J U N S U PA R K
A UNIFIED APPROACH TO THE UNDERSTANDING OF DECISION UNDER RISK
D I S S E R TAT I O N | 2017
Junsu Park
A UNIFIED APPROACH TO THE
UNDERSTANDING OF DECISION UNDER RISK
IMPLICATIONS FOR DIVERSITY MANAGEMENT
University of Eastern Finland Joensuu/Kuopio
2017
Park, Junsu
A unified approach to the understanding of decision under risk. Implications for diversity management.
Joensuu: Itä-Suomen yliopisto, 2017
ABSTRACT
Applying implicit social cognition to risk research, I performed a series of studies for the purpose of (a) cross-national and (b) inter-generational comparisons of implicit as well as explicit attitudes toward risk and (c) an assessment of the predictive validity of a newly developed implicit measure of stock investment.
Study 1 demonstrated a possibility of utilizing the implicit measures of risk attitude as a useful tool complementary to the existing self-report risk measures for understanding cross-cultural similarities and differences in risk. Study 2 replicated the utility of the implicit measures of risk attitude in the form of the known-group validity by showing the marked age differences that cannot be captured by the corresponding explicit measures. Study 3 illustrated that the implicit stock investment measure, designed to measure the associative strength of My stock investment with Aggressive- versus Conservative-related words, provided incremental variance in stock investment performance of undergraduates majoring in financial engineering when interacting with explicit measures of financial risk taking. The theoretical and practical implications of the findings are discussed.
Keywords: Risk, Risk Taking, Implicit, Explicit, Diversity, Differences
Park, Junsu
Yhtenäinen lähestymistapa ymmärrystä päätöksen vaaranalaisilta. vaikutuksista monimuotoisuuden hallinnassa
Joensuu: Itä-Suomen yliopisto, 2017
TIIVISTELMÄ
Avainsanat: Riski, Riskinotto, Implisiittinen, Täsmällinen, Monimuotoisuus, Erot
ACKNOWLEDGEMENTS
University of Eastern Finland, XX May 2017 Junsu Park
CONTENT
ABSTRACT ... 5
SUMMARY ... 6
ACKNOWLEDGEMENTS ... 7
1 INTRODUCTION ... 11
1.1 Research background and purposes ...11
1.2 The present study ...12
2 STUDY 1 ...
...142.1 Cross-national differences in explicit and implicit evaluations of risk...14
2.2 Method ...14
2.1.1 Sample ...16
2.1.2 Measures ...16
2.3 Results ...14
2.3.1 Construct convergence and divergence assessment ...16
2.3.2 National differences in explicit and implicit measures of risk ...20
2.3.3 Explicit and implicit evaluations of risk: gain vs. loss ...22
2.4 Discussion ...23
3 STUDY 2 ... 24
3.1 Age and risk taking ...24
3.2 Method ...25
3.1.1 Participants and procedure ...25
3.1.2 Measures ...25
3.3 Results ...26
3.3.1 Construct divergence assessment ...26
3.3.2 Age differences in explicit and implicit measures of risk ...31
3.3.3 Mediating role of psychosocial maturity on age - risk relationship....32
3.4 Discussion ...34
4 STUDY 3 ... 35
4.1 Concerns about self-report measures of attitudes toward stock investment ...35
4.2 Integrating explicit and implicit risk taking in stock investment ...35
4.3 Method ...38
4.1.1 Participants ...38
4.1.2 Procedure ...38
4.1.3 Measures ...38
4.4 Results ...40
4.4.1 Moderating effect of the ISI on stock investment performance ... 40
4.5 Discussion ... 43
5 IMPLICATIONS ... 43
6 SUGGESTIONS FOR FUTURE RESEARCH ... 44
7 REFERENCES ... 46
8 APPENDICES ... 50
9 PUBLICATIONS ... 54
LIST OF TABLES
Table 1. Correlations among all measures of risk for the pooled samples ...18
Table 2. Summary of confirmatory factor analyses results from the raw data across nations ...20
Table 3. Cross-national differences in risk measures ...21
Table 4. Correlations among all measures of demographics, personality, and risk across different age groups ...27
Table 5. Age differences in risk measures across different age groups ...31
Table 6. Correlations among all study variables ...41
Table 7. A moderating analysis of the ISI on rate of return...42
LIST OF FIGURES
Figure 1. Confirmatory factor analyses of risk construct divergence among the pooled samples ...19Figure 2. Confirmatory factor analyses of risk construct divergence for the Kore- an older age group ...29
Figure 3. Confirmatory factor analyses of risk construct divergence for the Korean younger age group ...30
Figure 4. The partially mediated model among the older age group ...33
Figure 5. The fully mediated model among the older age group ...34
Figure 6. Graphically illustrated interaction effect of the ISI and FRT-RC on rate of return ...43
1 INTRODUCTION
1.1 RESEARCH BACKGROUND AND PURPOSES
Over the past 70 years, attempts to understand how people make decision under risk have been elaborated by many scholars so that significant accounts have been documented through the different lens ranging from economic (expected utility theory; von Neumann & Morgenstern, 1944), cognitive (prospect theory; Kahneman
& Tversky, 1979), and affective (regret theory; Loewenstein, Weber, Hsee, & Welch, 2001), to motivational (regulatory focus theory; Scholer, Fujita, Stroessner, &
Higgins, 2010) perspectives. Such an eclectic approach to the analysis of decision under risk has provided a valuable contribution in the broad stream of management research including negotiations, emotion and motivation, human resource management, organizational risk – return relations, and strategic risk- taking behaviors (Holmes, Bromiley, Devers, Holcomb, & McGuire, 2011).
Nevertheless, two important concerns can be raised regarding their direct assessment of evaluations with standard self-report measures of risk; choice sets involving two risky prospects are a good example. First, the above studies present an over-reliance on a person’s conscious judgmental processes of evaluating expected utility or perceived value that s/he forms based on information about possible outcomes and their probabilities. However, if the way people process information about themselves and their environment operates not only in a deliberately monitored manner, but also in a spontaneously activated manner (Evans, 2008), participants’ self-reports of choices between risky prospects may not reflect their non-conscious or automatic evaluations of outcome and probability information due to lack of ability to report on these cognitive and affective processes that occur outside conscious control (i.e., introspective limits; Greenwald, Banaji, Rudman, Farnham, Nosek, & Mellott, 2002). Rather, the reports might reflect their conscious or controlled evaluative judgments about temporarily given information, that is, outcomes and probability that are not only anchored at a certain point, but also disparate across studies, which may limit generalizability of findings explored in any particular study.
Besides, earlier studies on implicit social cognition suggest that the cognitive processes operating outside of awareness (i.e., implicit processes) are also actively involved in the development of behavioral responses and psychological outcomes.
In fact, a meta-analysis of 152 studies has provided significant evidence for mutual incremental validity of explicit (self-report) and implicit (IAT; Implicit Association Test; Greenwald, McGhee, & Schwartz, 1998) measures, indicating that each measure provided unique variance in criterion behavior that was not accounted for by the other across different domains including gender orientation, personality
use, and clinical phenomenon ( = .18 and = .32, respectively – see Greenwald, Poehlman, Uhlmann, & Banaji, 2009). Also, recent works on behavioral finance provide support for the dual (i.e., explicit and implicit) cognitive approach to risk.
As an example, a transcendent model developed by Cheng (2010) suggests that an integration of unconscious thought into the existing explicit decision-making process potentially improves the quality of financial judgment and investment decision by compensating for inherent limitations (e.g., cognitive overload) of conscious thought on information processing capacity.
Another issue with self-report measures of risk relates to a potential susceptibility to response bias, which can be also applied to binary-choice between risky prospects. Normatively, people should base their choices on unbiased assessments of relevant information such as a description of possible outcomes and their probabilities in risky choices. However, a series of studies involving risk choices under six hypothetical scenarios has found both student and non-student participants to be biased in their evaluations of outcome and probability information in a way favorable to their preferred decision alternative, and subsequently, these biased evaluations are again used to update their preferences (DeKay, Patiño-Echeverri, & Fischbeck, 2009). In addition, social desirability bias has also been identified as one of the most common and pervasive factors that might diminish the accuracy of self-report tests in a variety of risk-related topics such as socially approved risk-taking and reckless behavior (Bradley & Wildman, 2002; Ronay & Kim, 2006), health-risk behavior (Brener, Billy, & Grady, 2003), consumer innovativeness (Tellis, Yin, & Bell, 2009), and sexual behavior (Kelly, Soler-Hampejsek, Mensch, & Hewett, 2013).
Taken together, the problems of introspective limits and response factors suggest that the construct of interest – in this case, risk – needs to be indirectly measured by using the IAT that is known to circumvent introspective limits (Egloff
& Schmukle, 2002), social desirability bias (Banse et al., 2001), and even voluntary distortion (Kim, 2003) to assess an additional construct of risk that is distinct, but related, to self-report assessments.
1.2 THE PRESENT STUDY
A primary goal of the present study is to introduce a new perspective known as implicit social cognition (Greenwald & Banaji, 1995) to risk research, hoping to complement extant economic, cognitive, affective, and motivational approaches to decision under risk by using two IAT variants that have been designed to target a domain-general and domain-specific risk construct – Implicit Risk Task (IRT; Ronay
& Kim, 2006) and Implicit Stock Investment (ISI; Park, Kim, & Oh, 2017), respectively. Encouraged by the fact that the IAT reflects the strength of automatic concept – attribute associations a person has experienced in their personal, societal, and cultural environment (Karpinski & Hilton, 2001; Olson & Fazio, 2004), I believe the implicit measures of risk may allow me to not only provide a thorough
comparison for cross-national and socio-generational differences in risk that have been documented from previous studies using self-report measures, but also improve explanatory power for individual variations in risk-taking behaviors.
For this reason, I perform a series of studies with different purpose. By comparing South Korean and Chinese (referred to as collectivists), and Australian (referred to as individualists) undergraduate samples, Study 1 analyzed: (i) whether construct divergence between explicit (self-report) and implicit (IRT) measures of risk is cross-nationally validated; and (ii) how the three national groups differ in their conscious and non-conscious evaluations of risk; and (iii) whether cross-national differences observed from a recent study using explicit measures (Park, Kim, & Zhang, 2016) are replicated in implicit measures of risk.
Study 2 also tested (i) construct divergence in a sample of Korean adults aged 30 – 50 years; and (ii) whether the IRT indeed captures socio-generational differences in risk, which have been hardly detected from self-report measures of risk, by comparing Korean undergraduate and adult samples. Finally, a predictive power of the ISI on stock investment performance was tested in Study 3 examining a sample of financial engineering-major undergraduates, with a particular emphasis on a complementary interplay between explicit and implicit measures of stock investment that is presumed to improve the quality of financial judgement and investment decision (Cheng, 2010).
2 STUDY 1
2.1 CROSS-NATIONAL DIFFERENCES IN EXPLICIT AND IMPLICIT EVALUATIONS OF RISK
Explicit risk. To characterize cross-national differences in one’s conscious evalu- ations of risk, a recent study using self-report measures of risk compared Australi- an, South Korean, and Chinese student samples and then found the three national groups to be reliably different (Park, Kim, & Zhang, 2016). The research results exhibited that Australian individuals were more pro-risk than were South Korean individuals, who in turn self-reported a greater pro-risk position than did Chinese individuals on two attitudinal measures of risk, represented by an abstract con- struct (i.e., Risk) and personally relevant risk activities (e.g., drink-driving). The authors explained the observed national differences in terms of both the culture theory (Douglas, 1992; Wildavsky, 1987) and Schwartz (1992)’s value theory that have often been adopted to explain cross-national differences in risk. Specifically, people from individualistic cultures (e.g., Australia, reviewed in Chai, Liu, & Kim, 2009) are socially facilitated to challenge uncertainty by taking responsibility for potential negative consequences of their decisions due to their cultural characteris- tics represented by few externally mandated social constraints (e.g., rules, laws, and group norms) on individual acts. These socio-cultural forces drive individualists to be self-assured, self-determined, and self-enhanced when expressing their opinions with regard to socially desired attributes (i.e., risk), thereby resulting in a pro-risk position on risk-relevant topics. On the other hand, people from strongly hierar- chical and bureaucratic cultures (e.g., South Korea and China, reviewed in Chai et al., 2009), which are characterized by many externally mandated social restrictions, are socially emphasized to behave in a way that regards individualistic acts as de- viant and aligns themselves with pre-existing rules, procedures, and customs of society. Hence, people from these cultures may find it difficult to express innova- tive ideas that go beyond complying with socio-culturally defined routines. For example, Yao and colleagues (2010) found Chinese employees to have difficulties in converting new ideas into innovative behaviors due to a heavy emphasis of socially desired norms – in this case, Zhong Yong (i.e., Doctrine of the Mean). The socio- cultural dissimilarity might lead to differences in uncertainty avoidance between individualistic and collectivist cultures (Hofstede, 1983), indicating a relatively higher level of risk tolerance among individualists than among collectivists.
As part of efforts to go beyond individualist – collectivist comparison in risk, within-collectivist cultural differences in risk between the South Korean and Chi- nese samples were further explained by Part et al. (2016) drawing on each country’s idiosyncratic history. Although South Korean and Chinese societies share hierar- chical cultural norms under one specific superordinate cultural typology (i.e., col-
lectivism), the Chinese individuals having less favourable attitudes toward risk than the South Korean individuals may be accounted for by the relatively belated China’s economic reform and opening-up program in which not only do foreign investors make various inroads to the Chinese market, Chinese people also run their private businesses since the late 1970s (Deng, 1994). These facts imply that Chinese individuals were restrained from initiating challenging activities involving risk due to the strict government controls under a centrally-administered economy.
This plausible explanation seems to be consistent with their findings.
Implicit risk. The comparison of risk between Australian, South Korean, and Chinese samples on self-report measures is of theoretical value because it suggests that individuals’ conscious evaluations of risk are colored by each country’s indige- nous sociocultural environments associated with risk. In addition to evaluating a person’s conscious judgmental thoughts on risk across countries, the present study further investigated how the three national groups differ in implicit (non-conscious or automatic) evaluations of risk by reproducing unpublished IAT data from the same Australian, South Korean and Chinese student samples as used in the previ- ous study (Park, Kim, & Zhang, 2016).
Insofar as individuals’ attitudes, beliefs, and norms result from the socio- cultural environment to which they have been exposed over long periods of time (Fiske, Kitayama, Markus, & Nisbett, 1998), such psychological products of the culture remain unconscious to people who hold them (Hofstede, Hofstede, &
Minkov, 2010) and thus might be well reflected by the automatic nature of the IAT task (Arkes & Tetlock, 2004; Karpinski & Hilton, 2001; Olson & Fazio, 2004). In fact, additional findings in the literature on implicit measures provide support for this argument. As an example, Swanson, Rudman and Greenwald (2001) found smokers to be ambivalent in terms of attitude – behaviour consistency, indicating that when asked explicitly, smokers revealed pro-smoking attitudes that are consistent with their smoking behavior, but surprisingly appeared negative in their implicit atti- tudes toward smoking so that these implicit attitudes are at odds with their smok- ing behavior. In the domain of racial prejudice, White and Black participants yield- ed IAT results that suggested that about 80% of White participants implicitly re- vealed preferences for White over Black while Black participants do not reveal such an implicit preference for the in-group (Nosek, Banaji, & Greenwald, 2002).
Similarly, if indeed people in different countries learns attitudes, beliefs, and norms regarding risk in a locally informed manner consistent with each country’s indigenous sociocultural norms (Park, Kim, & Zhang, 2016), different values of risk transmitted by the culture in each country should be well-learned and activated automatically when the potential gains and losses are associated with a concept of risk (Banaji, 2001; Lowery et al., 2001; Olson & Fazio, 2003). As a result, measures of implicit risk that are based on associations between a target concept (risk) and an attribute (gain vs. loss) may reflect widely shared cultural values of risk and should
be reflected in IAT performance. For this reason, this study proposes the following hypothesis:
Hypothesis 1 (H1): The implicit measures of risk exhibit the same pattern of cross- national differences in risk as reported from the previous study using the corresponding explicit risk measures where Australian, South Korean, and Chinese individuals were relia- bly ranked in order regarding favoring risk on their attitudes.
2.2 METHOD
2.2.1 Sample
The data for seven hundred and one subjects across three different nations (i.e., South Korea, Australia, and China) were reproduced with permission from the previously published studies, with a particular emphasis on the implicit measures of risk (i.e., the IRT-Global and the IRT-Unique). The samples consisted of 283 South Korean (135 males, mean age = 22.73, SD = 2.20; Kim & Park, 2010), 126 Australian (66 males, mean age = 22.7, SD = 5.7; Ronay & Kim, 2006), and 292 Chinese undergraduates (137 males, mean age = 21.04, SD = 2.25; Park, Kim, &
Zhang, 2016) who were recruited from a co-ed university in each country.
2.2.2 Measures
The same measures as employed in the previous study (Ronay & Kim, 2006) were applied to the other studies examining the South Korean (Kim & Park, 2010) and Chinese (Park, Kim, & Zhang, 2016) samples. They were originally developed in English, and were translated into Korean and Chinese. To optimize the translated questionnaires’ accuracy, a forward-backward translation process was employed following the translation technique recommended by Behling and Law (2000).
Sensation Seeking Scale. The SSS is a 40-item measure that assesses a person’s dispositional trait underlying risk-taking behaviors (Arnett, 1994). For this reason, this scale was adopted to examine whether there are shared variance between the SSS and the implicit measures of risk and showed a reliability of Cronbach’s alpha = .73 for the South Korean sample, α = .80 for the Australian sample, and α = .71 for the Chinese sample. Higher scores on the SSS indicate “the willingness to take physical and social risks for the sake of sensational experiences” (Zuckerman, 1979, p. 10).
Risk Global-Semantic Differential Scale. The RG-SDS is a self-reported attitudinal measure that assesses participants’ overall evaluations of a global construct of risk. They were asked to indicate their attitudes toward risk using 7- point scales that were anchored at either end by two polar opposite adjectives (e.g., loss vs. gain, mistake vs. success, penalty vs. benefit, waste vs. achieve, lose vs.
reward, cost vs. profit, and failure vs. win). This scale presented a reliability of Cronbach’s alpha = .90 for the South Korean sample, α = .85 for the Australian
sample, and α = .74 for the Chinese sample (see Appendix A). Higher scores on the RG-SDS show a more positive attitude toward risk.
Risk Unique-Semantic Differential Scale. The RU-SDS is a variant of the RG- SDS, in which the target, in this case, risk, is replaced with self-relevant risk activities. The RU-SDS was used because people are often involved in risk-taking behaviors (e.g., drink-driving) despite having a negative perception of overall (global) risk. Participants were instructed to select ten risky activities that were related to their lives from 34 risky behaviors, after which the ten items were reduced to the six activities most relevant to each participant. The six items were used as evaluation categories using the same format as used in the RG-SDS. This scale presented a reliability of Cronbach’s alpha = .95 for the South Korean sample, α = .95 for the Australian sample, and α = .88 for the Chinese sample (see Appendix B). Higher scores on the RU-SDS show a more positive attitude toward self-selected risk activities.
The Implicit Risk Task-Global. The IRT-Global (Ronay & Kim, 2006) is a variant of IAT that assesses an automatic evaluation of a global construct of risk. The IRT is regarded as a categorization task that involves stimulus items from RISK (i.e., risk) and BLANK (i.e., [ ]) categories as well as items from GAIN (i.e., gain, win, success, benefit, achieve, reward, and profit) and LOSS (i.e., loss, failure, mistake, penalty, waste, lose, and cost) categories. Respondents were informed to press either a left (“a”) or right (“5”) key to classify a stimulus word presented at the center of a computer screen as quickly and correctly as possible into one of the categories.
The IRT-Global procedure contains a sequence of five blocks (see Appendix C).
In the first and second blocks, the respondents were allowed to practice the discriminations of either RISK and BLANK stimuli or GAIN and LOSS stimuli. In the third block, the respondents sorted the stimulus items into two combined categories where the target concept and evaluative word shared the same key (e.g., RISK + GAIN for the right key, and [ ] + LOSS for the left key). In the fourth block, the same stimulus items for the target category as used in the first block were shown again, but their key assignment was switched. In the last block, the respondents did the same as in the third block, but one big difference was that pairings were switched (e.g., [ ] + GAIN for the right key, and RISK + LOSS for the left key). As a result, the effect of IRT-Global was generated by subtracting the mean performance for one combined task of “RISK + GAIN” and “[ ] + LOSS”
from that for the complementary combined task of “[ ] + GAIN” and “RISK + LOSS.” A negative IRT-Global effect indicates a participant’s greater strength of automatic association between risk and loss than between risk and gain. This scale presented a reliability of Cronbach’s alpha = .89 for the South Korean sample, α = .73 for the Australian sample, and α = .93 for the Chinese sample.
The Implicit Risk Task-Unique. The IRT-Unique (Ronay & Kim, 2006) is a variant of IRT-Global that assessed participants’ automatic evaluations of self-
selected risk activities. The respondents were asked to enter the six items (e.g., tattoo and snowboarding) that they considered most relevant to their lives, and then, the stimulus, “RISK,” in all blocks was automatically replaced with the six items. Except for this case, “RISK” and “[ ]” for the target concept and attribute words for “GAIN” and “LOSS” were kept identical to those used in the IRT-Global.
The IRT-Unique effect was calculated the same way as for IRT-Global. This scale presented a reliability of Cronbach’s alpha = .91 for the South Korean sample, α = .95 for the Australian sample, and α = .93 for the Chinese sample.
2.3 RESULTS
2.3.1 Construct convergence and divergence assessment
Pooling across cultures, the correlations among all measures employed in this study are shown in Table 1. Consistent with the finding from the previous study (Ronay
& Kim, 2006), the IRT-Global and the IRT-Unique were found to be significantly correlated with their parallel explicit measures – the RG-SDS, r = .14, p = 10-5 and RU-SDS, r = .29, p = 10-15, respectively. This study further found that both the IRT- Global and the IRT-Unique weakly but significantly correlated with the sensation seeking scale which has been reported as an important indicator of risk-taking (Wong & Carducci, 1991), r = .12, p = 10-3, and r = .11, p = 10-3, respectively.
Table 1. Correlations among all measures of risk for the pooled samples
Measure 1 2 3 4 5 6
Personality
1. IM -
2. SD .25** -
3. SSS -.25** .11** -
Explicit risk
4. RG-SDS -.11** .18** .20** -
5. RU-SDS -.05 .17** .12** .55** -
Implicit risk
6. IRT-G -.09* .02 .12** .14** .09* -
7. IRT-U -.09* .09* .11** .26** .29** .48**
Note. N=653. IM = Impression Management; SD = Self-Deception; SSS = Sensation Seeking Scale; RG-SDS = Risk Global – Semantic Differential Scale; RU-SDS = Risk Unique – Semantic Differential Scale; IRT-G = Implicit Risk Task – Global; IRT-U=
Implicit Risk Task – Unique; *** p < .001; ** p < .01; * p < .05; † p < .1
Coupled with the fact that the implicit measures of risk not only shared convergence with the corresponding explicit risk measures, but also showed theoretically expected associations with sensation seeking and openness to experience (except for the IRT-Unique – openness to experience relation), this study
decided to further explore a divergence of construct represented by the explicit and implicit measures of risk. As a summary of Table 1, utilizing r to z conversion (Campbell & Fiske, 1959), the correlations between measures of the different traits using same methods (method effects) were rexp =. 55 for the RG-SDS – the RU-SDS and rimp = .48 for the IRT-G – the IRT-U, respectively. In addition, the average correlation between measures of the same traits using different methods (i.e., between rRG-SDS – IRT-G and rRU-SDS – IRT-U) (trait effects) was rexp-imp = .22. The correlational data reflecting weaker trait effects than method effects suggested construct divergence between explicit and implicit measures of risk (Campbell &
Fiske, 1959). For this reason, this study has attempted to conduct two confirmatory factor analyses (CFAs) controlling for culture and gender variables to directly test whether the data for the explicit and implicit measures of risk were better suitable for a model with two factors (i.e., explicit and implicit risk; Figure 1, right panel) or one (risk; Figure 1, left panel).
Figure 1. Confirmatory factor analyses of risk construct divergence among the pooled samples
Note. RG-SDS = Risk Global – Semantic Differential Scale; RU-SDS = Risk Unique – Semantic Differential Scale; IRT-G = Implicit Risk Task – Global; IRT-U = Implicit Risk Task – Unique.
The results of the CFAs exhibited that a model in which explicit and implicit risk measures converged on a single construct (i.e., one-factor model) had a noticeably poor fit, χ2(11, N = 653) = 175.44 (p < .05), CFI = .78, NFI = .77, TLI = .44, RMSEA = .15, p(close fit) < .05. Conversely, a model that defined explicit and implicit risk as separate, but correlated, factors (i.e., two-factor model) showed a close-to-fair fit,
χ2(7, N = 653) = 25.43 (p < .05), CFI = .98, NFI = .97, TLI = .90, RMSEA = .06, p(close fit) = .20. A chi-square test of significance for the difference between the two nested models exhibited that the two-factor model provided a significantly better fit than did the one-factor model, χ2diff(4, N = 653) = 150.01 (p < .05). In the two-factor model, the estimated correlation between the two error terms of the explicit and implicit risk factors was .26 (p < .01).
Additional CFAs controlling for gender variable were performed to explore whether the divergent validity observed from the pooled data was replicated in each nation. Table 2 presents various fit indices and a summary of the chi-square difference test results comparing the nested models across nations.
Table 2. Summary of confirmatory factor analyses results from the raw data across nations
Nation South Korea Australia China
Model 1
factor 2 factors
1 factor
2 factors
1 factor
2 factors χ2
(d.f.)
42.37 (5)
9.54 (4)
13.19 (5)
1.43 (4)
57.75 (5)
5.55 (4)
CFI .78 .97 .80 1.00 .75 .99
NFI .95 .95 .76 .97 .74 .98
TLI .88 .88 .39 1.24 .25 .97
RMSEA p(closefit)
.16 (<.05)
.07 (.23)
.11 (.07)
.00 (.90)
.19 (<.05)
.04 (.55)
Δχ2 /Δd.f. 32.83/1*** 11.76/1*** 52.21/1***
Note. * p < .05, ** p < .01, *** p < .001
Most notably, a chi-square test of significance for the difference between the two nested models revealed a reliable pattern across South Korea, Australia, and China, indicating that the two-factor model yielded a significantly better fit than did the one-factor model.
2.3.2 National differences in explicit and implicit measures of risk
Table 3 presents the means and standard deviations for the RG-SDS, RU-SDS, IRT- G and IRT-U across Australian, South Korean, and Chinese samples and suggests cross-national differences in explicit and implicit measures of risk.
Table 3. Cross-national differences in risk measures
Measures Australians South Koreans Chinese F(df1, df2) Explicit Risk
RG-SDS 4.652a
(1.12)
3.642b (1.31)
2.893c (1.38)
78.73 ***
(2, 684)
RU-SDS 3.532a
(1.09)
3.138b
(1.02)
2.752c
(1.07)
25.30 ***
(2, 694) Implicit Risk
IRT-G -.135a
(.36)
-.183a (.41)
-.257b (.33)
5.12 **
(2, 673)
IRT-U -.131a
(.42)
-.292b (.40)
-.376c (.38)
15.08 ***
(2, 664) Note.* p < .05, ** p < .01, *** p < .001. Standard deviationsare denoted in parentheses below means. Means with differing subscripts within rows are significantly different across the three national groups at p < .05.
Explicit risk. To determine whether there were national differences between the three groups, a one-way analysis of variance controlling for gender variable1 was performed on the RG-SDS. There was a significant gender effect, F(1, 684) = 5.67, p = .02, ŋ2 = .01, suggesting that males generally had stronger pro-risk attitude than females. As expected, a significant main effect for nation was found, F(2, 684) = 78.73, p = 10-32, ŋ2 = .19. Follow-up planned comparisons revealed that the Australian participants exhibited stronger pro-risk attitudes than the South Korean participants, t(684) = 6.99, p = 10-13, d = .83, and Chinese, t(684) = 12.31, p = 10-32, d = 1.40. This pattern was also observed for the collectivistic cultural groups, such that South Korean participants were more risk-tolerant than Chinese participants, t(684)
= 6.80, p = 10-12, d = .56. This pattern was replicated in the RU-SDS. The test revealed a marginally significant main effect for gender, F(1, 694) = 3.52, p = .06, ŋ2 = .01, however as expected, a significant main effect for nation was noted, F(2, 694) = 25.30, p = 10-12, ŋ2 = .07: Australian versus South Korean individuals, t(694) = 3.42, p = 10-5, d = .37; Australian versus Chinese individuals, t(694) = 6.83, p = 10-12, d = .72;
South Korean versus Chinese individuals, t(694) = 4.38, p = 10-6, d = .37.
Implicit risk. The same question as raised in the explicit measures of risk was tested performing a one-way analysis of variance controlling for gender variable on
1 Unlike the previous study (Park, Kim, & Zhang, 2016), the present study included gender as a control variable in the analyses to cancel out the effect of gender on group differences in risk across the three national groups.
the implicit measures of risk. First on the IRT-G, the test results showed a significant main effect for gender, F(1, 673) = 5.52, p = .02, ŋ2 = .01, suggesting the IRT effect for males to be less negative than that for females. As expected, there was a significant main effect for nation, F(2, 673) = 5.12, p = .01, ŋ2 = .02. Follow-up planned comparisons revealed that Australian individuals showed less negative automatic evaluation of risk associated with loss than Chinese individuals, t(673) = 2.83, p = .01, d = .35, but not South Korean individuals, t(673) = 1.07, p = .29, d = .12. A similar pattern was observed for the collectivistic cultural groups, such that South Korean individuals showed less negative automatic evaluation of risk associated loss than Chinese individuals, t(673) = 2.39, p = .02, d = .19.
The national differences were highly salient to the IRT-Unique. A significant main effect was found for gender, F(1, 664) = 7.77, p = .01, ŋ2 = .01, suggesting males’
automatic evaluations of self-selected risk items associated with loss to be less negative than those of females. As expected, there was a significant main effect for nation, F(2, 664) = 15.08, p = 10-8, ŋ2 = .04. Follow-up planned comparisons revealed that Australian individuals showed less negative automatic evaluation of personally relevant risk items associated with loss than Chinese individuals, t(664)
= 5.48, p = 10-9, d = .62, as well as, South Korean individuals, t(664) = 3.53, p = 10-5, d = .40. This pattern was also reliable in South Korean versus Chinese individuals, t(664) = 2.48, p = .01, d = .22.
2.3.3 Explicit and implicit evaluations of risk: gain vs. loss
Explicit risk. To examine whether the participants lean toward gain or loss on their explicit evaluations of risk, a one sample t-test was performed across three national groups. On the RG-SDS, the test results indicated that the Australian indi- viduals’ evaluations of risk were found to be significantly closer to gain than loss, t(119) = 6.36, p = 10-10, d = .58, while the South Korean and Chinese individuals’
evaluations of risk were found to be significantly closer to loss than gain, t(276) = - 4.54, p = 10-7, d = .27, and t(290) = -13.68, p = 10-34, d = .80, respectively. Conversely, on the RU-SDS, the evaluations of all respondents were closer to loss than gain on self- selected risk activities: t(281) = -14.18, p = 10-36, d = .85 for the South Korean individ- uals, t(124) = -4.83, p = 10-7, d = .43 for the Australian individuals, and t(290) = -19.84, p = 10-57, d = 1.16 for the Chinese individuals.
Implicit risk. The same question as raised in the explicit measures of risk was tested performing a one sample t-test on the implicit measures of risk. The test re- sults revealed that all IRT-effects were in a negative direction, which is reliable across nations. In other words, the participants responded significantly faster for the groupings of “RISK + LOSS” and “[ ] + GAIN” combination than for the group- ings of “[ ] + LOSS” and “RISK + GAIN” combination, indicating stronger auto- matic association of risk with loss than gain: t(278) = -7.49, p = 10-14, d = .45 for the South Korean individuals, t(106) = -3.87, p = 10-5, d = .37 for the Australian individu- als, t(290) = -13.13, p = 10-32, d = .77 for the Chinese individuals. This pattern was also
consistently detected in the IRT-Unique, indicating stronger automatic associations of personally selected risk activities with loss than gain: t(268) = -12.01, p = 10-28, d = .73 for the South Korean individuals; t(106) = -3.25, p = 10-3, d = .31 for the Australian individuals; t(291) = -17.04, p = 10-46, d = 1.00 for the Chinese individuals.
2.4 DISCUSSION
I found risk to be a multi-faceted construct by showing a better fit for the two-factor model that defined explicit and implicit risk as distinct, but correlated, factors over the one-factor model where explicit and implicit risk converged on a single factor.
Also, the divergence of construct represented by explicit and implicit measures of risk was cross-nationally validated. These findings suggest the need for a multi- faceted (explicit and implicit) approach to the analysis of psychological responses to risk-relevant variables, which may allow us to reach a full understanding of within- and between-subject variations in risk. In fact, the Australian individuals considered risk as an opportunity for potential gains while the South Korean and Chinese individuals appeared negative toward risk on the explicit measure (RG- SDS), which is a different pattern observed from the corresponding implicit measure (IRT-Global) in which all participants, regardless of nationality, exhibited faster reaction times when risk was paired with loss versus gain. One plausible account for one’s stronger implicit cognitive associations of risk with loss than gain can be derived from the economic analysis of attitudes toward risk within the expected utility framework, suggesting that people are generally reluctant to take risks and thus their attitudes toward risks are expressed in the form of the utility function that is generally concave-shaped (i.e., risk-averse attitude; Blais & Weber, 2006; Elton, Gruber, Brown, & Goetzmann, 2009; Weber, Shafir, & Blais, 2004). This alternative explanation is in line with the findings of this study.
Besides within-subject variations in risk, it is also important to understand between-subject (cross-national) differences in implicit as well as explicit evaluations of risk. Despite the fact that the effect of gender was partialed out, the same pattern of cross-national risk differences as reported from the previous study (Park, Kim, & Zhang, 2016) was also shown in this study, indicating that risk was the most favored by the Australian individuals, followed by the South Korean and Chinese individuals on the explicit risk measures. Similarly, the systematic differences were partially but not fully captured by the implicit measures of risk (IRT-Global and IRT-Unique), exhibiting that the Australian individuals showed weaker negative evaluation of risk than do the South Korean individuals who in turn were less anti-risk than were the Chinese individuals – with the exception of the IRT-Global comparing the Australian and South Korean individuals. I surmise that these findings may suggest the possibility that the implicit measures of risk provide a reflection of cultural beliefs, norms, and values regarding risk that have been accumulated over a person’s lifetime through long-term societal exposure in
the national environment. For this reason, I attempted to perform a further investigation into a comparison of two Korean cohorts born an average of 16 years apart to clarify whether accumulated experience of risk taking each cohort group gained from the set period of time is indeed reflected in the implicit measures of risk and thus results in socio-generational differences in implicit evaluations of risk.
3 STUDY 2
3.1 AGE AND RISK TAKING
Consistent with real-life statistics reflecting higher rates of risk-taking behavior among adolescents and younger adults relative to older adults (Centers for Disease Control & Prevention, 2006), significant evidence provides support for the notion that risk taking decreases with age by showing them to be negatively correlated not only in a pooled sample of different age groups (Chaubey, 1974; Okun & Di Vesta, 1976), but also in a specific sample of managers and executives across different domains and companies (Vroom & Pahl, 1971; MacCrimmon & Wehrung, 1990).
However, these types of research findings appear somewhat at odds with relatively few age differences reported in cognitive factors that might account for why adolescents and younger adults are more involved in risk taking than older adults such as individuals’ evaluations of their vulnerability to risk, judgments of seriousness associated with potential consequences of risky behavior, and the ways of evaluating the relative costs and benefits inherent in risky activities (Beyth- Marom, Austin, Fischoff, Palmgren, & Jacobs-Quadrel, 1993; Millstein & Halpern- Felsher, 2002).
One plausible account for these puzzling findings is that the cognition-focused approach to age differences in risk overlook the influence of psychosocial factors (Steinberg, 2007). From a developmental neuroscience perspective, risk taking in the real world is determined by the use of both logical reasoning and psychosocial maturity (Steinberg, 2004). Contrary to logical-reasoning abilities that are considered fully mature by the age of 15 (Reyna & Farley, 2006), psychosocial maturity capacities – such as impulse control, emotion regulation, and delay of gratification that are negatively associated with risk-taking behaviors (Stanford, Greve, Boudreaux, Mathias, & Brumbelow, 1996; Romer, Duckworth, Sznitman, &
Park, 2010; Heilman, Crisan, Houser, Miclea, & Miu, 2010) – develop gradually throughout adulthood (Steinberg, 2004). For example, people are more likely to be exposed to social environments in which they should be self-disciplined, responsible for the welfare of in-group members (e.g., family and co-workers) and furthermore aligned with social norms and rules as they get older, married, and even promoted. These socio-emotional experiences accumulated over the past decade allow older adults to possess a higher level of psychosocial maturity that
younger adults. If indeed the implicit measures of risk well-capture a person’s non- conscious and automatic evaluation that reflects accumulated socio-emotional experiences associated with risk taking (Nosek, 2007), I expect older adults to exhibit stronger negative implicit associations with risk than adolescents and younger adults and thus marked age differences in risk might result. Accordingly, the present study further investigated how the older and younger adult groups differ in implicit and explicit evaluations of risk by comparing data from Korean young adult sample reported in two previous studies (Kim & Park, 2010; Park, Kim,
& Zhang, 2016) to newly collected data obtained from a Korean older adult sample.
For this reason, this study proposes the following hypothesis:
Hypothesis 2 (H2): The implicit measures of risk exhibit marked socio-generational differences between the two Korean cohorts, indicating older adults’ less favorable implicit evaluations of risk than adolescents and young adults.
3.2 METHOD
3.2.1 Participants and procedures
I recruited a total of sixty-two Korean adults (52 males, mean age = 38.85, SD = 7.64) who were enrolled in the university’s classroom-based MBA program as partici- pants in the older age group and then asked them to complete the same explicit (self-report) and implicit risk measures as used in the Study 1 along with three ad- ditional measures of risk choice – Choice Dilemma Questionnaire [CDQ] – and risk- relevant personality traits – Barratt Impulsiveness Scale [BIS] and NEO Five Factor Inventory-Conscientiousness [NEO FFI-C]. The newly collected data from the older age group were compared with the corresponding data for the younger age group (a total sample size of 283, 135 males, mean age2 = 22.73, SD = 2.20) that were repro- duced with permission from the previous study (Kim & Park, 2010) examining a Korean undergraduate sample to evaluate whether generational differences in risk are more salient to implicit measures of risk relative to explicit risk measures.
3.2.2 Measures
As noted above, the same scales (i.e., RG-SDS, RU-SDS, IRT-G, and IRT-U) were employed in Study 2 as used in Study 1, each measure presenting a reasonable level of reliability with Cronbach’s alpha coefficients over 0.6 (Loewenthal, 2004) ranging
2As expected, I found a significant age difference, indicating that the older adult participants were found to be significantly older than the corresponding younger adult participants, t(281) = 27.37, p = 10-82, d = 2.87. Throughout the current article, degree of freedom can vary owing to missing values on some measures including a
from .88 to .95 for both the older and younger age groups. Measures of risk choice and risk-relevant personality traits are explained in more detail below.
Choice Dilemma Questionnaire. The same variant of Kogan and Wallach’s (1964) CDQ as employed in the previous study (Kim & Park, 2010) was also used to evaluate age differences in risk-taking propensity under hypothetical uncertain situations. Participants were asked to provide the lowest probability of success they regarded as acceptable when advising the central person in each scenario on an alternative action. The choices were odds of 1, 3, 5, 7, and 9 chances of success out of 10 (scored as 1, 3, 5, 7, and 9, respectively), with one additional category (scored as 10) where participants could decline to recommend an alternative regardless of the probability of success. This scale presented a reliability of Cronbach’s α = .61 for the older age group and α = .60 for the younger age group. CDQ scores are calculated by summing the CDQ item scores. Lower CDQ scores indicate greater risk acceptance.
Barratt Impulsiveness Scale. Barratt’s (1994) Impulsiveness Scale is a 23-item questionnaire used to assess impulsivity as an indirect indicator of both impulse control – participants’ capacity for restraining their impulses – underlying psychosocial maturity and a variety of risk-taking behaviors (Stanford et al., 1996).
This scale showed a reliability of Cronbach’s alpha = .70 for the older age group and α = .73 for the younger age group. As with CDQ scores, BIS scores are also calculated by summing the BIS item scores. Higher scores on the BIS indicate a higher level of impulsiveness.
NEO FFI-Conscientiousness. I assessed conscientiousness as a proxy variable for psychosocial maturity using the 12-item conscientiousness scale from Costa and McCrea’s (1992) NEO Five Factor Inventory. Conscientiousness refers to a personality trait that is characterized by multiple facets such as self-discipline, dutifulness, industriousness, and impulse control, which were found to be theoretically associated with variables assumed to underlie psychosocial maturity – delay of gratification (Forstmeier, Drobetz, & Maercker, 2011), responsibility (Roberts & Bogg, 2004), emotional regulation (Larsen, 2000), and age (Jackson et al., 2009). This scale showed a reliability of Cronbach’s alpha = .87 for the older age group and α = .77 for the younger age group.
3.3 RESULTS
3.3.1 Construct divergence assessment
The correlations among all measures employed in this study are shown as a func- tion of the age group in Table 4.
Table 4. Correlations among all measures of demographics, personality, and risk across different age groups
Measure 1 2 3 4 5 6 7
Demographic
1. Age -
Psych. Maturity
2. BIS -.38*(-.29**) -
3. NEO FFI-C .44**(.20**) -.53**(-.61**) - Explicit risk
4. RG-SDS -.12(.06) -.18(-.12†) .26†(.08) -
5. RU-SDS .01(.03) .04(-.11) .24(.11) .36**(.39**) -
6. CDQ .09(.01) -.01(-.14*) -.07(.09) -.25*(.07) -.02(-.05) -
Implicit risk
7. IRT-G -.31*(-.03) .35*(.04) -.24(-.04) .26*(.18**) .07(.09) .02(-.004) -
8. IRT-U -.31*(-.09) .38*(-.04) -.48**(.07) .17(.25**) .23†(.30**) .06(-.05) .60***(.54***) Note. Numbers outside brackets refer to correlation coefficients in the older age group. BIS = Barratt Impulsivity Scale; NEO FFI-C = NEO Five Factor Inventory - Conscientiousness; RG-SDS = Risk Global Semantic Differential Scale; RU-SDS = Risk Unique Semantic Differential Scale; CDQ = Choice Dilemma Questionnaire; IRT-G = Implicit Risk Task – Global; IRT-U = Implicit Risk Task – Unique;
*** p < .001; ** p < .01; * p < .05; †p < .1
Older age group. To evaluate whether explicit and implicit measures of risk also diverge in a sample of Korean older adults as observed from younger adult sam- ples in different countries in Study 1, I performed bivariate correlations utilizing r to z conversion (Campbell & Fiske, 1959) and found method effects, rexp =. 36 for the RG-SDS – the RU-SDS and rimp = .60 for the IRT-G – the IRT-U, to be stronger than trait effects, rexp-imp = .25 (i.e., the average correlation between rRG-SDS – IRT-G and rRU-SDS –
IRT-U). Because the correlational data provide support for my expectation, two CFAs
controlling for gender were performed to directly test whether the data for the ex- plicit and implicit measures of risk were better suitable for a model with two factors (i.e., explicit and implicit risk; Figure 2, right panel) or one (risk; Figure 2, left pan- el).
Figure 2. Confirmatory factor analyses of risk construct divergence for the Korean older age group
Note. RG-SDS = Risk Global – Semantic Differential Scale; RU-SDS = Risk Unique – Semantic Differential Scale; IRT-G = Implicit Risk Task – Global; IRT-U = Implicit Risk Task – Unique.
The results of the CFAs exhibited that a model in which explicit and implicit risk measures converged on a single construct (i.e., one-factor model) had a noticeably poor fit, χ2(5, N = 62) = 10.46 (p = .06), CFI = .82, NFI = .77, TLI = .46, RMSEA = .11, p(close fit) = .13. Conversely, a model that defined explicit and implicit risk as sepa- rate, but correlated, factors (i.e., two-factor model) showed a close-to-fair fit, χ2(3, N
= 62) = 3.76 (p = .29), CFI = .98, NFI = .92, TLI = .87, RMSEA = .05, p(close fit) = .39. A chi-square test of significance for the difference between the two nested models revealed that the two-factor model had significantly better fit than the one-factor
model, χ2diff(2, N = 62) = 6.70 (p < .05). In the two-factor model, the estimated correla- tion between the two error terms of the explicit and implicit risk factors was .41 (p = .088). Overall, the divergence of risk construct represented by the explicit and im- plicit measures was detected in the Korean older age group.
Younger age group. As seen in the older age group, Table 4 exhibits the same pat- tern of construct divergence between explicit and implicit measures of risk was found in the younger age group, indicating stronger method effects, rexp =. 36 for the RG-SDS – the RU-SDS and rimp = .54 for the IRT-G – the IRT-U, than trait effects, rexp-
imp = .25 (i.e., the average correlation between rRG-SDS – IRT-G and rRU-SDS – IRT-U). For this reason, the same two CFAs as used in the older group were performed to directly test whether the data for the explicit and implicit measures of risk in the younger age group were also better suitable for a model with two factors (i.e., explicit and implicit risk; Figure 3, right panel) or one (risk; Figure 3, left panel).
Figure 3. Confirmatory factor analyses of risk construct divergence for the Korean younger age group
Note. RG-SDS = Risk Global – Semantic Differential Scale; RU-SDS = Risk Unique – Semantic Differential Scale; IRT-G = Implicit Risk Task – Global; IRT-U = Implicit Risk Task – Unique.
The results of the CFAs showed that a model in which explicit and implicit risk measures converged on a single construct (i.e., one-factor model) had a noticeably poor fit, χ2(5, N = 262) = 42.37 (p = 10-3), CFI = .78, NFI = .77, TLI = .35, RMSEA = .16, p(close fit) = 10-3. However, a model that defined explicit and implicit risk as sepa- rate, but correlated, factors (i.e., two-factor model) showed a close-to-fair fit, χ2(4, N
= 262) = 9.54 (p = .05), CFI = .97, NFI = .95, TLI = .88, RMSEA = .07, p(close fit) = .23. A chi-square test of significance for the difference between the two nested models
indicated that the two-factor model yielded a better fit to the data than did the one- factor model, χ2diff(1, N = 262) = 32.83 (p < .01). In the two-factor model, the estimat- ed correlation between the two error terms of the explicit and implicit risk factors was .43 (p = 10-3). Overall, the divergence of risk construct represented by the explic- it and implicit measures was also reliable in the Korean younger age group.
3.3.2 Age differences in explicit and implicit measures of risk
The means and standard deviations for the RG-SDS, RU-SDS, IRT-G and IRT-U between the older and younger age groups are shown in Table 5.
Table 5. Age differences in risk measures across different age groups
Measures Younger age group Older age group
Explicit Risk
RG-SDS 3.64 (1.31) 3.60 (1.38)
RU-SDS 3.13 (1.02) 3.19(.77)
Implicit Risk
IRT-G -.184 (.41) -.36(.42)
IRT-U -.30 (.39) -.42(.34)
Note. Standard deviations are denoted in parentheses below means.
Explicit risk. To examine whether age differences manifest themselves in explicit measures of risk, the older and younger age groups were compared by performing a univariate analysis of variance controlling for gender variable on the RG-SDS. The test exhibited no main effect for group, t(333) = 1.17, p = .24, d = .03, indicating that the older and younger adult participants were not found to differ in their explicit evaluations of risk. The absence of age differences was replicated when the same univariate analysis of covariance as noted above was performed using 1000 boot- strap samples on the RG-SDS, B = .23, p = .23, CI [-.15, .62]. In a similar vein, this pattern was reliably observed in the RU-SDS, t(338) = .12, p = .90, d = .05, indicating that the older and younger adult participants were not found to differ in their ex- plicit evaluations of self-relevant risk activities. Also, the absence of age differences was replicated when a total of 1000 bootstrap samples were used in the analysis, B = .02, p = .87, CI [-.23, .25].
Implicit risk. The same question as raised in the explicit measures of risk was tested performing a univariate analysis of variance controlling for gender variable on the IRT-G. The test yielded a significant main effect for group, t(334) = -3.44, p = 10-4, d = .44, suggesting that the older adult participants exhibited a stronger auto- matic association of risk with loss than did the younger adult participants. The sig-
nificant age differences were replicated when the same analysis as noted above was performed using 1000 bootstrap samples on the IRT-G, B = -.21, p = 10-3, CI [-.33, - .09]. Similarly, the IRT-U replicated the age differences as seen in the IRT-G when controlling for gender variable, t(327) = -3.00, p = 10-3, d = .34, suggesting that the older adult participants revealed a stronger automatic association of self-relevant risk with loss than did the younger adult participants. This pattern was reliably found when using 1000 bootstrap samples in the analysis, B = -.17, p = 10-3, CI [-.28, - .06].
3.3.3 Mediating role of psychosocial maturity on age – risk relationship Using structural equation modeling, further analyses were performed to evaluate whether age leads to a decrease in implicit risk through psychosocial maturity – in this case, impulsiveness and conscientiousness. Considering that the older and younger age groups are presumed to differ in psychosocial maturity and this theo- retical assumption was supported by the data3 in this study, the correlations among age, psychosocial maturity, and risk variables are explored by age groups in Table 4.
Older age group. Age was found to be significantly correlated with implicit measures of risk, r = -.31, p = .03 for the IRT-G and r = -.31, p = .03 for the IRT-U, as well as, psychosocial maturity variables, r = -.38, p = .01 for the BIS and r = .44, p = 10-3 for the NEO FFI-C. As is also seen from Table 4, both psychosocial maturity variables tended to be correlated with the two implicit measures of risk, r = .35, p
= .02 for the BIS – the IRT-G, r = .38, p = .01 for the BIS – the IRT-U, and r = -.48, p = 10-3 for the NEO FFI-C – the IRT-U, except for the NEO FFI-C – the IRT-G, r = -.24, p
= .12. Overall, these findings provided preliminary support for the expectation that the older adult participants’ implicit evaluations of risk decrease with age through psychological maturity.
Younger age group. As with the older age group, age was found to be significant- ly correlated with psychosocial maturity variables, r = -.29, p = 10-6 for the BIS and r
= .20, p = 10-3 for the NEO FFI-C, but not with implicit measures of risk, r = -.03, p
= .68 for the IRT-G and r = -.09, p = .16 for the IRT-U. Furthermore, both psychoso- cial maturity variables did not correlate significantly with implicit measures of risk, r = .04, p = .53 for the BIS – the IRT-G, r = -.04, p = .53 for the BIS – the IRT-U, r = -.04, p = .60 for the NEO FFI-C – the IRT-G, and r = .07, p = .32 for the NEO FFI-C – the IRT-U. Overall, these findings did not provide preliminary support for the expecta- tion that psychosocial maturity plays a mediating role of age – risk relation in the younger age group.
Impulsiveness as a mediator. Using the data for the older adult participants, I first estimated Model 1 that constrained the indirect path coefficients (i.e., age impul- siveness implicit risk) to be zero to look at whether the inverse age – risk relation
3 The older adult participants scored significantly lower than the younger adult participants on the BIS (impulsiveness), t(267) = 3.8, p = 10-5, d = .69, and scored significantly higher on the NEO FFI – C
that has been reported from previous studies using self-report risk measures is rep- licated in the implicit measures of risk (see Figure 4). As expected, the results showed a significant total effect of age on the latent implicit risk variable, β = -.42, p
= .01, and the fit indices of Model 1 exhibited a poor fit to the data (χ2 = 10.81, df = 3, CFI = .77, TLI = .24, NFI = .76, RMSEA = .21). Based on Model 1, Model 2 is devel- oped by releasing the indirect path coefficients (i.e., age impulsiveness implic- it risk) to test whether impulsiveness functions as a mediator of the relationship between age and implicit risk. The fit indices of Model 2 revealed a good overall fit to the data, χ2 = .057, df = 1, CFI = 1.00, TLI = 1.28, NFI = .99, RMSEA = .00, with a marginally significant direct effect of age on the latent implicit risk variable, β = -.29, p = .08. In addition, a chi-square test of significance for the difference between the two nested models revealed that the Model 2 had significantly better fit than the Model 1, χ2diff(2, N = 62) = 10.75 (p < .05). The findings suggest that impulsiveness partially mediated the relationship between age and implicit risk.
Figure 4. The partially mediated model among the older age group
Note. Numbers inside brackets refer to standardized coefficients in Model 1.
Numbers outside brackets refer to standardized coefficients in Model 2.
Conscientiousness as a mediator. The same analyses as used in impulsiveness were employed to examine whether conscientiousness acts as a mediator of the relation- ship between age and implicit risk in the older adult participants. The results showed a significant total effect of age on the latent implicit risk variable when the indirect path coefficients (i.e., age conscientiousness implicit risk) were con- strained to be zero (Model 1, see Figure 5), β = .42, p = .01. Also, the fit indices of Model 1 provided a poor fit to the data, χ2 = 16.90, df = 3, CFI = .66, TLI = -.15, NFI
= .66, RMSEA = .28. However, Model 2 that released the indirect path coefficients (i.e., age conscientiousness implicit risk) exhibited a non-significant direct effect of age on the latent implicit risk variable, β = .15, p = .30, with a good overall fit to the data, χ2 = 2.26, df = 2, CFI = .99, TLI = .97, NFI = .96, RMSEA = .05.
In addition, a chi-square test of significance for the difference between the two nested models revealed a better fit of Model 2 relative to Model 1, χ2diff(1, N = 62) =
14.64 (p < .05). The findings suggest that the relationship between age and implicit risk was fully mediated by conscientiousness in the older adult samples.
Figure 5. The fully mediated model for the older age group
Note. Numbers inside brackets refer to standardized coefficients in Model 1.
Numbers outside brackets refer to standardized coefficients in Model 2.
3.4 DISCUSSION
Consistent with the findings observed from the Korean younger adult sample in Study 1, I replicated construct divergence between explicit and implicit measures of risk in a sample of Korean older adults, suggesting the coexistence of explicit and implicit aspects of risk construct in a person’s mental representation. In addition, the implicit measures of risk appeared to perform better than the corresponding explicit risk measures in capturing age differences in risk, based on its relatively large effect size. These findings suggest that implicit evaluations of risk reflect accumulated socio-cultural experiences regarding risk-taking that can drive each cohort group to be homogenous in their attitudes toward risk (Rudolph & Zacher, 2015), thereby engendering marked age differences in risk that cannot be accounted for by explicit risk measures. Furthermore, a consideration of psychosocial maturity capacities (i.e., impulsiveness and conscientiousness) associated with risk-taking helps to explain why the marked age differences were seen only in the implicit measures of risk.
Another possible account for the observed age differences in this study was derived from a biochemical perspective. It relates to the hormonal changes during the aging process. More specifically, age is positively correlated with the levels of monoamine oxidase (MAO) that functions to degrade the three neurotransmitters – in this case, dopamine, norepinephrine, and serotonin. In fact, individuals who scored high on sensation seeking and impulsivity tend to have lower-than-average levels of MAO. As a person gets older, the increased levels of MAO are more likely to lower his or her willingness to take risk through degrading the three neurotransmitters that positively affect risk-taking behavior (Harlow & Brown, 1990). One final account from a behavioral perspective arises from the social
