A growing body of literature investigated the relationship between NSSI and affect with correlational and ANOVA-based analyses (e.g., Victor & Klonsky, 2014). However, no previous study was grounded in a nonlinear-dynamic systems framework where it is important to consider dynamic characteristics (such as fluctuations or critical slowing down) of affect and its relation to NSSI.

According to the synergetic approach to psychology, human change processes are characterized by non-linearity and non-stationarity of their dynamics (Schiepek & Strunk, 2010). When the relationship between input and output of a dynamic system is not directly proportional, the system is considered nonlinear (Stam, 2005). It is typical for nonlinear dynamic systems that, depending on the state of the system, even high levels of negative affect (input) do not lead to non-suicidal self-injury (output). In contrast, when the system is in an unstable state, even minor increases of negative affect can trigger self-injurious behavior. Nonstationarity, on the other hand, denotes changes in descriptive statistics, including (moving) averages and variance, over time. Nonstationarity is consistent with discontinuous phase transitions caused by the nonlinear mechanisms inherent in complex systems. Phase transitions are defined as a change from one pattern to another pattern, caused by internal and external conditions. Almost invariably, phase transitions are preceded and facilitated with critical instabilities (Schiepek & Strunk, 2010).

Critical instabilities preceding phase transitions can be assessed with the dynamic complexity (DC), which was developed for short time series typical in psychological research (Schiepek et al., 2014). DC is the product of two parameters, fluctuation (F) and distribution (D). F quantifies the intensity of change with respect to frequency and amplitude of a time-series (Figure 1). F is maximal when the variable in the time-series jumps between the minimum and maximum value as often as possible. D quantifies the scattering of the time-series within the range of possible values. D maximizes when values are irregular and chaotically distributed (for a detailed description and formula, see Schiepek & Strunk, 2010). Consequently, high DC indicates frequent, large, and irregular fluctuations (Olthof et al., 2020).

F and D and their product DC indicate critical instabilities in self-organizing processes and non-stationary phenomena in time series (Schiepek & Strunk, 2010). During periods of fluctuation, the system is destabilized and oscillates between old and new patterns until the system stabilizes in a new state (attractor) and variability decreases (Kelso, 1995; Thelen & Smith, 1994). Typically, DC is computed by looking at a window of data consisting of seven measurement points that are shifted from one day to the next throughout the time series. Applying this procedure produces a time series that includes DC values for each item (Schiepek & Strunk, 2010). A manifestation of nonlinear dynamics and phase transitions is the rise and fall in affect intensity over time. Individuals engaging in NSSI typically show high affective variability (Victor et al., 2021), which reflects the overall amplitude of affective changes. Affective changes (from positive to negative affective states) can be considered as order transitions. The typical high affect fluctuations in individuals engaging in NSSI could be recognized by the presence of more critical instabilities in their affect data. Therefore, we hypothesize that an order transition phenomenon like affective change is accompanied by phases of critical instabilities (fluctuations in affect) of the time series resulting in an increase of dynamic complexity in individuals with a history of NSSI compared to controls.

To summarize, research has proposed that NSSI is associated with intense negative affect and affective instability. Although the evidence has highlighted the role of emotion regulation in NSSI through negative reinforcement of aversive affect (e.g., Nock & Prinstein, 2004), some research has examined positively reinforcing effects (e.g., Jenkins & Schmitz, 2012).

Based on the literature described above, the present study is the first to apply the concept of DC to the description and analysis of NSSI. We are particularly interested in the following three hypotheses: First, we expect individuals engaging in NSSI to show a higher DC than controls without a history of NSSI. The resulting high DC in the NSSI group should be represented by frequent, large, and irregular fluctuations in time series. Second, we aim to examine mean levels of positive and negative affect in the NSSI group compared to controls. Previous research mostly focused on the role of negative affect and negative reinforcement in NSSI. Therefore, one could assume that participants with NSSI also show higher negative affect. However, we also wanted to investigate whether positive affect might be present as well. So, investigating DC in affective states that differ between individuals with and without a history of NSSI may be an opportunity to identify potential maintaining or risk factors of NSSI. Third, we investigate whether the average values of dynamic complexity change over time. We assume that the dynamic complexity should decrease during the course of the hospital stay, as therapeutic and pharmacological interventions should lead to the stabilization of the patients.

Ecological momentary assessment (EMA) enables an assessment of behavior and psychological processes in everyday life. EMA reduces the influence of recall biases, enhances research findings' ecological validity, and tracks changes in intra-individual psychological processes (Stone & Shiffman, 1994). Another advantage of EMA is the distinction between state affect and trait affect.

Inclusion criteria were a) suffering from a mental disorder; b) aged between 18 and 65 years; c) having adequate language skills in German to understand and follow instructions and give informed consent; d) non-suicidal self-injury in the last year for the NSSI-group and no such behavior in the control-group. Exclusion criteria were a) acute episodes of psychosis, psychotic or manic episodes, rapid cycling, and acute substance use; b) missing cognitive skills to complete assessment.

61 subjects selected according to the procedure described below met the required criteria, 33 (54.1%) of them did not show NSSI and constitute the control group (22 women, 11 men) with a mean age of 35.0 (SD = 10.73) years. In this group, 18 (54.5%) met the diagnostic criteria (ICD-10) of a depressive disorder, ten (30.3%) of an anxiety disorder, two (6.1%) of an obsessive-compulsive disorder, two (6.1%) of a personality disorder (avoidant and dependent personality disorder), and one (3.0%) of insomnia. The NSSI group consisted of 28 (45.9%) participants (26 females, two males) with a mean age of 26.69 (SD = 8.73) years. Of these, 12 (42.9%) met ICD-10 criteria for a depressive disorder, eight (28.6%) for a post-traumatic stress disorder, five (17.9%) for a primary diagnosis of borderline personality disorder, and three (10.7%) for an eating disorder. In addition, eight (28.6%) were diagnosed with comorbid borderline personality disorder. The diagnoses were initially assigned by the treating psychiatrist based on the clinical picture and confirmed in a second step by the researcher using the ICD-10 research criteria.

Subjects were recruited from an Austrian adult inpatient psychiatric unit (“Kardinal Schwarzenberg Klinik”) located in the southern area of the federal state of Salzburg. Patients meeting inclusion criteria were invited to participate in the study and were informed about the studies’ procedure in detail and written informed consent was obtained. Additionally, the phone number was recorded for being able to contact subjects during the study. Prior to commencement, the study procedure was approved by the Ethics Committee of the Province of Salzburg (EC number: 415-E). Only the third hypothesis was preregistered (see Bruckbauer-Schwed et al., 2023S), as the data for the first and second hypotheses had already been analyzed at the time of registration. Data collection started with a general questionnaire asking for age, gender, professional and family status and NSSI. With respect to that, subjects were asked if ever in their lives they had performed any NSSI-behavior; if yes, they were additionally asked about its frequency over their lifetime. Positive and negative affect was recorded by using the Positive and Negative Affective Schedule (PANAS) in an EMA-design over a period of at least six days. Participants received no compensation for their participation.

EMA was initiated by an invitation by a SMS-message. The questionnaire could be retrieved and completed in SoSci Survey (https://www.soscisurvey.de) on the subjects’ mobile phones. Due to technical problems, data acquisition had to be changed to the mobile PIEL App (Jessup et al., 2012). With a wake-up function subjects were prompted to answer their questionnaires at 6 p.m. daily taking approximately five minutes to be completed. For assistance participants could contact the researcher at any time during the study period. It was also possible to complete the questionnaire in a paper-pencil-format. Two participants choose this option for some of their entries.

The Positive and Negative Affective Schedule (PANAS; Watson et al., 1988; German: Krohne et al., 1996) consists of two 10-item scales, one for positive and one for negative affect. Participants rated, how they generally felt during each day with respect to each emotion on a 5-point Likert-like intensity scale (1 = very slightly/not at all, 5 = extremely). The internal consistency of the PANAS is high, with Cronbach alpha’s of 0.90 for positive and 0.87 for negative affect (Watson et al., 1988).

Dynamic complexity (DC) (Schiepek & Strunk, 2010) was computed for the PANAS´s positive and negative affect subscales. DC is a measure for identifying critical instabilities and fluctuations in short, coarse-grained, time series like those obtained in typical diary studies. It is calculated from two parameters: fluctuation intensity and distribution (see section on DC above). The fluctuation measure is highest when the time series fluctuates widely between its minimum and maximum, while the distribution measure is larger when it takes many different values. Both parameters are then multiplied by each other, resulting in a total dynamic complexity measure. Both parameters were calculated in moving windows. Like other studies using this method (e.g., Schiepek et al., 2014), we chose a moving window size of seven days. To identify local complexity peaks, we calculated the difference between the average and maximum complexity of each PANAS item. Peak complexity values for individual positive and negative affect items were averaged to obtain scale-level peak complexity values. We then used t-tests to compare the peak complexity values for the positive and negative affect scales of patients with NSSI to those of controls. Finally, to test whether average dynamic complexity values change over time, we fitted linear mixed models to predict dynamic complexity of the positive and negative affect scales from number of days since admission and group (NSSI vs. controls). Models included random intercepts for each participant and an interaction effect of the group by time. The linear mixed models were included to determine whether there were temporal trends in the development of dynamic complexity. Missing values were imputed using Kalman filter imputation.

Our study has several limitations. The first is that our sample size is not very large, which limits generalizability. On the other hand, the results suggest that intra-individual differences in dynamic complexity explain significantly more variance than differences in assessment duration or between-group differences. Future studies with larger samples and, thus, higher power to detect smaller effects, should be done to replicate our findings. A second major limitation is that we made only one measurement per day. Normally, affective changes occur several times per day, especially in emotionally unstable subjects. Consequently, a single assessment per day may not have been sufficient to capture the dynamics of emotional change. Typical ambulatory assessment studies therefore perform several measurements per day. However, this limitation may also be an advantage, as studies have found lower compliance rates in these designs. In their systematic review, Hepp et al. (2020) found an average compliance rate of 64.9% with prompts for more than one daily measurement, while a single daily measurement resulted in an average compliance rate of 78.9%. To minimize the risk of low response rates and early study termination, we chose to use a single measurement per day design. A third possible limitation is that participation in our study was not incentivized. Hepp et al. (2020), in their systematic review of daily life studies of NSSI, found that incentivized studies resulted in higher compliance rates. A fourth limitation is the assessment of the frequency of NSSI. We could only ask our participants for a retrospective estimate of the number of times they had injured themselves in their lifetime. This type of assessment is problematic because individuals may have had difficulties in recalling the exact number of lifetime self-injurious acts. In addition, many members of the NSSI sample have engaged in NSSI continuously over a long period of their lives. In this case, the exact number of NSSI events is likely to be impossible to assess. Due to the lack of compliance, it was not possible to assess individual acts of NSSI during the clinic stay or the survey period. Thus, we could not directly examine associations between acts or attempts of NSSI and affective states. As a result, we could not examine the relationship between emotions and NSSI acts, e.g., whether positive (or negative) affect occurs before or, in the case of a positive act, after an NSSI act. We were unable to assess the effects of the medications that participants received during their stay, as well as the effects of their psychological treatments on their affect fluctuations.

The results of our study illustrate that individuals with a history of NSSI show greater dynamic complexity for positive as well as negative affect and higher mean levels of positive affect. Future studies should explicitly consider this possibility and explore the dynamics of positive emotions in the context of NSSI more intensively, especially with larger samples, but also in individual case studies as a second very insightful strategy. The higher variance explained at the individual level suggests that the application of dynamic complexity is more meaningful when considered at the individual case level. Regarding prevention and treatment, future research should also focus on specific patterns of affect dynamics to identify early warning signs of critical instability and order transitions in individuals with a history of NSSI. Knowledge of these phenomena offers the possibility of early intervention in affective change. For example, van de Leemput and colleagues (2014) reported that specific dynamic patterns of emotions are associated with the onset and offset of episodes of major depression. In addition, anticipating and observing order transitions in affect may be useful for clinical assessment of mental disorders and treatment planning (e.g., dialectical behavior therapy (DBT)) to reduce affective instability).