How Emotional Intelligence Might Get You the Job: The Relationship Between Trait Emotional Intelligence and Faking on Personality Tests

ABSTRACT

This study examined trait emotional intelligence (EI) in relation to the ability to fake on personality tests. Undergraduate students (N = 129) were first instructed to fill out a personality inventory honestly, and subsequently in such a way as to maximize their chances of obtaining two distinctive job positions (lawyer and file clerk). Participants were able to change their scores in line with the hypothesized job profiles. Regression analyses showed that EI statistically predicted faking ability to an equal degree in both job scenarios. Finally, EI showed incremental validity over general mental ability and the Big Five personality traits in predicting the ability to fake. Possible implications of the results for the predictive validity of personality tests are discussed.

The widespread acceptance of the Big Five personality trait taxonomy has led to an increased use of personality assessments in personnel selection. This interest was fueled partly by meta-analyses showing the predictive power of personality on job performance and related outcomes (e.g., Barrick & Mount, 1991; Barrick, Mount, & Judge, 2001).

The upswing of personality assessments was accompanied by criticism on their use. The main concern is the possibility to distort responses on personality tests, for example, to answer in a socially desirable way (e.g., Ones, Viswesvaran, & Reiss, 1996). This may particularly be an issue in selection situations when the stakes are high. The tendency to respond in a socially desirable manner in order to maximize the chances of getting hired is often referred to as “faking” on personality measures (Mesmer-Magnus & Viswesvaran, 2006), yet others use the more neutral term “impression management” (J. Hogan, Barrett, & Hogan, 2007). Research has shown that individuals can and do provide a more favorable view of themselves on personality measures in selection procedures (Birkeland, Manson, Kisamore, Brannick, & Smith, 2006; Viswesvaran & Ones, 1999). Yet, whether this type of response distortion affects the predictive validity of personality tests remains rather unclear and is strongly debated in the literature (e.g., Cook, 2016; Dilchert, Ones, Viswesvaran, & Deller, 2006; R. Hogan, Hogan, & Roberts, 1996; Morgeson et al., 2007; Murphy & Dzieweczynski, 2005; Rosse, Stecher, Miller, & Levin, 1998; Tett & Simonet, 2011).

Earlier research using social desirability scales as an indication of faking has led some scholars to conclude that this type of response distortion has little effect on the predictive validity of personality constructs (e.g., Barrick & Mount, 1996; Hough, Eaton, Dunnette, Kamp, & McCloy, 1990; Li & Bagger, 2006). Controlling for scores on social desirability scales only has a small to negligible effect on the relation between personality constructs and work outcomes (Barrick & Mount, 1996; Li & Bagger, 2006; Ones et al., 1996; Schmitt & Oswald, 2006). On the other hand, social desirability scales have been criticized for their lack of construct validity: They appear to overlap with personality traits and thus seem to capture substantive and not error variance (Connelly & Chang, 2016; De Vries, Zettler, & Hilbig, 2014; Konstabel, Aavik, & Allik, 2006; McCrae & Costa, 1983; Ones et al., 1996; Uziel, 2010). In this regard, they do not seem to do what they were originally intended for, namely, detect faking (e.g., Holden, 2007, 2008).

More recent research indicated how faking can negatively affect the validity of personality test scores (Bing, Kluemper, Davison, Taylor, & Novicevic, 2011; Donovan, Dwight, & Schneider, 2014; Douglas, McDaniel, & Snell, 1996; Holden, 2007, 2008; O’Connell, Kung, & Tristan, 2011; Peterson, Griffith, Isaacson, O’Connell, & Mangos, 2011). In these studies, criterion validities were generally lower among participants who could be assumed to be faking (e.g., applicants) compared to those who were likely to respond truthfully (e.g., incumbents). Based on the suggestion that people differ in their amount of faking (McFarland & Ryan, 2000; Melchers et al., 2009; Viswesvaran & Ones, 1999; Zickar & Robie, 1999), scholars have warned that such differences can change the rank order of candidates in selection procedures and consequently influence the selection outcome, that is, who gets hired (Rosse et al., 1998).

Yet whether this poses a genuine problem for the test’s predictive validity—and the hiring organization—of course ultimately depends on whether those applicants who faked (and by result were hired) perform poorly on the job (Donovan et al., 2014). Some studies have indicated that this need not necessarily be the case by showing how faking can have a positive effect on personality validities by showing higher validities under conditions in which faking is likely to occur (e.g., Blickle, Momm, Schneider, Gansen, & Kramer, 2009; Ingold, Kleinmann, König, & Melchers, 2015; Klehe, Kleinmann, Nieß, & Grazi, 2014; Schmit, Ryan, Stierwalt, & Powell, 1995).

These contrasting findings might be due to the use of different definitions of what constitutes “faking”, different designs (within- or between-subject) and settings (e.g., instructed or naturally occurring faking), different strategies for classifying “fakers” (e.g., social desirability scales, magnitude of score changes or using “bogus” items), different focal constructs (e.g., multiple Big Five factors or one single trait), and different criteria (e.g., supervisor-rated performance or objective criteria). But more important, faking behavior appears to be a very complex phenomenon in which a multitude of factors simultaneously are at work (Komar, Brown, Komar, & Robie, 2008; Marcus, 2009). It is therefore imperative to identify what factors are responsible for the individual differences in faking behavior.

Several theoretical models have been developed to explain the variance in faking between individuals (Ellingson & McFarland, 2011; Goffin & Boyd, 2009; Marcus, 2009; McFarland & Ryan, 2000, 2006; Mueller-Hanson, Heggestad, Thornthon, 2006; Roulin, Krings, & Binggeli, 2016; Snell, Sydell, & Lueke, 1999; Tett & Simonet, 2011). Although the models differ somewhat in the antecedents or how these are labeled, they largely overlap, and three core elements can be identified: (a) (perceived) motivation to fake (b) (perceived) ability to fake and (c) (perceived) opportunity/risk of faking. The current study focuses on the second antecedent, namely, the ability to fake: Because faking ability and job performance have previously been linked (e.g., Viswesvaran & Ones, 1999), knowing what factors are related to the ability to fake might help us understand whether faking poses a threat to the predictive validity of personality tests.

That is, if factors influencing one’s ability to fake at the same time positively influence performance on the job, threats to validity by faking should be minimal. Scholars have indeed argued that how one behaves in selection procedures actually reflects one’s true social effectiveness, that is, genuine social skills and abilities (J. Hogan et al., 2007; Van der Linden, Te Nijenhuis, Cremers, Van de Ven, & Van der Heijden-Lek, 2014; Viswesvaran & Ones, 1999). In this line of argument, responding to a personality test is seen as a social interaction as any other where people try to convey a certain (i.e., positive) image of themselves and in which some will be more successful than others (R. Hogan, 1982, 2005; Marcus, 2009). This idea is supported by the fact that the same skills or traits that determine whether one is good at faking (or impression management) are likely to be the ones that are valuable and effective on the job (e.g., Blickle et al., 2008; R. Hogan & Shelton, 1998; Johnson & Hogan, 2006; Komar et al., 2008).

A wide range of conceptually similar but more or less distinctive constructs have been gathered under the collective term social effectiveness (see Ferris, Perrewé, & Douglas, 2002, for an overview). Each of these constructs relates to social competence and social abilities. Given that successful impression management or faking ability may depend on actual social knowledge and ability, one likely candidate among social effectiveness constructs is emotional intelligence (EI). Although scholars differ widely in their theoretical and operational definitions of EI, they all assume that high-EI individuals have the ability or tendency to act in socially effective ways (e.g., Ferris et al., 2002; O’Boyle, Humphrey, Pollack, Hawver, & Story, 2011).

The most important distinction in the current EI literature is between EI as an ability and EI as a trait. The former considers EI as a set of emotion-related cognitive abilities that can be assessed only through maximum performance tests, akin to the way cognitive intelligence would be measured. The trait conception considers EI as “a constellation of behavioral dispositions and self-perceptions concerning one’s ability to recognize, process, and utilize emotion-laden information” (Petrides, Frederickson, & Furnham, 2004, p. 278).

The present study adopts the trait EI model. The sampling domain of trait EI includes facets such as emotion perception, emotion expression, emotion regulation, adaptability, and social awareness/competence (Petrides, 2011). Using the trait EI model has the advantage that there are reliable and validated measures to assess the construct and studies and meta-analyses suggesting that trait EI is related to a range of other-rated and objective outcome measures such as health (Martins, Ramalho, & Morin, 2010); social behavior and leadership (Mavroveli, Petrides, Rieffe, & Bakker, 2007); and, most relevant to the present study, supervisor-rated and objective job performance (Joseph, Jin, Newman, & O’Boyle, 2015; Joseph & Newman, 2010; O’Boyle et al., 2011). As far as trait EI relates to understanding one’s own emotions and motivations and those of others, and entails the tendency to use this knowledge to be socially effective (e.g., Mavroveli et al., 2007; Sevdalis, Petrides, & Harvey, 2007), it is plausible that trait EI may also play a role in optimizing one’s chances of getting the desired job. Behaving effectively in selection settings has much to do with knowing what to say and do in order to create a favorable image of oneself in the eye of the hiring company (e.g., Levashina, Hartwell, Morgeson, & Campion, 2014; Melchers et al., 2009): High levels of trait EI may allow one to adequately achieve this.

Specifically, the effect of trait EI on faking ability can be summarized as follows. When applying for a job, the applicant finds him- or herself in an evaluative situation (Ellingson, Sackett, & Connelly, 2007; Kleinmann et al., 2011; Schmit & Ryan, 1993); unsure about what the hiring company is looking for exactly, the applicant has to interpret the situational cues at hand (e.g., personality test items or interview questions) in order to leave a good impression (Klehe et al., 2012; Roulin et al., 2016). In other words, when applying for a job it can be expected that one imagines what behavior is required or expected on the job and responds accordingly. It has been argued that people with higher levels of interpersonal skills are better at interpreting situational cues and using this information to choose the adequate response or type of behavior (Roulin et al., 2016). It therefore seems reasonable to expect EI-related social competences—that is, knowledge of what is expected in social situations and the ability to act accordingly—to affect one’s chances of providing the desirable responses during a selection situation.

The relation between EI and faking ability can conveniently be integrated within the existing theoretical models of faking behavior. Snell et al. (1999) and Ellingson and McFarland (2011) explicitly mentioned the influence of EI on faking but did not test it. Others have suggested a link between “analytical and behavioral skills” (Marcus, 2009) and “interpersonal skills” (Roulin et al., 2016), and faking ability. These traits or skills refer to social competences, which can be linked to EI. McFarland and Ryan (2000, 2006) proposed that knowledge of the measured constructs should be positively related to faking ability. This knowledge is related to EI in the sense that personality test items are assumed to serve as cues, which some individuals are better able to decipher: We hypothesize that higher levels of emotional and social knowledge will make it easier to grasp what an item is trying to convey. Finally, Goffin and Boyd (2009) posited a link between social astuteness, social intelligence and social skills, and faking ability. They constructed a decision tree that summarizes the faking process at the item level in six consecutive steps: Given their presumed higher level of social skills and knowledge, we can expect high-EI (vs. low-EI) individuals to be better able to identify whether a certain response is job relevant (Step 2) and whether the trait measured by the item would be (dis)advantageous on the job (Step 4; Goffin & Boyd, 2009, p. 157, Figure 2).

Research on other conceptually similar constructs provides initial evidence for this claim. For example, individuals’ ability to correctly perceive and interpret assessed performance criteria in selection procedures positively influences their performance herein (Holden & Jackson, 1981; Kleinmann et al., 2011; Raymark & Tafero, 2009). However, this ability is by definition context specific (Kleinmann et al., 2011; Melchers et al., 2009) and may be a more specific derivative of one’s more general level of EI. By a better understanding of the consequences of given actions in general (Sevdalis et al., 2007), high-EI individuals can be expected to know what behaviors are required across various contexts (e.g., at home, in school, or at work), with selection procedures being one of such contexts.

Although previous studies have examined the fakability of EI measures (Day & Carroll, 2008; Tett, Freund, Christiansen, & Coaster, 2012; Whitman, Van Rooy, Viswesvaran, & Alonso, 2008) and the consequences of faking on their validity (e.g., Choi, Kluemper, & Sauley, 2011), these studies did not yet take notice of EI as one of the potential factors that can actually influence faking behavior. After all, “high emotional intelligence may give individuals greater insight into the nuances or subtle ramifications of a given behavior, and this may allow them to choose more appropriate responses for a given situation” (Snell et al., 1999, p. 223). In light of the preceding, it can be expected that individuals high on trait EI are better able to fake on personality tests and to provide the desired responses that increase their chances of obtaining the desired job.

H1:

Trait EI is positively related to the ability to fake on personality tests in selection procedures.

Our second hypothesis concerns the incremental validity of EI over and above cognitive abilities and personality in predicting the ability to fake. General intelligence or general mental ability (GMA) has previously been found to be related to faking ability and impression management (Huffcutt, Roth, & McDaniel, 1996; Mersman & Shultz, 1998; Pauls & Crost, 2005; Vasilopoulos & Cucina, 2006). Adequate faking may be more cognitively demanding than responding honestly (Van Hooft & Born, 2012) and therefore require a prerequisite level of intelligence (Bing, Whanger, Davison, & VanHook, 2004); this notion is supported by studies finding personality scales to become more g-loaded under faking instructions (Bing et al., 2004; Mersman & Shultz, 1998) or when a more fake-resistant forced-choice item format is used (e.g., Christiansen, Burns, & Montgomery, 2005; Vasilopoulos, Cucina, Dyomina, Morewitz, & Reilly, 2006). Furthermore, GMA may be instrumental in one’s test-taking strategy (Snell et al., 1999) and in correctly understanding what is meant and measured by the items in a personality test (Pauls & Crost, 2005; Tett et al., 2012). Yet, for reasons just outlined, we expect trait EI to facilitate in assessing whether agreeing or disagreeing with an item will increase the chances of getting the desired job. Thus, although GMA as well as trait EI should be important in the explanation of the ability to fake, their effects should not fully overlap.

Whether and how GMA and EI influence one’s ability to fake is related to the debate in the literature on how distinctive social effectiveness constructs and GMA are (Ferris et al., 2002; Ferris, Witt, & Hochwater, 2001; Melchers et al., 2009). Conceptualized as a trait, the relations between EI and GMA appear to be weak (Petrides, Furnham, & Mavroveli, 2007). Thus, it can be expected that trait EI will account for variance in the ability to fake when controlling for the effects of GMA.

In addition, trait EI has often been criticized for showing too much overlap with “traditional” personality taxonomies such as the Big Five (e.g., MacCann, Matthews, Zeidner, & Roberts, 2003; Schulte, Ree, & Carretta, 2004) and a lack of incremental validity beyond such taxonomies. However, these accounts fail to realize that the overlap with personality factors such as the Big Five is in line with the conceptualization of trait EI as partly interwoven with personality traits (Petrides, 2011; Van der Linden et al., 2017). As such, strong correlations between trait EI and the personality traits that define it can and should in fact be expected (Petrides, Pérez-González, & Furnham, 2007). Despite this, several studies have shown that trait EI actually shows incremental validity over and above the Big Five in the prediction of, for example, life satisfaction, coping styles, and stress (Andrei, Siegling, Aloe, Baldaro, & Petrides, 2016; Petrides, Pérez-González, et al., 2007) and most important for the current study, job performance (O’Boyle et al., 2011). Based on this, trait EI can be assumed to entail affect-related variance (e.g., taking the perspective of others, social awareness) that is perhaps not directly encapsulated in existing personality taxonomies such as the five-factor model, variance that is positively related to the ability to fake.

Based on the previous discussion on the incremental validity of trait EI, our second hypothesis states the following:

H2:

Trait EI will explain variance in the ability to fake on personality tests in selection procedures over and above the variance explained by general mental ability and the Big Five personality traits.

The present study

The present study builds on the previous work of Raymark and Tafero (2009), who assessed the effects of individual differences on the ability to fake. In their between-subject design, participants were instructed to either “fake good” (i.e., try to leave the best possible impression) or respond so that it would assure them of getting them the position of an accountant. They found that the produced personality profiles differed considerably between the two conditions. Although they were not the first to discover that differential instructional sets result in different personality profiles (Furnham, 1990; Mahar, Colognon, & Duck, 1995; Martin, Bowen, & Hunt, 2002; Pauls & Crost, 2005), they concluded, “Further research is needed, examining a number of different jobs, to determine whether individual differences in ability to fake toward a job are stable across jobs” (p. 101). The present study directly addresses this question by using a within-subject design to investigate faking toward two very distinctive jobs and the role of trait EI in doing so. In addition, the incremental validity of trait EI over personality and GMA is assessed. Finding effects of EI on the ability to fake over and above the effects of personality and GMA would further emphasize the relevance of trait EI as a factor in the explanation of faking ability.

Method

Sample and design

Participants were undergraduate psychology students who participated for course credits. Of the 139 total respondents, 10 participants responded incorrectly to the manipulation check (see the Procedure section) and were excluded from the analyses. This resulted in a final sample of 129 participants (105 female = 81.4%), with a mean age of 20.39 years (SD = 3.64) ranging from 17 to 42 years old.

A key part of the study were two scenarios in which the participants were instructed to respond to a personality test in such a way that they would have the highest probability of getting the job, even though their answers to the items may not necessarily reflect how they really are. Henceforward, we refer to these two scenarios as the “faking scenarios.” For these scenarios, we largely followed the procedures as described by Raymark and Tafero (2009).

Faking scenarios

Analogous to Raymark and Tafero (2009), we extracted the job profiles for our scenarios from a study by Raymark, Shilobod, and Steffensmeier (2004). In their study, job profiles (i.e., required score patterns) were created after consulting test manuals and prior studies, and 30 experienced I/O-psychologists and 148 undergraduate students were asked to judge which personality facets of the NEO Personality Inventory–Revised (NEO–PI–R; Costa & McCrae, 1992) would be positively or negatively related to performance in a variety of jobs.

The profiles were constructed at the facet level, because some facets within higher order personality dimensions can be differentially related to different types of jobs (Raymark & Tafero, 2009). Therefore, focusing on facets captures subtle differences between jobs that would not be captured at the factor level. Moreover, previous studies have shown individuals to fake discriminatively, that is, increasing scores on some facets (or even items) and decreasing them on others (Donovan et al., 2014; Goffin & Boyd, 2009; Scherbaum, Sabet, Kern, & Agnello, 2013).

Table 1 shows the personality profiles of the jobs used in the scenarios. From the Raymark et al. (2004) study, we selected the lawyer and file clerk positions because these two profiles included the most facets for which the relations with performance were hypothesized to be in opposing directions. For example, the NEO-PI-R facet gregariousness is hypothesized to be positively related to performance in the lawyer position, whereas it is hypothesized to be negatively related to performance in the file clerk position. For compliance, a facet of agreeableness, on the other hand, the reverse pattern was expected.

Table 1. Job Profiles for Faking Scenarios.

File Clerk
Positive facets	Cooperation (A), Orderliness (C), Dutifulness (C), Self-discipline (C)
Negative facets	Gregariousness (E), Assertiveness (E), Activity (E), Excitement seeking (E), Adventurousness (O)
Neutral facets	Trust (A), Morality (A), Lack of Self-consciousness (ES), Emotionality (O)
Lawyer
Positive facets	Orderliness (C), Self-discipline (C), Gregariousness (E), Assertiveness (E), Activity (E), Lack of Self-consciousness (ES)
Negative facets	Trust (A), Morality (A), Cooperation (A)
Neutral facets	Dutifulness (C), Excitement seeking (E), Emotionality (O), Adventurousness (O)

Note. Self-consciousness is keyed in the positive direction. O = Openness to Experience; C = Conscientiousness; E = Extraversion; A = Agreeableness; ES = Emotional Stability. Positive and negative facets based on Raymark et al. (2004).

We also added facets that were positively related to performance in one job and unrelated to performance in the other (e.g., trust; see Table 1). This allowed us to test whether trait EI was particularly related to picking out the job-relevant traits and faking accordingly. Finally, we included facets for which the relation with performance was the same in both jobs (e.g., orderliness) for comparison purposes and in order to further complicate identification of the job-relevant traits. In total, the profiles consisted of 13 facets: three Conscientiousness facets, four Extraversion facets, two Openness facets, three Agreeableness facets, and one Emotional Stability facet (see Table 1).

The construction of such mixed profiles complicates producing the correct profiles (Raymark & Tafero, 2009). Rather than just scoring high on all facets (i.e., the simple heuristic “higher is better”), participants will need to carefully consider whether responding positively or negatively to a certain item will produce the correct profile. This will result in a more robust test for the effect of EI on faking behavior (Raymark & Tafero, 2009).

Measures

Personality test. Personality assessment was done with an adapted version of Johnson’s 120-item NEO-PI-R consisting of items from the International Personality Item Pool (IPIP; Goldberg et al., 2006). The IPIP is a public domain collection of items for the use of personality questionnaires. Johnson (2011) created a short 120-item questionnaire to reflect the facets and factors of the NEO-PI-R, with sufficient reliability and validity properties.

For most of Johnson’s 120 IPIP items we found the original Dutch version of the IPIP items as presented in Hendriks (1997). However, no Dutch versions were available for 49 of the items (40%). If that was the case, a different item belonging to the same facet was taken from the full 300-item version for which a Dutch version was available. When this method did not provide an alternative item, a Dutch item was chosen from the original study by Hendriks (1997) that most closely matched the content of the English item. For 10 items (8%), both methods did not lead to an alternative item, and those 10 were translated by the first author and checked by an undergraduate student: Any discrepancies were discussed and translations altered when needed. One item was added to each facet to improve facet reliabilities. The same methods for the addition of items as for obtaining Dutch versions of items were used, as previously discussed.

The final personality test thus included 150 (120 + (6 × 5 × 1)) items in a 5-point Likert scale format. An example item of the gregariousness facet is “Starts conversations,” an example item of the dutifulness facet is “Keeps his/her promises.” Reliability coefficients for the facets ranged from .39 (immoderation) to .80 (self-discipline), with an average of .68. Although four values were low (< .60), these specific facets (immoderation, intellect, liberalism, and sympathy) were all facets that were not relevant for the faking scenarios (see Table 2). Reliability coefficients for the Big Five dimensions were adequate to good in the current sample (.86, .88, .76, .84, and .87 for Emotional Stability, Extraversion, Openness, Agreeableness, and Conscientiousness, respectively).

Table 2. Mean Facet Scores in Honest Condition and Faking Scenarios.

	Honest			File Clerk								Lawyer
Facet	α	M	SD	M	SD	ΔM	t	d	H	α	rdd	M	SD	ΔM	t	d	H	α	rdd
Trust (A)	.71	16.96	3.00	18.43	3.04	1.47	−4.22	−0.48	NH	.76	—	13.04	3.46	−3.92	10.54	1.21	—	.72	.66
Morality (A)	.72	19.32	2.72	22.38	2.19	3.06	−10.63	−1.24	NH	.68	—	15.37	3.60	−3.95	10.79	1.23	—	.66	.63
Cooperation (A)	.63	17.53	2.83	20.81	2.37	3.28	−12.56	−1.25	+	.51	.36	12.93	3.38	−4.60	12.71	1.47	—	.71	.63
Orderliness (C)	.77	16.41	3.54	23.98	1.72	7.57	−23.12	−2.68	+	.79	.75	20.01	3.10	3.60	−8.65	−1.08	+	.73	.75
Dutifulness (C)	.64	19.88	2.37	23.36	2.05	3.48	−13.45	−1.57	+	.78	.66	20.07	3.51	0.19	−0.53	−0.06	NH	.76	—
Self-discipline (C)	.80	16.25	3.34	23.05	2.09	6.80	−19.49	−2.44	+	.63	.75	23.36	2.07	7.12	−20.06	−2.57	+	.77	.80
Gregariousness (E)	.68	17.25	3.21	13.33	3.20	−3.92	9.73	1.22	—	.70	.69	19.28	2.11	2.03	−6.13	−0.75	+	.29	.54
Assertiveness (E)	.67	15.49	3.00	15.45	3.10	−0.04	0.11	0.01	—	.70	.60	21.28	2.33	5.79	−16.71	−2.16	+	.47	.63
Activity (E)	.66	15.58	2.72	19.39	2.56	3.81	−12.03	−1.44	—	.50	.55	19.82	2.20	4.24	−13.87	−1.72	+	.19	.47
Excitement seeking (E)	.78	15.93	3.41	11.81	3.08	−4.12	9.83	1.27	—	.70	.76	19.89	2.70	3.96	−11.53	−1.28	NH	.63	—
Self-consciousness (ES)	.66	16.12	3.06	18.30	3.83	2.18	−5.18	−0.63	NH	.79	—	23.09	2.11	6.97	−21.35	−2.65	+	.70	.67
Emotionality (O)	.78	14.11	3.64	9.26	2.57	−4.85	12.57	1.54	NH	.67	—	8.22	2.06	−5.89	15.68	2.00	NH	.46	—
Adventurousness (O)	.61	14.76	2.70	12.52	3.05	−2.24	5.81	0.78	—	.56	.64	17.93	2.28	3.17	−10.53	−1.27	NH	.42	—

Note. All mean differences are significant (p < .01), except those in bold. Possible range of 5–25. Reliabilities of the difference scores were calculated using the formula: r_dd = (σ²_d- σ²_ed)/σ²_d where σ²_ed = σ²_h (1-r_hh) + σ²_f (1-r_ff) and h refers to the honest condition and f to the faking condition (McFarland & Ryan, 2006). O = Openness to Experience; C = Conscientiousness; E = Extraversion; A = Agreeableness; ES = Emotional Stability; H = hypothesized relationship; + = positive, — = negative; NH = no hypothesized relationship.

Trait EI

The short version of the Trait Emotional Intelligence Questionnaire (Petrides & Furnham, 2006) was used, which currently is the most well known and widely used measure of trait EI. It has the advantage over other self-report EI measures in that its development is explicitly based on a theoretical framework (Petrides, 2011).

The short version of the Trait Emotional Intelligence Questionnaire consists of 30 items and is designed to measure global trait EI (comprising four underlying factors: well-being, self-control, emotionality, and sociability). Questions are in a 7-point Likert-scale format. The Dutch version has been validated in two samples (Petrides et al., 2010). Example items are “Expressing my emotions with words is not a problem for me” and “I’m usually able to influence the way other people feel.” The internal consistency of the global trait EI scale in the current sample was good (α = .88; see Table 3).

Table 3. Means, Standard Deviations, Reliability Coefficients, and Intercorrelations Between Variables.

General mental ability

GMA estimates of the participants were collected independently from the current study but in about the same period. As part of their curriculum, students completed three tests of the Dutch intelligence test series by Drenth, Van Wieringen, and Hoolwerf (2001): They consecutively completed the Verbal Analogies Test, Numerical Capacity (number series) Test, and Nonverbal Analogies (abstract reasoning) Test. All tests were completed in small groups (12 students) in silence, under the supervision of a test assistant. Students were given 40, 30, and 20 min for the verbal, number series, and abstract reasoning test, respectively, as indicated in the test manual. All participants completed a consent form stating that their results could be used for research purposes. Ability test scores could not be collected for all participants, leading to 102 participants for whom complete information was available.

A principal component analyses was conducted on the scores on the three ability tests. A clear single factor indicating GMA, emerged, explaining 56.56% of the variance with factor loadings of .78, .69, and .78 for the verbal, numerical, and nonverbal reasoning test, respectively. Factor scores were saved using the regression method as an indication of participants’ GMA.

Ability to fake

The ability to fake was operationalized as the difference between the participants’ score on a facet from the faking scenario and their honest score on that facet (Raymark & Tafero, 2009). This difference score was keyed positive when the change in scores from the honest to the faking scenario was in the expected direction and keyed negative when it was in the opposite direction. There were nine ability to fake scores (i.e., keyed difference scores) for each scenario, one for each facet for which a specific relation with the job position was expected (see Table 1). These difference scores within each scenario were summed, resulting in two overall ability to fake scores: (a) the ability to fake toward the file clerk profile and (b) the ability to fake toward the lawyer profile. Higher scores reflected a higher ability to fake.

Opportunity to fake

Participants with a lower honest score on a facet that needed to be faked upwardly to get the job (e.g., cooperation that was hypothesized to be positively related with the file clerk position) have more opportunity to fake than people with a higher honest score on that facet (Goffin & Boyd, 2009). Conversely, people with higher honest scores on a facet with a negative relation with a job (e.g., trust in the lawyer scenario; Table 1) have more room to fake than people with lower honest scores on this facet. Therefore, analogous to the procedure followed by Raymark and Tafero (2009), we obtained a person’s opportunity to fake score by taking the difference between the participants’ honest score on a facet and the optimal score on that facet in the faking scenarios. For the positive facets, the optimal score was the highest possible score. For the negative facets, the optimal score was the lowest possible score. Again, in total there were nine opportunity to fake scores for the lawyer scenario and nine for the file clerk scenario (one for each job relevant facet; see Table 1). As for the ability to fake score, within each scenario these nine scores were summed to form two overall opportunity to fake scores, with higher scores implying more opportunity to fake.

Procedure

All materials were administered online. Participants could complete the questionnaires (except the cognitive ability measures) at their own convenience from any computer. At the beginning of the survey and before the start of the first personality and EI measure, we provided the instructions to answer as honestly as possible to reflect the participant’s true score. Thereafter, the test battery consisted of the two faking scenarios. The honest condition should therefore be seen as the “baseline” measurement, whereas the faking scenarios served as a task or “test” in the true sense—in which one could either be good or bad. Operationalized in this way, that is, as a maximum performance test, score changes (faking) truly reflect an ability in the current study.

The instructions for the scenarios included a formal job description from a Dutch website similar to O*NET (“werk.nl”), including regular tasks and duties. This was done to ensure that all participants had the same background information on the jobs. Possible biasing effects of differences in job familiarity or knowledge (Goffin & Boyd, 2009; Raymark & Tafero, 2009; Snell et al., 1999) were therefore eliminated with this procedure. To create an incentive, participants were informed that for every 50 participants, the person who did best in producing the correct job profiles was selected and rewarded with €10 (approximately US$11).

After completing all questionnaires, participants received a manipulation check, consisting of an item asking them to mark the two ways in which they had responded to the personality questionnaires in the faking scenarios. The response options were: As honest as possible, Applying for the position of file clerk, As socially desirable as possible, Applying for the position of lawyer, and Applying for the position of accountant. Those participants (10 in total) who marked one or more false options were excluded from the analyses. As a final check, data screening analyses as described by Meade and Craig (2012) were conducted to assess the quality of our data, which showed that there were no influential outliers or no indications of irregular data points.¹

Statistical analyses

Participants’ overall ability to fake, overall opportunity to fake, and trait EI scores were used as input for regression analyses in order to test H1. Analyses were done separately for the file clerk and lawyer position, with the dependent variable being the overall ability to fake score. The overall opportunity to fake was included in the first step and trait EI in the second step.

To test the incremental validity of trait EI over GMA and the Big Five (H2), hierarchical regression analyses were conducted. Opportunity to fake, GMA, and each of the Big Five dimensions were simultaneously entered in Step 1, and EI was entered in Step 2. Changes in R² between Step 1 and Step 2 were scrutinized for evidence in support of H2. Both hypotheses were tested against a one-sided p value of .05, given the directional nature of our hypotheses.

A few notes should be made here on issues of multicollinearity in our tests of H1 and H2. A difference score is by definition correlated with the components from which it is derived (Burns & Christiansen, 2011; Pedhazur & Schmelkin, 1991). Thus, if trait EI is highly correlated with a personality facet in the honest condition, then this shared variance between EI and the facet itself could be responsible for any relation found between EI and the difference score based on this facet. We therefore first scrutinized the strength of the relations between EI and honest facet scores before proceeding to our test of H1.

Similar multicollinearity issues pertain to our test of H2 because the Big Five scores are by definition incorporated in the opportunity to fake score (i.e., the optimal facet score minus honest facet score). However, multicollinearity is an issue only for the interpretation of the beta weights of the predictors, not for the proportions of explained variance of (a set of) predictors (Kutner, Nachtsheim, & Neter, 2004). Thus, for our test of H2, we merely looked at increments in explained variance rather than at the specific effects of the individual predictors on the outcome.

Note that the opportunity to fake and ability to fake scores are difference scores. Some scholars have cautioned against the use of difference scores due to issues with unreliability and statistical artefacts such as the aforementioned autocorrelations with the measures from which they are derived (Burns & Christiansen, 2011; Edwards, 1994; Edwards & Parry, 1993). In the present study, however, these issues are addressed in the following ways. First, when a Participant × Treatment interaction is expected, difference scores are appropriate to use (McFarland & Ryan, 2000; Tisak & Smith, 1994). This was true in our case, as we hypothesized differences in score changes (i.e., faking ability) between people in the faking conditions (i.e., the “treatment”) based on individual differences (i.e., EI). Second, the issue of potential lower reliabilities of difference scores is partly mitigated because we use variables based on aggregated difference scores. Third, the limitation of spurious correlations is solved by controlling for the opportunity to fake (essentially keyed honest scores). This transforms our dependent variable from a “raw” difference score into a “regression-adjusted” difference score (Burns & Christiansen, 2011). As such, the dependent variable, faking ability, can be interpreted as the part of the scores in the faking condition that cannot be explained by the honest score, hereby identifying who changed his or her score more (or less) than expected based on their initial trait standing (Burns & Christiansen, 2011).

In addition to reporting the results based on difference scores, we included polynomial regression analyses (Edwards, 1994; Edwards & Parry, 1993). Polynomial regression averts the problems associated with difference scores and has the ability to give insight into the (in)congruence between variables (e.g., honest-faking scores) and a third variable (e.g., EI). Guided by the work of Edwards and colleagues (Edwards, 1994, 2002; Edwards & Parry, 1993), we compared the explained variance of a linear model and a quadratic model in hierarchical regression analysis, both for the file clerk and lawyer scenario:

EI = b₀ + b₁FS + b₂HS + e

EI = b₀ + b₁FS + b₂HS + b₃FS² + b₄FSHS + b₅HS² + e

where FS = score in faking scenario and HS = score when asked to respond honestly. If the quadratic model shows a significant increase in explained variance, the coefficients of this model are used to draw and interpret a response surface representing the relation between the honest scores and faking scores simultaneously, and EI. The stationary point and the principal axes are essential elements of this response surface (Edwards & Parry, 1993) because they respectively indicate the point where the slope is zero in all directions and the lines along which the upward/downward curvature is greatest/smallest. Prior to the polynomial regression analyses, all facet scores were rescaled to its original 5-point scale, keyed in the same positive direction and summed within each scenario. These sum scores were then mean centered to facilitate interpretation of the regression model and response surface.²

Results

Descriptive statistics

Table 2 shows the mean facet scores in the honest and the faking conditions. Whether the participants actually faked—and if so, in the right direction—can be inferred from the differences in mean facet scores in the honest and faking conditions. Replicating the findings by Raymark and Tafero (2009), participants were overall able to produce the required job profiles quite well. Scores on the cooperation facet, for example, were significantly elevated in the file clerk condition, t(128) = 12.56, p < .001, whereas significantly lowered in the lawyer condition, t(128) = −12.71, p < .001. All facets were faked in the hypothesized direction in the lawyer condition. In the file clerk condition, only activity was faked in the nonpredicted direction, t(128) = 12.03, p < .001, whereas the change in assertiveness between the honest and faking condition was not significant, t(128) = −.11, p = .92. Across both scenarios, changes in scores were largest for positive facets, followed by the negative facets and smallest for the neutral facets (i.e., for which no clear relation with the job positions were expected), as indicated by their mean effect sizes d (Table 2; 1.89, 1.08, and 1.06, respectively). Overall, these effect sizes differed significantly from one another, F(2, 23) = 5.30, p = .01, although the only significant differences were between the positive facets, on one hand, and the neutral (p = .03) and negative facets (p = .03), on the other.

Overall, participants were more inclined to increase their scores than to lower them, a finding also reported elsewhere (e.g., Raymark et al., 2004). Perhaps it is easier to identify desirable as opposed to undesirable traits for a job, or perhaps participants deemed it useful to, in case of doubt, go for the general socially desirable direction. In line with this tendency, the neutral facets were mostly faked upward, with the exception of emotionality. However, this specific finding can probably be explained by the fact that the items of the emotionality facet (e.g., “has crying fits”) were likely to reside at the lower end of the social desirability spectrum. Of interest, scores on dutifulness, a facet of conscientiousness, were not significantly altered in the lawyer condition, t(128) = .53, p = .60. Dutifulness was assumed to be a neutral facet in the lawyer condition: Although it measures aspects such as keeping promises and being truthful, which can generally be expected to be favorable assets for any type of job, based on these findings it seems as if participants were not sure whether being dutiful would ensure them of getting the lawyer position.

Table 2 shows that in the faking conditions, the reliabilities of the facets generally became lower, and in some cases were rather low (e.g., .19 for activity in the lawyer condition). Facet reliabilities have generally been shown to decrease under faking conditions (McCann, 2013; Ziegler, Danay, Schölmerich, & Bühner, 2010). We elaborate on this in the Limitations and Future Studies section. It also has to be noted here that we created the overall ability to fake measures by summing over multiple difference scores, hereby increasing the reliability of these composite measures (Raymark & Tafero, 2009).

Table 3 shows the means, standard deviations, and correlations between all variables. Coefficient alphas, where applicable, are shown on the diagonal.

The correlation between EI and GMA was positive but not significant (r = .14), in line with their respective conceptualizations as a trait and an ability. As noted previously, of special interest from a multicollinearity perspective are the correlations between EI and the facets in the honest condition, because these facets were used to create our dependent variable. The maximum correlation between EI and the facets in the honest condition was .51 (with the lack of self-consciousness facet), whereas most correlations were much lower (average absolute r = .26). Thus, although trait EI and the facets in the honest condition showed some overlap, their shared variance did not appear to be problematic for our test of H1.

Initially, there was a nonsignificant relation between the faking abilities on the two jobs. However, this appeared to be due to differences in the opportunity to fake (i.e., the possibility to go up or down the scale), which often was in the opposite direction in the two jobs (see Table 1). After controlling for opportunity to fake their respective opportunity to fake scores, the correlation between the faking ability scores in the two jobs became r = .40 (p < .001). This shows two things. First, it confirms that the participants showed some level of stability in the extent to which they faked. Second, it underlines the effect of the opportunity to fake and the need to control for this when testing hypotheses (Raymark & Tafero, 2009).

Hypothesis testing

EI and faking ability (H1)

Controlling for the opportunity to fake, trait EI was significantly related to the overall ability to fake (i.e., the sum of the job-related difference scores) for both job positions (file clerk: β = .13, p = .02; lawyer: β = .15, p = .01; see Table 4, Model 1b). The beta values found for the clerk and the lawyer did not significantly differ from each other, Z(b_laywer – b_{file clerk}) = −.28, p = .78, indicating a similar influence of EI on the level of faking in both scenarios. This supports the predicted positive relation between trait EI and faking ability (H1). GMA related to overall ability to fake in the expected direction, but the effect did not reach significance (Table 4, Model 2: β_{file clerk} = .08, p = .13; β_laywer = .09, p = .09).³

Table 4. Summary of Hierarchical Regression Analyses for Variables Predicting the Ability to Fake.

Polynomial regression analyses (H1)

In the file clerk condition, the addition of the quadratic and interaction terms (Step 2) did not result in a significant increase of explained variance of EI (Table 5). Thus, in the file clerk condition, faking was best described by a linear model, indicating a significant main effect of EI on faking (β = .17, p = .04) independent of honest scores. This finding further supports H1.

Table 5. Results From Polynomial Regression Analyses Predicting Trait Emotional Intelligence.

	File Clerk						Lawyer
Model	b	SE	β	R^2	ΔR^2	ΔF	b	SE	β	R^2	ΔR^2	ΔF
Step 1				.13**		9.60**				.23**		18.31**
Faking score	.39	.19	.17*				.38	.17	.17*
Honest score	−.71	.18	−.33**				1.02	.18	.45**
Step 2				.14**	.01	.39				.28**	.05*	2.83*
Faking score	.41	.20	.18*				.49	.18	.22**
Honest score	−.73	.18	−.34**				1.06	.18	.47**
Faking score squared	.23	.45	.04				.50	.38	.11
Honest score squared	−.39	.44	−.08				−1.16	.48	−.19*
Faking × Honest score	.14	.68	.02				−.48	.64	−.06

*p < .05. **p < .01, two-sided.

In the lawyer condition, the quadratic and interaction term in Step 2 accounted for a significant portion of additional variance in trait EI beyond the linear terms (Step 1). In this model, the coefficients for the linear effect of the faking score and the linear and quadratic effects of honest score were significant (Table 5). The surface plot corresponding to the full model showed a complex saddle-shaped surface. The stationary point was found at X = −.25 and Y = .51. The first principal axis ran along the line Y = .47 – .14X, whereas the second principal axis was found well outside the range of the honest and faking scores at Y = 2.24 + 7.02X.

The response surface indicated that, in line with H1, at a given honest score, higher faked scores were generally associated with higher EI levels. However, this was less so the case when honest scores were high: EI was more strongly associated with higher faked scores at lower honest scores, whereas at high honest scores, the surface was essentially flat.

Incremental validity of EI over GMA and Big Five personality traits (H2)

EI showed incremental validity over the effects of GMA and personality in the prediction of faking ability (Table 4). Entering EI (Model 3b), after including both GMA and the Big Five (Model 3a), resulted in a significant increase in explained variance (ΔR²_{file clerk} = .03, p = .02 and ΔR²_lawyer = .04, p = .00).⁴ In contrast, adding the Big Five simultaneously in Step 2 after entering GMA and EI in Step 1 did not result in significant increases in explained variance in the ability to fake (ΔR²_{file clerk} = .02, p = .56 and ΔR²_lawyer = .05, p = .08). Thus, trait EI adds unique variance beyond personality in the explanation of the ability to fake, but the reverse could not be supported.

Additional analyses: Trait EI and faking on the facet level

In addition to the overall ability to fake, the effect of EI on the ability to fake on each of the specific facets separately was investigated. Due to the relatively moderate reliabilities of the difference scores (Mdn_lawyer = .63 and Mdn_{file clerk} = .66), however, these results should be interpreted with due caution.

Significant relations between EI and faking on the specific facets are reported in Table 6. For most individual facets, no significant effects of trait EI on score changes were found. For the relations that were significant, EI appeared to influence the ability to fake on lack of self-consciousness, morality, and cooperation in the lawyer condition and on gregariousness, excitement seeking, and cooperation in the file clerk condition. The strengths of the relations between EI and faking ability were roughly equal across facets (.12 < β < .18).

Table 6. Summary of Hierarchical Regression Analyses for Variables Predicting the Ability to Fake on the Facet Level.

		File Clerk							Lawyer
Dependent Variable	Independent Variable	b	SE	β	rxy.z	R^2	ΔR^2	F	b	SE	β	rxy.z	R^2	ΔR^2	F
Ability to fake
Lack of self-consciousness	OTF – Lack of self-consciousness								1.08	.07	.89**	.81	.69	.01*	139.28**
	TEI								.78	.34	.13*	.20
Morality	OTF – Morality								.79	.11	.52**	.53	.30	.03*	26.96**
	TEI								1.16	.49	.18**	.21
Cooperation	OTF – Cooperation	.69	.07	.66**	.67	.46	.02*	53.73**	.86	.11	.59**	.59	.35	.01*	34.35**
	TEI	.60	.31	.13*	.17				.78	.47	.12^†	.15
Gregariousness	OTF – Gregariousness	.95	.10	.66**	.66	.52	.01*	69.20**
	TEI	.92	.49	.13*	.17
Excitement seeking	OTF – Excitement seeking	1.01	.09	.72**	.73	.59	.01*	92.34**
	TEI	.84	.46	.11*	.16

Note. OTF = opportunity to fake; TEI = trait emotional intelligence; r_xy.z = partial correlation controlling for other variable in the model.

^† p = .05, one-sided. *p < .05. **p < .01.

There are three important things to note about these results. First, EI was related to faking on facets that were (hypothesized to be) faked both upward and downward. In other words, EI was not merely related to “faking good” (i.e., faking toward the socially desirable end of the scale) but also related to one’s ability to “fake bad.” Second, EI had an effect on faking on the cooperation facet, which was faked in opposing directions in the two scenarios as expected (see Table 2). Third, EI was related to faking on job-relevant traits only, as no relations were found between EI and changes in scores on facets for which no relationship were proposed (Table 1). For example, morality was expected to be faked downward in the lawyer position, whereas no relation was hypothesized in the file clerk position. Although scores were altered in both scenarios (Table 2), only a significant effect of EI was found in the lawyer condition. Again, however, no firm conclusions should be derived from these results.

Discussion

The present article showed that trait EI was related to the ability to fake on personality tests and that the effects were similar when asked to fake toward two jobs with divergent profiles. That is, when the job indicated that higher scores on certain traits were better, high-EI individuals shifted their scores more strongly in that direction than low-EI individuals. Similarly, when a job indicated that lower scores on certain traits might be better, higher EI was associated with stronger effects in that direction. The nature of the effects and the effect sizes were roughly in line with those found for other traits known to be related to the ability to fake, such as openness (Raymark & Tafero, 2009), integrity (McFarland & Ryan, 2000), and efficacy of self-presentation (Pauls & Crost, 2005). In addition, trait EI showed incremental validity over GMA and the Big Five personality dimensions in its relation with the ability to fake. Finally, although not the main focus of this study, in the present sample no significant effect of GMA on the ability to fake was found. Because the participants were all university students (and from the same study, psychology) there may have been a restriction of range in intelligence, reducing relationships with criteria (Hunter & Schmidt, 2004). Using the standard deviations of the population from the test manual, we tested whether this was true. Corrected for range restriction, the correlation between GMA and faking ability (independent of faking opportunity) increased from .11 to .14 (p = .08, one-sided) in the file clerk condition and from .13 to .17 (p = .04, one-sided) in the lawyer condition. Thus, homogeneity of the sample may have partly been responsible for the nonsignificant relation between GMA and faking ability.

Important to note, our main findings were robust across methods of analyses: The regression-based results were confirmed in polynomial regression analyses, a sophisticated technique gaining popularity in faking research (e.g., Peterson et al., 2011). Results from the lawyer condition implied that the effect of EI on faking was more pertinent when honest scores were low compared to when honest scores were high. A possible theoretical explanation could be that with high honest scores, faking ability becomes less relevant, because there may be less need to fake in order to leave a good impression (Goffin & Boyd, 2009). Put differently, when honest scores are low and self-presentation is required to get the job, we can expect individual differences in EI-related competences to become more pronounced. However, there could also be a methodological explanation: The lawyer scenario included three extraversion facets—with a relatively strong positive relation to EI—which needed to be faked upward. Thus, those with higher honest scores on those facets are also more likely to have higher EI scores, reducing the likelihood of finding a positive effect of EI on faking for this particular group of individuals. Given that the findings did not generalize across scenarios, this explanation appears to be likely. Nonetheless, the results do emphasize that honest scores are an important factor in faking behavior and therefore should not be overlooked (Goffin & Boyd, 2009; Marcus, 2009; McFarland & Ryan, 2000).

Overall, the present study has implications for both the EI literature and the faking literature, and it connects these two fields theoretically. First, EI appeared to be a relevant factor in the ability to fake across jobs. We found that individuals who can fake toward the profile of one job are also able to fake toward the other to a considerable degree (r = .40 with respective opportunities to fake controlled). This result supports previous findings implying that some people are consistently better at faking or impression management than others (Ingold et al., 2015; Klehe et al., 2014; McFarland & Ryan, 2000). The present study provided initial evidence that EI can be partly responsible for such individual differences in faking ability. In the sense that trait EI reflects dispositions related to genuine social competence or social skills that can be expected to have positive effects on the job, practitioners might be less worried about the effects of EI on faking personality tests in terms of their the predictive validity. Put differently, practitioners trying to safeguard personality tests from faking may be throwing the baby out with the bathwater; by doing so, valid trait variance related to self-presentational skills they might actually seek in employees will be removed. This is not improbable, given that trait EI has been shown to be related to job performance (Côté & Miners, 2006; Joseph et al., 2015; Joseph & Newman, 2010). For example, in a large meta-analysis, O’Boyle et al. (2011) showed that trait EI has criterion validity for job performance of around .26. Moreover, trait EI measures showed an incremental validity of 5.2% (p < .05) above and beyond intelligence and personality. Obviously, this study provides only one piece of the puzzle: One would need information on both performance in selection procedures and on the job to arrive at more firm conclusions about this dual effect of EI. Promising evidence comes from research that has shown that the ability to identify selection criteria, a concept similar to EI, is positively related to both performance in selection procedures and on the job (Kleinmann et al., 2011).

The findings of the present study are in line with impression management (J. Hogan et al., 2007) or self-presentational (R. Hogan, 1982; Marcus, 2009) theories of faking, which state that people are always concerned with the impression they make on others in social interactions, and in this regard responding to a personality test in selection procedures is no exception. Imperative here is the argument that individuals with higher levels of EI—indicative of one’s general social effectiveness—would engage and be more effective in impression management not only during a selection procedure but also in their everyday (social) life. This could be the reason why high-EI (vs. low) individuals are rated as being more prosocial by others (Petrides, 2011).

Finally, our study has implications for the faking literature at large. Scores on one and the same facet were raised in one scenario, whereas lowered in the other, in line with the job profiles (Raymark et al., 2004). This implies that providing a higher score on a facet does not need to increase one’s chances of getting the job; whether this will be the case will depend on the job for which one is applying. In fact, elevating a score on one and the same facet could increase one’s chances of getting hired for one job but decrease them for the other. In addition, increasing scores on nonrelevant traits would not increase someone’s chances of getting the job at all. Fake good (e.g., Dunnett, Koun, & Barber, 1981; Mersman & Shultz, 1998) and fake-as-the-best-applicant-for-a-“desired job” (e.g., McFarland & Ryan, 2000) cannot account for these effects: Such studies consider “higher scores” to be “better,” but the present study showed that this is not necessarily the case. Therefore, we agree with Raymark and colleagues (Raymark et al., 2004; Raymark & Tafero, 2009) that in faking studies a priori hypotheses on which facet (or factor) scores will be altered and in which particular directions are essential in correctly estimating the effects of faking in selection procedures.

Limitations and future studies

When interpreting the results, the lowered reliabilities of the facets under the faking conditions warrant some consideration. They stand in contrast to reliabilities reported in fake good studies, in which consistently choosing for the most socially desirable response option generally results in increases in internal consistencies (e.g., Douglas et al., 1996). In the current study, apparently, participants scrutinized each individual item as to whether it would increase or decrease their chances of getting hired and answered accordingly. Investigation of the items of facets with low reliabilities in the faking scenarios confirms this idea. For example, for the gregariousness facet in the lawyer condition, the item “Likes large parties” was solely responsible for the low reliability value. Of interest, the mean score on this item was lower in the faking condition compared to the honest condition, whereas mean scores on the remaining four items were higher. Clearly, participants considered a preference for large parties (perhaps indicative of irresponsibility or lack of self-control) as undesirable for getting the job while endorsing the other items (e.g., “Starts conversations”) as desirable. This indeed suggests that participants very carefully selected their response to each individual item (see Donovan et al., 2014, for similar results).

Further, given our student sample, the potentially limited generalizability of the results should be taken into account. Students may lack the relevant knowledge needed when applying for a job, and are perhaps less motivated to do well in the faking scenario compared to when a real job is at stake (Mersman & Shultz, 1998). On the other hand, we presented all participants with the same job information and provided a monetary incentive, which can be assumed to at least partly deal with the aforementioned limitations. Also in relation to generalizability of findings, some have argued that instructed faking studies are limited in terms of validity because they reflect maximum rather than actual faking behavior (Smith & Ellingson, 2002). Although this might be true, the aim of the present study first and foremost was to confirm that EI is related to the ability to fake in the first place. Nevertheless, additional research with samples from the working population in real-life settings is needed to confirm the present findings.

The present study has exclusively focused on the influence of EI on the ability to fake. Models of faking behavior include other aspects such as motivational (e.g., the need for the job) and situational factors (e.g., possible verification of test results) as well (e.g., Goffin & Boyd, 2009; McFarland & Ryan, 2006; Roulin et al., 2016). The fact that individuals with higher levels of trait EI are better able to fake does not necessarily mean that, in real life, they would also have an increased tendency to do so. In this sense, it would be interesting to investigate how EI and motivational and situational factors interplay in the prediction of faking behavior.

On the predictor side, the present study focused on trait EI; alternatively, future studies could investigate how faking ability relates to other social effectiveness constructs, for example, the ability to identify the criteria in selection procedures (Kleinmann et al., 2011). Furthermore, although we demonstrated incremental validity of EI over GMA and the Big Five with respect to faking ability, there are other traits against which incremental validity could be tested (e.g., core-self evaluations; Joseph et al., 2015; Kluemper, 2008). Finally, given the theoretical debate on how EI should be conceptualized, the relation between faking ability and trait EI vis-à-vis ability EI should be tested empirically.

Conclusion

The present study provides new insights into the factors associated with faking. Specifically, it showed that EI positively relates to the ability to fake on personality tests across two distinctive jobs. The findings from the present study may provide some reassurance for those who tend to be pessimistic on the use of personality tests: Yes, personality tests can be faked, but it might be that successfully doing so requires the same skills and knowledge that also lead to a genuinely better performance on the job. This does not necessarily mean that faking will never pose a problem. The notion that people can fake all the more calls for an adequate assessment of people’s traits and skills, as well as the specific conditions influencing whether people actually do fake.

Acknowledgments

This research was supported by Ixly, a test-publishing company from Utrecht, the Netherlands. The authors have no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Notes

¹ We thank an anonymous reviewer for this suggestion.

² Technically, independent variables should be scale centered prior to the analyses (Edwards, 1994; Shanock, Baran, Gentry, Pattison, & Heggestad, 2010); however, in our case this resulted in considerable problems of multicollinearity. Mean centering of the variables solved this issue (Aiken & West, 1991) without altering the substantive interpretations of the response surface.

³ Analyses including GMA were performed on a subsample (79%) of the total sample (see the Measures section). The relation between EI and the overall ability to fake scores controlling for their respective opportunity to fake scores did not change when the subsample was used instead of the total sample. Furthermore, the excluded participants did not differ from the subsample in terms of their opportunity adjusted overall faking ability scores—file clerk: t(127) = –.48, p = .63; lawyer: t(127) = 1.28, p = .20—and level of TEI, t(127) = –.47, p = .64.

⁴ To further address the multicollinearity issues, we also ran models including the Big Five traits but excluding the opportunity to fake scores from the analyses: Results were identical.