Studying the intensity of and temporal variability in emotional states is especially relevant in research involving adolescents, since they are at a developmental stage characterized by numerous internal and external changes that may produce fluctuations in mood states (Hardy, Bukowski, & Sippola Reference Hardy, Bukowski and Sippola2002). This instability in relation to dynamic psychological processes may play an important role in some forms of psychopathology especially with regard to internalizing problems (Helbig-Lang, Lang, Petermann, & Hoyer, Reference Helbig-Lang, Lang, Petermann and Hoyer2012). Fluctuations in mood states have been associated with negative life outcomes (Salum, Desousa, do Rosário, Pine, & Manfro,2013), drug use (Comulada, Lightfoot, Swendeman, Grella, & Wu, Reference Comulada, Lightfoot, Swendeman, Grella and Wu2015), and stress (Walsh, Basu, & Monk, Reference Walsh, Basu and Monk2015). The intensity of negative mood states has also been linked to alcohol (Ahonen, Nebot, & Giménez, 2017; Martínez-Hernáez, Marí-Klose, Julià, Escapa, & Marí-Klose, Reference Martínez-Hernáez, Marí-Klose, Julià, Escapa and Marí-Klose2012) and tobacco use (Martínez-Hernáez, Marí-Klose, Julià, Escapa, Marí-Klose, & DiGiacomo, Reference Martínez-Hernáez, Marí-Klose, Julià, Escapa and Marí-Klose2012).

However, despite the importance of reliably assessing the intensity of and temporal variability in mood states, the latter are not always well captured by traditional retrospective assessment methods, which tend to be the most commonly used. Regarding the intensity of psychological symptoms, most self-report surveys require respondents to consider their previous mood state after varying lengths of time have elapsed. This kind of retrospective appraisal is subject to potential bias, since it may be influenced by both the quality of memory and the impact of recent life events (Ebner-Priemer & Trull, Reference Ebner-Priemer and Trull2009; Raselli & Broderick, Reference Raselli and Broderick2007; Trull & Ebner-Priemer, Reference Trull and Ebner-Priemer2009). Thus, adolescents may overestimate their problems and this would lead them to report higher levels of psychological symptoms when assessed using conventional questionnaires. Margraf, Taylor, Ehlers, Roth, and Agras (Reference Margraf, Taylor, Ehlers, Roth and Agras1987) noted three decades ago that the number of symptoms reported in questionnaire surveys was higher than in diaries, suggesting that symptom reports tend to be exaggerated retrospectively. A study by Mokros (Reference Mokros1993) showed a clear discrepancy between the reporting of psychological symptoms during a week and the recall of these symptoms at the end of the week.

As regards emotional variability over time, it is argued that psychopathology is best seen as a dynamic and constantly changing construct (DeVries, Reference DeVries1992). However, the dynamics of psychological symptoms “has been neglected for some time, likely because of the inability of cross-sectional and retrospective reports to uncover the ebb and flow of symptoms” (Ebner-Priemer, Eid, Kleindienst, Stabenow, & Trull, Reference Ebner-Priemer, Eid, Kleindienst, Stabenow and Trull2009, p. 195), and this seriously limits the ability to accurately characterize, understand, and change behavior in real-world settings.

Ecological momentary assessment (EMA) offers a way of overcoming these problems, since it involves the repeated assessment of mood and behavior in the here and now, in both natural settings and real-time, thus allowing the dynamics of and fluctuations in psychological states to be observed over a certain period of time (Shiffman, Stone, & Hufford, Reference Shiffman, Stone and Hufford2008). Indeed, one of the most promising applications of EMA lies in the possibility of detecting fluctuations over time (Wenze & Miller, Reference Wenze and Miller2010). EMA methods were developed partly in response to the limitations of retrospective recall, which is hampered by the fact that our recollections are not simply inaccurate but may often be systematically biased (Shiffman et al., Reference Shiffman, Stone and Hufford2008). For example, people are more likely to retrieve negatively valenced information when they are in a negative mood (Clark & Teasdale, Reference Clark and Teasdale1982). This implies that there will be discrepancies between EMA-based and recall-based assessments of the same period (Shiffman et al., Reference Shiffman, Stone and Hufford2008).

In adolescent samples, EMA has been used to evaluate a wide variety of problematic behaviors associated with everyday life, including drug use (Comulada et al., Reference Comulada, Lightfoot, Swendeman, Grella and Wu2015), smoking relapse (Shiffman et al., Reference Shiffman, Balabanis, Gwaltney, Paty, Gnys, Kassel and Paton2007), and binge eating (Wegner et al., Reference Wegner, Smyth, Crosby, Wittrock, Wonderlich and Mitchell2002). Silk, Steinberg, and Sheffield Morris (2003) found that the intensity and lability of levels of emotion in everyday life were associated with elevated depressive symptoms and poor adjustment in adolescents. Accordingly, children and youth with major depressive disorder reported both more intense and labile negative affect than did a control group (Silk et al., Reference Silk, Forbes, Whalen, Jakubcak, Thompson, Ryan and Dahl2011). A study by Tan et al. (Reference Tan, Forbes, Dahl, Ryan, Siegle, Ladouceur and Silk2012) found that in real contexts and during everyday life anxious youth do not experience more intense momentary negative emotions than do control youths. However, the variability of emotions has been reported to predict the development of anxiety disorder symptoms (Neumann, van Lier, Frijns, Meeus, & Koot (Reference Neumann, van Lier, Frijns, Meeus and Koot2011)) and emotional distress (Aldinger et al., Reference Aldinger, Stopsack, Ulrich, Appel, Reinelt, Wolff and Barnow2014). The systematic review by Waltz, Nauta, and Aan Het Rot (2014) of 34 studies using EMA highlighted its benefits for the study of anxiety disorders, including generating insight into the temporal variability of symptoms.

These research findings underline the importance of early detection in relation to the intensity of and temporal variability in the mood states of adolescents, especially given their propensity to experience internalizing problems (McLaughlin, Hilt, & Nole-Hoeksema (Reference McLaughlin, Hilt and Nolen-Hoeksema2007). While there are important differences in the prevalence of such problems by country (Austin & Chorpita, Reference Austin and Chorpita2004) some multinational studies indicate that Spanish community adolescents show the highest rate of internalizing problems (Rescorla et al., Reference Rescorla, Achenbach, Ivanova, Dumenci, Almqvist, Bilenberg and Verhulst2007). Ahonen et al. (2017) stated that around 13% of Spanish boys and 21% of Spanish girls reported negative mood sates.

In light of the above, the current study uses smartphone-based EMA to assess in a community sample of Spanish adolescents both the intensity of and variability in three kinds of internalizing psychological problem. In a previous study that examined the use of EMA with more or less the same sample as used here, we showed that it is a feasible method with community adolescents (Magallon-Neri, Kirchner-Nebot, Forns-Santacana, Calderon, & Planellas, 2016). Our aim here is to explore some of our earlier data in greater depth, specifically as regards the role of migratory status, the level of psychological problems, and temporal variability in psychological problems. The overall objective was to analyze adolescents’ daily perceptions about sadness/depression, anxiety, and somatic problems in real-time and in an ecological setting so as gain greater insight into their day-to-day emotional experiences. This general objective was broken down into two specific aims. The first was to examine the perceived day-to-day intensity of the three kinds of internalizing symptom (i.e., sadness/depression, anxiety, and somatic problems), as reported using smartphone-based EMA, and second, to analyze the reported levels according to gender, migratory status (Spanish nationals vs. Latin American immigrants), and level of psychological problems (normal vs. risk groups). Our hypothesis was that the immigrant population, the at-risk population, and girls would report higher levels of these internalizing emotional states. As regards the immigrant population, adapting to a new culture and to the changes that follow from having left one’s country of origin are factors that may impact on the emotional states of adolescents, heightening their experience of psychological problems, especially those of the internalizing kind (Janssen et al., Reference Janssen, Verhulst, Bengi-Arslan, Erol, Salter and Crijnen2004) such as depression (Fandrem, Sam, & Roland, Reference Fandrem, Sam and Roland2009) and somatizations (Vieno, Santinello, Lenzi, Baldassari, & Mirandola, Reference Vieno, Santinello, Lenzi, Baldassari and Mirandola2009). We also expected to find an association between the sadness/depression, anxiety, and somatic states recorded using EMA and the internalizing symptoms reported via conventional paper-and-pencil questionnaires such as the Youth Self Report (YSR) (Achenbach & Rescorla, Reference Achenbach and Rescorla2001). More specifically, we hypothesized that participants with high levels of symptoms according to the YSR would also report higher levels of problems via EMA, as compared with those participants who showed a low level of symptoms on the conventional measure. However, we also expected to find that the EMA method would indicate lower overall levels of problems than would the conventional paper-and-pencil questionnaire, due to the fact that the bias effect associated with the former will be lower (Shiffman et al., Reference Shiffman, Stone and Hufford2008).

With respect to gender, previous research using traditional methodology indicates that males and females differ in their levels of specific types of psychological problems, with females reporting higher levels of internalizing symptoms (Ahonen et al., Reference Ahonen, Nebot and Giménez2007; Forns et al., Reference Forns, Kirchner, Gómez-Maqueo, Landgrave, Soler, Calderón and Magallón-Neri2015; Zahn-Waxler, Shirtcliff, & Marceau, Reference Zahn-Waxler, Shirtcliff and Marceau2008). In our previous aforementioned study (Magallon-Neri et al., Reference Magallón-Neri, Kirchner-Nebot, Forns-Santacana, Calderón and Planellas2016) we observed a higher level of sadness/depression, anxiety, and somatic concerns among girls. As the present sample is almost the same, similar results are expected.

The second aim was to examine via EMA the variability in sadness/depression, anxiety, and somatic states across a one-week period. Once again, we sought to analyze the results in relation to gender, migratory status, and level of psychological problem. The hypothesis here was that females, the immigrant population, and adolescents in the risk group would show greater variability in their sadness, anxiety, and somatic states over the period of one week. Previous studies using EMA have reported that males and females differ in the mean level and variability of internalizing problems (Rusby, Westling, Crowley, & Light, Reference Rusby, Westling, Crowley and Light2013; Silk et al., Reference Silk, Steinberg and Sheffield Morris2003), with girls tending to show greater day-to-day fluctuations in problems of this kind. Likewise, adolescents at risk of psychological problems are more sensitive to everyday stressors (Palmier-Claus et al., Reference Palmier-Claus, Myin-Germeys, Barkus, Bentley, Udachina, Delespaul and Dunn2011), and they show greater lability of negative emotions, especially sadness (Silk et al., Reference Silk, Forbes, Whalen, Jakubcak, Thompson, Ryan and Dahl2011).

Method

Participants

The convenience sample comprised 90 students (61.1% girls) recruited from two high schools in Barcelona and ranging in age from 12 to 18 years (M = 14.61, SD = 1.64). In terms of their migratory status, 68.5% were Spanish nationals and 31.5% were first-generation immigrants from Latin America (6.9% from Argentina, 13.8% from Bolivia, 3.5% from Chile, 10.3% from Colombia, 34.5% from Ecuador, 3.5% from Paraguay, 10.3% from Peru, and 17.2% from the Dominican Republic). From among the immigrant school population we selected exclusively those from Spanish-speaking Latin America so as to ensure that they shared a common language with their Spanish peers. This sample was used in order to address the first aim. For the analysis underpinning the second aim, and in line with the recommendations of various authors (Chen, Bundy, Cordier, Chien, & Einfield, Reference Chen, Bundy, Cordier, Chien and Einfield2015; Palmier-Claus, Dunn, & Lewis, Reference Palmier-Claus, Dunn and Lewis2012), we excluded from this sample those students whose response rate using the smartphone device was less than 33% of the overall response rate, which in this study equated to ≤ 13 assessment moments (range 1–35). After applying this criterion, the sample for this final analysis comprised 86 adolescents (58.8% girls) with an age range between 12 and 18 years (M = 14.52, SD = 1.64). In terms of their migratory status, 67.1% were Spanish nationals and 32.9% were first-generation immigrants from Latin America. Attrition analysis revealed no significant differences between excluded and included participants on the main independent study variables: gender, χ²(1, N = 90) = .340, p > .05; migratory status, χ²(1, N = 90) = 3.43, p > .05; level of psychological symptoms, χ²(1, N = 90) = .149, p > .05.

Measures

Sociodemographic data and covariates. A survey sheet was created to gather basic data on age, gender, school year, and migratory status.

Smartphone app based on EMA. Using EMA in conjunction with a smartphone app (Acer Liquid Z–200 based on Android 4.0) we assessed three kinds of problems associated with the internalizing spectrum: sadness/depression (Right now, do you feel sad or depressed?), anxiety (Right now, do you feel nervous or worried?), and somatic problems (Right now, do you have a headache or feel sick?). These items were chosen to reflect the content of the items with the highest factor loading on the corresponding DSM-Oriented scales of the YSR (all above 0.74). Each item was scored on a 5-point Likert-type scale, ranging from 1 (not at all) to 5 (very much). The assessment schedule was five times per day (between 9 a.m. and 9 p.m.), semi-randomly, for seven days (35 possible moments for each participant). Participants were required to respond within three minutes of hearing the reminder alarm from the app. If the user did not begin to answer within this period of time the app stopped the alarm and blocked the unit. Regarding reliability of the EMA in this study, the overall Cronbach’s coefficient for the three items related to internalizing problems (sadness/depression, anxiety, and somatic problems) was (α = .61). If only the sadness/depression and anxiety items were considered, the alpha was .70 (for more details, see Magallón-Neri et al., Reference Magallón-Neri, Kirchner-Nebot, Forns-Santacana, Calderón and Planellas2016). Figure 1 shows app screenshots for the smartphone app used in the current study.

Figure 1. App Screenshots of Smartphone.

Internalizing problems. The traditional paper-and-pencil instrument used to establish the normal and at-risk groups was the Spanish version (Abad, Forns, & Gómez, Reference Abad, Forns and Gómez2002) of the Youth Self-Report (YSR) (Achenbach, & Rescorla, Reference Achenbach and Rescorla2001, Reference Achenbach and Rescorla2007). This questionnaire measures psychological distress in children and adolescents between 11 and 18 years old through a list of 112 items that represent emotional and behavioral problems. Each item is scored on a 3-point scale: 0 (not true), 1 (somewhat or sometimes true), or 2 (very or often true). For the present study we used only the Internalizing scale, and we segmented the initial sample (n = 88) into two groups according to the T scores obtained, labeling the two groups as normal group (T score ≤ 59) and at-risk group (T score ≥ 60), following the criteria of Achenbach and Rescorla (Reference Achenbach and Rescorla2001). Cronbach’s alpha for the Internalizing scale in the present sample was .86.

Procedure

We began by contacting six state schools in two districts of the metropolitan area of Barcelona; one of these districts had a large immigrant population, while the other had a typical proportion of immigrants (13.8%). After explaining the nature of the research to the management boards of the schools, two of the six agreed to participate in the study. The low acceptance rate was due to concerns that school heads had about the possible impact that EMA might have on classroom management or on the dynamic between participating and non-participating adolescents, and/or because some parents did not wish to be responsible for their child being in possession of a smartphone for a week. In the two schools that agreed to take part we hosted an information session for adolescent students, encouraging them to participate and giving them an informed consent form to be signed by their parents or legal guardians; the latter were also sent a letter setting out the aims of the study and encouraging them and their children to take part. Once permission had been obtained we administered the paper-and-pencil questionnaire in sessions lasting approximately 90 minutes. In these same sessions the use of the smartphone devices and the nature of EMA were explained. Upon delivery of a device to each participant, he or she was assigned an individual alphanumeric code so as to protect their identity in case of loss or theft of the device. It was explained to participants that they would have the smartphone for a whole week and that the device would ring five times a day at random times; they were asked to respond on as many occasions as possible each day. Each participant signed an agreement regarding the care and proper use of the smartphone device, and they were given contact details should they encounter any technical problems during the experiment. Technical assistance was requested by 5.6% of participants, mainly due to a malfunction of the device or app.

Data analysis

The present study fulfills the majority of technical characteristics that must be met by studies using EMA methodology and which are stipulated in the Checklist for Reporting EMA studies (CREMAS) (Liao, Skelton, Dunton, & Bruening, Reference Liao, Skelton, Dunton and Bruening2016). Data were analyzed using SPSS 23, Mplus 5.1, and STATA 14.2. Regarding the intensity of sadness/depression, anxiety, and somatic problems, significant differences between the groups for each comparison (i.e., by gender, migratory status, and risk level) were examined using the Student’s T test for independent samples, after first applying Levene’s test for equality of variances. Cohen’s d was calculated as a measure of the effect size. Differences between the different groups in the percentages of responses for sadness/depression, anxiety, and somatic problems were tested with χ², applying the Monte Carlo exact test when needed and calculating the standardized adjusted residuals.

Regarding variability, trajectories were estimated with longitudinal growth curve models (Singer & Willett, Reference Singer and Willett2003). Growth models allow for flexible handling of time and account for non-independence from repeated measures on each participant. In addition to the usual linear regression parameters, growth models can have a random intercept, capturing the variability between participants in starting points (in our models, individual differences in baseline affective, anxiety, and somatic symptoms) and random slopes, and capturing the variability between participants in change over time (in our models, the trajectories of affective, anxiety, and somatic symptoms). The models also allow the intercept and slope to be correlated, which indicates the degree of correlation between an individual’s starting point and his/her change over time. For example, a strong positive correlation would indicate that the higher a participant’s baseline level of affective symptoms, the more this level would increase over time.

Several models were tested. First, we estimated an unconditional model to characterize the overall trajectory of affective, anxiety, and somatic symptoms. All models included a random intercept and time slope that were allowed to correlate. Time was coded as 0, 1, 2, and so on. Second, we ran a conditional model, entering time, gender, migratory status, and level of psychological problems into all models. Finally, we tested whether each covariate interacted with time (time x gender, time x migratory status, and time x level of psychological problems). If the interaction term was not significant, we dropped it and report estimates from the final model without the interaction. Effects were considered statistically significant at p < .05.

Results

Intensity of Subjective Perceptions of Momentary Sadness/Depression, Anxiety, and Somatic States

Table 1 shows mean raw scores for momentary ratings of sadness/depression, anxiety, and somatic problems according to gender, migratory status (Spanish nationals vs. Latin American immigrants), and level of psychological problems (normal T < 60 on YSR vs. risk groups T ≥ 60 on YSR). Girls obtained significantly higher mean raw scores than boys for momentary sadness/depression, t = 8.75; p < .001, d = 0.41; anxiety t = 5.06; p < .001, d = 0.4; and somatic problems, t = 6.02; p < .001, d = 0.28. However, Cohen’s d indicated a small effect size in all cases. Similarly, the risk group obtained significantly higher mean raw scores than the normal group for sadness/depression, t = 4.40; p < .001, d = 0.26); anxiety, t = 6.33; p < .001 d = 0.43; and somatic problems, t = 7.14; p < .001, d = 0.53. Here, the effect sizes (Cohen’s d) were between small and moderate. Migratory status was not associated with any significant differences in sadness/depression, t = 1.69; p > .05; anxiety, t = 1.58; p > .05; or somatic problems, t = 1.75; p > .05.

Table 1. Frequencies and Descriptive of Momentary Sadness/Depression, Anxiety and Somatic Problems. Counts and Percentages of Momentary Responses, Means and SD of EMA Raw Scores according to Gender, Migratory Status and Level of Psychological Problems according to YSR

Notes:

(a) This percentage corresponds to the number of responses/ number maximum of potential responses (n = 1225)

(b) This percentage corresponds to the number of responses/ number maximum of potential responses (n = 1925)

(c) This percentage corresponds to the number of responses/ number maximum of potential responses (n = 2135)

(d) This percentage corresponds to the number of responses/ number maximum of potential responses (n = 1015)

(e) This percentage corresponds to the number of responses/ number maximum of potential responses (n = 2625)

(f) This percentage corresponds to the number of responses/ number maximum of potential responses (n = 525)

Despite these differences, however, and as shown in Table 1, the mean scores for sadness/depression, anxiety, and somatic problems fall between a minimum value of 1.17 and a maximum of 1.69 (range 1–5; mode = 1), independently of gender, migratory status, and level of psychological problems. At the qualitative level, the percentage of responses with a value of 1 on the Likert scale (i.e., not at all) was 80.0% for sadness/depression, 79.3% for anxiety, and 84.9% for somatic problems. The percentage of responses with a value of 2 was 9.6% for sadness/depression, 9.8% for anxiety, and 6.7% for somatic problems. The percentage of responses with a value of 5 (i.e., very much) was 1.7%, 1.4%, and 1.7% respectively. The values 3 and 4 in no case exceeded 8% of respondents. These data indicate a low rate of internalizing problems reported over a week by these adolescents.

Notably, the value of 1 (not at all) was the most often reported by adolescents classified as at risk according to the YSR (69.9% for sadness/depression, 61.4% for anxiety, and 70.2% for somatic problems), although the proportions were significantly lower than those for their peers in the normal group: χ²(4, 1936) = 33.83, p < .001 (standardized adjusted residual = 5.0) for sadness/depression; χ²(4, 1936) = 73.21 p < .001 (standardized adjusted residual = 8.4) for anxiety; and χ²(4, 1932) = 119.08, p < .001 ( standardized adjusted residual = 8.3) for somatic problems. In addition, the risk group gave, in comparison with the normal group, a significantly higher proportion of responses with a value of 5 (very much): 3.8% (standardized adjusted residual = 3.3), 3.5% (standardized adjusted residual = 3.37), and 7.1% (standardized adjusted residual = 8.4) for sadness/depression, anxiety, and somatic problems, respectively. The fact that the risk group and the normal group, as classified by means on the Internalizing scale of the YSR, differed significantly in their EMA ratings for sadness/depression, anxiety, and somatic problems supports the convergent validity of the latter method.

One-Week variability in internalizing problems

Model for sadness/depression problems. First, we tested a growth model with time, but no covariates, in order to characterize the overall trajectory. For this baseline model the estimate for the intercept was b = 1.42 and the random effect was σ_intercept = .33, indicating that the overall mean intercept was 1.42 and the standard deviation of individual subject variability about that mean was 0.33. The overall time effect was statistically significant, b = 0.0004, 95% CI [0.0001, 0.0008], p = .001, but not meaningful.

The random effect for the time slope was σ_time = .011, which is small relative to the mean, suggesting that there is no substantial individual variability in the linear time slope. The estimated correlation between the random intercept and the time slope parameters was –.73, indicating that subjects who were above the mean at baseline tended to have more negative time slopes.

Next, we tested whether each of the covariates predicted the trajectory of sadness/depression. There was a significant effect of gender, b = .21, 95% CI [0.19, 0.23], p < .001, suggesting that girls self-reported more affective symptoms than did boys. Furthermore, participants with psychological problems experienced more affective symptoms over the course of the study period, b = .09, 95% CI [0.06, 0.11], p < .001. Although the interactions of time x migratory status and time x level of psychological problems reached statistical significance, the associated effect sizes indicated that they were not meaningful.

Model for anxiety problems. For anxiety, we tested an unconditional growth model characterizing the overall trajectory. For this model, the estimate for the intercept was b = 1.45 and the random effect was σ_intercept = .33, indicating that the overall mean intercept was 1.45 and the standard deviation of individual subject variability about that mean was 0.33. Though the overall time effect was statistically significant it was unimportant, b = –.001, 95% CI [0.0001, 0.006], p < .001. The random effect for the time slope was σ_time = .012, suggesting a small individual variability in the linear time slope. The estimated correlation between the random intercept and the time slope correlation parameters was –.61, indicating a more negative slope for subjects who were above the mean at baseline.

A conditional model was then tested, adding covariates into a second model. Here we did find a significant effect of gender, b = .08, 95% CI [0.06, 0.09], p < .001; migratory status, b= –.07, 95% CI [–0.09, –0.05], p < .001; and level of psychological problems, b = .18, 95% CI [0.16, 0.21], p < .001. Specifically, girls, Latinos, and participants self-reporting psychological problems experienced more anxiety over the study period. However, although all the interactions tested were statistically significant, the effect sizes indicated that they were not meaningful.

Model for somatic problems. An unconditional model with no covariates was tested first. For this model, the estimate for the intercept was b = 1.28 and the random effect was σ_intercept = .37, indicating that the overall mean intercept was 1.28 and the standard deviation of individual subject variability about that mean was 0.37. The overall time effect was not meaningful, b = .004, 95% CI [0.0004, 0.005], p < .001. There was a small individual variability in the linear slope (σ_time = .012). The estimated correlation between the random intercept and the time slope was –.44, indicating that participants who were above the mean had a more negative slope.

Finally, we used a conditional model to test the contribution of covariates to the prediction of somatic problems. Here we found a significant effect of gender b = .08, 95% CI [0.06, 0.09], p < .001); migratory status, b = –.04, 95% CI [–0.09, –0.05], p < .001; and level of psychological problems, b = .15, 95% CI [0.12, 0.17], p < .001. Specifically, girls, Latinos, and participants self-reporting psychological problems experienced more somatic problems over the study period. Although the interactions tested reached statistical significance, the effect sizes indicated that they were not meaningful (see Table 2).

Table 2. Longitudinal Growth Curve Models for Variability of Sadness/Depression, Anxiety and Somatic Problems