 47.2% for responders), and less likely to have attended 1 to 3 years of college (14.1% for nonresponders vs. 25.8% for responders). Our overall response rate of 27% is similar to those in the Methodology Reports of the National Institutes of Health/National Cancer Institute’s Health Information National Trends Surveys (HINTS), which recruits nationally representative samples of Americans who are asked about their use of health information.

A total of 803 participants (379 men, 424 women) completed measures at T1 and 200 participants completed the measures at T1, T2, and T3, resulting in a retention rate from T1 to T3 of 25%. While the retention rate may seem low, it should be noted that the original study (at T1) did not include planned follow-up assessments. However, once funding became available, these participants were re-contacted and asked to complete follow-up interviews at T2 and T3.

The analytic sample for the present paper comprises the individuals who provided data at each of the three waves of data collection (N = 200). Those who completed the T3 interview, compared to those who did not (N = 603), were significantly older (59.30 vs. 54.93 years old, t(443.55) = 4.11, p < .001, 95% CI 2.28-6.46; Cohen’s d = 0.31). However, there was no gender difference (χ²(1) = 2.34, p = .13, OR = .77, 95% CI .56-1.07). After adjusting for age and gender, the retained participants were more likely than non-retained ones to have college education (χ²(1) =7.27, p = .007, OR .64, 95% CI .46–.89), no difference in reporting “poor” health (χ²(1) = 1.28, p = .26, OR 0.69, 95% CI 0.36–1.32), no difference in having an income greater than $30k (χ²(1) = 1.94, p = .16, OR .78, 95% CI .55–1.11), and no difference in whether married or living with partner (χ²(1) = .01, p = .93, OR .99, 95% CI .71–1.37).

Measures

A modified version of the NEO Five-Factor Inventory-Form S (25-items; Costa & McCrae, 2004) was used to assess the Five Factor Model (Big Five) personality dimensions. The scales assessed Openness to Experience (e.g., “Enjoy hearing new ideas”), Conscientiousness (e.g., “Am always prepared”), Extraversion (e.g., “Am skilled at handling social situations”), Agreeableness (e.g., “Have a good word for everyone”), and Neuroticism (e.g., “Often feel blue”). Each of these personality traits was measured with five items. Participants responded on a Likert-type scale (1 = strongly disagree to 5 = strongly agree). We summed the item scores to get the score for each trait. Higher scores indicate more of the trait. These items are from a widely-used instrument with internal consistency in a prior study using the 80-item version ranging from .65 to .91 in an African American sample (Collins & Gleaves, 1998). Internal consistencies of the scores were good for the current sample: Conscientiousness α = .75, Extraversion α = .71, Openness to Experience α = .71, Neuroticism α = .79, and Agreeableness α = .73.

Prior research suggests that there is little difference in the five-factor structure between European Americans and African Americans (Collins & Gleaves, 1998; Savla et al., 2007). Both convergent and discriminant validities have been established with other versions, including the prediction of various aspects of job performance and MMPI scores (Costa, 1996; Costa et al., 1986; Piedmont & Weinstein, 1994). Costa et al. (1986) found evidence for convergent validity: (a) NEO-PI openness was positively associated with MMPI intellectual interests; (b) conscientiousness positively correlated with MMPI religious orthodoxy; (c) extraversion positively correlated with MMPI surgency/extraversion; and (d) neuroticism was positively correlated with MMPI neuroticism. Costa et al. (1986) found evidence for discriminant validity: NEO-PI openness was not correlated with MMPI psychoticism; (b) conscientiousness was not related with MMPI intellectual interests; (c) extraversion was not correlated with MMPI cynicism; and (d) neuroticism was not correlated with MMPI surgency/extraversion.

The Interpersonal Support Evaluation List (ISEL-12; Cohen, 2008) was used to assess perceived social support (e.g., “I feel there is someone I can share my most private worries and fears with”). The scale consisted of 12 items. Participants use a four-point Likert-type scale (1 = definitely false to 4 = definitely true) to respond. Scores can range from 12 to 48, with higher scores indicating more support. Prior research has indicated that internal reliability estimates have ranged from .75 to .90 across four studies, and were generally above .80 (Cohen, 2008). Reliability for the current sample was also good (α = .88).

Convergent and discriminant validity have been demonstrated for the ISEL-12. Convergent validity was evidenced by the ISEL-12 correlating with various other social support measures, including the Locke-Wallace Marital Adjustment Scale (Locke & Wallace, 1959), the Social Network Diversity Index (Cohen et al., 1997), the Social Network Integration Scale (Merz et al., 2014), and the number of people in one’s social network (Cohen, 2008). Discriminant validity has been evidenced through the low or nonsignificant correlations with the Profile of Mood States Scales (McNair et al., 1971; anxiety, depressed, anger, and calm) in addition to a low correlation with a global engagement scale (Cohen, 2008).

Physical functioning was assessed using the physical component summary (PCS) scores of the Medical Outcomes Short Form SF-12 (Ware et al., 1996). This is a popular health-related quality of life instrument that assesses physical functioning (e.g., Does your health now limit you in these activities? If so, how much? Moderate activities such as moving a table, pushing a vacuum cleaner, bowling, or playing golf’). Most items use a five-point Likert-type response format (e.g., 1 = none of the time, 5 = all of the time) to measure how much health problems interfere with daily functioning. The PCS scores were extracted using standard- norm-based algorithms (Ware et al., 2002). An advantage of norm-based scoring is that it allows for meaningful comparisons across scales (Ware et al., 2007). The resulting score has population means of 50 and standard deviations of 10. Scores can range from 0 to 100. Higher scores on the PCS indicate better functioning. Prior research has indicated that test–retest reliability for the SF-12 was acceptable at 0.89 over two weeks (Ware et al., 2002) and 0.72 over one year (Hayes et al., 2017). Item reliability for the current sample was good (α = .89).

Convergent and discriminant validity have been established for the SF-12 physical functioning component. Convergent validity has been demonstrated with the SF-12 positively correlating with the SF-36 (its longer, prior validated counterpart; Ware, et al., 1996) and a perceived health scale (Hayes et al., 2017; Kathe et al., 2018). Further, the changes in SF-12 physical functioning scores were correlated with the changes in back pain intensity, and for patients whose back pain improved, there was a significant increase in the follow-up physical component summary scores as compared to the baseline (Luo et al., 2003). Finally, there is evidence for known groups validity (groups known to differ in their physical and/or mental health as clinically defined; Ware et al., 1996). Discriminant validity has been demonstrated in that the SF-12 physical functioning scores were not highly correlated with perceived mental health, with the Kessler Mental Health Scale (StatisticsSolutions, n.d), the number of psychiatric hospitalizations in the past year, current substance abuse disorder, or the age participants first had emotional/psychiatric problems (Hayes et al., 2017; Salyers et al., 2000).

A standard demographic module was administered. The items assessed several participant characteristics, including gender, age, education, marital status, employment status, and household income before taxes. For education, participants were asked, “What is the highest grade or year of school you completed?” Responses were coded as one of eight options: Never attended school or only attended kindergarten, grades 1 through 8 (elementary), grades 9 through 11 (some high school), grade 12 or GED (high school graduate), college 1 year to 3 years (some college or technical school), college 4 years or more (college graduate), master’s degree, doctoral or other advanced degree (e.g., law, medicine). To assess marital status, we asked “What is your marital status?” and the options were as follows: never been married, currently single, currently married, living with a partner, separated or divorced, widowed. To assess employment, we asked, “Do you work for pay outside of the home?” The options were full time employed, part-time employed, not currently employed, retired, receiving disability. To assess income, participants were asked, “What is the total income of everyone in your household per year, before taxes?” with 8 response categories (coded 1 – 8; less than $5,000, $5,001 - $10,000, $10,001 - $20,000, $20,001 - $30,000, $30,001 - $40,000, $40,001 - $50,000, $50,001 - $60,000, more than $60,000).

Data Analyses

Data analyses were performed using SPSS, including the PROCESS macro. PROCESS is used for observed variable path analysis-based mediation and moderation analyses, among other purposes (Hayes, 2018). For mediation analyses, PROCESS is used to estimate model coefficients, standard errors, t- and p-values, confidence intervals using ordinary least squares regression, and direct and indirect effects. The path from the predictor variable (T1 personality trait) to the mediator (T2 social support) is usually designated the “a” path. The path from the mediator (T2 social support) to the outcome variable (T3 physical functioning) is usually designated the “b” path. The path from the predictor (T1 personality trait) to the outcome (T3 physical functioning) is usually designated the “c” path. When the mediator (T2 social support) is taken into account, the “c” path is then called the “c prime” or c’ path. We conducted bootstrapping (5,000 repeated samples with replacement) to further examine the mediation when present. We also examined the indirect effect and the corresponding 95% confidence intervals. The indirect effect is the effect of a predictor variable (T1 personality trait) on an outcome variable (T3 physical functioning) that passes through the mediator variable (T2 social support). An indirect effect is calculated by multiplying the paths that constitute the effect (a x b). The magnitude of the indirect effect indicates the extent of mediation through the relevant mediator variables. Mediation is indicated when the confidence interval of the indirect effect does not cross zero.

Results

Preliminary Analyses

We first examined the distributions of the study variables, which indicated distributions acceptable for the planned analysis. Though the social support data were somewhat negatively skewed (−1.50; SE = .18), this pattern was deemed not to impact the analysis due to the use of bootstrapping (Hayes, 2018). We also checked for and found a small number of outliers (range of 2–5 across the major variables). To be conservative, we conducted the mediation analyses with the original skewed social support data and another analysis with a log transformation of (skewed) social support data (Mertler & Vannatta, 2010) and did this both with and without outliers. Because the findings did not differ, we reported the model using all the data and the non-transformed social support data.

Of the participants who completed the survey, missing data resulted in smaller sample sizes (range from 154-160 out of the original 200 in the analytic sample) in the mediation analyses and there was no apparent pattern to the missing data. The PROCESS macro discards cases missing on any variable (listwise deletion), which ensures the analysis is not using different subsets of the data for the model estimates (Hayes, 2018). Finally, we checked to see which, if any, demographic variables correlated with the physical functioning outcome variable. As a result, we controlled for income (assessed at T1) as well as T1 physical functioning in the mediation analyses. We used an a priori alpha of .05. For the mediation indirect effect, we set an a priori completely standardized indirect effect size of .06 (Clark et al., 2019).

Because we used only 25 items from the NEO Five-Factor Inventory–Form S, we performed a factor analysis using IBM SPSS Statistics Version 26. Bartlett’s test indicated the assumption of sphericity could be assumed, χ2(300) = 6114, p < .001. Evidence of sampling adequacy was found to be in the meritorious range (Kaiser-Meyer-Olkin, KMO = .9; Kaiser, 1974). Given that the data appeared acceptable for the planned exploratory factor analysis, we then analyzed the 25 items using the maximum likelihood model extracting five factors to examine the five-factor structure. Because the interfactor correlation matrix contained absolute values larger than .32, we chose to use direct oblimin rotation (Tabachnick & Fidell, 2012). This factor structure was similar to the traditional five-factor structure, with only one item not loading onto its original factor (the openness item of “enjoy hearing new ideas” loaded onto the agreeableness factor). Factor 1 (agreeableness) accounted for 28.21% of the variance before rotation. Factor 2 (neuroticism) accounted for 9.58% of the variance before rotation. Factor 3 (conscientiousness) accounted for 5.71% of the variance before rotation. Factor 4 (openness) accounted for 5.13% of the variance before rotation. Factor 5 (openness) accounted for 4.95% of the variance before rotation. Intercorrelations between the five factors ranged from −.50 to +.51. Given the similarity of the current analysis to the original five-factor solution, we decided to use original five-factor structure.

Means, standard deviations, and intercorrelations for the primary study variables are displayed in Table 2. Higher T3 physical functioning was significantly related to T1 higher conscientiousness (r = .20, p = .006), higher T1 extraversion (r = .17, p = .023), and lower T1 neuroticism (r = −.18, p = .013). T3 Physical functioning was not related to T1 openness (r = .14, p = .053) and not related to T1 agreeableness (r = .05, p = .515). Higher T2 social support was significantly related to higher T1 openness to experience (r =. 34), higher T1 conscientiousness (r = .31), higher T1 extraversion (r = .24), higher T1 agreeableness (r = .26), and lower T1 neuroticism (r = −.27), all at the p < .001 level. Finally, higher T3 physical functioning was significantly related to higher T2 social support (r = .24, p = .001).

Table 2.

Correlations, Means, and Standard Deviations of Study Variables

Variable	M (SD)	OpenT1	ConscT1	ExtravT1	AgreeT1	NeuroT1	SocSupT2	PhysFuncT3	PhysFuncT1
OpenT1	20.32 (2.86)
ConscT1	19.98 (2.91)	.50***
ExtravT1	20.35 (2.75)	.55***	.48***
AgreeT1	21.08 (2.59)	.49***	.52***	.54***
NeuroT1	9.77 (3.26)	−.24***	−.21***	−.12	−.19**
SocSupT2	42.24 (6.12)	.34***	.31***	.24***	.26***	−.27***
PhysFuncT3	43.34 (11.64)	.14	.20**	.17*	.05	−.18*	.24***
PhysFuncT1	45.09 (11.28)	.13	.10	.09	−.11	−.23***	.19**	.56***
IncomeT1	5.12 (2.29)	.03	−.04	.02	−.13	−.20*	.06	.35***	.33***

Note. N ranges from 161 to 199. OpenT1 = Openness to experience at Time 1; ConscT1 = Conscientiousness at Time 1; ExtravT1 = Extraversion at Time 1; AgreeT1 = Agreeableness at Time 1; NeuroT1 = Neuroticism at Time 1; SocSupT2 = Social Support at Time 2; PhysFuncT1 = Physical Functioning at Time 1; PhysFuncT3 = Physical Functioning at Time 3. For Income, the mean of 5.12 reflects an income in the $30,001-40,000.

^*p < .05

^**p < .01

^***p < .001

Mediation Analyses

Overview

We conducted separate mediation analyses using ordinary least squares path analyses (Hayes, 2018; Preacher & Kelley, 2011) for each of the five personality traits, using the PROCESS macro of SPSS without centering the variables. We controlled for T1 income as well as T1 physical functioning. We tested whether the T1 personality trait was related to the outcome of T3 physical functioning. We examined whether the T1 personality trait was associated with the hypothesized mediator, T2 social support. We tested the relationship between T2 social support and T3 physical functioning. We examined both the T1 personality trait and T2 social support in the equation predicting T3 physical functioning to examine whether the relationship between the T1 personality trait and T3 physical functioning decreased or became nonsignificant, indicating partial or full mediation.

Openness to Experience

T1 openness did not predict T3 physical functioning (b = 0.17, t(155) = 0.60, p = .548). However, we found that higher T1 openness predicted higher T2 perceived social support (b = 0.80, t(155) = 4.93, p < .001). Further, we found that higher T2 social support significantly predicted better T3 physical functioning (b = 0.27, t(154) = 2.07, p = .040). With both T1 openness and T2 social support in the equation predicting T3 physical functioning, T1 openness did not predict T3 physical functioning (b = −0.05, t(154) = −0.18, p = .855). The indirect effect was further tested using a bootstrapping estimation approach with 5,000 samples. These results indicated that the indirect coefficient was not significant; the lower and upper bounds of the confidence interval crossed zero (b = 0.22, SE = 0.13, 95% CI = −0.0052, 0.4894). The completely standardized indirect effect of T1 openness on T3 physical functioning was .05.

Conscientiousness

Higher T1 conscientiousness significantly predicted better T3 physical functioning (b = 0.74, t(155) = 2.85, p = .005). Further, we found that higher T1 conscientiousness predicted T2 perceived social support (b = 0.69, t(155) = 4.30, p < .001). T2 social support did not significantly predict T3 physical functioning (b = 0.17, t(154) = 1.35, p = .177). With both T1 conscientiousness and T2 social support in the equation predicting T3 physical functioning, the effect for conscientiousness on T3 physical functioning was significant but decreased (b = 0.62, t(154) = 2.25, p = .026). The indirect effect was further tested using a bootstrapping estimation approach with 5,000 samples. These results indicated that the indirect coefficient was not significant; the lower and upper bounds of the confidence interval crossed zero (b = 0.12, SE = 0.10, 95% CI = −0.0532, 0.3479). The completely standardized indirect effect of T1 conscientiousness on T3 physical functioning was .03.

Extraversion

T1 extraversion was a marginally significant predictor of T3 physical functioning (b = 0.59, t(150) = 1.97, p = .050), such that higher T1 extraversion marginally predicted better T3 physical functioning. In addition, higher T1 extraversion predicted higher T2 perceived social support (b = 0.63, t(150) = 3.43, p = .001). T2 social support did not significantly predict T3 physical functioning (b = 0.20, t(149) = 1.52, p = .131). When both T1 extraversion and T2 social support were in the equation predicting T3 physical functioning, the T1 extraversion effect decreased and did not significantly predict T3 physical functioning (b = 0.46, t(149) = 1.50, p = .136. The indirect effect was further tested using a bootstrapping estimation approach with 5,000 samples. These results indicated that the indirect coefficient was not significant; the lower and upper bounds of the confidence interval crossed zero (b = 0.13, SE = 0.10, 95% CI = −0.0482, 0.3538). The completely standardized indirect effect of T1 extraversion on T3 physical functioning was .03.

Agreeableness

T1 agreeableness did not significantly predict T3 physical functioning (b = 0.47, t(156) = 1.61, p = .108). Further, higher T1 agreeableness significantly predicted higher T2 perceived social support (b = 0.67, t(156) = 3.75, p < .001). T2 social support did not significantly predict T3 physical functioning (b = 0.23, t(155) = 1.83, p = .070). When both T1 agreeableness and T2 social support were in the equation predicting T3 physical functioning, the effect for T1 agreeableness was still not statistically significant (b = 0.31, t(155) = 1.03, p = .303). The indirect effect was further tested using a bootstrapping estimation approach with 5,000 samples. These results indicated that the indirect coefficient was not significant; the lower and upper bounds of the confidence interval crossed zero (b = 0.16, SE = .10, 95% CI = −0.0242, 0.3842). The completely standardized indirect effect of T1 agreeableness on T3 physical functioning was .04.

Neuroticism

T1 neuroticism did not significantly predict T3 physical functioning (b = −0.10, t(155) = −0.43, p = .664). Higher T1 neuroticism significantly predicted lower T2 perceived social support (b = −0.47, t(155) = −3.14, p < .002). Higher T2 social support significantly predicted better T3 physical functioning (b = 0.27, t(154) = 2.13, p < .035). With both T1 neuroticism and T2 social support in the equation predicting T3 physical functioning, T1 neuroticism did not significantly predict T3 physical functioning (b = 0.02, t(154) = 0.09, p < .926). The indirect effect was further tested using a bootstrapping estimation approach with 5,000 samples. These results indicated that the indirect coefficient was not significant; the lower and upper bounds of the confidence interval crossed zero (b = −0.13, SE = 0.08, 95% CI = −0.3125, 0.0015). The completely standardized indirect effect of T1 neuroticism on T3 physical functioning was −.04.

Secondary Analyses

Which Personality Traits Uniquely Predict Social Support over Time?

Similar to Swickert et al. (2010), we examined which of the Five Factor Model personality traits uniquely predicted subsequent social support over time. Using hierarchical regression, we entered the control variable of T1 income in step 1 and entered the five T1 personality traits in step 2. The outcome variable was T2 social support. Higher T1 openness to experience (β = 0.24, t = 3.10, p = .002) and lower T1 neuroticism (β = −0.19, t = 2.97, p = .003) uniquely predicted higher T2 social support (Table 3).

Table 3.

Hierarchical Multiple Regression Analyses for Personality Traits at Time 1 Predicting Social Support at Time 2

	Social Support at Time 2
Predictor	β	R2	ΔR2	F	df	t	p
Step 1		.02	.02	5.53	1, 238		.02
Income	0.15					2.35	.02
Step 2		.21	.18	10.01	6, 233		<.001
Openness	0.24					3.10	.002
Conscientiousness	0.10					1.40	.16
Extroversion	< 0.01					0.06	.95
Agreeableness	0.03					0.47	.64
Neuroticism	−0.19					−2.97	.003

Note. N = 240; df = degree of freedom.

Do Personality Traits and Social Support Uniquely Predict Physical Functioning Longitudinally?

Using the same variables as in the mediational analyses, we explored which T1 personality traits uniquely predicted T3 physical functioning and whether T2 social support significantly added predictive ability over time. Using regression analysis, we entered variables in three steps. In step 1, we entered the control variables of T1 income (coded 1-8 for the eight response categories described in Table 1) and T1 physical functioning. In step 2, we entered the five T1 personality traits. At step 3, we entered T2 social support. The outcome variable was T3 physical functioning. Among the personality variables, only T1 conscientiousness uniquely predicted T3 physical functioning (β = 0.18, t = 2.13, p = .035) such that higher T1 conscientiousness was related to better T3 physical functioning. T2 social support in step 3 did not uniquely predict T3 physical functioning nor significantly added to the amount of T3 physical functioning variance accounted for (Table 4).

Table 4.

Hierarchical Multiple Regression Analyses for Personality Traits at Time 1 and Social Support at Time 2 Predicting Physical Functioning at Time 3

	Physical Functioning at Time 3
Predictor	β	R2	ΔR2	F	df	t	p
Step 1		.29	.29	31.08	2, 149		<.001
Income	0.19					2.52	.01
Physical Functioning at T1	0.45					6.11	<.001
Step 2		.34	.05	10.67	7, 144		<.001
Openness	−0.13					−1.44	.15
Conscientiousness	0.20					2.29	.02
Extroversion	0.02					1.21	.23
Agreeableness	−0.02					0.23	.82
Neuroticism	−0.06					−0.21	.83
Step 3		.35	.01	9.57	8, 143		<.001
Social Support at T2	0.10					1.24	.22

Note. N = 152; df = degree of freedom.

Discussion

The purpose of the present study was to determine whether social support mediated the relationship between personality traits and physical functioning over five years using a longitudinal design in a national sample of African American adults. We assessed the Five Factor Model personality traits at T1, social support at T2 (2.5 years after T1), and physical functioning at T3 (5 years after T1). Prior cross-sectional research suggests that social support is at least a partial mediator of the Five Factor personality traits and health (Clark et al., 2019). No prior longitudinal studies with African American population have examined the relationships between these variables. The personality-support-health relationship is especially important given the health disparities between African Americans and Europeans Americans (National Center for Health Statistics, 2007) and the potential implications for informing intervention development.

We found no evidence that T2 social support mediated the relationship between any of the T1 Five Factor personality traits and T3 physical functioning. These longitudinal findings are inconsistent with the cross-sectional results of Clark et al. (2019) using only T1 data, who found evidence for at least partial mediation by social support for all of the five factor traits. Other prior research also found evidence for social support mediating the personality-health connection, although those studies examined different traits and different measures of health and followed participants for shorter durations than the current study (e.g., Shen et al., 2004; Staniute et al., 2015). The relatively long duration between the time periods (2.5 years between T1 and T2, and another 2.5 years between T2 and T3) may account for the lack of significant results in the mediational analyses and, relatedly, for the relatively small correlations between variables (Table 2). It may be that over time people’s social networks and related social support change or dissipate. Also, more proximal factors (e.g., T3 personality and social support) may influence T3 physical functioning than T2 social support from 2.5 years earlier or T1 personality traits from 5 years earlier.

Regarding the personality-physical functioning association, contrary to prior research (e.g., Aiken-Morgan et al., 2014), we found weak relationships between T1 personality traits and T3 physical functioning in the mediational models and in the zero order correlations (Table 2). The five-year period between T1 and T3 possibly accounts for the differences. With respect to the personality-social support relationship, our mediational analyses were that those higher on the T1 traits of openness to experience, conscientiousness, extraversion, and agreeableness, and lower on neuroticism reported higher perceived T2 social support 2.5 years later. These results are fairly consistent with prior research (e.g., Swickert et al., 2010).

In relation to the social support-physical functioning connection, we found some evidence that higher T2 social support was related to better T3 physical functioning, although the associations were sometimes weak or non-significant in the mediational models. The 2.5-year period between T2 social support and T3 physical functioning, and related changes in T2 social support may account for the weaker relationships between these variables.

Further, our failure to replicate prior research may relate to how African Americans might uniquely experience social support, health, and personality. Prior research suggests the importance of religious support, religious capital as well as social capital (including interconnectedness and community participation) to the health of African Americans (Holt et al., 2015; Holt et al., 2013), coping with illness (Ha et al., 2011), and caring for older extended family members (Dilworth-Anderson & Goodwin, 2005).

In the secondary analyses, we attempted to replicate Swickert et al. (2010)’s examination in which personality traits uniquely predicted social support. While Swickert et al. (2010) found, using a cross-sectional design, that higher extraversion and lower neuroticism uniquely predicted higher social support, we found that higher T1 openness and lower T1 neuroticism uniquely predicted higher T2 social support. However, even though only openness and neuroticism uniquely predicted social support in the current study, the correlations depicted in Table 2 indicated that all of the five T1 traits (higher openness, conscientiousness, extraversion, agreeableness, and lower neuroticism) were weakly to moderately related to higher T2 social support. Swickert et al. (2010) found that all five personality traits were correlated with social support. The differences between our findings and those of Swickert et al. may be due to differences in sample and in the personality measure utilized. Swickert et al.’s sample was recruited from a college with most participants 18-26 years old and only 14% identified as African American, whereas we recruited a sample of African American adults with an average age of 59 years old. Swickert et al. used the 60-item version of the Five Factor traits whereas we used a 25-item version, which was more manageable for a telephone survey. And, as noted, Swickert et al. used a cross-sectional design while we used a longitudinal design.

Additionally, in the secondary analyses, we examined whether T1 personality traits and T2 social support uniquely predicted T3 physical functioning (e.g., Strickhouser et al., 2017). We found that only T1 conscientiousness uniquely predicted T3 physical functioning, such that higher T1 conscientiousness predicted higher T3 physical functioning. Conscientiousness includes developing and carrying out plans, and accomplishing tasks without delay. These characteristics are consistent with engaging in behaviors essential for good physical functioning (e.g., healthy diet, regular exercise, making and keeping physician appointments). T1 conscientiousness was also moderately correlated with the other four T1 personality traits and T2 social support (Table 2), so those higher in conscientiousness also tended to be higher in openness, extraversion, agreeableness, and social support, and lower in neuroticism. Further, T1 traits of higher conscientiousness, higher extraversion, and lower neuroticism and higher T2 social support weakly correlated with higher T3 physical functioning (Table 2). The secondary analyses and correlational findings suggest that while there was little evidence for longitudinal mediation, there is evidence that some T1 personality traits predict T2 social support, and both T1 personality traits and T2 social support predict, if not always uniquely, T3 physical functioning.

Limitations of the Present Study

The present study has several limitations. Participants’ self-reports may have been affected by inaccurate recall or social desirability concerns. Our sample was not representative of the entire African American population and excluded persons without home telephones. Further, we were only able to examine T1 personality traits predicting T2 support and predicting T3 physical functioning due to the lack of availability of data on all the variables at all three time periods. For example, social support or physical functioning may influence personality trait change.

Another limitation involved personality trait measurement. Factor analyses of the personality items did not perfectly replicate the original five-factor structure (van der Linden et al., 2010), with one item loading on a different factor. However, prior research has replicated the five-factor structure with African American samples (Collins & Gleaves, 1998; Savla et al., 2007). The difference may have been due to our use of a shorter assessment form, self-presentation concerns, or cultural variables related to this mostly older African American sample. Additional validity studies on the Five Factor Model measures with African American samples are needed.

Future Directions, Applications, and Conclusions

Future research should evaluate personality-health models using different health indicators (e.g., cardiovascular disease) as other personality-health researchers have done (e.g., Aiken-Morgan et al., 2014). Future research should examine the different dimensions of social support such as religious and informational support. In addition, the diversity of the social network (Eng et al., 2002), number of social roles, or degree of social integration (Crittenden et al., 2014) may shape the personality-health relationship.

Other potential personality-health mediators, such as health habits, appraisal, stress, and coping (Sanderson, 2013), should be examined as they have not been thoroughly explored in African American samples. Finding no evidence for the mediating role of social support in the personality-physical functioning relationship over five years (but has been found in shorter duration and cross-sectional studies) is important as further research is needed to explore the duration over which social networks are more, versus less, effective. More research is needed on why social support is less effective over time (e.g., perhaps the social network becomes less stable or dissipates over longer periods of time).

Using cluster analysis to examine the efficaciousness of certain combinations of personality traits may be helpful. For example, perhaps the group most likely to show social support-related health benefits are those who have a combination of high conscientiousness, moderate extraversion, and low neuroticism. A more cognitive approach to this research would be to prime certain personality traits and test in a true experiment with randomization to conditions (i.e., prime and no prime) and examine the effects on social support and health outcomes. This type of research would be particularly helpful in determining causal processes.

The current study’s findings may be useful to community members and healthcare providers when considering the role of social support in treatment and prevention efforts. These findings may be particularly relevant for health promotion efforts for African Americans, a group that historically has battled discrimination, health disparities, and has relied on their social network for advancement (Hudson et al., 2016). Similar to recommendations made by Clark et al. (2019), we suggest that interventions targeting individuals’ personality and social skills may help people learn to leverage available social support. Interventions could train participants in diversifying their social network, eliciting assistance from others, and accepting support when offered (Wingood et al., 2013). Additionally, these interventions can help participants develop skills in conflict management with social network members (Chapman et al., 2014). Some Africentric interventions already include relevant training (Gilbert et al., 2009). These interventions build on the cultural importance of neighborhood institutions (e.g., becoming involved in the church aerobics class). Our results also suggest that individuals with lower social support might benefit from interventions to increase conscientiousness and extraversion or decrease neuroticism (see Piedmont, 2001 for an example with a predominantly African American sample; Roberts et al., 2017). These changes in certain personality traits may facilitate communalism and, consequently, improve health behaviors (Chapman et al., 2014). Further, because the mediation relationships did not hold up over our five-year study, we suggest that relapse prevention efforts (“booster sessions”) over time may be especially important. Relapse prevention efforts (e.g., Lyons et al., 2019) can help clients maintain personality-guided cognitions and behaviors relevant to eliciting needed health-related social support or teach new behaviors as the individual’s situation changes over time (e.g., finding a new jogging partner after relocation to a new city). Research on interventions is essential and randomized trials would be especially helpful in examining the effectiveness of interventions over time.

Kern and Friedman (2017) noted that theory driven personality-health research is consistent with patient-centered care and can inform risk models, decision making, and treatment options. Treatment options could be tailored more directly to those most likely to respond, saving time and cost. Such treatment could be tailored based on personality and social support. For example, a healthcare provider would focus on planning, adherence, and elicitation of social support for someone low in conscientiousness. This information could be constantly updated and inform future patient-provider interactions.

In conclusion, the current findings provide additional information on the complex interplay of social and individual processes relating to health. Culturally sensitive interventions that cultivate conscientiousness and openness to experience and help individuals capitalize on these traits can help them leverage their social support to engage in health behaviors and achieve better health outcomes.