Introduction

Assessment of attitudes on socially sensitive subjects using traditional self-report measures is susceptible to faking [1,2]. The Implicit Association Test (IAT) aims to reflect test participants’ automatic evaluations of an attitude object. It is a reaction time-based, computerized sorting task intended to conceal the true measurement target, and thus potentially more resistant to faking [3–5]. Research has shown that IATs are especially useful in the assessment of attitudes related to socially sensitive subjects (e.g., stereotyping or prejudice) [6]. Faking an IAT on the very first exposure appears to be almost impossible [1,2,7], but evidence is accumulating that IATs can be faked to some extent when participants are allowed to become familiar with the test format [7,8].

All previous studies have explicitly asked participants to try faking their IAT score [2,7–10]. Successful faking has been assessed using both within- and between-subject designs [2,10], and in some studies participants were given explicit instructions on how to fake the test (e.g., slowing of responses in sensitive trials and reacting as quickly as possible in others) [7,8]. In summary, extant studies indicate that successful IAT faking increases when participants are given repeated opportunities to fake the test, and that provision of an effective faking strategy is an even more effective aid to faking [2,9].

Instructing participants to fake test scores in order to assess the vulnerability of the test to faking, or to learn how to detect faking attempts, is a method which has also been used in other research areas e.g. in memory malingering [11,12]. In these studies participants are usually assigned to a faking or a non-faking condition by the experimenter. This approach secures internal control on the participants’ group membership [13]. Other studies from the same research area however suggest that participants who have been instructed to fake poor performance on memory tests tend to over-exaggerate their memory deficits [12,14].

Taking our cue from this line of research it is suggested that the same could happen when participants are asked and instructed to fake an IAT score: results from instructed faking attempts (a behavior that is ethically acceptable in this case) might differ from uninstructed, hidden faking attempts. We believe that this difference might account for conflicting data on the IAT’s vulnerability to faking [7,15] as well as for ambiguous findings with respect to the IAT’s predictive validity [5].

As indicated by other research from memory malingering, it is difficult to determine whether people who are involved in litigation are faking or not. The presence of an objective, external incentive for faking is one important criterion for classifying a claimant’s performance as fraudulent [16]. However, with regard to our study on IAT faking, an experimental manipulations should be used that provides participants with an implicit incentive to hide their true attitudes. They should be incentivized to alter their responses in order to achieve or avoid a certain consequence. Faking motivated in this way is likely to differ from faking elicited by a researcher’s request. As cheating on a test is regarded a socially condemned behavior, participants should believe that their attempt to fake has a fair chance to remain undetected.

Athletes’ attitudes to doping are subject to social pressure in a way that they are expected to be strongly anti-doping. Direct assessments with attitude questionnaires are heavily biased by socially desirable responding [17]. Attitudes to doping are therefore a suitable substrate in which to investigate the IAT’s susceptibility to faking. In the following study we tested the hypothesis that in an experimental condition in which an implicit incentive to fake was present, participants would be capable of faking an attitude on a questionnaire, the Performance Enhancement Attitude Scale (PEAS) [18], but not on a response time-based measure, the doping Brief IAT (BIAT) [19].

Method

Results

Descriptive Statistics

Descriptive statistics for the PEAS, BIAT and the impression management scale are given in Table 1. Two participants admitted to having tried faking the PEAS. All participants denied trying to fake the doping BIAT. Two-third (66.67%) of participants in the incentive-to-fake condition correctly remembered the name of the anti-doping program, indicating that most were aware of the bogus training program. An independent samples t-test showed that participants in both conditions had similar dispositions of socially desirable responding, t(63) = 0.24, p =. 81, so that this variable was not added as a covariate to the subsequent MANOVA.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 1. Descriptive statistics for attitudes and strategic responding in groups.

https://doi.org/10.1371/journal.pone.0118507.t001

Main Analysis

The MANOVA revealed a significant omnibus effect with medium effect size, F(2, 62) = 3.72, p =. 05, η² =. 10. Subsequent univariate tests showed that the PEAS scores, F(1, 63) = 6.17, p =. 02, η² =. 09, but not the doping BIAT scores, F(1, 63) = 0.57, p =. 45, η² =. 01, were significantly lower in the incentive-to-fake group than in the control group. Standardized mean scores for the PEAS and the doping BIAT are presented as a function of experimental condition (i.e. in the two groups) in Fig 1.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 1. Z-standardized scores (boxplots with data point distributions) from the response-time based (BIAT) and questionnaire (PEAS) doping attitude measures in the two experimental conditions.

https://doi.org/10.1371/journal.pone.0118507.g001

Discussion

This is the first study to investigate BIAT faking using an implicit incentive-to-fake condition instead of explicitly instructing participants to try faking the test. In our experiment half of the participants were informed that detection of an unacceptably lenient doping attitude would lead to an additional arduous subsequent task. These participants then displayed significantly more negative evaluations of doping in a doping attitude questionnaire than control participants in a no-treatment condition. The experimental manipulation did not produce similar differences in our participants’ attitudes. On the pictorial doping BIAT, participants given an implicit incentive to fake, displayed attitudes to doping similar to those of control participants without the implicit incentive to fake. These effects can be attributed to our experimental manipulation since assignment to conditions was random.

Because of the subtle manipulation we used we are unable to determine whether or not participants actually tried to fake their tests. It is possible that our participants were able but unwilling to fake the doping BIAT or that they were willing but unable. Fear of being caught could have decreased our participants’ motivation to fake their tests. But, most crucially, no matter whether participants were unable or unwilling to fake, only the response-time based attitude scores showed up to be unaffected by the experimental treatment. This pattern of results is similar to results from a study by Petróczi and colleagues [25] who found that those assumed to lie about their doping behavior (based on discrepancy between self-reported absence from doping and hair analysis) produced extreme negative test results across all questionnaire measures (PEAS, projected prevalence of doping, perceived pressure to dope, etc.) but not so much on the IAT. With regard to our study, the hypothesized and finally observed group difference on the questionnaire suggests that faking occurred. Since participants cannot be expected to truthfully answer whether or not they tried to fake any of the two tests, there is no way of determining post-hoc whether or not participants tried to alter their responses on the BIAT. It could be argued that participants were motivated to alter their responses only on the PEAS; however, research on memory malingering has demonstrated that claimants producing suspicious results on one symptom validity test also tend to show suspicious performance on other validity tests and cognitive measures [26]. Further research is needed to disentangle differential motivation-to-fake vs. capacity-to-fake effects though.

All participants were informed that both the pictorial doping BIAT (the response time based attitude test) and the PEAS (the attitude questionnaire) were accurate measures of attitudes to doping. This deception was necessary for the experimental manipulation to work although both measures are far from perfect measures and although one should not use these tests for individual diagnostics in fact. Hiding the measurement target has been regarded as one reason for the relative robustness of indirect measures against faking in previous studies [4,27]. In our study however, knowing what latent construct the BIAT test supposedly measured, did not result in successful faking. It seems that, at least under a subtle incentive to fake, it is not what the BIAT measures but how it is measured that is responsible for its robustness against faking. Implicitly incentivized faking attempts might lower the predictive validity of direct measures of attitudes related to socially sensitive subjects but, according to our data, indirect tests that aim to measure participants’ automatic evaluations are unlikely to be affected.

Most of the extant findings on IAT faking are based on ‘instructed faking’ paradigms with participants being supplied with explicit guidance on how to fake. Research on memory malingering has shown that participants who are instructed to malinger tend to perform differently from claimants who are suspected of malingering [12,14]. The same may hold true for attitude testing, especially when participants work on response time-based indirect tests. Our aim was to establish and use an implicit ‘incentive-to-fake’ paradigm making it possible to investigate the possibility of faking and faking effects more realistically. Our results suggest that (B)IATs could be less prone to faking than suggested by earlier research. The IAT’s vulnerability to faking might has been overestimated so far; provided that other researchers are able to replicate our finding.

In most previous experiments, unlike in real life settings, there was no reason for participants to fear the consequences of faking being detected. Faking in real world situations is therefore likely to be more subtle, since fakers do not want to get caught and do not necessarily regard their performance as faking in the first place. Telling participants that it was possible to detect faking would have enhanced the external validity of the paradigm; the manipulation would have provided both a situational incentive to fake and a disincentive (threat of punishment) for detectable faking then. Future studies should investigate how response patterns change when rewards and punishment for faking are manipulated in conjunction. Participants may adopt a different faking strategy when they believe that faking might be detected, but they are also less likely to make the attempt [28].

Conclusion

This study provides first evidence that BIAT faking (and presumably also IAT faking more generally) is less likely than implied by the extant literature. A situational manipulation that was sufficient to alter participants’ responses on an attitude questionnaire did not lead to alterations in scores on a reaction time-based measure of attitude to doping. In our view, this finding supports the claim that the BIAT is difficult to fake [27]. Valid, robust tests are a prerequisite of social science research directed at understanding and preventing doping. While it has to be acknowledged that a multitude of situational and personal factors affect doping behavior [30], indirect measures such as the (B)IAT hold promise in explaining incremental variance in doping behavior [25]; the more so as research in doping-related automatic cognition is still in its infancy [31]. The pictorial doping BIAT used here has already been shown to statistically significant discriminate between positive and negative biochemical doping test results [22]. It is important to note however that results from indirect tests intend to tap automatic human thinking processes and should not be seen as “doping tests”. Evidence of a response time-based indirect attitude test’s robustness in a realistic faking scenario supports its potential utility as a valid research instrument that helps to deepen our understanding of doping-related cognition and—as a consequence—to develop (even) more effective doping prevention interventions.

Supporting Information