A systematic review of daily life studies on non-suicidal self-injury based on the four-function model

Abstract

Non-suicidal self-injury (NSSI) is a prevalent, impairing, and trans-diagnostic behavior that can be comprehensively assessed in daily life studies. We conducted a systematic literature review of 35 Ambulatory Assessment and Daily Diary studies of NSSI, to achieve three aims. First, we reviewed descriptive evidence on NSSI acts. On average, studies observed 1.6 acts per participant, but evidence regarding methods, pain, and context was sparse. Second, we reviewed evidence on NSSI urges. On average, studies reported 4.3 urges per participant. Urges were also associated with increased negative affect and predicted later acts. Third, we reviewed evidence on the Four-function Model of NSSI. Eight studies partially supported negative intrapersonal reinforcement, showing increased negative affect pre NSSI, but, of these, only four studies supported decreased negative affect post NSSI. Additionally, only three studies supported positive intrapersonal reinforcement, showing decreased positive affect pre and increased positive affect post NSSI. Evidence for the interpersonal functions was limited to two studies and inconclusive. We recommend assessing the intensity, frequency, and context of acts and urges, as well as pain and urge duration in future studies. We also recommend follow-up prompts after acts and urges to better track affect trajectories, and a detailed assessment of interpersonal events.

Introduction

Non-suicidal self-injury (NSSI) is defined as deliberate, self-inflicted damage of body tissue without suicidal intent (e.g., Klonsky, 2011) and has a lifetime prevalence of approximately 5% in the general population (Swannell, Martin, Page, Hasking, & St John, 2014). Individuals with and without additional psychopathology engage in NSSI, but NSSI is more prevalent within clinical populations (e.g., Briere & Gil, 1998). NSSI appears to be largely trans-diagnostic, affecting individuals with a range of psychopathology, including anxiety and mood disorders, psychosis, eating disorders, and personality disorders (Bentley, Cassiello-Robbins, Vittorio, Sauer-Zavala, & Barlow, 2015; Nock, Joiner, Gordon, Lloyd-Richardson, & Prinstein, 2006). Importantly, NSSI not only correlates with suicidal behavior, but is also a specific risk factor for suicide attempts (Klonsky, May, & Glenn, 2013; Ribeiro et al., 2016; Victor & Klonsky, 2014). Beyond the personal burden, self-harm, including NSSI and that with suicidal intent, entails substantial health care and economic costs due to increased morbidity and mortality (Kinchin, Doran, Hall, & Meurk, 2017; Tsiachristas et al., 2017), which further underlines the need for research on why and how people self-injure.

In an effort to explain why people self-injure, multiple theoretical models of NSSI have been proposed over the past two decades. One of the earliest models is the Experiential Avoidance Model (Chapman, Gratz, & Brown, 2006), which proposes that NSSI is primarily performed to avoid aversive emotional experiences (i.e., negative reinforcement). Extending this, the Emotional Cascade Model (Selby, Franklin, Carson-Wong, & Rizvi, 2013) proposes that individuals engage in NSSI to distract themselves from positive feedback loops of negative affect (NA) and rumination (i.e., emotional cascades). Thus, this model incorporates negative thoughts in addition to NA, but still posits negative reinforcement as a central component. In addition to negative reinforcement, the Four-function Model (Nock, 2009; Nock & Prinstein, 2004) includes a positive reinforcement component, suggesting NSSI can also serve to induce positive states. It further distinguishes between interpersonal and intrapersonal functions. Some factor-analytic work suggests that individuals who engage in NSSI may more clearly distinguish between intra- and interpersonal functions than between positive and negative reinforcement functions (Klonsky, Glenn, Styer, Olino, & Washburn, 2015). This may suggest that a simpler two-factor model may be sufficient for understanding why individuals self-injure. However, there are theoretical grounds for distinguishing positive and negative reinforcement (Nock & Prinstein, 2004) and, although individuals may not as readily distinguish between positive and negative reinforcement, they may still experience effects from NSSI that can be meaningfully categorized along this dimension.

Extending this previous work, two recent models have additionally focused on understanding why individuals first decide to engage in NSSI. Both the Cognitive-Emotional Model of NSSI (Hasking, Whitlock, Voon, & Rose, 2017) and the Benefits and Barriers Model (Hooley & Franklin, 2018) propose that positive and negative reinforcement (i.e. the ‘benefits’ of NSSI) primarily explain the maintenance of NSSI, but not why individuals first decide to engage in this self-destructive behavior. To fill this gap, the Cognitive-Emotional Model describes how cognitions, specifically cognitive representations of NSSI, expectations about the outcome of NSSI, and self-related cognitions such as self-efficacy expectations predict who initiates NSSI. Similarly, Hooley and Franklin suggest in their model that most people have innate barriers that keep them from engaging in NSSI, such as an innate aversion to pain, and that these barriers have to be lowered before someone engages in NSSI for the first time. While the attention to barriers of NSSI is likely important, the majority of existing empirical studies have focused on testing the ‘benefits’ of NSSI, which are most comprehensively summarized in the Four-function Model of NSSI (Nock, 2009). Thus, we chose this model as the focus for the present review, while acknowledging that several components are also integrated within other models and that barriers are not represented.

The Four-function Model (Nock, 2009) proposes that NSSI serves four functions that can be either intrapersonal or interpersonal, and that are either negatively or positively reinforcing. Intrapersonal negative reinforcement suggests that NSSI serves to alleviate aversive intrapersonal states, such as NA. The counterpart, intrapersonal positive reinforcement, suggests that NSSI serves to generate positive internal states, for instance a sense of euphoria or thrill. The two interpersonal functions comprise NSSI that is performed to influence the behavior of others, or the individual’s relationship with others. Interpersonal negative reinforcement comprises NSSI with the intent to reduce undesired behavior from another person or undesired interactions with them, such as ending a conflict or avoiding being confronted for a mistake. Interpersonal positive reinforcement, in contrast, suggests that individuals engage in NSSI to elicit positive behavior from others or positive interactions with them, such as gaining attention or comfort. Past studies have extensively tested the model in a cross-sectional framework and this evidence has been summarized in several reviews and meta-analyses (e.g., Bentley, Nock, & Barlow, 2014). Adding to this, an increasing number of recent studies have tested components of the model in intensive longitudinal designs, which overcome several of the limitations of cross-sectional work. Below, we summarize advantages of intensive longitudinal studies and argue why these are ideally suited to test models of NSSI.

Intensive longitudinal designs comprise an assessment of NSSI in the context of daily life and in near real-time via methods such as Ambulatory Assessment (AA, Trull & Ebner-Priemer, 2013). In AA, participants report on constructs of interest (e.g., NSSI events, affect) via smartphone several times a day over a period of days or weeks. Daily diary (DD) studies are a specific type of AA wherein participants report on their experiences once daily. Using AA and DD to capture NSSI overcomes a number of limitations associated with laboratory or cross-sectional self-report methods. First, most laboratory studies have used experimenter- and not self-administered pain stimuli as proxies for NSSI. The few studies that have used self-administered pain stimuli employed electric shock or cold stimuli, which do not closely match the methods typically used by those who self-harm (see Ammerman, Berman, & McCloskey, 2018 for a review of laboratory studies on NSSI). AA and DD studies overcome this problem by capturing NSSI as it occurs in real life. Second, the ability of daily life studies to reduce memory biases is particularly relevant in the case of NSSI, because memory biases are intensified when participants report on highly emotional events (e.g., Blaney, 1986), and this likely reduces the accuracy of cross-sectional self-report of past NSSI episodes. Third, theoretical models of NSSI make relatively fine-grained assumptions about the effects NSSI has over time and these can be ideally tested in an AA framework. For instance, most models propose that some individuals use NSSI to alleviate NA. To test this, NA must be assessed before, during, and after NSSI, as well as at comparison occasions without NSSI. In conclusion, AA and DD are uniquely suited for testing theoretical models of NSSI and we therefore summarize the available AA and DD evidence in the present up-to-date and systematic review.

The present review

We note that some of the articles we review were also included in two previous reviews of NSSI in daily life (Hamza & Willoughby, 2015; Rodríguez-Blanco, Carballo, & Baca-Garcia, 2018). However, the aims and scope of the current review differ substantially from previous ones, thus giving cause for this update and extension. The review by Hamza and Willoughby included 18 studies published until January 2014 and assessed the link between NSSI and emotion regulation. The review included both AA and laboratory studies and comprised only seven studies that specifically addressed NSSI in a daily life context. The review also did not specifically examine evidence for the Four-function Model or for NSSI urges. The second review by Rodríguez-Blanco et al. (2018) included 23 AA studies published until September 2017, addressing the daily life context of NSSI and changes in constructs from pre to post NSSI. The reviewed studies partly overlap with the current review and the authors describe findings on NSSI and affect or interpersonal variables, but these were not systematically evaluated with regard to a theoretical model of NSSI. Moreover, the review did not provide detailed descriptive data to characterize the phenomenology of NSSI acts and urges and their association, nor provide specific recommendations for future AA studies. The present review had three specific aims. First, we provide a descriptive summary of NSSI acts in daily life. We consider the descriptive stage an important part of studying a phenomenon and believe that much can be learned about NSSI from carefully considering aspects of the behavior itself (e.g. severity, type) and its daily life context (e.g. when, where, and with whom NSSI acts occur).

Second, in addition to NSSI acts, we review evidence on NSSI urges. NSSI urges are highly prevalent in individuals with NSSI history (Kamphuis, Ruyling, & Reijntjes, 2007) and their importance is emphasized in the DSM-5 criteria for NSSI disorder, which require “prior to engaging in the act, a period of preoccupation with [NSSI] that is difficult to control” and “thinking about self-injury that occurs frequently, even when it is not acted upon” (American Psychiatric Association, 2013, p. 803). This suggests that urges may have negative outcomes beyond those related to the actual act, for instance binding cognitive or self-regulatory resources. Moreover, the criteria directly imply that urges occur ‘prior to engaging in the act’, which suggests a temporal precedence that can be ideally tested in a daily life context. Urges, in this way, may provide a valuable target for interventions. Moreover, in contrast to NSSI acts, urges are relatively unconstrained by environmental factors, in that an individual may experience an urge in situations where engaging in NSSI may not be possible (e.g. in the workplace). Thus, studies examining NSSI urges may reveal associations with relevant contextual factors that may be obscured in the case of acts.

Third, we aimed to provide a systematic review of daily life evidence regarding the Four-function Model of NSSI (Nock, 2009; Nock & Prinstein, 2004). To this end, we summarize previous evidence for the intrapersonal and interpersonal functions and discuss the evidence for each.

Search strategy and literature extraction

In conducting this review, we followed the PRISMA guidelines (Tricco et al., 2018). We included only individual, empirical studies and employed the following eligibility criteria: a) measurement of NSSI acts or urges in the moment or during the day (using AA or DD methodologies), b) written in English language, c) published in a peer reviewed journal, d) accessible through PubMed or PsychInfo, e) published prior to December 5, 2019. We excluded experimental, interventional, and clinical trials and samples comprising individuals with psychosis or developmental disorders.

The search terms used to extract articles are presented in Table A1. We extracted 78 articles through PubMed and 66 articles through PsychInfo on December 5, 2019. After removal of duplicates, articles were screened, first by title and then by abstract content. After this screening procedure, two raters analyzed 38 full-text articles in detail for final inclusion. In case of a disagreement, a third rater read the article and inclusion or exclusion was decided by consensus. After this extraction process, 32 full-text articles were included. We then conducted a ‘cited by’ search, which revealed three additional articles. The flowchart depicted in Figure A.1 illustrates this process.

The final selection included 35 articles published between 2009 and 2019. Ten studies used DD and 26 studies used AA methods (Kleiman, Turner, Chapman, & Nock, 2018 reported both a DD and an AA study). Overall, 11 studies assessed only NSSI acts, two studies assessed only NSSI urges, and 22 studies assessed both. Although each study addressed different research questions, several studies used the same sample or a subsample of another study. Taken together, 24 independent samples are reviewed in this article and we indicate sample overlap in the tables. When we report statistics such as average study duration in the following sections, these were based on samples, not studies. In the following sections, we review evidence for the three specific aims and provide short summary sections with main findings and conclusions after each section.

Study designs and sample characteristics

Below, we summarize the reviewed studies with regard to the overall study period, sample size, the different time-points at which data was collected, compliance rates, and demographic structure. Table A.2 provides a detailed overview of the study designs.

Sample size and sampling frequency

The total number of participants included in all reviewed studies combined was 1,727. Sample sizes ranged from 21 to 255 with an average sample size of M = 71.9 participants (SD = 54.1), and a median sample size of Md = 52.5. The duration of AA and DD studies was similar, with both types of studies ranging between 5 and 21 days; on average 13.4 days in AA studies (SD = 5.9), and 17.5 days in DD studies (SD = 5.9). The median study duration was 14.0 days in both cases.

In DD studies, participants typically provided only one diary report at the end of the day, whereas AA studies differed in both the number of assessments, as well as the type of prompts included. Seventeen out of 18 AA samples included prompts that were presented at random time-points throughout the day, or semi-random prompts that were presented within pre-specified timeframes. The number of random prompts within a day ranged between four and 10 prompts with an average of M = 6.14 random prompts per day (SD = 1.75). Other than random prompts, seven out of 18 samples used fixed prompts that were presented at specific times throughout the day (e.g. morning reports directly after wake-up) and seven samples included event-contingent prompts that participants self-initiated to provide reports right after an NSSI act or urge.

The total number of included study days (across all participants) in the reviewed DD studies was an average 623.6 days (SD = 74.6, Md = 613 days). The total number of prompts that was included in the reviewed AA studies varied significantly depending on sample size and sampling scheme, with longer study periods and more frequent within-day sampling leading to a higher number of included prompts. Only 10 of 18 unique AA samples reported the total number of included prompts (for details, see Table A.2). For the studies that reported this number, included prompts ranged from 428 to 11,172 prompts with an average completed 3,434.4 prompts (Md = 2,429.5, SD = 3,431.5).

Compliance

Compliance rates in DD and AA studies typically indicate the percentage of prompts or days that participants completed on average. Five out of the six unique DD samples provided average compliance rates, resulting in an overall compliance across the reviewed studies of 78.9% of days (SD = 6.9). Horowitz and Stermac (2018) did not provide an average compliance value but stated that ‘nearly 70% of participants completed all 21 days’. In the reviewed AA studies, average compliance rates for random prompts were reported for only 10 out of 18 samples. Compliance rates ranged from 30% to 91.1%, with an average compliance across studies of 64.9% of prompts (SD = 18.2, Md = 70.0). Instead of reporting overall compliance rates, Kranzler et al. (2018) reported that 40 out of the 47 participants in their sample completed at least 80% of prompts. Similarly, Nock, Prinstein, and Sterba (2010) reported that 25 out of 30 participants in their sample had a 100% compliance rate. Compliance rates were not reported for the remaining six samples (see Table A.3). Compliance was lower than that observed in a review of suicidal thoughts and behaviors (71.1%; Kleiman & Nock, 2018) and severe mental disorders (78.7%; Vachon, Viechtbauer, Rintala, & Myin-Germeys, 2019). However, removing the two samples with the lowest compliance rates (Ammerman, Olino, Coccaro, & McCloskey, 2017; Armey, Crowther, & Miller, 2011), which were outliers with substantially lower compliance than the remaining samples, compliance rose to 75.2%.

Most studies did not exclude participants before calculating compliance, with several exceptions. Five studies reported that they excluded participants who did not participate in the AA or DD part of the study (i.e., never completed an AA assessment). Four other studies reported that they excluded participants who provided less than a certain number of assessments: in two studies, participants were excluded if they did not respond to at least 20% of assessments, in one study if they did not respond to at least 2% of assessments, and one study excluded participants with a compliance that was more than 2SD below the sample mean. Compliance in these studies was adequate to good and did not substantially differ from the average compliance of approximately 65% that we computed for the reviewed AA studies overall. Thus, compliance thresholds are unlikely to have substantially inflated the reported compliance rates. For details on compliance thresholds, see Table A.3.

Beyond the impact of compliance thresholds, the reported compliance rates could also have been impacted by how participation was incentivized and whether participants could earn more money at higher levels of compliance. Information on this was available for 17 of the 22 unique samples we reviewed. One sample offered no compensation and 4 samples (23.5%) included a flat fee or course credit as compensation. The remaining 11 samples incorporated some form of compliance contingent compensation. Specifically, 5 (29.4%) studies compensated participants per completed entry and 6 (35.3%) studies compensated participants with a flat fee and an additional bonus payment if they achieved a high compliance (usually >80%). Thus, around two thirds of studies specifically incentivized compliance. To assess whether this resulted in higher compliance rates, we computed a point-biserial correlation between incentivisation scheme (compliance dependent vs. not) and compliance rate. The correlation was large, though non-significant due to the small number of studies, with r = 0.50, df = 12, p = .071. Though we had minimal power to detect an effect, this indicates at least a trend that studies that incentivized compliance tended to observe higher compliance rates.

Gender

The majority of studies comprised primarily women, with the exception of Humber, Emsley, Pratt, and Tarrier (2013), who assessed male inmates. Three studies recruited only women (Anestis et al., 2012; Muehlenkamp et al., 2009; Pearson et al., 2016). Excluding these four studies (where participants of only one gender were recruited), the remaining samples comprised an average of 76.2% women (SD =18.98, Md = 83.5%). This does not well reflect prevalence rates for NSSI, which are about equal for men and women in the general population and only slightly higher for women in clinical samples (for meta-analyses, see Bresin & Schoenleber, 2015; Swannell et al., 2014). However, some argue that the higher prevalence estimates for women in clinical samples that were observed by Bresin and Schoenleber may be attributable to the fact that men tend to use forms of self-harm that are less frequently assessed in NSSI studies, such as object or wall-punching (Kimbrel, Calhoun, & Beckham, 2017). Indeed, object and wall punching was not included as a form of NSSI in the meta-analysis by Bresin and Schoenleber. Moreover, Kimbrel and colleagues argue that women are over-represented in clinical samples because they are more likely to seek out mental health treatment and this could further inflate observed gender differences in NSSI. The same issue may have resulted in over-sampling of women in the reviewed studies, 68.18% of which recruited at least a proportion of their participants through treatment centers. However, this still would not fully account for the high prevalence of women in the reviewed studies, as Bresin and Schoenleber (2015) only found a slightly higher prevalence of women in clinical samples. Ultimately, the existing AA work on NSSI pertains primarily to women with NSSI and it is unclear how well these findings translate to men who self-harm.

Beyond the over-sampling of women, only few studies reported whether they assessed gender in a non-binary manner. Only Hughes et al. (2019) reported that 2% of their participants identified as transgender. As recent findings show lifetime prevalences of NSSI for transgender individuals from the community (i.e. not recruited through clinical settings) of more than 50% (Jackman, Dolezal, Levin, Honig, & Bockting, 2018), the under-sampling of these individuals appears especially problematic (see also Fox, Millner, Mukerji, & Nock, 2018, for a discussion). Due to the lack of details on how gender was assessed, it is unknown whether samples included only participants that identified within the categories of binary gender, or whether non-binary individuals lacked the option to choose a different category.

Race and ethnicity

The reviewed samples varied considerably with regard to the amount of detail they reported on participant race or ethnicity. Five out of 24 samples included no details regarding participant race and ethnicity. The only category that was consistently reported across the remaining 19 samples was a White or Caucasian category, which comprised an average of 66.6% of participants across studies (SD = 23.7, Md = 64.0). In eight samples, this group comprised more than 80% of participants. Another 10 samples included a category for African, African American, African Canadian, or Black. Across these 10 samples, an average of 22.7% of participants (SD = 21.6, Md = 19.0) endorsed this category. Participants could select a category for Asian, East Asian, South Asian, or Asian American in 10 of the reviewed samples. Across these 10 samples, this group comprised an average of 17.1% participants (SD = 17.0, Md = 10.3). Participants in eight samples indicated whether they identified as Latinx or Hispanic. On average, 7.8% of participants in these samples endorsed this category (SD = 6.5, Md = 6.3). Other categories that were assessed in individual studies included Native American, Native Canadian, Middle Eastern, Pacific Islander, and Multiracial (see Table A.2).

As described above, information on race or ethnicity was lacking for five out of 24 samples and the majority of studies included a high number of White or Caucasian individuals. It is of considerable importance to provide full information on race and ethnicity and to discuss the associated limitations with regard to generalizability. This appears especially important in light of recent evidence suggesting that race and ethnicity are important moderators for understanding differences within and between studies on NSSI (see Gholamrezaei, De Stefano, & Heath, 2017, for a review).

Age

Five of the reviewed 24 samples included participants under 18 years old, whereas the remaining 13 samples comprised only adult participants. The average age in samples that included underage participants was 17.4 years (SD = 1.3, Md = 17.3), and individual age ranged from 12 to 25. The average age in adult samples was 26.2 years (SD = 6.7, Md = 25.0) and individual age ranged from 18 to 58. Thus, the adult samples included mostly young adults, which parallels evidence on a prevalence peak for NSSI in adolescence and early adulthood (Muehlenkamp, Claes, Havertape, & Plener, 2012; Swannell et al., 2014). However, NSSI is still substantially prevalent in the adult population and a recent review suggests that even adults 60 years of age and older are considerably affected by it (Troya et al., 2019). Thus, the reviewed studies under-represent middle-aged and older adults.

Mental health status

Nine out of the reviewed 24 unique samples recruited participants with a specific type of psychopathology. Six samples comprised participants with a formal borderline personality disorder (BPD) diagnosis and three samples comprised individuals with bulimia nervosa. Participants in the remaining 15 samples were selected for having a history of NSSI. However, this does not mean that participants could not also meet criteria for mental disorders. Eight of the 15 samples that did not explicitly recruit participants with a formal diagnosis conducted clinical interviews and reported diagnoses for their participants (diagnostic information was unavailable for participants in the remaining seven samples). Pooling these with the above described nine studies that recruited participants with BPD and bulimia nervosa allowed us to review the distribution of diagnostic categories across studies.

BPD was assessed in 10 samples and on average 61.5% (SD = 35.1) of participants met criteria for BPD in these samples (28.0% in the studies that did not specifically recruit BPD). Mood disorders (predominantly major depressive disorders) were assessed in 12 samples and affected on average 53.5% (SD = 23.1) of the sample. Anxiety disorders were assessed in 10 samples and affected on average 47.5% (SD = 20.8) of the sample, though the type of anxiety disorder assessed and the prevalence of specific anxiety disorders differed substantially. Similarly, eating disorders were also assessed in 10 samples and affected 48.9% (SD = 47.9) of participants in these samples. Lastly, substance use disorder diagnoses were established in eight samples and showed a prevalence of 17.5% (SD = 12.1), on average. Thus, while the majority of included samples did not recruit participants with a specific diagnosis, most samples comprised participants with substantial amounts of psychopathology beyond NSSI. We summarize the most prevalent diagnostic categories in Table A.2.

The inclusion of individuals with psychopathology in some but not in other samples is also reflected by the studies’ different recruitment strategies. Overall, 27.27% of studies recruited their participants only in clinical settings such as outpatient clinics, 36.36% of studies recruited their participants only from the community, and 40.91% recruited participants from both clinical and community settings (see Table A.3 for details). Thus, more than two thirds of studies recruited at least some proportion of their participants in clinical settings, which likely accounts for the high prevalence of psychopathology in the reviewed samples.

NSSI history

In addition to the nine samples that recruited individuals with a formal diagnosis that incurs risk for NSSI (i.e., BPD, bulimia nervosa), the remaining samples recruited individuals based on a history of NSSI. One set of samples included individuals with recent NSSI thoughts or behaviors. Specifically, three samples included participants with NSSI in the past two weeks, and four samples included participants with NSSI urges in the past two weeks or last month. Of the latter four samples, two also required lifetime history of NSSI or past year acts. A second set of samples required NSSI thoughts or behaviors in the past year (three samples), and a third set recruited participants based on a lifetime history of NSSI (two samples, see Table A.2 for details).

Summary of study designs and sample characteristics

The present daily life evidence on NSSI is primarily based on the study of White women, who are in their early twenties, and who have some form of current psychopathology, which points to a clear need for replication in more diverse samples. As epidemiological studies suggest that NSSI is about equally prevalent in women and men (Swannell et al., 2014) and particularly frequent in minority populations, including people of color and the LGBTQ+ community (Fox et al., 2018; Gholamrezaei et al., 2017; Jackman et al., 2018), the strong focus on samples of young White women substantially reduces the generalizability of the reviewed findings. Methodologically, most studies chose an AA design and typically sampled for two weeks with six random daily prompts. Participants in AA studies completed on average two thirds of random or fixed prompts, and several studies relied on additional event-contingent reporting to capture NSSI events or urges. Studies with event-contingent assessment tended to observe more acts than studies with only random sampling.¹

Aim 1: Review of the phenomenology of NSSI acts

Table 1 provides detailed descriptive information for NSSI acts and their daily life context. The majority of studies assessed NSSI acts dichotomously, as present or absent. Across the five DD studies, four assessed NSSI dichotomously over ‘the past day’. In AA studies, 11 out of 16 samples assessed NSSI dichotomously as present or absent ‘since the previous prompt,’ ‘just now,’ or similar. The remaining five samples used either a sum score across several different NSSI methods since the last prompt or created a dichotomous NSSI variable from items assessing specific NSSI methods. One exception of note is Kranzler et al. (2018), who had participants first report whether they had engaged in NSSI and, if they had, the number of individual times (e.g., the number of individual cuts). They found considerable variance, with participants reporting a range of 1–25 individual NSSI acts.

Table 1.

Descriptive statistics, method and context for NSSI acts

Study	NSSI assessment	N (%) participants reporting acts	N total acts	M acts per participant	Method	Context of NSSI and pain associated with NSSI
Ammerman et al. (2017)	NSSI (y|n) since last call	N = 13 (26%)	/	/	77.8% banging head 60.3% poking/biting 39.6% cutting 22.2% burning 51.9% other	/
Andrewes et al. (2016) + Andrewes et al. (2017)	NSSI (y|n) since last prompt	N = 24 (22%)	N = 52	M = 0.49 across all M = 2.17 across participants w ≥1 act	70% cutting 15% scratching 6% hitting object/self 2% biting 2% burning 2% strangling	29% on weekends 33.3%: 10– 2 pm 33.3%: 2– 6 pm 33.3%: 6– 10 pm
Anestis et al. (2012)	NSSI Sum of: Cutting (y|n), burning (y|n), hitting (y|n), head banging (y|n) time frame n.r.	/	/	NSSI Sum: M = 0.28, SD = 0.97	Percentages n.r.	/
Armey et al. (2011) + Armey et al. (2012)	NSSI (y|n) since last prompt	N = 17 (47%)	N = 22	M = 0.61 across all M = 1.29 across participants w ≥1 act	29.4% cutting 23.5% wound picking 23.5% scrape skin 11.8% beat/ hit self 11.8% biting 0 % burning	Time spent planning: none at all, a few seconds, a few minutes, <1 hour, <1 day, 1 – 2 days, >2 days 94.1% reported less than 1 h other percentages n.r.
Bresin et al. (2013)	NSSI (y|n) for that day		N = 9	M = 0.13 across participants w ≥1 act	/	/
Czyz et al. (2019)	NSSI (y|n) for that day	N = 15 (44%)	N = 70	M = 2.06 across all M = 4.7 across participants w ≥1 act	/	38.6% afternoon (12 pm- 6 pm) 22.9% nighttime (12 am- 6 am) 20.0% evening (6 pm- 12 am) 18.6% morning (6 am- 12 pm)
Hochard et al. (2015) + Hochard et al. (2019)	NSSI (y/n) for that day since wake-up Pooled with urges	For that day: N = 16 (22%) Since wake-up: N = 11 (15%)	For that day: N = 9 Since wake-up: N = 19	For that day: M = 0.13 across all M = 0.56 across participants w ≥1 act Since wake-up: M = 0.26 across all M = 1.73 across participants w ≥1 act	/	/
Houben et al. (2017) + Vansteelant et al. (2017)	NSSI (y/n) since last prompt	N = 30 (94%)	N = 88	M = 2.75 across all M = 2.93 across participants w ≥1 act	/	/
Horowitz et al. (2017)	NSSI (y/n) for that day	N = 16 (42%)	N = 46	M = 1.21 across all M = 2.88 across participants w ≥1 act	/	/
Kranzler et al. (2018) + Selby et al. (2019) + Hughes et al. (2019)	NSSI (y/n) since last prompt	N = 40 (85%)	N = 442 in 145 episodes M number of acts per episode = 3.05 (SD = 3.65)	Counting episodes: M = 3.09 across all M = 3.63 across participants w ≥1 act	40.7% cutting 32.4% punching 17.9% scratching 9.7% biting 9.0% burning	Pain assessed 0 – 10: pain before, during, after NSSI episode (retrospectively) M before = 1.94, SD = 2.77 M during = 4.51, SD = 3.10 M after = 3.37, SD = 2.91 Modal duration 1 – 30 mins
Law et al. (2015)	“I hurt myself on purpose in the last 60 minutes” (0 – 5)	/	/	/	/	/
Lear et al. (2019)	NSSI (y|n) for that day	N = 20 (43%)	N = 48	M = 1.02 across all M = 2.40 across participants w ≥1 act	50.0% wound manip. 41.7% Cutting 27.1% scratching 18.8% pinching 12.5% pulling Hair 10.4% hitting self All other < 5%	Mean time of NSSI acts 3:30 pm Pain assessed 0-10: Pain during NSSI M = 2.96 (SD = 2.1)
Muehlenkamp et al. (2010)	any 1 or more of: Cutting (y|n), Burning (y|n), Hitting (y|n), Head banging (y|n) since last prompt	N = 19 (15%)	N = 55	M = 0.42 across all M = 2.90 across participants w ≥1 act	Individual frequencies for different methods not reported	/
Nock et al. (2010) + Selby et al. (2014)	NSSI act (y|n) since last prompt	N = 26 (87%)	N = 104	M = 3.47 across all M = 4.00 across participants w ≥1 act	/	Who with? 49.0% alone 16.3% peer 16.3% friend 9.6% mother 5.8% father 5.8% stranger 3.8% sibling 1.9% other	Doing what? 21.2% socializing 20.2% resting 19.2% homework 17.3% music 14.4% tv/games 15.4% recreation 13.5% eating 4.8% drugs 3.8% alcohol
Pearson et al. (2016)	NSSI (y|n) of: Cutting (y|n) Burning (y|n) Hitting (y|n) Head banging (y|n) Scratching (y|n) Since last prompt	N = 19 (14%)	/	/	Frequencies not reported	/
Selby et al. (2013)	NSSI (y|n) since last prompt	N = 7 (35%)	N = 25	M = 1.25 across all M = 3.57 across participants w ≥1 act	52% picking skin 24% cutting 24% scratching 12% hitting self	Amount of time between urge and act: < 1min (68%), 1 – 5 min between (20%), 10 – 30 min (12%) Pain during NSSI assessed (y/n): Reported yes for 20 acts (80%)
Shingleton et al. (2013)	NSSI (y|n) just now	N = 24 (80%)	N = 83	M = 2.77 across all M = 3.46 across participants w ≥1 act	/	/
Snir et al. (2015)	NSSI (y|n) since last prompt	N = 29 (29%)	N = 110	M = 1.11 across all M = 3.79 across participants w ≥1 act	/	/
Turner et al. (2016ab, 2018) + Kleiman et al. (2018) + Miller et al. (2019)	NSSI act (y|n) retrospective report in the evening for three different episodes of the day	N = 31 (52%)	N = 107	M = 1.78 across all M = 3.45 across participants w ≥1 act	29.9% scratching 16.8% cutting 15.9% hitting 15.9% other 4.7% biting	/
Victor & Klonsky (2014)	No. of NSSI acts indicate y|n for 12 NSSI methods drawn from ISAS for that day	/	/	/	frequencies for individual methods n.r.	/
Zaki et al. (2013)	NSSI act (y|n) since last prompt			M = 1.0 acts in BPD group, SD = 2.1, range 0–8	/	/

Note. N.r. = not reported.

The number of observed NSSI acts varied considerably across studies. To obtain a fuller picture of how prevalent acts were, we reviewed the percentage of participants in each study that reported any acts, the average number of acts per participant when considering all included participants in each study, and the average number of acts per participant when considering only participants that showed at least one act during the study period. The percentage of participants that showed at least one act ranged from 14.3% to 93.8% and, on average, 46.1% of participants in each sample reported NSSI (Md = 42.5, SD = 25.6). Thus, the majority of all assessed participants did not report any NSSI acts. The average number of acts across all participants was 1.60 (Md = 1.23, SD = 1.06) and this number increased to 2.82 (Md = 2.93, SD = 1.23), when considering only participants that reported any act.

Eleven samples assessed which NSSI method participants used for each specific NSSI act. However, only seven samples reported this data. The method with the highest endorsement rate across studies was skin picking or wound manipulation. This was endorsed for an average 41.8% of NSSI acts and was the most commonly endorsed method in two samples (Armey et al., 2011; Lear, Wilkowski, & Pepper, 2019), but was only assessed in three samples. Cutting, in contrast, was assessed in all seven samples and endorsed for an average 37.4% of NSSI acts. It was also the most commonly endorsed method in three samples (Andrewes, Hulbert, Cotton, Betts, & Chanen, 2016; Armey et al., 2011; Kranzler et al., 2018). The next most common method was hitting oneself (incl. head banging), which was endorsed for 23.8% of acts and was the most common method in one sample (Ammerman et al., 2017). Scratching was assessed in six samples and endorsed for an average of 22.9% of acts and was the most endorsed method in also one sample (Turner, Cobb, Gratz, & Chapman, 2016). Lastly, biting was assessed in 5 samples and was endorsed for an average 17.7% of acts across studies, followed by burning the skin, which was endorsed in 8.3% of acts across four samples.

Other aspects of NSSI acts, such as how severe or painful the injury was, were infrequently assessed. Only three samples assessed the experience of pain during NSSI (Kranzler et al., 2018; Lear et al., 2019; Selby et al., 2013). This is of particular interest because pain is assumed to be a central component of NSSI and a number of experimental studies find that individuals with a history of NSSI report reduced pain experience on pain induction tasks (Ammerman et al., 2018). The reviewed studies found that most individuals report some pain (vs. complete analgesia) during NSSI, but that the pain intensity was generally mild (i.e. ranging around 3 on a 0–10 scale).

Turning to context of NSSI acts, few studies have utilized the potential of AA and DD methods for characterizing when and where NSSI acts take place. Only four out the reviewed 24 samples have assessed contextual factors surrounding NSSI acts. Due to this clear gap, the previous evidence on contextual factors is difficult to synthesize as it relies on a small number of findings. In detail, two samples suggest that NSSI may be somewhat more likely to occur in the afternoon (Czyz, Glenn, Busby, & King, 2019; Lear et al., 2019), while a third found an even distribution across the day (Andrewes et al., 2016). In the only other sample to assess context, Nock et al. (2010) assessed what youth were doing before they engaged in NSSI and who they were with, finding that they typically reported being alone prior to NSSI and engaging in largely solitary activities (e.g., listening to music). Overall, more detailed work on the context of NSSI is needed to identify when and where NSSI typically occurs.

Aim 1 summary

The majority of studies assessed NSSI dichotomously and did not assess the severity of the NSSI act. On average, only slightly less than half of the participants included in each study reported any NSSI act. Thus, the majority of included participants actually showed no such behavior during the assessment period. However, when aggregating across participants within studies, studies observed an average of 1.6 NSSI acts per person and when considering only those participants that showed any NSSI during the study period, this number increased to 2.8. Across study periods that were a median of two weeks long, this suggests that individuals may not often engage in NSSI more than once a day. However, the range of observed events varied considerably across studies and, surprisingly, not all studies reported how many participants engaged in NSSI or the total number of acts. Methods were infrequently assessed and reported but the available data points to skin picking, cutting, hitting oneself, and scratching as commonly used, whereas other methods such as burning and biting were infrequently endorsed. Even though pain is assumed to be a central component of NSSI, it was infrequently assessed, with the available evidence suggesting individuals report low levels of pain but not complete analgesia. Lastly, the daily life context of NSSI acts is under-researched and it remains largely unclear whether NSSI is more likely to occur under certain conditions. Preliminary evidence suggests NSSI acts could be more likely when individuals are alone and later during the day, but more research is warranted to further characterize the context of NSSI acts.

Aim 2: Review of daily life findings on NSSI urges

The second aim of the present review was to summarize the daily life evidence on NSSI urges or thoughts, which were assessed in 14 samples (see Table 2; 10 AA, 4 DD). For simplicity, we will use the term ‘NSSI urges’ to refer to either thoughts or urges. Almost all samples assessed NSSI urges dichotomously as absent or present and, in four samples, participants rated NSSI urges both as absent/present and continuously in terms of intensity (Kranzler et al., 2018; Lear et al., 2019; Nock et al., 2010; Scala et al., 2018). In three samples (Humber et al., 2013; Turner, Cobb, et al., 2016; Victor, Scott, Stepp, & Goldstein, 2019), NSSI urges were assessed only continuously in terms of intensity. Studies also varied in terms of the period of time participants were asked to retrospect over when assessing NSSI urges. In AA studies, participants reported on urges ‘since the previous assessment,’ ‘right now’ or similar. Moreover, participants were able to use event-contingent reports to self-report urges in one study (Nock et al., 2010). In DD studies, participants reported the presence of an urge over the past day. Allowing participants to report any urge since the last assessment or over the entire day increases the probability that urges will be reported, which may be important given that even intense urges may be short lived. However, this comes with the limitation of reduced information about the temporal ordering of NSSI urges with respect to other phenomena of interest.

Table 2.

Descriptive statistics for NSSI urges and context.

Study	Urge assessment	N (%) participants w urges	N total urges	M acts per participant	Intensity, duration, and context
Andrewes et al. (2016) + Andrewes et al.(2017)	NSSI thought (y|n) since last prompt	N = 70 (65%)	N = 330	M = 3.08 across all M = 4.71 across participants w ≥1 urge	Event prior to thought? 46% ‘nothing’ 21% interpers. conflict 10% memories 8% rumination 6% stressful event 6% suicidal thoughts 2% eating, 2% bored	Occurring when? 22% on weekends 38% between 10 am - 2 pm 42% between 2 pm - 6pm 23% between 6 pm - 10 pm
Bresin et al.(2013)	NSSI urge (y|n) for that day	N = 67 (100%)	N = 47	M = 1.11 across all M = 1.11 across participants w ≥1 urge	/
Horowitz et al.(2017)	NSSI urge (y/n) in the last 24 hours	N = 35 (92%)	N = 148	M = 3.90 across all M = 4.23 across participants w ≥1 urge	/
Humber et al.(2013)	NSSI urge intensity (0–10) right now	/	/	/	intensity (1–10) M = 2.5, SD = 2.0
Kranzler (2018) Selby (2019) + Hughes (2019)	NSSI thoughts (y/n) since last prompt	N = 47 (100%)	N = 536	M = 11.4 across all M = 11.4 across participants w ≥1 urge	intensity (0 –10) M = 5.9, SD = 2.5 Duration and ability to resist assessed but n.r.
Lear et al. (2019)	NSSI urge (y/n) + intensity (0–6) since last diary	N = 45 (96%)	N = 210	M = 4.47 across all M = 4.67 across participants w ≥1 urge	intensity (0–6) M = 1.1 SD = 1.8	Strongest urges occurred on average around 3:30 pm
Nock et al. (2010) + Selby et al.(2014)	NSSI thought (y|n) since last prompt/self-initiated + intensity (0–4) + duration Both assessed when acts followed and when they did not	N = 28 (93%)	N = 344	M = 11.47 across all M = 12.29 across participants w ≥1 urge	Intensity (no act | with act) absent 1.7% | 0.0% mild 25.2% | 1.0% moderate 38.5% | 18.4% severe 25.2% | 32.0% very severe 9.4% | 48.5% duration (no act | with act) < 5 sec 5.0% | 16.5% 5–60sec 0.8% | 20.4% 1–30 min 46.2% | 40.8% 30–60 min 15.4% | 13.6% 1–5 hr 15.4% | 7.8%, >5 hr 2.9% | 1.0%	Doing what? 31.3% social 22.9% resting 13.8% music 7.7% homework 13.3% gaming 10.8% recreation 11.3% eating 2.9% drugs 2.5% alcohol	Who with? 38.3% alone 29.6% peer 12.9% friend 11.7% mother 6.7% father 5.8% stranger 2.9% sibling 0.8% other
Scala et al. (2018)	NSSI urge (y/n) since last prompt	N = 17 (31%)	N = 84	M = 1.05 across all M = 3.85 across participants w ≥1 urge	intensity assessed but n.r.	/
Shingleton et al.(2013)	NSSI thought (y/n) right now	N = 28 (93%)	N = 249	M = 8.30 across all M = 8.89 across participants w ≥1 urge		6.3% eating 30.4% with friends
Snir et al.(2015)	NSSI urge (y|n) since last prompt	N = 27 (27%)	N = 104	M = 1.05 across all M = 3.85 across participants w ≥1 urge		/
Turner et al.(2016a) + Turner et al.(2016b) + Turner et al.(2018) + Kleiman et al.(2018) + Miller et al.(2019)	NSSI urge scale retrospective ratings for three episodes of the day ABUSI (total sum score ranging from 0 ‘no urge’ to 29 ‘ex- tremely high urge’)	Fleeting thoughts: N = 55 (92%) Persistent thoughts: N = 29 (48%) Intense urges: N = 39 (65%)	Fleeting thoughts: N = 355 Persisten thoughts: N = 81 Intense urges: N = 125	Fleeting thoughts: M = 5.92 across all M = 6.46 w ≥1 urge Persistent thoughts: M = 1.35 across all M = 2.79 w ≥1 urge Intense urges: M = 2.08 across all M = 3.21 w ≥1 urge	Occurring when? Persistent thoughts: 9.9% morning 36.3% midday 53.8% evening Intense urges: 11.8% morning 38.6% midday 49.7% evening	Doing what? 47.1% resting 35.3% work 27.5% social 25.5% TV 23.5% eating 21.6% music 17.6% recreation 2% drugs 2% alcohol	Where? 75% alone 4.2% w friends 6.3% w family 8.3% work/school 8.3% other public setting
						Events before urge started? 35.3% argument/conflict 27.5% felt isolated 23.5% disappointed s/o 21.6% let down by s/o 21.6% rejected 19.6% criticized/ put down 15.7% couldn’t spend time w. s/o 13.7% lost s/o important 13.7% had a new demand 13.7% talked about upsetting x 9.8% financial problems 9.8% health problems
Victor et al.(2018)	NSSI urge intensity (1–5), dichotomized in urge (y|n) since the last prompt	N = 28 (45%)	N = 127	M = 2.05 across all M = 4.45 across participants w ≥1 urge		Rejection and criticism precede urges: at t-1 they significantly predict urge probability at t0 Alcohol use and drug use not significantly associated with urge reporting
Zaki et al.(2013)	NSSI urge (y|n) since last prompt	/	/	/	M = 1.53 urges per BPD participant, SD = 3.49, range 0–15	/

To obtain a fuller picture of urge prevalence in the reviewed samples, we again distinguished the average percentage of participants that reported urges in each study, the average number of urges per participant, and the average number of urges per participants that reported at least one urge. Across studies that assessed urges, an average of 70.7% (Md = 78.5, SD = 28.2) of participants reported at least one urge during the assessment period. Considering all included participants, each participant reported an average 4.33 urges (Md = 2.58, SD = 3.88), and considering only those that reported at least one urge, each participant reported on average 5.24 urges (Md = 4.61, SD = 3.13). In six samples, participants also rated the intensity of their urges. Mean intensity was reported for four of these, but unfortunately was assessed using a different scale in each. In two samples, evidence points to urges being of a mild intensity, on average, (Humber et al., 2013; Lear et al., 2019), whereas two samples point to urges being of moderate intensity (Kranzler et al., 2018; Nock et al., 2010). In two samples, participants reported the duration of the urge, with only one sample reporting this information and finding that about half of NSSI urges persisted between one and 30 minutes (Nock et al., 2010).

Paralleling findings for acts, evidence on the daily life context of NSSI urges was very limited. Two samples examined the time of day participants experienced urges, one finding urges occurred at 3:30 pm on average (Lear et al., 2019), and the other finding they were most common in the evening (Andrewes et al., 2016). Four studies included at least one question to assess what participants were doing before experiencing an urge. One of these focused solely on eating behaviors due to the target population (Shingleton et al., 2013). Of the remaining, participants typically reported ‘resting’ or ‘doing nothing’, while a considerable proportion also endorsed a context of socializing or interpersonal conflict (Andrewes et al., 2016; Nock et al., 2010; Turner, Cobb, et al., 2016). Two of these samples assessed who participants were with when experiencing NSSI urges, finding that participants were often alone or with a friend or peer.

NSSI urges as predictors of NSSI acts

Among the 15 samples that measured NSSI urges, seven samples comprised descriptive information or tests on the association between NSSI urges and NSSI acts. This is a particularly important association to test as further evidence on the temporal association between urges and acts may help identify ideal time windows for mobile health intervention. At the day level, urges were significantly associated with NSSI acts in two studies (Ammerman et al., 2017; Lear et al., 2019), such that days with a higher number of urges or greater urge intensity entailed greater risk for same-day NSSI. At the momentary level, one study found that NSSI urges were positively associated with next-moment acts (Kranzler et al., 2018). Two studies found a similar association, but only when participants also reported NA, interpersonal conflict, or when they were alone (Nock, 2009; Turner, Cobb, et al., 2016). Turner, Cobb, et al. (2016) also detailed that the probability of NSSI urges leading to acts depended on the strength of the urge. They found that ‘fleeting thoughts’ of NSSI led to acts on only 23.4% of days on which they occurred, whereas ‘persistent thoughts’ did so on 37% of days. ‘Intense urges’ led to NSSI in more than half the number of days on which they were endorsed (57.6%).

Several studies also reported on the temporal association between urges and acts. Most studies point to a short time frame, finding the majority of acts occur less than a minute after the initial thought, with most other acts occurring within 30 minutes (Nock, 2009; Selby et al., 2013; Turner, Cobb, et al., 2016). However, Andrewes et al. (2016) found that the time-lag between first urge and act could be much longer with a median lag of 34.9 hours (range = 0 – 122 hr), although it is unclear what explains these sustained urges. However, they also found that changes in NA led to additional thoughts about NSSI, which occurred a median of 2.5 hours prior to NSSI (range = 0 – 27.79 hr). These findings support the hypothesis that NSSI acts are often preceded by urges in daily life and are not entirely impulsive acts that happen unannounced. Further, Andrewes et al. (2016) notwithstanding, urges and acts appear to often be in close temporal proximity to one another.

Aim 2 summary

NSSI urges were less frequently assessed than NSSI acts and not all studies reported how many participants experienced any urges or the total number of urges reported. Across studies that reported these details, approximately 70% of participants reported at least one urge, suggesting that studies were relatively successful in recruiting samples well suited to address research questions on urges. Participants who endorsed any urges experienced on average 5 urges across the study period (a median of 14 days). Data on average urge intensity was difficult to compare due to differing scales across studies but indicated that reported urges were most often mild or moderate. We also aimed to review the context of urges, but evidence was sparse. Some preliminary evidence points to urges being more likely later during the day, during sedentary activities, and in the presence of others. Studies also provided evidence for a close association between urges and acts, such that urges precede acts, typically by less than 30 minutes.

Aim 3: Review of evidence for the Four-function Model of NSSI

In the following sections, we review the evidence for intrapersonal and interpersonal negative and positive reinforcement as conceptualized in the Four-function Model of NSSI (Nock, 2009; Nock & Prinstein, 2004). We review both self-reported evidence (i.e. how often participants endorsed NSSI for intra/interpersonal effects) and evidence on the association between NSSI and affect or interpersonal variables (see Table 4).

Table 4.

Evidence on intrapersonal and interpersonal functions of non-suicidal self-injury.

Study	Affect and interpersonal items	Intrapersonal negative	Intrapersonal positive	Interpersonal function
Ammerman et al. (2017)	PANAS NA (M) Current	Not supported: Daily NA was not associated with NSSI	/	/
Andrewes et al. (2016) N = 107	PANAS NA (M) PANAS PA (M) Current	NA increased pre NSSI NA decreased post NSSI (quadratic curve)	PA decreased pre NSSI PA increased post NSSI (quadratic curve)	/
		Function ‘affect regulation’ self-endorsed 45% events Function ‘anti-dissociation’ self-endorsed in 5% events	Function ‘sensation seeking’ self-endorsed in 5% of events
Andrewes et al. (2017) N = 107	PANAS NA (M) incl. distressed item PANAS PA (M) ‘Negative complex emotions’ (NCE): Number of NA items rated >2 Current	NCE increased pre NSSI NCE decreased post NSSI (quadratic curve) distress increased pre NSSI distress decreased post NSSI (quadratic curve)	/	/
Armey et al. (2011) N = 36	PANAS NA (M) PANAS PA (M) PANAS-X guilt (M) PANAS-X hostility (M) Exploratory: Irritable, angry, loathing item Current	NA & guilt increased pre NSSI NA & guilt decreased post NSSI (quadratic curve) Exploratory: ‘Angry’ increased pre NSSI ‘Angry’ decreased post NSSI (quadratic curve)	Not supported: PA did not show significant pattern surrounding NSSI	/
Horowitz et al. (2017) N = 38	/	Function ‘affect regulation’ most commonly endorsed		Functions ‘setting interpersonal boundaries’, ‘revenge’, ‘influencing others’, ‘establishing autonomy’ all self-endorsed for 0% of events
Houben et al. (2017) N = 30	NA M of: angry, depressed, anxious, stressed PAM of: happy, relaxed Scale 0 – 100; current	NA increased pre NSSI NA was high during NSSI Not supported: NA increased post NSSI	Not supported: PA decreased post NSSI	/
Hughes et al. (2019) N = 47	NA M of: sad, angry, frustrated, overwhelmed hurt/rejected, guilty, physically numb, empty/ numb, anxious/ afraid, lonely, ashamed Scale 0 – 10; current	NA increased pre NSSI Overwhelmed increased pre NSSI Anxious increased pre NSSI	/	/
Kranzler et al. (2018) N = 47	NA M of: overwhelmed, sad, frustrated, angry, hurt/ rejected, ashamed, anxious/afraid, lonely, embarrassed, empty/numb, guilty, physically numb PA M of: content, relieved, proud, rush or a high, excited, satisfied, happy calm/relaxed Scale 0– 10; current	NA increased pre NSSI NA decreased post NSSI Exploratory: Sad, angry, overwhelmed, lonely, frustrated, hurt, anxious items decreased post NSSI Function ‘to get rid of bad or negative feelings’ self-endorsed for 53.8% of events	Lag PA not associated w NSSI PA increased post NSSI Exploratory: happy, content, satisfied, proud, relieved, calm items increased post NSSI Function second most endorsed	/
Law et al. (2015) N = 255	NA M of: Irritable, angry, ashamed, guilty Scale 0 – 5; last 60 minutes	NA, angry, irritable high during NSSI Lagged NA not associated w NSSI	/	/
Lear et al. (2019)	PANAS-X guilt (M), hostility (M), sadness (M), fear (M) Created subsets of 4 items for each scale For the past day	Daily guilt was associated with increased probability for daily NSSI Daily hostility, sadness, fear were not	/	/
Muehlenkamp et al. (2009) N = 131	PANAS NA (M) subset PANAS PA (M) subset Time frame not reported	NA increased pre NSSI (linear trend) No change in NA following NSSI	PA decreased pre NSSI PA increased post NSSI (quadratic curve)	/
Nock et al. (2010) N = 30	Only reported as a context for urges, see urge table	Function endorsed in 64.7% of events Specifically ‘reducing/escaping’: - anxiety (34.8%), sadness (24.2%), anger (19.7%) - bad thought (28.8%), bad memory (13.6%)	Function self-endorsed in 24.5% of events	Interpersonal negative function self-endorsed in 14.7% of events Interpersonal positive function self-endorsed in 3.9% of events
Selby et al. (2014) N = 30	/	/	Participants self-reported engaging in NSSI to: Feel something (35% of events) Feel satisfaction (20% of events) Feel stimulation (16% of events)	/
Shingleton et al. (2013)	/	Function endorsed in 4% of events ‘get rid of anxiety/ bad thoughts’	/	/
Snir et al. (2015) N = 94	NA M: tense, disappointed, afraid, sad, angry, irritated Scale 0– 4; time frame not reported	Not supported: NA did not show a significant pattern surrounding NSSI	Function ‘feeling generation’ endorsed on average in 47% of events in the BPD group and 18% in the APD group	AB increased pre NSSI AB decreased post NSSI (quadratic curve, in APD group)
	Avoidant behavior (AB) M: Cancelled/avoided social plans, avoided conflict by keeping quiet, isolated self Yes/ No scale; since last entry Rejection/isolation (RI)M: lonely, isolated, abandoned, rejected, accepted (re), my needs are being met (re) Scale 0– 4; time frame not reported	Function ‘emotion relief endorsed on average in 52% of events in the BPD group and 27% in the APD group		RI increased pre NSSI RI decreased post NSSI (quadratic curve, in APD and BPD group) Function ‘interpersonal avoidance’ endorsed on average in 6% of events in the BPD group and 9% in the APD group Function ‘interpersonal communication’ endorsed on average in 12% of events in the BPD group and 17% in the APD group
Turner et al. (2016a) N = 60	MDMQ valence, calmness, energetic arousal Bipolar scale; > 0 negative valence and high arousal Conflict dayM of ‘Test of Negative Social Exchange’ Scale 0– 17 (number of negative events) DayM of ‘Goldsmith Social Support Scale’ Scale 1 (very unsupportive) – 7 (very supportive) All retrospective for morning, midday, evening	Function ‘to get rid of thoughts or feelings’ endorsed in 67.3% of events	Function ‘to feel something’ endorsed in 14.3% of events	Conflict was increased on days with NSSI Conflict did not decrease on days post NSSI Social support increased on days post NSSI that was revealed to others Function ‘to escape people or task’ endorsed in 16.3% of events, ‘to communicate’ in 2% of events
Turner et al. (2016b) N = 25	Sad/worthless, overwhelmed, scared/anxious, angry at self, self-hatred, angry at other, rejected/hurt, numb/nothing Retrospective for feelings right before NSSI	Increased feelings of being rejected / hurt were reported pre NSSI	/

Note. PANAS = Positive negative affect scale with a range of 1 (very slightly/ not at all) to 5 (extremely), NA = negative affect, PA = positive affect, BPD = Borderline Personality Disorder, APD = Avoidant Personality Disorder.

Evidence for intrapersonal negative reinforcement

Participants in eight unique samples self-reported that they engaged in NSSI to reduce aversive internal states, which describes negative intrapersonal reinforcement. In all but one of these samples, this was the most commonly endorsed function. Specifically, participants indicated they performed NSSI to regulate their affect, ‘get rid of negative feelings’, or obtain ‘emotion relief’ for, on average, 44.8% of NSSI acts (SD = 22.4). Notably, this includes the study by Shingleton et al. (2013), which was an outlier reporting only 4% endorsement of intrapersonal negative reinforcement. Nock et al. (2010) further assessed self-reported intrapersonal negative reinforcement for specific types of NA and found that adolescents with mood, anxiety, and substance use disorders most commonly endorsed wanting to reduce anxiety (34.8% of acts), followed by sadness (24.2%), and anger (19.7%). Additionally, participants in their study specifically endorsed the goal of reducing negative thoughts (28.8%) and bad memories (13.6%). Two further studies coupled the question about NA reduction with negative thoughts (Shingleton et al., 2013; Turner, Cobb, et al., 2016). Adults with anxiety and mood disorders or BPD in the study by h Turner and colleagues reported NSSI with the goal to ‘get rid of thoughts or feelings’ in 67.3% of events, but adolescents with depression or generalized anxiety disorder in the study by Shingleton and colleagues endorsed this in only 4% of cases. Lastly, youth with BPD in the study by Andrewes et al. (2016) endorsed NSSI with the goal to end dissociation in 5% of cases.

Beyond self-reported functions, several studies also assessed the association between daily life NSSI and aversive inner states. Predominantly, these studies focused on the association between NSSI and momentary or daily NA. Intrapersonal negative reinforcement in this case would imply that a) NA is elevated prior to NSSI, and that it b) decreases from pre to post NSSI, which c) increases the probability of individuals using NSSI to reduce NA in the future. While no AA study has directly assessed component c), both increased levels of NA prior to NSSI and decreased NA post NSSI have been observed. Increased NA prior to NSSI acts was found in eight out of 10 studies. Most studies examined general NA (Andrewes et al., 2016; Armey et al., 2011; Houben et al., 2017; Hughes et al., 2019; Kranzler et al., 2018; Muehlenkamp et al., 2009). With regard to specific types of NA, one study found elevated levels prior to NSSI for a ‘negative complex emotions’ index, which comprised the number of NA items that were rated above 2 on a 1 to 5 scale (Andrewes, Hulbert, Cotton, Betts, & Chanen, 2017), for the PANAS-X guilt scale (Armey et al., 2011), and for a number of individual NA items, including ‘distressed’ (Andrewes et al., 2017), ‘angry’ (Armey et al., 2011), ‘overwhelmed’ and ‘anxious/afraid’ (Hughes et al., 2019) and ‘feeling rejected or hurt’ (Turner, Yiu, Claes, Muehlenkamp, & Chapman, 2016)². In contrast, Law, Fleeson, Arnold, and Furr (2015) and Snir, Rafaeli, Gadassi, Berenson, and Downey (2015) examined whether NA at one report was associated with NSSI acts at the next report and found no significant association.

A decrease in NA post NSSI was observed in four out of seven studies. Specifically, Andrewes and colleagues (2016, 2017) found a decrease in mean NA following NSSI, as well as a decrease in negative complex emotions and ‘distressed’ affect. Armey et al. (2011) observed decreases in NA and guilt, as well as ‘angry’ affect post NSSI, as did Kranzler et al. (2018) for general NA and the specific items ‘sad, angry, overwhelmed, lonely, frustrated, hurt, anxious’. In contrast, Houben et al. (2017) did not observe a decrease post NSSI, but rather found that NA continued to increase after the NSSI event. Muehlenkamp et al. (2009) and Snir et al. (2015) both found no change in NA following NSSI.

Two important methodological aspects to consider when interpreting these findings are the time frame for which change in affect was measured and the number of NSSI acts that were observed. Regarding time-frames, samples can be split into those that assessed affect proximal to the NSSI act and studies that considered affect across the whole day or even across days. Of the latter, four of the reviewed studies modelled change in NA for time frames of 10 hours or more. Andrewes et al. (2016, 2017) modelled NA 15 hours prior to and following NSSI, and observed both increases of NA prior to and decreases post NSSI that followed a quadratic pattern in youth with BPD. Similarly, Armey et al. (2011) modelled NA up to 20 hours prior to and 20 hours post NSSI and also found the predicted quadratic trajectory in a sample of college students with lifetime NSSI history. Contrasting this, Snir et al. (2015), modelled changes in affect up to 10 hours before and after NSSI and found no discernable pattern for NA surrounding NSSI acts in participants with BPD or APD. Andrewes et al. (2016) observed 52 NSSI acts in their sample, Armey et al. (2011) observed 22 events, and Snir et al. (2015) observed 110 events. Thus, while findings for longer time frames from two samples support negative intrapersonal reinforcement, the negative findings by Snir et al. (2015) are based on a larger number of acts than both other studies combined, and should therefore be weighed equally. Looking at short time periods, Kranzler et al. (2018) found that NA levels 2–3 hours prior to NSSI predict NSSI engagement, and that NSSI predicts NA decrease at the next prompt (again around 2–3 hours later) in youth with depression or BPD. Importantly, these findings are based on a large number of 143 individual NSSI episodes and thus provide strong evidence for intrapersonal negative reinforcement. This is contrasted by Muehlenkamp et al. (2009), who observed a linear trend for increased NA pre NSSI, but no changes in NA in the 4 hours following NSSI, based on 55 acts and in adult women with bulimia nervosa. Likewise, Houben et al. (2017) found increased rather than decreased NA approximately 1.5 hours following 88 NSSI acts in inpatients with high BPD features and depression scores.

In addition to the association between NSSI acts and negative affect, the reviewed studies also provide evidence on urges and affect. The association between NSSI urges and affect was assessed in 17 of the reviewed studies (12 samples). Most of these focused on NA, though the exact types of NA assessed varied considerably (see Table 3). All but one study (Scala et al., 2018) found that NSSI urges were positively associated with some form of NA. This included general NA scales such as the PANAS NA scale, specific negative emotion scales such as the PANAS-X guilt scale, emotion constructs such as negative complex emotions (Andrewes et al., 2017), and even individual emotion items such as ‘overwhelmed’ (Kranzler et al., 2018). As summarized in Table 3, positive associations between NSSI urges and NA were present at the day level, such that days characterized by more NA entailed higher risk for NSSI urges, and also at the momentary level. This included both concurrent and lagged associations, meaning that NA was elevated at time-points when urges were reported and that NA also prospectively predicted the occurrence of urges or urge strength at later time points. Contrasting this, evidence on the trajectory of NA after an urge was sparse and less conclusive. Only Snir et al. (2015) modelled the course of NA surrounding urges and found that NA increased before the urge, continued rising after and then faded.

Table 3.

Evidence on the association between NSSI urges and affect and the association between NSSI urges and NSSI acts

Paper	Affect	Association with affect	Association with acts
Ammerman et al. (2017)	PANAS NA (M) Current	/	Daily urges were positively associated with the probability to engage in NSSI that day
Andrewes et al. (2016) + Andrewes et al. (2017)	PANAS NA (M) PANAS PA (M) Short-form with 5 items each ‘Negative complex emotions’ (NCE): number of NA items rated >2 Current	Significant positive predictors of thought occurrence: - changes in NA - changes in PA - NCE - ‘distressed’ item from NA scale	Md = 34.9 hr (range 0–122 hr) passed from initial thoughts about NSSI to NSSI act
Bresin et al. (2013)	PANAS-X NA (M) PANAS-X guilt (M) PANAS-X sadness (M) Daily	Significant predictors of daily urge occurrence: - daily NA - daily sadness (not guilt) - (trait) negative urgency x daily sadness	/
Hochard et al. (2015)	PANAS-NA (M) Current	Significant predictors of NSSI urge in the morning: - NA	/
Humber et al. (2013)	Anger affect (1–10) sum of: irritable, wound up, frustrated, angry, touchy, annoyed Distress (1–10) sum of: hopeless, trapped, low, isolated, helpless, lonely, worthless Current	Significant positive predictors of concurrent urges: - anger affect - anger ex-in - anger ex-out (also associated with NSSI urges one prompt later)	/
Kranzler et al. (2018) + Hughes et al. (2019)	NA (0–10) sum of: sad, angry, frustrated, overwhelmed, hurt/emotionally rejected, guilty anxious/afraid, lonely, ashamed, empty/numb, physically numb PA (0–10) sum of: content, relieved, proud, experiencing a rush or a high, excited, satisfied calm/relaxed, happyCurrent	Significant positive t−1 predictors of t0thought intensity - NA - anxious/afraid - overwhelmed Significant negative t−1 predictor of t0 NSSI thought intensity - PA	NSSI thought intensity at t−1 positively predicted NSSI act probability at t0
Lear et al. (2019)	Guilt, Hostility, Sadness, and Fear subscales of PANAS X (1–5)	Significant positive predictors of daily urge intensity - guilt - sadness	NSSI urge intensity was positively associated with same day NSSI act probability
	Daily	- hostility
Nock et al. (2009) + Selby et al. (2014)	Sad/worthless, overwhelmed, scared/anxious, angry at self, self-hatred, angry at another, rejected/hurt, numb/nothing (scale not reported) Retrospective: what led to the thought?	What were you feeling (before NSSI thoughts)? 37.9/39.8% sad/worthless 33.8/45.6% overwhelmed 31.3/32.0% scared/anxious 21.7/48.5% angry at self 21.7/42.7% self-hatred 23.3/35.0% angry at another 15.0/34.0% rejected/hurt 9.2/21.4% numb/nothing	NSSI thoughts were significantly positively associated with the probability for NSSI acts when thoughts were preceded by: - being alone - feeling sad/worthless - feeling angry at self - feeling self-hatred - feeling at another person - feeling rejected/hurt - feeling numb/nothing
			Amount of time that passed from thought to NSSI act: For individuals with an APR motivation (automatic positive reinforcement) mostly 1–30 min For non-APR mostly < 1 min
Scala et al. (2018)	NA (0–100) M of: irritable, sad, frightened, angry Current	Significant predictors of NSSI urge occurrence - interaction of NA (high) x self-concept clarity (low) - NA alone not significant	/
Selby et al. (2013)	/	/	Amount of time that passed from urge to NSSI act: < 1min (68%) 1 – 5 min (20%) 10 – 30 min (12%)
Shingleton et al. (2013)	Open question: “What led to the NSSI thoughts?” Retrospective	What led to the NSSI thought? % cases of all NSSI thoughts 8.2% criticism/ insult 21.1% feeling rejected/hurt 38.2% feeling sad/worthless	/
Snir et al. (2015)	NA (0–4) M of: tense, disappointed, afraid, sad, angry, irritated Current	Significant predictor of urge occurrence: - NA NA increased before the urge, continued rising after and then faded in a linear-quadratic-cubic pattern	/
Turner et al. (2016a) + Turner et al. (2016b) + Turner et al. (2018)	Sad/worthless, overwhelmed, scared/anxious, angry at self, self-hatred, angry at another, rejected/hurt, numb/nothing Retrospective for morning, afternoon, evening	Retrospective feelings reported for ‘right before’ urge: overwhelmed (58.8%) sad/worthless (49.0%) rejected/hurt (45.1%) angry at self (41.2%) self-hatred, scared/anxious (39.2%) angry at others (25.5%) numb/nothing (21.6%)	NSSI thoughts were significantly positively associated with the probability for NSSI acts when thoughts were preceded by: - arguments/conflicts - feeling rejected/hurt Time passed between NSSI thought and act on average between 1 and 30 min - Fleeting thoughts led to acts on 23.4% of days they occurred - Persistent thoughts led to acts on 37% of days they occurred - Intense urges led to acts on 57.6% of days they occurred
Victor et al. (2108) N = 62	‘Intemalizing’ NA (1–5) M of: ashamed, guilty, scared, lonely, sad ‘Externalizing’ NA (1–5) M of: hostile, irritable, angry at others, annoyed, mad Past 15 min	Significant positive t−1predictors of t0 NSSI urge occur- rence: - internalizing NA (externalizing NA significant when entered alone, not significant when tested against internalizing NA)	/

Note. PANAS = Positive negative affect scale with a range of range 1 (very slightly/ not at all) to 5 (extremely), NA = negative affect, PA = positive affect, BPD = Borderline Personality Disorder, APD = Avoidant Personality Disorder.

Evidence for intrapersonal positive reinforcement

Participants self-endorsed performing NSSI to increase desired internal states in a number of studies, but overall this function was less commonly endorsed than intrapersonal negative reinforcement. Performing NSSI with the desired effect of ‘feeling something’ was endorsed by participants in three studies, specifically in 14.3% of NSSI events in the study by Turner, Yiu, et al. (2016), in 25% of events in the sample reported by Nock et al. (2010) and in 35% in Selby et al. (2013). Turner and colleagues as well as Selby and colleagues assessed adults with mixed psychopathology, whereas Nock assessed adolescents, also with different types of psychopathology. Participants in the sample collected by Selby et al. also reported NSSI with the motivation to ‘feel satisfaction’ (20%) or ‘feel stimulation’ (16%). Additionally youth with BPD endorsed ‘sensation seeking’ for 5% of acts (Andrewes et al., 2016) and adults with BPD or APD endorsed ‘feeling generation’ as a motive in 32.5% of events across groups (Snir et al., 2015).

Beyond self-reported motives, five studies tested whether PA decreased prior to and increased post NSSI. Two studies found decreased levels of PA pre NSSI and an increase post NSSI that followed a quadratic trend (15 hour time-frame in youth with BPD: Andrewes et al., 2016; 4 hour time-frame in adult women with bulimia nervosa: Muehlenkamp et al., 2009). Partly corroborating this, Kranzler et al. (2018) also observed an increase in PA 2–3 hours following NSSI, but did not find decreased PA prior to NSSI in youth with depression or BPD. Not supporting positive reinforcement, Armey et al. (2011) did not find a significant pattern of PA in the 20 hours pre and post NSSI in college students and Houben et al. (2017) found decreased PA approx. 1.5 hours following NSSI in inpatients with BPD features.

Associations between NSSI urges and positive affect (PA) were only assessed in two studies. Andrewes et al. (2016) found a positive correlation between the timing of the initial thought about NSSI and a change in PA. In line with this, PA negatively predicted NSSI urge intensity one prompt later in the study by Kranzler et al. (2018).

Evidence for the interpersonal function of NSSI

The interpersonal function of NSSI suggests individuals engage in NSSI to influence others or to create desired outcomes in interactions with others. Four of the reviewed studies assessed self-reported functions of interpersonal negative reinforcement³. Participants endorsed the function ‘To escape a task or people’ in approximately 15% of NSSI events in two studies (adolscents with mixed psychopathology: Nock et al., 2010; adults with mixed psychopathology, including BPD: Turner, Yiu, et al., 2016), and the function ‘interpersonal avoidance’ in around 8% of NSSI events across a BPD and an APD group (Snir et al., 2015). The same studies also assessed self-reported functions pertaining to interpersonal positive reinforcement. Endorsement rates for the function ‘interpersonal communication’ ranged from 2% and 4% in studies by Nock et al. (2010) and Turner, Yiu, et al. (2016), to 12% in the BPD group and 17% in the APD group in the study by Snir et al. (2015). Lastly, Horowitz and Stermac (2018) assessed the functions ‘influencing others’, ‘getting revenge’, ‘establishing autonomy’, and ‘setting interpersonal boundaries’, in community individuals with NSSI history, which all showed a close to zero endorsement.

Beyond self-reported interpersonal functions, only two of the reviewed 35 studies assessed interpersonal constructs and how they relate to NSSI in daily life. Snir et al. (2015) assessed perceived rejection/isolation from others at each assessment and found that it increased prior to NSSI and decreased post NSSI in a quadratic trend. This was observed for both the BPD group and the APD group in their sample, and provides evidence for interpersonal negative reinforcement. Adding to this, Turner, Cobb, et al. (2016) assessed how conflict and social support related to NSSI on a daily basis in adults with anxiety and mood disorders or BPD. They hypothesized that interpersonal conflict would decrease on days following NSSI that was revealed to others, thus eliciting the desired interpersonal negative reinforcement of NSSI. However, while interpersonal conflict was elevated on days with NSSI, conflict did not decrease on days following revealed NSSI. The authors further tested the association between social support and NSSI. They found that social support increased on days following NSSI that was revealed to others, which supports the notion that NSSI can elicit interpersonal positive reinforcement.

Aim 3 summary

First, although a substantial number of daily life studies tested part of the Four-function model of NSSI (Nock, 2009), no study tested the whole model and the overall evidence was mixed. This is in stark contrast to evidence from cross-sectional studies which have largely supported the four different components of the model (e.g., Bentley et al., 2014). It is possible that the designs that were used in the reviewed studies were not ideal to test the model (see aim 4) or that the dynamics of NSSI in daily life are much more complex than previously assumed. Overall, the discrepancy between cross-sectional and AA evidence underlines the need to test theoretical models in real life.

The most evidence was available for intrapersonal negative reinforcement, which participants in the reviewed studies highly endorsed in the moment as their intended function. The evidence on whether NSSI actually had the intended effect of reducing NA was, however, less clear. Intrapersonal negative reinforcement of NSSI implies that NA should be elevated prior to NSSI, decrease significantly post NSSI, and thus increase the probability of using NSSI in the future. The reviewed studies provided some evidence for an increase of NA pre NSSI, whereas the evidence for a reduction in NA post NSSI was less conclusive. Evidence from only four studies pointed to a reduction in NA post NSSI, though with varying time frames ranging from one to more than 20 hours. Moreover, contrasting this, evidence from three other studies suggested that NSSI does not predict changes in later NA or even a further NA increase. Thus, while participants often intended for their NSSI to reduce NA, this did not necessarily take place. Future research is also needed to more explicitly test whether the probability to engage in NSSI increases after successful regulation of NA, as this most closely models negative reinforcement.

Evidence for intrapersonal positive reinforcement and the interpersonal function of NSSI was sparse and any conclusions have to be considered preliminary until further investigation. Three studies at least partly corroborated intrapersonal positive reinforcement of NSSI whereas two studies did not, and participants self-endorsed this function substantially less often than negative reinforcement. Further, only two studies assessed the interpersonal function, even though 33–56% of self-harming individuals explicitly endorse interpersonal functions for their self-harm in cross-sectional studies (Taylor et al., 2018). Evidence from these two studies suggests that the probability of negative interpersonal events (rejection, conflict) is elevated prior to NSSI, but only one study demonstrated a decrease post NSSI. Lastly, one study suggested that social support may increase post NSSI. Given the dearth of research on the interpersonal functions of NSSI and positive intrapersonal reinforcement, as well as inconclusive evidence for intrapersonal negative reinforcement, more AA work is needed to rigorously test theoretical models of NSSI such as the Four-function model as such evidence may diverge substantially from cross-sectional findings. For example, some cross-sectional work examining self-reported functions of NSSI has not found evidence for distinct negative and positive reinforcement functions (Klonsky et al., 2015). The existing AA evidence base is not yet sufficiently developed to determine the value of distinguishing negative and positive reinforcement in understanding the momentary experiences of individuals who engage in NSSI.

Discussion

In the following sections, we provide recommendations for future studies that were derived from each of the different topics we reviewed: Daily life phenomenology of NSSI acts (aim 1), evidence on NSSI urges (aim 2), and evidence on the Four-function model of NSSI (aim 3).

Implications from aim 1: Refining the descriptive stage in daily life NSSI research

One aspect that was overlooked in most of the reviewed studies is that daily life studies can be a rich source of descriptive data that may increase our understanding of the nature of NSSI. With regard to NSSI acts, we suggest four central constructs future studies should consider assessing more closely: the intended function of NSSI, the severity of the injury, the frequency of NSSI acts within an episode, and the experience of pain.

Several studies assessed the intended function of NSSI in the moment and this should be a standard part of AA protocols assessing NSSI. However, none examined whether NSSI has the effects individuals intended, for instance by assessing whether those that report engaging in NSSI to reduce NA actually see a decrease of NA post NSSI. This could also help explain repeated events within a day, examining whether individuals continue to engage in NSSI if the initial act did not have the desired effects. Moreover, assessing the intended function for each event could help determine whether different motives are associated with different methods or degree of severity. For instance, NSSI with the goal of ending dissociation/intrusions could be associated with greater injury severity.

Second, assessing the severity of the injury resulting from NSSI is important, because severity may moderate associations between NSSI and other constructs of interest, such as NA. For instance, a superficial cut may incur a smaller reduction in NA than a deep incision. However, the assessment of NSSI severity remains a challenge as no validated scale currently exists. Further, severity becomes even more difficult to gauge when comparing across different methods. In addition to the need to devise self-rating scales for the daily life context, the inclusion of photos of the wounds taken with the study phone in AA studies may help researchers rate wound severity externally. However, researchers must carefully think through potential ethical concerns related to injuries where a photo indicates the need for immediate medical attention.

Third, there is a need for a more fine-grained assessment of NSSI frequency. Most reviewed studies only assessed whether any NSSI event was present or not. In contrast to this, Kranzler et al. (2018) assessed repeated NSSI behaviors within distinct episodes and observed 143 episodes with 442 individual acts. Importantly, the authors were able to determine some variables that increased the number of behaviors within an episode, such as experiencing little to no pain or higher NA prior to the episode. Beyond this, there are likely certain methods that entail repetitive NSSI behavior rather than others. Indeed, method, severity, and frequency could be closely interlinked. To give an example, the administration of 10 deep cuts to the arm in quick succession may have a different effect on NA than one superficial scratch. In most of the reviewed studies, both examples would currently be characterized as only one NSSI event and important differences, either in type or magnitude of the effects they incur, would be missed.

Fourth, there is a need to assess the amount of pain caused by an NSSI act. Theoretical models of NSSI assume that pain is a central component for the effects and maintenance of the behavior (see Selby et al., 2019 for an overview). For instance, pain has been discussed as having a distractory function and thus helping to end negative cascades (Chapman et al., 2006), and pain offset following NSSI is theorized to negatively reinforce the behavior. However, it is unclear to what degree individuals tend to experience pain during NSSI. Pain may correlate substantially with the severity of the injury, but individual differences in pain thresholds, pain appraisal, chronification of NSSI, and even phenomena such as dissociation may also impact pain experience. As with severity, pain may be an important moderator of any psychological effects NSSI has and may be predictive of different patterns of NSSI in daily life. For instance, as demonstrated by Kranzler et al. (2018), pain could be a predictor of repetitive acts. Overall, the assessment of pain seems vital to further understand the phenomenon that is NSSI.

Implications from aim 2: Refining research on NSSI urges

NSSI urges are important to study because they may have negative consequences separate from those of NSSI acts (e.g. binding cognitive and self-regulatory resources) and because different qualities of the urge (e.g. intensity) may help predict which urges result in acts and thus inform preventive measures. This may be particularly meaningful for the development of mobile health interventions to address NSSI. However, as reviewed above, most studies only assessed urges in a dichotomous fashion (present/ absent), and did not assess urge intensity, duration, or context, which renders it difficult to learn more about the transition from urges to acts. We suggest that future studies should include a careful assessment of urge intensity on a continuous scale, for instance ranging from ‘no urge’ to ‘extreme urge’. Moreover, researchers should consider assessing the duration of an urge, as more persistent urges may be more likely to lead to NSSI and, perhaps, more severe NSSI. Additionally, assessing urges after NSSI occurs (e.g., via scheduled follow-up prompts programmed to occur at intervals following report of NSSI) would make it possible to track the course of urges and which context variables influence that course. For instance, if reduction in NA was the intended goal of NSSI but NA persists, urges may likewise persist after the act and then lead to further acts. As such, urges may be important in understanding repetitive NSSI or chains of several events within a day.

Overall, it is important to note that the reviewed studies observed substantially more urges than acts, and studies that assessed the association between both demonstrated that most urges did not ultimately result in acts. Thus, the evidence for urges reviewed herein primarily seems to comprise urges that were not acted upon and these may be systematically different from urges that do eventually result in acts. For instance, resisted urges may be of shorter duration or lower intensity, or occur in contexts where NSSI is not feasible, as the study by Turner, Cobb, et al. (2016) suggested. This again underlines a need to carefully characterize the phenomenology of urges and adds the important aspect of distinguishing between urges with and without subsequent acts. At the same time, this may also add valuable information on acts, as it would allow to further distinguish acts that were preceded by urges from those that were not.

Implication I from aim 3: Increasing the number of observed NSSI events

Aim 3 of the present review was to summarize the evidence on the Four-function Model of NSSI (Nock, 2009; Nock & Prinstein, 2004) and revealed that the daily life evidence for the model was largely inconclusive. This is in stark contrast to a multitude of reviews and meta-analyses finding substantial evidence at a cross-sectional level (e.g., Bentley et al., 2014) and in laboratory tasks (Ammerman et al., 2018). Therefore, more daily life research testing theoretical models of NSSI is warranted and the manner of NSSI assessment in AA and DD studies should be improved. Specifically, to further test theoretical models of NSSI in daily life, studies must capture a sufficiently large number of acts, as low base rates restrict the interpretability of results and may be behind some of the contradictory effects we found (e.g. due to low statistical power, chance of over- or underestimation of the association between NSSI and related constructs etc.).

The primary way to achieve this is to ensure a sampling of participants that show NSSI acts during the study period. As described above, many of the included studies were not successful at achieving this, as less than half of all included participants reported any NSSI act during the assessment timeframe. When looking in more detail at what distinguished studies that were more successful at recruiting participants that engaged in NSSI from those that were not, it becomes evident that the former tended to sample individuals with a recent history of NSSI acts or urges. When coding whether studies required recent (up to last 2 months) NSSI acts/urges or not and computing a point-biserial correlation with the prevalence of NSSI acts, this revealed medium sized correlations between having this inclusion criterion and the percentage of participants in a study that reported any acts (r = .46, df = 13, p = .083) and the average number of acts across the included participants (r = .51, df = 11, p = .075). With regard to urges, we observed a slightly different pattern, with a small and non-significant correlation between a recency inclusion criterion and the percentage of participants in a study that reported any urges (r = 0.26, df = 8, p = .472), but a medium sized correlation with number of urges across participants (r = 0.59, df = 10, p = .042). We assume this was likely the case because studies that assessed urges at all generally observed urges across a very high number of participants, on average in 70% of their sample and often more. Thus, there could have been a ceiling effect with regard to percentage of the sample that reported urges. At the same time, the effect of the recruiting strategy was substantially correlated with number of observed urges, suggesting that including participants with recent NSSI is also beneficial for observing a large number of urges. Again we would like to point out that these correlations are all based on a very small n and that we present them only to illustrate patterns and trends that we observed descriptively, without claiming that they provide well-powered inferential tests of the associations.

This suggests that imposing an NSSI recency criterion for recruitment likely increases the probability of including participants that actually show NSSI acts during the study period. However, as frequent NSSI is a relatively rare behavior among adults (Swannell et al., 2014), recruiting individuals with recent NSSI can be difficult. Based on epidemiological data, the probability of finding participants with recent NSSI should be higher in clinical than in general populations (e.g., Briere & Gil, 1998). Thus, we also assessed the association between NSSI prevalence and whether the sample was recruited through clinical or community settings (for details, see Table A.3). With regard to recruitment, we observed a medium-sized, though non-significant, correlation between recruitment in clinical settings and percentage of participants in a sample that reported any NSSI act during the study (r = .38, df = 13, p = .165). The association between clinical recruitment and average number of observed NSSI acts across all participants was significant and large (r = .60, df = 11, p = .031), as was the association between clinical recruitment and number of NSSI acts in participants with at least one act (r = .75, df = 12, p = .002). We note that these correlations are all based on a very small number of studies and should be interpreted with this in mind.

Thus, it appears likely that sampling participants with recent NSSI would increase the probability of sampling participants that show NSSI acts throughout the AA or DD study period, and sampling individuals through clinical settings would likely increase the probability of observing multiple acts per person. However, additional psychopathology could impact associations between NSSI and constructs of interest, such as affect. Additionally, sampling individuals with frequent and recent NSSI experience prevents examination of processes involved in the transition from first engagement in NSSI (i.e. voluntary action) to a chronic trajectory (i.e. habitual behavior) (Selby et al., 2019). Therefore, future studies should also consider the impact of NSSI history. Individuals with a longer NSSI history may potentially receive reduced benefits in terms of reinforcement because they have, in essence, developed tolerance. However, these individuals may also, as a result of repeated NSSI, have formed the strongest associations between, for instance, NA and NSSI. As a result, they may be even more likely to engage in NSSI following NA, despite experiencing smaller NA reductions.

A second way to increase the number of observed NSSI acts or urges is to extend the time of participation. This makes it possible to observe multiple acts or urges even when the sample consists of individuals who do not engage in NSSI regularly. While this option may seem more feasible than recruiting participants with very frequent and recent self-harm, it also increases costs, as well as participant burden and, thus, the probability of drop-out and reduced compliance.

In AA studies, a third way to increase the probability of sampling a high number of NSSI events is to include event-based prompts that participants self-initiate following acts or urges, which reduces the possibility of events going unreported. To achieve an even more fine-grained temporal resolution than a combination of random and event-based prompt would allow, researchers should further consider follow-up prompts. These can be scheduled in varying density around an NSSI act or urge (e.g. three follow-ups 10 minutes apart) and allow the monitoring of changes in constructs of interest such as intrapersonal and interpersonal variables that are theorized to be associated with NSSI. This would help address the relative dearth of research on the immediate consequences of NSSI and allow for a more conclusive test of theoretical models.

A possible concern of including follow-ups is that participants may be less likely to report NSSI acts, because they know that doing so will result in needing to answer additional prompts. While this is ultimately an empirical question, we hold it unlikely that the inclusion of follow-up prompts would lead to under-reporting of events or decreased compliance, as long as follow-ups are kept relatively short and are not overly frequent. NSSI urges and acts were still relatively rare events when considering each participant individually – as described above, participants on average reported 5 urges and 3 acts across the study period. Therefore, additional follow-up prompts would also occur rarely and we deem it unlikely that these would be aversive enough to keep participants from reporting NSSI events. Moreover, a recent study (which is still under review, but the authors have made a preprint available at https://psyarxiv.com/zf4nm/) found that compliance in AA designs tends to decrease with prompt length (i.e. the number of questions asked per prompt) rather than with sampling frequency. Including follow-up prompts would increase sampling frequency but not individual prompt length. Therefore, we would argue that brief follow-up prompts are likely a valuable and important addition to future AA studies that should not overly decrease compliance.

Implication II from aim 3: Test the interpersonal function of NSSI

While there was substantial investigation of the intrapersonal function (esp. negative reinforcement), there were only two studies that directly assessed the interpersonal function of NSSI in daily life (Snir et al., 2015; Turner, Cobb, et al., 2016). As 33–56% of individuals with NSSI report that they use NSSI for interpersonal functions at least some of the time in cross-sectional studies (Taylor et al., 2018), this points to a clear need for additional research. It is possible that this function has rarely been studied because it creates more methodological challenges than the assessment of affect in relation to NSSI. First, there appears to be a lack of existing inventories that capture a range of both negative and positive interpersonal events. The two studies reviewed herein only assessed rejection and conflict as negative events and assessed interpersonal support as the only positive event. Second, interpersonal events may be less proximal to the NSSI act itself than, for example, NA and therefore harder to capture, or the timeline may be less clear. For instance, a fight with a spouse during breakfast may incur NSSI urges that are sustained throughout the workday until NSSI is performed after the end of the workday. In this example, many would consider the interpersonal stressor the trigger for the NSSI event, but it would be more distal than the temporal effects of NA that is felt before the event. As such, interpersonal events may often even underlie NA that results in acts but existing research has not examined this. A detailed assessment of negative and positive interpersonal events and a frequent sampling of these could shed further light on the interpersonal functions of NSSI.

General summary and conclusions

This review provided an updated summary and synthesis of daily life studies on NSSI, including 35 AA and DD studies published until December 2019. First, we showed that few studies reported details on the phenomenology and daily life context of NSSI acts. Evidence on the methods used for NSSI pointed to skin picking, cutting, and hitting as the most frequently endorsed methods and limited evidence for the experience of pain suggested most participants felt some pain during NSSI, but at mild levels. Evidence on daily life contexts of NSSI was sparse and mixed, rendering it largely unclear whether NSSI is more likely to occur at certain times or in certain environments. Beyond assessing context, we recommended a more consistent assessment of the intended function, frequency, severity, and associated pain of NSSI in future studies as this could help further characterize the nature of NSSI in daily life and shed light on the association between NSSI and other constructs of interest (e.g. as affect), which could be moderated by properties of the act.

Second, we demonstrated that urges, too, are poorly characterized in daily life studies and little is known about their phenomenology and contexts they typically occur in. Therefore, we suggested an assessment of urge intensity, duration, and context in future studies. We concluded that there was compelling evidence that urges predict an increased risk for NSSI acts at the daily and momentary level, that urges temporally precede acts, and that urges are associated with elevated levels of NA.

Third, we aimed to review the daily life evidence for the Four-function model of NSSI (Nock, 2009; Nock & Prinstein, 2004). Based on 16 studies, we concluded that participants highly endorsed intrapersonal negative reinforcement as a motive for NSSI in daily life and that there was compelling evidence for elevated levels of NA prior to NSSI. The evidence for a reduction in NA following NSSI was mixed with four studies providing supporting evidence and three studies contradicting evidence. Intrapersonal positive reinforcement was tested in only five of the reviewed studies. Three studies found decreased levels of PA prior to NSSI but only two studies reported an increase of PA post NSSI, rendering the current evidence on intrapersonal positive reinforcement inconclusive. Likewise, the daily life evidence on interpersonal functions of NSSI is to be considered inconclusive as only two studies directly tested NSSI in relation to interpersonal variables. As discussed, this implies a clear need for future studies to examine more closely the interpersonal function of NSSI.

Beyond the various research implications we laid out in detail, the reviewed evidence also suggests a number of clinical implications. The evidence on the association between NSSI urges and acts suggests there is usually substantial ‘warning,’ in the form of urges, before an NSSI act takes place. This implies that, despite how individuals sometimes describe and experience NSSI, NSSI acts do not generally happen as an immediate impulse without warning. Rather, there seems to be a definite time window in which urges are present. This underlines the need to teach patients how to closely monitor their urges and apply intervention strategies, such as urge surfing. Additionally, this finding suggests the potential of mobile health interventions for NSSI, which could leverage the existence of this window by delivering an immediate intervention when patients report urges. Second, the reviewed evidence showed that urges and acts are both closely associated with NA. Thus, in addition to the discussed need for interventions that help manage urges, a focus on managing NA will remain central to treating NSSI. As most existing treatments that address NSSI, such as Dialectical Behavior Therapy (DBT, Linehan, 1993), already have a strong focus on affect regulation skills, this mostly underlines that treatments already well reflect the affective dynamic of NSSI. At the same time, the reviewed evidence also suggests that there may be a discrepancy between self-endorsed goals of NSSI (e.g., to reduce NA) and the actual effect NSSI acts have. It is possible that, due to rather long inter-prompt intervals, most studies failed to capture the immediate benefits of NSSI (e.g. NA reduction in the minutes after an act) and rather reflected medium-term effects (e.g. 1–2 hours after an act), showing that benefits are generally not sustained for long. While careful replication and further disentanglement of the timelines is needed, providing feedback to patients about how effective, or ineffective, they find NSSI acts, both immediately and in the longer term, may help increase motivation to cease the behavior. Overall, daily life studies, such as the ones reviewed herein, are a rich source of basic research on NSSI that may help inform and refine both traditional treatment approaches and ultimately open the door for mobile health interventions that are applied in the moment they are needed to prevent NSSI.

Highlights.

• We systematically reviewed 35 daily life studies on non-suicidal self-injury (NSSI)

• We summarize evidence on NSSI acts and urges and the four-function model of NSSI

• On average, studies observed 1.6 NSSI acts and 4.6 urges per participant over 14 days

• Supporting intrapersonal negative reinforcement, negative affect was high pre NSSI

• Evidence on intrapersonal positive and interpersonal reinforcement was inconclusive

Author Biographies

Dr. Johanna Hepp received her Doctorate from Heidelberg University in 2019 and is currently working as a postdoctoral researcher in the Clinic for Psychosomatics and Psychotherapeutic Medicine at the Central Institute of Mental Health in Mannheim, Germany.

Her full biography is accessible at https://www.researchgate.net/profile/Johanna_Hepp

Dr. Ryan W Carpenter received his PhD from the University of Missouri in 2018 and is currently working as a postdoctoral research fellow at the Center for Alcohol and Addiction Studies at Brown University, USA.

His full biography is accessible at https://www.researchgate.net/profile/Ryan_Carpenter

Lisa M. Störkel received her Master’s of Science from Heidelberg University in 2016 and is currently completing her Doctorate in the Clinic for Psychosomatics and Psychotherapeutic Medicine at the Central Institute of Mental Health in Mannheim, Germany.

Her full biography is accessible at https://www.researchgate.net/profile/Lisa_Stoerkel

Sara E. Schmitz received her Master’s of Science from Freiburg University in 2015 and is currently completing her Doctorate in the Clinic for Psychosomatics and Psychotherapeutic Medicine at the Central Institute of Mental Health in Mannheim, Germany.

Her full biography is accessible at https://www.researchgate.net/profile/Sara_Schmitz4

Prof. Dr. Christian Schmahl is the director of the Clinic for Psychosomatics and Psychotherapeutic Medicine and head of the research group on experimental psychopathology at the Central Institute of Mental Health in Mannheim, Germany.

His full biography is accessible at https://www.researchgate.net/profile/Christian_Schmahl

Dr. Inga Niedtfeld is head of the research group on emotion regulation and social cognition in the Clinic for Psychosomatics and Psychotherapeutic Medicine at the Central Institute of Mental Health in Mannheim, Germany. Her full biography is accessible at https://www.researchgate.net/profile/Inga_Niedtfeld