Revealed Traits: A Novel Method for Estimating Cross-Cultural Similarities and Differences in Personality

Potential Benefits

Below, we elaborate upon potential benefits of revealed trait methodology for studying cross-cultural variation in personality.

Differences in Conscientiousness-Related Acts

We estimated cultural differences and similarities in personality via two methods. First, we replicate previously reported East–West differences in self-reported personality using more typical abstract trait ratings (Benet-Martínez & Karakitapoglu-Aygün, 2003; Gosling et al., 2003; McCrae et al., 1998; Schmitt et al., 2007). We then attempt to replicate these cultural differences using what we have referred to as revealed trait estimates. Because cultural variation in Conscientiousness has spurred much discussion (e.g., Heine et al., 2008; Oishi & Roth, 2009), we focused on action tendencies specifically relevant to Conscientiousness-related traits.

Method

Materials

Following the general framework of revealed preferences methodologies, we describe (a) the development of relevant stimuli (i.e., action scenarios), (b) the coding of the action scenarios on key features (action characterizations and expected effects), and (c) the collection of participants’ responses to the action scenarios and abstract trait items.

Development of action scenarios

To collect a broad range of stimuli, we asked participants at both Wake Forest University (WFU) and Singapore Management University (SMU; Samples 1 and 2) to describe situations in which they or someone else performed an action that exemplified a target personality trait. These actions were then adapted into two sets of action scenarios: actions specifically relevant to Conscientiousness and actions relevant to other traits from the HEXACO dimensions (Ashton & Lee, 2007). The Conscientiousness and “HEXAO” sets are used in Studies 1 and 2 of this article, respectively.

We briefly describe the development of these action scenarios below (see Wood, Tov, & Costello, 2015, for more details). Research assistants at both WFU and SMU adapted the free-response descriptions into action scenarios that could be rated more consistently across participants. An example of an organized action nominated by one participant was as follows: “My roommate is not very neat and places things everywhere around the room. If his things or trash are in my space [I] place them on his desk or clean up the mess upon sight.” This example was adapted into the following item:

You have a roommate that frequently leaves trash all over the room. Today you get home to find a mess he/she left. You could wait for him/her to clean up the trash, but he/she may not get around to it for a while. How likely would you be to clean up his/her mess yourself? (Scenario #144 in Supplemental Table S1)

The nominated actions were ultimately adapted to shorter items that had (a) less than 350 characters in length, (b) an explicit or strongly implied alternative action, (c) enough information to understand the implications of performing the action, and (d) retention of important aspects of the original scenario. Finally, actions were reviewed by research assistants from WFU and SMU to ensure their cultural appropriateness.

Study 1 focuses on actions that were originally nominated for three antonymous pairs of Conscientiousness-related traits: (a) dependable/reliable and undependable/unreliable, (b) organized/neat and disorganized/messy, and (c) careful/cautious and impulsive/spontaneous. This resulted in the 150 action scenarios. The full list of action scenarios used in Studies 1 and 2 is provided in Supplemental Table S1.

Action feature coding

The target action in each scenario was coded on two key features: action characterizations and expected effects.

Action characterizations

Samples 3, 4, 5, and 6 read the items generated above, in the general format of “You are in [situation X], You [perform action Y].” Samples 3 and 4 then indicated the extent to which the target action should be characterized along 10 dimensions selected to span traits central to the Big Five and HEXACO frameworks. Samples 5 and 6 rated these same 150 action scenarios along 13 additional dimensions central to the Big Five and HEXACO frameworks. For each dimension, participants rated whether the action was best described by one characteristic or its antonym (e.g., bold/assertive vs. submissive/unassertive) on a scale ranging from 1 = very [Characteristic A], to 4 = neither [Characteristic A] nor [Characteristic B], to 7 = very [Characteristic B]. Subsequently, 4 was subtracted from all scores resulting in a scale from −3 to +3, with 0 indicating that the target action was not characterized by either term.

The 150 action scenarios were randomly divided into three subsets of 50. Each participant only rated one subset to minimize fatigue. To ensure data quality, ratings were eliminated from subsequent analyses if they had corrected item-total correlations lower than .35, which indicated that participants were responding randomly. For Sample 3, this rule resulted in eliminating two participants for a total of 27, or nine for each subset. For Sample 4, this rule resulted in the elimination of three participants for a total of 34, or 12 for the first subset and 11 for the second and third. For Sample 5, this rule resulted in eliminating 16 participants for a final sample of 55 WFU participants for the 13 additional ratings: 16 for the first subset, 15 for the second subset, and 24 for the final subset. For Sample 6, one participant was eliminated from the first subset for a final sample of 35 SMU participants for the 13 additional ratings: 12 for the first subset, 12 from the second subset, and 11 from the third subset.

Reliability coefficients were obtained by calculating Cronbach’s alpha on a restructured dataset treating individual raters as items, and actions as the unit of analysis; when data are structured this way, alpha is equivalent to the Intraclass Correlation Coefficient (ICC) from a two-way random-effects model (MacLennan, 1993). These alpha values can be regarded as providing the correlation that would be expected if these averages were correlated with a new group of raters of equal size. Action characterizations were highly reliable, with average reliability coefficients ranging from a high of .95 for careful to a low of .52 for creative, M(α) = .83, and so we took the average rating within each sample. Supplemental Table S2 contains the reliability coefficients for the characterization ratings and the average reliability for each characterization.

Supplemental Table S3 shows how characterizations made by the Singaporean and U.S. samples correlated with each other. For Study 1, the raw correlations were quite high, ranging from a low of .60 for creative to a high of .95 for dependable, M(r) = 0.84. Supplemental Table S3 also shows the estimated correlations adjusted for unreliability by dividing by the square root of the reliabilities reported above (e.g., J. Cohen, Cohen, West, & Aiken, 2003). These adjusted correlations indicate the expected correlations between rater groups if both groups had used a very large (conceptually infinite) number of raters. The cross-sample estimates of the correlations adjusted for unreliability were all very close to 1.0 (rs from 0.81 to 1.26); M(r) = 1.01. These high adjusted correlations suggest that there are at best relatively small differences in how actions were characterized along the 23 dimensions examined across the two samples. In other words, the rank order of the actions described as revealing different traits (e.g., assertive vs. unassertive) was extremely similar across the American and Singaporean samples, indicating that the traits examined did not appear to manifest themselves in different ways across the scenarios assessed in the two cultures. Therefore, we took the average of each sample’s mean characterization ratings, weighting the mean from each sample equally despite their different sample size. Supplemental Table S4 shows the revealed trait analyses using characterizations from each sample separately.

Expected effects

Eleven research assistants read the action scenario items and coded them along 21 expected effect dimensions (Wood, Tov, & Costello, 2015). Specifically, raters read the scenario and action descriptions and the prompt “How much would doing this (vs. the alternative) alter the potential/possibility of the following outcomes?” using a 5-point scale ranging from 1 = “Greatly Increase” to 3 = “No expected change” to 5 = “Greatly Decrease.” We subtracted 3 from all scores resulting in a scale from −2 to +2, and were then reversed such that higher scores were associated with greater expected effects on the indicated dimension.

Self-report measures

Samples 7 and 8 completed two self-report measures enabling us to compare cross-cultural variation in personality traits as assessed by traditional versus revealed trait methods.

Abstract personality ratings

To assess differences in self-rated personality trait terms, Samples 7 and 8 completed the Inventory of Individual Differences in the Lexicon (IIDL; Wood, Nye, & Saucier, 2010), an inventory designed to measure a broad range of traits regularly reflected by terms in the English lexicon. Participants rated the extent to which 61 synonymous adjective pairs (e.g., “dependable, reliable”) accurately describe them from 1 (very uncharacteristic) to 5 (very characteristic), with a midpoint of 3 (neither characteristic nor uncharacteristic). Two additional adjective pairs were included: “hard-working, productive” and “cheap, stingy” (see Table 2 for the full list of adjective pairs).

OPEN IN VIEWER

Table 2.

Average Endorsement of Abstract Self-Ratings for Each Sample.

No.	IIDL item	Study 1 (Consc. set)			Study 2 (HEXAO set)
		WFU M	SMU M	d	WFU M	SMU M	d
Conscientiousness
1	Practical, sensible	4.19	3.92	0.38	4.10	3.74	0.38
2	Competent, capable	4.29	3.68	0.99	4.26	3.60	0.80
3	Dependable, reliable	4.43	3.90	0.75	4.31	3.96	0.47
4	Disorganized, messy	2.59	3.15	–0.48	2.57	2.68	−0.09
5	Undependable, unreliable	1.69	1.95	−0.29	1.60	1.74	−0.18
6	Hardworking, productive	4.19	3.36	0.96	4.17	3.64	0.58
Extraversion
7	Sociable, outgoing	3.88	3.59	0.30	3.83	3.30	0.46
8	Excited, enthusiastic	4.01	3.64	0.47	3.88	3.70	0.18
9	Assertive, bold	3.46	2.85	0.61	3.34	2.94	0.35
10	Happy, joyful	4.12	3.77	0.49	4.07	3.83	0.32
11	Funny, amusing	3.98	3.69	0.35	3.99	3.62	0.42
12	Affectionate, loving	4.17	3.91	0.34	4.33	3.85	0.53
13	Loud, noisy	2.84	2.92	−0.07	2.55	2.47	0.07
14	Brave, adventurous	3.72	3.16	0.60	3.65	3.23	0.42
15	Bashful, shy	2.73	3.04	−0.26	2.68	3.47	–0.72
16	Confident, self-assured	3.76	3.31	0.47	3.75	3.23	0.50
Emotional stability
17	Stable, well adjusted	3.97	3.54	0.61	3.80	3.70	0.12
18	Relaxed, calm	3.73	3.45	0.34	3.77	3.51	0.26
19	Positive, optimistic	3.86	3.65	0.24	3.97	3.81	0.19
20	Afraid, scared	2.28	2.75	–0.49	2.27	2.77	–0.42
21	Tense, anxious	2.98	3.11	−0.12	2.64	2.79	−0.12
22	Sad, unhappy	2.13	2.40	−0.30	1.83	2.47	–0.66
23	Temperamental, touchy	2.62	2.92	−0.30	2.25	2.87	–0.51
24	Crabby, grouchy	2.28	2.25	0.03	2.03	2.32	−0.28
25	Lonely, lonesome	2.41	2.82	–0.41	2.22	2.94	–0.65
Agreeableness
26	Kind hearted, caring	4.33	3.92	0.64	4.31	4.00	0.35
27	Giving, generous	3.95	3.77	0.26	4.14	3.72	0.51
28	Pleasant, agreeable	4.07	3.95	0.21	4.20	4.00	0.25
29	Polite, courteous	4.45	4.08	0.56	4.35	3.96	0.49
30	Thankful, grateful	4.33	3.98	0.53	4.25	4.06	0.21
31	Truthful, honest	4.33	3.71	0.83	4.18	4.02	0.23
32	Selfish, self-centered	2.44	2.40	0.05	1.93	2.32	−0.37
33	Unfriendly, cold	1.70	2.13	–0.48	1.49	2.11	–0.54
34	Inconsiderate, rude	1.74	1.89	−0.17	1.48	1.89	–0.48
35	Egotistical, conceited	2.15	2.32	−0.17	2.08	2.28	−0.17
36	Cruel, abusive	1.32	1.43	−0.19	1.27	1.43	−0.21
37	Dominant, controlling	3.02	2.87	0.14	2.61	2.89	−0.23
38	Angry, hostile	1.85	2.19	−0.35	1.79	2.02	−0.24
Intellect/openness
39	Creative, imaginative	3.75	3.49	0.27	3.54	3.45	0.08
40	Smart, intelligent	4.27	3.37	1.26	4.21	3.38	0.93
41	Radical, rebellious	2.47	2.59	−0.12	2.32	2.34	−0.02
42	Skilled, skillful	3.91	3.44	0.66	3.65	3.17	0.46
43	Conservative, traditional	2.97	2.91	0.04	2.95	3.11	−0.15
44	Close minded, narrow minded	2.04	2.25	−0.22	1.94	2.11	−0.16
Physical/evaluative
45	Awkward, clumsy	2.94	3.01	−0.06	2.84	3.13	−0.24
46	Exciting, fascinating	3.58	3.44	0.17	3.65	3.26	0.39
47	Influential, prominent	3.42	2.99	0.48	3.45	2.74	0.71
48	Likable, well liked	4.14	3.81	0.50	4.01	3.83	0.23
49	Admirable, impressive	3.67	3.15	0.69	3.89	3.13	0.92
50	Great, terrific	3.62	3.16	0.63	3.73	3.49	0.26
51	Lucky, fortunate	3.81	3.76	0.06	3.85	3.72	0.14
52	Wealthy, well to do	3.15	2.67	0.48	3.25	2.45	0.73
53	Weird, strange	3.29	2.85	0.40	2.97	2.96	0.01
54	Ordinary, average	2.88	3.47	–0.62	2.85	3.64	–0.71
55	Good looking, attractive	3.67	3.16	0.66	3.73	3.23	0.59
56	Beautiful, pretty	3.48	2.78	0.73	3.41	2.89	0.63
57	Youthful, young	4.23	3.80	0.61	4.14	3.89	0.28
58	Well, healthy	4.14	3.71	0.53	4.09	3.87	0.24
59	Slim, slender	3.16	2.98	0.16	3.31	2.87	0.35
60	Tired, exhausted	3.34	3.19	0.14	3.28	3.21	0.06
61	Short, little	2.34	2.55	−0.16	2.30	2.32	−0.02
62	Cheap, stingy	2.28	2.10	0.19	2.19	2.34	−0.13
63	Feminine, unmasculine	3.09	2.99	0.09	3.18	2.83	0.28
n for each sample		86	90		217	47

Note. The above table depicts the average self-rating of 63 trait adjectives for each sample and the effect size of the difference between our samples (Cohen’s d). Cohen’s d values which are significant at p < .05 are underlined. Cohen’s d values which are significant at p < .01 are additionally shown in bold. WFU = Wake Forest University; SMU = Singapore Management University; HEXAO = Honesty-Humility, Emotionality, eXtraversion, Agreeableness, Openness.

Action-likelihood ratings

Samples 7 and 8 read the 150 action scenarios, and rated how likely they were to perform each action. For instance, in the messy roommate example, participants were asked, “How likely would you be to clean up his or her mess yourself?” with likelihood ratings of 1 = “Less than 10% chance,” 2 = “About 25% chance,” 3 = “50% chance,” 4 = “About 75% chance,” and 5 = “More than a 90% chance.” We calculated reliability for these ratings by treating individuals as items and actions as the unit of analysis. Action-likelihood ratings were highly reliable (both αs = .97), so we created group means by taking the sample average for each action. Group means were transformed to a 0 to 100 “percentage of maximum possible” (POMP) metric (P. Cohen, Cohen, Aiken, & West, 1999).

Prior to conducting substantive analyses, several participants were removed based on indications that they completed some or all of the survey randomly, which reduces statistical power (Maniaci & Rogge, 2014). Given evidence that response time serves as a strong indicator of careless responding (DeSimone, Harms, & DeSimone, 2015), participants were removed if they completed the survey in less than 30 min, and additionally showed low agreement with the normative profile of responses to either the action scenarios or IIDL (rs < .30; for more details, see Wood, Tov, & Costello, 2015). This process resulted in a final sample of 176 participants (WFU n = 86; SMU n = 90; see Samples 7 and 8 in Table 1).

Results and Discussion

Cultural similarities and differences in revealed traits and revealed preferences

We first discuss the overall similarity in action-likelihood profiles across the Singaporean and U.S. samples. We then examine how action likelihoods relate to action characterizations (revealed traits) and expected effects (revealed preferences) in both cultures.

Similarity in overall response profiles

As seen in Figure 1, there was an extremely high degree of similarity in the actions that the U.S. and Singaporean participants reported they were likely to perform (r = .90). Thus, this broad analysis suggests that our samples are much more similar than different with regard to how they report they will act across a wide array of hypothetical situations. For instance, in both samples, people reported being highly unlikely to throw a ping pong paddle at the wall when losing a game of ping pong (Action 129 in Supplemental Table S1, M_WFU = 19.5, M_SMU = 18.9), and highly likely to double check their class schedule before the first day of classes (Action 68, M_WFU = 93.3, M_SMU = 88.6).

OPEN IN VIEWER

Figure 1.

Overall similarity in endorsement of actions in Studies 1 and 2.

There were nonetheless actions with moderate discrepancies in endorsement across cultures. For instance, the Singaporean sample reported that they were considerably more likely to bring only one notebook to school to use for multiple classes (Action 121, M_WFU = 31.4, M_SMU = 59.7), whereas the U.S. sample reported that they were considerably more likely to clean their messy room before going on a trip (Action 97, M_WFU = 81.0, M_SMU = 58.1).

Interestingly, in contrast to abstract trait ratings, there was little evidence of differences in scale use for the action-likelihood ratings. Whereas the U.S. sample showed significantly more variability than the Singaporean sample across the mean ratings of the 63 IIDL items as noted above (SD_WFU = 0.84, SD_SMU = 0.62), no difference was observed in the variability across the mean ratings of the 150 actions (SD_WFU = 0.76, SD_SMU = 0.71 in original scale units ² ), Levene’s test F(1, 298) = 0.79, p = .375. This suggests that action-likelihood ratings may be less susceptible to scale use extremity differences than abstract trait ratings. ³

Revealed trait analyses

Individuals in both cultures were fairly similar in how they report they are likely to act across a diverse set of situations. We can use the revealed trait analyses to investigate what sorts of actions both groups indicate they are likely to perform.

Cross-cultural similarities

Table 3 shows the correlation between the action characterization ratings and mean action-likelihood ratings for each sample. We discuss correlations that are particularly large in magnitude (|rs| ≥ .40). This effect size threshold was chosen because it is between the bounds of what is traditionally considered a medium and large effect size (J. Cohen, 1992). The following can thus be interpreted as some of the most commonly expressed traits in both cultures. In descending magnitude, participants in both samples report being more likely to perform actions that are characterized as normal (row #15 in Table 3), dependable (#4), intelligent (#5), courteous (#11), traditional (#19), modest (#12), likable (#13), competent (#16), careful (#3), kind hearted (#6), giving (#14), trusting (#17), industrious (#1), and truthful (#7). In Big Five or HEXACO terms, participants from both cultures report being more likely to perform actions that reflect Conscientiousness, Agreeableness, and Honesty/Humility-related characteristics.

OPEN IN VIEWER

Table 3.

Correlation Between Action Characterization and Mean Endorsement of Performing Action.

No.	Action characterization	Conscientiousness set (Study 1)			HEXAO set (Study 2)
		WFU	SMU	Diff.	WFU	SMU	Diff.
1	Industrious, hardworking	.47	.40	.07	.48	.37	.09
2	Organized, neat	.33	.30	.03	.60	.51	.09
3	Careful, cautious	.57	.63	−.06	.46	.55	–.09
4	Dependable, reliable	.68	.64	.04	.64	.57	.07
5	Intelligent, smart	.64	.66	−.02	.72	.64	.08
6	Kind hearted, caring	.53	.56	−.03	.60	.51	.09
7	Truthful, honest	.47	.50	−.03	.30	.31	−.01
8	Confident, self-assured	–.05	–.14	.09	.17	.00	.17
9	Bold, assertive	−.12	−.15	.03	–.01	–.12	.11
10	Outgoing, sociable	.15	.07	.08	.29	.08	.21
11	Courteous, polite	.67	.69	.02	.67	.56	.11
12	Modest, humble	.66	.66	.00	.60	.55	.05
13	Likable, pleasant	.65	.64	.02	.66	.53	.13
14	Giving, generous	.56	.56	.00	.58	.48	.10
15	Normal, usual	.79	.82	−.03	.82	.75	.07
16	Competent, capable	.62	.60	.02	.62	.47	.15
17	Trusting, unsuspicious	.55	.52	.03	.41	.30	.11
18	Calm, relaxed	.35	.41	−.06	.43	.31	.12
19	Traditional, conventional	.66	.70	−.04	.34	.33	.01
20	Exciting, fascinating	–.10	–.21	.11	.09	–.07	.16
21	Narrow minded, close minded	–.05	.04	–.09	–.27	–.10	–.17
22	Creative, imaginative	.06	.01	.05	.23	.08	.15
23	Happy, joyful	.23	.18	.05	.40	.26	.14

Note. Action characterization values used in the analyses represented in this table combined ratings performed by SMU and WFU participants. The Diff. column for each Study (columns 5 and 8) contains the differences in the correlation between mean endorsement of each sample and action characterization ratings. Underlined values indicate correlations which are significantly different between our samples using Steiger’s (1980) test for dependent correlations (p < .05). WFU = Wake Forest University; SMU = Singapore Management University; Diff. = Difference; HEXAO = Honesty-Humility, Emotionality, eXtraversion, Agreeableness, Openness.

Cross-cultural differences

To test for differences in revealed traits across the two samples, we applied Steiger’s (1980) test of dependent correlations. As seen in Table 3, analyses revealed a stronger tendency among U.S. participants to endorse actions as a function of how much they were characterized as outgoing (row #10 in Table 3) and confident (#8). Singaporean participants showed a stronger tendency to endorse actions as a function of how much they were characterized as narrow minded (#21), and a stronger tendency to endorse actions less as a function of how much they were characterized as exciting (#20).

Revealed preferences

We can use revealed preferences analyses to indirectly estimate the weight of specific expected effects on reported action likelihoods across cultures. The revealed preferences can help explain why we observe similarities and differences in revealed traits, as they have been previously demonstrated to be important in characterizing actions with particular trait concepts and are closer to features central to processing accounts of personality variation (Wood, Tov, & Costello, 2015). As with the revealed trait analyses, we first discuss those expected effects which had a large impact on action likelihoods in each of the two samples (|r| ≥ .40), and then discuss expected effects which had relationships with action likelihoods that differed significantly across the two samples.

Cross-cultural similarities

Table 4 shows the results from the revealed preferences analyses. Participants in both samples reported being more likely to perform actions that were expected to fulfill commitments (row #2), match others’ preferences (#1), and gain social status (#11). Both samples reported being less likely to perform actions that were expected to result in social rejection for the self (#3) or others (#4). The shared preference for the three former expected effects and aversion to the latter two may underlie the revealed similarity in kindness and dependability reported above as these five expected effects have previously been found to be related to characterizing actions as kind and dependable (Wood, Tov, & Costello, 2015). There was also a common preference for actions expected to enhance predictability (#15) and a common aversion to actions expected to surprise others (#5). Both of these expected effect dimensions were found by Wood and colleagues to be important in characterizing actions as careful, which may underlie the similarity in revealed carefulness reported above. Finally, both samples showed a preference for actions expected to meet achievement goals (#18), which was previously found to be central to characterizing actions as industrious; the similarity in revealed industriousness may thus reflect a common desire to meet achievement goals. Taken together, people in both cultures seem to preferentially perform actions expected to have prosocial effects (e.g., fulfill commitments), or that further their own more agentic concerns (e.g., gain social status), and these preferences may manifest themselves in the high degree of similarity in revealed traits.

OPEN IN VIEWER

Table 4.

Correlation Between Expected Effects Dimensions and Mean Endorsement of Performing Actions.

No.	Expected effects	Conscientiousness set (Study 1)			HEXAO set (Study 2)
		WFU	SMU	Diff.	WFU	SMU	Diff.
1	Doing what other(s) would prefer	.63	.60	.03	.45	.41	.04
2	Fulfilling your commitments	.64	.61	.03	.51	.47	.04
3	Experiencing rejection	−.60	−.59	−.01	−.42	−.47	.05
4	Someone else experiencing rejection	−.45	−.47	.02	−.43	−.39	−.04
5	Acting in way surprising to others	−.61	−.66	.05	−.38	−.38	.00
6	Drawing attention to self	−.35	−.41	.06	−.21	−.25	.04
7	Expressing own wants/values	–.34	–.26	–.08	−.21	−.19	−.02
8	Effort expended	.42	.33	.09	.14	.05	.09
9	Someone having work in the future	−.21	−.18	−.03	−.18	−.21	−.03
10	Learn new things	.29	.25	.04	.09	.05	.04
11	Gaining social status	.48	.45	.03	.21	.17	.04
12	Be more connected to other(s)	.38	.33	.05	.33	.27	.06
13	Being exposed to social situations	.05	−.01	.06	.07	–.04	.11
14	Having work to do in future	−.04	.01	−.05	.03	.02	.01
15	Being in a predictable situation	.46	.52	−.06	.27	.34	−.07
16	Having order/structure in environment	.22	.23	−.01	.35	.38	−.03
17	Experiencing sensory stimulation/excitement	–.28	–.41	.13	–.13	–.22	.09
18	Meeting achievement goals	.44	.41	.03	.10	.12	.02
19	Give accurate info of what know/feel	.16	.20	−.04	−.07	−.05	−.02
20	Experiencing physical harm	–.05	–.20	.15	.04	−.01	.05
21	Someone experiencing physical harm	–.21	–.35	.14	−.24	−.22	.02

Note. Underlined values indicate correlations which are significantly different between our samples using Steiger’s (1980) test for dependent correlations (p < .05). WFU = Wake Forest University; SMU = Singapore Management University; Diff. = Difference; HEXAO = Honesty-Humility, Emotionality, eXtraversion, Agreeableness, Openness.

Cross-cultural differences

Even though the associations between expected effects and action likelihoods were highly similar across our samples, there were still some notable differences. As seen in Table 4, the Singaporean participants reported being less likely to perform actions that were expected to result in physical stimulation or excitement (#17), and expending effort (#8) than U.S. participants; these action effects were previously found by Wood and colleagues to be highly related to outgoing and confident actions. Thus, revealed differences in outgoingness and confidence may be driven by differences in preferences for experiencing excitement and conserving effort. Singaporean participants also showed a larger avoidance of actions expected to increase risk of physical harm or pain to the self (row #20) or others (row #21) than U.S. participants. Somewhat surprisingly, U.S. participants reported being less likely to perform actions that were expected to express one’s own wants or values (#7), which was previously found by Wood and colleagues to be principally related to boldness. The difference in preferences for expressing wants and values does not appear to be manifested in revealed trait differences, as none of the traits that showed revealed differences were found by Wood and colleagues to relate strongly to this expected effect.

Revealed Traits With a More Diverse Set of Actions

Study 1 demonstrated a great deal of similarity in how U.S. and Singaporean participants report being likely to act in specific situations. This similarity extends to the traits those actions express as well as the expected effects of those actions. However, we also observed subtle differences such as lower revealed trait estimates of outgoing and confident and related expected effects among Singaporean participants, relative to U.S. participants.

One notable limitation of Study 1 is its emphasis on Conscientiousness-related action scenarios. It is unclear whether the similarities and differences identified are robust across a more diverse set of actions. Study 2 attempts to replicate Study 1 using a broader set of action scenarios targeting the other five factors of the HEXACO structure (i.e., Honesty–Humility, Emotionality, extraversion, Agreeableness, and Openness; “HEXAO”). Outside of this difference in stimuli, the second study was a direct replication of Study 1.

Method

Results and Discussion

Much More Similar Than Different

Although we did find some cross-cultural differences in action likelihoods across samples from the two populations, these differences were qualified by very strong similarities. The magnitude of similarity (both rs = .89), as shown clearly in Figure 1, was much closer to being indistinguishable (r = 1) than to being independent (r = 0), and far surpass traditional benchmarks for strong correlations in the social sciences (J. Cohen, 1992). These results suggest that cross-cultural similarity is high not only at the broader level of values (Schwartz & Bardi, 2001) but also at the level of action tendencies.

The revealed trait analyses shed further light on the nature of these similarities. In both studies, 10 traits (i.e., action characterizations) were highly related (|rs| ≥ .40) to action likelihoods for participants in Singapore and the United States. Both groups were considerably more likely to perform actions characterized as normal, intelligent, courteous, likable, dependable, competent, kind-hearted, modest, giving, and careful. In terms of the Big Five or HEXACO trait structure, this suggests that people in both cultures reported being more likely to perform actions that express Conscientiousness, Honesty–Humility, and Agreeableness.

Turning to the revealed preferences analyses, three expected effects were highly related (|rs| ≥ .40) to action likelihoods for Singaporean and U.S. participants across both studies. Participants from both groups reported being more likely to act in ways that increase the likelihood of matching others’ preferences and fulfilling commitments; and decrease the likelihood of experiencing rejection. People from both groups also reported being more likely to act in ways that decrease the likelihood of someone else being rejected and surprising others, though each of these fell just short of the effect size threshold in Study 2.

The observed similarities in revealed preferences are consistent with theories that propose certain characteristics to be valued across cultures (Baumeister & Leary, 1995; Ryan & Deci, 2000; Schwartz & Bardi, 2001). The need to belong has been postulated as a universal need; and the strong, negative correlation between the possibility of eliciting rejection and action likelihood across both groups in this study further supports that view. In addition, the observed role of matching others’ preferences, limiting others’ experience of interpersonal rejection, and fulfilling commitments in shaping action likelihoods is consistent with the high ranking accorded to benevolence values in cross-national studies of self-reported values (Schwartz & Bardi, 2001).

By identifying commonalities in how members of different groups calibrate their actions on the basis of expected effects, revealed preference analyses may suggest possible links between universal needs and values, and similarities in personality trait levels across cultures. Likewise, given that expected effects are likely closer to the ends that people pursue with their actions and are central to characterizing actions by a particular trait concept (Wiggins, 1997; Wood, Tov, & Costello, 2015), they provide a connection between structural and processing accounts of personality (McCabe & Fleeson, 2012, 2016; Wood, Gardner, & Harms, 2015). Indeed, the three expected effects dimensions that showed strong preferences across cultures in both studies have been found to correlate strongly with characterizing actions as kind and dependable (Wood, Tov, & Costello, 2015)—characteristics within the domains of Conscientiousness and Agreeableness. Our results suggest that the theorized universal values of belongingness and benevolence may manifest themselves in the cross-cultural tendency to behave in ways that are Conscientious and Agreeable.

Revealing Cultural Differences

A major purpose of this project was to test whether or not East–West differences in Conscientiousness-related traits found in previous research (Benet-Martínez & Karakitapoglu-Aygün, 2003; Gosling et al., 2003; McCrae et al., 1998; Schmitt et al., 2007) replicate using a novel method. That is, do we see multimethod convergence for this finding? The answer to this question differs across our two studies. There were no significant differences between the Singaporean and U.S. samples with regard to revealed Conscientiousness-related trait levels in Study 1. However, differences were observed in Study 2, which estimated revealed traits using a broader set of actions. Three of the four Conscientiousness-related revealed traits were higher for U.S. participants (industrious, organized, and dependable) and one was higher for Singaporeans (careful). It is worth noting that our self-report measures did find the typical East–West differences in Conscientiousness in both studies, and so the lack of consistent Conscientiousness differences using revealed trait methodology does not appear to be due to an idiosyncratic lack of Conscientiousness differences within our samples.

On the contrary, cultural differences in revealed Extraversion were more consistent with past findings. U.S. participants were more likely than Singaporean participants in both studies to perform actions characterized as Extraverted (e.g., confident, outgoing, exciting). This replicates past cultural differences using standard self-report personality scales (e.g., Benet-Martínez & Karakitapoglu-Aygün, 2003; Schmitt et al., 2007). Moreover, unlike Conscientiousness, nation-level Extraversion has demonstrated straightforward criterion validity (negative correlation with suicide rate; Oishi & Roth, 2009), suggesting that differences in Extraversion are generally less suspect than Conscientiousness.

As demonstrated, one strength of the revealed trait method is that it provides an opportunity to assess multimethod convergence. How can we make sense of divergences between revealed and direct estimates of personality, such as those observed in these studies? Divergences between these methods may point to novel, more nuanced interpretations of cross-cultural differences. For example, perhaps U.S. individuals encounter more situations in which they can demonstrate their dependability than Singaporean individuals, even if the latter would choose the dependable acts as often if they were in the same situations. Importantly, the revealed trait methods used a standard set of situation–action pairs, which should remove differences driven by rates of encountering trait-relevant situations (see Lievens, 2017, for a similar argument for the related SJT method). This is in contrast to typical act frequency approaches, which confound action enactment with the level of exposure to situations (e.g., reported frequencies of J. J. Jackson et al.’s, 2010, item “Miss a meeting” will be higher on average for people who have relatively more meetings independent of relative levels of Conscientiousness; Buss & Craik, 1983, take a similar approach to J. J. Jackson and colleagues). More generally, rates of encountering trait-relevant situations should be unlikely to affect revealed traits differences, whereas they are likely to affect both traditional act frequency approaches and self-reported personality using abstract trait items or short sentences. Future research could tease apart these possibilities by asking participants how often they encounter the situations in the action scenarios.

The results also illustrate how the nature of revealed trait differences can be further clarified by inspecting revealed preferences for expected effects. As mentioned previously, expected effects can help bridge the gap between values, structural models of personality, and processing accounts of personality. In both studies, U.S. participants reported being more likely to perform actions that required effort and that were stimulating. These differences in expected effects may explain revealed differences in Extraversion-related characteristics such as outgoing, confident, and exciting: Expending effort and experiencing stimulation are important to characterizing actions with such terms (Wood, Tov, & Costello, 2015), and may reflect key processing features of Extraversion (i.e., reward sensitivity; Denissen & Penke, 2008). The same expected effects may explain the revealed trait differences in Conscientiousness: Expending effort is central to trait tendencies that were stronger for U.S. participants (i.e., industrious and organized), and seeking stimulation correlates negatively with the trait tendency that was stronger for Singaporean participants (i.e., careful; Wood, Tov, & Costello, 2015). Thus, it may be that U.S. participants choose less careful actions due to a greater preference for stimulation, and Singaporeans choose less industrious and organized actions to conserve effort. These analyses provide evidence that East–West differences in Extraversion, and Conscientiousness to a lesser extent, may be due to differential preference for expending effort and experiencing excitement. These differences are also consistent with research showing a greater preference among European Americans for experiencing high-arousal positive affect and influencing the social environment relative to people of Chinese descent (Tsai, Knutson, & Fung, 2006).

Limitations and Future Directions

The clearest limitation of the revealed trait methodology presented here is that it does not index how people actually behave in these situations, but how they report they will behave. These indexes will certainly differ to some extent, and socially desirable responding can bias self-reports of hypothetical actions in much the same manner as abstract trait ratings (Fleeson, 2009; Peeters & Lievens, 2005). Socially desirable responding is a distinct possibility in our data, given that the cross-cultural similarities were in large desirable characteristics (kindness, dependability, etc.). To the extent that this is true, it is possible that the data do not suggest that people actually behave similarly across cultures, but instead that behaviors are similarly desirable across cultures. While this would go against the interpretations presented here, this would in itself be an interesting finding. It would further support the universality of values that other researchers have found or theorized (Baumeister & Leary, 1995; Ryan & Deci, 2000; Schwartz & Bardi, 2001), if not the actualization of these values in behavior. Notwithstanding these issues, the cross-cultural differences in revealed trait estimates of Extraversion-related characteristics are unlikely to be due to differences in socially desirable responding, as characteristics within the domain of Extraversion tend to be less evaluative or desirable (indeed, the lowest of the Big Five; John & Robins, 1993).

Outside of socially desirable responding, there are other reasons to question whether the results presented here reflect real differences in behavior. Indeed, it may be that action-likelihood responses reflect what people want or intend to do, rather than what they would actually do. Even so, intentions to act a certain way do predict how people actually act, though this prediction is not perfect (Ajzen, 1991). Moreover, methodologies similar to revealed traits, such as the SJTs, appear to predict real behaviors over and above typical self-reported personality inventories (Lievens, 2017; Motowidlo et al., 2006). Thus, responses to hypothetical scenarios may be valid for the prediction of behavior in similar, real-life situations. A clear avenue for future research is to explore how estimates of revealed traits predict actions taken in real situations. A key moderator may be the resemblance of the scenarios to everyday situations. According to Robinson and Clore (2002), people typically respond to hypothetical scenarios by considering how they have acted in similar situations. When scenarios are completely unfamiliar, however, responses are more likely to be based on general beliefs about oneself or possibly interpersonal comparisons.

Drawing on past theoretical approaches (Buss & Craik, 1983), the revealed traits approach assumes that personality traits are “revealed” by enacting trait-related behaviors in relevant situations. However, traits can be revealed additionally through cognitive and affective reactions to particular situations (e.g., Funder, 2013; Wilt & Revelle, 2015). We did include actions nominated for (and characterized by) trait terms which are often considered less behavioral (e.g., calm/relaxed). Nevertheless, it is possible that this approach is less suited to assessing traits that largely reflect cognitive or emotional tendencies. The generalizability of this approach for studying a range of traits deserves further attention.

Another concern is whether the scenarios themselves are representative of the cultures tested and understood in the same manner. The clearest example of an action that was not equally applicable across cultures in the present study was the aforementioned scenario involving “mooning” a friend in class, which was endorsed much more highly by Singaporean students than by American students, almost certainly due to Singaporean students not understanding the meaning of this slang term. It is likely that there were more subtle differences in the understood meaning of other scenarios presented to participants. Nonetheless, most scenarios appeared to be familiar or relatable to students in both cultures. First, the cross-cultural differences observed here were generally replicated regardless of whether actions were characterized by American or Singaporean students (see Supplemental Table S4). Second, the latent correlation between action characterizations made by the two cultures was near unity (i.e., correlations averaged approximately 1.0 after adjusting for rater unreliability), implying that the trait-relevant implications of each action understood similarly across cultures (e.g., the actions seen as indicating assertiveness or impulsivity by students in an American university were understood in largely the same manner by students in a Singaporean university, and vice versa).

We also took measures to address representativeness in the design phase of the study. Participants from both groups nominated the actions that ultimately became the action scenarios used in this study. In addition, research assistants from both the United States (the WFU research team) and Singapore (the SMU research team) reviewed the stimuli. These procedures ensured that the bulk of scenarios were generally familiar to participants from both cultures. Of course, the nature of different cultural groups is that they are often exposed to different social contexts, some of which are thought to be responsible for cultural or geographical variation in personality and behavior (e.g., Rentfrow, Gosling, & Potter, 2008). Thus, future development of revealed traits stimuli could include additional ratings by participants indicating how often they have encountered each situation described in the action scenarios to more formally assess the familiarity, meaningfulness, and cultural appropriateness of each item.

Although we have taken steps to maximize and ensure cross-cultural comparability of the scenarios, other methods could have been employed to evaluate the measurement invariance of the items. Current approaches to invariance testing (e.g., multigroup confirmatory factor analysis [CFA]) assume that several items are indicators of a target latent construct (e.g., the Extraversion subscale of the Revised NEO-Personality Inventory; Costa & McCrae, 2008). One difficulty of applying these analytical methods to revealed trait items is that the actions presented in a given scenario are not conceptualized as pure indicators of a single trait. Instead, a single action can be characterized in several different ways, and thus express multiple traits and expected effects. We believe that such an approach offers a unique perspective on trait covariation. Nevertheless, future researchers could apply multigroup CFA to revealed trait items. In particular, strict factorial invariance across groups may be required for comparisons between observed variances and covariances that are of interest in revealed trait analyses (Gregorich, 2006). However, two caveats are in order. First, it is likely that items will load onto more than one latent construct (e.g., dependable and organized). Second, the complexity of such models is likely to require sample sizes that are much larger than obtained in the present research. For example, Marsh et al. (2010) employed a sample of 1,570 respondents to test the gender invariance of a 60-item measure of Big Five personality traits. In their model, all items were allowed to load onto all five latent constructs.

An important step for future research is to investigate the utility of this method in cultural groups beyond the groups studied here. As we only examined a single group within each culture (i.e., students from a single university), it is possible that the extent of similarity found in these studies is limited to university students, or even to the specific universities from which the samples were drawn. To make more general conclusions regarding the reality and generality of cultural similarities and differences, it is certainly preferable to obtain multiple samples from each culture, as estimates will regularly fluctuate as a function of where the sample is drawn (e.g., college student vs. nonstudent adults in the United States; Peterson, 2001). Relatedly, this method may prove useful in comparing subgroups within a single culture. For instance, using this method alongside more typical self-report methods to investigate geographical variation in personality within the United States could be useful, as comparisons of self-reported trait levels across regions of the country have produced counterintuitive results that may reflect similar methodological artifacts as ones that affect cross-cultural comparisons (see Wood & Rogers, 2011). It may also be worth exploring how this methodology can be used to compare groups in very different cultures or contexts, where extremely different types of actions may be performed to manifest a particular type of trait such as industriousness or assertiveness (e.g., ballerinas, soldiers in the Israeli Defense Forces, preschoolers, Maasai women, astronauts). In addition to identifying the actions that reveal one’s trait level in different groups, the ability to correlate action likelihoods with abstract action characterizations in principle should allow the actions and trait levels of people from very different cultures to be more meaningfully compared.