University of Groningen A Dynamical Approach to Psychological Resilience Hill, Yannick

(1)

A Dynamical Approach to Psychological Resilience

Hill, Yannick

DOI:

10.33612/diss.144252644

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: 2020

Link to publication in University of Groningen/UMCG research database

Citation for published version (APA):

Hill, Y. (2020). A Dynamical Approach to Psychological Resilience. University of Groningen. https://doi.org/10.33612/diss.144252644

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.

(2)

548915-L-bw-Hill 548915-L-bw-Hill 548915-L-bw-Hill 548915-L-bw-Hill Processed on: 30-9-2020 Processed on: 30-9-2020 Processed on: 30-9-2020

Processed on: 30-9-2020 PDF page: 45PDF page: 45PDF page: 45PDF page: 45

Variations in Self-reported

Protective Factors of Resilience

in Athletes

This chapter is based on:

Hill, Y., Meijer, R. R., Van Yperen, N. W., Michelakis, G., Barisch, S., & Den Hartigh R. J. R. (in press). Non-ergodicity in protective factors of resilience in athletes. Sport, Exercise and Performance Psychology.

(3)

Abstract

A large body of literature on resilience in sports has focused on identifying different protective factors, which help an individual to adapt to stressors. However, the link between protective factors and athletes’ resilience may vary across different measurement moments. In addition, we know little about the extent to which findings at a group-level are generalizable to individuals. This study aims to test whether 1) the relationships between protective factors and resilience vary over time and 2) the statistics of the repeated assessments at the group-level generalize to the individuals. Therefore, we conducted a diary study in which we asked athletes to report on a selection of different protective factors (i.e., perceived social support, perfectionism, confidence, and motivation) and on resilience for 21 days. Our analyses indicate that the relationships between protective factors and resilience vary across the 21 days. Furthermore, we found a lack of generalizability of group-level to individual-level statistics in protective factors and resilience. Given the observed variability in the relationship between protective factors and resilience as well as the lack of group-to-individual generalizability, we suggest to place the individual at the level of analyses of resilience. Future research should collect time-series data to capture temporal patterns that can indicate changes in resilience over time in individual athletes.

(4)

3

1 | Introduction

When doing sports at a competitive level, setbacks such as injuries or losing matches occur frequently for athletes. In order to continue the engagement in sports and reach high levels of performance, it is essential that athletes have the ability to bounce back or recover from these stressful events (Smith et al., 2008). Returning to the previous level of performance following an adverse event is called resilience (Fletcher & Sarkar, 2013; Galli & Gonzalez, 2015; Hill et al., 2018a). For decades, scholars have been interested in the underlying determinants of resilient responses in athletes (e.g., Fletcher & Sarkar, 2012). Researchers have searched for protective factors, such as trait-like characteristics and mental processes, which can explain how resilient athletes are (for reviews, see Fletcher & Sarkar, 2012; Galli & Gonzalez, 2015; Sarkar & Fletcher, 2014). For instance, in a qualitative study with current and former high-level athletes, Galli and Vealey (2008) found that resilience in athletes was associated with various factors, such as achievement motivation, social support, confidence, and coping.

To date, researchers have not yet discovered specific (sets of) protective factors that can consistently and reliably predict resilience as an individual difference variable across time and individuals (Hill et al., 2018a). This may be explained by the idea that resilience is a complex process, which emerges from (ongoing) dynamic interactions between the person and the environment rather than isolated factors or simple interactions between them (e.g., Egeland et al., 1993; Fletcher, 2018; Fletcher & Sarkar, 2012; Galli & Vealey, 2008; Hill et al., 2018a, 2018b). Accordingly, different adverse events may require different adaptive processes in order to return to the previous level of functioning. As a simple example, in response to a lost match, confidence in one’s abilities may help to return to the previous level of functioning for the next match. However, when encountering some physical discomfort, limiting physical activities on the short-term, psychological coping strategies, such as seeking social support, may be more appropriate. Thus, due to the constant interaction with the environment, athletes constantly adapt, which, in turn likely causes variations within the athletes psychological or physiological states and the environment. Thereby, a circular process of adaptation may emerge (Balagué et al., 2017), which implies that a psychological characteristic may vary across measurement moments within a certain range inherent to the characteristic of interest. Due to variations in the underlying factors of resilience, the ability to bounce back or recover from these stressful events may vary over time and across contexts accordingly (Masten & Abradovic, 2006). Furthermore, because these variations result from specific person-environment interactions, findings that are based on group-level data may not capture the variations that the individuals within the sample actually display (Liu et al., 2006).

(5)

So far, there is no empirical study that has systematically assessed variations of rela-tionships between protective factors and resilience. To fill this void, we designed a study in which we tracked different protective factors in athletes over a time span of three weeks. Thereby, we shifted our perspective from establishing (static) relations between protective factors and resilience to detecting whether variations between protective factors and resilience occur. The specific aims of the current study are to test a) Do relationships between a selection of protective factors and resilience vary across measurement days? And b) Do the statistics of the repeated assessments of the protective factors at a group-level “generalize” to the individual processes?

1.1 | Varying Interactions between Protective Factors

Although the notion of possible variations in relationships between protective factors and resilience is well accepted among scholars, Hill et al. (2018b) concluded that many studies in this domain are still based on measurements at a single time-point. For example, when studying resilience, a typical qualitative or experimental design would assess protective factors at a single point in time before a stressor occurs. This single time-point is (implicitly) assumed to be representative for the link between the protective factor and resilience across time, both within and across individuals (Hill et al., 2018b). However, if protective factors vary across measurement moments, a single time-point may not be representative for the discovered relationships between protective factors and an individual’s ability to bounce back from the stressful event. Accordingly, the interpretation of the relationship between a given protective factor and resilience may be different if assessed at a different point in time. For example, at a given time-point, a researcher may find a strong relationship between perceived social support and resilience. The researcher may thus conclude that this characteristic serves as a protective factor. Yet, this relationship may be weak or even reversed at another point in time (Hamaker, 2012; Simpson, 1951). As an illustration, although perceived social support has been found to predict resilience in athletes in some studies (e.g., Galli & Vealey, 2008), Mummery, Schofeld, and Perry (2004) found that in competitive swimmers, independence of social support was associated with higher levels of resilience.

The notion of variations over time is not restricted to the relationship of protective factors and resilience; the relationships between different protective factors may also be subject to variations over time. Imagine that for a given match, an athlete may feel very confident, which also increases the athlete’s intrinsic motivation (i.e., playing the sport for fun). Thus, confidence and (intrinsic) motivation may fuel each other to help the athlete to be resilient. However, when anticipating a match against a far superior opponent, an athlete may not feel very confident before the match, yet still highly intrinsically motivated to engage in the challenging event and less afraid to make

(6)

3

a mistake during the match (i.e., adaptive perfectionism; Stoeber et al., 2006). In this

case, (intrinsic) motivation and confidence would not interact to facilitate the athlete’s resilience, but instead (intrinsic) motivation and adaptive perfectionism. Hence, when determining snap-shot relationships between protective factors and resilience, we need to take statistical uncertainty into account. From this point of view, one may argue that a true score representing the actual relationship between factors may be approached by averaging increasing numbers of participants and measurements across time.

However, there are also theoretical and practical reasons to take variations across time and individuals more seriously. Recent research has shown that changes in fluctuations can signal losses of resilience in various complex biological systems, rather than being the source of undesirable noise (Hill et al., in press; Kelso, 2010; Scheffer et al., 2009, 2012; Van de Leemput et al., 2014). Thus, tracking an athlete’s perceived resilience and different protective factors over time may yield insight into these functional patterns. In addition, individual scores per time-point are averaged across participants in order to yield a group-level summary of the data (Liu et al., 2006). In the following step, this averaged trajectory would be analyzed to identify distinct patterns, such as stationarity or variability of the process. Interestingly, a recent empirical study found that repeated measures of different psychological variables cannot be generalized from groups to individuals (Fisher et al., 2018). Furthermore, informative patterns of the dynamics of the process found on the individual-level may be masked and eliminated when averaging across several individuals (Liu et al., 2006).

Indeed, research has shown that group-level processes can often not be generalized to processes on the individual level (i.e., which has been referred to as the ergodicity problem; Fisher et al., 2018; Molenaar & Campbell, 2009). This effect has been found, among others, in the domain of (motor) development (Liu et al., 2006; Molenaar, 2004; Van Geert, 2014), the Big Five Model of Personality (Hamaker et al., 2005), and neural networks (Medaglia et al., 2011). More specifically, models based on group-level trajectories can only be extended to individuals under specific conditions. These conditions are homogeneity and stationarity of the process, which are rarely given in social sciences (Hamaker et al., 2005; Molenaar, 2004). An illustrative example has been provided by Hamaker (2012). That is, on the group level, one may expect to find a negative relationship between typing speed and number of typos, because on average, experts usually type faster and more accurately than novices. However, the relationship is reversed on the individual level. If individuals type more quickly, they are more likely to increase the number of typos they make. Therefore, the statistics on the group level indicating a negative relationship between speed and typos is reversed and thus differs on the individual level. Similarly, in terms of resilience, on the group-level, researchers may find that individuals with higher levels of confidence show higher levels of resilience.

(7)

However, on the individual level, increasing levels of confidence may have no effect at all on their (perceived) resilience or may even lead to complacency due to overconfidence and consequently undermine resilience. This common lack of projectability of group-level findings to the individual group-level is called non-ergodicity (Fisher et al., 2018).

To date, the assumption of generalizability of group findings to the individuals has been largely ignored, in particular in resilience research. If findings from a group cannot be extended to the individuals, applying models that focus on patterns within individuals rather than the averaged group, may be more appropriate. Therefore, testing for the similarity of group-level and individual-level statistics in time-serial data is an important step, because it may contradict various assumptions of statistical procedures and change the interpretation of group-level results.

1.2 | The Current Study

In this study we aimed to provide first empirical insights into the variations in the relationships between protective factors and resilience, and of the degree to which group-level time-series of protective factors and resilience can be “generalized” to the individual-level (cf. Fisher et al., 2018). For this aim, we designed a diary study (e.g., Blaauw et al., 2016; Van der Krieke et al., 2016), during which participants were asked to fill out questionnaires assessing resilience and a variety of protective factors, assumed to be relevant and reliable predictors of resilience across individuals and time. Note that in order to be able to measure resilience daily (even in the absence of a stressor), we used a questionnaire that focuses on an individual’s perceived ability to “bounce back” from a stressor (Smith et al., 2008). For the protective factors, we focused on confidence, motivation, perceived social support, and adaptive perfectionism (Sarkar & Fletcher, 2014). Confidence as a protective factor describes the extent to which an athlete possesses necessary capacities to excel in their sport (Beattie et al., 2011). Motivation means to be moved to act (1) because individuals find the task interesting and enjoyable in its own right (intrinsic motivation), or (2) because the activity is subjectively associated with external rewards (extrinsic motivation; e.g., Ryan & Moller, 2017). Perceived social support refers to the degree to which an individual believes that others in their social environment would provide assistance if requested (Freeman et al., 2011). Perfectionism denotes individuals’ tendency to set high standards for themselves. The two main dimensions being personal standards (i.e., striving towards high personal performance expectations), and discrepancy (i.e., the perceived gap between one’s current and desired performance, Rice et al., 2014).

Following the idea that protective factors change across measurement moments, we anticipated that correlations between protective factors and resilience vary (Hypothesis

(8)

3

1). This means that relationships between the same variables may be interpreted as low,

medium, or high, depending on the time of measurement. Furthermore, we expected that the group-level trajectories of protective factors may not be generalizable to individuals (i.e., non-ergodicity, Fisher et al., 2018). Specifically, Hypothesis 2 states that statistics such as medians, standard deviations, fluctuation rates, and correlations, computed at the group-level differ from (i.e., are not identical to) the same statistics computed on the individual-level. Finding considerable variations between protective factors and resilience and a lack of generalizability from group-level statistics to individual-level statistics, reinforces the need for future studies which focus on individual patterns of resilience over time.

2 | Method

2.1 | Participants

A total of 111 university students, who engage in competitive sports on a regular basis (i.e., at least once per week) signed up to participate in the current study in exchange for course credits. Forty-nine (44.14%) participants did not adhere to the study protocol either by not completing the full 21 days or by not filling in the daily questionnaire regularly (e.g., several questionnaires in one day or intervals larger than 2 days between assessments) and were removed from the sample. The final sample consisted of 62 participants (21 males, 41 females), with 91.94% being 22 years of age or younger. On average, the participants practiced their specific sport 2.81 (SD = 1.94) times per week.

2.2 | Materials

For the current study, we used an online platform, where participants were able to complete the daily assessments of the different questionnaires at any time of the day using either a computer or their smartphones. Each daily assessment contained the exact same items for the different protective factors: confidence, motivation, perceived social support, and perfectionism, as well as for resilience. Each item of the protective factors’ assessment was indicated with a visual analogue scale (Reips & Funke, 2008) of 100 points ranging from 0 (strong disagreement) to 100 (strong agreement).

2.2.1 | Confidence

Confidence was measured by the Trait Robustness of Sports-Confidence Inventory-8 developed and validated by Beattie and colleagues (2011). It consists of eight items, which ask how strong a person is agreeing to different statements (e.g., If I make a mistake it

(9)

has quite a large detrimental effect on my self-confidence). In our study, there was a high group-based correlation between scale scores in subsequent measurement points (day 1 compared to day 2, day 2 compared to day 3, and so forth) with the lowest correlation over the 20 comparisons being equal to .88 (95% CI [.81, .93]).

2.2.2 | Motivation

Motivation was measured by the two 4-item subscales from the Sport Motivation Scale-6 (SMS-6, Mallett et al., 2007) that provide the largest distance on the intrinsic-extrinsic motivation continuum, that is, external regulation and intrinsic motivation (e.g., Vansteenkiste et al., 2010). External regulation was assessed by items, such as ‘I play sports for the prestige of being an athlete’. Sample items for intrinsic motivation included ‘I play sports for the excitement I feel when I am really involved in the activity’. The lowest group-based correlation between the scale scores in subsequent measurement points was .76 (95% CI [.63, .85]) for external regulation and .71 (95% CI [.56, .82]) for intrinsic motivation.

2.2.3 | Perceived social support

Perceived social support was measured with the 16-item Perceived Available Support in Sports Questionnaire (PASS-Q, Freeman et al., 2011). The four 4-item subscales of the PASS-Q represent emotional support (e.g., If needed, to what extent would someone show concern for you?), esteem support (e.g., If needed, to what extent would someone enhance your self-esteem?), informational support (e.g., If needed, to what extent would someone give you tactical advice?), and tangible support (e.g., If needed, to what extent would someone help with travel to training and matches?). In the current study, the smallest group-based correlation between scale scores in subsequent measurements for all subscales exceeded .76 (95% CI [.63, .85]) over the 20 comparisons except for the first two observations of informational support (r = .69, 95% CI [.53, .80]).

2.2.4 | Perfectionism

Perfectionism was assessed with the Short Almost Perfect Scale (SAPS, Rice, Richardson, & Tueller, 2014). This scale captures two major dimensions of perfectionism (e.g., Van Yperen & Hagedoorn, 2008). Personal Standards (4 items) reflect the essential and positive aspects of perfectionism (Slaney et al., 2002). A sample item is, ‘I have high expectations of myself.’ Perceived Discrepancy between standards and criteria of success in meeting those standards was measured using the Discrepancy subscale (4 items). A sample item is, ‘I hardly ever feel that what I’ve done is good enough.’ The subscale for Personal Standards showed a group-based correlation of at least .86 (95% CI [.78, .91])

(10)

3

between the scale scores in subsequent measurements, while the Discrepancy subscale

showed a correlation of at least .74 (95% CI [.60, .84]) between subsequent observations.

2.2.5 | Resilience

Resilience was assessed using the 6-item Brief Resilience Scale (Smith et al., 2008), which strongly focuses on the process of returning to the previous level of functioning. A sample item includes ‘I tend to bounce back quickly after hard times’. The scale showed a group-based correlation between scale scores in subsequent measurement points of at least .84 (95% CI [.75, .90]) over the 20 comparisons.

2.3 | Procedure

After receiving the approval of the local ethical committee of the University’s Psychology department, the study was activated in the University’s online research platform. Before the participants were able to start the study, they were informed that they would be asked to fill out questionnaires on a daily basis assessing confidence, motivation, perceived social support, perfectionism, and resilience for a period of 21 days yielding 21 measurement points for each construct. Each assessment was estimated to take about five to seven minutes. In order to proceed with the first questionnaire, the participants had to indicate their informed consent, demographics, and their email address in order to automatically receive the URL for the following questionnaire. The first and all subsequent questionnaires assessed the protective factors and resilience. After completing the daily survey, the participants received the link to the survey of the upcoming day. Following the final questionnaire (i.e., day 21), participants were thanked for their participation, debriefed on the purpose of the study, and granted their research credits.

2.4 | Data Analysis

To test for variations in correlations between the protective factors and resilience across measurement moments (Hypothesis 1), we computed the group-level correlations between all subscale scores of the protective factors and resilience scores for each of the 21 measurement points. Thereby, we created a time-series of 21 correlation coefficients between resilience and each subscale of the protective factors. For example, the correlation of the 62 participants between resilience score and confidence score was computed based on the participants’ responses on day 1, then the correlation between the two variables was computed based on the participants’ responses on day 2, and so forth. Then, the smallest and largest correlations in these time-series were identified to determine the absolute distance between them. This distance reflects the range of different correlations between the given constructs over the assessment period. In social

(11)

sciences a difference of .2 in the correlation is considered large as it distinguishes the interpretation of the correlation strength from small (.1) to medium (.3) and to strong (.5, Cohen, 1988). In addition, we computed the 95% confidence intervals (CIs) using Fisher z-transformation.

In order to test for the non-ergodicity in the protective factors and resilience trajectories (Hypothesis 2), we first computed an average trajectory for each protective factor and resilience by computing the daily group mean to derive the group-level statistics. Specifically, we computed the mean, median, standard deviation, and stability (i.e., mean square of successive differences, MSSD, Von Neumann et al., 1941) for each averaged trajectory.

Furthermore, we computed the mean group-level correlations between the protective factors and resilience based on the group-level correlation time-series (see Hypothesis 1). Next, we computed the individual-level statistics by computing the mean, median, standard deviation, and MSSD for each individual trajectory based on the 21 measurement points for each individual before averaging these statistics across the sample. In order to compare the correlations of the individual-level to the group-level, we calculated the individual-level correlations between the protective factors and resilience based on the 21 measurement points of each individual. The resulting individual-level correlations were then averaged across the sample.

Note that group-level statistics were calculated by first collapsing individual scores into a group-level trajectory, before the analysis, while individual-level statistics are first analyzed for each individual before the results are collapsed into group scores. Mathematically, this results in the same mean values, but may yield different values for other descriptive statistics, such as medians or standard deviations. Because ergodic processes are marked by homogeneity of group-level and individual-level statistics, the hypothesis of non-ergodicity is supported when the group- and individual-level statistics differ from each other. That is to say that the statistics for the group means are not equal to the mean statistics for the individuals in the group. However, because it may be argued that just because the values are not exactly equal to one another, they may not differ in statistical terms. Therefore, we also conducted t-tests to assess whether the distribution of the individual scores comes from the group-level mean.

3 | Results

Hypothesis 1 was that correlations between protective factors and resilience show varia-tions over the assessed period of 21 days. We found that 4 out of 9 (44.44%) subscales

(12)

3

show variations in correlation above .2, a range of which is typically interpreted as different

in strength (Cohen, 1988). For example, assessing the correlation between resilience and esteem support on day 20 would yield a moderately strong, positive relationship between these factors (r = .33, 95% CI [.09, .54]) while the variables would be considered weakly related on day 4 (r = .11, 95% CI [-.14, .35]). Note, however, that large confidence intervals also indicate that considerable variations can be expected in the relationship between resilience and esteem support following a single measurement point based on our sample size (see Figure 4). Furthermore, in five subscales, the interpretation of the correlation strength changed depending on the day of the assessment (for patterns of variations across measurement moments, see Figure 4). These findings suggest that the interpretation of the relationship between a given protective factor and resilience measured at a single point in time may interpreted differently when measured at a different point in time.

Figure 4 | Time-series of correlations and their confidence intervals for each protective

factor subscale (A confidence, B external regulation, C intrinsic motivation, D emotional support, E esteem support, F informational support, G tangible support, H standards, and I discrepancy) and resilience. The dotted lines display the interpretational threshold for small, medium, and large correlations (Cohen, 1988). In 5 time-series (B, E, F, G, and I) fewer than 90% of the observed correlations fall within the same interpretation thresholds.

Hypothesis 2 stated that the statistics computed for averaged group-level trajectories are not identical to the statistics computed for individual-level trajectories. As shown

0 0,1 0,2 0,3 0,4 0,5 0,6 0,7 0,8 0,91 1 3 5 7 9 11 13 15 17 19 21 Corr elation Days -0,2 -0,1 0 0,1 0,2 0,3 0,4 0,5 0,6 1 3 5 7 9 11 13 15 17 19 21 Corr elation Days -0,2 -0,10 0,1 0,2 0,3 0,4 0,5 0,6 0,7 1 3 5 7 9 11 13 15 17 19 21 Corr elation Days -0,2 -0,10 0,1 0,2 0,3 0,4 0,5 0,6 0,7 1 3 5 7 9 11 13 15 17 19 21 Corr elation Days A D G B C E H F I -0,3 -0,2 -0,10 0,1 0,2 0,3 0,4 0,5 0,6 1 3 5 7 9 11 13 15 17 19 21 Corr elation Days -0,4 -0,3 -0,2 -0,10 0,1 0,2 0,3 0,4 0,5 1 3 5 7 9 11 13 15 17 19 21 Corr elation Days -0,2 -0,1 0 0,1 0,2 0,3 0,4 0,5 0,6 1 3 5 7 9 11 13 15 17 19 21 Corr elation Days -0,5 -0,4 -0,3 -0,2 -0,1 0 0,1 0,2 0,3 1 3 5 7 9 11 13 15 17 19 21 Corr elation Days 0 0,1 0,2 0,3 0,4 0,5 0,6 0,7 1 3 5 7 9 11 13 15 17 19 21 Corr elation Days

(13)

in Table 1, we observed divergent patterns of group-level and individual level medians, standard deviations, and MSSDs. For example, the fluctuation rate (i.e., MSSD) of the daily mean scores (i.e., the fluctuation rate of the means) is not equal to the mean fluctuation rate for the individual trajectories. Similarly, we found different outcomes for correlations between protective factors and resilience when being computed at a level and individual-level (Table 1). For example, we observed a medium strong group-level association between resilience and informational support (r = .32), which indicates that individuals who display a higher level of informational support, have higher levels of resilience. However, the individual-level association between resilience and informational support might be interpreted as non-existent (r = .04), which means that within individuals the two variables are unrelated at a given moment in time. Moreover, the t-tests revealed that only for the centrality measures (i.e., mean and median) as well as 22.22% (2/9) of the correlations the averaged statistic may represent the mean value of the individual-level distributions (ps < .01). This supports the notion that except for the means and medians, the statistics obtained on the group-level differ from the ones obtained on an individual level.

4 | Discussion

Previous research on resilience has put a major focus on identifying different psychological characteristics and processes (i.e., protective factors) that protect individuals from the negative effects of stressors (Fletcher & Sarkar, 2012, 2013). The first aim of the current study was to test whether the relationships between the protective factors and resilience vary across measurement moments. In line with the notion that single time-point assessments may yield unreliable predictions of resilience (Hill et al., 2018b), our results suggest that correlations between different protective factors and between protective factors and resilience show considerable variation across measurement moments. This means a correlation may, on average, be considered as high at a given point in time and low-to-medium or even absent at a different point in time.

Statistically, variations in correlations can be expected due to uncertainty of the measurements and the sample of a given study. However, these variations may also be due to natural fluctuations within athletes’ psychological states (Den Hartigh, Van Geert et al., 2016; Gernigon et al., 2015; Nowak & Vallacher, 1998; Vallacher et al., 2015; Van Geert, 1991, 1994, 2009) and can lead to different interpretations of relationships between different variables at different points in time. Interestingly, the variations we found occurred despite using scales based on rather trait-like conceptualizations of the different constructs. These variations may become even more profound when assessing state-like variables such as emotions and cognitions. Hence, for future research, the

(14)

3

Table 1

| S

tatistics for gr

oup

-level and individual-level analyses for pr

otective factors and r

esilience. Mean Median SD MSSD Corr elation Ind./ Gr oup Ind. Gr oup Ind. Gr oup Ind. Gr oup Ind. Gr oup Conf idence 50 .33 50 .7 4 50 .93 7.43** 1.81 64 .27** .66 .33** .70 Motivation Extrinsic Motivation 35 .48 34 .95 35 .53 8.55** .96 106 .37** 1.63 .05 .11 Intrinsic Motivation 75 .36 76 .21 76 .09 7.59** 2.73 82 .25** 1.35 .07 .04 Per

ceived Social Support

Emotional Support 81.24 81.99 81.66 6.29** 1.57 62 .8** .76 .05* .20 Esteem Support 72 .6 73.27 73.33 7.75** 2.29 84 .89** 1.81 .06** .24 Informational Support 62 .4 62 .99 62 .91 8.27** 2.31 95 .04** 2.57 .04** .32 Tangible Support 55 .55 55 .85 56 .73 8.59** 2.42 96 .62** 1.42 .10** .28 Per fectionism Personal S tandar ds 68.92 69.25 69.14 6.93** .97 77 .76** 2.02 .00* -.13 Discr epancy 51.87 52 .05 51.85 8.55** .89 112 .72** 2.24 .11** .42 R esilience 54 .76 55 .17 54 .58 7.04** 1.18 70 .51** 1.78 - -* p < .01 ** p < .001

(15)

variability of correlations between protective factors and resilience across time, but also at a given measurement moment within a sample, should be addressed explicitly. The second aim of the study was to test whether statistics calculated for group-level data extend to individual-level analyses. Our findings are in line with the non-ergodicity of dynamic processes in psychology (Fisher et al. 2018). This means that collapsing individual time-series into a single, group-level trajectory masks properties of the individual-level dynamics, such as fluctuations (Liu et al., 2006). For example, a medium correlation between a protective factor and resilience based on group-level data may indicate that if the protective factor becomes stronger, individuals have higher levels of resilience. This implies that an intervention may be aimed at improving the protective factor to increase resilience of individuals within the group. However, if the relationship between the protective factor and resilience is weak or absent at an individual-level, the intervention would prove to be ineffective. That is, increases in the protective factor are unrelated to changes in resilience over time. This means that interpretations and models based on group-level findings need to be interpreted with caution and accurately labeled group-level effects.

Combined with the findings that correlations between different protective factors and resilience may vary across measurement moments, the lack of group to individual generalizability of protective factors suggests that fluctuations need to be taken seriously, rather than treated as undesirable noise (Kelso, 2010). Indeed, research has shown that time series of psychological and behavioral states in many cases fluctuate according to a pattern that cannot be characterized as random error (e.g., De Ruiter et al., 2015; Delignières et al., 2004; Harrison & Stergiou, 2015; Pincus, 2014; Pincus et al., 2019). Similarly, different studies have shown that changes in fluctuation patterns of various biological systems can indicate changes in their resilience (Hill et all., in press; Scheffer et al., 2009, 2012; Van de Leemput et al., 2014). These studies apply a dynamical systems approach, which may be a fruitful avenue for resilience research.

4.1 | Theoretical Implications

A dynamical system consists of several components, which constantly interact with each other and the environment (e.g., Nowak & Vallacher, 1998; Vallacher & Nowak, 1997). The current state of a system emerges from the ongoing interactions and cannot be reduced to individual components (i.e., interaction-dominance, Den Hartigh et al., 2017; Van Orden et al., 2003). Due to the circular interaction between the system and the environment over time, the current state of a system has developed in an iterative process from the previous states of the system and simultaneously serves as input for every future state (e.g., Den Hartigh, Van Dijk et al., 2016; Gernigon et al., 2015;

(16)

3

Nowak & Vallacher, 1998; Vallacher et al., 2015; Van Geert, 1991, 2009). Therefore, a

dynamical system is continuously undergoing change, and shows natural fluctuations in physiological and psychological states over time. The dynamics of people’s physiological and psychological development is typically individual-specific (Den Hartigh, Van Dijk et al., 2016), and people may respond to events in their own way (Den Hartigh, Hill et al., 2018).

The properties of variation over time and idiosyncrasy in trajectories appear to match characteristic features of resilience. Hill and colleagues (2018a) therefore suggested to apply the toolbox of the dynamical systems perspective to study people’s resilience. Specifically, (nonlinear) time-series analyses can be conducted on repeated measurements of individuals’ psychological or behavioral states (e.g., Araújo et al., 2015; Den Hartigh et al., 2017). Such techniques may reveal distinct patterns that inform about a system’s resilience (e.g., Kiefer & Myer, 2015) or when a system’s resilience is lost (Hill et al., in press; Scheffer et al., 2009, 2012). For instance, Kiefer and Myer (2015) found that resilient movement systems show an optimal blend of stability and flexibility in their temporal organization. Furthermore, individual-level fluctuations may signal re-organization of structural components within a system in order to produce functional behavioral or structural adaptations (Kiefer et al., 2018). Moreover, increasing fluctuations in a psychological or performance variable may be linked to resilience losses (Hill et al., in press). This approach thus shifts the focus from identifying isolated sets of factors that may predict resilience, to assessing temporal patterns that inform about the adaptability of human psychology and behavior.

4.2 | Practical Implications

In line with the findings of the current study and the proposed application of dynamical systems theory to studying resilience, there are two major implications for practitioners. First, following the lack of generalizability from group-level data to individuals, coaches need to pay close attention to how protective factors and resilience relate to each other in each individual athlete. To reiterate, only because researchers may find a positive association between a given protective factor and resilience at a group-level, does not necessarily mean that changes in this protective factor are related to changes in resilience within a particular individual athlete (cf. Hamaker, 2012). However, coaches can utilize the literature to identify potential factors, which may be applicable to their athletes. The task of the coach is then to identify what factors can be tailored to their athletes, and how.

Second, protective factors and resilience should be assessed repeatedly over time. Ideally, practitioners create a time-series of dense, repeated measurements of their variables

(17)

of interest (Araújo et al., 2015; Den Hartigh et al., 2017). Such time-series would allow practitioners to track whether the relationships between variables change over time in different individuals and whether the observed fluctuations show meaningful patterns, which can inform about changes in an athlete’s resilience (Hill et al., in press; Kiefer & Myer, 2015). This may allow practitioners to assess how, for example, protective factors and resilience interact over time within an individual. Consequently, interventions, which target the necessary variable tailored to each individual, can be designed. Furthermore, Hill and colleagues (2018a) argued that identifying critical changes in time-serial patterns may indicate when interventions need to be implemented. Therefore, collecting and analyzing individual time-series of athletes foster “wise” interventions that targeted to relevant variables, tailored to each individual, and potentially timed to prevent resilience losses (cf. Walton, 2014).

4.3 | Limitations and Future Directions

The design of the current study has two limitations. First, the findings of the current study are based on 21 assessment points of relatively few participants. Although the drop-out rate for this assessment range already approached 50%, increasing the number of data-points allows for more rigorous assessment of time-variation in respondents’ behavior, even for relatively small samples (e.g., Delignières & Marmelat, 2014; Den Hartigh et al., 2015). Therefore, future studies may aim to extend the number of data points, to allow for different time series techniques that inform about time-varying patterns in the data. The second limitation of the current study centers around the use of self-report measures. All the measures of the protective factors included in this study are based on self-reports. Although there is often no better method to measure psychological variables, the quality of self-report data is not always evident. For example, when presented with a questionnaire containing only answers, but no actual questions, participants can show a positive response bias for items, such as “yes”, “true”, or “agreed” (Berg & Rapaport, 1954; De Jonge & Slaets, 2005; Van Heerden & Hoogstraten, 1979; Verma, et al., 1980). Nevertheless, the diary study was a deliberate choice for testing the hypotheses as this design allows for continuous assessment of psychological variables over an extended period. Future studies may include data from different sources, including psychological, behavioral, physical and physiological parameters (Blaauw et al., 2016).

5 | Conclusion

In conclusion, the current study provides first empirical support for the notions that (1) the relationship between different protective factors and protective factors with

(18)

3

resilience may vary across measurement moments, and (2) trajectories of protective

factors and resilience are not generalizable from averaged group data to individuals. These conclusions are in line with the conceptualization of resilience as a complex process, which emerges from dynamic interactions between multiple components (Fletcher, 2018; Hill et al., 2018a). Accordingly, we discussed the dynamical systems approach to studying resilience as it accounts for the fluctuations of the protective factors over time. The dynamical systems approach is tailored to capturing dynamic non-ergodic processes. Based on our findings and the fit between them and the theoretical underpinnings of dynamical systems theory, we propose to apply nonlinear time series analyses to studying resilience in sports and beyond. The application of a dynamical systems approach also extends to practical implications as coaches need to become aware that group-level research may not be applicable to individual athletes and that fluctuations over time are not to be ignored as mere error, but may contain functional information about their athletes.

(19)