Participants for the present study were high school juniors who were recruited through their local schools for an ongoing longitudinal study of personality, cognitive, biological, and life stress risk factors for emotional problems in late adolescence. All students in their junior year at two local high schools (in suburban Chicago and Los Angeles) were initially invited to participate. A total of 1,977 students completed the 22-item Neuroticism scale of the Eysenck Personality Questionnaire–Revised (EPQ-R; S. B. Eysenck, Eysenck, & Barrett, 1985
), which was used to screen potential participants for level of neuroticism. Students scoring in the upper tertile of the EPQ-R N scale were over-selected as potential participants, in order to increase the number of participants at risk for future psychopathology (Alloy et al., 2006
; Costello et al., 1996
). Using this method, 923 individuals were invited to participate. Of the 923 invited students, 520 initially agreed to participate, of whom 491 completed baseline assessments, including assessment of Axis I psychopathology using the Structured Clinical Interview for DSM-IV-TR, non-patient edition (SCID-I/NP; First, Spitzer, Gibbon, & Williams, 2002
). Of those participants completing baseline assessments, 375 were randomly selected to participate in a cortisol sampling assessment, 278 of whom eventually completed that portion of the study.
Additional exclusionary criteria were applied in selecting the final sample for analysis, and included the following: pregnancy in the third trimester; a past or present endocrine disorder (e.g. Cushing's syndrome, adrenal tumors, Addison's disease); presence of psychotic symptoms during the diagnostic clinical interview; use of corticosteroid-based medications; diagnosis of a current clinical mood disorder, PTSD, or generalized anxiety disorder; or extensive missing data. None of the participants in the current study reported being pregnant. Three individuals were removed due to psychotic symptoms and 13 due to the use of corticosteroid-based medications. Two participants were excluded due to poor reporting of cortisol sampling times, and five participants did not complete the personality questionnaires. Finally, 25 participants who met criteria for current major depression (n
= 12), generalized anxiety disorder (n
= 13), dysthymia (n
= 3), or post-traumatic stress disorder (n
were excluded, to ensure that potential effects of personality could not be accounted for by the current psychopathology that is itself typically associated with HPA-axis dysregulation (e.g., Hirschfeld et al., 1989
Despite the removal of participants with current emotional disorders, the final sample (n = 230) did not significantly differ from the larger study's original sample (n = 491) on neuroticism (t = 1.520, p = .129) or on introversion (t = 0.880, p = .379). There were also no significant differences between the current sample and the original sample on age (t = -0.335, p = .737). Regarding race/ethnicity, the current sample of participants was more likely to be Hispanic than participants in the original sample (t = -1.959, p = .051), and was less likely to be African American (t = 2.72, p = .007). The current and original samples did not differ in terms of their representation of any other races/ethnicities (i.e., Caucasian, Asian, Native American, Pacific Islander). The current sample (n = 230) scored an average of 1.82 points higher (t(1974) = -5.702, p < .000) on the study's initial N measure (the EPQ-R; range = 0 – 22), as compared to the screening sample (n = 1977). This difference was likely due to the effect of oversampling for high N. Group differences between the current sample and the screening sample were not significant for age (t = -0.444, p = .657) or for any racial/ethnic group.
The results reported herein were based on data from the remaining 230 participants (170 females and 60 males). The unequal gender representation in the present sample is attributable to several causes. First, females were slightly more likely that their male counterparts to complete the initial screening questionnaire (i.e., 56% female, 44% male). Second, females, on average, tend to score higher on measures of neuroticism than do males (Goodwin & Gotlib, 2004
), and because the study overselected for high N individuals, more females were inherently invited to participate than males (i.e., 63% of those who were invited were female). Third, of those invited, females were more likely than males to agree to participate (i.e., 65 % of the invited females agreed to participate, vs. 52% of the invited males).
At the time that cortisol sampling was completed, participants were 16–18 years old (M = 17.1, SD = 0.4). The sample was diverse, consisting of Caucasian Americans (47%), Hispanic or Latino Americans (23%), African Americans (15%), Asian Americans / Pacific Islanders (8%), and other ethnicities (7%). Fifty-eight percent scored in the upper tertile on the EPQ-R N scale, 23% were in the middle tertile, and 19% were in the bottom tertile on N.
All participants in the present study were administered the Structured Clinical Interview for DSM-IV-TR, non-patient edition (First et al., 2002
) following a semi-structured life stress interview (Hammen et al., 1987
). Following the two interviews, participants completed a battery of questionnaires either following the interview session or during a meeting scheduled shortly thereafter. Participants received $40 for their participation in these tasks.
Approximately 2 months (M
= 57.8 days, SD
= 42.8 days) following questionnaire completion, participants completed the three-day period of salivary cortisol collection. Participants were mailed cortisol collection supplies, written instructions, and programmable wristwatches. Students were instructed to provide a saliva sample immediately upon waking, 40 minutes after waking, immediately before bedtime, and in response to three wristwatch beeps that were programmed to sound approximately 3, 8, and 12 hours after waking. This schedule (totaling 18 collections) was devised to provide the most informative description of participants' cortisol rhythm, while avoiding the increase in cortisol levels following meals (Follenius, Brandenberger, & Hietter, 1982
During the three days of cortisol assessment, students also completed Experience Sampling Method diary reports (ESM; Csikszentmihalyi & Larson, 1987
). Students were instructed to report on the situations and emotions that they experienced immediately prior to the cortisol sampling. For the present analyses, particular variables from the ESM data were theoretically chosen as potential mediators of associations between personality and diurnal cortisol patterns. These ESM variables included: social environment at the time of the cortisol sample (alone, not alone, want to be alone, with family member, with peer, with friends, with significant other, with someone else); mood at the time of the cortisol sample (happy, friendly, cheerful, cooperative, alert, caring, relaxed, active, productive, tired, nervous, lonely, frustrated, worried, irritable, stressed, sad, determined, feeling good about self); and the presence, severity, and topic (self, family, peer, friend, significant other) of any current stressor at the time of sampling. Individual mood state items were averaged to form composite measures of momentary positive moods (i.e., happy, friendly, cheerful, cooperative, alert, caring, relaxed; α = .922) and momentary negative moods (i.e., nervous, lonely, frustrated, worried, irritable, stressed, sad; α = .903). Additional data and analyses based on the full range of ESM data are discussed elsewhere (Doane et al., 2008
; Mor et al., 2007
Eysenck Personality Questionnaire-Revised (EPQ-R), Neuroticism scale
Potential participants for the present study completed a 22-item4
version of the EPQ-R Neuroticism scale (S. B. Eysenck et al., 1985
). Items included yes/no questions concerning mood, anxiety, and energy levels. In a 24-item version of the EPQ-R N scale, Eysenck et al. reported internal consistencies of .85 for females and .88 for males. Using Cronbach's alpha to assess reliability, the 22-item EPQ-R N scale in the present study yielded appropriate internal consistency for a trait measure (α = .78). Coefficient omegahierarchical
) was estimated at 0.66 for the present sample (Uliaszek et al., 2007
International Personality Item Pool NEO-PI N scale (IPIP)
Participants also completed the 60-item IPIP NEO-PI N scale as a second measure of N (International Personality Item Pool, 2001
). Individuals were asked to rate all items on a 5-point scale (1 = “Very Inaccurate”, 5 = “Very Accurate”), according to the items' accuracy in describing their own behaviors. Items include both positively and negatively coded statements such as “Get irritated easily”. All six IPIP NEO-PI subscales have been found to yield adequate internal consistency (α > .77), and correlate substantially with their corresponding NEO-PI-R subscale (r >
.7). Internal consistency was again estimated using Cronbach's alpha, indicating high reliability for the IPIP NEO-PI N scale (α = .94), and coefficient omegahierarchical
was estimated at .67 (Mor et al., 2007
Behavioral Inhibition System (BIS)
Students were also administered the Behavioral Inhibition System scale (Carver & White, 1994
). The 7-item BIS scale is measured on a 4-point Likert scale and is a part of a larger 20-item BIS/BAS questionnaire, assessing both behavioral inhibition and behavioral activation characteristics. The BIS is generally considered to reflect the construct of trait anxiety, one of the known facets of neuroticism (Costa & McCrae, 1995
). Anxiety and neuroticism in general have been positively associated with the BIS (Gray, 1994
). Specifically, the BIS is associated with the tendency to fear and avoid potentially averse situations, and to respond to threat with increased arousal and sensitivity (for a review, see Revelle, 1995
). BIS scale items include statements such as “If I think something unpleasant is going to happen, I usually get pretty ‘worked up.’” Carver and White (1994)
reported an 8-week test-retest reliability of .66 and demonstrated significant convergent/divergent validity of the BIS as a marker of neuroticism (also see Zinbarg & Mohlman, 1998). Internal consistency for the BIS scale was estimated with Cronbach's alpha (α = .75).
Big Five Mini-Markers
To provide a fourth measure of neuroticism, as well as a single measure of introversion, participants in the current study completed the Big 5 Mini-Markers (referred to hereafter as the “Big 5”; Saucier, 1994
). In the 40-item, homogenously keyed Big 5 questionnaire, each of the five personality factors is represented with 8 questionnaire items. Individuals rated all items on a 9-point Likert scale (1 = “Extremely Inaccurate”, 9 = “Extremely Accurate”), also according to the items' accuracy in describing the rater's behaviors. Items corresponding to N included such characteristics as “Moody” or “Relaxed”, and items representing I included traits such as “Energetic”, “Talkative”, “Bashful”, or “Withdrawn”. All of the five factors on this scale have been reported to exhibit strong reliability (α > .77). For the Big 5 N scale in the present study, internal consistency was strong (α = .81).
Although the present analyses assessed I using only the Big 5 I scale, there is evidence to support the reliability of this measure. Saucier reported alpha estimates of .80 for the Big 5 I scale (Saucier, 1994
), and correlations between .91 and .96 with Goldberg's more extensive I scale have been reported (Goldberg, 1992
). It is important to note, however, that the Big 5 Mini-Markers assessment of I includes only items that reflect characteristics of sociability and withdrawal. Items reflecting other putative facets of I, such as low positive emotionality (Costa & McCrae, 1995
), are not examined with this measure. Using the present data, internal consistency for the Big 5 I scale was similarly strong (α = .83).
Computations of N and I
To make use of the multiple measures of neuroticism (N) that were available for the present dataset, a composite measure of N was calculated. This was computed by averaging z
scores from the EPQ-R N scale, IPIP, BIS, and Big 5 Mini-Markers N subscale5
. To confirm that these four measures of N (EPQ-R N, IPIP, BIS, Big 5 N) could be regarded as reflecting a unitary construct, intercorrelations were computed. Correlation coefficients between pairs of measures ranged from: r
= 0.49 (EPQ-R N and BIS; p
< .0001) to r
= 0.76 (IPIP and Big 5; p
< .0001). To confirm that the BIS (a measure of trait anxiety—one of the facets of N) was appropriately included in the present measure of N rather than I, intercorrelations were computed between the BIS and the N and I scales of the Big 5 Mini-Markers. The correlation between the N scale on the Big 5 Mini-Markers and the BIS was significant, r
= .559, p
< .0001. The correlation between the I scale on the Big 5 Mini-Markers and the BIS was also significant (r
= -.288, p
< .0001) but was more comparable to the correlation between the N and I scales on the Big 5 Mini-Markers (r
= -.286, p
< .0001). A recent report (Griffith et al., 2007
) determined that a unifactorial model provided a good fit to the observed covariances among the four scales of N (CFI = 1.0, RMSEA = .04, SRMR = .01). Introversion (I) was calculated by obtaining reverse scores of the raw values on the Big 5 E subscale, and then standardizing this measure.
Health and Demographic Measures
Participants in the present study completed a health questionnaire reporting on their personal medical history and health behaviors, including self-reported medical conditions, chronic medical conditions (particularly those that could affect cortisol levels), intake of caffeine and nicotine, waking and bed time, and use of prescribed medications (e.g., corticosteroids, oral contraceptives). Participants also provided information on health-related variables that are putatively associated with cortisol (e.g., S. Cohen et al., 2006
; Kudielka & Kirschbaum, 2003
; Meulenberg et al., 1987
). These health-related variables (i.e., waking time and bed time; use of caffeine, nicotine, and oral contraceptives) were not considered exclusionary criteria but were included as covariates in statistical analyses, along with additional demographic variables (i.e., gender, ethnicity). Ethnicity was reported using dummy (0, 1) variables, for participants who identified as African-American, Asian-Pacific Islander, or Hispanic-Latino (Caucasian was the excluded group). Due to prior findings that diurnal cortisol slopes were flatter for African-American males (DeSantis et al., 2007
), ethnicity by gender interactions were also included.
Cortisol Sampling and Assay Procedures
To obtain samples, participants expelled saliva through a straw into a sterile 2 mL cryogenic vial (Schwartz, Granger, Susman, Gunnar, & Laird, 1998
). No salivary stimulants were used. Participants were asked to cap the vials of their saliva samples tightly, label them with date and time of day, and refrigerate them as soon as possible. Once returned via postal mail or by way of a frequently-checked drop box at participants' schools, saliva samples were stored at −20 degrees Celsius until assayed. Investigations have confirmed that cortisol concentrations remain stable when mailed in an unfrozen state (e.g., Clements & Parker, 1998
Samples were later shipped to the Biochemisches Labor at the University of Trier, Germany. Samples were assayed in duplicate, using a competitive solid phase time-resolved fluorescence immunoassay with fluorometric endpoint detection (DELFIA). Fluorescence was detected using a DELFIA-Fluorometer (Wallac, Turku, Finland). Inter- and intra-assay coefficients of variation are 7.1% – 9.0% and 4.0% – 6.7%, respectively.
Description of key variables
Key predictor variables were the composite N and I scores described previously. Because scores on these measures approximated normal distributions, no transformations were applied. Gender was also a predictor variable, but was considered a primary variable of interest only in the context of its interaction with personality. All values of personality variables (that were multiplied by gender to create interaction terms) were converted to z
scores for subsequent analyses, in order to reduce multicollinearity due to non-essential ill-conditioning among interaction variables (J. Cohen, Cohen, West, & Aiken, 2003
). As described earlier, demographic and health-related variables were regarded as covariates. In the final analyses, N and I were entered in models simultaneously, such that associations could be attributable to the unique effects of N and I, rather than the effects of their shared variance.
The criterion variables in the present analyses were 3 parameters characterizing the diurnal cortisol rhythm: level of cortisol at wakeup time, size of cortisol awakening response (CAR)
and slope of the diurnal curve
. Although the precise meanings of each cortisol parameter (wakeup level, CAR, and diurnal slope) are not yet fully understood, we consider these three parameters to be relatively distinct in their origins and interpretations. Cortisol level during wakeup is typically more genetically determined than cortisol measured in the afternoon or evening (Clow et al., 2004
). The size of the CAR is thought to be heavily genetically influenced as well, but is also related to current chronic stress, in addition to perceived or expected stress on the specific days that the CAR is measured (Adam et al., 2006
; Clow et al., 2004
; Gunnar & Vazquez, 2001
). Finally, a flattening of the diurnal cortisol slope can result, in part, from the experience of acute stress or negative affect on the specific days of cortisol testing (Adam et al., 2006
). Flatter slopes also appear to be partially entrained over time by a history of chronic psychosocial strain (Adam et al., 2006
; Gunnar & Vazquez, 2001
Estimates of the three cortisol parameters of interest were obtained using multilevel growth curve modeling, as described below. In order to correct for positive skew, a base e logarithmic transformation was applied to the cortisol values prior to use in our models. No significant cortisol outliers were identified in the data.
Multilevel Modeling of Diurnal Cortisol Rhythms
A 3-level multilevel analysis (Raudenbush & Bryk, 2002
; Singer & Willett, 2003
) was performed to model the shape of the diurnal cortisol rhythm, and to examine the associations between the independent variables and the key parameters defining individual differences in diurnal cortisol rhythms (wakeup values, size of the CAR, and slope from wakeup to bedtime). The three levels represented data at the moment level (cortisol levels across the day), nested in the day level (waking times and bedtimes on the days of testing), nested in the person level (individual differences in personality, health, and demographic variables). Multilevel modeling was selected because it allowed us to simultaneously model the shape of the diurnal cortisol rhythm for each individual, and to examine predictors of those rhythms (Adam & Gunnar, 2001
; Adam et al., 2006
). This method also adjusts for correlated error within each level of data, whereas OLS models do not (Adam & Gunnar, 2001
; Hruschka, Kohrt, & Worthman, 2005
To provide estimates of diurnal cortisol rhythm from the moment-level cortisol data, a Level 1 model was fitted with the following parameters:
(1) Level-1 Model: Cortisol = π0 + π1 * (Time Since Waking) + π2 * (Time Since Waking)2 + π3 * (CAR) + e
In the model above, the intercept term (π0) represents the latent estimate of each participant's cortisol level at wakeup, estimated across all 3 days of cortisol collection. The π1 coefficient (associated with the linear term of “Time Since Waking”) represents the latent estimate of the slope of each participant's diurnal cortisol curve at wakeup. The π2 coefficient (associated with the quadratic term), was included to test for curvilinear relationships. The “Time Since Waking” variable represents the time of day that the cortisol sample was taken, modeled as the number of hours since the participant had woken on that particular day. Finally, the π3 coefficient is a latent estimate of the size of the cortisol awakening response. By including a dummy variable (0, 1) indicating which samples were part of the CAR (taken 40 minutes post-awakening), π3 reflects how much the 40 minute post-awakening samples deviated above the expected cortisol value at that time of day (based on the typical diurnal decline for that individual).
Demographic and health-related covariates were included at either the person level (Level 3) or the day level (Level 2). Covariates that varied with each day (including wakeup time and bedtime) were included at Level 2 of the hierarchical analysis. The general model for these day-level covariates was as follows:
(2) Level-2 Model: π0 to π3 = βi0 + βij * (Day Level Covariates) + rij
Covariates that varied with each participant (but remained consistent from day to day) were included at Level 3. These person-level covariates included participants' ethnicity, as well as use of caffeine, nicotine, and oral contraceptives. In addition, Level 3 included the primary predictor variables of N and I, as well as the N * gender and I * gender interaction terms (with male = 1). The general model for these person-level variables was as follows:
(3) Level-3 Model: βi0 + βij = γij0 + γijk * (Person Level Covariates) + γijk * (Personality Variables) + γijk * (Gender by Personality Interaction Variables) + uijk
To test for potential mediators of any significant associations between personality and diurnal cortisol patterns, ESM diary variables were added to the model (describing participants' average mood states and social experiences across the days of cortisol testing, as described above). ESM data were reported at the moment level but aggregated to the person level, in order to test their potential mediation of the person-level personality effects. All momentary ESM variables of interest were therefore averaged for each participant across all moments of cortisol sampling. These variables were added to the Level 3 model. The general model for these analyses is listed below.
(4) Level-3 Model: βi0 to βij = γij0 + γijk * (Person Level Covariates) + γijk * (Personality Variables) + γijk * (Gender by Personality Interaction Variables) + γijk * (Averaged Daily Experiences) + uijk