An exploratory approach to the implicit and explicit assessment of risk: Benefits of their combined use for the prediction of risk attitudes and risky behavior

doi:10.21203/rs.3.rs-2083319/v1

Background

Exclusive reliance on declarative measures of risk attitude may lead to an incorrect overall picture of the construct and its relations with behavioral outcomes. This study approached these concerns from a social cognition perspective by looking at implicitly assessed attitudes toward risk measured using the implicit association test (IAT). We conducted two studies to identify the benefits of using the IAT in studying risk attitudes.

Methods

In Study 1, we recruited 62 students enrolled in an MBA program (older group) and asked them to complete the explicit and implicit attitudinal measures of the risk test, along with measures of risk-related personality traits. Newly collected data obtained from the older group were compared with corresponding data from their younger counterparts reported in a previous study. In Study 2, 53 students majoring in financial engineering (FE) were recruited and asked to complete explicit and implicit measures of financial risk attitudes. A week later, they were instructed to join a simulated real-world stock trading competition.

Results

Study 1 found theoretically expected age differences more manifest in implicit measures of risk attitude than conceptually corresponding explicit measures, further identifying psychosocial maturity as a mediator in an inverse age – implicit risk attitude relationship. Study 2 demonstrated that the prediction of individual differences in stock return performance is improved when the explicit risk control and implicit stock investment operate in a complementary manner; implicitly aggressive students who majored in FE had higher rates of return on stock when they were predisposed more explicitly to pursuing control over expected future returns.

Conclusion

These results suggest that implicit measurement approach serves as a useful complement to more traditional, explicit ones. It could propose a new perspective that offers a basis for providing insights into risky behavior and supplementing existing knowledge of human behaviors.

risk

risk attitudes

explicit and implicit

age and risk taking

stock investment

Risk attitude plays a pivotal role in human decision-making processes and has long been studied as the main predictor of daily risk-taking behavior. The literature has tried to understand people’s attitudes toward risk through different lenses, including economic [1], cognitive [2], affective [3], and motivational perspectives [4]. However, most studies on risk attitudes rely heavily on self-report methods which limit their contribution to risk research in organizational settings [5].

Although most self-report methods have robust psychometric properties with acceptable levels of validity and reliability, several concerns have been raised. Respondents may find it difficult to report their actual values because of their introspective limits [6]. Another concern arises from susceptibility to response bias [7]. Respondents might be reluctant to provide an accurate description of themselves as risk-takers or risk-haters due to a fear of being stigmatized [8]. Thus, exclusive reliance on declarative measures of risk attitude may lead to an incorrect overall picture of the construct and its relations with behavioral outcomes.

This study approached these concerns from a social cognition perspective by looking at implicitly assessed attitudes toward risk measured using the implicit association test (IAT) [6]. The IAT assesses the relative strength of associations, indexed by reaction times in sorting target stimuli with positive and negative attributes generated to represent one’s implicit attitude toward a target concept. The IAT’s advantage is that it enables us to tap into an additional aspect of attitude that is automatic while limiting the negative impacts of self-reporting [7]. Greenwald et al. [9] and Kurdi et al. [10] supported this argument with two observations in their meta-analyses: (a) both IAT and self-reporting measures have complementary incremental validity, whereby each contributes a unique variance in explaining different aspects of criterion behavior, and (b) the social sensitivity of the topic exerts a more substantial influence on the predictive validity of self-reporting than IAT measures.

Despite the IAT’s merits, few studies have provided evidence of reliable methods for indirect assessment of a domain-independent risk attitude. Ronay and Kim [11] adapted the IAT to assess implicit attitudes toward risk as a global concept by measuring the strength of automatic associations between target words (i.e., RISK and BLANK) and gain- and loss-related attributes (e.g., GAIN and LOSS). They found that implicitly assessed attitudes toward risk play a significant role in explaining individual differences in risk-taking in the Balloon Analogue Risk Task (BART). Similarly, Traczyk and Zaleskiewicz [12] applied an IAT that assessed a general attitude toward risk by measuring the associative strength between target words (i.e., RISKY and SAFETY) and positive and negative attributes (e.g., POSITIVE and NEGATIVE). The proposed IAT explained additional variance in self-reported risk-taking behaviors that explicit attitudinal measures of risk cannot capture. Although the two IATs discussed above differ in the targets and attribute stimuli used, they provide reliable evidence supporting the incremental predictive validity in the field of risk attitudes. Such results could be interpreted by drawing on implicit social cognition phenomena [13], whereby self-reported and implicitly assessed attitudes toward risk can be related—albeit distinctively separate—constructs.

In the literature, two crucial questions require further investigation. First, there is mixed evidence about the determinants of risk attitudes—in this case, age—when traditional survey methods are used. Second, few studies have determined under what conditions the two main aspects of attitudes toward risk may complement each other to provide a more accurate forecast of risky behavior than taking one alone into account. To address these issues, we conducted two studies to identify the benefits of using the IAT in studying risk attitudes. In Study 1, we tested whether age differences in attitudes toward risk were especially pronounced in implicit measures, hoping to explain puzzling inconsistencies in the literature on age differences in directly stated attitudes toward risk. To delve deeper into possible mechanisms accounting for the expected age differences in implicit risk attitudes, we also measured individual differences with an emphasis on psychosocial maturity. In Study 2, we analyzed the conditions under which implicitly assessed evaluations of risk in response to stock investment were complementary to self-reported evaluations in the predictions of stock return rate.

Age and risk-taking

A traditional approach to examining the relationship between age and risk attitude is to directly ask people about their evaluations of risk using various types of surveys. Early studies that took this approach found that risk taking lessened with age in a sample of managers and executives when using a Likert-type scale and scenario-based questionnaire [14]. Rutledge et al. [15] conducted a smartphone-based experiment in which respondents’ attitudes toward risk were elicited using a hypothetical lottery choice task involving a risky and a safe option; they found that younger respondents chose the risky option more often than their older counterparts, at least in the domain of gains.

However, other findings from gains studies are inconsistent with these results. For instance, when asked to choose between a certain gain and a risky option, older and younger participants did not differ in the binary choice outcome [16]. This pattern was also observed in perceptions of risk [17] and in evaluation of the costs and benefits inherent in risky activities [18]. One compelling explanation for these inconsistent findings is derived from developmental neuroscience, which emphasizes the influence of psychosocial factors on real life risk-taking [19]. Contrary to logical reasoning abilities, which fully mature by 15 years of age, psychosocial maturity such as impulse control [20], emotion regulation [21], and delayed gratiﬁcation [22], which are negatively associated with risk behaviors, develop gradually during adulthood. Given that implicit attitudes are defined as “introspectively unidentified traces of past experience” [13, p. 7], the IAT should reflect risk-related associations acquired from the different experiences of older versus younger adults in psychological maturity and consequently capture marked age differences.

Methods

Participants and procedure

We recruited 62 students enrolled in an MBA program (52 males, mean age = 38.85 years, SD = 7.64; older group) and asked them to complete the Barratt impulsiveness scale and the NEO Five Factor Inventory-Conscientiousness as a proxy variable for psychosocial maturity, with impression management as a covariate. Following this, they completed the explicit and implicit attitudinal measures of the risk test. To explore how aging is associated with attitudes toward risk and what mechanisms can account for implicit risk attitudes among different age groups, we compared newly collected data obtained from the older group with corresponding data^[1]from a sample of 284 South Korean undergraduates (135 males, mean age = 22.73 years, SD = 2.20; younger group) reported in a previous study (redacted for review).

Measures

Risk Global-Semantic Differential Scale (RG-SDS)

The RG-SDS [11] is a self-reported attitudinal measure that assesses participants’ overall evaluation of the concept of risk. Participants indicated their attitudes on a seven-point scale (see additional file 1), with a Cronbach’s α of .89. Higher scores on the RG-SDS indicates more positive attitudes toward risk.

Risk Unique-Semantic Differential Scale (RU-SDS)

The RU-SDS [11] is a variant of the RG-SDS in which the target (i.e., risk) is replaced with self-relevant risk activities. The RU-SDS was used because people can display risk-taking behaviors despite a negative perception of the overall (global) risk. Participants self-selected 10 relevant activities from 34 risky behaviors (see additional file 2) and then reduced these ten to the six most relevant to each participant, which were used as evaluation categories in the same format as the RG-SDS. Higher scores on the RU-SDS indicates positive attitudes toward self-selected risk activities. The Cronbach’s α was 0.95.

Implicit Risk Task-Global (IRT-Global)

The IRT-Global is a variant of the IAT that provides indirectly assessed implicit evaluations of risk as an “abstract” construct by measuring the relative strength of automatic associations between the attributes, “GAIN” and “LOSS,” and the target concepts of “RISK” and “[ ]” [11], following the same paradigm and method used by Kim [23]. For example, if a person shows a stronger positive evaluation about risk than a negative evaluation, s/he will associate risk with gain more easily than with loss, and consequently perform the classification task more quickly and accurately when RISK is paired with GAIN attributes compared to LOSS.

In the IRT task, participants were asked to categorize the target stimuli (i.e., RISK and open square bracket—in this case, [ ]) while simultaneously discriminating between GAIN- and LOSS-related words (i.e., attribute stimuli^[2]) as fast as they could and with minimal errors. The IRT-Global consists of a sequence of five blocks (see additional file 3). In Block 1, participants practiced the attribute discrimination by sorting items into “GAIN” and “LOSS” categories. In Block 2, they repeated this for the target concept discrimination by sorting stimuli into the “RISK” and “[ ]” categories. In Block 3, participants classified items into two combined categories, each including the attribute and target concept that were assigned to the same response key in the preceding two blocks (e.g., RISK + GAIN for a right (“5”) key and [ ] + LOSS for a left (“a”) key). In Block 4, the key assignment for the target dimension was switched. Block 5 was complementary to Block 3, but it used the reverse key assignment to the one practiced in the previous block (e.g., [ ] + GAIN for the right key and RISK + LOSS for the left key). As a result, an overall IRT-Global score was generated by subtracting the mean reaction time in the test trials of the combined task (i.e., Block 3) from the mean reaction time in the test trials of the reversed combined task (i.e., Block 5); higher IRT-Global scores indicated faster associations of risk with gain rather than loss. The Cronbach’s α was 0.81.

Implicit Risk Task-Unique (IRT-Unique)

The IRT-Unique is an adaptation of the IRT-Global that measures an individual’s positive and negative associations with personally relevant risks [11]. The six items participants chose automatically replaced the stimulus “RISK” in all the blocks, which was the only difference between the IRT-Global and IRT-Unique. The overall IRT-Unique score was calculated using the same method as the IRT-Global; higher IRT-Unique scores indicated faster associations of self-selected risk activities with gain than loss. The Cronbach’s α was 0.87.

Barratt Impulsiveness Scale (BIS)

The BIS is a 23-item questionnaire that assesses impulsivity as a component of both impulse control and risk-taking behavior [24]. Higher scores on the BIS indicate greater impulsiveness. The Cronbach’s α was 0.77.

NEO Five Factor Inventory-Conscientiousness (NEO FFI-C)

We adopted the 12-item scale from Costa and McCrae’s [25] NEO Five-Factor Inventory to assess conscientiousness as a personality trait characterized by self-discipline, industriousness, and impulse control. These factors were found to be strongly associated with delayed gratification [22], emotional regulation [21], and age [25], which are theoretically assumed to underpin psychosocial maturity. Higher scores on the NEO FFI-C indicate greater conscientiousness. The Cronbach’s α was 0.80.

Impression Management (IM)

The 20-item scale for IM was based on Paulhus’ [26] Balanced Inventory of Desirable Responding, which set out to negate the influence of IM while responding to various risk-related attitudes and personality traits. The Cronbach’s α was 0.74.

Results

Implicit–explicit divergence in the risk-attitude construct

Table 1 presents the correlations among Study 1’s measures across different age groups. Two confirmatory factor analyses were first performed across age groups to directly test if the correlational data for the explicit and implicit measures of risk attitude were better suited for a model with two factors (Figure 1, right side) or one (Figure 1, left side). According to the test results, the two-factor model shows a better fit, with a significant chi-square difference in the older group, χ²_diff (2, N = 62) = 6.66 (p < 0.05), as well as in the younger group, χ²_diff (1, N = 283) = 32.45 (p < 0.05) (see additional file 4 for further details about the confirmatory factor analyses results). These results suggest that self-reported and implicitly assessed attitudes toward risk are conceptually divergent across both age groups.

Divergent findings on age

A univariate analysis of variance controlling for gender^[3] was performed to detect the age differences for explicit and implicit measures.

Table 1 Correlation coefficients of the main variables for Study 1 across age groups (n = 62[284])

Measure	1	2	3	4	5	6	7	8
Demographics
1. Age	-
2. Gender	-.50*(-.38*)	-
3. IM	.41**(.06)	-.13(.13^†)	-
Psychosocial maturity
4. BIS	-.38(-.29**)	-.01(.21**)	-.42(-.22)	-
5. NEO FFI-C	.44(.20)	-.24(-.18**)	.43**(.11^†)	-.53*(-.61*)	-
Explicit risk
6. RG-SDS	-.12(.06)	-.12(-.19**)	.07(.00)	-.18(-.12^†)	.26^†(.07)	-
7. RU-SDS	.01(.03)	-.11(-.10)	.08(.02)	.04(-.11)	.24(.09)	.36(.38*)	-
Implicit risk
8. IRT-Global	-.31*(-.03)	-.08(-.09)	-.25(-.05)	.35*(.04)	-.24(-.04)	.26(.17*)	.07(.10)	-
9. IRT-Unique	-.31*(-.09)	-.08(-.18**)	-.50**(.00)	.38*(-.04)	-.48**(.07)	.17(.23***)	.23^†(.32***)	.60*(.54*)

Numbers outside (inside) parentheses indicate the correlation coefficients in the older (younger) group.

Gender was coded as 1 for male and 2 for female (^†p < 0.1; * p < 0.05; ** p < 0.01; *** p < 0.001)

Explicit risk. The test on cross-age comparisons showed similar patterns across two explicit measures of risk. The older (M = 3.60, SD = 1.38) and younger (M = 3.64, SD = 1.31) groups did not differ in the RG-SDS, t(336) = -1.13, p = 0.26, d = 0.03. Nor was there a difference in the RU-SDS for the older (M = 3.19, SD = 0.77) versus younger (M = 3.13, SD = 1.02) groups, t(341) = -0.18, p = 0.86, d = 0.05.

Implicit risk. A scoring algorithm introduced by Greenwald, Nosek, and Banaji [27] was used for the IRT. In this study, all statistical analyses were conducted using the scoring method (D score). While the two age groups were implicitly negative about risk, both globally and uniquely, the older group (M = -0.36, SD = 0.42) more strongly associated risk with loss than the younger group (M = -0.18, SD = 0.41) on the IRT-Global, t(337) = -3.47, p = 10^-3, d = 0.44. The same pattern also applied to the IRT-Unique for the older (M = -0.42, SD = 0.34) versus younger (M = -0.29, SD = 0.39) group, t(327) = -3.08, p = 10^-3, d = 0.34.

Mediating role of psychosocial maturity

The bivariate correlations provide support for the mediating role of psychosocial maturity in the relationship between age and implicit risk attitudes. In the older group, age correlated significantly with two IRTs (r = −0.31, p = 0.03 for the IRT-Global and r = −0.31, p = 0.03 for the IRT-Unique) and two personality traits (r = −0.38, p = 0.01 for the BIS and r = 0.44, p = 0.003 for the NEO FFI-C). Notably, the two implicit measures and two personality trait measures were closely associated, with an average correlation of r = 0.36 (p = 0.02). However, this pattern was not present in the younger group. To examine if implicit risk attitude decreased due to psychosocial maturity in the older group, we compared Model 1, which constrained the two paths of psychosocial maturity (age à psychosocial maturity à implicit risk) to zero, with Model 2, in which the indirect path coefficients were unconstrained (Figure 2).

The test results reveal a better fit with Model 2, χ² (7, N = 62)= 8.82 (p = .27), CFI = 0.98, TLI = 0.91, NFI = 0.92, RMSEA = 0.06, than Model 1, χ² (9, N = 62)= 17.31 (p < 0.05), CFI = 0.90, TLI = 0.68, NFI = 0.84, RMSEA = 0.12, with a significant χ²_diff(2, N = 62) = 8.49 (p < 0.05). In addition, further analysis using bias-corrected bootstrapping (k = 5,000) was performed in order to compute a 95 % confidence interval (CI) to test the significance of the indirect effect. The test results revealed a significantly negative indirect effect of age on implicit risk (β = -0.19, 95% CI [-3.185, .000]).

Discussion

Study 1 identified the underlying mechanism that can explain these differences while demonstrating the role of psychosocial maturity as a mediator. The fact that older adults had more negative implicit attitudes toward risk than younger adults can be interpreted by drawing an environmental association model [28], which suggests that exposure to an environment with different developmental opportunities may steer each age group toward heterogeneous implicit evaluations of risk. This interpretation is further supported by our finding that older adults’ implicit attitudes toward risk becomes more negative with age (i.e., through psychological maturity). Overall, these findings demonstrate the implicit measure (e.g., the IAT) to be reliable, valid, and helpful in investigating theoretical phenomena that cannot be tapped into by self-reporting. Accordingly, in Study 2, we sought to extend the utility and generality of our findings by examining whether the combined use of implicit and explicit evaluations of risk would offer additional benefits in the domain of finance (namely, attitudes toward stock investment), where financial investment companies rely heavily on self-reports to learn about customers’ attitudes toward how much risk to bear in their investments. As noted earlier, deciding where, when, and how much to invest in stocks requires people to analyze large amounts of market information that can be processed either deliberately or spontaneously [29], which may ultimately form a base for the development of an individual's explicit and implicit attitudes toward how aggressively or conservatively valued one’s own investments in stocks are.

However, respondents’ self-reports of their willingness to take risks in stock investment may not reflect automatic evaluations of one’s own investments due to the limitations of explicit assessment methods. For this reason, we conducted an additional study to tap into explicit and implicit attitudes toward stock investment and determine whether indirectly assessed implicit evaluations of stock investment provide gains in behavioral outcome predictions; this is measured using the rate of return on stocks when working synergistically with self-reported (i.e., explicit) evaluations.

[1] Parts of the data from the younger group have been used in a previous publication. However, this publication does have a very different focus (group discussion effect in risk decision making) by presenting group-level analyses.

[2] The stimuli consisted of 14 items, seven in each of the two attributes (i.e., GAIN and LOSS), which are the same as used in the RG-SDS.

[3] The effect of gender was found to vary across the measures (see additional file 5 for more details).

Predictive models of implicit–explicit attitudes and their utility in stock investment

Three main predictive models were derived from the literature: an additive model that occurs when the two types of attitudes predict a unique portion of the variance in the criterion [9,10], a double-dissociation model corresponding to situations where implicit and explicit attitudes are often considered particularly successful in the prediction of spontaneous and deliberate behaviors [30,31], and a multiplicative model, whereby implicit and explicit attitudes interact synergistically to affect behavior. Compared with the former two, the third model is the least tested, given its relatively frequent occurrence. However, since the mid-2000s, there have been several attempts to directly test the predictive power of the implicit–explicit attitude interplay for a range of behaviors. In his pioneering work on health-related behavior, Perugini [32] provided empirical insights into the interactive relationship between self-reported and implicitly assessed attitudes toward smoking by demonstrating how predictions about whether someone was a smoker were more accurate when respondents had positive attitudes, both explicitly and implicitly, toward smoking. Subsequently, the multiplicative effect of implicit and explicit attitudes has been replicated across different domains such as aggression [33] and bullying [34].

Although the multiplicative model seems to be relatively uncommon, we conjecture that it might be useful for the prediction of behavioral outcomes in the domain of stock investment; that is, stock return performance would be better predicted by combining self-reported (explicit) and indirectly assessed (implicit) evaluations of one’s own investment in stocks in a multiplicative fashion. A theoretical framework of financial decision-making proposed by Cheng [35] reinforces our expectation. His framework goes one step further by emphasizing the synergetic interplay between conscious thought, characterized by explicit reasoning processes that require conscious attention, and unconscious thought, conceived as implicit associative mechanisms that operate automatically or with limited conscious accessibility. Given the relatively small capacity of conscious thought (i.e., 40 to 60 bits per second) [36] compared to that of the entire human thought system, Cheng [35] proposed that unconscious thought can help reduce some of the judgmental heuristics (e.g., availability and representativeness) and behavioral biases (e.g., overconfidence and attraction effect) that people are subject to when making complex decisions on financial matters. This is due to its enormous processing capacity, which improves the quality of financial and investment decisions in general. Nordgren, Bos, and Dijksterhuis [37] provided indirect evidence for this view by examining whether a sequential combination of conscious and unconscious thoughts helps make better decisions on complex problems. The authors followed the same procedures and research design (i.e., immediate, conscious, and unconscious thought conditions^[4]) as in the standard unconscious thought paradigm [38], except for including a newly created condition; in this case, participants deliberated on which apartments they would be willing to rent for a given time and were then distracted in the same interval, allowing conscious and unconscious thought to occur sequentially. Results showed that participants were more likely to choose the best apartment when they engaged in periods of both conscious and unconscious thought, compared to the other three conditions.

These facts suggest that unconscious thought is of particular value when one is faced with complex decisions that require a substantial amount of information to be processed simultaneously [39], such as stock investment decisions. We argue that measuring one’s automatic evaluations of stock investment with the IAT would help more accurately reflect unconscious^[5] features of indirectly assessed attitudes, which cannot be controlled by motivation and ability [40]; as such, when considered simultaneously with implicitly assessed attitudes toward stock investment, their interplay with self-reported attitudes would provide a valid basis for making better decisions and improving performance in stock investment.

Implicit–explicit attitudes toward investment

Similarly, other studies have attempted to apply the IAT to explore the attitude–behavior relationship in risk-related domains such as risky flight behavior [41], gambling [42], and engagement in extreme sports [12]. Another example is Park, Kim, and Oh [43], who adapted the IAT to provide a measure of implicit attitudes toward how aggressively or conservatively valued one’s own investments in stocks were—namely, the Implicit Stock Investment (ISI), in which the strength of automatic associations between the target (i.e., My stock investment and BLANK) and aggressive- and conservative-related attributes (e.g., AGGRESSIVE and CONSERVATIVE) are measured as reaction times. Using a sample of students majoring in financial engineering (FE), they found that the ISI captures unique incremental variance beyond existing self-report measures of financial risk attitudes, while predicting risk-taking in a laboratory-based behavioral task (i.e., BART). Next, to understand the influence of implicit attitudes on real-life risky behavior, we sought to explore a potential link between FE major students’ attitudes toward their investments in stocks and their stock return performance. We expected these two variables to be negatively related, indicating that the FE major group’s overall rate of return on stocks would decrease, as they would respond more quickly when associating “My stock investment” with “Aggressive” than with “Conservative.” There is a growing body of evidence for this idea, suggesting that investors who report being more aggressive and tolerant of risk are more likely to make risky investments such as equities and options [44]; thus, they hold relatively fewer diversified stock portfolios [45]. Indeed, investors who allocate investments to specific industries and stock characteristics perform worse in their annual rates of return than investors with well-diversified portfolios [46]. Hence, implicitly assessed attitudes toward stock investments in the form of the bipolar IAT are expected to negatively affect the overall rate of return in the FE major group.

In line with the multiplicative model of implicit–explicit attitudes, the influence of the evaluative stock investment associations measured by the ISI on the rate of return is assumed to be conditional: we expect that the influence of implicit evaluations of one’s investment in stocks on the rate of return depends on their explicit evaluations of the level of control they are willing to exert over financial risk. Lampenius and Zickar [47] provided supporting evidence that risk control played a crucial role in predicting investment portfolio selection. In their first study, the authors provided a newly developed, self-report measure of attitudes toward financial risk, Financial Risk-Taking (FRT), that was elicited from two major factors: speculative risk (SR) and risk control (RC). Whereas SR is an internal force that causes individuals to accept greater risks for maximizing returns on investment, RC refers to a driving force that leads them to prioritize the predictability of future expected returns, cash flows, and monetary status when making investment decisions; thus, people prone to high levels of RC are more likely to prefer low-yield but relatively safe investment options, as they want to control their returns. Lampenius and Zickar [47] further investigated the predictive power of these two factors on portfolio selection in their second study examining students who majored in finance. They found that when asked to choose the preferred one from four pairs of portfolios formed by a combination of multiple assets (e.g., cash, stocks, options, and bonds), participants who scored highly in SR and RC preferred a portfolio comprised of a balanced (i.e., diversified) mix of high, medium, and low risk/return on assets.

Hence, by modelling RC as the moderator, we can expect implicit attitudes toward stock investment to positively impact the rate of return when associated with high RC. For instance, an individual harboring strong, aggressive associations with stock investment may perform better in the rate of return on stocks provided he/she becomes more explicit in controlling returns.

Method

In this study, we used a convenience sample of Korean students majoring in FE by recruiting them in collaboration with a Korean university, which established the FE department with student quotas of 40 every year. This is because the number of FE major students in the research participant pool was largely limited for reasons such as the FE department in existence less than four years and a leave of absence for mandatory military service.

Participants and procedure

Fifty-three students majoring in FE (43 males, age M = 20.53, SD = 0.64) participated in the study. Each participant received course credits for completing explicit (the FRT) and implicit (the ISI) measures of financial risk attitudes. One week after completing the surveys, they were asked to join a four-week simulated real-world stock trading competition organized by a large securities company. To encourage active participation, students were paid $5 each. At the end of the competition, participants who ranked first through sixth in the overall rate of return on their investments received additional monetary rewards ($100 for first place, $50 for second place, $30 for third and fourth place, and $20 for fifth and sixth place).

Measures

Financial risk-taking

The FRT measure has two 5-item subscales assessing speculative risk (FRT-SR) and risk control (FRT-RC) (see additional file 6) [47]. This study rates FRT-SR and FRT-RC on a Likert scale ranging from 1 (strongly disagree) to 5 (strongly agree). The Cronbach’s α were 0.65 and 0.57^[6], respectively.

Implicit stock investment

We used the ISI to assess implicit attitudes toward stock investment by estimating the relative strength of automatic associations between the attributes, “AGGRESSIVE” and “CONSERVATIVE,” and the target concepts of “My STOCK INVESTMENT” and “[ ]” [43].

As with the IRT task, the ISI procedure also included a sequence of five blocks, each of which involved a categorization task where the participants were instructed to sort the target stimuli (i.e., MY STOCK INVESTMENT and open square bracket, [ ]) while simultaneously discriminating between AGGRESSIVE- and CONSERVATIVE-related words (i.e., attribute stimuli) (see Figure 3). In Block 1, the participants practiced attribute discrimination by sorting the items into the “AGGRESSIVE” and “CONSERVATIVE” categories. In Block 2, the same was done for discrimination by sorting the stimuli into “MY STOCK INVESTMENT” and “[ ]” categories.

In Block 3, they sorted the items into two combined categories, each including the target concept and attribute that shared the same response key in the preceding two blocks (e.g., “MY STOCK INVESTMENT” + “AGGRESSIVE” for the right (“5”) key and “[ ]” + “CONSERVATIVE” for the left (“a”) key). In Block 4, stimulus items for the target dimension used in Block 2 were illustrated again but their assignment key was switched. In Block 5, the participants repeated Block 3 actions with switched pairings (e.g., “[ ]” + “AGGRESSIVE” for the right key and “MY STOCK INVESTMENT” + “CONSERVATIVE” for the left key). An overall ISI score (D) was generated by subtracting the mean reaction time in test trials of the combined task (i.e., Block 3) from that in the test trials of the reversed combined task (i.e., Block 5); higher ISI scores indicate faster associations of stock investment with aggressive rather than with conservative behavior. The Cronbach’s α was 0.78.

Control variables

A brief survey was conducted to gather participants’ demographic information and their perception of external factors that could affect trading decisions, such as the investment information, stock market forecasting, international and domestic economic fluctuations, and time spent on stock trading. These factors were rated on a five-point scale from “not at all influential” (1) to “extremely influential” (5) to reflect their perceived influence on participants’ trading decisions during the competition. Along with year of study (ranging from 1 to 3 years), gender (coded female = 0; male = 1), and the average number of stocks in their portfolios as control variables, these factors were included in the analysis.

Results

Enhanced interaction of ISI with FRT-RC

Table 2 presents the correlations between the study variables. As expected, the ISI was negatively correlated with the overall rate of stock return, r = -0.24, p = .079, although it was marginally significant. However, this pattern was not reliable for FRT-RC, r = -0.16, p = .26, and FRT‑SR, r = -0.11, p = .43.

To explain the additional variance due to the interaction between ISI and FRT-RC, we employed a moderated multiple regression analysis, entering control variables in step 1, a set of predictors (i.e., FRT-RC, FRT-SR, and ISI) in step 2, the other two interactions (i.e., ISI × FRT-SR and FRT-RC × FRT-SR) in step 3, and the ISI–FRT-RC interaction in step 4.

Table 2. Correlation coefficients among the main variables for Study 2 (n = 53)

Measure	1	2	3	4	5	6	7	8	9	10
Individual demographics
1. Gender	-
2. Year of study	-.21	-
3. Investment information	-.10	.05	-
4. Market forecasting	-.40**	.05	.17	-
5. Economic condition	-.20	.002	.27^†	.16	-
6. Time spent investing in stocks	.12	.05	.47**	.07	.15	-
7. Number of stocks in portfolio	.29^*	.02	.19	-.21	-.28^†	.25	-
Explicit risk
8. FRT-RC	-.04	-.03	-.15	.06	-.07	-.32*	.14	-
9. FRT-SR	-.05	.09	-.01	-.07	.11	.24	-.01	-.24^†	-
Implicit risk
10. ISI	-.15	.14	-.21	-.004	-.05	-.25	-.26^†	-.16	-.11	-
Performance
11. Rate of return	-.18	.05	-.13	-.04	.04	.13	.14	.15	-.03	-.24^†

†p < .10, * p < .05, ** p < .0

As shown in Table 3 (Model 2), the three main predictors were found to be non-significant: b = 0.58, se = 0.65, p = 0.38 for FRT-RC; b = -0.15, se = 0.57, p = 0.80 for FRT-SR; b = -6.15, se = 4.04, p = 0.14 for ISI. However, critically and as theoretically expected, a significant two-way interaction effect was found only for ISI × FRT-RC (Model 4), β = 0.53, t = 2.63, p = 0.01. In addition, we conducted a bias-corrected bootstrap test with 5,000 samples to verify the moderating effect, resulting in a significantly positive interaction effect of ISI and FRT-RC on the rate of return (b = 4.17, se = 2.31, 95% CI [0.16, 8.82]).

Table 3. Moderating analysis of ISI on rates of return

Variables	Rate of return on stock
	1	2	3	4
Step 1: Control variables
Gender	-.30	-.30	-.29	-.33^†
Year of study	.06	.07	.04	.17
Investment information	-.35^†	-.37^†	-.41^†	-.47*
Market forecasting	-.08	-.12	-.12	-.26
Economic fluctuations	.13	.10	.14	.16
Time spent investing in stocks	.24	.29	.31	.31
Number of stocks in portfolio	.28	.15	.17	.24
Step 2: Main effects
FRT-RC		.17	.13	.22
FRT-SR		-.05	-.06	.07
ISI		-.25	-.16	-.12
Step 3: Interaction 1
FRT-RC*FRT-SR			.12	.05
FRT-SR*ISI			-.14	.13
Step 4: Interaction 2
FRT-RC*ISI				.53*
R²	.18	.28	.30	.44
DR²		.10	.02	.14
DF		1.34	.44	6.91

Standardized coefficients are presented (^† p < 0.10; * p < 0.05)

To better understand this interaction, we plotted and tested simple slopes using values one standard deviation above and below the mean of FRT-RC.

As Figure 4 shows, a positive relationship is evident between ISI and return performance under a high level of FRT-RC (b = 8.25, se = 3.54, p = 0.03), indicating that FE students who were implicitly more aggressive about stock investments had better rates of return when they were explicitly predisposed to pursuing control over expected future returns. In contrast, the relationship was significantly negative under a low level of FRT-RC (b = -13.81, se = 2.34, p = 10^-7), indicating that implicitly aggressive FE students had lower rates of return on stock when they were predisposed less explicitly to pursuing control over expected future returns.

Discussion

Study 2 provided evidence for additional benefits of the combined use of financial risk attitudes, demonstrating a complex link moderated by the FRT-RC between the ISI and stock return performance. The interaction was fully crossed, suggesting that stock return performance hinges on the levels of implicit and explicit attitudes toward financial risk. That is, being implicitly more aggressive about stock investment improved rates of return on stock when participants are more predisposed to exercising conscious control over expected future returns and lowered the return rates when they declare themselves to be less explicit in controlling financial risk. Our findings suggest that self-reported and implicitly assessed attitudes toward financial risk, as measured by the FRT-RC and ISI, complement each other in terms of predicting overall rates of return.

[4] Most studies following the unconscious thought paradigm involved three conditions: immediate, conscious, and unconscious. All participants, regardless of condition, encountered a complex decision problem that required them to make a choice between multiple options (e.g., apartment A-D), each with different combinations of positive (e.g., a very nice area) and negative (e.g., rather noisy) attributes. The participants were then asked either to select one of the apartments immediately (the immediate condition) or think about the decision for a couple of minutes (the conscious thought condition), or perform a distraction task (assumed to elicit unconscious thought while directing their attention away from the decision problem; the unconscious thought condition) and then make a decision.

[5] In their review of studies on attitudes, Gawronski et al. [40] found that people are unaware of how their implicit attitudes can impact psychological/behavioral processes outside of conscious awareness and suggested that such unconscious influences can be interpreted as evidence for unconscious features of indirectly assessed implicit attitudes.

[6] The FRT-RC’s reliability coefficient was re-estimated using McDonald's Omega (ω) index, a newly recommended indicator which is less biased than Cronbach's alpha [48], indicating a ω of 0.62 (ω values above 0.6 indicate adequate internal consistency).

Implicit risk attitudes as complementary assets

The utility of using implicit measures of risk to provide new insights into age differences in risk attitudes is confirmed in Study 1, indicating that older adults are more negative in their implicit attitudes toward risk, both as a global concept and as a response to personally relevant risk activities, than younger adults. Furthermore, we shed light on the reasons the older group had strong risk–loss associations in the IRTs, demonstrating that they became implicitly more negative about risk as they aged (via gradual increases in levels of psychological maturity, which are represented by conscientiousness and impulsiveness). These findings suggest that IAT-based measures of risk attitude may reflect the strength of risk-related associations that each age group has been exposed to in their psychosocial environments.

It is worth noting that the mediation effect observed in older adults was not replicated in the younger group due to a lack of correlations between psychosocial maturity and implicit risk attitudes. One potential explanation for the absence of these relationships in the younger group can be inferred from their developmental status in psychosocial maturity. Compared to older adults, young adults are considered to have unstable psychosocial maturity and more vulnerability to external forces [49]. Thus, the influence of psychosocial maturity on risk judgment may not be salient enough to become internalized and automatized in the younger group compared to the older group. The transitional, flexible, and formative nature of young adults during psychosocial maturity makes it difficult for them to foster strong associations of risk with their developmental experiences in maturity, thereby resulting in non-significant correlations between measures of IRTs and psychosocial maturity.

Despite inconsistency in the mediation effects across the two age groups, the findings that reflect marked age differences for the IRTs are consistent with our suggestion that the implicit measurement approach serves as a useful complement to more traditional, explicit ones. It proposes a new perspective that offers a basis for interpreting the mixed findings in the literature, which used age as a determinant of self-reported risk attitudes.

Extension Of The Implicit–explicit Interplay To Financial Risk Attitudes

Study 2 aimed to examine the influence of stock investment-related associations, as measured by the ISI, and their interplay with explicit evaluations of financial risk, measured with the FRT-RC and FRT-SR, on the overall rate of return on stocks in an FE student sample. Despite the FRTs being considered predictors along with the ISI, we found only the FRT-RC x ISI interaction to have a significantly positive relationship with the rate of return, suggesting that the FRT-RC acts as a dispositional boundary condition under which the negative ISI–stock return rate relationship would turn positive. Specifically, compared with their low FRT-RC counterparts, FE students who were strongly inclined to exercise explicit control over future expected returns benefited from becoming more implicitly aggressive about their investments in terms of the stock return rate. These findings provide clear support for the fact that the multiplicative model, among the three predictive models of implicit–explicit attitudes, does work, even in the prediction of risky financial behavior. The interactive pattern can be explained by two independent but interactive systems of information processing—one thoughtful (a reflective system) and one more automatic (an impulsive system)—that jointly determine social behavior [50]. The key message in the reflective-impulsive model (RIM) is that human behavior is not categorized as purely reflective or impulsive, but is instead guided by a joint function of self-reported and implicitly assessed attitudes that consider the consequences of reflective (effortful, intended, and controlled) and impulsive (effortless, spontaneous, and automatic) processes, respectively. Consistent with the prediction of the RIM, Niemand and Mai [51] found one's counterfeit buying behavior to be better explained when their attitudes toward pirated products, measured both explicitly and implicitly, were congruent in a positive direction. Our results provide further evidence supporting the RIM by demonstrating that the prediction of individual differences in stock return performance can be markedly improved when self-reported and implicitly assessed attitudes toward financial risk-related concepts—risk control and stock investment—operate in a multiplicative manner.

Implications

Our results have several implications for risk research. Study 1 achieved the additional benefit of adopting implicit measures of risk, documenting profound age differences in implicitly assessed attitudes toward risk and the underlying mechanism for why implicit risk attitudes become negative with age. Notably, the explicit–implicit divergence of risk construct using structural equation modeling noted in Study 1 adds new information to the risk and social cognition literature. Furthermore, Study 2 expands on previous works that have focused largely on the incremental predictive power of IAT-based measures of risk attitude above explicit measures in determining risky behaviors by demonstrating what happens when implicit and explicit aspects of attitudes toward financial risk-related concepts interact complementarily in financial investment behavior.

Future studies are necessary to determine whether the findings can be replicated in Western or other Asian samples to ensure greater external validity. Furthermore, the factors that shed further light on the black box, the murky chain of links between the ISI–FRT-RC interaction and stock investment performance, should be further investigated. Despite these remaining shortcomings, the comprehensive assessment of risk attitudes presented in this study could provide insights into risky behavior and supplement existing knowledge of human behaviors in the field of finance.

IAT

Implicit association test

RG-SDS

Risk global-semantic differential scale

RU-SDS

Risk unique-semantic differential scale

IRT-Global

Implicit risk task-global

IRT-Unique

Implicit risk task-unique

BIS

Barratt impulsiveness scale

NEO FFI-C

NEO five-factor inventory-conscientiousness

IM

Impression management

FE

Financial engineering

ISI

Implicit stock investment

FRT-SR

Financial risk taking-speculative risk

FRT-RC

Financial risk taking-risk control

RIM

reflective-impulsive model.

Acknowledgments

This research was conducted as part of the first author's doctoral dissertation. We gratefully acknowledge the members of the Center for Interdisciplinary Research in Business and Psychology and Global Management for their invaluable assistance in the completion of this research.

Author contributions

JP reviewed the literature, conducted the analysis, and wrote the manuscript. DYK interpreted the findings, edited and revised the manuscript. Both authors critically reviewed and approved the final manuscript.

Funding

This work was supported by a grant funded by the National Research Foundation of Korea (2017R1E1A2A02022040).

Availability of data and materials

The data and additional files described in this article are openly available in the Open Science Framework at https://osf.io/23qg8/

Ethics approval and consent to participate

The protocol complies with the ethical standards of the 1964 Helsinki declaration and its later amendments. Because this research was not biomedical, not interventional, and focused on existing B.S./M.S. programs, formal ethical approval was not required for this type of study, in agreement with Article 13 of the Enforcement Decree of The Bioethics and Safety Act (No. 852, 23/03/2013): for reference, see https://www.law.go.kr/. The research protocol was reviewed and approved by the first author’s dissertation committee members from Ajou University, University of Eastern Finland, University of Lappeenranta, and SungKyunKwan Univesity. The present research was implemented with a proper procedure based on APA ethical standards, and informed consent was obtained from all individual participants included in the study.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.

Von Neumann J, Morgenstern O. Theory of games and economic behavior. Princeton, NJ: Princeton University Press; 2007.
Kahneman D, Tversky A. Prospect theory: An analysis of decision under risk. Econometrica. 1979;47: 263.
Loewenstein GF, Weber EU, Hsee CK, Welch N. Risk as feelings. Psychol Bull. 2001;127: 267–286.
Scholer AA, Zou X, Fujita K, Stroessner SJ, Higgins ET. When risk seeking becomes a motivational necessity. J Pers Soc Psychol. 2010;99: 215–231.
Holmes RM Jr, Bromiley P, Devers CE, Holcomb TR, McGuire JB. Management theory applications of prospect theory: Accomplishments, challenges, and opportunities. J Manage. 2011;37: 1069–1107.
Greenwald AG, McGhee DE, Schwartz JL. Measuring individual differences in implicit cognition: the implicit association test. J Pers Soc Psychol. 1998;74: 1464–1480.
Kim D-Y. Voluntary controllability of the implicit association test (IAT). Soc Psychol Q. 2003;66: 83.
Kelly CA, Soler-Hampejsek E, Mensch BS, Hewett PC. Social desirability bias in sexual behavior reporting: evidence from an interview mode experiment in rural Malawi. Int Perspect Sex Reprod Health. 2013;39: 14–21.
Greenwald AG, Poehlman TA, Uhlmann EL, Banaji MR. Understanding and using the Implicit Association Test: III. Meta-analysis of predictive validity. J Pers Soc Psychol. 2009;97: 17–41.
Kurdi B, Seitchik AE, Axt JR, Carroll TJ, Karapetyan A, Kaushik N, et al. Relationship between the Implicit Association Test and intergroup behavior: A meta-analysis. Am Psychol. 2019;74: 569–586.
Ronay R, Kim D-Y. Gender differences in explicit and implicit risk attitudes: a socially facilitated phenomenon. Br J Soc Psychol. 2006;45: 397–419.
Traczyk J, Zaleskiewicz T. Implicit attitudes toward risk: the construction and validation of the measurement method. J Risk Res. 2016;19: 632–644.
Greenwald AG, Banaji MR. Implicit social cognition: Attitudes, self-esteem, and stereotypes. Psychol Rev. 1995;102: 4–27.
MacCrimmon KR, Wehrung DA. Characteristics of risk taking executives. Manage Sci. 1990;36: 422–435.
Rutledge RB, Smittenaar P, Zeidman P, Brown HR, Adams RA, Lindenberger U, et al. Risk taking for potential reward decreases across the lifespan. Curr Biol. 2016;26: 1634–1639.
Mamerow L, Frey R, Mata R. Risk taking across the life span: A comparison of self-report and behavioral measures of risk taking. Psychol Aging. 2016;31: 711–723.
Reyna VF, Farley F. Risk and rationality in adolescent decision making: Implications for theory, practice, and public policy. Psychol Sci Public Interest. 2006;7: 1–44.
Beyth-Marom R, Austin L, Fischhoff B, Palmgren C, Jacobs-Quadrel M. Perceived consequences of risky behaviors: Adults and adolescents. Dev Psychol. 1993;29: 549–563.
Steinberg L. Risk taking in adolescence. Curr Dir Psychol Sci. 2007;16: 55–59.
Stanford MS, Greve KW, Boudreaux JK, Mathias CW, L. Brumbelow J. Impulsiveness and risk-taking behavior: comparison of high-school and college students using the Barratt Impulsiveness Scale. Pers Individ Dif. 1996;21: 1073–1075.
Larsen RJ. Toward a science of mood regulation. Psychol Inq. 2000;11: 129–141.
Forstmeier S, Drobetz R, Maercker A. The delay of gratification test for adults: Validating a behavioral measure of self-motivation in a sample of older people. Motiv Emot. 2011;35: 118–134.
Kim D-Y. The implicit life satisfaction measure. Asian J Soc Psychol. 2004;7: 236–262.
Monahan J, Steadman HJ, editors. Impulsiveness and aggression. Violence and mental disorder: Developments in risk assessment.
Jackson JJ, Bogg T, Walton KE, Wood D, Harms PD, Lodi-Smith J, et al. Not all conscientiousness scales change alike: a multimethod, multisample study of age differences in the facets of conscientiousness. J Pers Soc Psychol. 2009;96: 446–459.
Paulhus. Balanced inventory of desirable responding (BIDR). Acceptance and Commitment Therapy Measures. 1988. Available: https://www.ueb.ro/psihologie/files/Psihologie_pozitiva_teste_si_scale.pdf#page=41
Greenwald AG, Nosek BA, Banaji MR. Understanding and using the Implicit Association Test: 1. An improved scoring algorithm. PsyArXiv. 2016. doi:10.31234/osf.io/acgxd
Karpinski A, Hilton JL. Attitudes and the implicit association test. J Pers Soc Psychol. 2001;81: 774–788.
Evans JSBT. Dual-processing accounts of reasoning, judgment, and social cognition. Annu Rev Psychol. 2008;59: 255–278.
Asendorpf JB, Banse R, Mücke D. Double dissociation between implicit and explicit personality self-concept: the case of shy behavior. J Pers Soc Psychol. 2002;83: 380–393.
Egloff B, Schmukle SC. Predictive validity of an implicit association test for assessing anxiety. J Pers Soc Psychol. 2002;83: 1441–1455.
Perugini M. Predictive models of implicit and explicit attitudes. Br J Soc Psychol. 2005;44: 29–45.
Wiers RW, Beckers L, Houben K, Hofmann W. A short fuse after alcohol: implicit power associations predict aggressiveness after alcohol consumption in young heavy drinkers with limited executive control. Pharmacol Biochem Behav. 2009;93: 300–305.
van Goethem AAJ, Scholte RHJ, Wiers RW. Explicit- and implicit bullying attitudes in relation to bullying behavior. J Abnorm Child Psychol. 2010;38: 829–842.
Cheng PYK. Improving financial decision making with unconscious thought: A transcendent model. J Behav Financ. 2010;11: 92–102.
Wilson TD. Strangers to ourselves. London, England: Belknap Press; 2009. doi:10.4159/9780674045217
Nordgren LF, Bos MW, Dijksterhuis A. The best of both worlds: Integrating conscious and unconscious thought best solves complex decisions. J Exp Soc Psychol. 2011;47: 509–511.
Dijksterhuis A. Think different: the merits of unconscious thought in preference development and decision making. J Pers Soc Psychol. 2004;87: 586–598.
Acker F. New findings on unconscious versus conscious thought in decision making: Additional empirical data and meta-analysis. Judgm Decis Mak. 2008;3: 292.
Gawronski B, Hofmann W, Wilbur CJ. Are “implicit” attitudes unconscious? Conscious Cogn. 2006;15: 485–499.
Molesworth BRC, Chang B. Predicting pilots’ risk-taking behavior through an implicit association test. Hum Factors. 2009;51: 845–857.
Dislich FXR, Zinkernagel A, Ortner TM, Schmitt M. Convergence of direct, indirect, and objective risk-taking measures in gambling. Z Psychol Z Angew Psychol. 2010;218: 20–27.
Park J, Kim D-Y, Oh S. Explicit and implicit stock investment: Differences in psychological characteristics and risky behavior between college students majoring in financial engineering or business. Curr Psychol. 2020;39: 1954–1969.
Kapteyn A, Teppa F, Others. Subjective measures of risk aversion and portfolio choice. Citeseer; 2002.
Dorn D, Huberman G. Talk and action: What individual investors say and what they do. Rev Financ. 2005;9: 437–481.
Goetzmann WN, Kumar A. Equity portfolio diversification. Rev Financ. 2008;12: 433–463.
Lampenius N, Zickar MJ. Development and validation of a model and measure of financial risk-taking. J Behav Financ. 2005;6: 129–143.
Hayes AF, Coutts JJ. Use omega rather than cronbach’s alpha for estimating reliability. But…. Commun Methods Meas. 2020;14: 1–24.
Gardner M, Steinberg L. Peer influence on risk taking, risk preference, and risky decision making in adolescence and adulthood: an experimental study. Dev Psychol. 2005;41: 625–635.
Strack F, Deutsch R. Reflective and impulsive determinants of social behavior. Pers Soc Psychol Rev. 2004;8: 220–247.
Niemand, Mai. Cold rationale and hot associations in product piracy consumption: The joint impact of implicit and explicit attitudes. Marketing theory and applications: Annual American. 2013.

No competing interests reported.

BMCAdditionalfilescombined.docx

An exploratory approach to the implicit and explicit assessment of risk: Benefits of their combined use for the prediction of risk attitudes and risky behavior

Status:

Version 1

Abstract

Background

Methods

Results

Conclusion

Figures

Introduction

Study 1

Study 2

General Discussion

Implicit risk attitudes as complementary assets

Extension Of The Implicit–explicit Interplay To Financial Risk Attitudes

Implications

Conclusion

Abbreviations

Declarations

References

Additional Declarations

Supplementary Files

Status:

Version 1