No evidence of “healthy neuroticism” in the Hawaii Personality and Health Cohort

Abstract

Background.

Heterogeneity in the effects of trait neuroticism on mortality has inspired recent theories of “healthy neuroticism,” or the possibility that neuroticism can lead people down either healthy or unhealthy behavioral pathways. The logical extension of this theory is that some construct – perhaps another trait, financial resource or health-relevant situation – changes the relationship between neuroticism and health. The other possibility is that different components of neuroticism lead to different health behaviors, and therefore different outcomes.

Purpose.

The current study systematically examines the relationship between child and adult neuroticism and various health indicators including perceptions of health, behaviors, health outcomes and biomarkers of health. Finally, we examine both potential moderators of the associations with neuroticism and examine its facet-structure.

Methods.

The current study utilizes data from the Hawaii Longitudinal Study of Personality and Health, which includes both adult (IPIP-NEO) and childhood (teacher-reported) measures of personality and SES, as well as a variety of health outcomes, from self-reported health and health behavior to biological markers, such as cholesterol and blood glucose levels. Sample sizes range from 299 to 518.

Results.

The relationship between neuroticism and health was not consistently moderated by any other variable, nor were facets of neuroticism differentially related to health.

Conclusions.

Despite a systematic investigation of the potential “paths” which may differentiate the relationship of neuroticism to health, no evidence of healthy neuroticism was found.

One catalyst for the increasing recognition that personality traits influence health outcomes has been the greater awareness and adoption of common trait taxonomies, allowing researchers the ability to compare findings across samples. For instance, research founded within the Big Five taxonomy has allowed researchers to demonstrate consistent linkages between traits like conscientiousness and neuroticism and health outcomes, typically showcasing the benefits associated with the former and detriments for those individuals higher on the latter (1–4). However, this summary of the literature ignores a substantial number of studies which find null (5–7) or protective effects of neuroticism on health (8–12). Though commonly cast as a universally maladaptive trait for health, researchers also have suggested the possibility that neuroticism may prove beneficial in some circumstances, leading to the concept of “healthy neuroticism” (13). The current study sought to comprehensively examine the evidence for healthy neuroticism using data from the long-running Hawaii Longitudinal Study of Personality and Health (14). Specifically, we sought to examine four potential influences on whether neuroticism may prove adaptive for an individual’s health and health behavior: (i) the individual’s level of conscientiousness, (ii) whether the individual reports having a chronic disease, (iii) the individual’s socioeconomic status, and (iv) if the associations with health differ based on the facet of neuroticism. Though the personality literature has often referred to the notion of “healthy” neuroticism (13), and thus we focus on that terminology here, it is worth noting that each of these tests will help indicate whether neuroticism is at least less maladaptive in some situations; in so doing, our intent was to conduct a comprehensive investigation into when and in which contexts the role of neuroticism on health may differ. Each of these possibilities reflect previous theoretical and empirical literature on the role of neuroticism on health outcomes, and each can potentially lead to healthy neuroticism effects.

Conscientious Neuroticism

A common tendency in research on personality and health is to focus on traits as independent factors influencing health. Such a perspective fails to properly consider that any one individual’s personality reflects a constellation of Big Five (and other) traits, and that association with any single trait may vary depending on a person’s standing on another trait. With respect to neuroticism, one common theory has focused on the interaction between neuroticism and conscientiousness, insofar that it may benefit individuals to be relatively high on both dimensions (15,16). The rationale behind conscientious neuroticism reflects the notion that conscientious individuals are prone toward healthy behavior (17), and this effect may be only further amplified when those individuals also are somewhat anxious or vigilant by disposition, which in turn will keep individuals focused on promoting health.

Evidence for this claim has been found for specific health outcomes and health behaviors. For instance, researchers using the Midlife in the United States data have found a conscientiousness-by-neuroticism interaction when predicting the inflammation biomarker Interleukin-6 (15). In addition, this interaction held when predicting smoking behavior following the onset of disease (18). However, even these studies have shown evidence against conscientious neuroticism; for instance, this interaction did not hold when predicting smoking for the general population without disease (18), suggesting the need to consider additional moderators.

The Moderating Role of Chronic Disease

One possibility is that the role neuroticism plays in health and health behavior is altered in the presence of chronic disease. Individuals may respond to the onset of chronic illness with an increase in health-related vigilance. Alternatively, if individuals have not experienced past disease onset, it is less important for them to focus on identifying symptoms or signs of disease progression. Neuroticism has been linked more broadly to higher symptom reporting (19,20), in line with the central argument for why the trait could prove “healthy” in some instances (13), namely that higher rates of symptom reporting could result from heightened health vigilance. Unfortunately, outside of that one study (18), research has failed to address whether neuroticism’s impact on health and health behavior differs based on chronic disease status.

The Moderating Role of Socioeconomic Status

Central to the theory that neuroticism could be adaptive is that vigilant or anxious individuals are able to enact healthier behaviors when confronted with health concerns (13). Neuroticism has been linked to greater health care costs (21) and more doctor visits (22), presumably related to the associations between the trait and greater vigilance and symptom reporting. However, it is unclear whether these associations would hold across socioeconomic strata, insofar that individuals from poorer backgrounds may not have the resources available for these frequent visits. Moreover, even if neuroticism could promote healthier behaviors such as exercise and activity engagement, these opportunities are limited for those without sufficient resources. Similarly, neuroticism may fail to promote health for less educated individuals, who may have limited knowledge of best health practices. Support for this claim comes from recent work predicting inflammatory makers (23). In that study, neuroticism proved a risk factor for greater inflammation among low-SES participants, but was related to a reduced risk for inflammation in those higher in SES. In addition, higher neuroticism has been linked to greater longevity among women with higher socio-economic status (24). The current study sought to replicate and extend this interaction by testing whether it holds across different health and health behavior measures.

Differing Facets of Neuroticism

A final possibility reflects another common issue resulting from the movement in health psychology toward broader trait dimensions, namely that these dimensions reflect constellations of lower-order traits, or facets, that may not all work in tandem. Neuroticism is comprised of facets that reflect different aspects of emotional instability, including depression, anxiety, anger or hostility, immoderation, and self-consciousness (25). However, as noted above, the notion of a “healthy neurotic” is largely reflective of the anxiety component of the trait domain (13), as anxiety may encourage vigilance to potential health conditions, and to situations that may prove health-averse. Contrarily, one would not expect depression or hostility to promote health and health behavior, and indeed, both depression (26) and hostility (27) have been linked instead to greater mortality risk. That said, work has shown that hostility and anger, facets typically seen as negative for health, may prove less maladaptive for some groups and in some contexts (28,29); accordingly, though we focus on anxiety as it proves central to the past discussion of healthy neuroticism (13), we investigate all facets to consider the potential for differential influences on health outcomes.

Indeed, this work is in line with a broader literature that has demonstrated the utility of “moving beyond” the Big Five when predicting health outcomes. For instance, regarding neuroticism, work suggests that the impulsivity / immoderation component may be the facet most associated with obesity (30,31), likely due to inappropriate eating habits. However, multiple facets appear negatively linked to self-rated health (32), including anxiety, the facet most associated with “healthy neuroticism.” Though it appears important to consider neuroticism at the facet level, little support exists for the facet predictions aligned with the healthy neuroticism concept.

The current study

While we are not the first study to seek evidence of healthy neuroticism, the current study may have the highest likelihood of identifying such evidence, if it exists, for two reasons. First, the current study contains both adulthood and childhood personality measures. Concurrent assessments of predictors and outcomes often have the largest effect sizes, making them easier to detect. However, in the case of personality and cumulative health outcomes, traits often only have small effects at any single time point present moment, with more easily observed effects accruing over time (4, 33). For example, an adult smokes a single cigarette in response to high levels of neuroticism, having a small impact on one’s health. In contrast, if a teenager chooses to initiate a smoking habit in response to high levels of neuroticism, the resulting sustained behavior may have an enormous impact. In addition, many health habits are set in late adolescence and early adulthood, and these choices are likely influenced by personality at that age. By examining both adulthood and childhood personality, we leveraged both the possibilities of stronger associations due to concurrent assessment and stronger associations due to the accrual of behaviors over time.

Second, the current study utilized multiple measures of health. As with many studies of personality and health, we examined how traits may predict self-rated health and health behavior. Additionally, we predicted major chronic disease status. Finally – and unlike the majority of studies – we employed biomarker data as objective outcomes. By examining a large variety of health indicators, we are able to not only increase the likelihood of identifying healthy neuroticism, but we are also able to better understand the mechanisms involved. For example, if we identify evidence for healthy neuroticism with respect to the clinical biomarkers but not behaviors, it would provide evidence that researchers need to move beyond lifestyle behaviors when explaining the value of healthy neuroticism, such as considering stress reactivity.

Methods

Participants

Participants came from the Hawaii Personality and Health Cohort. This cohort was originally assessed between 1959 and 1967, when teachers in entire elementary-school classrooms on two Hawaiian Islands assessed children (N = 2,418) on their personality traits at the end of the school year, one time for each child. In 1998, efforts to find these children, now middle-aged adults, were begun. When each member of the original cohort was located (i.e., rolling recruitment), he or she was invited to join the study. Since joining, participants have completed one or more of seven questionnaires (Q1–Q7) and, from 2003 onward, have been invited to attend a half-day medical and psychological examination. In addition to the childhood personality assessment, measures used here were drawn from items in Q5 (adult personality; assessed in 2008), Q6 (self-rated health, smoking, alcohol use, and chronic disease status; assessed in 2013), and Q7 (exercise and flossing; assessed in 2015), and the clinic visit (physical health outcomes; assessed at age 50).

To be included in the subsample for this study, participants had to have completed either the Q5 adult personality questionnaire or the childhood personality questionnaire and at least one of these portions of the study: Q6, Q7 or the clinical assessments. This resulted in a subsample of 1,115 participants (51% female). Additional demographic information is presented in Supplementary File 1, Table 5. (Supplementary Files can be found at osf.io/bdgs7.) However, the sample sizes for various analyses differ, given which participants completed which surveys. Specifically, data were collected in a longitudinal study, in which questionnaires were mailed to participants at non-regular intervals across decades. If a participant chose not to respond to a questionnaire – because they moved, were busy, were tired of participating – the result was missingness on multiple questions in a systematic way. We chose to not to limit our sample to only participants with all data, as this would greatly limit our power; we also chose not use to a multiple imputation approach, as missingness was not completely at random. Moreover, we think it is likely that the data are missing not at random, and the use of multiple imputation could exaggerate existing bias and produce results that are not representative of this population. The largest sample size was for regressions of systolic blood pressure on the interaction of childhood neuroticism and conscientiousness (N = 802), and the smallest was for the regression of alcoholic behaviors on the facets of neuroticism (N = 544). We note that even the smallest sample sizes are well-powered, if we can assume the interaction of neuroticism and other variables to be large enough to predict substantial differences between studies. Sample sizes for each regression are listed in Supplementary File 1, Tables 7–22.

Table 5.

The associations of general neuroticism and facets with health variables.

Factor loadings from bifactor structural equation model
Item	General Neuroticism	Anxiety	Anger	Depression	Self-consciousness	Immoderation	Vulnerability
Worry about things	.32	.50
Fear for the worst	.49	.41
Am afraid of many things	.55	.31
Get stressed out easily	.53	.59
Get angry easily	.33		.77
Get irritated easily	.45		.63
Lose my temper	.42		.69
Am not easily annoyed	.26		.26
Often feel blue	.65			.40
Dislike myself	.73			.05
Am often “down in the dumps”	.70			.54
Feel comfortable with myself (–)	.51			−0.02
Find it difficult to approach others	.46				.51
Am afraid to draw attention to myself	.29				.41
Only feel comfortable with friends	.37				.53
Am not bothered by difficult social situations (–)	.16				.12
Go on binges (overeating, gambling, etc.)	.36					.35
Rarely overindulge (–)	.14					.45
Easily resist temptations (–)	.06					.47
Am able to control my cravings (–)	.29					.47
Panic easily	.48						.38
Get overwhelmed by events	.49						.58
Feel that I can’t deal with things	.61						.15
Remain calm under pressure (–)	.35						.40
Coefficient estimates from regression models
Self-rated health	−0.55* [−0.72, −0.38] p < .001	−0.58 [−1.92, 0.77] p = .401	0.01 [−0.25, 0.27] p = .944	0.70 [−0.46, 1.86] p = .239	−0.25 [−0.57, 0.07] p = .129	−0.68* [−1.17, −0.20] p = .005	4.63 [−4.93, 14.19] p = .342
Heart condition	−0.010 [−0.06, 0.05] p = .827	0.02 [−0.28, 0.32] p = .895	0.00 [−0.07, 0.06] p = .890	0.10 [−0.13, 0.33] p = .400	0.05 [−0.03, 0.14] p = .201	0.02 [−0.09, 0.12] p = .753	0.64 [−1.64, 2.92] p = .580
Number chronic illnesses	0.10 [−0.05, 0.24] p = .203	0.3 [−0.52, 1.12] p = .475	0.02 [−0.16, 0.20] p = .819	−0.16 [−0.78, 0.46] p = .616	0.16 [−0.09, 0.40] p = .204	0.48* [0.15, 0.80] p = .004	−1.52 [−7.05, 4.02] p = .591
Ever smoked	0.06 [−0.02, 0.14] p = .145	−0.1 [−0.51, 0.31] p = .632	0.03 [−0.07, 0.12] p = .581	−0.13 [−0.42, 0.16] p = .369	−0.02 [−0.14, 0.11] p = .811	0.14 [−0.02, 0.30] p = .082	−1.24 [−5.02, 2.54] p = .520
Number cigarettes	0.18* [0.04, 0.32] p = .013	−0.25 [−0.97, 0.47] p = .495	0.08 [−0.07, 0.24] p = .307	0.04 [−0.49, 0.58] p = .870	−0.05 [−0.26, 0.16] p = .614	−0.01 [−0.28, 0.26] p = .923	−1.02 [−6.49, 4.45] p = .715
Ever drank	0.31 [−0.34, 0.96] p = .351	2.59 [−4.05, 9.23] p = .445	0.02 [−0.41, 0.46] p = .914	0.72 [−0.72, 2.16] p = .328	1.44 [−1.65, 4.53] p =.360	2.11 [−2.49, 6.7] p = .368	−4.12 [−13.64, 5.40] p = .396
Alcohol (days per week)	−0.08 [−0.4, 0.24] p = .609	−0.41 [−2.34, 1.52] p = .678	0.18 [−0.26, 0.63] p = .423	0.42 [−1.18, 2.02] p = .607	0.05 [−0.44, 0.54] p = .834	0.05 [−0.66, 0.75] p = .898	3.99 [−9.01, 16.99] p = .547
Alcohol (drinks per sitting)	0.14 [−0.11, 0.38] p = .286	−1.05 [−2.9, 0.80] p = .267	0.17 [−0.21, 0.54] p = .385	0.49 [−1.05, 2.03] p = .533	−0.18 [−0.61, 0.25] p = .412	0.28 [−0.35, 0.91] p = .388	4.84 [−8.07, 17.76] p = .462
Alcohol (binge per week)	0.24 [0.00, 0.47] p = .046	−1.30 [−3.41, 0.82] p = .229	0.17 [−0.25, 0.58] p = .433	0.66 [−1.16, 2.48] p = .478	−0.16 [−0.60, 0.29] p = .496	0.37 [−0.30, 1.04] p = .281	6.61 [−8.91, 22.13] p = .404
Exercise (strenuous)	−0.28 [−0.61, 0.05] p = .098	−0.33 [−2.03, 1.37] p = .704	0.09 [−0.14, 0.33] p = .425	−0.33 [−0.73, 0.06] p = .100	−0.25 [−0.54, 0.05] p = .098	−0.36 [−0.8, 0.08] p = .107	0.63 [−1.39, 2.64] p = .542
Exercise (moderate)	−0.02 [−0.3, 0.26] p = .874	0.80 [−1.17, 2.78] p = .426	0.01 [−0.23, 0.25] p = .925	−0.42* [−0.82, −0.03] p = .035	−0.05 [−0.33, 0.24] p = .739	−0.44 [−0.89, 0.01] p = .057	−0.90 [−3.28, 1.48] p = .457
Exercise (mild)	−0.10 [−0.38, 0.18] p = .488	0.41 [−1.00, 1.83] p = .567	0.01 [−0.19, 0.21] p = .926	–0.33 [−0.67, 0.00] p = .048	−0.11 [−0.35, 0.13] p = .390	−0.41* [−0.79, −0.03] p = .036	−0.28 [−1.95, 1.40] p = .747
Flossing	0.05 [−0.23, 0.33] p = .720	−0.15 [−1.55, 1.26] p = .839	−0.08 [−0.28, 0.11] p = .403	0.29 [−0.05, 0.63] p = .098	0.05 [−0.19, 0.29] p = .685	0.37* [0.01, 0.73] p = .042	0.29 [−1.4, 1.97] p = .739
Blood glucose levels	0.26 [−0.03, 0.54] p = .080	1.20 [−0.81, 3.22] p = .242	−0.03 [−0.24, 0.17] p =.745	−0.71 [−1.80, 0.38] p = .199	0.00 [−0.36, 0.36] p = .985	0.36 [−0.15, 0.87] p = .163	−1.55 [−3.97, 0.88] p = .211
Cholesterol/HDL ratio	−0.03 [−0.32, 0.26] p = .833	1.03 [−0.67, 2.73] p = .236	−0.05 [−0.25, 0.15] p = .631	−0.26 [−1.12, 0.61] p = .559	0.00 [−0.30, 0.31] p = .988	0.60* [0.08, 1.11] p = .025	−1.36 [−3.36, 0.64] p = .182
Systolic blood pressure	0.05 [−0.20, 0.30] p = .707	0.20 [−0.84, 1.25] p = .702	0.08 [−0.06, 0.22] p = .273	−0.40 [−1.02, 0.22] p = .210	0.01 [−0.23, 0.24] p = .957	0.32 [−0.04, 0.68] p = .957	−0.64 [−1.98, 0.70] p = .082
Waist to hip ratio	0.34* [0.09, 0.59] p = .008	−0.01 [−0.97, 0.96] p = .990	−0.01 [−0.17, 0.15] p = .927	0.35 [−0.20, 0.90] p = .209	0.24 [−0.05, 0.53] p =.105	0.99* [0.43, 1.55] p = .001	−0.13 [−1.33, 1.07] p =.834

Regression coefficients and 95% confidence intervals are presented. All models control for gender and age.

^*indicate p < .05.

Attrition analyses suggested that participants who chose not to fill out certain questionnaires differed in important ways from those who did. Participants who did not complete the adult personality questionnaire were more likely to be female (54% versus 47%, t(584) = −2.13, p = .034) and less educated (M_participated = 7.02, M_withdrew = 6.52, t(408) = 3.66, p < .001). Participants without health data from Q6 (self-rated health, smoking, alcohol use, and chronic disease status) were less educated (M_participated = 7.09, M_withdrew = 6.52, t(654) = 4.71, p < .001), less conscientious as children (M_participated =0.05, M_withdrew = −0.08, t(830) = 2.00, p = .046), and had parents with lower SES (M_participated = 0.05, M_withdrew = −0.12, t(417) = 2.13, p = .034). Participants without health data from Q7 (exercise and flossing) were less educated (M_participated = 6.99, M_withdrew = 6.51, t(330) = 3.37, p = .001) and had parents with lower SES (M_participated = 0.05, M_withdrew = −0.21, t(234) = 2.89, p = .004). Participants who did not have biomarker data did not differ on any variables of interest from those who did. At this time, researchers at the Oregon Research Institute have archived the child data from the Hawaii and Personality Cohort, and portions of the adult follow-up assessments (14). Additional waves of data will be added to these as resources become available.

Measures

Adult Neuroticism.

In 2008, participants completed the 120-item IPIP-NEO from the International Personality Item Pool (34, 35), in which participants are asked to rate how well a statement describes them on a scale from 1 (Not at all) to 5 (Very much). Neuroticism was measured using 24 statements, including “I worry about things” and “I become overwhelmed by events” (α = .88). Trait levels of neuroticism were measured by averaging responses to these items (M = 2.57, SD = 0.55). Neuroticism was standardized for these analyses.

Facets of adult neuroticism.

Six facets of neuroticism can be estimated using these items: anxiety, anger, depression, self-consciousness, immoderation and vulnerability. Each of these was measured using four of the 24 items in the neuroticism scale. Facets were estimated using a bifactor approach, described below.

Childhood neuroticism.

For the children in the present sample, teachers assessed their entire elementary school classrooms (Grades 1, 2, 5, or 6; M age 10 years) by rank-ordering children on each of a comprehensive set of 43–49 personality attributes, using a 7 or 9-step quasi-normal distribution. Definitions developed by teachers for each attribute were provided. These were converted to standardized scores, and orthogonal factor scores for the Big Five were derived for within each subsample of children using all available items in each subsample (34). The average reliability of the neuroticism factor scores across subsamples was estimated at α = .68 (35). Previous analyses of the childhood and adulthood assessments of personality in this sample indicate that the neuroticism measures are unassociated with each other (35).

Self-Rated Health.

Participants rated their own general health, compared to others of their same age and sex, on a scale from 1 (Poor) to 5 (Excellent; M = 3.36, SD = 1.00 (36).

Health Behaviors.

In 2013, participants reported whether they had (1) or had not (0) ever smoked cigarettes (48% said yes). Current smokers reported the number of cigarettes they typically smoked per day and non-smokers were assigned a value of 0 for this variable (M = 0.34, SD = 0.91).

Participants also reported whether they had (1) or had not (0) consumed alcohol at any point in their lives (86% said yes). Current drinkers reported on how many days of the week they drank alcohol (M = 1.68, SD = 1.97), how many drinks they consumed when they drank (M = 1.29, SD = 1.62), and how many days a week they engaged in binge drinking (defined as five or more drinks in one sitting; M = 0.58, SD = 1.49), and non-drinkers were assigned a value of 0 for each of these variables.

In 2015, participants reported on how frequently they engaged in strenuous exercise (defined as feeling your heart beat rapidly), moderate exercise (defined as not exhausting) and mild exercise (defined as minimal effort) (37). Responses were coded as zero times (1), one or two times (2), three or four times (3), five or six times (4) or seven or more times a week (5). On average participants engaged in strenuous exercise 1.98 times a week (SD = 1.19), moderate exercise 2.69 times a week (SD = 1.28), and mild exercise 2.93 times a week (SD = 1.34). Participants also reported on how frequently they flossed: never (0), sometimes (1) or always (2; M = 1.59, SD = 0.80).

Chronic disease status.

In 2013, participants reported whether they had ever been diagnosed with the following chronic diseases or health events: heart attack, heart disease, stroke, thyroid disease, migraine, cancer, chronic fatigue syndrome, Type I diabetes, Type II diabetes. The total number of diseases reported was a primary outcome (M = 0.71, SD = 0.98). In addition, we examined whether a participant had been diagnosed with any heart condition (heart attack or heart disease; 13.4%). This condition was chosen for closer consideration due to its relatively large base rate, compared to the other conditions, and the substantial impact it has on American mortality rates.

Objectively assessed health.

Starting in 2003 (M_age = 51.00, SD = 2.93), participants were invited to a half day clinic assessment which included the collection of urine and blood samples that were tested for various biomarkers (38). The following biomarkers were used as individual outcomes: fasting blood glucose (M = 100.68, SD = 26.14), total cholesterol/HDL ratio (M = 4.29, SD = 1.53), systolic blood pressure (M = 122.39, SD = 15.72), and waist-to-hip ratio (M = 0.88, SD = 0.09). These outcomes were standardized. We control for participants’ age at the clinic visit in analyses.

Moderators.

Adult conscientiousness was measured in 2008 using 24 statements from the IPIP-NEO, including “I complete tasks successfully” and “I am always prepared” (α = .88; M = 4.01, SD = 0.46). Neuroticism and adult conscientiousness were negatively correlated (r = −.51). Typically, it is advised that variables be uncorrelated in moderation analyses; however, this interaction was examined in prior studies of healthy neuroticism (15,18) despite correlations between the variables. Childhood conscientiousness was assessed using the same procedures described above for childhood neuroticism (average α = .77) (35).

Educational attainment was used as a proxy for adult socio-economic status (SES). Education was assessed on a scale from 1 (8^th grade education or less) to 9 (professional or postgraduate degrees). The current sample was well educated (M= 6.89, SD = 1.81) All participants had completed at least junior high. Childhood SES was a linear combination of father’s education, mother’s education (both assessed on a scale from 1 [8^th grade or less] to 9 [professional or postgraduate degree]) and whether the parents owned a home before the participant had entered school (1 – Yes, 0 – No). Each variable was standardized, and then the average of the three standardized variables was used as the participant’s score.

Chronic disease was a binary variable defined as whether a participant had been diagnosed with one of the chronic diseases listed above (47%). We acknowledge that this variable encompasses a wide range of health problems, including heart disease, diabetes, and chronic fatigue. However, we wanted to use a variable that broadly represented having major health problems – as those studies which find protective effects of neuroticism represent broad populations of older adults – and because we would expect any major health problem to incite anxiety in a person high in neuroticism. When using this moderator, we did not include number of chronic conditions or heart disease status as outcomes, given these are too theoretically and numerically similar to the moderator.

Analyses

All analyses were performed in R (version 3.4.2) (39). We used the pscl package (40) to estimate the zero-inflated Poisson models and the lavaan package (41) to estimate the bifactor models. For all analyses, we included every participant who had provided the data on the relevant constructs. Because of the longitudinal nature of the study, not every participant had data at every wave. In some cases, such as for the adult personality assessment or self-report, some participants were simply unavailable during data collection, In the case of the clinic data, data collection was targeted at about age 55, as opposed to a specific year, so some participants were not yet eligible for participation. For transparency, we reported the sample size available for the models run (see Supplementary File 1, Tables 7–22).

To move beyond the simple significant-nonsignificant binary and better describe the evidentiary value of these results, we have presented the exact p-values for all coefficients of interest in our regression results. In other words, we presented the exact p-values for the multiple regressions with neuroticism or childhood neuroticism as predictors, the exact p-values for the interaction terms in the moderation models, and the exact p-values for the regression results in the bifactor models. By presenting these p-values, a discerning reader will be able to adequately assess the evidentiary value of any significant results, as well as examine the distribution of results across all analyses.

Question 1: To what extent is neuroticism associated with health (i.e., self-rated health, health outcomes, health behaviors and clinical indicators of health)?

To assess the extent to which neuroticism is related to health, we examined both bi-variate correlations and coefficient estimates from generalized linear models. For each outcome, we estimate a model with adult neuroticism – because concurrent measurements are likely to yield the largest effect sizes – and a separate model with childhood neuroticism – which may have cumulative influences on health accruing over long spans of time. Generalized linear models controlled for gender and age at the clinic assessment. For outcomes that were normally distributed (i.e., self-rated health, moderate and mild exercise, flossing, clinical indicators), linear models were estimated using ordinary least squares. For binary outcomes (i.e., heart conditions, ever smoking, and ever drinking alcohol), binary logistic regression were used. For most count outcomes (number of cigarettes smoked, number of drinks per day, days drinking per week, and number of days of binge drinking per week), absence of behavior was over-represented, so we used zero-inflated Poisson regression models. As a reminder, these models estimated the relationship between the predictors and both the probability of the outcome (e.g., probability the outcome is greater than or equal to 1) and also the relationship of the predictors to the count for cases when the outcome is greater than 0. We presented the results of both models. For strenuous exercise, we used a generalized linear model with a negative binomial distribution were used. For number of chronic diseases, we used a Poisson distribution, because models using the negative binomial failed to converge in a few instances.

Question 2: Does conscientiousness moderate the association of neuroticism with health?

Next, we examined the extent to which conscientiousness moderated the association of neuroticism on health. To do so, we added conscientiousness and the interaction of conscientiousness and neuroticism to each of the generalized linear models described above. When adult neuroticism was the predictor, adult conscientiousness was chosen as the moderator; when childhood neuroticism was the predictor, childhood conscientiousness was the moderator. We used a broad interpretation of healthy neuroticism: in these models, a significant interaction term in any of these models was interpreted as an example of “healthy neuroticism.” This encompassed both instances when neuroticism had no association with health and instances when neuroticism was protective.

Question 3: Does SES moderate the association of neuroticism with health?

We examined the extent to which SES moderated the association of neuroticism with health. Here we added SES and the interaction of SES and neuroticism to the models described in Question 1. When adult neuroticism was the predictor, adult education was chosen as the moderator. When childhood neuroticism was the predictor, childhood SES was used as the moderator. If the interaction term in any of these models is significant, it would suggest that SES is a “healthy neuroticism” moderator.

Question 4: Does chronic disease moderate the relationship between neuroticism and health?

Here we examined whether having been diagnosed with a chronic disease status moderates the association of neuroticism with health. We added the binary disease status variable and the interaction of this variable with adult neuroticism to the models described in Question 1. We also chose not to examine number of chronic diseases or likelihood of having a heart condition as outcomes, given they were substantially overlapping with the moderating variable. If the interaction term in any of these models is significant, it would suggest that chronic disease status is a “healthy neuroticism” moderator.

Question 5: Do facets of neuroticism predict health, and if so, do they predict health in opposing directions?

To determine whether facets of neuroticism had differing effects on health, we used a bifactor model approach. In this model, all items were set to load on general trait neuroticism. Simultaneously, each item was set to load on its corresponding facet. We constrained the covariance between general neuroticism and each of the facets to be 0. This approach has been used in other studies of personality facets and outcomes (38,42,43), and allowed us to specify facet constructs that are “pure” of the shared variance across facets, and in turn provided a better opportunity to test whether the facets hold unique associations with the outcomes of interest. Fit for this model was very good (RMSEA = .04, CFI = .952). However, some loadings of items onto their assigned facets were poor. For each outcome, we used this model and added a set of regression pathways by regression the health outcome onto latent general neuroticism and each latent facet simultaneously. If any of the pathways from personality (neuroticism and its facets) to health are positive (i.e., indicating greater levels of a facet or trait are linked to better health), it would suggest that some facets of neuroticism are “healthy”.

Results

Question 1: To what extent is neuroticism associated with health (i.e., self-rated health, health outcomes, health behaviors and clinical indicators of health)?

First, we examined the bivariate correlations of adult neuroticism and childhood neuroticism with health outcomes, behaviors and indicators, as shown in Table 1. Sample sizes ranged from 558 (outcome: ever drink alcohol) to 802 (systolic blood pressure). Adult neuroticism was associated with worse self-rated health (r = −.37, p < .001), increased likelihood of having been diagnosed with a heart condition (r = .08, p = .033), more chronic diseases (r = .15, p < .001), more cigarettes smoked (r = .09, p =.016), less strenuous exercise (r = −.10, p = .014), and larger waist-to-hip ratios (r = .08, p =.038). Childhood neuroticism was not associated with any of the health variables. Next, we used linear models to estimate the relationship of adult neuroticism and childhood neuroticism to health, controlling for gender, birth year and highest level of education. These results are shown on the right side of Table 1. Adult neuroticism was associated with self-rated health (b = −0.59, 95% CI [−0.73, −.46], p < .001), number of chronic diseases (b = 0.25 [ 0.07, 0.42], p = .005), more cigarettes smoked (b = 0.61 [0.07, 1.18], p = .030), greater likelihood of consuming alcohol (b = 0.44 [0.02, 0.89], p = .044), larger amounts of binge drinking (b = 0.53 [0.05, 1.02], p = .030), and waist-to-hip ratio (b = 0.17 [0.06, 0.28], p = .003).

Table 1.

The zero-order and modeled associations of neuroticism and childhood neuroticism with health variables.

	Zero-order correlations (r estimate)		Linear models controlling for covariates (b estimate)
	Adult Neuroticism	Child Neuroticism	Adult Neuroticism		Child Neuroticism
			Est [95% CI]	p-value	Est [95% CI]	p-value
Self-rated health	−.37* [−.44, −.31]	−.06 [−.11, .02]	−0.59* [−0.73,−0.46]	<.001	−0.08* [−0.16,−0.01]	.026
Heart condition	.08* [.00, .16]	−.03 [−.11, .03]	0.21 [−0.25, 0.67]	.362	−0.01 [−0.24, 0.22]	.954
Number chronic diseases	.15* [.07, .22]	.02 [−.04, .08]	0.25* [0.07, 0.42]	.005	0.07 [−0.03, 0.17]	.110
Cigarettes ever	.07 [.00, .15]	.02 [−.06, .10]	0.25 [−0.05, 0.55]	.100	0.07 [−0.08, 0.22]	.350
Cigarettes (count model)	.09* [.02, .16]	−.01 [−.07, .06]	0.11 [−0.22, 0.45]	.498	−0.08 [−0.24, 0.08]	.334
Cigarettes (zero model)	–	–	−0.61* [−1.08,−0.13]	.013	−0.12 [−0.35, 0.11]	.307
Alcohol ever	.05 [−.04, .13]	−.07 [−.14, .00]	0.45* [0.02, 0.89]	.044	−0.10 [−0.31, 0.11]	.333
Alcohol days/week (count model)	−.03 [−.11, .05]	−.02 [−.08, .06]	0.03 [−0.08, 0.15]	.570	0.01 [−0.05, 0.07]	.825
Alcohol days/week (zero model)	–	–	0.13 [−0.02, 0.46]	.442	0.05 [−0.12, 0.21]	.585
Alcohol drinks/sitting (count model)	.01 [−.08, .09]	−.05 [−.11, .01]	0.17* [0.02, 0.32]	.023	0.03 [−0.05, 0.11]	.451
Alcohol drinks/sitting (zero model)	–	–	0.29 [−0.16, 0.74]	.209	0.08 [−0.14, 0.29]	.476
Alcohol binge days/week (count model)	.05 [−.03, .12]	−.02 [−.08, .04]	0.24* [0.02, 0.46]	.032	−0.05 [−0.16, 0.06]	.370
Alcohol binge days/week (zero model)	–	–	−0.24 [−0.67, 0.18]	.257	−0.11 [−0.32, 0.10]	.300
Strenuous exercise	−.10* [−.17,−.03]	−.04 [−.12, .04]	−0.11 [−0.22, 0.00]	.052	−0.01 [−0.06, 0.04]	.669
Moderate exercise	−.05 [−.11, .02]	−.03 [−.11, .06]	−0.09 [−0.24, 0.05]	.206	−0.03 [−0.10, 0.05]	.487
Mild exercise	−.06 [−.12, .02]	.00 [−.06, .07]	−0.10 [−0.25, 0.05]	.179	0.00 [−0.07, 0.07]	.944
Flossing	.06 [−.01, .13]	−.05 [−.12, .02]	0.11 [−0.03, 0.25]	.138	−0.03 [−0.10, 0.05]	.491
Blood glucose	.08 [−.01, .16]	−.03 [−.09, .03]	0.14 [0.00, 0.27]	.053	−0.02 [−0.09, 0.05]	.565
Cholesterol/HDL ratio	.01 [−.05, .08]	−.04 [−.1, .03]	0.02 [−0.13, 0.17]	.769	−0.01 [−0.09, 0.06]	.690
Systolic blood pressure	.01 [−.07, .09]	.03 [−.03, .10]	0.02 [−0.12, 0.16]	.772	0.05 [−0.02, 0.12]	.170
Waist-to-Hip ratio	.08* [.01, .16]	−.07 [−.13, .00]	0.17* [0.06, 0.28]	.003	0.01 [−0.04, 0.06]	.743

Zero-order correlations are presented in the first two columns with bootstrapped 95% confidence intervals. Modeled associations use ordinary least squares, binary logistic regression or negative binomial regression, depending upon the outcome. Unstandardized regression coefficients and 95% confidence intervals are presented. All models control for gender and age at the time of outcome measurement.

^*indicates p < .05. Exact p-values for the regression models are presented.

Question 2: Does conscientiousness moderate the association of neuroticism with health?

In only one of the nineteen models did adult conscientiousness moderate the association of adult neuroticism with health. The results of all models are shown in Table 2. In this model, as adult conscientiousness increased, adult neuroticism had a stronger, positive relationship with binge drinking (b = 1.26 [0.18, 2.53], p = .022). In only one of the nineteen models did childhood conscientiousness moderate the association of childhood neuroticism with adult health. In this model, higher childhood conscientiousness strengthened the relationship between higher childhood neuroticism and greater waist-to-hip ratios. (b = 0.01 [0.001, 0.012], p = .025). Sample sizes ranged from 558 (ever smoked cigarettes) to 802 (systolic blood pressure). To conserve space in the current paper, and to avoid confusing readers, we do not provide estimates of slopes at different levels of moderators here. However, we have plotted the simple slopes at one standard deviation above and below the means, and these plots are available in the Supplementary Files 1 and 3.

Table 2.

Conscientiousness as a moderator of neuroticism and childhood neuroticism.

Outcome	Adulthood predictors				Childhood predictors
	Neur	Con	Neur x Con		Neur	Con	Neur x Con
	Est [95% CI]	Est [95% CI]	Est [95% CI]	p	Est [95% CI]	Est [95% CI]	Est [95% CI]	p
Self-rated health	−0.54 [−0.70, −0.38]	0.13 [−0.06, 0.32]	0.11 [−0.17, 0.38]	.451	−0.08 [−0.16, −0.01]	0.03 [−0.04, 0.11]	−0.01 [−0.08, 0.06]	.865
Heart condition	0.23 [−0.31, 0.75]	0.08 [−0.55, 0.72]	0.80 [−0.22, 1.95]	.149	−0.01 [−0.24, 0.22]	0.05 [−0.19, 0.30]	0.10 [−0.12, 0.31]	.353
Number chronic illnesses	0.23 [0.02, 0.43]	−0.05 [−0.29, 0.20]	−0.02 [−0.37, 0.35]	.927	0.07 [−0.02, 0.16]	−0.07 [−0.16, 0.03]	0.02 [−0.06, 0.11]	.619
Ever smoked	0.17 [−0.17, 0.52]	−0.18 [−0.60, 0.23]	0.04 [−0.57, 0.65]	.899	0.08 [−0.07, 0.23]	−0.32 [−0.49, −0.16]	−0.02 [−0.17, 0.12]	.751
Number cigarettes (count model)	0.05 [−0.35, 0.46]	−0.05 [−0.56, 0.45]	0.57 [−0.31, 1.44]	.203	−0.10 [−0.27, 0.07]	−0.02 [−0.17, 0.14]	−0.08 [−0.23, 0.08]	.323
Number cigarettes (zero model)	−0.61 [−1.16, −0.05]	0.06 [−0.62, 0.75]	0.22 [−0.90, 1.34]	.695	−0.13 [−0.37, 0.11]	0.23 [−0.01, 0.47]	0.01 [−0.22, 0.24]	.916
Ever drank	0.43 [−0.07, 0.96]	0.05 [−0.56, 0.66]	0.42 [−0.45, 1.22]	.328	−0.09 [−0.30, 0.12]	0.06 [−0.18, 0.28]	−0.14 [−0.33, 0.05]	.143
Alcohol (days per week) (count model)	−0.04 [−0.18, 010]	−0.17 [−0.35, 0.01]	0.20 [−0.07, 0.46]	.147	−0.01 [−0.05, 0.07]	−0.03 [−0.09, 0.04]	−0.04 [−0.10, 0.02]	.212
Alcohol (days per week) (zero model)	0.04 [−0.34, 0.43]	−0.24 [−0.72, 0.23]	0.38 [−0.36, 1.11]	.315	0.04 [−0.13, 0.20]	−0.01 [−0.19, 0.17]	0.06 [−0.10, 0.22]	.452
Alcohol (drinks per sitting) (count model)	0.12 [−0.06, 0.30]	−0.17 [−0.40, 0.06]	0.26 [−0.09, 0.61]	.141	0.03 [−0.05, 0.10]	−0.07 [−0.16, 0.01]	−0.06 [−0.14, 0.02]	.115
Alcohol (drinks per sitting) (zero model)	−0.04 [−0.18, 0.10]	−0.17 [−0.35, 0.01]	0.20 [−0.07, 0.46]	.119	0.01 [−0.05, 0.07]	−0.03 [−0.09, 0.04]	−0.04 [−0.10, 0.02]	.815
Alcohol (binge per week) (count model)	0.26 [0.02, 0.50]	−0.28 [−0.61, 0.05]	1.51* [0.73, 2.29]	< .001	−0.09 [−0.20, 0.03]	−0.11 [−0.22, 0.01]	−0.11 [−0.22, 0.01]	.072
Alcohol (binge per week) (zero model)	−0.09 [−0.60, 0.42]	0.15 [−0.45, 0.75]	0.27 [−0.87, 1.41]	.643	−0.12 [−0.33, 0.09]	0.12 [−0.10, 0.34]	0.01 [−0.11, 0.31]	.358
Exercise (strenuous)	−0.05 [−0.18, 0.07]	0.11 [−0.04, 0.27]	−0.04 [−0.23, 0.17]	.700	−0.01 [−0.06, 0.04]	−0.02 [−0.07, 0.04]	0.02 [−0.03, 0.07]	.403
Exercise (moderate)	−0.09 [−0.27, 0.08]	0.00 [−0.20, 0.21]	0.12 [−0.15, 0.39]	.389	−0.03 [−0.10, 0.05]	−0.05 [−0.13, 0.02]	0.01 [−0.06, 0.07]	.856
Exercise (mild)	−0.01 [−0.19, 0.16]	0.19 [−0.02, 0.39]	0.08 [−0.20, 0.35]	.584	0.00 [−0.07, 0.07]	−0.06 [−0.13, 0.02]	−0.01 [−0.07, 0.06]	.859
Flossing	0.05 [−0.12, 0.22]	−0.11 [−0.31, 0.09]	−0.16 [−0.42, 0.11]	.254	−0.02 [−0.10, 0.05]	−0.05 [−0.12, 0.03]	−0.02 [−0.09, 0.04]	.527
Blood glucose levels	0.08 [−0.08, 0.24]	−0.12 [−0.32, 0.07]	−0.04 [−0.32, 0.24]	.785	−0.02 [−0.10, 0.05]	−0.08 [−0.16, −0.01]	0.00 [−0.07, 0.07]	.919
Cholesterol/HDL ratio	0.00 [−0.18, 0.17]	−0.06 [−0.28, 0.15]	0.09 [−0.23, 0.40]	.587	−0.02 [−0.09, 0.05]	−0.11 [−0.18, −0.04]	0.02 [−0.04, 0.09]	.521
Systolic blood pressure	0.10 [−0.06, 0.26]	0.16 [−0.03, 0.36]	0.24 [−0.04, 0.52]	.087	0.05 [−0.02, 0.12]	−0.03 [−0.10, 0.04]	−0.01 [−0.07, 0.06]	.778
Waist to hip ratio	0.15 [0.01, 0.28]	−0.05 [−0.21, 0.11]	0.01 [−0.21, 0.24]	.901	0.01 [−0.05, 0.06]	−0.10 [−0.15, −0.04]	0.06* [0.01, 0.10]	.025

Neuroticism, childhood neuroticism and moderators are all standardized, allowing for easier interpretation of the simple slopes. Regression coefficients (standardized when the outcome is continuous and the model is estimated using ordinary least squares) and 95% confidence intervals are presented. All models control for gender and age at outcome measurement.

^*indicates p < .05. Stars are only used indicate significant interaction terms; we do not interpret significance of the simple slopes (e.g., the coefficients for neuroticism and conscientiousness) because they are dependent upon the level of the other predictor and only included to allow for better interpretation of the interaction. Exact p-values for the interaction terms are presented to allow for estimation of the evidentiary value of these findings.

Question 3: Does SES moderate the association of neuroticism with health?

In none of nineteen models did adult SES moderate the association of adult neuroticism with health. In only two of nineteen models did childhood SES moderate the association of childhood neuroticism with adult health. Specifically, this interaction was associated with the binary portion of the alcoholic beverages per sitting model (b = 0.24, OR = 1.27, 95% CI [1.01, 1.60], p = .040) and the count model of number of times per month the participant engaged in binge drinking (b = 0.13, 95% CI [0.00, 0.26], p = .045). We note that both of these p-values are between .04 and .05 and thus hold little evidentiary value. In both cases, the relationship between childhood neuroticism and alcohol consumption was stronger at higher levels of childhood SES. Sample sizes for the adult predictor models ranged from 562 (alcoholic drinks per sitting) to 593 (flossing). Sample sizes for the childhood predictor models ranged from 636 (cholesterol/HDL ratio) to 693 (flossing). Results for all models are shown in Table 3.

Table 3.

Socio-economic status as a moderator of neuroticism and childhood neuroticism.

Outcome	Adulthood predictors				Childhood predictors
	Neuroticism	SES	N x SES		Neuroticism	SES	N x SES
	Est [95% CI]	Est [95% CI]	Est [95% CI]	p	Est [95% CI]	Est [95% CI]	Est [95% CI]	p
Self-rated health	−0.55 [−0.69, −0.42]	0.22 [0.13, 0.3]	0.00 [−0.15, 0.14]	.958	−0.09 [−0.16, −0.02]	0.29 [0.19, 0.38]	−0.05 [−0.13, 0.04]	.306
Heart condition	0.23 [−0.25, 0.71]	−0.01 [−0.28, 0.29]	−0.05 [−0.55, 0.47]	.860	0.00 [−0.24, 0.23]	−0.31 [−0.61, −0.01]	−0.08 [−0.37, 0.21]	.593
Number chronic illnesses	0.24 [0.06, 0.42]	−0.13 [−0.23, −0.02]	−0.17 [−0.35, 0.01]	.110	0.09 [−0.01, 0.18]	−0.16 [−0.28, −0.05]	0.04 [−0.07, 0.15]	.489
Ever smoked	0.19 [−0.13, 0.51]	−0.54 [−0.74, −0.34]	−0.26 [−0.64, 0.10]	.167	0.08 [−0.07, 0.24]	−0.23 [−0.43, −0.04]	0.16 [−0.02, 0.36]	.090
Number cigarettes (count model)	0.10 [−0.27, 0.47]	−0.12 [−0.35, 0.12]	0.10 [−0.31, 0.50]	.637	−0.08 [−0.25, 0.09]	0.04 [−0.16, 0.24]	0.03 [−0.12, 0.17]	.720
Number cigarettes (zero model)	−0.44 [−0.96, 0.07]	0.58 [0.29, 0.87]	0.36 [−0.17, 0.89]	.181	−0.22 [−0.46, 0.03]	0.51 [0.24, 0.78]	−0.23 [−0.48, 0.02]	.068
Ever drank	0.40 [−0.05, 0.87]	−0.21 [−0.51, 0.06]	−0.04 [−0.57, 0.46]	.881	−0.09 [−0.31, 0.12]	−0.15 [−0.42, 0.12]	−0.08 [−0.34, 0.18]	.537
Alcohol (days per week) (count model)	0.06 [−0.07, 0.20]	0.11 [0.02, 0.19]	0.00 [−0.17, 0.16]	.968	0.01 [−0.05, .08]	0.11 [0.04, 0.18]	0.00 [−0.06, 0.07]	.946
Alcohol (days per week) (zero model)	0.08 [−0.26, 0.42]	−0.14 [−0.35, 0.07]	0.05 [−0.34, 0.43]	.809	0.06 [−0.11, 0.23]	−0.19 [−0.37, −0.01]	0.15 [−0.03, 0.33]	.092
Alcohol (drinks per sitting)(count model)	0.13 [−0.03, 0.29]	−0.18 [−0.28, −0.09]	−0.09 [−0.27, 0.08]	.293	0.01 [−0.07, 0.08]	−0.10 [−0.19, −0.01]	0.02 [−005, 0.12]	.443
Alcohol (drinks per sitting) (zero model)	0.16 [−0.31, 0.62]	−0.37 [−0.63, −0.11]	−0.10 [−0.57, 0.37]	.671	0.09 [−0.13, 0.31]	−0.37 [−0.63, −0.11]	0.24* [0.01, 0.47]	.040
Alcohol (binge per week) (count model)	0.26 [0.04, 0.49]	0.03 [−0.11, 0.16]	0.01 [−0.23, 0.26]	.924	−0.04 [−0.15, 0.08]	0.01 [−0.12, 0.13]	0.13* [0.00, 0.26]	.045
Alcohol (binge per week) (zero model)	−0.12 [−0.55, 0.32]	0.38 [0.13, 0.62]	−0.01 [−0.46, 0.44]	.963	−0.09 [−0.30, 0.12]	0.12 [−0.11, 0.35]	0.13 [−0.10, 0.35]	.269
Exercise (strenuous)	−0.11 [−0.22, 0.00]	0.00 [−0.06, 0.07]	−0.01 [−0.13, 0.11]	.892	0.00 [−0.05, 0.05]	0.00 [−0.06, 0.07]	−0.06 [−0.12, 0.00]	.070
Exercise (moderate)	−0.11 [−0.27, 0.04]	0.01 [−0.08, 0.09]	0.13 [−0.04, 0.29]	.127	−0.01 [−0.09, 0.06]	0.07 [−0.02, 0.16]	−0.06 [−0.15, 0.03]	.180
Exercise (mild)	−0.12 [−0.28, 0.03]	0.00 [−0.08, 0.09]	0.13 [−0.03, 0.29]	.119	0.01 [−0.06, 0.08]	0.08 [−0.01, 0.17]	−0.04 [−0.13, 0.05]	.388
Flossing	0.09 [−0.06, 0.23]	−0.10* [−0.18, −0.01]	−0.04 [−0.20, 0.11]	.573	−0.02 [−0.09, 0.05]	−0.13* [−0.22, −0.04]	0.07 [−0.02, 0.15]	.147
Blood glucose levels	0.10 [−0.04, 0.25]	−0.13* [−0.21, −0.05]	−0.05 [−0.19, 0.10]	.539	−0.02 [−0.10, 0.05]	−0.15* [−0.24, −0.06]	0.04 [−0.06, 0.13]	.450
Cholesterol/HDL ratio	−0.02 [−0.18, 0.14]	−0.13* [−0.22, −0.04]	0.04 [−0.12, 0.21]	.597	−0.02 [−0.10, 0.06]	−0.05 [−0.14, 0.04]	0.00 [−0.09, 0.09]	.993
Systolic blood pressure	−0.02 [−0.17, 0.12]	−0.12* [−0.20, −0.04]	0.13 [−0.02, 0.28]	.101	0.03 [−0.04, 0.11]	−0.09* [−0.17, 0.00]	0.06 [−0.03, 0.14]	.207
Waist to hip ratio	0.13* [0.01, 0.25]	−0.14* [−0.21, −0.08]	0.05 [−0.07, 0.17]	.393	0.00 [−0.06, 0.06]	−0.11* [−0.18, −0.05]	0.04 [−0.03, 0.11]	.229

Neuroticism, childhood neuroticism and moderators are all standardized, allowing for easier interpretation of the simple slopes. Regression coefficients (standardized when the outcome is continuous and the model is estimated using ordinary least squares) and 95% confidence intervals are presented. All models control for gender and age at outcome measurement.

^*indicates p < .05. Stars are only used indicate significant interaction terms; we do not interpret significance of the simple slopes (e.g., the coefficients for neuroticism and SES) because they are dependent upon the level of the other predictor and only included to allow for better interpretation of the interaction. Exact p-values for the interaction terms are presented to allow for estimation of the evidentiary value of these findings.

Question 4: Does chronic disease moderate the relationship between neuroticism and health?

Chronic disease status was indicated by a binary variable, 0 (no chronic diseases) or 1 (one or more chronic diseases). Results are shown in Table 4. We did not examine the health outcomes of heart condition or number of chronic diseases, as these were confounded with the moderator. In only two of the sixteen models did adult chronic disease status moderate the association of adult neuroticism with health. This interaction was associated with the likelihood of drinking alcohol ever in the lifespan (b = 1.27, OR = 3.56 [1.46, 8.89], p = .005) and the count model of number of times per month the participant engaged in binge drinking (b = 0.70, 95% CI [0.25, 1.15], p = .002). In both cases, the interaction suggested that while neuroticism was unrelated to alcohol consumption in the absence of a chronic condition; however, when a chronic condition was present, neuroticism was positively associated with alcohol consumption. Sample sizes ranged from 484 (cholesterol/HDL ratio) to 588 (number of cigarettes smoked).

Table 4.

Chronic disease as a moderator of neuroticism.

Outcomes	Neuroticism	Chronic Disease	N x Chronic Disease
	Est [95% CI]	Est [95% CI]	Est [95% CI]	p
Self-rated health	−0.47 [−0.66, −0.29]	−0.47 [−0.61, −0.32]	−0.15 [−0.41, 0.12]	.270
Ever smoked	0.11 [−0.30, 0.53]	0.15 [−0.18, 0.48]	0.26 [−0.34, 0.86]	.405
Number cigarettes (count model)	−0.12 [−0.60, 0.37]	0.03 [−0.38, 0.44]	0.40 [−0.27, 1.06]	.243
Number cigarettes (zero model)	−0.46 [−1.18, 0.27]	−0.39 [−0.95, 0.17]	−0.21 [−1.18, 0.77]	.680
Ever drank	−0.16 [−0.78, 0.47]	−0.41 [−0.90, 0.08]	1.27* [0.38, 2.19]	.005
Alcohol (days per week) (count model)	0.08 [−0.08, 0.23]	−0.14 [−0.29, 0.00]	−0.07 [−0.31, 0.18]	.588
Alcohol (days per week) (zero model)	0.27 [−0.20, 0.74]	0.49 [0.12, 0.86]	−0.39 [−1.06, 0.28]	.251
Alcohol (drinks per sitting) (count model)	0.10 [−0.09, 0.30]	−0.12 [−0.30, 0.06]	0.18 [−0.13, 0.49]	.252
Alcohol (drinks per sitting) (zero model)	0.30 [−0.37, 0.96]	0.57 [ 0.03, 1.10]	−0.12 [−1.06, 0.82]	.805
Alcohol (binge per week) (count model)	−0.02 [−0.31, 0.26]	−0.25 [−0.52, 0.03]	0.70* [0.25, 1.15]	.002
Alcohol (binge per week) (zero model)	−0.43 [ −0.99, 0.13]	0.21 [−0.26, 0.69]	0.42 [−0.46, 1.29]	.351
Exercise (strenuous)	−0.07 [−0.22, 0.08]	−0.07 [−0.20, 0.05]	−0.05 [−0.28, 0.18]	.676
Exercise (moderate)	−0.03 [−0.24, 0.17]	−0.17 [−0.34, 0.00]	−0.07 [−0.38, 0.24]	.655
Exercise (mild)	0.04 [−0.17, 0.24]	0.04 [−0.13, 0.21]	−0.28 [−0.59, 0.03]	.080
Flossing	0.11 [−0.09, 0.30]	0.15 [−0.01, 0.31]	0.02 [−0.27, 0.32]	.875
Blood glucose levels	0.01 [−0.20, 0.23]	0.46 [0.29, 0.63]	0.19 [−0.11, 0.49]	.224
Cholesterol/HDL ratio	0.04 [−0.20, 0.28]	0.13 [−0.05, 0.32]	0.05 [−0.28, 0.38]	.783
Systolic blood pressure	−0.02 [−0.24, 0.19]	0.32 [0.16, 0.48]	0.06 [−0.23, 0.36]	.676
Waist to hip ratio	0.07 [−0.10, 0.24]	0.41 [0.28, 0.54]	0.14 [−0.09, 0.38]	.237

Neuroticism is standardized and chronic disease status is binary (0 = no chronic disease, 1 = one or more chronic diseases). Regression coefficients (standardized when the outcome is continuous and the model is estimated using ordinary least squares) and 95% confidence intervals are presented. All models control for gender and age at outcome measurement.

^*indicates p < .05. Stars are only used indicate significant interaction terms; we do not interpret significance of the simple slopes (e.g., the coefficients for neuroticism and health) because they are dependent upon the level of the other predictor and only included to allow for better interpretation of the interaction. Exact p-values for the interaction terms are presented to allow for estimation of the evidentiary value of these findings.

Question 5: Do facets of neuroticism predict health, and if so, do they predict health in opposing directions?

To examine the associations with specific neuroticism facets, we turned to a bifactor model, which defines unique latent constructs for general neuroticism and specific facets, allowing for an investigation of whether facets are associated with health outcomes after accounting for their shared variance (the general factor). Both the loadings of variables on to latent factors and the results of the regressions with the bifactor variables can be found in Table 5. Finally, we found no evidence that facets of adult neuroticism showed opposing associations associated with health and health behaviors. Instead, general neuroticism and two facets, immoderation and vulnerability, were both associated with worse health. General neuroticism was associated with worse self-rated health (b = −0.55 [−0.72, −0.38], p < .001), smoking a greater number of cigarettes (b = 0.18 [0.04, 0.32], p = .013), more frequent binge drinking (b = 0.24 [0.00, 0.47], p = .046) and a greater waist-to-hip ratio (b = 0.34 [0.09, 0.59], p = .008). Immoderation was associated with worse health behaviors and worse health. Specifically, immoderation was associated with less moderate exercise (b = −0.42 [−0.82, −.0.03], p = .035), less mild exercise (b = −0.33 [−0.67, −.0.00], p = .048). Vulnerability was associated with worse self-rated health (b = −0.68 [−1.17, −.0.20], p = .005), a greater number of chronic illnesses (b = 0.48 [0.15, .0.80], p = .004), less mild exercise (b = −0.41 [−0.79, −.0.03], p = .036), high cholesterol to HDL ratios (b = 0.60 [0.08, 1.11], p = .025) and a higher waist-to-hip ratio (b = 0.99 [0.43, 1.55], p = .001). Vulnerability was also associated with greater flossing (b = 0.37 [0.01, .0.73], p = .042; although, here we again note the p-value is between .04 and .05). All other facets were either negatively associated with health and health behaviors, or showed no association. Sample sizes ranged from 544 (ever drink alcohol) to 590 (flossing).

Discussion

The current study sought to identify indicators of healthy neuroticism by examining multiple potential moderators of neuroticism and health and the role of facets of neuroticism, which may have competing associations with health. Even when examining both adulthood and childhood personality and using a variety of health indicators, there was no evidence that the relationship between neuroticism and health depends on levels of conscientiousness, socio-economic status, or chronic disease status. Moreover, there is no evidence that facets of neuroticism are associated with health in differing directions.

We examined nineteen interactions for each combination of neuroticism and moderators, with only a few significant effects found. We chose not to interpret these significant effects because we do not believe they represent true population effects but are instead the result of sampling variability and chance. This point is underscored by the fact that some of these effects are in the direction opposite of what would have been predicted. For example, greater childhood conscientiousness is associated with stronger relationships between child neuroticism and adult waist-to-hip ratio, suggesting neuroticism is unhealthier at higher levels of conscientiousness, rather than less. This is in contradiction to other studies of neuroticism and conscientiousness (15,18). Irregularities across these studies may be due to differences in inclusion criteria; the current study removed a greater number of participants to maintain consistency across analyses. If this is the case, contradictions in results underscores the current study: these interactions may be non-existent or too small to reliably detect. In summation, the current study provided no substantial evidence to support the healthy neuroticism theory.

Statistical analyses cannot disprove a theory, and we acknowledge that there is still much to explore regarding the role of neuroticism on health. For example, many descriptions of healthy neuroticism refer to the adaptability of vigilance (13,15,18,44,45), yet little research on trait vigilance and its relationship to personality can be found (46). Similarly, neuroticism is widely accepted to be a trait that refers to fluctuating patterns of emotion and reactivity (47). Examinations of emotional response and the ways in which emotion may be adaptive could also illuminate pathways from neuroticism to longevity. The potential benefits of neuroticism also may require delving deeper into a given context; for instance, it may be important to consider whether neuroticism predicts diabetes-specific adherence behavior (i.e., checking insulin counts) for individuals with diabetes. Such tests may provide better fidelity for considering the effects of neuroticism for individuals with a chronic condition. Finally, researchers may wish to consider whether traits such as anxiety hold non-linear associations with health, as shown for personality and performance (48), and whether trait moderations of neuroticism may occur at the facet level (e.g., organization, industriousness, and self-control instead of conscientiousness). Indeed, personality and health researchers may need to think outside the box and beyond the self-report questionnaire to untangle these relationships.

However, we do caution the creative researcher. Theories of healthy neuroticism arose from the need to address heterogeneity in findings from a literature defined by large panel studies. This heterogeneity may be due to an important psychological variable or process that is inconsistently sampled across studies, or this heterogeneity may be due to random noise. We will never be able to say with certainty that it is the latter, so we may need to decide on standards of credibility for the former. Along these lines, we advocate for emphasis on base rates of potential moderators or processes. The examination of the high-conscientiousness, high-neuroticism combination is unlikely to yield many significant results, given that these traits tend to be negatively correlated with each other (15). Panel studies may be needed to garner sample sizes sufficiently large to examine this combination, but this should influence our interpretation of the phenomenon as a whole. That is, if these large studies are needed to isolate these individuals, then this combination is an unlikely candidate for explaining heterogeneity across studies.

It is unclear why other studies have found interactions between neuroticism and variables of interest while the current study did not. The current study is limited by its relatively small sample size, compared to other studies of neuroticism and mortality, and we may lack the power to detect these interactions. It may be that published studies suffer from file drawer problems. More specifically, by focusing on single outcomes – such as mortality or smoking – instead of providing a comprehensive investigation, these studies fail to see the larger pattern of null results. Indeed, in the current study, one of our 19 interactions between child neuroticism and conscientiousness was statistically significant. If we had chosen to examine only that model, the current study might provide more favorable evidence for a moderator-based theory. In addition, it would be valuable to have an earlier self-assessment of neuroticism that occurred prior to the accumulation of the health conditions, and some reliabilities for the childhood measure evidenced less-than-ideal reliability. Finally, we were unable to examine some of the outcomes previously studied, such as interleukin-6 or mortality (13,15). Perhaps any effects of healthy neuroticism are more limited to these outcomes.

Overall, the current study searches for and ultimately fails to find any indication of healthy neuroticism, even when examining the most likely moderators of neuroticism and health, neuroticism across the lifespan, and the facets of neuroticism. Future research should consider other pathways through which neuroticism may influence health, including state-level processes.