Despicable me Masselink, Maurits
DOI:
10.33612/diss.102140763
IMPORTANT NOTE: You are advised to consult the publisher's version (publisher's PDF) if you wish to cite from it. Please check the document version below.
Document Version
Publisher's PDF, also known as Version of record
Publication date:
2019
Link to publication in University of Groningen/UMCG research database
Citation for published version (APA):
Masselink, M. (2019). Despicable me: self-esteem and depressive symptoms among adolescents and young adults. Rijksuniversiteit Groningen. https://doi.org/10.33612/diss.102140763
Copyright
Other than for strictly personal use, it is not permitted to download or to forward/distribute the text or part of it without the consent of the author(s) and/or copyright holder(s), unless the work is under an open content license (like Creative Commons).
Take-down policy
If you believe that this document breaches copyright please contact us providing details, and we will remove access to the work immediately and investigate your claim.
Downloaded from the University of Groningen/UMCG research database (Pure): http://www.rug.nl/research/portal. For technical reasons the number of authors shown on this cover page is limited to 10 maximum.
Download date: 27-06-2021
5
CHAPTER 5
Dynamic Relationships Between Self- esteem, Pleasure, Sadness, and Social Experiences During Daily Life
Masselink, M.
Bennik, E.C.
Oldehinkel, A.J.
aan het Rot, M.
Jeronimus, B.F.
Van Roekel, E.
Submitted for publication
ABSTRACT
Concurrent and temporal associations between self-esteem, social interaction, and sadness and lack of pleasure (the two core symptoms of depression) were investigated during daily life. Two independent Ecological Momentary Assessment studies (Dataset 1: N = 69, Mean age = 21 years, 80% female; Dataset 2: N = 790, Mean age = 39 years, 84% female) were used in which self- esteem, sadness, pleasure, social interaction quantity (Dataset 1) and social interaction appraisal (Dataset 2) were measured three times a day during thirty days. A Dynamic Structural Equation Model (DSEM) estimated concurrent and temporal associations simultaneously. Significant concurrent associations were found between self-esteem, sadness, pleasure, time spent talking, time spent alone, and the desire to be alone when in social company. Only the concurrent associations between sadness and time spent talking / time spent alone did not reach the smallest effect of interest (β = 0.10). Pleasure showed stronger concurrent associations with self- esteem and social interaction variables than sadness. With respect to temporal associations, only the temporal association from pleasure to self-esteem was equal to our smallest effect size of interest. On the one hand, our findings suggest an interplay between concurrent self-esteem, pleasure, sadness and social experiences, but, on the other hand, suggest few associations over six-hour measurement intervals. Future research may test temporal associations between social interaction, self-esteem, and depression across different time intervals.
5 Low self-esteem is frequently mentioned in relation to depression. Studies investigating the
temporal order of the association between self-esteem and depression generally support a vulnerability model in which low self-esteem is viewed as a precursor of depression (Abramson & Metalsky, 1989; Beck, 1967; Metalsky, Joiner, Hardin, & Abramson, 1993; see for a meta-analysis Sowislo & Orth, 2013). Other studies support the scar model in which low self- esteem is a consequence of depression (Shahar & Davidson, 2003; Sowislo & Orth, 2013), or report bidirectional associations that support both models (Steiger et al., 2015). Bidirectional associations would align with the cognitive reactivity hypothesis, which postulates reciprocity between depressed mood and negative cognitions about the self (Teasdale, 1988).
The bidirectional relation between low self-esteem and depression may be (partially) explained by the quantity and subjective appraisal of social interactions. This role of social interactions would align with a wide literature on the interpersonal origins of depression (e.g, Brown, Andrews, Harris, Adler, & Bridge, 1986; Hames, Hagan, & Joiner, 2013; Segrin, 2011; Watson & Andrews, 2002). Different theories highlight the role of belongingness needs, loneliness (Cacioppo, Grippo, London, Goossens, & Cacioppo, 2015), and social acceptance, and social rank (Price, Sloman, Gardner, Gilbert, & Rohde, 1994). More specifically, the sociometer theory connects self-esteem with social interactions. Sociometer theory postulates the so-called sociometer as social warning system, via which self-esteem decreases after social rejection and increases after social acceptance (Leary, 2005). Reductions in state self-esteem after a perceived social rejection, are postulated to motivate efforts to improve interpersonal relationships (Leary
& Baumeister, 2000).
Several experimental studies found that social rejection and acceptance can result in changes in self-esteem (Leary, 2005). The premise that low self-esteem activates behavior aimed at restoring social belongingness is crucial to sociometer theory, but lacks a firm empirical basis.
Instead, low self-esteem seems to be associated with social avoidance, which may lead to a further decrease in self-esteem (Marshall et al., 2014; Reitz, Motti-Stefanidi, & Asendorpf, 2016;
Srivastava & Beer, 2005; Stinson et al., 2008). A similar dynamic pattern has been found for the relation between depression and social avoidance and rejection (Allen & Badcock, 2003; Nolan et al., 2003; Sentse et al., 2010). Depressed mood typically co-occurs with hypersensitivity to social threat and the inhibition of risk-seeking social behaviors (Allen & Badcock, 2003). While mild depressed mood is expected to motivate adaptive behaviors that prevent social loss, more severe depression levels can lead to withdrawal from social interactions (Moulds, Kandris, Starr, &
Wong, 2007; Rubin, Burgess, & Coplan, 2011), even with close others (Brown, Strauman, Barrantes- Vidal, Silvia, & Kwapil, 2011). Social withdrawal may subsequently increase depressed mood even further via fewer positive social experiences (Steger & Kashdan, 2009). Also, more interpersonal difficulties and social rejection (Coyne, 1976; Joiner, 1999) predict higher depressive symptom levels (Kendler, Hettema, Butera, Gardner, & Prescott, 2003; Slavich et al., 2010).
In sum, there appear to be dynamically patterned associations between self-esteem, depressed mood, and the quality and quantity of social interaction (e.g., Brown et al., 1986).
However, this literature is predominantly based on prospective associations over months and years at the group level. In contrast, studies of intra-individual dynamic processes between self- esteem, depressed mood, and social interactions at the scale of hours and days in people in their natural context remain scarce (see Liu, Xie, & Lou, 2018 for a review), but are essential to investigate these patterned temporal associations within persons (Fisher et al., 2018).
Ecological momentary assessment (EMA) is a research method in which self-report data are measured at many points in time in participants’ everyday lives. This offers highly appealing advantages, such as reduced retrospective memory biases, high ecological validity and the ability to reliably examine temporal associations within persons (Aan het Rot, Hogenelst, &
Schoevers, 2012; Bolger, Davis, & Rafaeli, 2003; Shiffman et al., 2008). EMA studies have found a link between self-esteem and depression in never-depressed pregnant women (Franck et al., 2016) and between self-esteem and positive and negative affect in first-year students (Nezlek &
Plesko, 2003). An EMA study of the prospective associations between changes in self-esteem and depression during the day, reported that self-esteem predicted depressive symptoms in 35% of their participants (Clasen et al., 2015), whereas 28% of participants reported a reverse association, and 7% a bidirectional association, and in 30% of the participants self-esteem and depression showed no temporal relationship. These EMA results are in line with the aforementioned literature and underscore the substantial individual differences in the link between self-esteem and depression.
EMA studies also supported the link between the quantity and quality of social interactions and depressed mood (Nezlek, Imbrie, & Shean, 1994; Steger & Kashdan, 2009) or between social interactions and self-esteem (Denissen, Penke, et al., 2008). However, the observed associations were small in size and no day-to-day temporal associations were found between state self- esteem and social interaction quantity or quality with friends or family (Denissen, Penke, et al., 2008). Importantly, the unique relation between state self-esteem, depression symptoms (i.e.
sadness and lack of pleasure) and the quantity and subjective appraisal of social interaction remain virtually unknown, because previous studies did not include all variables in one model.
THIS STUDY: SELF-ESTEEM, SOCIAL INTERACTIONS, AND DEPRESSION
Using two EMA datasets we aimed to unravel the concurrent and temporal associations between self-esteem, depressive symptoms, and social interaction variables during daily life. We focused on sadness and a lack of pleasure (anhedonia) because they are the core symptoms of Major Depressive Disorder (MDD; DSM-5; American Psychiatric Association, 2013) which are thought to represent high negative affect (NA) and low positive affect (PA), respectively.
5 In EMA Dataset 1, we focused on concurrent and temporal (i.e. lagged) associations among
self-esteem, two measures of social interaction quantity (i.e. time spent talking and time spent alone), and sadness and lack of pleasure as symptoms of depression. Based on the literature linking self-esteem to social avoidance, we hypothesized lower self-esteem to be concurrently and temporally associated with more time alone and less time talking. Further, based on the cognitive reactivity hypothesis, self-esteem was hypothesized to be negatively related to sadness and positively to pleasure –concurrently and temporally . Furthermore, sadness was hypothesized to be negatively related to social interaction quantity whereas pleasure was positively related to social interaction quantity. All of the temporal effects were hypothesized to be present in both directions.
In Dataset 2, we aimed to test the same predictors but use appraisal of social interaction as the outcome variable, instead of interaction quantity. We hypothesized lower self-esteem and higher levels of sadness to be concurrently and temporally associated with a stronger desire to be alone when in company and a weaker desire to be in company while being alone. Inverse relationships were expected for pleasure. All of the temporal effects were again hypothesized to be present in both directions.
DATASET 1 METHOD
Participants and procedure
Data for this study came from the control group (N = 69) of the No Fun No Glory Study (van Roekel, Masselink, et al., 2016), which were collected between April and July 2015. The No Fun No Glory study consisted of a large screening survey among 2,937 young adults between 18 and 24 years old from the north of the Netherlands. Detailed sampling information is provided elsewhere (van Roekel, Masselink, et al., 2016). The No Fun No Glory study has been registered in the Dutch Clinical Trial Register (NTR5498) and was approved by the Medical Ethical Committee from the University Medical Center Groningen (no. 2014/508).
Out of this sample, participants with persistent anhedonia were selected for participation in an EMA protocol, together with a control group without anhedonia, who were matched on age, educational level and sex. Exclusion criteria for participation were inability to complete an electronic questionnaire three times a day; professional treatment for psychiatric problems; use of psychotropic medication; epilepsy; and pregnancy. As the No Fun No Glory study included a skydive intervention for some participants, there were a few additional exclusion criteria:
loose prostheses; height of more than two meters; weight of more than 95 kg; inability to raise one’s legs 90 degrees; cardiovascular complaints or problems; significant visual or hearing
impairments, and prior experience with skydiving, bungee jumping, or base jumping7. The control group participants included in the EMA protocol had to report an at least moderately high pleasure level (i.e. > 50th percentile), which was rated as similar to or higher than what was normal for that individual. Pleasure level was assessed with an item from the Domains Of Pleasure Scale (Masselink, van Roekel, Heininga, Vrijen, & Oldehinkel, 2019). An accompanying item assessed how this level compared to the participant’s normal pleasure level. The 114 control group participants received an information letter and were invited to participate in the No Fun No Glory study via email, of whom one participant did not meet the inclusion criteria, 21 declined participation, 22 did not respond, and one participant dropped out, resulting in a total sample of 69 participants (mean age = 21 years, standard deviation = 2, 80% female). Most of the participants reported a high educational level (59%, middle 38%, low 3%).
After participants provided written informed consent, an introductory meeting was scheduled with a research assistant in the University Medical Center Groningen. During this meeting exclusion criteria were checked, the study procedure was explained, and all EMA items were discussed with the participant to ensure understanding. Subsequently, their EMA schedule started with three measures a day, six hours apart (e.g., 9:00h, 15:00h, 21:00h). The timing of the EMA schedule was determined in consultation with the participant. At each designated time point, participants received a text message with a link to the online questionnaire on their smartphone, if desired also via email. Participants were instructed to complete the questionnaires on their smartphone, and not on tablets, laptops or desktop computers. The questionnaire had to be completed within two hours after the first prompt, and reminders were send after one and 1.5 hours. It took most participants on average three minutes to complete the questionnaire.
After 30 days, participants returned to the University Medical Center Groningen for a debriefing session and the questionnaire prompts were stopped. Because most participants had the debriefing meeting a few days after the 30 day-period, the data collection period lasted 32 days on average (range 28-36 days). The participants received 75 euros for completing at least 80% of the EMA questionnaires and filling in two monthly questionnaires not relevant for the present study. Adherence was excellent, with on average 92.79% (SD = 4.83) of all questionnaires completed, or on average 89 questionnaires per participant (SD = 6.91, range 68-101). Although measurements were six hours apart, the exact interval ranged between 4-8 hours due to the allowed time window of 2 hours to fill in the questionnaires.
7. Only 12.8% (N=376) of the 2,937 participants initially screened was unwilling to perform a tandem skydive. This group did not differ significantly from those who were unsure about their willingness to perform a tandem skydive (N=802) or willing to perform a tandem skydive (N=1,759) in terms of sex, age, trait positive affect or depressive symptoms (V.E. Heininga, Ahles, Van Roekel, Mezulis, & Oldehinkel, 2017).
5 Materials
All EMA variables (see Table 1) were measured using Visual Analogue Scales (VAS) on which response scales ranged from 0 to 100. Self-esteem and time spent alone were measured over the period since the last measurement. The amount of time spent alone was only asked when the participant indicated having been alone since the previous measurement occasion, otherwise this variable was constrained to zero. During morning assessments, momentary pleasure and sadness were measured, whereas during afternoon and evening assessments pleasure and sadness were measured retrospectively (i.e. since the previous measurement occasion).
Statistical analyses
The pre-registered hypotheses and analyses of this study are provided on the Open Science Framework (https://osf.io/db647). Prior to running the analyses, we ran Monte Carlo simulation power analyses to check whether the power would be sufficient to detect expected and relevant effects. We specified a sample size of 69 individuals each providing 90 measurements, resulting in a total of 6210 simulated observations. We assumed standardized effects of 0.20 for the autoregressive effects and 0.10 for the cross-lagged and concurrent effects8. The effect size of 0.10 corresponds to what we consider the Smallest Effect Size Of Interest (SESOI) to be interpreted (Lakens, 2014). This SESOI of 0.10 is comparable to the commonly used threshold for a small correlation coefficient (Cohen, 1992). The autoregressive effect of 0.20 is not a SESOI but has previously been reported for self-esteem (Denissen et al., 2008; Masselink et al., unpublished manuscript), and we generalized this magnitude to all autoregressive effects. The results of the power analyses showed that our study would be highly powered (~100%) to find the specified autoregressive and cross-lagged effects (scripts and output provided on https://osf.io/db647).
Linear trends were removed from all variables for each individual in the dataset (using Stata 15) because trends can lead to inflated autocorrelations (Rovine & Walls, 2006). Hypotheses were tested using Dynamic Structural Equation Modelling (DSEM) of the Mplus 8.2 program. A multivariate DSEM multilevel model was fit (Level 1: assessments, Level 2: individuals) adjusted for unequal time intervals between measures by specifying a time interval of 6 hours using the Mplus TINTERVAL option. Model convergence was established by inspecting the Potential Scale Reduction (PSR) value. A stable PSR value below 1.02 was used to indicate convergence (Brown, 2015; Hoofs, van de Schoot, Jansen, & Kant, 2017). We planned to run a minimum of 20.000 iterations, and to double the iterations each time consistent convergence was not reached. In our DSEM analyses variables were automatically within-person centered when lagged variables were included. In the same random slope and random intercept model all variables were included as dependent variables and predicted by their lagged value and the lagged values
8. The specified effect of .10 for the concurrent associations was accidently left out of the preregistration.
TABLE 1. Item Description of EMA items used in Datasets1 and 2 ConceptEnglish translationOriginal Dutch questionResponse range (scoring 0-100) Dataset 1 Self-esteemI was pleased with myselfIk was tevreden over mezelf“Not at all” to “Very much” SadnessI feela / felt gloomybIk voela / voelde me somberb“Not at all” to “Very much” PleasureI feel pleasure at this momenta/ I experienced pleasure since the last assessmentbIk voel me nu plezieriga / Ik heb sinds de laatste meting plezier gehadb“Not at all” to “Very much” Time spent aloneTime I spent aloneTijd die ik alleen heb doorgebracht“Very little” to “Very much” Time spent talkingHow much have I been talking to other peopleHoeveel heb ik gepraat met andere mensen“Not at all” to “Very much” Dataset 2 Self-esteemI feel valuedIk voel me gewaardeerd“Not at all” to “Very much” I feel confidentIk voel me zelfverzekerd“Not at all” to “Very much” I feel I fall shortcIk heb het gevoel tekort te schietenc“Not at all” to “Very much” SadnessI feel gloomyIk voel me somber“Not at all” to “Very much” PleasureI feel cheerfulIk voel me opgewekt“Not at all” to “Very much” Desire to be alonedI would rather have been with othersdIk was liever in gezelschap geweestd“No, preferably not” to “Yes, certainly” Desire to be in companyI would rather have been aloneIk zou liever alleen zijn geweest“No, preferably not” to “Yes, certainly” Note. a Asked in the morning. b Asked in afternoon and evening. c Removed from scale for the analyses. d Not included in analyses due to too many missing data.
5 of all other variables. All random effects were correlated with each other and with the fixed
effects. The concurrent associations were based on the correlations between the residuals of the temporal associations.
We controlled for multiple testing by applying a False Discovery Rate (FDR) method (Benajmini & Hochberg, 1995), separately to the temporal and concurrent results. To calculate the corrected significant threshold an alpha was set (.05) and all p-values were ranked from low to high. For each ranked test, a FDR threshold was calculated using the formula:
FDR derived significance threshold = 0.05
number of tests / ranking
The FDR threshold belonging to the lowest ranked significant p-value which had a p-value below its FDR threshold was used as the new cut-off to determine significance.
RESULTS
Means, between-person SDs, and within-person SDs of the EMA items are depicted in Table 2.
Before running the DSEM models, we calculated the Intra Class Correlations (ICC) of the five variables in the model. ICCs indicate how much of the variance is explained by between-person differences, and consequently 1-ICC indicates how much of the variance is explained by within- person fluctuations. Within-person fluctuations explained 56% of the variance of self-esteem, 68% of the variance of sadness, 70% of the variance of pleasure, 75% of the variance of time spent talking, and 87% of the variance of time spent alone. The DSEM model converged well:
after 2000 iterations the PSR remained below 1.01 and it was 1.001 after 20.000 iterations.
Concurrent associations
The standardized results of the concurrent associations are listed in Table 3. The FDR-derived significant threshold remained at .05. All concurrent associations were significant (all p≤ .004).
Only the association of sadness with time spent alone and the association of sadness with time spent talking were below the SESOI of 0.10, which means all but two of the concurrent hypotheses could be confirmed.
TABLE 2. Means, Standard Deviations, and Within-person Standard Deviations in Datasets1 and 2 Overall mean (SD)Between-person SDWithin-person SD1 - intraclass correlation Dataset 1 (N = 69) Self-esteem64.48 (14.87)9.8911.13.56 Sadness12.70 (13.19)7.4810.89.68 Pleasure63.54 (17.00)9.2914.26.70 Time spent alone34.31 (31.57)11.3629.46.87 Time talking52.35 (23.33)11.7720.16.75 Dataset 2 (N = 938) Self-esteem58.37 (17.71)13.6711.54.42 Sadness23.19 (21.57)15.1715.40.51 Pleasure55.69 (20.47)12.6516.30.62 Desire to be aloned25.43 (22.44)12.9318.13.66 Desire to be in company36.52 (26.74)18.7817.28.46 Note. d Not included in analyses due to too many missing data.
5
TABLE 3. Concurrent Standardized Associations Between Variables in Dataset 1 and 2 Self-esteem β [95% CI]Sadness β [95% CI]Pleasure β [95% CI]Time spent talking β [95% CI] Dataset 1 Sadness-0.26 [-0.29, -0.24] Pleasure0.39 [0.37, 0.42]-0.29 [-0.31, -0.26] Time spent talking0.20 [0.17, 0.22]-0.07 [-0.10, -0.05]0.24 [0.21, 0.26] Time spent alone-0.14 [-0.16, -0.11]0.04 [0.01, 0.07]-0.19 [-0.21, -0.16]-0.47 [-0.49, -0.45] Dataset 2 Sadness-0.47 [-0.48, -0.46] Pleasure0.60 [0.60, 0.61]-0.52 [-0.53, -0.51] Desire to be alone-0.30 [-0.31, -0.29]0.24 [0.23, 0.25]-0.29 [-0.30, -0.28] Note. Standard deviation of the posterior was 0.01 for all estimates in Dataset 1, and 0.003-0.005 in Dataset 2. Values in bold were significant at p < .01, all other values significant at p< .001
TABLE 4. Temporal Standardized Associations Between Variables in Dataset 1 and 2 Self-esteemSadnessPleasureTime spent talkingTime spent alone β [95% CI]p-valueβ [95% CI]p-valueβ [95% CI]p-valueβ [95% CI]p-valueβ [95% CI]p-value Dataset 1a Self-esteem t-10.16 [0.13, 0.20]<.001-0.06 [-0.10, -0.03] .0010.09 [0.05, 0.12]<.0010.05 [0.02, 0.08].004-0.04 [-0.08, -0.00].036 Sadness t-1-0.06 [-0.09, -0.03].0020.13 [0.10, 0.17]<.001-0.05 [-0.09, -0.01].0140.01 [-0.03, 0.05].5040.01 [-0.04, 0.05].834 Pleasure t-10.09 [0.06, 0.13]<.001-0.04 [-0.07, -0.00].0400.07 [0.03, 0.10]<.0010.05 [0.01, 0.09].012-0.04 [-0.08, -0.01].026 Time spent Talking t-1-0.01 [-0.05, 0.03].638-0.02 [-0.06,0.02].2680.04 [0.01, 0.08].0260.08 [0.05, 0.12]<.001-0.03 [-0.06, 0.01].184 Time spent alone t-1-0.03 [-0.07, 0.00].0500.00 [-0.03, 0.04].852-0.08 [-0.12, -0.05]< .001-0.08 [-0.11, -0.04]<.0010.19 [0.15, 0.22]< .001 Dataset 2b Self-esteem t-10.16 [0.14, 0.17]<.001-0.06 [-0.08, -0.05]<.0010.06 [0.05, 0.08]<.001-0.03 [-0.05, -0.02]<.001 Sadness t-1-0.06 [-0.07, -0.05]<.0010.15 [0.14, 0.17]<.001-0.08 [-0.09, -0.07]<.0010.03 [0.02, 0.05]<.001 Pleasure t-10.10 [0.08, 0.11]<.001-0.10 [-0.12, -0.09]<.0010.16 [0.14, 0.17]<.001-0.06 [-0.08, -0.04]<.001 Desire to be alone t-1-0.01 [-0.03, 0.00].1400.02 [0.001, 0.03].044-0.00 [-0.02, 0.01].6300.09 [0.07, 0.10]<.001 Note. For all associations the standard deviation of the posterior was 0.02 in Dataset 1 and 0.01 in Dataset 2 a The false discovery rate derived significance threshold was .032 b The false discovery rate derived significance threshold was .041
5 Temporal associations
The results of the analyses investigating the temporal associations are listed in Table 4. The FDR- derived significance threshold for the temporal associations was .032, thus only smaller p-values were considered significant. Apart from most autoregressive results, all significant temporal associations were below the SESOI of 0.10, indicating that our temporal hypotheses could not be confirmed.
DATASET 2 METHODS
Participants and procedure
The HowNutsAreTheDutch (HND) crowdsourcing sample was recruited from the general population of the Netherlands by means of radio broadcasts, television, local podium discussions, newspapers, and magazines. Participants were invited to visit the website www.HoeGekIs.nl and report on their mental health in a cross-sectional study and/or a longitudinal EMA study. The HND study protocol was assessed by the Medical Ethical Committee of the University Medical Center Groningen and exempted from review under the Medical Research Involving Human Subjects Act (in Dutch: WMO), because it concerned a non-randomized open study targeted at anonymous volunteers in the general public (registration number M13.147422).
All details on procedures, participants, and measures are provided elsewhere (van der Krieke et al., 2016, 2017). Participants completed the electronic EMA questionnaires in their natural environments, three times a day for 30 days, resulting in a maximum of 90 assessments per individual. Assessments were prompted at equidistant time points with a six-hour interval in between, with the exact time points depending on participants’ sleep-wake schedule.
Participants received a text message on their mobile phone with a link to a questionnaire. They were asked to fill out the questionnaire immediately after the alert, or, if impossible, within one hour, after which the questionnaire could no longer be accessed. This study includes the 938 individuals who completed the EMA protocol between May 22nd, 2014 (launching date of the EMA study) and December 31th, 2017 (end of fourth-year of the study). The inclusion criteria for the present study were an age of at least 18 years and providing at least two responses on all study variables (1185 started the EMA study, 247 participants who provided insufficient data were excluded). The mean number of EMA responses was 56.72 (SD = 25.57). The distribution of the number of completed questionnaires was bi-modal as many participants dropped out in the first 20 assessment moments, possibly because they signed up out of curiosity, rather than
out of the motivation to finish the whole measurement period (see van der Krieke et al., 2017, for a detailed analysis). Another substantial part, around 38% of participants, filled in at least 75% of the measurements.
The sample was 84% female, and the mean age was 39 years (SD = 13). Most participants were highly educated (77%, middle educational level 13%, low educational level 2%, other form of education or no information available 9%). Of all participants, 98% had the Dutch nationality, 2% had another nationality (mostly Belgian).
Measures
The items used to assess self-esteem, pleasure, sadness, and social company are outlined in Table 1. Pleasure was operationalized as “cheerfulness” but is henceforth described as pleasure.
The desire to be alone variable was only measured when participants had been in company of others for the most time since the last measurement moment. The desire to be in company, was only measured when participants had been alone for the most time since the last measurement moment. For self-esteem we calculated the within-person reliability while controlling for the dependence of measurements over multiple assessments as described by Nezlek (2017). Across the three items, the within-person reliability was .41, but removing the negatively worded item
“I feel I fall short” raised it to .51. We therefore constructed a mean self-esteem score using the two positively worded items only.
Statistical analyses
The hypotheses and analyses of this study were pre-registered on the Open Science Framework (https://osf.io/hmnxw). We repeated the analytical strategy as used for Dataset 1. After individual linear trends were removed the reciprocal associations between self-esteem, pleasure, sadness, and desire to be alone or in company were tested in a DSEM model. We used a TINTERVAL of six hours between measurements to control for unequal time intervals caused by the delay in filling in the questionnaires and the night interval.
Previous EMA studies on the HND data showed that individuals filling in more questionnaires did not differ in terms of personality (van der Krieke et al., 2017) or on the tested EMA associations (Snippe et al., 2018). Nevertheless, we performed a sensitivity analysis in which we compared results of those who filled in 75% or more of the measurements with those who filled in less.
RESULTS
Means, between-person SDs, and within-person SDs of the EMA items are depicted in Table 2.
Within-person fluctuations explained a substantial percentage of variance in self-esteem (42%), sadness (51%), pleasure (62%), desire to be in company (46%) or to be alone (66%).
5 As the DSEM model with all five variables did not converge we had to deviate from our
preregistered plan of analysis. Non-convergence may reflect missing data on the social appraisal variables, especially on the desire to be in company, which, on average, had missing data on 72 occasions per participant. This data was missing either via missed assessments or because participants had not been alone. A better converging model was derived by removing the desire to be in company variable from the model, and simplifying the model by removing the correlations between the random effects (leaving only the correlations among the fixed effects). The PSR remained below 1.10 after 3000 iterations, and stabilized between 1.07-1.08 after iterations 58000-80000, with a PSR value of 1.076 after the last iteration. This value is below the default and commonly used convergence criterion of 1.10 (Gelman, Carlin, Stern, & Rubin, 2014), but higher than the stricter PSR < 1.02 criterion that we preregistered.
Concurrent associations
The results of the concurrent associations are listed in Table 3. All concurrent results were significant at p<.001, in the direction as hypothesized and above the SESOI, meaning that all the concurrent hypotheses were supported.
Temporal associations
The temporal associations are listed in Table 4. The FDR-derived significant threshold for the temporal associations was .041. Apart from three of the four autoregressive results, two temporal associations were significant and equal to the SESOI. Pleasure had a positive effect on self-esteem (β = 0.10) and a negative effect on sadness (β = -0.10).
Sensitivity analyses
Repeating the trimmed model with 454 participants who completed ≥ 75% of all assessments converged well (40.000 iterations, which fluctuated between a PSR of 1.015 -1.025 after the first 20000 iterations up to PSR= 1.017 after the last iteration) and supported similar conclusions as those in the complete model using all observations (see Supplementary Table S1 and Table S2 for details).
POST HOC ANALYSES IN DATASETS 1+2
As can be seen in Table 3, the concurrent associations between self-esteem and pleasure (β = 0.39 and 0.60, respectively) appeared stronger than the concurrent associations between self- esteem and sadness (β = 0.26 and 0.47, respectively). Because significant effect size differences might shed an interesting light onto the association between self-esteem and depression, we
extended our models with a difference coefficient (standardized absolute self-esteem/sadness coefficient – standardized absolute self-esteem/pleasure coefficient). In both datasets, the concurrent association between self-esteem and pleasure was significantly stronger than the association between self-esteem and sadness (both β = 0.13, p< .001).
Because all temporal associations were close to the SESOI their differences were not investigated.
DEVIATIONS FROM PRE-REGISTRATIONS
Our analyses deviated from the two preregistrations in three ways. First, we failed to pre- register our FDR corrections for multiple testing to prevent type I errors. The specification of our SESOI may mitigate concerns somewhat as different correction methods do not impact the interpretation of the results. Second, our preregistration accidently omitted prior knowledge of the data; the research questions in Dataset 1 were partly tested in the context of a summer school on network analyses in 2016. A description of these exploratory analyses can be found in the supplement (S3). The Dataset 2 variables self-esteem and the desire to be alone or to be in company have been used in a master thesis to examine the relation between state self-esteem and the social company variables using univariate multi-level analyses supervised by the second author. However, the results presented in the master thesis did not reach the first author who drafted the design and fit the models in the current study. Third, reliability and convergence issues necessitated modifications of the measures, PSR criterion, and model tested in Dataset 2, as outlined in the results and limitation sections of this paper.
DISCUSSION
We examined the dynamic associations between self-esteem, pleasure, sadness, and social experiences in one model. These types of associations have rarely been investigated at the level at which such processes typically occur, i.e. within individuals within days (Fisher et al., 2018).
Our study yielded three key findings: (a) concurrently, self-esteem and pleasure were associated with social experiences, both in terms of time spent talking / time spent alone (which relates to interaction quantity, Dataset 1) and in terms of the desire to be alone during social interaction (Dataset 2), whereas sadness only had a meaningful association with the desire to be alone, (b) within-person temporal associations were mostly negligible in size, and (c), in both datasets, the concurrent association between self-esteem and pleasure was stronger than the concurrent association between self-esteem and sadness. These findings are now discussed in more detail.
Previous research has shown that social experiences are associated with self-esteem (Marshall et al., 2014; Reitz et al., 2016) and depressive symptoms (Hames et al., 2013; Rubin et al., 2011), and possibly play an important role in the mechanisms that underlie the associations between
5 self-esteem and depressive symptoms. In Dataset 1 we investigated the role of interaction
quantity, via time spent talking and time alone. When participants reported lower self-esteem than their average level, they also reported to have spent more time alone, and having talked less. We did not differentiate between the characteristics of the company which may influence these effects. For example, Denissen et al. (2008) found small associations between self-esteem and the amount of time interacting with close family members, and remote friends, but not with a partner, other family members or close friends.
In our model, pleasure was concurrently associated with talking more and spending less time alone, but sadness lacked meaningful associations with both social interaction quantity variables. Perhaps, in non-clinical samples, positive affect propels a desire to engage in social activities rather than that negative affect drives avoidance (Watson, 2000). Avoidance tendencies may only surface in clinical samples, i.e. at higher symptom levels of depression.
In Dataset 2 we investigated whether self-esteem, sadness, and pleasure were associated with the appraisal of being in social company. When participants were in company and reported more sadness or less pleasure, they also reported a stronger desire to be alone. This finding is in line with the EMA study of Brown et al. (2011), who also found that depressive symptoms were associated with a desire to be alone, and aligns with previous research linking depressive symptoms to the motivation to withdraw from social interactions (Allen & Badcock, 2003).
When participants were in company and reported lower self-esteem than their average level, they also reported a stronger desire to be alone. The concurrent association between self-esteem and the desire to be alone mirror earlier findings showing that low self-esteem is associated with social avoidance, possibly as an attempt to avoid further harm to self-esteem or due to the experience of social problems (Heimpel et al., 2006; Masselink, Van Roekel, et al., 2018). The findings contradict sociometer theory’s premise that low self-esteem motivates behavior to restore self-esteem, although this may also vary as a function of the time frame. Arguably, it is sensible to retreat temporarily when there are indications of lowered relational value (i.e., lower self-esteem), to evaluate the situation from a safe distance, before one instigates behavior aimed at restoring the relationship and thereby self-esteem (Horwitz & Wakefield, 2007). The actions taken need to be successful in restoring self-esteem, and even if successful, this process takes time. Given that all temporal associations from and to the desire to be alone were smaller than the SESOI, this might take more time than our six-hour measurement interval.
Our post-hoc results, showing stronger concurrent within-person association between self-esteem and pleasure than between self-esteem and sadness, are informative about the association between self-esteem and depression. The results suggest that this association is mostly driven by the association between self-esteem and anhedonia, rather than the other core symptom of MDD, depressed mood. In the first dataset, the difference in strength of associations may have been an artifact of within-person variance differences between the sadness and pleasure measures, but the effect replicated in the second dataset without apparent differences
in variances. Future studies may try to replicate the stronger associations between self-esteem and pleasure than between self-esteem and sadness, and examine these associations across various time scales. Nezlek and Plesko (2003), for example, reported the opposite pattern using two measures per week during 10 weeks, namely, self-esteem was more strongly related to negative affect than to positive affect.
We used a SESOI at a standardized effect size (beta) of 0.10 to indicate which effect is meaningful and deserves interpretation. With sufficient statistical power –due to many observations per individual and many individuals– most effects will become significant, which does not equal relevance. The SESOI may be arbitrary and context dependent, but can guide the interpretation of our work, and was chosen to be equivalent to the commonly used threshold for a small correlation coefficient (Cohen, 1992). Standardized effect sizes are not equal to correlation coefficients when multiple predictors are used, and interpreting them as correlation coefficients overestimates the explained variance by each predictor, depending on the degree of multicollinearity between the predictors (Dudgeon, 2016). A standardized effect size of 0.10 corresponds to about 1% explained variance, which, in this study, we considered to be the bare minimum to be relevant.
The application of the SESOI rendered most concurrent within-person associations between self-esteem, pleasure, sadness and social experiences meaningful, and discarded most lagged associations. Theoretically, small within-person effects may accumulate to more meaningful within-person effects over time, and reinforce each other (Caspi, Roberts, & Shiner, 2005;
Teasdale, 1988). However, the size of the autoregressive effects in our studies (range 0.07-0.19) indicate that our participants tended to quickly return to their mean scores, which does not support long-lasting reinforcing processes at play. This suggests that, over the six-hour interval studied, most of the temporal effects were negligible. An exception was the temporal effect of pleasure on self-esteem as found in Dataset 2. This suggests that a lack of pleasure may increase negative self-referential thinking (Teasdale, 1988). The lack of temporal associations between self-esteem and the social variables contradict both sociometer theory and theories predicting that self-esteem is associated with social avoidance behavior (Baumeister et al., 1989; Heimpel et al., 2006).
However, it is too early to dismiss the hypotheses that self-esteem, pleasure, sadness and social experiences influence each other over time, because it is possible that the hypothesized temporal within-person effects (Table 1) occur across different time intervals. Future studies can explore the possibility that dynamic interactions between self-esteem, pleasure, sadness, and social experiences operate at different time scales than the one we studied. It is possible that the dynamic interactions between self-esteem, pleasure, sadness and social experiences occur on a different short time interval, such as over days (but see Denissen, et al., 2008 for null findings from day-to-day), but it is also possible that the dynamic interactions mostly occur on the trait level (cf. Fleeson & Jayawickreme, 2015). These options can be studied using longitudinal studies
5 with multiple short EMA measurement bursts (Sliwinski, 2008) and continuous time modelling
which can deal with associations that operate at different time-intervals within one model (van Montfort, Oud, & Voelkle, 2018). In addition, self-reinforcing vicious cycles between lower self- esteem, decreased social interaction quantity and appraisal, and depressive symptoms may only occur after a certain severity threshold is exceeded (a “tipping point”; e.g. Kunnen, De Ruiter, Jeronimus, & Van der Gaag, 2019; van de Leemput et al., 2014), or in the presence of specific cognitive vulnerabilities (Teasdale, 1988).
Strengths and limitations
Strong points of our study include our pre-registered models and power calculations based on expected effect sizes, using a SESOI, the use of multivariate DSEM models in which unequal time-intervals between measurement moments was taken into account and replication of some findings across studies. Several limitations need to be mentioned as well. First, our results pertain to samples of young Dutch adults (Dataset 1) who experienced above-average pleasure and education and Dutch adults (Dataset 2) including a majority of higher-educated women, which may impede direct generalization to the general population. Second, we used a mix of retrospective and in-the-moment formulated items (see Table 2) which hampers the interpretation of concurrent (not truly in the moment) and temporal associations (in reality somewhere between 1 and 2 lags). Third, in Dataset 2 we excluded a negatively worded item from the self-esteem measure to increase reliability, leaving only two positively worded items which may limit construct coverage. Although items should not be removed just to increase the reliability index (Hoekstra, Vugteveen, Warrens, & Kruyen, 2018), mixing negatively and positively worded items can result in validity issues (Barnette, 2000; DiStefano & Motl, 2006). Fourth, we averaged within-person effects rather than fitting individual models and random effects indicated individual variation in the strength of effects (see Fisher et al., 2018). Fifth, in Dataset 2 we were not able to carry out our analyses as planned due to too many missing values on the variable desire to be in company, which had to be dropped, and the final model did not reach our predetermined convergence criterion of a PSR < 1.02 (although it was below the commonly applied criterion of < 1.10). Sixth, there is possible ambiguity in the interpretation of the desire to be alone measure. A desire to be alone may reflect both a social avoidance motivation or an indication of the quality of passed interactions. Finally, we measured pleasure in Dataset 1 and cheerfulness in Dataset 2. This may on the one hand constrain the comparability of results across the studies: pleasure may be more related to positive experiences coming from activities, whereas cheerfulness may relate more to an overall positive affective state. On the other hand, the similarity of the findings between measures suggest that the results were not an artifact of the used measures and thus speak for the robustness of the findings.
CONCLUSION
Using the data of two EMA studies in Dutch adults, we showed that low self-esteem co-occurs with sadness and lack of pleasure during daily life. Self-esteem, pleasure, and sadness were all related to social experiences such as spending time talking or spending time alone. Most temporal associations fell below our smallest effect size of interest, and exceptions did not replicate across the datasets, indicating that self-esteem, pleasure, sadness and social experiences do not affect each other much across six hours. Future studies may determine whether there are more relevant time intervals to study the dynamic associations between self-esteem, depression, and social experiences. Post-hoc results indicate that one of the core symptoms of depression, lack of pleasure, was more strongly connected to self-esteem than sadness, and may be the primary drive behind the link between self-esteem and depression.
5
SUPPLEMENTAL MATERIAL TABLE S1. Concurrent Standardized Associations Between Variables in Dataset 2 of Participants Who Filled in >75% of measures Self-esteem β [95% CI]Sadness β [95% CI]Pleasure β [95% CI] Sadness-0.46 [-0.47, -0.45] Pleasure0.60 [0.59, 0.60]-0.51 [-0.52, -0.50] Desire to be alone-0.29 [-0.30, -0.28]0.23 [0.22, 0.24]-0.29 [-0.30, -0.27] Note. Standard deviation of the posterior was 0.004-0.007. All values significant at p< .001 TABLE S2. Temporal Standardized Associations Between Variables in Dataset 2 of Participants Who Filled in >75% of Measures Self-esteem β [95% CI]p-valueSadness β [95% CI]p-valuePleasure β [95% CI]p-valueDesire to be alone β [95% CI]p-value Self-esteem t-10.17 [0.15, 0.19]<.001-0.05 [-0.07, -0.04]<.0010.07 [0.05, 0.08]<.001-0.04 [-0.06, -0.02]<.001 Sad t-1-0.06 [-0.07, -0.04]<.0010.17 [0.15, 0.18]<.001-0.08 [-0.10, -0.06]<.0010.04 [0.02, 0.06]<.001 Pleasure t-10.10 [0.08, 0.12]<.001-0.10 [-0.12, -0.09]<.0010.17 [0.15, 0.19]<.001-0.05 [-0.07, -0.03]<.001 Desire to be alone t-1-0.001 [-0.02, 0.02].9120.01 [-0.01, 0.03].188-0.002 [-0.02, 0.02].8420.10 [0.08, 0.12]<.001 Note. For all associations the standard deviation of the posterior was 0.01 a The false discovery rate derived significance threshold was .041
S3. Prior Analyses Dataset 1
Analyses during a summer school on network analysis in 2016 were conducted on the combined anhedonic and control group (the current paper only included the control group) of the No Fun No Glory Study (for details see van Roekel et al., 2016). The analyses were done univariate using the mlVAR package of R instead of multivariate with DSEM in Mplus. The total sample was divided into high and low trait self-esteem based on a proxy measure of trait self-esteem to investigate differences in associations between high/low trait self-esteem individuals. Overlapping variables included self-esteem, sadness and time spend talking. Variables included in the earlier analyses but not in the current are time spend with friends and joyfulness. The design of the study was changed because we perceived the design in which an anhedonic and control group would be combined and subsequently divided using a proxy variable of trait self-esteem to come along with too many limitations hampering interpretation of the findings.
