Self-Disclosure is a process of communication through which one person reveals themself to another [15]. It comprises everything an individual chooses to tell the other person about themself, making themself known to the other [16].

Being able to purposively self-disclose to another is widely recognised as an important component of a young person’s development. Some psychologists would say that self-disclosure serves not only as an indicator of a healthy personality, but also as a means by which one may be achieved [17]. Seminal research shows that young infants are capable of some management of self-disclosure, internalising their feelings whilst choosing to display something different to those around them [18]. Maturation improves self-monitoring skills enabling children to reflect on the consequence and purpose of disclosing aspects of themself in different contexts and to different people [19]. Self-disclosure is “intrinsically rewarding” [20] and associated with positive relationship development [21, 22]. In contrast, the intentional keeping of secrets is associated with both intra- and inter-personal costs, including stunted psychosocial development [23].

Self-disclosure is likely to include the sharing of feelings as well as personal ideas and thoughts. To function healthily, it is suggested a person needs at least one significant relationship in which they can disclose to a very significant degree [24]. Self-disclosure between peers and with trusted adults provides young people with foundational experiences that equip them with the skills to form appropriate levels of intimacy within their later relationships [25].

While self-disclosure is beneficial to healthy adolescent functioning, and particularly the initiation, maturation and sustaining of close relationships [26, 27], the ability to purposively choose to hold back from disclosing aspects of ourselves to others is also an important developmental goal of childhood and adolescence [19, 21]. Indeed, choosing when to disclose, what to disclose and to whom is critical in the formation of healthy relationships [25]. For instance, sharing overly personal information may lead to social rejection [21]. Sharing negative experiences and emotions can lead to emotional reactivation and prolong the negative state [27]. Within relationships, the reciprocal sharing of problems and negative feelings (“co-rumination”) can be associated with depressive symptoms [28].

High self-disclosure also has implications for personal safety, particularly online. Increasingly, adolescents use the internet both to interact with their peers and form relationships with people they have never met in person [29]. Dangers of inappropriate self-disclosure online include cyberbullying, harassment, and possible victimisation [30]. Children and adolescents who develop a strong bias towards high disclosure may lack the self-monitoring to avoid putting themselves, or others, at risk.

We trust our own qualities, skills, ideas and opinions when we place value upon who we are. We recognise ourselves as distinct and expect others to notice, value and respect our contribution in the world. In trusting our qualities, skills, ideas and opinions, we exhibit a favourable evaluation of ourselves. In questioning these qualities we exhibit a less favourable evaluation of ourselves; we see ourselves as less distinct, and are less expectant of being noticed and valued in the world.

Trust of Self is a model of relational self-definition rather than self-esteem. Whilst self-esteem describes an intrinsic sense of self-worth [31], self-definition is related to self-concept and describes the degree to which one sees oneself as distinct from others around you. Self-definition incorporates the concept of closedness and autonomy. To have a high Trust of Self is to be strongly self-defined, independent of, and closed to others’ evaluations. To have a low Trust of Self is to be weakly self-defined, dependent on, and open to others’ evaluations.

Though the two concepts are distinct, self-esteem can contribute to self-definition. Leary and colleagues provide evidence that children with higher self-esteem have an implicit assumption that they are relationally acceptable to others; when faced with critical feedback, these children are more likely to rebuff it [32]. Recent reviews and meta-analyses affirm that healthy self-esteem is longitudinally associated with positive relationship development [33, 34].

Within our bi-polar model of self-regulation, biases towards either high or low Trust of Self may be associated with poor social-emotional development. High Trust of Self may be informed by a child’s appropriate understanding of their qualities, skills, ideas and opinions, yet it can also be informed by an inflated, superior view of self. In the same way, a child’s low Trust of Self may be influenced by an inaccurate perception of their qualities, skills, ideas and opinions which diminishes their sense of value.

People with elevated views of themselves sometimes exhibit socially undesirable interpersonal behaviours such as interrupting and talking over others [35]. They can alienate others who see them as haughty, conceited or snobbish [36], and may exhibit narcissistic traits which contribute to behavioural problems [37] that are, in turn, associated with negative psychosocial outcomes [38].

Conversely, children with lower Trust of Self may have an implicit assumption that they are less relationally acceptable. This may manifest as greater rejection sensitivity, which is detrimental to psychological well-being [39]. When faced with critical feedback, these children may be more likely to assume some deficit on their part, potentially manifesting in a set of defensive behaviours which manage the perceived threat of rejection [40]. Children who anticipate rejection may adapt their behaviour to increase relational acceptance. As such, they might offer different and less stable social performances to different audiences [41].

A central component of children’s psychosocial functioning is the development of interpersonal trust [42]. When we trust others, we see them as supportive of our needs. We believe that the requests they make of us are of good intent and will benefit us, so we are responsive to their requests. By extension, children’s interpersonal trust facilitates positive relationships with peers and teachers [43] and is associated with school adjustment [44].

Attachment theorists suggest that the degree to which we trust or question others is shaped by our early attachments with caregivers which form working models of future relationships [18, 45]. However, these internal working models can change in response to impactful life events and aspects of the family dynamic including parent-child communication, family conflict, and parental separation [46-48].

Research indicates that a child’s trust of their caregiver to meet their needs has a significant impact on a child’s subsequent emotional self-regulation [48, 49]. Over time, effective emotional self-regulation gives children and adolescents a repertoire of skills including the ability to find healthy strategies to deal with uncomfortable feelings, exhibit impulse control, delay gratification, inhibit inappropriate responses in social situations, and resist distraction [50, 51, 43, 44].

However, children who have an implicit trust in the availability of others are not necessarily advantaged. If a parent leaps in to protect their child from conflict or distress, this over-involvement may deny their child the opportunity to develop self-confidence and leadership skills [52] and may contribute to the emergence of depression and anxiety [53]. If an adolescent becomes expectant of a teacher’s responsibility to instantly support them they are unlikely to develop the self-efficacy essential for high performance [50].

Low Trust of Others can also have deleterious effects. Children who experience their early caregivers as less emotionally available are likely to exhibit a reluctance to rely on others to acknowledge, support and regulate their emotional state [48]. They are more likely to develop a repertoire of self-reliant strategies, which may have a limiting impact on their social-emotional development [48, 54] and later self-regulation [50, 51].

Indeed, there is a body of research to support that both polar high and low levels of interpersonal trust are detrimental to psychosocial development [55]. Rotenberg and colleagues have demonstrated that, compared to children in the middle range of trust beliefs, those with very high or low levels of trust exhibit lower social skills, face greater peer rejection, experience more interpersonal aggression, and spend more time alone [56]. These findings support the “centralist approach to trust”, in which trusting “too much” or “too little” is associated with poor emotional and social functioning [57].

The task of managing the interplay between change and stability is something that we do every day from early childhood [58]. An infant uses a range of seeking behaviours to seek proximity from their caregiver, while also detaching from them to explore their world [59].

Healthy development requires change and risk, as well as stability and predictability. Teaching children and adolescents to optimally self-regulate their response to change and risk is crucial in supporting their mental health, social competencies and achieving good learning outcomes [60].

Children and adolescents with a fixed bias towards high Seeking Change will direct their attention towards what they do not yet have. Their drive for change may be fuelled by a desire for novel sensation seeking: to experience something new and exciting, or feel a sense of thrill or anticipation [61, 62]. Such young people may struggle to develop perseverance and stamina when tasks are challenging, or sustain attention and focus [63]. They may exhibit poor impulse control, and be less able to monitor and regulate their emotional responses [64], which contributes to poor school adjustment [65]. They may struggle to delay gratification with negative consequences; studies evidence that children and adolescents who struggle to delay gratification are more likely to use cigarettes, alcohol, and drugs and underperform academically [66, 67]. They may gravitate towards highly arousing or risky situations [68], something that the teenage brain is particularly predisposed towards, especially in the company of their peers [69].

In contrast, some children and adolescents exert their influence to reduce change and risk; they seek security, stability, predictability and consistency. In seeking to reduce change, a child or adolescent exerts control to make their world more predictable. Children who develop a fixed bias toward reducing change may withdraw from novel situation and may be timid, fearful, and shy with unfamiliar people [70], exhibiting a heightened risk of developing social anxiety [71, 72]. Such children may lack the resilience to cope with risk, change or uncertainty [73]. They may also be more prone to internal rumination and fixating on concerns, which ultimately will increase their levels of anxiety [74].

Students were first presented with the General assessment. To begin, students followed instructions on a digital dashboard which prompted them to create an imagined space in their minds. Students listened through headphones to the following script being read slowly with pauses to allow time for the subject to visualise their imagined space:

Imagine yourself in a place that you know, or a place that you can imagine. You can choose anywhere you like. Take a look around. What can you see? What can you hear? What can you smell? Reach out your hand and touch something. What does it feel like?

Stop for a moment and look around. Choose a part that you would like to keep for yourself. From now on we are going to call this YOUR SPACE.

What can you put around the edge of YOUR SPACE to show it is yours? Perhaps a rope, a fence, a line, a ditch, a wall or anything that you can think of. What is happening outside your space? What is happening inside your space?

Having generated this imagined space, the subject is then prompted with imagined scenarios by the dashboard. These scenarios were designed to trigger relevant self-regulatory responses related to the four tasks of the conceptual model. The structure of the prompts and the response options are in the form of a question followed by a six-item Likert scale. For example,

Would a visitor looking at your space know that it belongs to you?

(Definitely not / No / Not really / Maybe / Yes / Definitely)

Four items of such structure were developed to assess each of the four tasks, giving 16 items in all:

• Items for Trust of Self explore the distinction and definition of the subject’s imagined personal space from the space they imagine around it. For example, Is your space different to what is outside your space? and Something changes outside your space. Does your space change too?
• Items for Trust of Others explore the degree to which another person would be trusted in the subject’s imagined space. For example, Someone in your space asks you to do something, will you do it? and You need something in your space; do you want other people to help you get it?
• Items for Self-Disclosure explore to the degree to which the subject keeps aspects or parts of their space private from those who might enter. For example, Someone comes into your space; will you let them see what you are thinking? and Imagine you can choose to keep part of your space private; how much would you keep private?
• Items for Seeking Change explore the degree to which the subject is interested in and open to their space changing. For example, How often do you like things to change in your space? and Do you like trying things you haven’t done before in your space?

Having responded to the 16 General assessment items, students then listen to a second recording to prime them for the School assessment stage:

Now imagine that you are back in your space. Imagine that you can see those from your school in your space with you. What do you see those in your school doing? What do you see yourself doing in your space with your school? How are you feeling about your school in your space with you?

The subject was allowed to imagine whatever aspect or version of school they chose when primed by the visualisation recording. The environment of school was deliberately not prescribed more narrowly, for example, to a particular class or peer group. The aim of this was to obtain as subject-generated a response as possible. Students are then presented with the same 16 questions as in the General assessment, with item text modified to reference the context of school.

For example, in the first stage, one item reads: Would a visitor looking at your space know that it belongs to you? In the second stage, this item reads: Would people in your school know that this space belongs to you, just by looking at it?

Students’ 32 item responses for the two stages of Steer Tracking generate raw scores which are then computed to give a score from 0-15 for each of the four tasks, separately for the General and School assessment. Items are reversed in a random pattern across the assessment to reduce the effects of passive responding; the pattern differs slightly for the instrument used with 8-12 and 13-18 year students. In addition, algorithms check patterns of scores against likely manipulation or disengagement (for example, repeated scores across the assessment, unlikely sequences of scores or rapid speed). Where the algorithm detects such patterns, the software shows students a message inviting them to re-complete the assessment. Items are equally weighted. There are no fake items.

The degree to which a score deviates from the numerically median factor score of 7.5 indicates that the student self-regulates that task either higher (7.5-15) or lower (0-7.5). This deviation reflects the degree of bias a student is exhibiting in that task. The comparative difference in General versus School scores can be attributed to the impact of the inserted priming context, providing a measure of the impact the context of School on a student’s self-regulation (Fig. 2).

Students, for example, who obtain a score of 4.5-11 will have answered most items with maybe/sometimes. Their responses indicate that how they manage that task would be contingent on the situation. According to the conceptual model, these students show an ability to flexibly self-regulate; they pay attention to the situation they are in and adapt or adjust their tasks according to the needs of that context.

In contrast, students who obtain a score of 0-3 or 12-15 will have answered the items with definitely/ definitely not. Their responses indicate they would always tend to manage that task in the same way, regardless of the situation. These students, according to the conceptual model, show less ability to self-regulate – are less likely to adapt or adjust their tasks according to the needs of that context.

Low bias scores of 0-3 and high bias scores of 12-15 were determined as cut-offs for flagging as ‘polar’ on the basis of statistical and practical considerations. Statistically, across pilot samples, these represented roughly the 85^th percentile. Practically, teachers need to prioritise a manageable percentage of students, and feedback supports this proportion to flag. Students flagged as having polar biases in more than two factors were flagged as ‘highest priority’. In addition, the tool ranks numerical discrepancy between School and Generalised scores; this enables teachers to locate the context of risk. For example, polar biases exhibited in School but not General scores would indicate that the context of school was a source of dysregulation. By contrast, polar biases in the General assessment but not in School scores would signify concerns relating to domestic or personal factors.

The teacher interface (Fig. 3) presents data in a simple row-by-row array, using numbers, colours and icons to indicate students of priority concern. Options to rank, sort and filter students optimise the identification process. Interactive student names allow teachers to open further sub-screens to expand a particular student’s data and obtain verbal explanations of associated risks. From these screens, teachers can complete an action planning process by which they select suggested school actions which may be put in place to better support that particular student.

Fig. (2). Illustration of how a subject’s factor scores may adjust from their generalised self-regulation to in-school self-regulation.

Fig. (3). Example illustration of the Steer Tracking teacher interface.

Data were collected from 2,829 students across three primary and six secondary schools in England. Of these, five schools were fee-paying and four were state funded. Of the fee-paying schools, three were day schools and two were mixed day and boarding students. The sample included 991 girls and 1180 boys aged 8 to 12 years old, and 167 girls and 491 boys aged 13 to 18 years old. Students were not asked to disclose further demographic information (e.g., ethnicity or socioeconomic status).

Data collection took place in school classrooms during 2013 and 2014, under the supervision of teachers. Students completed the Steer Tracking General and School assessments. Schools were approached to participate in the study directly by the authors. Schools were selected on their willingness to participant in research as part of a process to develop a new tool to support student social-emotional development. Schools and students were not paid to participate, and Steer Tracking was provided free of charge. Schools received a study report providing wellbeing conclusions at a population level for their school and had the option to view assessment data at an individual student level.

Schools signed a data sharing agreement through which the school gave consent as a surrogate for minors (article 16 Declaration of Helsinki). The data were collected pre-GDPR. Schools used their own policies to determine how to communicate participation with parents. Students were given the choice to opt out of the assessments and if so, were provided with alternative activities by their teachers. Students understood that their responses would be visible at an individual student level to the school.

Students completed the computerized Steer Tracking School and General assessments, as described in section 1.4.

Data were analysed using Xlstat to confirm invariance and descriptive statistics of score distributions by age and gender. Xlstat was also used to extract Promax Rotated Factor Loadings for the four latent factors.

Descriptive statistics are presented in Table 1. First and second standard deviations for the two main age groups (ages 8 to 12 and ages 13 to 18) are provided for assessment scores. Distributions behave within normal parameters, with approximately 95% of scores falling within 2 standard deviations of the mean. Mean and median scores are also approximately equal. It is appropriate to use mean and first and second standard deviations as representations of spread to provide norms for age and gender. Distribution norms show variance by each age, gender and assessment type. This indicates that developmental, gender and ecological factors may contribute to a student’s self-regulation of the four tasks.

Within the Age 8-12 group, 46.8% of the variance was accounted for by four latent factors, each of which showed an eigenvalue of >2 (Table 2). The remaining 53% was accounted for by more than 16 latent factors (Fig. 4). Factor loadings for items across the four principle latent factors show relatively strong associations with the conceptual constructs of the data model: Trust of Self, Trust of Others, Self-Disclosure and Seeking Change. Factor loadings for items of Trust of Others and Self-Disclosure were strongest overall, but also exhibited some cross-factor loading (Table 2). Overall factor divergence was moderate, indicating that some items could be refined to improve factor discrimination (Fig. 5).

Fig. (4). Scree plots of percentage variance explained by principal components for (a) students aged 8-12 and (b) students aged 13-18.

Within the Age 13-18 group, 42.4% of the variance was accounted for by four latent factors, each of which showed an eigenvalue of >2 (Table 3). In other characteristics, factor loadings reflected similar patterns to the 8-12 age group (Table 3), with moderate factor divergence (Fig. 5).

Fig. (5). Intercorrelation between items (a) students aged 8-12 and (b) students aged 13-18.

Data were collected from 2518 students aged 12 to 18 years old across 16 secondary schools in England. Of these, fourteen were fee-paying schools and two were state-funded. Of the fee-paying schools, six were day schools, and eight were mixed day and boarding students. The sample included 48% girls and 52% boys. Students were not asked to disclose further demographic information (e.g., ethnicity or socioeconomic status).

Data collection took place in school classrooms between November and December 2014, under the supervision of teachers. Schools were invited to participate in the study directly by the authors. Schools were selected on the basis of an existing interest in research, as indicated by participation in previous studies. The range of types of school (fee-paying, day and boarding) reflected a market mix in the UK. Schools and students were not paid to participate, and Steer Tracking was provided free of charge. Schools were offered a study report and were invited to attend a webinar explaining over-arching wellbeing conclusions at the school level. Schools chose to participate based on their understanding of the potential advances the assessment might make in supporting students’ social-emotional development.

Schools signed a data sharing agreement through which the school gave consent as a surrogate for minors (article 16 Declaration of Helsinki). The data was collected pre-GDPR. Schools used their own policies to determine how to communicate participation with parents. Students were given the choice to opt out of the assessments and if so, were provided with alternative activities by their teachers.

Students completed the Steer Tracking General and School assessments. Students were informed that their Steer Tracking assessments may be seen by their teachers as well as researchers. Students were then asked to respond to three questions about their wellbeing. Students understood that their responses to these questions would be strictly anonymous and would not be seen by their school. Students understood that they would not have the opportunity to view their personal assessment results.

In addition to the Steer Tracking General and School assessments described in section 1.4, students responded to three questions about their wellbeing which were written by the researchers for the purpose of this study.

• Bullying: Have you been bullied in the past year? (Never / No / Sometimes / Yes / Often)
• Self-harm: Have you self-harmed or thought about self-harming in the past year? (Never / No / Sometimes / Yes / Often)
• School Pressure: How able are you to manage the pressure of work at school? (Well / Quite well / Ok / Not very well / Not well)

Theoretically, we did not expect any single factor to be associated with pressure/ bullying/ self-harm. Real-world use of the tool suggested that patterns across the factors would be more likely to show specific associations to these risks.

A Support Vector Machine (SVM) model was therefore used as a method of testing the fit of many models of the combined four factors to students experiencing bullying, thinking about or engaging in self-harm, or struggling with school pressure. SVM is a supervised machine learning model that analyses data using learning algorithms to solve classification and regression tasks.

To begin, binary variables were generated based on responses to the three wellbeing questions:

• Bullying: 0 = Never, No, Sometimes 1 = Yes, Often
• Self-harm: 0 = Never, No 1 = Sometimes, Yes, Often
• School Pressure: 0 = Well, Quite well 1 = Ok, Not very well, Not well

The model was first trained on the dataset with a Radial Basis Function (RBF) kernel to account for non-linear data (i.e. both high and low Steer Tracking scores are expected to be associated with bullying, self-harm, and school pressure). The parameters of the model were determined using 10-fold cross-validation. The SVM model was then applied to the non-training data.

The model was cross-validated to test for the probability of accurately classifying students who belong to either category for each wellbeing variable (e.g., bullying/no bullying) based on STEER Tracking factor scores; consistency between the two measures indicates a low chance of false readings.

The model achieved an 80% accuracy (83/78% both classes) in identifying students who had experienced bullying in the past year. Experiencing bullying was associated with a significantly higher level of poor self-regulation and specifically with low Self-Disclosure. Students who had a bias towards low Self-Disclosure, as well as a high degree of dysregulated bias (a high deviation from the mean across their item scores), were more likely to report bullying.

The model achieved an 80% (82/78% both classes) accuracy in identifying children who endorsed self-harm thoughts or behaviours. Self-harm was associated significantly with overall poor self-regulation as well as with low Self-Disclosure and high Seeking Change. Students who show a combination of these factors (high Seeking Change, low Self-Disclosure and high dysregulation) were the most likely to report self-harm.

The model achieved an 83% (88/77% both classes) accuracy in identifying children who were not coping well with pressure at school. Struggling to cope with pressure was statistically associated with overall poor self-regulation as well as, specifically, with Self-Disclosure and Seeking Change. Students who displayed a bias toward low Self-Disclosure or low Seeking Change, or a high degree of dysregulated bias (a high deviation from the mean across their item scores) reported that they managed less well with the pressure experienced at school.

Cross-validation checks compared the accuracy of both exhibited and non-exhibited conditions (e.g., bullying/no bullying). Across the three wellbeing variables, the difference between conditions was small (bullying 83/78, self-harm 82/78), though slightly higher for school pressure (88/77) with both positive and negative predictions showing high accuracy, thus largely eliminating the explanation of false positive/negative readings. Thus, the assessment exhibited power 8 times greater than statistical chance of correctly identifying students who are experiencing bullying, self-harm, or struggling to cope with school pressure.