Trajectories of nonsuicidal self-injury during adolescence
Abstract
Objective
Although nonsuicidal self-injury is a public health concern, there is little information on how it changes across adolescence or what contributes to stability or change. We aimed to identify trajectories of stability and change in self-injury from ages 13 to 17 years, and to identify interpersonal and intrapersonal correlates that differentiate between trajectories of stability and change.
Method
We used five annual waves of cohort-sequential data, targeting 7th and 8th graders attending all public schools in three municipalities in central Sweden. The data were gathered via questionnaires, using a multi-item measure of non-suicidal self-injury and assessing negative experiences at home, in school, with peers, and in romantic settings, as well as intrapersonal issues (internalizing symptoms and difficulties with emotional, and behavioral regulation). The analytic sample was 3195 adolescents (51.7% boys, 48.3% girls; ages 12–16 years at T1, M = 13.61; SD = 0.66), most of whom were born in Sweden (88.6%) to at least one parent of Swedish origin (77.4%).
Results
Latent growth curve modeling revealed three self-injury trajectories: a stable-low, a low-increasing, and an increasing-decreasing trajectory. Adolescents in the stable-low class reported the best overall adjustment at ages 13 and 16. Comparatively, adolescents in the other two classes reported similar levels of difficulty interpersonally and intrapersonally. Where they differed, the increase-decrease class fared worse than the low-increasing class.
Conclusions
This study suggests the need to frame self-injury as having multiple directions of development during adolescence and develop theory that aligns with differential patterns of self-injury development.
1 INTRODUCTION
A public health concern that tends to emerge during adolescence is non-suicidal self-injury (e.g., Westers & Culyba, 2018; hereafter self-injury), defined as intentional destruction of body tissue without suicidal intentions (Nock & Favazza, 2009). This seems to be particularly relevant in Sweden, where more than a third of adolescents have recounted self-harming within the year before their reports (e.g., Zetterqvist et al., 2013). Despite an increased effort among scholars to better understand this behavior and the aspects associated with it, there is next to no information on how self-injury changes across adolescence nor information about which factors contribute to the change and stability of self-injury behaviors. Our aim was to examine the trajectories of change of self-injury and associated interpersonal experiences and intrapersonal challenges during adolescence.
Self-injury includes behaviors such as skin cutting or self-battery. In line with an extensive line of research (e.g., Gratz, 2003; Klonsky, 2007; Nock & Prinstein, 2004), adolescents use this behavior for different reasons. However, its affect regulation function seems to be the most common (Gratz, 2003; Taylor et al., 2018). Where this function is concerned, several models can be applied, including the emotional cascade model (Selby & Joiner, 2009), the experiential avoidance model (Chapman et al., 2006), or the cognitive-emotional model (Hasking et al., 2017). They all postulate that self-injury is a coping strategy that helps adolescents to manage emotions experienced as overwhelming or otherwise difficult to handle. Hence, in the search for precursors of self-injury, researchers have focused on personal experiences that generate strong, negative emotions and disrupt the development of healthy emotion-regulation strategies, as well as on individual characteristics and behaviors that contribute to difficulties managing these emotions. These precursors can be divided, roughly, into interpersonal experiences and intrapersonal characteristics or behaviors (Wang et al., 2016).
1.1 Development of self-injury during adolescence
Self-injury has its onset in adolescence, usually between 12 and 14 years of age (e.g., Cipriano et al., 2017; Gandhi et al., 2018). According to several studies, the prevalence of self-injury peaks around 15 years of age (e.g., Barrocas et al., 2012; Gandhi et al., 2018), decreasing later on in early adulthood (e.g., Plener et al., 2015). However, some studies using community samples indicated a later onset of self-injury, somewhere between the ages of 17 and 20 (e.g., Gandhi et al., 2018; Plener et al., 2016). These findings suggest that there are different patterns of self-injury development. However, knowledge of how self-injury changes across adolescence is rather scarce.
That is, there are only a handful of longitudinal studies, most of which used between two (e.g., Wu et al., 2021) and three waves of data (e.g., Gandhi et al., 2019; Guerry & Prinstein, 2009 with data at ages 9, 15, and 18; Steinhoff et al., 2021). Although useful, these studies did not provide enough data to examine nonlinear trajectories during adolescence, nor did they cover most of the period of adolescence. In addition, they focused on mean-level change and did not account for subgroups of adolescents, whose development may not look anything like the average trajectory.
To our knowledge, only three studies used more than three waves of data and assessed trajectories of change across different subgroups of adolescents. All three used samples from China. The first and second studies used the same data and reported the same trajectories (Barrocas et al., 2015; Giletta et al., 2015). Using eight assessments from grades 10 to 12, covering mid- to late-adolescence, these researchers found three trajectories. One trajectory indicating negligible levels dropping to no self-injury (69% of the sample), another trajectory class starting at moderately low levels and decreasing (26%). A final, small class (~5%) started relatively high and increased (called chronically high). Barrocas et al. reported that these classes were distinguished by gender and variables associated with difficulty regulating emotions. Specifically, being classified with the low trajectory was predicted by identifying as a girl, reporting less lifetime depression, ruminating less, and having a less negative attributional style, compared with the other two trajectories. Only attributional style differentiated the moderate and chronic classes, with higher levels predicting inclusion in the chronic class. In the second paper, Giletta et al. (2015) reported that in addition to symptoms of depression differentiating the class of low levels from trajectories with moderate and high levels, peer victimization (overt and relational) differentiated low and moderate trajectories from the high/chronic trajectory.
The other study, which covered most of adolescence, used a short-term (1 year), three-wave, accelerated longitudinal design (Wang et al., 2016). In this study, the data were grouped by age, so that trajectories from age 12 to age 17 years could be examined. Using latent growth curve analysis, they identified four trajectories. One trajectory, labeled negligible, represented most of the adolescents in the study (74.6%) and had initial levels that indicated little self-injury and then dropped to none. A second trajectory, accounting for 12.8% of the sample, had very low levels, dropping close to none by age 17. The third trajectory, representing 10.8% of the sample, started at moderate levels, followed by decreases. A final, every small group (1.9%) had high levels that fluctuated across the age range studied, with the levels dropping overall. The researchers sought to differentiate between these trajectories, examining correlates of change. They found that girls were over-represented in trajectory groups with non-negligible levels of self-injury. They also examined aspects of initial levels of intraindividual functioning, including depressive and anxiety symptoms, self-criticism, and impulsivity, as well as two interpersonal variables, namely unstable relationships and parental criticism. They found that most of their variables only distinguished the negligible trajectory from the others. Of note, only impulsivity illuminated differences between others. Specifically, adolescents whose most likely classification was in the moderate trajectory reported more impulsivity than those in the experimental trajectory. Self-criticism, by contrast, did not predict differences.
The commonality between these studies was that the bulk of the samples were classified in trajectories that had little or no self-injury across the time of the studies, indicating that most adolescents do not engage in self-injury. Both also found trajectories with relatively high levels, although only Barrocas et al. found a trajectory that increased. In addition, both examined interpersonal experiences and intrapersonal characteristics as predictors, but with inconsistencies. For example, they found opposing patterns of gender differences. However, both found few predictors that differentiated between the classes with higher levels of self-injury. In short, more information is needed to find out what trajectories exist and what predicts variation in trajectories of self-injury.
1.2 Theoretical models
An important step in understanding self-injury is to document the variations in its change across adolescence, where person-centered analyses reveal variation that variable-centered analyses do not. Focus on variation is important because small subgroups of individuals whose self-injury may increase or decrease across adolescence can be easily obscured by large groups. In the case of self-injury, large numbers of adolescents who engage in negligible amounts of self-injury seem to drive the average trajectory, masking greater change that may be nonlinear. As there are no theoretical models to define when change is likely, we drew from previous studies and from developmental models that suggest when experiences with others (in school, with peers, with parents) may be especially challenging and when adolescents may have the greatest difficulties managing their emotions. Models of parent-adolescent relations suggest that the relationship changes most in early adolescence, when the frequency of conflicts with parents increase (Laursen & Collins, 1994). Early to middle adolescence is also a period when many adolescents begin to develop romantic relationships, which can be particularly challenging (Welsh et al., 2003). Later, relationship difficulties with parents and peers tend to diminish, coinciding with adolescents developing better regulatory and coping strategies. These descriptions suggest that early adolescence may be a period of particular vulnerability for self-injury, with increases into middle adolescence and dropping thereafter. This would be consistent with the mean-level trajectories reported by Barrocas et al., 2012) and Gandhi et al., 2018). However, a smaller number of adolescents may continue to have difficulties, as indicated by the trajectory with increased rates found by Baroccas et al. (2015). It is most likely, though, that individual variation in experiences and regulatory abilities contribute to the timing of increases and decreases in self-injury. Thus, it is important to identify potential sources of variation in self-injury trajectories. This step—being able to distinguish between different trajectories—is crucial for understanding who is likely to desist on their own and who may benefit from prevention or intervention.
Without theory guiding when change in self-injury should occur, we drew from the functional models (e.g., Hasking et al., 2017) that others have used. According to these models, environmental stimuli often trigger the emotional, cognitive, and behavioral pathways that lead to self-injury. Specifically, the social-ecological approach (Heilbron et al., 2014) suggests that problematic interactions within social contexts, such as conflicts with parents, friends, and romantic partners, are perceived as particularly stressful and generate negative emotional experiences that might lead to the use of self-injury as a strategy to cope with such experiences. In support of this idea, studies on different community samples of adolescents identified interpersonal conflicts with parents, friends, or romantic partners as correlates of self-injury (e.g., Latina et al., 2022; Mirsu-Paun & Oliver, 2017; Turner et al., 2016). Other negative experiences may be forms of punitive, aggressive, or manipulative behaviors directed at adolescents. These experiences include parents engaging in psychological control (Huang et al., 2022) or directing unregulated anger at the adolescent (Wang et al., 2016), which could disrupt parent-adolescent bonds, which is predictive of self-injury (Victor et al., 2019). In addition, being the victim of relational and physical/overt aggression from peers (Giletta et al., 2015) could trigger self-injury.
In addition to the stress produced by these experiences, strong emotional responses to stressful events may be difficult to handle for adolescents who do not have adequate strategies for managing their emotional arousal. It is thought that the inability to dampen arousal or engaging in behaviors that amplify arousal leads to the use of self-injury as a coping strategy. Accordingly, rumination is one of the central processes linking emotional experiences to self-injury in the emotional cascade model (Selby & Joiner, 2009). Rumination and other forms of repetitive negative thinking amplify arousal because there is repeated focus on negative aspects of experiences. Rumination tends to focus on events past and present, whereas worry focuses on potential (future) events. Studies support such cascade models, where adolescents’ self-injury was associated with more rumination (e.g., Hoff & Muehlenkamp, 2009; Nagy et al., 2022; Sorgi-Wilson et al., 2022). We speculate that worry, another perseverative cognition focused on anticipating negative events, may have similar ties to self-injury, as worry and rumination are similar in processes, but differ in their time orientation (Watkins et al., 2005).
In addition, and in line with some functional model of self-injury (Gratz & Roemer, 2004; Hasking et al., 2017), effective emotion regulation requires the ability to control impulses and behaviors. This seems to be the case for adolescents who engage in self-injury, who present higher self-reported impulsivity (e.g., Hamza et al., 2015; Lockwood et al., 2020) and anger dysregulation (Latina & Stattin, 2016) compared to those who do not engage or engage in less self-injury.
The tendency to gravitate towards negative and repetitive thoughts seems to be related to other forms of psychopathology that are linked to self-injury, especially forms of internalizing (i.e., depression and anxiety, Hosseinichimeh et al., 2018; Watkins & Roberts, 2020; Young & Dietrich, 2015; and self-criticism, Gadassi Polack et al., 2021).
All in all, the current theoretical and empirical literature on self-injury seem to point at specific interpersonal and intrapersonal factors that could differentiate between adolescents who might engage in a self-injury trajectory and those who will not. Based on the above-mentioned findings, the identification of individual and contextual features that differentiate self-injury trajectories is crucial to create targeted interventions for the group of adolescents whose risk is the greatest.
1.3 The present study
The present study had two aims. First, we aimed to identify trajectories of self-injury development across adolescence. Existing evidence suggests that we should expect between two and five trajectories. We hypothesized that we would find one very large group of adolescents whose self-injury was consistently negligible or quite low. We viewed three other groups as also likely: one that might reflect average trajectories (Barrocas et al., 2012), peaking in middle adolescence, dropping through to late adolescence; and two others that increase across adolescence, but vary in both initial levels and rate of change. These latter two trajectories would replicate the findings of Wang et al. (2016). A final trajectory, implied by studies involving late adolescents, might emerge where self-injury does not rise until after middle adolescence, perhaps due to nonnormative experiences.
Our second aim was to identify individual and contextual processes that would differentiate between trajectories of self-injury. Applying models linking challenging experiences and regulatory difficulties to self-injury (e.g., Chapman et al., 2006; Hasking et al., 2017; Nock & Prinstein, 2004) we targeted the two broad categories linked with self-injury. For interpersonal experiences, we examined stressors at home, in school, with peers and romantic partners, with special attention to experiences of feeling alienated from parents and abusive behaviors from parents and peers. For intrapersonal characteristics and behaviors, we included gender, adolescents’ internalizing (depression, anxiety, and low self-esteem), repetitive negative thinking (rumination and worry), and impulsivity. As a social form of coping that is similar to rumination in its focus on negative events, we also included co-rumination. As the trajectory analyses were rather exploratory, we only speculated that increases in self-injury would be seen when the individuals with that trajectory were experiencing greater difficulties with interpersonal experiences and in regulating themselves. We also speculated that girls may be over-represented in trajectories involving higher levels of self-injury, based on some findings that girls report more self-injury than do boys (Bresin & Schoenleber, 2015).
2 METHODS
2.1 Design and procedures
The data for this study came from the 3 Cities Study (TreStadStudien), a cohort-sequential study of Swedish adolescent in three municipalities in central Sweden. The data were collected from students who were in 7th and 8th grade at the first assessment (in Sweden, typically between ages 12 to 15 years). Annual data collections continued for 4 years (5 waves of data), during the spring, from the beginning of February and to the beginning of June.
Before the first data collection, ethics approval was obtained from the Uppsala Regional Ethics Board (# 2013-384). Each year, parents were informed about the study through the mail and given the opportunity to decline their adolescents’ participation in the study (i.e., passive consent) via postage-paid cards, email, or by phone. Less than half a percent denied consent (nT1 = 121, nT3 = 10, nT3 = 65, total n = 196).
Data collection took place at school, during class hours, supervised by trained assistants, without teachers in the rooms. Adolescents were informed of the study's purpose, what their participation entailed, and their rights as research participants (including the right to decline or to terminate participation at any time). Students were given a total of 3 h to complete a two-part questionnaire, with a 30-minute break in the middle, when they were provided with refreshments. Each class was given a 300 kronor (~30 Euro) honorarium for their class trip funds. Annually, the research team returned to collect data from the students who had previously participated and any new students who wished to participate.
2.2 Measures
We used assessments from all five waves for self-injury and from the first four waves for correlates. In Table 1, information is provided about the number of items, range of possible responses, sample items, estimates of reliability, and descriptive statistics.
| Measure | Items | Items range | Sample item | T | α | M | SD | N |
|---|---|---|---|---|---|---|---|---|
| Non-suicidal self-injury | 9 | “Zero times” (0) to “More than five times” (6) | In the last 6 months, have you purposely cut your wrists, arms, or some other part of your body (without intending to kill yourself)? | 1 | .92 | 0.13 | 0.45 | 2313 |
| 2 | .88 | 0.16 | 0.52 | 2513 | ||||
| 3 | .93 | 0.16 | 0.57 | 1866 | ||||
| 4 | .88 | 0.15 | 0.51 | 1951 | ||||
| 5 | .99 | 0.14 | 0.48 | 2141 | ||||
| Interpersonal experiences | ||||||||
| Stressors | ||||||||
| Family-related stress | 4 | “Not at all stressful” (1) to “Very stressful” (5) | In the last 6 months, how stressful has it been for you to have arguments at home? | 1 | .84 | 1.74 | 0.79 | 2342 |
| 2 | .86 | 1.81 | 0.85 | 2507 | ||||
| 3 | .88 | 1.82 | 0.90 | 2870 | ||||
| 4 | .87 | 1.86 | 0.89 | 1952 | ||||
| Romantic relationship stress | 3 | “Not at all stressful” (1) to “Very stressful” (5) | In the last 6 months, how stressful has it been for you to break up with your boy/girlfriend? | 1 | .80 | 1.31 | 0.66 | 2306 |
| 2 | .82 | 1.38 | 0.77 | 2479 | ||||
| 3 | .87 | 1.42 | 0.83 | 1854 | ||||
| 4 | .85 | 1.41 | 0.83 | 1941 | ||||
| Peer-related stress | 4 | “Not at all stressful” (1) to “Very stressful” (5) | In the last 6 months, how stressful has it been for you to fit in with peers? | 1 | .80 | 1.36 | 0.59 | 2350 |
| 2 | .78 | 1.36 | 0.58 | 2517 | ||||
| 3 | .80 | 1.33 | 0.57 | 1873 | ||||
| 4 | .75 | 1.37 | 0.57 | 1956 | ||||
| School-related stress | 3 | “Not at all stressful” (1) to “Very stressful” (5) | In the last 6 months, how stressful has it been to keep up with schoolwork? | 1 | .85 | 2.57 | 1.08 | 2355 |
| 2 | .85 | 2.91 | 1.14 | 2516 | ||||
| 3 | .88 | 2.86 | 1.20 | 1872 | ||||
| 4 | .87 | 2.93 | 1.58 | 1957 | ||||
| Peer victimization | ||||||||
| Relational victimization | 6 | “Never” (1) to “Daily” (5) | Other kids say mean things behind my back. | 1 | .83 | 1.41 | 0.53 | 2341 |
| 2 | .85 | 1.43 | 0.57 | 2494 | ||||
| 3 | .86 | 1.35 | 0.53 | 1857 | ||||
| 4 | .84 | 1.33 | 0.51 | 1929 | ||||
| Physical victimization | 4 | “Never” (1) to “Daily” (5) | Other kids hit me. | 1 | .82 | 1.20 | 0.45 | 2344 |
| 2 | .84 | 1.16 | 0.43 | 2490 | ||||
| 3 | .87 | 1.14 | 0.44 | 1855 | ||||
| 4 | .83 | 1.10 | 0.35 | 1932 | ||||
| Negative experiences with parents | ||||||||
| Maternal alienation | 6 | “Almost never or never true” (1) to “Almost always or always true” (5) | I don't know who I can depend on. | 2 | .78 | 1.86 | 0.74 | 2505 |
| 3 | .79 | 1.84 | 0.75 | 1865 | ||||
| 4 | .82 | 1.84 | 0.74 | 1926 | ||||
| Paternal alienationa | 2 | .80 | 2.01 | 0.82 | 2447 | |||
| 3 | .80 | 1.97 | 0.81 | 1818 | ||||
| 4 | .84 | 1.99 | 0.84 | 1884 | ||||
| Maternal coldness-rejection | 6 | “Never” (1) to “Often” (3) | When you do something your mother doesn't like, is she silent and cold towards you? | 1 | .75 | 1.28 | 0.37 | 2306 |
| Paternal coldness-rejectiona | 1 | .76 | 1.29 | 0.37 | 2189 | |||
| Maternal angry outbursts | 5 | “Never” (1) to “Often” (3) | When you do something that your mother doesn't like, does she get very mad and have an angry outburst? | 1 | .88 | 1.42 | 0.50 | 2298 |
| Paternal angry outburstsa | 1 | .87 | 1.44 | 0.51 | 2184 | |||
| Maternal dependency-oriented control | 8 | “Strongly disagree” (1) to “Strongly agree” (5) | My mother is only happy with me if I rely exclusively on her/him for advice. | 2 | .82 | 1.96 | 0.81 | 2492 |
| 3 | .85 | 1.89 | 0.80 | 1851 | ||||
| 4 | .84 | 1.84 | 0.76 | 1917 | ||||
| Paternal dependency-oriented controla | 2 | .86 | 1.77 | 0.80 | 2406 | |||
| 3 | .83 | 1.70 | 0.77 | 1793 | ||||
| 4 | .86 | 1.63 | 0.73 | 1853 | ||||
| Intrapersonal characteristics | ||||||||
| Internalizing | ||||||||
| Depressive symptoms | 6 | “Not at all” (0) to “A lot” (3) | I didn't sleep as well as I usually sleep. | 1 | .95 | 1.83 | 0.72 | 2342 |
| 2 | .90 | 2.13 | 0.79 | 2518 | ||||
| 3 | .91 | 2.12 | 0.79 | 1863 | ||||
| 4 | .92 | 2.29 | 0.86 | 1950 | ||||
| Anxiety symptoms | 5 | “None” (0) to “All the time” (4) | In the past week, how often did you avoid situations, places, objects, or activities because of anxiety or fear? | 1 | .87 | 1.59 | 0.69 | 2332 |
| 2 | .89 | 1.71 | 0.80 | 2515 | ||||
| 3 | .88 | 1.76 | 0.82 | 1868 | ||||
| 4 | .90 | 1.87 | 0.88 | 1945 | ||||
| Self-esteem | 10 | “Not true at all” (1) to “True” (4) | On the whole, I am satisfied with myself. | 1 | .88 | 3.10 | 0.62 | 2319 |
| Regulation issues | ||||||||
| Rumination | 10 | “Never” (1) to “All the times” (4) | When I'm feeling sad, I think about the things that have happened over and over again. | 1 | .93 | 2.47 | 0.92 | 2320 |
| 2 | .93 | 2.82 | 0.96 | 2512 | ||||
| 3 | .94 | 2.88 | 1.01 | 1866 | ||||
| 4 | .93 | 3.03 | 0.95 | 1943 | ||||
| Worry | 13 | “Not at all true” (1) to “Always true” (4) | When I'm under pressure, I worry a lot. | 1 | 82 | 1.90 | 0.50 | 2324 |
| 2 | .84 | 2.05 | 0.56 | 2504 | ||||
| 3 | .85 | 2.06 | 0.57 | 1868 | ||||
| 4 | .85 | 2.14 | 0.58 | 1941 | ||||
| Difficulty controlling anger | 5 | “I don't agree at all” (1) to “I completely agree” (4) | When you get REALLY ANGRY with someone, you feel like you are lacking control over yourself. | 1 | .73 | 1.88 | 0.54 | 2306 |
| 2 | .75 | 1.88 | 0.67 | 2520 | ||||
| 3 | .78 | 1.86 | 0.70 | 1865 | ||||
| 4 | .75 | 1.80 | 0.64 | 1953 | ||||
| Impulsivity-urgency | 11 | “I don't agree at all” (1) to “I completely agree” (4) | I have trouble resisting my cravings (e.g., for sweets). | 1 | .83 | 1.90 | 0.54 | 2338 |
| 2 | .82 | 1.92 | 0.55 | 2520 | ||||
| 3 | .83 | 1.89 | 0.57 | 1872 | ||||
| 4 | .83 | 1.90 | 0.55 | 1957 | ||||
| Co-rumination with friends (problem talk) | 4 | “Not at all true” (1) to “Really true” (5) | When I have a problem, me and my friend talk a long time about it. | 1 | .92 | 2.51 | 1.11 | 2313 |
| 2 | .91 | 2.79 | 1.18 | 2509 | ||||
| 3 | .93 | 3.01 | 1.15 | 1865 | ||||
| 4 | .92 | 2.86 | 1.13 | 1941 | ||||
- Note: Before stacking, the following were the sample sizes at each wave:
- a Scales for fathers used the same items and responses, changing “mother” to “father.”
2.2.1 Nonsuicidal self-injury
Self-injury was measured using an abbreviated version of the Deliberate Self-Harm Inventory (Lundh et al., 2007), with an added stem to ask for nonsuicidal injuries only (Tilton-Weaver et al., 2019). Adolescents responded to items about the extent to which they had inflicted nonlethal injuries on themselves 6 months before data collection, without suicidal intent.
2.2.2 Stressful interpersonal experiences at home, at school, and with peers
2.2.2.1 Stressful experiences
Four sources of stress were examined using the Adolescent Stress Questionnaire (ASQ, Byrne et al., 2007): family-related stress (arguments at home), romantic relationship stress (difficulties in these relationships) peer-related stress, (tapping pressure to fit in with peers), and school-related stress (tapping difficulties related to school performance).
2.2.2.2 Parent alienation
After the first wave, we assessed the degree to which adolescents felt alienated from their mothers or fathers (or stepparent/guardian; reporting on each parent separately), using items from the Inventory of Parent and Peer Attachment (IPPA; Armsden & Greenberg, 2009).
2.2.2.3 Peer victimization
The Short-Personal Experiences Checklist (PECK; Hunt et al., 2012) was used to measure peer victimization by relational and physical aggression. Participants indicated the number of times in the month before data collection that they have been relationally or physically victimized.
2.2.2.4 Negative experiences with parents
We examined four forms of parenting behavior associated with maladjustment. At time 1, we assessed coldness-rejection (aka love withdrawal) and angry outbursts. Coldness-rejection was assessed using 6 items from Persson et al. (2004) and angry outbursts used items from Tilton-Weaver et al. (2010).
After wave 1, parents’ dependency-oriented control was measured, using items from Soenens et al. (2010).
2.2.3 Intrapersonal characteristics and behaviors: Internalizing and regulatory difficulties
2.2.3.1 Internalizing
We assessed symptoms of depression and anxiety, as well as (low) self-esteem as a sign of internalizing. We used a version of the Center for Epidemiological Studies-Depression (CES-D; Radloff, 1977) scale that was revised for use with adolescents and translated into Swedish (CES-DC, Swedish; Olsson & von Knorring, 1997). This scale includes affective, cognitive, and somatic symptoms associated with depression. We used the OASIS scale (Overall Anxiety Severity and Impairment; Norman et al., 2006) to assess symptoms of anxiety and Rosenberg's (1965) measure for self-esteem.
2.2.3.2 Regulation difficulties
To represent difficulties regulating emotions and behavior, we assessed rumination, worry, difficulty controlling anger, impulsivity (urgency), and co-rumination. The tendency to ruminate was measured using the Children's Response Styles Scale (CRSS; Rood et al., 2010) and to worry using an adaptation of the Penn State Worry Questionnaire for Children (PSWQ-C; Chorpita et al., 1997). Difficulty managing anger was assessed with a scale developed for the project from which these data were drawn (Tilton-Weaver et al., 2019). We used the Urgency subscale of Whiteside and Lynam (2001) UPPS scale (Urgency, Premeditation, Perseverance, Sensation Seeking, Positive Urgency, Impulsive Behavior Scale) to assess adolescents’ tendencies to have difficulty refraining from acting on their impulses. Finally, the items focusing on negative events (problem talk) from Rose's (2002) scale were used to assess adolescents’ co-rumination with friends, a social form of ruminative-like behavior.
2.3 Sample
The first year of data collection, 3262 students were targeted for inclusion. Of these, 2769 students participated (77% response rate; 47.3% girls, 53.7% boys, Mage = 13.75 years, SD = 0.64). After T1, 2961 provided information at T2 (target n = 3352; 88%), 3002 provided information at T3 (target n = 4308; 70%), and 3258 at T4 (target n = 4049; 80%).
To maximize power, we retained students who had at least two waves of data. As analysis involved stacking by age, we examined the distribution of ages across waves. As there was insufficient samples of some ages (12-year-olds at T1; 18+ at waves 3, 4, and 5), we retained individuals who were between 13 and 17 during assessments. Thus, the analytical sample consisted of 3195 adolescents (51.7% boys, 48.3% girls). Of these, 31% participated in all five waves, 19% participated in at least four waves, 24% participated in at least three waves, and 26% participated in only two waves of data collection.
The analytic sample was composed mainly of adolescents born in Sweden (88.6%). Most had mothers born in Sweden (73.6%), with smaller numbers born in other Scandinavian countries (2.0%), in other European countries (6.9%) or outside of Europe (17.5%). Similarly, most of their fathers were Swedish (73.9%), from elsewhere in Scandinavia (2.5%), elsewhere in Europe (6.9%) or outside of Europe (17.6%). A large minority reported experiencing parent separation or divorce at least once during data collections (30.7%). Economically, measures tapping disposable income suggested that most lived in families with some disposable income, such as having more than one car (96.7%), annual holiday travel (95.4%), and having multiple computers at home (98.2%), although only 26% reported having their own bedroom.
2.4 Missing data and attrition analysis
The data were stacked, so that assessments were by age, rather than by wave. Before stacking, less than 5% of individual data were missing at any wave. Nonetheless, a Little's test indicated that the data were likely not missing at random (χ2 = 38,141.05, p < .001). Analyses of the individuals included in the study showed that adolescents in the sample who did not provide information at T1 were more likely to be foreign born (r = .06, p < .001), have less disposable income (r = −.08, p < .001), and were less likely to be living with both parents (estimated as percentage of waves living with both, r = −0.24, p < .001). Nonparticipation at T2 was related to having a Swedish mother or father (both r = −.07, p < .001), and living with both parents (r = −.23, p < .001). Nonparticipation at T3 was associated with having a mother or father of foreign origin (both r = .04, p = .02), having parents that had separated or divorced at some point in the study (r = .06, p < .001), and living in nonintact households (r = −.32, p < .001). At T4 and T5, nonparticipation was tied to being foreign born (rT4 = .11, rT5 = .12, p s < .001), having foreign-born parents (mothers: rT4 = .15, rT5 = .16, p s < .001; fathers: rT4 = .18, rT5 = .17, p s < .001), having less disposable income (rT4 = −.08, rT5 = −.09, p s < .001) and living in non-intact households (rT4 = −.23, rT5 = −.28, p s < .001).
Where self-injury was concerned, participation at T1 or T2 was unrelated to self-injury at any age (r s ranged from | .001| to | .04 | ). Participation at T3 was related to self-injury at ages 13 (r = .05, p = .02) and 14 (r = .07, p < .001). Participation at T4 was related to self-injury at ages 13 (r = .04, p = .04), 14 (r = .05, p = .01), 15 (r = .05, p = .02); and 17 (r = .07, p < .001). Finally, participation at T5 was related to self-injury at ages 13 (r = ,09, p < .001), 14 (r = .08, p < .001), 16 (r = .07, p < .001), and 17 (r = .08, p < .001). In all cases, significant correlations indicated that nonparticipation at that age was related to higher levels of self-injury. This may have contributed to some variance restriction in self-injury.
In terms of the characteristics of those that were not part of the sample (due to missing data at two or more waves), those who were excluded were more often foreign born (r = .17, p < .001), have foreign born mothers and fathers (rm = .11, rf = .15; for both, p < .001), experience a divorce during the course of the study (r = −.09, p < .001), live in non-intact households (r = −.35, p < .001), and have less disposable income (r = −.14, p < .001). Overall, this suggests that attrition was associated with being an immigrant and in a family with less social and economic stability.
2.5 Analytic plan
We analyzed our data in Mplus (v. 8.7, Muthén & Muthén, 1998), using established procedures for estimating trajectories and modeling mixtures of trajectories. Using the data stacked by age, we first conducted an overall growth curve analyses from ages 13 to 17, using MLR and FIML to estimate the intercept, linear, and quadratic slopes. These analyses were evaluated using the standard fit indices (nonsignificant χ2; CFI and TLI > 0.90; RMSEA and SRMR < 0.05).
For the growth curve mixture models, we used the procedures outlined by Masyn (2013), modeling two to five trajectories and examining changes in the overall fit (AIC, aBIC), classification (entropy, average posterior probabilities or aPP), as well as values of the LMR-aLRT tests and replication of the model log-likelihood (to ensure the results were not local maxima).
To examine correlates for the resulting trajectories, we used the BCH procedures recommended by Asparouhov and Muthén (2014). We examined correlates at the starting point of the trajectories (age 13) and at a later age, chosen after identifying the trajectories. As there are multiple tests, we compensated for increase error rates by setting the α level for the BCH omnibus tests at .01.
3 RESULTS
3.1 Preliminary analyses
We examined the descriptive statistics for the model and predictor variables (provided in Table 1). Most were skewed, with the bulk indicating positive experiences and adjustment, as would be expected in a community sample. Physical aggression was very platykurtic, with little variation in values. We removed this variable from analyses, as Type II error was likely. As expected, (and can be seen in the supplementary Tables), self-injury had some rank-order stability over time, as indicated by zero-order correlations ranging from 0.48 to 0.56 at adjacent waves. Also as expected, most of our predictors were related to self-injury. Co-rumination was the obvious exception, weakly related between ages 13 and 15, but nonsignificant later.
3.2 Growth curve analyses
We first fit a basic latent growth curve model to the data, to examine the shape and variances for a single trajectory. This model fit was very good (χ2 [6] = 5.88, p = .44; CFI, TLI = 1.00; RMSEA = 0.00; SRMR = 0.02) and suggested that the general trajectory started low (M = 0.13, SE = 0.01, p < .001), with a slight increase (M = 0.03, SE = 0.01, p = .002) that had a small, but significant curve (b = −.01, SE = 0.00, p = .01). More importantly, there was sufficient variance in the slopes to indicate that there was sample heterogeneity that should be examined (variances ranged from 0.01 to 0.16, all p < .001).
3.3 Latent growth mixture models
Next, we fit latent growth mixture models, starting with two trajectories and modeling up to five. As can be seen in Table 2, the significant LMR-aLRT for two trajectories suggests that there may be more classes needed to fit the data, whereas the nonsignificant LMR-aLRT for three, four, and five trajectories suggested that three trajectories were sufficient. Although two of the classes were smaller than what is recommended by standard rules-of-thumb, the entropy and posterior probabilities indicated that these classifications were still very good. Moreover, we expected small trajectories, as higher rates of self-injury are infrequent in community samples. Thus, we retained the three-class solution for further analysis.
| # Trajectories | LL | AIC | aBIC | Entropy | aPP | LMR-aLRT | % smallest |
|---|---|---|---|---|---|---|---|
| 2 | 4770.34 | 9.564.68 | 9599.39 | .994 | .990 | 4997.23** | 4 |
| 3 | 3913.07 | 7858.13 | 7904.40 | .985 | .970 | 1663.03 | 3 |
| 4 | 3152.07 | 6344.14 | 6401.98 | .983 | .953 | 1476.25 | 2 |
| 5 | 2581.10 | 5210.20 | 5279.61 | .985 | .958 | 1042.36 | 1 |
The largest class was composed of adolescents whose self-injury was low and relatively stable throughout the ages sampled (stable-low trajectory) (see Table 3, Figure 1). As expected, this was the largest class, with 94% of the sample having posterior probabilities that placed them in this class. A second class, composed of 3% of the sample, started out relatively low, but rose exponentially in self-injury with no evident plateau (low-increasing trajectory; similar to the trajectory found by Wang and colleagues, 2016). This trajectory is consistent with what we speculated might emerge. A third small class, again composed of 3% of the sample, had the inverted U pattern seen in other studies, increasing from ages 13 to 15, then dropping (increase-decrease trajectory). We did not find the two trajectories that would have been similar to Wang and colleagues (2016), with either substantial fluctuations or low and rising levels.
| Trajectory | Intercept | Linear slope | Quadratic slope | ||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Est. | SE | t | p | r2 | Est. | SE | t | p | r2 | Est. | SE | t | p | r2 | |
| Stable-low (94%) | 0.08 | .01 | 10.41 | <.001 | .99 | −0.02 | .01 | −1.87 | .06 | .78 | 0.00 | .00 | 0.95 | .34 | .47 |
| Low-increasing (3%) | 0.34 | .14 | 2.53 | .01 | .87 | 0.03 | .19 | 0.18 | .85 | .03 | 0.12 | .05 | 2.33 | .02 | .84 |
| Increase-decrease (3%) | 1.22 | .23 | 5.25 | <.0001 | .97 | 1.56 | .24 | 6.62 | <.001 | .98 | −0.42 | .05 | −8.17 | <.001 | .99 |
3.4 Trajectory differences in stressful interpersonal experiences, Age 13
We next examined trajectory differences at Age 13, in correlates representing interpersonal experiences. These are reported in the top half of Table 4.
| Trajectories | |||||||||
|---|---|---|---|---|---|---|---|---|---|
| Stable-low | Low-increasing | Increase-decrease | Omnibus Wald χ2 | ||||||
| Variable at age 13 | M | SE | M | SE | M | SE | p | ω2 | |
| Interpersonal experiences | |||||||||
| Stressors | |||||||||
| Family-related stress | 1.71 | .02 | 2.07 | .11 | 2.61 | .17 | 38.02 | <.001 | .13 |
| Romantic relationship stress | 1.28 | .01 | 1.54a | .12 | 1.92a | .17 | 19.20 | <.001 | .10 |
| Peer-related stress | 1.33 | .01 | 1.72a | .12 | 1.98a | .15 | 29.82 | <.001 | .12 |
| School-related stress | 2.44 | .03 | 2.86a | .18 | 3.21a | .25 | 14.58 | .001 | .11 |
| Peer victimization | |||||||||
| Relational aggression | 1.39 | .01 | 1.62 | .10 | 2.06 | .15 | 25.63 | <.001 | .11 |
| Negative experiences with parents | |||||||||
| Coldness-rejection | |||||||||
| Mother | 1.28a | .01 | 1.34a | .05 | 1.48a | .08 | 8.22 | .02 | .06 |
| Father | 1.28a | .01 | 1.29a | .05 | 1.55a | .10 | 8.30 | .02 | .06 |
| Angry outbursts | |||||||||
| Mother | 1.41a | .01 | 1.54ab | .08 | 1.65b | .09 | 9.52 | .009 | .07 |
| Father | 1.43a | .01 | 1.48a | .07 | 1.75 | .10 | 10.04 | .007 | .07 |
| Intrapersonal characteristics | |||||||||
| Gender | Boys (54%) | Girls (81%)a | Girls (72%)a | 94.76 | <.001 | .17 | |||
| Internalizing | |||||||||
| Depressive symptoms | 1.79 | .02 | 2.32 | .12 | 2.80 | .16 | 62.85 | <.001 | .17 |
| Anxiety symptoms | 1.55 | .02 | 2.06 | .12 | 2.45 | .15 | 53.95 | <.001 | .16 |
| Self-esteem | 3.14 | .01 | 2.71 | .09 | 2.28 | .11 | 72.62 | <.001 | .19 |
| Regulation issues | |||||||||
| Rumination | 2.43 | .02 | 2.90a | .12 | 3.09a | .15 | 34.78 | <.001 | .13 |
| Worry | 1.88 | .01 | 2.14a | .07 | 2.33a | .10 | 33.19 | <.001 | .13 |
| Difficulty controlling anger | 1.85 | .01 | 2.05a | .09 | 2.20a | .13 | 10.63 | .005 | .07 |
| Impulsivity-urgency | 1.88 | .01 | 2.21a | .08 | 2.42a | .10 | 45.83 | <.001 | .15 |
| Co-rumination with friends | 2.51a | .03 | 2.59a | .21 | 3.11a | .22 | 7.57 | .02 | .06 |
- Note: Trajectories with shared subscripts do not significantly differ. Parent behaviors and self-esteem were measured only at T1.
Where stressors were concerned, adolescents in the stable-low class reported the lowest levels of stress at age 13. Two patterns emerged for the other trajectories. For family-related stressors, levels were higher for the low-increasing class, and highest for the increase-decrease class. For romantic relationships, peers, and school stress, the levels were higher than the stable-low class, but similar for adolescents in both low-increasing and increase-decrease classes (Table 5).
| Variable at age 16 | Trajectories | ||||||||
|---|---|---|---|---|---|---|---|---|---|
| Stable-low | Low-increasing | Increase-decrease | Omnibus Wald X2 | ||||||
| M | SE | M | SE | M | SE | p | ω2 | ||
| Interpersonal experiences | |||||||||
| Stressors | |||||||||
| Family-related stress | 1.78 | .02 | 2.45 | .18 | 3.03 | .17 | 68.20 | <.001 | .19 |
| Romantic relationship stress | 1.38 | .02 | 1.99a | .17 | 2.14a | .21 | 26.10 | <.001 | .12 |
| Peer-related stress | 1.33 | .01 | 1.92a | .13 | 2.09a | .14 | 49.81 | <.001 | .16 |
| School-related stress | 2.85 | .03 | 3.65a | .19 | 3.96a | .18 | 56.42 | <.001 | .17 |
| Victimization | |||||||||
| Relational aggression | 1.33 | .01 | 1.78a | .11 | 1.76a | .13 | 27.76 | <.001 | .12 |
| Negative experiences with parents | |||||||||
| Alienation from parent | |||||||||
| Mother | 1.80 | .02 | 2.28a | .16 | 2.49a | .16 | 27.59 | <.001 | .12 |
| Father | 1.95 | .02 | 2.51a | .16 | 2.90a | .16 | 47.33 | <.001 | .16 |
| Dependency-oriented control | |||||||||
| Mother | 1.85 | .02 | 2.17a | .16 | 2.32a | .16 | 13.18 | .001 | .08 |
| Father | 1.65a | .02 | 1.75ab | .13 | 2.13b | .15 | 10.11 | .006 | .07 |
| Intrapersonal characteristics | |||||||||
| Internalizing | |||||||||
| Depressive symptoms | 2.16 | .02 | 3.29 | .15 | 3.71 | .13 | 199.89 | <.001 | .32 |
| Anxiety symptoms | 1.75 | .02 | 2.70 | .16 | 3.25 | .17 | 117.46 | <.001 | .25 |
| Regulation issues | |||||||||
| Rumination | 2.90 | .02 | 3.84a | .14 | 3.78a | .13 | 86.73 | <.001 | .21 |
| Worry | 2.07 | .01 | 2.67a | .10 | 2.71a | .10 | 79.18 | <.001 | .20 |
| Difficulty controlling anger | 1.81 | .02 | 2.45a | .13 | 2.73a | .11 | 90.74 | <.001 | .22 |
| Impulsivity-urgency | 1.87 | .01 | 2.60a | .10 | 2.80a | .09 | 161.58 | <.001 | .29 |
| Co-rumination with friends | 2.51a | .03 | 2.59a | .21 | 3.11a | .22 | 7.57 | .02 | .04 |
- Note: Trajectories with shared subscripts do not significantly differ. Alienation and parent control were not measured at T1.
Regarding victimization, we found significant differences in being the victim of relational aggression. These were similar to the adjustment differences, with adolescents in the stable-low class reporting the least amount of victimization, more victimization reported by those in the low-increasing class, and the highest levels for the increase-decrease class.
Our analysis of negative parenting behaviors showed no significant differences for either mothers’ or fathers’ coldness-rejection. However, differences were found for angry outbursts, with the patterns differing slightly across parents. For mothers, the lowest levels were reported by adolescents in the stable-low class and the highest with the increase-decrease class. Adolescents in the low-increasing class fell in the middle, not significantly differing from either of the other two classes. For fathers’ anger, the lowest levels were reported by adolescents in both the stable-low and low-increasing classes, with significantly higher levels reported by those in the increase-decrease class.
To summarize, the stable-low class consistently reported the lowest levels of negative or challenging interpersonal experiences. In many ways, the other two classes did not differ significantly; this was the case for stressors with romantic partners, peers, and in school. These two classes with change differed only in terms of family-related stress, relational victimization, and fathers’ angry outbursts, with higher levels reported for the increase-decrease class.
3.5 Trajectory differences in intrapersonal characteristics and behaviors, age 13
These results are reported in the lower half of Table 4. These results showed that there were some gender differences across the trajectories, with girls over-represented in the two trajectories involving increases in self-harm.
Overall, the results suggest that the stable-low trajectory was associated with better psychological adjustment and regulatory abilities than the other two trajectories. Symptoms of depression and anxiety were lowest for the stable-low trajectory, followed by higher levels for the low-increasing trajectory and highest levels for the increase-decrease trajectory. The trajectories were similarly ranked in self-esteem, with the highest levels associated with the stable-low trajectory, and progressively lower levels for the low-increasing and increase-decrease trajectories, respectively.
Where emotional and behavioral regulation were concerned, similar patterns emerged for rumination, worry, difficulty controlling anger, and impulsivity-urgency: the lowest levels were seen in the stable-low trajectory, with higher, but similar levels for the other two trajectories. Significant differences were not found for co-rumination with friends.
In summary, the stable-low grouping was consistently the trajectory with values indicating the best adjustment. The low-increasing class differed from the increase-decrease class in that adolescents in the increase-decrease class reported more internalizing and lower self-esteem, more family-related stress, more frequent relational victimization, and more parental rejection and anger than the adolescents with the low-increasing trajectory.
3.6 Trajectory differences in stressful interpersonal experiences, age 16
We repeated the analysis of correlates at age 16. We chose this age point because it came after the peak for the increase-decrease trajectories, while levels were still rising for the low-increasing trajectories.
Where stressors were concerned, the patterns seen for age 13 were repeated at age 16. The low-stable class reported the lowest levels of stress, across all dimensions. Family-related stress differed significantly across all classes, with the highest reported by those in the increase-decrease class. The remaining types of stress (with romantic relationships, peers, and school) were reportedly higher, but similar for adolescents in the low-increasing and increase-decrease classes.
Being the victim of relational aggression showed significant differences. This pattern showed a shift from age 13. Adolescents in the stable-low class reported the least amount of victimization, whereas the adolescents in the other two classes reported higher, but similar levels.
At age 16, reports of feeling alienated also differed across trajectories, with the lowest levels of alienation reported by stable-low adolescents. Alienation from mothers and fathers was similarly higher in both the low-increasing and increase-decrease classes.
In terms of dependency-oriented control, there were differences across trajectories and across parents. Adolescents in the stable-low class reported having mothers whose control was lower than the other adolescents. The others reported higher, but similar levels of maternal control. By comparison, only the stable-low and low-increasing trajectories differed on fathers’ control, where adolescents in the stable-low class reported significantly less paternal control than did the adolescents in the increase-decrease classes.
In summary, some of the patterns were the same as seen at age 13. Only family-related stress differentiated between all three classes. As seen at age 13, the stable-low trajectory consistently reported the lowest levels of negative interpersonal experiences.
3.7 Trajectory differences in intrapersonal characteristics and behaviors, age 16
The analyses of internalizing symptoms revealed that, much like at age 13, the adolescents in the stable-low class reported the lowest levels of depression and anxiety, followed by adolescents in the low-increasing class reporting more symptoms and those in the increase-decrease class reporting the most symptoms.
Regarding regulation issues, rumination, worry, difficulty controlling anger, and impulsivity-urgency had similar patterns. Across these issues, the significant differences were between the adolescents with stable-low trajectories, whose values were the lowest, and the adolescents with either low-increasing or increase-decrease trajectories, whose levels were higher, but statistically similar to each other. As was the case at age 13, we found no significant differences in co-ruminating with friends at age 16.
In short, the patterns that were seen at age 13 were repeated at age 16, with only symptoms of depression and anxiety differentiating between all three trajectories. Again, the low-stable class reported the best functioning.
4 DISCUSSION
The first aim of our study was to examine variation in self-injury as it changes across adolescence. Like others, we found a trajectory of stably low levels of self-injury which represented a sizable portion of our sample (like Barrocas et al., 2015; and Wang et al., 2016), and a trajectory suggesting that a small proportion of the sample increases (to age 15) and then drops (like Barrocas et al., 2012). Divergent from others, we also found a small subset that starts at relatively low levels and increases through to at least age 17. This last trajectory, although small, warrants replication and closer attention, as it suggests that there are individuals at risk for continuing self-injury in late adolescence and potentially, early adulthood.
Our second aim was to explore potential explanations for the variation in development. Although there is currently no theoretical model explaining the maintenance or changes in self-injury across adolescence, we borrowed from models built on the functions of self-injury, dividing predictors of self-injury into interpersonal experiences that induce negative emotions and intrapersonal issues that make the arousal more difficult to manage (e.g., Selby & Joiner, 2009). From this perspective, increases in self-injury would be related to problematic interpersonal experiences and intrapersonal deficits, whereas drops in injury would be related to better experiences and functioning. This was the case—to a point. For example, the adolescents with stably low levels of self-injury tended to have the least problems interpersonally and intrapersonally. They were also more likely to be boys than girls. However, most of the issues we examined did not differentiate between adolescents whose trajectories increased without abating from those whose self-injury increased and then decreased. As might be expected, self-injury increasing from age 13 to 15 was associated with more interpersonal difficulties and more dysregulated behavior (except for co-rumination).
However, we expected that at age 16, these same variables would differentiate those who increase (low increasing) from those that were decreasing in self-injury (i.e., the increase-decrease trajectory). Most did not. At age 13, only family-related stress, relational victimization, symptoms of depression and anxiety, and self-esteem differentiated the trajectories involving change. As could be expected, the increase-decrease class (which had the highest levels of self-injury) had significantly more difficulties with respect to these issues than the low-increasing class. At age 16, the picture was the same, even though self-injury was decreasing for the increase-decrease class and continuing to rise in the low-increasing class. This raises some important questions, including what should differentiate trajectories of increasing self-injury from decreasing self-injury. As the inflection for the increase-decrease class was at age 15, one might think that this reflected improvements in regulation that tend to occur in from early to middle adolescence. This does not appear to be the case, as many of the interpersonal experiences we analyzed were equally problematic across these classes. Likewise, adolescents in these classes appeared to have equivalent levels of regulation. It is possible is that the experiences and regulation of adolescents in the increase-decrease trajectory are not improving and they have turned to other forms of coping, albeit equally ineffective. They may opt for strategies other than self-injury because of its disadvantages (e.g., visible physical scars). These other strategies may be the types of behaviors that tend to co-occur with self-injury, such as disordered eating (e.g., Cucchi et al., 2016; Sorgi et al., 2021), substance use (e.g., Bresin & Mekawi, 2020; Sorgi et al., 2021), or aggression towards others (e.g., Daukantaitė et al., 2019; Sorgi et al., 2021). These findings suggest the need to carefully consider other maladaptive regulatory and coping strategies in future research to explore why drops in self-injury do not seem to coincide with improvements in interpersonal experiences or gains in regulatory abilities.
It is also possible that other factors predict change in self-injury, such as the beliefs that Hasking et al. (2017) describe. According to their cognitive-emotional model, outcome expectancies and self-efficacy beliefs play a central role in when adolescents self-injure to regulate negative emotions. Accordingly, adolescents who believe self-injury will lead to negative outcomes, such as physical pain or social rejection, are unlikely to self-injure. Adolescents who believe self-injury would result in positive outcomes, such as decreased negative emotions, increased positive emotions, or help from loved ones, and believe that they have the ability to self-injure, they are more likely to view self-injury as a viable coping strategy, as long as these beliefs remain stable. Such beliefs, and changes in beliefs, could differentiate increasing and decreasing rates. Empirical findings about the link between self-injury-specific expectancies, self-efficacy, and frequency of self-injury seem to support such theoretical explanation (Hasking & Rose, 2016). As such beliefs are behaviorally specific, combining them with expectancies for other strategies (e.g., self-medication with substances) may also be enlightening.
Finally, we found no evidence that adolescents in these classes differed in co-rumination. We focused on items that tap repeatedly talking about problems, presumably a negative focus akin to rumination, but with others. It is possible that differences did not emerge because talking to friends about problems, even if negative in focus, allow friends to help them feel better, gain validation, problem solve, or simply feel supported. However, empirical evidence suggests that adolescents who co-ruminate with friends tend to increase in negativity and related symptoms, even if they experience improved relationships (Rose et al., 2014). As these are variable-centered findings, perhaps those who are suffering the most only experience the “bright side” of the co-rumination trade-off.
5 IMPLICATIONS FOR THEORY
The findings implicate theory in two ways. First, they draw attention to the need to frame self-injury as having multiple developmental trajectories during adolescence. The diversity in trajectories requires further exploration to understand the conditions that lead to either increase or abatement of self-injury. Studying such trajectories will require unique data sets, with a large sample that is followed over the course of adolescence. Although three patterns were detected in this study, we argue that rarer trajectories, such as those found by Wang et al. (2016) need to be considered and examined to fully understand the change in self-injury during adolescence.
Second, our results draw attention to the need to continue developing theoretical accounts that underscore both painful interpersonal experiences and intrapersonal vulnerabilities. Unclear from this investigation is why the two smaller trajectories change in velocity or direction over time. Theoretical attention is needed to fully understand why the changes occur and the extent to which the changes signal the potential for additional health risks.
Such theoretical frameworks could also draw attention to the need to consider when adolescents replace self-injury with other strategies, adaptive or maladaptive. Integrating these issues with adolescents’ reasons for self-harming could strengthen our understanding of how self-injury manifests during adolescence.
6 IMPLICATIONS FOR PRACTICE
Our results highlight the chronic nature of self-injury for some adolescents starting already at age 13. Maladjustment in different psychological, emotional, and social areas seems to explain the course of this development. Based on these findings, and on the fact that self-injury tends to emerge around age 12 (e.g., Cipriano et al., 2017), it is paramount to develop and introduce preventive interventions in late childhood and early adolescence. Such interventions should primarily target regulation strategies, as these represent a transdiagnostic factor that appear to affect individual problems (e.g., rumination, worry, anger dysregulation, and impulsivity).
Our study identified both interpersonal and intrapersonal vulnerabilities associated with trajectories of increasing self-injury. These findings suggest that clinicians should not only focus on current engagement in self-injury, but also on the risk factors that could help identify future risks for self-injury. By developing screening questions that address self-injury, as well as these inter- and intrapersonal risk factors, clinicians will have more tools for identifying those adolescents who might persist in this conduct. Given that self-injury is a strong risk factor for suicidal intention and behaviors (Griep & MacKinnon, 2020) and that the risk of death by suicide is highest during the first 6 months after a self-injury episode (e.g., Cooper et al., 2005), such screening tools have the potential to target and prevent suicidal attempts.
7 STRENGTHS, LIMITATIONS, AND FUTURE DIRECTIONS
Our study has some important strengths, including the use of a multi-item measure of self-injury and established measures of correlates, within a cohort-sequential designed that covered ages 13–17 years. Our sample was large enough to identify rather rare trajectories. Nonetheless, it would have been preferable to have longitudinal data that covered the emergence of self-injury for those whose levels were already rising, through to early adulthood, when self-injury frequently abates. Our measures were also all from the same adolescent reporters. Although the use of multiple reporters for psychosocial distress, cognitions, and perceptions may not be as useful as self-reports, it would have been optimal to have reports of parenting behavior from parents and perhaps even reports of peer experiences from peers or teachers. Our analysis of missing data and attrition also suggests some selection bias that may have contributed to the underestimation of self-injury. This may be particularly true for immigrant adolescents, who often have fewer resources than their Swedish peers.
These issues call for continuing our efforts, to see if our trajectories can be replicated. Further, it will be important to broaden the scope of interpersonal and intrapersonal issues, as well as behaviors, which could differentiate between rarer, but important trajectories. Determining if these trajectories are specific to Sweden, where internalizing rates among adolescents have risen in recent decades (Bremberg, 2015), would also be important. Such information would help further guide prevention and treatment efforts, by identifying what contributes to the types of trajectories our study revealed.
ACKNOWLEDGMENTS
This study was made possible by access to data from the Three Cities Study, a longitudinal research program at the department of Law, Psychology, and Social work at Örebro University, Sweden, and funded by the Swedish Research Council for Sustainable Development (FORMAS) to the Principal Investigator and steering committee of the 3 Cities Study (diary # 2012-10609-21848-41/2012-65) for data collection and management. Funding was also provided by a grant from the Swedish Research Council for Health, Working Life, and Welfare (FORTE) to the first author (diary #: 2021-00100).
CONFLICTS OF INTEREST
The authors declare no conflicts of interest.
ETHICS STATEMENT
Ethical approval was obtained by the PI of the 3 Cities Study prior to data collection and applies to the research in this paper (Uppsala Regional Ethics Board, dnr. 2013-384).
Open Research
DATA AVAILABILITY STATEMENT
Research data are not shared.
