Introduction

Adolescents have a differential tolerance to risk and ambiguity than both children and adults simply because this is their way of learning (Levy et al., 2012). Scientists have predicted that risk-taking behavior is necessary to help facilitate the adolescent’s social move toward independence, this evolutionary phenomenon occurring in young persons around the age of puberty (Steinberg et al., 2008). Increased reward-seeking behavior renders adolescents more prone to risky decision-making relative to their younger peers, with engagement in higher-risk activities where the outcomes are often negative (Albert et al., 2013; Mulye et al., 2009; Pharo et al., 2011). Extensive research on social (Guyer et al., 2016; Tomé et al., 2012), emotional, and cognitive factors (Steinberg, 2005) associated with risky decision-making in young people suggest that these individuals make suboptimal choices because of critical neurodevelopmental changes taking place, specifically the prefrontal cortex and its interconnections which are slow to mature (Luciana, 2013). Furthermore, young people are also extremely vulnerable to mood disorders such as anxiety and depression during adolescence, factors which further influence their attitudes towards risk, sensation seeking, and impulsivity (Hickie et al., 2013; Steinberg, 2008, 2010, 2004; Steinberg and Cauffman, 1996; Steinberg and Morris, 2001; Steinberg et al., 2008, 2009). Contemporary research demonstrates that prolonged mood states, such as anxiety, have been shown to correlate with impaired decision-making (Alvares et al., 2014; Caplin and Leahy, 2001; Charpentier et al., 2017; Harle et al., 2017; Larquet et al., 2010; Mukherjee and Kable, 2014; Scott et al., 2014; Weinrabe et al., 2020; Weinrabe and Hickie, 2021; Wu, 1999). Young individuals with severe mood disorders have even greater neuropsychological impairment, impacting brain regions that in turn affect decision-making (Hermens et al., 2015, 2013).

To better understand youth decision-making, it is critical to evaluate subjective attitudes toward risk and ambiguity. Many studies have evaluated risk in healthy adolescents (Blankenstein et al., 2016; Tymula et al., 2012; Van Den Bos and Hertwig, 2017; Van Duijvenvoorde et al., 2016), but few studies have evaluated risk and ambiguity using economic decision-making tasks in clinical populations—especially youth with emerging mood disorders. Research comparing healthy adults to healthy adolescents under fMRI whilst completing a monetary task that evaluated responses to gains and losses revealed that an adolescents’ brain activation patterns showed a reduction in the mesolimbic circuitry (right ventral striatum and right amygdala) when compared to adults in response to anticipating gains but found no difference between adult brains during reward notification (Blakemore, 2018). From this and other clinical research we know that the adolescent’s ability to navigate the complexity of the decision-making process itself is problematic (Steinberg, 2010). When evaluating specifically risk and ambiguity in the context of severe mental disorders, it was found that decision-making was severely impaired, (Cisler et al., 2019, 2023; Fujino et al., 2016, 2017; Sonuga‐Barke et al., 2016) and anxiety-disordered patients were more risk averse than non-clinical controls (Charpentier et al., 2017).

In the current study, to evaluate risk and ambiguity attitudes in young people with emerging mood disorders, we used decision-making tasks widely applied in economics research (Andersen et al., 2008, 2014; Andersen and Teicher, 2008; Cox and Harrison, 2008; Ellsberg, 1961; Holt and Laury, 2002; Tymula et al., 2012; Weger and Sandi, 2018). Risk is defined as the willingness to accept offers and gambles when the person knows the precise odds of each outcome (Knight, 2012). The difference between risk and ambiguity stems from how much information is available at the time the decision occurs (Glimcher, 2011; Levy et al., 2012). We hypothesize that participants experiencing anxiety, especially heightened trait-like anxiety, would avoid risky decision-making, making our anxious participants more risk averse; and those who rated high on depression scores, would be more risk tolerant as compared to those experiencing anxiety. In light of recent studies, there is a need to understand more clearly other impact factors (beyond mainly age) that influence the young decision-maker.

Method

Participants

This study had a total of 30 participants (19 males, 11 females, mean age = 19.2 years, age range = 16–25 years, standard deviation = 2.23 years), recruited from Youth Services Clinics across two regions in New South Wales (NSW), Australia. These clinics were selected to ensure an accurate representation of youth in the region’s varied socio-economic and cross-cultural population. Recruitable subjects were evaluated by the on-site clinician at the Youth Services Clinics to qualify and participate in the experiment. Criteria for inclusion in this study were: (i) aged between 16 and 25 years, (ii) seeking professional help primarily for a depressive (unipolar or bipolar) syndrome, (iii) sufficient fluency in the English language to complete the decision-task, (iv) no history of the neurobiological disease (e.g., head trauma), (v) lack of any intellectual and/or developmental disability, (vi) no allergies or dietary sensitivities, (vii) abstaining from drug and alcohol use for 48 h prior to the appointment, and (viii) willingness to participate in two experimental sessions: on the day they signed up for the study and approximately 6–8 weeks later. Excluded from the study were those participants insufficiently fluent in the English language to be able to partake in the economic tasks or psychological assessment, had a known history of neurobiological disease (e.g., head trauma), disability (intellectual and/or developmental), or those with allergies or dietary sensitivities. Study investigators received clinician-referred potential participant information using the research study’s University’s Human Research Ethics Committee approved referral letter, which contained de-identified personal information, no information on proposed treatment for each person, and their mood scores for this visit to the clinic. Investigators then issued potential participants with a Participants Information Statement providing clear information about the study content, its aims, and objectives. Upon agreeing to participate in the Youth Choice Study, a Participant’s Consent Form was issued, and consent was obtained from each study participant. The University of Sydney’s Human Research Ethics Committee approved the study, and the methods and data confidentiality were carried out in accordance with the relevant guidelines and regulations. All data was de-identified during statistical analysis. Due to the strict medical privacy policy held by the clinics where the study was conducted (as determined by the Australian Government), study investigators were not kept informed whether patients received treatment, or what kind of interventions were being used (pharmacological or otherwise).

Table 1 presents the demographic information of the study participants. 30 participants completed Session 1 and 25 completed both sessions. The attrition rate (N = 5 at Session 2) is expected for the study because patients visiting the clinic may be too ill to return for their subsequent visit to the mental healthcare clinic. Unreported analysis indicates that subjects did not vary significantly from the rest of the group completing Session 2 (available upon request).

Table 1 Demographic information.
Full size table

Procedure

Clinical factors. Three well-known questionnaires were applied: Kessler’s Psychological Distress Scale (10-item) (K10) (Kessler et al., 2003), Quick Inventory of Depressive Symptomatology Adolescent (17-item) (QIDS-A17) (Rush et al., 2003) specifically designed for youth (Bernstein et al., 2010), and The Somatic and Psychological Health Report (12-item) (SPHERE-12) questionnaire (Hickie et al., 2001). At Session 1, and before the decision-making component of the study started, participants met with their clinicians. At Session 2, 6–8 weeks later, participants met directly with the researcher, who was uninformed whether participants were receiving clinical treatment on the day.

K10 assessed participants’ severity of distress on the day of the study. Scores were between 0–50, where 10–19 means ‘Likely to be well’, 20–24 ‘Likely to have a mild disorder’, 25–29 ‘Likely to have a moderate disorder’, and 30–35 ‘Likely to have a severe disorder’. When rated >30 a severe mental disorder is present (Scott et al., 2013)

QIDS-A17 evaluated how severe the study participant’s depressive symptoms were on the day they arrived for testing and included symptoms present for the previous seven days. The QIDS-A17 total score can vary between 0–27, where 0–5 means ‘Not Depressed’, 6–10 ‘Mild’, 11–15 ‘Moderate’, 16–20’ Severe’, and >21 ‘Very Severe’.

SPHERE-12 measuring physiological and psychological distress was divided into two parts: 6 PSYCH items and 6 SOMA items and participants scored themselves on how distressed they felt across those two primary areas in the past weeks. PSYCH refers to psychological distress, and SOMA refers to physiological distress. When scoring positive on PSYCH and/or SOMA, it will be rated as being either Level 1 (Type 1), reflecting participants with the most symptoms present; Level 2 (Type 2) where participants report on the PSYCH subscale or psychological symptoms only; or Level 2 (Type 3) where participants report on the SOMA subscale or somatic symptoms only; and lastly, “No Symptoms”, where participants score neither a psychosocial nor somatic symptomatology and is not rated as having any symptoms to support a diagnosis.

Decision-making. We assessed each person’s attitude towards risk and ambiguity using a robust and widely used experimental task, well-known to evaluate subjective risk preferences in economic studies. At both time points in the study, participants accessed a computer where they received written instructions and practical training for the task, prior to the researcher starting the task.

The overall decision task consisted of 60 choices that tested the participant’s preference for risk and ambiguity. The choices that made up the overall task consisted of thirty separate questions and involved the participant making a choice between one of two monetary options: an immediate payout amount of $5, or a lottery option that might pay more than $5, but potentially no payout. This experiment using the Ellsberg Paradox ensured probabilities were always clear in the risky task (using two distinct colors), whereas for the ambiguous task, an occluder purposefully covered the outcome of probabilities making the choices obscured (Ellsberg, 1961). For each choice, participants were presented with purposefully randomized options to evaluate probabilities and payoffs for the chances of winning (between $5 and $41). The ambiguity of choices factored in a 50% probability of winning for each of the three different chances of winning, but this was occluded and not known by the participants. The choices were randomized by the software each time presented online.

After the tasks were completed the software program randomly picked a trial from the 60 questions in total. The subject was instructed to request that the researcher determine whether the payout was to be paid on the day of a lottery trial. If the payout was a for sure amount, participants received the payout on the day. If the subject picked a lottery, the participant chose from an actual container with two colored props, using the color of the prop to be allocated that amount. Importantly, it is known that if payments are hypothetical and not realized on the day of the study, people will not make authentic choices (Holt and Laury, 2002). If the software detected that a participant kept making the same choices, their decision-making data would not be included as these are known by economists to not be an accurate representation (Konovalov and Krajbich, 2019; Tymula et al., 2012). The tasks were repeated 6–8 weeks later, with final layouts realized at the end of the session.

Participants each took home approximately AU$17–AU$18 on average for the whole experiment. The experiment took no more than 45 min to complete at each time point (90 min in total). The study used E. prime2 Professional (Schneider and Zuccolotto, 2012) computer software to capture all decision-making tasks.

Analytic Approach Measuring Economic Preferences. We applied a well-known economic model that is used to calculate risk and ambiguity tolerances (Levy et al., 2012). We use the choices from the participants, with approaches from previous economic studies (Levy et al., 2010; Tymula et al., 2012), to show how the utility function is effectively applied to evaluate risk and ambiguity tolerances within subjects:

$$U\left( {p,A,v} \right) = \left( {p - \beta * \frac{A}{2}} \right) * v^\alpha$$

“where v is the dollar amount, p is the winning probability, A is the ambiguity level, α is a measure of the risk attitude, and β is a measure of ambiguity attitude” (Tymula et al., 2012). Practically, this expected utility prediction experiment measures risk and ambiguity by using a count of the number of risky and ambiguous choices each participant made in the experiment. To estimate these choices accurately, we first calculated the times each person decided on the lottery amount instead of the “for certain” amount available at each risk level in our experiment. We then estimated the effect of ambiguity on perceived winning probability on each person’s mood score.

For our statistical analyses, we then had to approximate how risk or ambiguity-tolerant each person was during the study. When calculating risk, an individual can range between the following: risk neutral (α = 1); preferred a risky choice to a safe one (α > 1); preferred to avoid risk (α < 1). When calculating ambiguity (β) those who avoid ambiguity would show β > 0, those preferring ambiguity (β < 0), and those who neither preferred nor avoided ambiguity (β = 0) Fig. 1.

Fig. 1: Economic preference curves.
figure 1

Graphs display within subject choice behavior for risk and ambiguity options. Both choices display the pooled sample’s preference for choosing the lottery option over the for sure amount, and the exact amount of reward they prefer (various % probability options).

Full size image

Statistical analysis

We then used multiple regression analysis, across both Session 1 and Session 2. The outcome of the preference proxy (e.g., risk or ambiguity tolerance) is standardized to a mean of 0 and a standard deviation of 1. The mood states are either the continuous proxy of prolonged mood states (e.g., K10, QIDS-A17, or SPHERE-12) or as a binary measure (0,1), which indicates clinically relevant health problems. The control variables are age and gender. For study replication purposes, a well-known economic model (Schurer, 2015) was applied:

$${{{y}}}_{{{\mathrm{i}}}} = \alpha _1 + \alpha _2{{{\mathrm{Mood}}}}\,{{{\mathrm{State}}}}_{{{\mathrm{i}}}} + \alpha _3{{{\mathrm{Age}}}}_{{{\mathrm{i}}}} + \alpha _4{{{\mathrm{Female}}}}_{{{\mathrm{i}}}} + \epsilon _{{{\mathrm{i}}}}$$

Here, yi is the outcome of interest: either “risk tolerance” or “ambiguity tolerance.” The subscript i refers to each individual participant; Moodi is a proxy for the mood state of interest (K10, or QIDS-A17 or SPHERE-12); Agei is a continuous measure of age, Femalei is a binary indicator for whether the participant is female; and ϵi is the error term. Standard errors are estimated using the Huber/White method which is robust to non-constant variance. Parameters that need an estimation are defined by the Greek letters: α1 measures the mean outcome, if all other values of mood state, age, and female, were to be set to zero. Therefore, α2 measures the difference in Y for a unit increase in the mood measures, α3 measures the difference in Y for each additional year of age or “age gradient” in preferences, and α4 is the difference in outcomes between females and males, or “sex gradient” in preferences. α2 is the main parameter of interest in this model. Statistical analyses were performed using STATA/MP Version 14.2. All tests evaluated groups at both sessions.

Experiment timeline and demographics

Participants first completed their psychological questionnaires and decision-making tasks and then the demographic questionnaire. All this information was captured before they received their payment. Participants also filled out demographic questionnaires to evaluate any correlations to age, gender, education level, father’s, mother’s/guardian’s education (used as a proxy for this group), postcode, and self-assessed wealth (ratings will determine where the participants state they sit on a scale from rich to very poor). Participants filled in psychological questionnaires upon arrival and thereafter completed the experiment’s decision-making component. Participants were asked to participate in two sessions, Session 1 (after meeting with their clinician) and the same Session 2 again around six to eight weeks later. At no time, due to strict ethics protocol, were the investigators informed whether participants were receiving clinical treatment, and if so, what kind of treatment was being applied, even though study participants had been recruited for this 6–8-week period.

Results

Clinical profile

We conducted data analyses of 30 patients in Session 1 and calculated that 25 participants were rated as experiencing a mental health condition. In total, 15 (50%) participants scored high across all three psychological questionnaires combined: “severely” anxious (K10 ≥ 30), “severely” depressed (QIDS-A17 ≥ 16), and “Level 1 (Type 1)” for SPHERE-12 (PSYCH ≥ 2 and SOMA ≥ 3). Based on psychological scores, we estimated that 5 (17%) participants self-rated as having no condition.

In Session 2, taking attrition into account (n = 25), results showed that 20 participants rated as having a mental health condition, with 5 (20%) self-rated as having no condition. Across all three mood questionnaires, 11 (44%) participants rated: “severely” depressed (QIDS-A17 ≥ 16), severely” anxious (K10 ≥ 30), and “Level 1 (Type 1)” for SPHERE-12 (see Table 2).

Table 2 Mood scores at Sessions 1 and 2.
Full size table

When analyzing the battery of psychological scores separately for K10, QIDS-A17, and SPHERE-12 it was estimated that at baseline, 1 participant was rated highly on K10 only, no others were scored high on QIDS-A17 only and 2 participants rated as “Level 1 (Type 1)” for SPHERE-12 only. When evaluating scores 6–8 weeks later, one change occurred on the QIDS-A17 only scores, with the addition of 1 participant who rated as “severely” depressed (QIDS-A17 ≥ 16). However, no changes occurred on the K10 and SPHERE-12 scores (see Table 2). These results suggest that in general, most participants experienced dysregulated emotion for the duration of our study.

Decision-making

Individual level

We found a positive correlation between risk (ρ = 0.68) and ambiguity (ρ = 0.64) scores across sessions. In other words, if someone makes an ambiguous choice, they are more likely to make a riskier choice as well. In Session 1, participants had a mean of approximately 14 risky choices (SD 4.6), and a mean of approximately 11 ambiguous choices (SD = 7.6). In Session 2, participants’ mean increased to approximately 13 ambiguous choices (SD = 8.5), but risk increased by 1–15 choices (SD = 6.5). This correlation implies that the measure has stability and validity as it indicates that decision-making remained consistent over the 6–8 weeks study period.

Group-level

In Fig. 2, we illustrate the percentage of times that this group of young people chose risky and ambiguous lottery choices over the two-time points. Figure 2 shows that overall, this population chose riskier, and specifically more ambiguous lotteries during Session 2, than when first arriving for mental health care and participating in this study. In other words, most participants avoided taking any kind of risk the first time they completed the experiment compared to the second time. We also found that on average, when making risky choices, the group chose the higher winning probability (75%) more often than the lower winning probabilities. This suggests that the participants distinguished between riskier choices in a consistent and monotonic sense, where they chose more risky bets that had a higher probability of success. However, this does not appear to be the case for ambiguous choices, where participants had similar ambiguity attitudes for lotteries that were the highest and lowest probabilities.

Fig. 2: Risk and ambiguity attitudes in Sessions 1 and 2.
figure 2

In the left figure, we show the percentage of times the clinical group chose a lottery, reflected by its winning probability (25%, 50%, or 75%), preferred to a for sure offer ($5) in Session 1 (horizontal axis) and Session 2 (vertical axis). In the right figure, the group’s average ambiguity attitude is compared at the same three winning probabilities between sessions. Group differences between risk and ambiguity attitudes for N = 30 at Session 1 and N = 25 at Session 2.

Full size image

Fitting the expected utility model to the choices from participants, we estimated the pooled risk (α = 0.55, SE = 0.05) and ambiguity (β = 0.25, SE = 0.04) parameters of the model, suggesting that our participants were overall risk-averse and preferred ambiguity (see Supplementary Material for modeled choices for each individual).

Group-level regression analysis: risk tolerance and mood

Table 3 presents our estimation of results applying a multiple regression analysis for observations across both Session 1 (N = 30) and Session 2 (N = 25). Panels A1 and A2 report the results for risk tolerance using a continuous measure for each mood score in Sessions 1 and 2, respectively. Presented are estimated coefficients, obtained from the model described above. Although the estimated coefficients are like zero in a statistical sense, interesting qualitative patterns emerge, mainly that this economic model at a group level suggests that participants behaved averse to risk. To evaluate this more closely, we first test the association between risky choices and mood state applying a regression analysis. Qualitatively, we found the group’s K10 results maintain their sign across Sessions 1 and 2, which suggests that anxiety lowers risky decision-making. In other words, the group’s scores demonstrate that when anxiety is present, there is risk-avoidant behavior. This result is presented in the data when evaluating a pooled sample, results that are directionally consistent with the economic model of risk aversion (see Supplementary Material).

Table 3 Predictors of risk tolerance, for different proxies of mood states.
Full size table

Group-level regression analysis: evaluating ambiguity tolerance (see Supplementary Material)

We applied the same multiple regression analysis as above, using a sample of observations across both Session 1 (N = 30) and Session 2 (N = 25), evaluating ambiguity and mood for our sample of study participants. Overall, we find no statistically significant relationship between mood and ambiguity tolerance when using conventional levels of significance (α = 0.05) across both Sessions 1 and 2. However, we find some evidence in Session 1 of a direct association between depression scores and ambiguous choices, but they are not statistically significant in Session 2. Furthermore, we note that overall, there is more ambiguous decision-making in Session 2, yet not visible in the regressions. This suggests that either some learning process emerges, or other factors influence how people choose during the experiment that increases ambiguous choices, whilst making it independent of depression (see the “Discussion” section).

Discussion

Youth research suggests that healthy adolescents, more so than healthy young adults, are risk-seeking (Gullone et al., 2000) and that a range of factors directly contribute to this behavior (Pfeifer et al., 2011). Economic decision-making studies, not controlling for mood, demonstrate an alternative finding, that adolescents are more averse to risk than adults, when ambiguity is present, making them more risk-tolerant in these situations (Tymula et al., 2012) Nevertheless, it is the negative consequences of risky decisions, that are problematic for the young person (Eaton et al., 2006). For ill populations experiencing psychopathologies, negative decision-making is due to alternations in fear, learning, and alternation neurocircuitry (Harle et al., 2017; Hartley and Phelps, 2012). Research studies reporting on the correlation between dysregulated emotions (anxiety and depression), and impaired economic decision-making found that a positive relationship existed between these key variables (Weinrabe and Hickie, 2021). For example, when working with a clinically depressed population this study found that patients accepted more unfair offers more of the time than did healthy control groups (Harle et al., 2010). Another more recent economic study, although not reporting on the causes of the symptoms of mood disorders, claims that adolescents who suffer from severe anxiety symptoms systematically violate principles of normative decision-making (Weinrabe et al., 2020).

Few empirical studies address emotional dysregulation in a natural setting, depending on mood-changing laboratory experiments to bring about certain emotional states. Importantly, negative emotional states do not impact decision-making in the same way as prolonged mood states—such as trait anxiety and depression. Trait anxiety, a serious influencer of stress-induced depression is known to negatively influence a person, leading to behavioral and physiological changes (Weger and Sandi, 2018). Our study addressed how risk tolerant a within-subject level group that was experiencing prolonged mood, as validated by a referring clinician and according to self-report mood scores. Study participants were also seeking professional care for an emerging mood disorder. Our sample’s risk attitudes are like studies in the economic decision literature that suggest—unhealthy youth populations are averse to risk as measured by the same or similar experiments used in this study (Hartley and Phelps, 2012; Maner et al., 2007). Although taken from a small sample size, our results suggest that overall, those young people self-rated as experiencing an emerging mood disorder, specifically trait anxiety, were overall more risk averse in their decision-making. Our results also show that young people in this age group who take risks are more prone to making ambiguous choices, or ambiguity tolerant. We consider that our results show no statistically significant relationship between our other key variables, for example, depression scores and risk. This highlights that other factors influencing the young person, either on the day of the study or overall, are affecting our results. Authors suggest that although there is tremendous value in this type of research (the clinical setting), there are multiple challenges to rigorously evaluate mood disorders, specifically anxiety and its impact on decision-making, using economic paradigms (Murphy et al., 2001; Paulus and Yu, 2012). These authors suggest that multiple external factors impact the patients before or during treatment, confounding the effect on the study results (Paulus and Yu, 2012). We consider that these and other factors are playing a role, making it more challenging to answer the question in this specific study. Are these findings generalizable beyond monetary gambles? Are economic tasks sensitive to people not only with mood disorders but especially so for youth at a certain stage of brain development?

It could be argued that in addition to a young person experiencing anxiety leading to negative psychological bias, another factor has a major impact on the person’s capacity to make well-informed decisions (Murphy et al., 2001, 1999), i.e., task comprehension. Each young person decides based on their subjective value or utility function, and how they estimate risk and uncertainty depends on these processes at work (Harbaugh et al., 2002). Does this play a significant role, especially so in youth, in relation to calculating the probability of gains versus losses? For example, our experimental setup gave participants three choices: picking a for-sure amount, predicting a risky amount where the probability was clear, or predicting the probability of winning a lottery amount where the probability was ambiguous. Thus, the option available to choose a certain amount was always clearer than the risky (and ambiguous) choices. Is it possible that the young person had difficulty estimating the probability of winning because of their IQ score, limited experience, and/or because ambiguous choices are challenging at this age anyway? Was this especially so when trying to differentiate the outcome of each gamble, therefore making less risky choices during the experiment?

Probability weighting function in youth is a recognized limitation of other decision-making studies, where studies claim the results vary considerably (Lang and Betsch, 2018). A more comprehensive behavioral economic study, evaluating risk and ambiguity attitudes in healthy youth aged 10–25, included a choice experiment that assessed participants’ understanding of the overall economic task (Blankenstein et al., 2016). This study did not control mood using any psychological questionnaires, but when evaluating choice violations, it found that most participants were economically rational—unrelated to age. In other words, when given the chance at a lottery offering a higher amount, participants choose a certain amount (avoided risky choices) and consistently do so, over the course of the experimental task. Healthy adolescents in Blankenstein’s study, like participants in our study who self-rated as severely anxious, were also risk averse. Another famous study reporting on risk attitudes across different age groups, specifically evaluating probability weighting of the choices in children, young adults, and adults, makes the bold claim that “younger participants’ behavior over both losses and gains appear consistent with a tendency to underweight low-probability events and overweight high-probability ones” (Harbaugh et al., 2002, p. 63). In other words, the authors found that a lean towards adulthood, made the individual more risk neutral, and the younger they were more risk-taking, the latter especially so when the perceived yield of the choice was a high-probability prospect over gains (Harbaugh et al., 2002, p. 53). Overall, what their study demonstrates is the change in attitudes towards risk over the healthy person’s lifetime (Harbaugh et al., 2002),

What these mixed results reinforce is the importance of the context in which the decisions are being made. In a large review studying the impact of stress on decision-making in multiple contexts, these findings are further validated (Starcke and Brand, 2012). Young people may therefore struggle with the decision process itself (Berns et al., 2008), and when having to make them in a variety of uncertain situations that often present themselves in day-to-day life. Decision-making of any kind, but especially ambiguous decision-making, requires that affective and cognitive processing work in sync, and economic scholars have long suggested that people in general struggle with decisions where the probability of a choice outcome cannot be estimated (Ellsberg, 1961; Knight, 2012). One recent scientific study using Magnetic resonance imaging (MRI), exploring social decision-making in 61 adolescents with trauma who were experiencing internalizing symptoms, found that individuals suffered a diminished prospective of mental simulation of reward because the mechanism associated with learning is impacted (Cisler et al., 2023, pp. 6910–6920). This had a major influence on their capacity to make more rewarding decisions. Trauma in this context refers to one or multiple factors associated with psychological, emotional, and/or physical abuse and neglect experienced in developmental years (Cisler et al., 2019).

Furthermore, decision-making continually changes because of multiple variables that need to be in place for the decision process to occur effectively. Regulated mood is not the only factor that impacts the person’s decision outcomes, other factors have a major influence—such as the biologically developing neural mechanisms. Whether choice offerings are real versus hypothetical is another example that demonstrates the subtle biases associated with the process (Holt and Laury, 2002); and for example money itself, as compared to food items as payment outcomes for individuals (Chung et al., 2019).

Limitations

Our study has a few limitations. We did not compare our findings to healthy controls, run the study longer than the 6–8-week period, nor did we collect longitudinal data over a long-term period. Due to mental health privacy laws, we were not allowed to include an analysis in this study of whether or not our study participants were in treatment, what kind of treatment they were receiving, and for how long. Future studies could compare risk and ambiguity attitudes during this crucial neurodevelopmental stage and evaluate whether or not the young person experiencing dysregulated emotions outgrew their risky attitudes and/or their mood scores improved over the course of this time frame. This is due to multiple factors, but the influence of receiving clinical treatment. Future studies could separate the possible conflation between ‘mood disordered’ and ‘healthy adolescent’ behavior using indirect measures, such as those used in economic paradigms. Our current study only included economic tasks to evaluate behavior. Future studies could include a more age-sensitive and comprehensive battery of tasks, such as cognitive tasks used in contemporary and clinical neuropsychological studies24 that could aim to better understand whether other issues are presenting, either related to or unrelated to an emerging mood disorder.

Another limitation of this experiment was our small sample size and that we used psychological self-reporting to evaluate dysregulated mood. The sample size was estimated this size for three reasons: firstly, this was a pilot study testing economic methodologies in a clinical population in an Australian youth clinic for the first time; secondly, our aim was to precisely duplicate the Harbaugh economic experiment (Harbaugh et al., 2002) with a similar sample size for young adults, and compare these results, before moving to a much larger trial; and thirdly, due to constraints associated with recruitment of clinical youth study participants arriving for mental health support, it was challenging to have a bigger sample size participate. Future studies could target larger cohorts of young people with clinical diagnoses in larger clinical facilities and longitudinally, to allow for generalizability of our results. We could apply the Mental Health Clinical Staging Model (Scott et al., 2013), where earlier signs and symptoms of mood disorders can be identified in young help-seekers and evaluate risk and ambiguity attitudes at each of the staging levels. Furthermore, our study did not screen out those participants who were prescribed anti-depressive medication and/or who were receiving other forms of treatment, or for how long, and whether some or all participants were in fact receiving treatment during our study’s research period—all factors that could further confound our results.

Where our study differs from other research using the same economic tasks to capture risk and ambiguity preferences, especially in youth decision-making, is that our focus was on the impact of prolonged mood states—anxiety (trait) and depression in a clinical setting, as opposed to a laboratory setting. There has been much research conducted using economic methodologies in healthy youth populations—such as university student populations, in the United States of America and in Europe, giving our pilot study a foundation to build on. When reviewing the literature, we identified that even when conducting studies in these healthy populations, or populations where the study was not screening for mental health disorders, few mentioned biological factors that could be influencing their findings, such as youth in this age group are biologically still developing. In conclusion, this pilot study forms part of a growing number of empirical studies that aim to investigate how the decision-making of young people experiencing dysregulated emotion is affected when exposed to risk. The impact of mood disorders on decision-making in general is complicated—even with technological advances in science shedding light on this topic. Anxiety and depression, especially during adolescence, have become more prevalent in our society with negative decision-making consequences, in some cases tragically leading to suicide (Twenge et al., 2018). Our empirical findings emphasize the relevance of investigating youth risk attitudes, not in opposition to other fields, but in conjunction with a broader research schema. This interdisciplinary research may have clinical implications as well as support policymakers.