Correlates and Characteristics of Adolescents' Encoded Emotional Arousal During Family Conflict

Abstract

Associations between adolescents' range of fundamental frequency, cortisol output, and self-reported emotional experience were examined during problem discussions with parents. Participants are a community-based sample of 56 boys and girls in a longitudinal study on conflict exposure. Results reveal that higher aggregate levels of range of fundamental frequency are associated with higher cortisol output and higher levels of self-reported negative emotions for boys and girls. Additionally, greater cortisol output is significantly associated with a slower time-to-peak of range of fundamental frequency for girls and with significantly less variability in range of fundamental frequency for boys. Implications of results for emotional development in adolescents, measurement, and modeling are discussed.

Emotional arousal of both parents and children during family conflict is a well-documented correlate of numerous domains of intrapersonal and interpersonal functioning. For example, higher levels of family conflict-related emotional arousal are linked to poorer psychological adjustment (e.g., Birchwood, Hallett, & Preston, 1989) and impaired social functioning (e.g., Gottman & Katz, 1989). Greater emotional arousal during family conflict is also associated with poorer physical health outcomes such as more symptoms of physical morbidity (see Kiecolt-Glaser & Newton, 2001 and Margolin & Gordis, 2000 for reviews). Finally, higher levels of emotional arousal during couple conflict are also associated with increased risk for more frequent and intense detrimental relationship behaviors such as physical aggression (e.g., Gottman et al., 1995) and the demand/withdraw interaction pattern (Baucom et al., 2011). Compared with the robust literature on parents' emotional arousal during family conflict, there has been little attention to children's emotional arousal. The lack of literature on children's emotional arousal during family interaction is even more pronounced in the study of nonverbal expression of emotional arousal. To address this gap, the current study examines associations between a vocal indicator of emotional arousal (i.e., range of fundamental frequency) and more standardly used measures of emotional arousal, namely cortisol output and self-reported negative emotions. We investigate associations between these measures of emotional arousal in adolescents during discussions with their parents about conflictual issues.

Fundamental Frequency as a Measure of Emotional Arousal Across Developmental Stages

Encoded emotional arousal is a well-accepted vocal measure of emotional arousal and is typically measured using fundamental frequency (f₀; see Juslin & Scherer, 2005 for a review). F₀ refers to the lowest frequency harmonic produced during the phonation phase of speech production and is highly correlated with perceived pitch. Numerous indices of f₀ have been linked to the vocal expression of emotional arousal in adults and infants. Greater f₀ range, higher mean f₀, higher maximum f₀, and greater standard deviation of f₀ all indicate higher levels of emotional arousal in adults (for reviews, see Juslin & Scherer, 2005; Russell, Bachorowski, & Fernandez-Dols, 2003). Likewise, higher mean f₀, higher maximum f₀, and greater standard deviation of f₀ of infant crying have all been linked to higher levels of infant distress (e.g., Zeskind & Marshall, 1988; Porter, Miller, & Marshall, 1986).

Though there is a strong basis of empirical support for several f₀-related parameters, current methodological recommendations and recent empirical evidence suggest that f₀ range is likely to be the most robust f₀-related parameter of encoded arousal. Juslin and Scherer (2005) reviewed over 100 articles examining acoustic correlates of emotional expression in adults and concluded that f₀ range is the cleanest f₀-related index of emotional arousal. Likewise, Busso, Lee, and Narayanan (2009) recently found that f₀ range conveys the largest amount of emotional information out of 43 different indices of f₀ in adult speech. We therefore examine f₀ range as a measure of adolescent's vocally encoded emotional arousal and do so using four different aspects of f₀ range (overall range, time-to-peak range, linear trends over time and oscillatory behavior over time) to conduct as thorough an examination as possible.

Significant vocal changes that occur during adolescence also highlight the importance of exploring potential sex differences in the links between f₀ range and emotional arousal. Vocal folds lengthen and thicken during puberty, resulting in decreased mean f₀ for both sexes. For example, Hollien, Green, and Massey (1994) found boys' average speaking f₀ to drop from over 230 Hz at age 13 to approximately 115 Hz at age 15. Additionally, greater differentiation develops between male and female voices during puberty. Prior to puberty, the average mean f₀ of boys' and girls' speech is nonsignificantly different; however, after puberty, the mean f₀ of males' speech is significantly lower than that of females (e.g., Perry, Ohde, & Ashmead, 2001). In contrast to the observed decreases in mean f₀, f₀ range appears to increase during puberty and to be more variable than either the prepubertal or post-pubertal periods (e.g., Hollien et al., 1994). We therefore also examine sex differences in vocal behavior and encoding patterns.

Examining expressive aspects of adolescents' internal emotional experiences may reveal information with implications for emotional development. As other authors have noted (e.g., Sroufe, 1997), one of the key aims of research on emotional development is to document the form and organization of emotional reactions across the life span. For example, physiological states concomitant with facial expressions of emotion are known to change across developmental stages (e.g., Tsai, Levenson, & Carstensen, 2000). It is currently unknown whether, and if so how, adolescents' vocal characteristics reflect internal emotional experiences and if these patterns of association are consistent across developmental stages. Establishing physiological and subjective correlates of adolescents' vocal expression would create a foundation for future comparison with similar associations in different developmental stages. Focusing specifically on expressive and experiential aspects of emotion in adolescence could lead to new insights into developmental trends in the vocal expression of emotional arousal (Sroufe, 1997).

Coherence Across Emotional Systems During Interpersonally Salient Interactions

The widely varying methodologies used to establish associations between f₀–related parameters and emotional arousal in adulthood and infancy suggest that there are a number of possible methods that could be employed to examine f₀ range during adolescence. One notable omission from existing studies of f₀ and emotional arousal is the lack of evidence of direct associations between any f₀–related parameter and other continuous measures of emotional arousal. Perhaps one reason for this lack of evidence is that theories of emotion differ in the extent to which they suggest that emotional expression, physiological state, and subjective experience are convergent. Most modern models of emotion, such as component process models, suggest at least some significant coordination across different facets of emotion (e.g., Scherer, 2000). Component process models suggest that the coordination of emotional subsystems arises largely from the impact of attributions for physiological arousal on emotional expression, subjective experience, and action tendencies (Grandjean, Sander, & Scherer, 2008). It is important to note that though these models suggest a general coherence across emotional subsystems, they also highlight the importance of differential levels of coordination across emotional subsystems both over time as well as between individuals (Scherer, 2009).

Existing studies of f₀–related parameters and emotional arousal do not employ sufficient focus on interpersonally significant, social interactions. Some studies examine f₀–related parameters during salient communications, such as during the conversations of aircraft pilots and air traffic controllers in the moments before the crash of an aircraft (Williams & Stevens, 1969). However, the relevance of those interactions arises not from the significance of the relationship between the two interactors but rather from the importance of the situation in which they are interacting. In contrast, interactions between parents and children represent both situationally and relationally significant interactions. Conflict between parents and children during adolescence is a particularly important interaction context for many of the primary developmental tasks of adolescence, such as establishing greater independence from parents. Evidence linking f₀ range to physiological and subjective arousal would therefore be of particular value during interpersonally significant social interaction, such as conflictual family discussions.

Summary and Time-Varying Aspects of F₀ Range During Family Discussions

A final important consideration for examining f₀ range as a measure of emotional arousal during family discussions is the options for characterizing f₀ range over the entirety of the discussion. Our examination of four methods for characterizing f₀ range (described in Table 1) is designed to allow for comparison with existing studies of f₀ range during infant and adult vocalization as well as to explore previously unexamined aspects of f₀ range. Many published studies aggregate f₀ range over the entirety of an interaction. Consistent with these studies, we first examine associations between a summary value of f₀ range for the entire discussion and physiological and self-reported arousal.

Table 1.

Methods for Characterizing F₀ Range

Index	Construct	Calculation of index
Aggregate f0 range	Overall amount of vocally encoded arousal expressed during the entire discussion	The aggregate measure of f0 range was calculated by subtracting the minimum f0 value for the entire conversation from the maximum f0 value for entire conversation.
Time-to-peak f0 range	Latency of the maximum increase in vocally encoded arousal	Time-to-peak f0 range was calculated as the number of seconds from the beginning of the conversation to the start of the talk turn containing the highest f0 range value.
Linear trend in f0 range	Escalation of vocally encoded arousal over time	Separate f0 range scores were created for each talk turn and analyzed using a linear growth curve model.
Oscillatory behavior of f0 range	Regulation or control of vocally encoded arousal over time	The second derivative of f0 range (${\mathrm{x}}_{\mathrm{ij}}^{"}$) was created using the Local Linear Approximation method (Boker & Nesselroade, 2002). Each individual's data was first “detrended” to remove the effect of the intercept and the linear effect of time. The second derivatives for each individual's detrended data were generated using simple difference scores and analyzed using dynamic oscillator models.

Although it is valuable to examine f₀ range parameters at the aggregate level, momentary fluctuations in f₀ range parameters also convey meaningful information about emotional states (e.g., Busso et al., 2009). We therefore also examine three time-varying aspects of f₀ range in relation to physiological and subjective arousal. One of these aspects is the amount of time elapsed before maximum f₀ range is exhibited (i.e., time-to-peak f₀ range). A faster time-to-peak mean f₀ during infant crying is associated with greater perceived distress of the infant (Protopapas & Eimas, 1997). While we therefore expect a faster time-to-peak f₀ range to be similarly related to greater physiological and self-reported arousal for both boys and girls, it is possible that gender differences additionally exist in time-to-peak f₀ range. Women's emotional responses to negative stimuli typically last longer than do men's responses (Germans Gard & Kring, 2007). If these emotional responses are encoded in f₀ range, increases in girls' f₀ range for each negative stimulus may last longer than increases of equal magnitude in boys' f₀ range. This gender difference is significant for examining emotional processes during family conflict because most unresolved family conflict involves frequent and repeated exposure to negative stimuli. Gender differences in the duration of responses to negative emotional stimuli combined with the frequent and recurring negative stimuli typical of family conflict result in a situation where girls may not have sufficient time for a response to subside before another negative stimulus is experienced. Girls' responses may therefore pile up on one another and result in girls' peak f₀ range occurring significantly later than boys' peak f₀ range.

In addition to their probable associations with time-to-peak f₀ range, physiological and self-reported arousal may also be related to specific contours of f₀ range over time, such as linear and oscillatory trends (e.g., Busso et al., 2009). Though some studies link various contours of f₀-related parameters to emotional arousal (see Juslin & Scherer, 2005 for a review), there is no current consensus in how to best model f₀ contours. In order to include as much acoustic and temporal information as possible, we propose the combined use of linear and oscillatory models to capture the major sources of variability in f₀ range over time. We therefore examine longitudinal variation in moment-to-moment fluctuations in f₀ range using both a standard linear growth curve approach and dynamic oscillator models.

In addition to the statistical benefit they offer when combined with linear models, dynamic oscillator models are advantageous for a number of other statistical and conceptual reasons. The primary added statistical benefit of dynamic oscillator models is that emerging evidence shows that, as a general class of models, dynamic models are particularly well suited to capturing and describing variability in cyclical social behavior (Guastello, Pincus, & Gunderson, 2006). Dynamic oscillator models in particular (e.g., Boker & Nesselroade, 2002) have proven beneficial for modeling emotional processes generally (e.g., Chow et al., 2005) as well as those occurring during family interaction specifically (e.g., Guastello et al., 2006).

The conceptual benefit of dynamic oscillator models is that they are consistent with key ideas in arousal theory (Yerkes & Dodson, 1908) and current models of emotion regulation (e.g., Butler & Gross, 2009). Arousal theory suggests that there are optimal levels of arousal for individuals during engaging activities (i.e., individual set-points) and that individuals are most comfortable when they experience arousal at their optimal level. Current models of emotion regulation (i.e., Butler & Gross, 2009) similarly suggest that there is intrinsic motivation to regulate arousal so that it is as close to an optimal level as possible and that there are meaningful differences in how well individuals are able to regulate their arousal. Because the specific parameterization proposed by Boker and Nesselroade (2002) models precisely this form of variability, individual differences in rate of return to set-points can be interpreted as individual differences in regulatory processes.

Cortisol as a Measure of Physiological Arousal

Similar to the numerous options for examining temporal and summary measures of f₀ range, there are also many physiological measures that could be used to assess emotional arousal during family conflict. We use cortisol output because: (a) it is widely used to assess physiological state during couple and family interaction (e.g., Ditzen et al., 2007), (b) it is an established measure of emotional arousal, and, (c) it offers a number of important advantages over other physiological measurement options. Though cortisol is most often referred to as a measure of stress response, it is also linked to negative emotional and mood states. For example, elevated cortisol levels are typically observed in individuals with depressive disorders, and higher levels of depressive symptoms are associated with higher levels of cortisol output (see Kiecolt-Glaser, McGuire, Robles, & Glaser, 2002). Cortisol also has a number of measurement properties that make it particularly advantageous for examining f₀ range during family conflict. First, cortisol is one of least intrusive physiological measures that can be obtained (Gunnar & Quevedo, 2007). Cortisol levels during a family conflict can be measured by collecting saliva samples following the conflict. This process maximizes the ecological validity of the family conflict and is in sharp contrast to invasive and distracting methods necessary for collecting many other physiological measures of emotional arousal. Second, cortisol is also known to be responsive to repeated exposure to the same stressor, particularly when the stressor is experienced as uncontrollable (Baum, Cohen, & Hall, 1993). This aspect of cortisol makes it particularly appealing for assessing emotional arousal during unresolved family conflict, which is often highly repetitive and frequently occurring (Margolin, Christensen, & John, 1996). A final source of support for examining links between cortisol and f₀ range comes from evidence of between person associations in mean f₀ and cortisol output. Women with lower perceived power have greater cortisol output when interacting with children with higher mean f₀ voices than when interacting with children with lower mean f₀ voices (Lin, Bugental, Turek, Martorell, & Olster, 2002). There has yet to be an empirical examination of within person association between cortisol and f₀ range.

Hypotheses

Based on the theory and research discussed above, we propose four hypotheses. First, greater aggregate f₀ range will be associated with higher levels of self-reported negative emotions and with greater cortisol output. Second, faster time-to-peak f₀ range will be associated with higher levels of self-reported negative emotions and with greater cortisol output. Also, girls will show slower time-to-peak f₀ range than boys. Third, higher levels of self-reported negative emotions and greater post discussion cortisol output will be associated with a greater linear increase in f₀ range over time. Fourth, we will explore associations among self-reported negative emotions, post discussion cortisol output, and the oscillatory behavior of f₀ range over time. Directional predictions are not possible given the lack of past empirical findings.

Method

Participants

Participants (N = 56 adolescents) are a subsample of 83 youth that completed the fourth assessment in a longitudinal study of the impacts of multiple forms of violence exposure on individual development and family functioning. Primary eligibility criteria at initial recruitment included the conditions that both parents and one child lived together for at least the past three years and that the child was between 9 and 10 years of age. Of the 27 youth not included in the current study, one was excluded because study procedures were completed at home, one was excluded because of a medical condition that invalidated cortisol data, 11 were excluded because they did not provide cortisol data, three were excluded because their cortisol values were consistently out of range, and 11 were excluded because of equipment malfunction and poor recording quality. The subset of families included in the current study did not significantly differ from the entire sample on demographic variables.

Adolescents in this sample ranged from 13.83 to 18.58 years old, with a mean age for boys of 15.16 years (SD = 0.63) and a mean age for girls of 15.27 years (SD = 0.71). The racial and ethnic composition of this sample is diverse and reflective of the major metropolitan area from which adolescents and their parents were recruited; 20% of adolescents self-classified as African American, 11% Asian or Pacific Islander, 21% as Caucasian, 38% as Hispanic or Latino/a, and 11% as multiethnic. Mothers and fathers reported an average of 14.42 and 14.79 years of education, respectively. Families reported wide ranging annual household income levels (13% less than $25,000; 7% between $25,000 and $50,000; 39% between $50,000 and $100,000; 41% greater than $100,000). Parents reported cohabiting for an average of 13.7 years and having an average of 2.6 children. Forty-six percent of adolescents were female (n = 26) and 54% were male (n = 30).

Procedures

Adolescents completed a 3- to 4-hr laboratory-based assessment session with either or both of their parents. After obtaining consent from parents and assent from adolescents, all participating family members completed computerized questionnaires in separate rooms. Families then participated in videotaped, 15-min conflict discussions. To determine high-conflict areas for discussion, family members rated 35 topics (e.g., chores, privacy, etc.) for the amount of conflict and the level of distress associated with each topic. In simultaneous interviews, one of three research assistants then met with each family member to identify the most salient topics and to prime that person to present her or his position during the conflict discussion. The three research assistants then selected three topics for discussion that represented significant issues for the family. Family members were then instructed to discuss the selected topics, starting with the one rated as most conflictual. Families were further instructed to interact as they do at home and each family member was encouraged to get his or her point across.

Saliva samples were collected over the course of the assessment, and we focus on the two samples that assess cortisol output during family discussion (see Saxbe, Baucom, Spies, & Margolin, in press for additional details about cortisol procedures). Because it takes 15 to 20 minutes for cortisol levels to register in saliva (Kirschbaum & Hellhammer, 1994) and family conflict discussions were 15 minutes long, the sample collected immediately after the family discussion and the next sample collected 10 minutes later best capture physiological arousal during family discussion. Compliance with instructions to neither eat nor smoke for an hour before the visit and to not consume alcohol or caffeine for 24 hours was confirmed both through verbal questioning and a questionnaire. Families were compensated $175 for participation in this assessment that also included a battery of other questionnaires and procedures.

Measures

Encoded arousal

Encoded arousal was indexed using adolescents' f₀ range during family conflict discussions. Audio recordings of family conflict discussions were manually decomposed into separate files for each family member. Acoustic features of the adolescents' files were then analyzed in 0.25 second intervals using Praat (Boersma & Weenink, 2011; see Owren & Bachorowski, 2007 for a review of methods for assessing vocal features of emotion). Obvious errors in pitch extraction were corrected using procedures outlined by Johnstone, van Reekum, Hird, Kirsner, and Scherer (2005). Plots of f₀ range values were first visually inspected for signs of obvious errors in pitch extraction. Errors, which occurred for three participants, were corrected by reanalyzing the respective audio recordings and reinspecting plots of f₀ range values. A single reanalysis was sufficient for correcting all errors. Time varying indices of f₀ range were analyzed using f₀ range values for each talk turn, a commonly used conversational unit in which one person is the speaker. F₀ range was calculated for each talk turn by first subtracting the minimum f₀ value for the entire discussion from the maximum f₀ value for each 0.25 second interval and then averaging those resultant values over each talk turn. Calculation of the aggregate index of f₀ range and each time varying index of f₀ range is described in Table 1.

Salivary cortisol

Cortisol output was indexed using two samples—one collected immediately after the end of the family conflict discussion and the second collected 10 minutes later—and calculated by finding the area under the curve between the two points with respect to ground (Pruessner, Kirschbaum, Meinlschmid, & Hellhammer, 2003). We chose to use this aggregate measure of cortisol output during the family conflict because it is most analogous to f₀ range; both indices reflect overall levels of arousal as well as individual differences in overall levels of arousal. Saliva samples were analyzed by Salimetrics, LLC, who provide an enzyme immunoassay with a lower limit of sensitivity of 0.003 μg/dl, and intraassay and interassay coefficients of variation of 3.5% and 5.1%, respectively. Each saliva sample was assayed twice, and analyses were repeated if any pair of results differed by > 7%. Cortisol values were strongly skewed so a log transformation was applied prior to analysis (Singer & Willett, 2003). In light of the diurnal pattern of cortisol, all analyses of cortisol data controlled for time of day (Smyth et al., 1997).

Self-reported negative emotions

Adolescents reported on the emotions they experienced during the family conflict by completing the 14-item Post-Discussion Questionnaire (PDQ; Spies, Margolin, Susman, & Gordis, 2011) immediately after the family conflict discussion. Eight of these items (angry, frustrated, anxious, irritated, embarrassed, ignored, upset, and afraid) were summed to create a negative emotion score (Cronbach's alpha = .85). Negative emotion scores were strongly skewed so a log transformation was applied prior to analysis (Singer & Willett, 2003).

Age

Age was measured in months using parent-reported birthdate.

Puberty

Pubertal status was assessed using parent report on three items that assess the degree of growth spurts, skin changes, and body hair growth on a 4-point scale (Saxbe & Repetti, 2009). Higher scores reflect more advanced pubertal status, and the sample mean of 6.42 (SD = 1.45) indicates that adolescents were midway through puberty on average.

Data Reduction and Analysis

All hypotheses were tested using multiple regression. Hypotheses 1 and 2 were tested using regressions run in Stata, Release 11 (StataCorp, 2009); hypotheses 3 and 4 were tested using multilevel models run in Hierarchical Linear Modeling, Version 7 (Raudenbush, Bryk, & Congdon, 2010). All predictors were grand mean centered prior to analysis and creation of interaction terms. Separate models were run using age and pubertal status to control for adolescent's physical development; results from the two sets of models are identical in terms of the direction and significance of predictors of interest. Results for models using age are presented in this article; results for models using pubertal status are available from the first author. Time of day was included as an additional covariate in models involving cortisol. Separate models were run for cortisol and for self-reported negative emotions for all hypotheses.

Results

Table 2 presents means and standard deviations of as well as correlations between all variables. As expected, f₀ range was significantly correlated with total cortisol output for boys, r(30) = 0.45, p = .012; a correlation of similar magnitude, yet nonsignificant, emerged for girls, r(26) = 0.38, p = .055. Similarly, f₀ range was significantly correlated with self-reported negative emotions for girls, r(26) = 0.40, p = .043; a correlation of similar magnitude, yet nonsignificant, emerged for boys, r(30) = 0.32, p = .082. Also as expected, time-to-peak f₀ range was significantly correlated with total cortisol output for girls, r(26) = 0.49, p = .010. F₀ range was significantly correlated with boy's pubertal status, r(30) = 0.50, p = .007. Finally, total cortisol output was significantly correlated with girl's age, r(26) = −0.42, p = .034. No other correlations emerged as significant.

Table 2.

Descriptive Statistics for Aggregate F₀ Range, Time-to-Peak F₀ Range, Negative Emotion, Cortisol, and Covariates

	Correlations
	F0 range	Time-to-peak f0 rangesec	Negative emotionln	Total cortisol outputln	Age	Pubertal status	Mean	SD
F0 range	—	.26	.40*	.38†	−.17	.02	155.35	15.44
Time-to-peak f0 rangesec	.09	—	.01	.49*	.05	−.07	474.16	241.49
Negative emotionln	.32†	.13	—	.06	.08	.02	1.83	1.00
Total cortisol outputln	.45*	−.17	.31†	—	−.42*	−.21	3.74	0.42
Age	−.27	.00	.05	−.20	—	.11	15.27	0.84
Pubertal status	.50**	.20	.32†	.05	.10	—	7.04	1.49
Mean	90.23	504.15	1.77	3.83	15.16	5.89
SD	10.18	247.81	0.99	0.92	0.63	1.20

Note. Descriptive statistics are presented for log transformed self-reported negative emotion and cortisol variables and for females above the diagonal/males below the diagonal.

^†p < .10.

^*p < .05.

^**p < .01.

We examined Lowess smoothed plots of f₀ range for participants in the current study to see if they appeared to have linear and oscillatory characteristics. Figure 1 displays these plots for one boy and one girl from the current study. F₀ range appears to have both linear and oscillatory components and to differ between individuals as well as over the course of the discussions. Standard linear growth models and dynamic oscillator models therefore appear to be reasonable methods for characterizing pitch contours in this sample.

Figure 1.

Lowess smoothed plots of f₀ range over time.

Associations Among Aggregate F₀ Range, Time-to-Peak F₀ Range, Self-Reported Emotion, and Cortisol

Table 3 presents results of regressions used to test hypothesis 1 where aggregate f₀ range was regressed onto each predictor, sex, the sex by predictor interaction, and covariates. As depicted in Figure 2, greater aggregate f₀ range was significantly associated with higher levels of self-reported negative emotions, β = 0.14, p = .007, and with greater cortisol output, β = 0.20, p = .009. The main effect of sex was also significant in both models, β = 0.93, p < .001, indicating that girls' f₀ range was significantly higher than boys'.

Table 3.

Regression Analyses of Aggregate F₀ Range and Time-to-Peak F₀ Range With Self-Reported Negative Emotions and Total Cortisol Output

	Aggregate f0 range						Time-to-peak f0 range
	Self-reported negative emotions			Total cortisol output			Self-reported negative emotions			Total cortisol output
Variable	B	SE B	β	B	SE B	β	B	SE B	β	B	SE B	β
Intercept	122.97	1.64	—	123.24	1.67	—	489.57	33.63	—	498.02	31.74	—
Sex	64.78***	3.31	0.93	65.81***	3.35	0.94	−31.74	67.43	−0.07	−19.64	63.60	−0.04
Predictor	4.77**	1.69	0.13	9.69**	3.54	0.20	15.93	34.44	0.06	144.24*	67.23	0.43
Sex × predictor	3.31	3.37	0.05	7.74	6.65	0.08	−30.33	68.76	−0.06	360.34**	126.37	0.54
Age	−3.92	2.29	−0.08	−1.12	2.46	−0.02	7.58	46.70	0.02	45.33	46.82	0.14
Time of day	—	—	—	0.65	0.88	0.04	—	—	—	2.96	16.81	0.03

^*p < .05.

^**p < .01.

^***p < .001.

Figure 2.

Plots of summary associations between f₀ range, cortisol, and self-reported negative emotions.

Table 3 also presents results of regressions used to test hypothesis 2 where time-to-peak f₀ range was regressed onto each predictor, sex, the sex by predictor interaction and covariates. A significant main effect emerged for cortisol on time-to-peak, β = 0.43, p = .037. However, this main effect was qualified by a significant cortisol by sex interaction, β = 0.54, p = .006. Computing simple slopes for this interaction indicates that greater cortisol is associated with later time-to-peak for girls only, B = 324.41, p = .009, and that no significant association exists between cortisol and time-to-peak for boys, B = 35.93, p = .49. None of the predictors were significant in the model predicting time-to-peak from self-reported negative emotions.

Linear Trend in F₀ Range Over Time

Table 4 presents results of multilevel models used to test hypothesis 3 where f₀ range was regressed onto the main effects and interactions for time, sex, emotional arousal, and covariates. The equations below describe these models:

• Level-1:
  $${\mathrm{F}}_0={\pi }_{0\mathrm{i}}+{{\pi }_{1\mathrm{i}}}^{\ast }\left(\text{time}\right)+{\mathrm{e}}_{\mathrm{ti}}$$
• Level-2:
  $${\pi }_{0\mathrm{i}}={\beta }_{00}+{{\beta }_{01}}^{\ast }\left(\text{sex}\right)+{{\beta }_{02}}^{\ast }\left(\text{predictor}\right)+{{\beta }_{03}}^{\ast }(\text{sex}\times \text{predictor})+\text{covariates}+r_{0\mathrm{i}}$$
  $${\pi }_{1\mathrm{i}}={\beta }_{10}+{{\beta }_{11}}^{\ast }\left(\text{sex}\right)+{{\beta }_{12}}^{\ast }\left(\text{predictor}\right)+{{\beta }_{13}}^{\ast }(\text{sex}\times \text{predictor})+\text{covariates}+r_{1\mathrm{i}}$$

Table 4.

Regression Table for Linear Model of F₀ Range Over Time

	Self-reported negative emotions			Total cortisol output
Variable	B	SE B	β	B	SE B	β
Intercept, β00	63.01	88.42	—	18.11	101.07	—
Predictor, β01	7.89	4.58	0.12	9.05	8.48	0.07
Sex, β02	68.39***	9.18	0.57	69.32***	9.16	0.82
Predictor × sex, β03	2.08	9.20	0.02	−3.02	17.03	−0.02
Age, β04	3.23	5.83	0.04	6.06	6.29	0.07
Time of day, β04	—	—	—	0.11	2.34	0.00
Time, β10	0.14	0.09	0.57	0.05	0.11	0.20
Predictor, β11	0.01	0.00	0.08	0.02*	0.01	0.09
Sex, β12	−0.01	0.01	−0.04	0.00	0.01	0.00
Predictor × sex, β13	0.00	0.01	0.00	0.03	0.02	0.10
Age, β14	−0.01	0.01	−0.56	0.00	0.01	0.00
Time of day, β14	—	—	—	0.00	0.00	0.00

Note. Robust standard errors are reported.

^*p < .05.

^***p < .001.

A significant main effect for sex emerged on the intercept in both models (β = 0.57, p = .001). This effect indicates that girls' f₀ range was greater than boys' f₀ at the beginning of the discussions. Figure 3 depicts the significant cross-level interaction between cortisol and time (β = 0.82, p = .040 for cortisol). A similar trend emerged for self-reported negative emotions (β = 0.12, p = .070). These effects are in the hypothesized directions (i.e., higher cortisol or more self-reported negative emotions, greater increase in f₀ range over time). No other main effects or interactions emerged as significant.

Figure 3.

Plots of linear effects of time on f₀ range for males and females in the top and bottom quartiles of cortisol output.

Oscillatory Behavior of f₀ Range Over Time

Table 5 displays results of dynamic oscillator models used to test hypothesis 4, and the equations below describes these models:

• Level-1:
  $${\mathrm{x}}_{\mathrm{ij}}^{"}\left({\text{change in slope of f}}_0\right)={{\eta }_{\mathrm{ix}}}^{\ast }{\mathrm{x}}_{\mathrm{ij}}\left({\mathrm{f}}_0\right)+{{\xi }_{\mathrm{ix}}}^{\ast }{\mathrm{x}}_{\mathrm{ij}}^{’}\left({\text{slope of f}}_0\right)+{\mathrm{e}}_{\mathrm{ij}}$$
• Level-2:
  $${\eta }_{\mathrm{ix}}={\mathrm{c}}_{00}+{{\mathrm{c}}_{01}}^{\ast }{\left(\text{predictor}\right)}_{\mathrm{i}}+{{\mathrm{c}}_{02}}^{\ast }{\left(\text{sex}\right)}_{\mathrm{i}}+{{\mathrm{c}}_{03}}^{\ast }{(\text{predictor}\times \text{sex})}_{\mathrm{i}}+\text{covaritates}+{\mathrm{u}}_{0\mathrm{i}};$$
  $${\xi }_{\mathrm{ix}}={\mathrm{c}}_{10}+{{\mathrm{c}}_{11}}^{\ast }{\left(\text{predictor}\right)}_{\mathrm{i}}+{{\mathrm{c}}_{12}}^{\ast }{\left(\text{sex}\right)}_{\mathrm{i}}+{{\mathrm{c}}_{13}}^{\ast }{(\text{predictor}\times \text{sex})}_{\mathrm{i}}+\text{covaritates}+{\mathrm{u}}_{1\mathrm{i}};$$

Table 5.

Regression Table for Dynamic Oscillator Models of F₀ Range

	Self-reported negative emotions			Total cortisol output
Variable	B	SE B	β	B	SE B	β
Current level of f0 (η), c00	−0.02***	0.00	−0.12	−0.02***	0.00	−0.12
Predictor, c01	0.00	0.00	0.00	0.00	0.00	0.00
Sex, c02	−0.01	0.00	−0.02	−0.01	0.00	−0.02
Predictor × sex, c03	0.00	0.00	0.00	0.02*	0.01	0.04
Age, c04	0.00	0.00	0.00	0.00	0.00	0.00
Time of day, c05	—	—	—	−0.00	0.00	0.00
Slope of f0 (ξ), c10	−1.14***	0.01	−0.92	−1.14***	0.01	−0.92
Predictor, c11	0.00	0.02	0.00	−0.03	0.02	−0.02
Sex, c12	0.05*	0.03	0.02	0.05*	0.02	0.02
Predictor × sex, c13	0.00	0.03	0.00	−0.08	0.05	−0.02
Age, c14	0.02	0.01	0.25	0.00	0.01	0.02
Time of day, c15	—	—	—	0.01	0.01	0.00

Note. Robust standard errors are reported.

^*p < .05.

^**p < .01.

^***p < .001.

At level-1, the second derivative of f₀ range (${\mathrm{x}}_{\mathrm{ij}}^{"}$), which represents change in the slope of f₀ range (or the acceleration of f₀ range), is predicted by current level of f₀ range, x_ij, and the slope of f₀ range, ${\mathrm{x}}_{\mathrm{ij}}^{’}$. Random effects, u_0i and u_1i, are included at Level-2 to allow for between youth differences in associations.

A baseline version of this model including only intercepts and random effects at level-2 was run prior to addition of any predictors to allow examination of the basic oscillatory contours of f₀ over time. Significant η (β = −0.12, p < .001) and ξ (β = −0.92, p < .001) parameters emerged in this model. The significant η parameter indicates that arousal varies in a regulated fashion with the greatest values of ${\mathrm{x}}_{\mathrm{ij}}^{"}$ occurring when x_ij is largest. The significant ξ parameter must be combined with η to be interpreted. Because they are of the same sign, this parameter indicates that youth's f₀ range returns more rapidly to a set-point later in the discussion. The random effects for both η and ξ emerged as significant (χ²(55) = 239.16, p < .001 for u₀₀; χ²(55) = 142.56, p < .001 for u₁₀), indicating significant remaining between-youth variability.

When main effects, interactions, and covariates were added to this model, a significant cortisol by sex interaction emerged on η (β = 0.04, p = .05). Computing simple slopes for this interaction indicated that greater cortisol is associated with a more rapid return to set-point of f₀ range for boys only (B = −0.01, p = .02). No significant association exists between cortisol and return to set-point of f₀ range for girls (B = 0.01, p = .19). A significant sex effect on ξ also emerged in both models (β = 0.02, p = .05). Computing simple slopes for this interaction indicated that though both boys' and girls' f₀ range varied less about their set-points later in the discussions, the effect is larger for boys (B = −1.17, p < .001) than for girls (B = −1.12, p < .001). Combining the significant sex by cortisol effect on η with the sex effect on ξ indicates that boys' f₀ range varies less about their set-points than does girls' and that when boys' f₀ range does vary from their set-point, it returns to it faster when boys have a greater cortisol output.

Discussion

This study examined associations between adolescents' f₀ range, cortisol output, and self-reported negative emotions during the interpersonally and situationally significant context of family conflict discussions. When examining aggregate measures of arousal, results linked higher levels of f₀ range to higher levels of cortisol output and self-reported subjective negative emotions for boys and girls. When examining time-varying measures of arousal, results found higher levels of cortisol output to be related to steeper linear increases in f₀ range over time for boys and girls (with a similar trend emerging for self-reported negative emotions); additionally, higher cortisol output was linked to a slower time-to-peak f₀ range for girls. Finally, exploratory analyses revealed that boys' f₀ range oscillates to a significantly lesser degree than girls' and that this effect is magnified when boys have higher cortisol output. This collection of results provides empirical support for the use of f₀ range as a measure of adolescents' emotional arousal. It also adds to evidence describing how internally experienced arousal is expressed in multiple aspects of adolescents' emotional expression and highlights sex differences in temporal dimensions of vocal behavior.

Range of Fundamental Frequency as a Measure of Adolescents' Emotional Arousal During Interpersonally Significant Interaction

The results of this study help to clarify how emotional arousal is encoded in several aspects of adolescents' f₀ range. At the most coarse level of granularity, greater f₀ range is associated with greater cortisol output and higher levels of subjective negative arousal; at a finer level of analysis, greater cortisol output is associated with increased f₀ range over time. In contrast, short-term oscillations in f₀ range generally were not associated with either cortisol output or subjectively experienced arousal. Taken together, these results indicate that f₀ range, whether measured in the aggregate or over time, can be interpreted as an index of encoded emotional arousal whereas moment-to-moment variation in f₀ range is not strongly tied to adolescents' emotional arousal.

There are several possible explanations for why short-term fluctuations in f₀ range were not associated with cortisol output and subjectively experienced arousal. One possibility is that variations in f₀ range may convey additional information beyond the speaker's emotional arousal. For example, mean f₀ generally rises at the end of a question and it is possible that variations in f₀ range are also used to convey syntax. A second possibility is that adolescents' f₀ range may be affected by their parents' vocal behavior. Numerous studies (e.g., Gregory & Hoyt, 1982; Gregory, Webster, & Huang, 1993) find significant convergence in interactants' vocal characteristics (including mean f₀) and for the level of convergence in mean f₀ to be an index of the perceived quality of the interaction (Gregory et al., 1993). It is therefore likely that adolescents' f₀ range is related to their parents' f₀ range and that the strength of that effect varies across families based on the quality of their interactions. A final possibility is that variability in f₀ range about the mean is influenced by other aspects of emotion, such as emotion regulation (e.g., Butler & Gross, 2009). Indirect support for this final possibility can be derived from the sex differences that emerged in the temporal characteristics of f₀ range.

Sex Differences in Temporal Characteristics of Range of Fundamental Frequency

Sex differences in the temporal characteristics of f₀ range were generally consistent with hypotheses and have implications for emotional functioning. The slower return to set-point for girls relative to boys is consistent with Germans Gard and Kring's (2007) finding that women's emotional response to negative stimuli typically lasts longer than does men's response. Part of the reason why girls' peak f₀ range occurred later than boys' may be because girls' emotional responses were more sustained than boys, as evidenced by girls' slower return to set-point relative to boys. Girls' slower return to set-point may have resulted in a form of emotional escalation where girls' subsequent responses occurred before earlier responses subsided.

Boys displayed significantly less variability in f₀ range than did girls and this difference was magnified when boys' cortisol output was high. Specifically, when boys' cortisol output was high they returned to baseline levels of f₀ range more quickly than did girls. This finding may be similar to men's tendency to display less variability in facial affect when they are more physiologically aroused during marital conflict (Gottman, Levenson, & Woodin, 2001). Boys' tendency to display less variability in vocal affect when they have greater cortisol output may reflect a form of suppression or emotional dampening. In adults, suppression of overall nonverbal affect has been linked to heightened physiological arousal (e.g., Gross & Levenson, 1993; Butler et al., 2003). Though the current study did not include a measure of suppression, boys are more likely to try to control emotional expression during communication than are adolescent girls (e.g., Brody, 1999) and adolescent boys display less emotion in family interactions than do adolescent girls (Brody, 1996). Additionally, youth report using suppression significantly more often than other forms of emotional regulation and report using suppression at significantly higher levels than do adults (John & Gross, 2004). F₀ range is open to a degree of conscious control; yet, whether boys' restricted variability of f₀ range is an indication of suppression awaits future investigation.

Temporal Characteristics of Range of Fundamental Frequency Shared by Boys and Girls

In addition to uncovering sex differences in the temporal characteristics of f₀ range, results also revealed that linear increases in f₀ range are similarly related to cortisol output for boys and girls. More specifically, steeper increases in f₀ range are associated with greater cortisol output during the conflict discussion. To the best of our knowledge, this evidence is the first demonstration that any form of temporal variability in f₀ range is indicative of physiological arousal during interpersonal interaction.

In considering this finding, it is important to take the different time course and sampling rates of f₀ range and cortisol into account. F₀ range changes on a second by second basis whereas changes in cortisol output typically take 15 to 20 minutes to register in saliva. Likewise, f₀ range was sampled at a much higher resolution (second by second) than was cortisol (twice over a 10-minute period). These aspects of the two measures and of the study methodology have implications for understanding the results of the current study. The results of the current study indicate that linear trends in f₀ range over a conflictual family discussion are associated with cortisol output measured across the same period of time. This finding can be interpreted as showing that total cortisol output is reflected in the slope of f₀ range over time. It is important to note that this finding cannot be interpreted to indicate that cortisol and f₀ range are associated at any specific point in time. Rather, it shows that overall trends in cortisol output and overall linear trends in f₀ range are associated.

It is also important to note that these sex consistent findings in temporal course occurred within the context of significant sex differences in f₀ range observed throughout the family discussion. Girls' f₀ range was significantly higher than boys' f₀ range at the beginning of the discussion, and this sex difference persisted throughout the remainder of the interaction. This sex difference may be an artifact of increased differentiation in male and female voices during puberty. Alternatively, it may also be an indication that adolescents were aroused when they began the family discussions. It is well documented that sex differences in vocal characteristics, such as f₀ range, are amplified in emotionally arousing contexts (e.g., Juslin & Scherer, 2005). It is possible that adolescents already were anticipating the conversation would go poorly and were aroused as a consequence. However, this sex difference at the start of the family discussion does not appear to reflect a differential association with physiological arousal. Additionally, there does not appear to be a sex differences in the association between the rate of change of f₀ range and cortisol output during the corresponding time period. Figure 3 depicts these sex differences in the intercept and linear trends in f₀ range across time for boys and girls in the top and bottom quartiles of cortisol output. This pattern of findings further highlights the robustness of the association between general trends in f₀ range and overall cortisol output, as well as the importance for controlling for sex differences in mean levels when examining longitudinal variation in f₀ range.

Limitations

There are several limitations to bear in mind when considering the results of the current study. First, one measure of physiological arousal, cortisol, was examined. Patterns of physiological response to interpersonal stressors are complex and variable both between individuals as well as between physiological systems within individuals (Cacioppo, Tassinary, & Berntson, 2007). Although it is possible to conclude that physiological arousal measured by cortisol response is encoded in f₀ range, replication of these findings using other physiological measures (e.g., heart rate or electrodermal activity) is warranted. Second, the summary nature of the cortisol output and self-reported negative emotions measures may have impacted study findings. Part of the reason that stronger evidence emerged for encoding of cortisol output and self-reported negative emotions at the aggregate level of f₀ range may be that all three measures were summary measures. It is possible that different patterns would emerge if more variable measures of physiological arousal and self-reported negative emotions were examined. Third, only one measure of vocally encoded arousal, f₀ range, was examined. Other vocal parameters related to fundamental frequency (e.g., mean f₀ and standard deviation of f₀) as well as related to other parts of the speech signal (e.g., intensity and speech rate) have been found to convey emotional information. Examination of a wider range of vocal parameters would allow for a more detailed analysis of multiple ways that cortisol and self-reported negative emotions are encoded in the voice. Finally, parents' vocal behavior was not considered. Though it is likely that parents' f₀ range is related to children's f₀ range, there is little reason to think that parents' f₀ range alters associations between children's f₀ range, cortisol output, and self-reported negative emotions.

Summary and Future Directions

The current study finds strong evidence that cortisol output (and to a lesser extent self-reported negative emotions) is encoded in adolescent's f₀ range and highlights the value of examining both summary and time-varying indices of vocal behavior. The current study also identifies several avenues for future exploration of emotion and family processes. Given the lack of empirical investigation of vocal behavior relative to other forms of emotional expression (such as facial expressions), exploration of associations between f₀ range and arousal in other physiological systems would be valuable. It would also be valuable to explore other predictors that may account for remaining variability in adolescent's f₀ range. In addition to proximal variables, such as parents' vocal behavior, there are a number of distal family characteristics known to impact important emotional processes in children such as domestic aggression (Gordis, Margolin, & John, 1997) and metaemotion (Gottman, Katz, & Hooven, 1997) that may also relate to adolescent's vocal behavior. F₀ range therefore appears not only to offer a valid and nonintrusive measure for emotional arousal during social interaction but also a sensitive measure for better understanding important emotional processes in families.