#142798 | AsPredicted

'Personality traits (un)related to dishonest behavior'
(AsPredicted #142798)


Author(s)
This pre-registration is currently anonymous to enable blind peer-review.
It has 3 authors.
Pre-registered on
09/06/2023 03:37 AM (PT)

1) Have any data been collected for this study already?
It's complicated. We have already collected some data but explain in Question 8 why readers may consider this a valid pre-registration nevertheless.

2) What's the main question being asked or hypothesis being tested in this study?
The research question is whether (self-reports of) certain (groups of) personality traits are associated with behavioral dishonesty.

Hypotheses:
1 a-d: Big Five Agreeableness as measured by the (a) NEO-FFI, (b) BFAS, (c) BFI-2, and (d) IPIP-50 is not negatively associated with dishonesty
2 a-d: HEXACO Honesty-Humility will account for incremental variance in dishonesty beyond Big Five Agreeableness as measured by the (a) NEO-FFI, (b) BFAS, (c) BFI-2, and (d) IPIP-50

3 a-e: The dark traits (a) Machiavellianism, (b) Narcissism (as per the SD3), (c) Narcissism (as per the NARQ-short), (d) Psychopathy, and (e) Sadism are each (separately) positively associated with dishonesty
4 a-e: The dark traits (a) Machiavellianism, (b) Narcissism (as per the SD3), (c) Narcissism (as per the NARQ-short), (d) Psychopathy, and (e) Sadism do not account for incremental variance in dishonesty beyond the Dark Factor of Personality

5 a-c: The impression management (aka other-deception or lie) scales from the (a) BIDR, (b) EPQ lie-scale, and (c) MMPI are not positively associated with dishonesty

6 a&b: The measures of trait self-control (low impulsivity) from the (a) Brief Self-Control Scale and (b) BIS-15 are not negatively associated with dishonesty

3) Describe the key dependent variable(s) specifying how they will be measured.
The key dependent variable is the (latent) probability of dishonesty across time and context as measured by two incentivized probabilistic cheating paradigms, the mind game and the coin-toss task (administered in temporal separation from each other and from measured personality traits)

'Independent' variables are (manifest) scores on the following self-report scales (order as per the hypotheses specified above):
- Big Five Agreeableness: NEO-FFI (Costa & McCrae, 1992); BFAS (DeYoung et al., 2007); BFI-2 (Soto & John, 2017); IPIP-50 (Goldberg, 1992); additionally (see the exploratory analyses in 8): IPIP-NEO (Maples et al., 2015); AG+ (Crowe et al., 2018)
- Machiavellianism: SD3 (Jones & Paulhus, 2014)
- Narcissism: SD3 (Jones & Paulhus, 2014); NARQ–short (Leckelt et al., 2018)
- Psychopathy: SD3 (Jones & Paulhus, 2014)
- Sadism: SSIS (O'Meara et al., 2011)
- Impression management: BIDR (Paulhus, 1991); EPQ lie-scale (Eyesenck & Eyesenck, 1991); MMPI (Hathaway & McKinley, 2000)
- Self-control: Brief Self-Control Scale (Tangney et al., 2004); BIS-15 (Spinella, 2007)

Control variables (cf. Hypotheses 2 and 4):
- HEXACO Honesty-Humility: HEXACO-60 (Ashton & Lee, 2009)
- Dark Factor of Personality: D70 (Moshagen et al., 2020)

4) How many and which conditions will participants be assigned to?
In the coin-toss task, participants are randomly assigned to two different incentive types (money vs. avoiding work); however, this manipulation bears no relevance for the present hypotheses, which are correlational in nature. Thus, we are not interested in the effect of the manipulation. Still, we will include condition as a control factor in our regression analyses to account for possible mean-level differences in the level of dishonesty.

5) Specify exactly which analyses you will conduct to examine the main question/hypothesis.
To test our hypotheses, we will fit modified, hierarchical, logistic regression models to predict whether participants give a "yes"-response in each of the two cheating paradigms (see Heck et al., 2018). The model accounts for the fact that a certain proportion of respondents (i.e., p = 1/8) reports "yes" honestly.
All models will include random intercepts for participants as well as fixed effects, namely, (1) an effect-coded factor for the two cheating paradigms (mind game vs. coin-toss task) and (2) an effect-coded factor for the two types of incentive in the coin-toss task (money vs. avoiding work).
Hypotheses 1, 3, 5, and 6 will be tested by including a third fixed-effects covariate in the model, namely, (3) the trait of interest (using standardized scores). We will fit a separate regression model for each trait/scale.
Hypotheses 2 and 4 are tested by including a fourth predictor, namely, (4) the control variable Honesty-Humility or the Dark Factor (also standardized), respectively.
The Bayesian hierarchical model is fitted in Stan using MCMC sampling. The stochastic association of a trait with dishonest behavior is operationalized by the odds ratio (OR), with OR = 1 denoting the absence of an association (i.e., the null hypothesis H0). We will assess whether our hypotheses are supported by checking whether the corresponding 95% credible interval (CI) includes the value OR = 1.
Prior distributions are slightly adapted compared to Heck et al. (2018) to ensure that the prior-predictive distribution of the model (i.e., predictions for new, simulated data) is more plausible. Specifically, we will assume a logistic distribution (with scale = 1) for the intercept "mu" at the logit scale, a multivariate normal distribution (with scale = 0.5) for the standardized regression coefficients, and a half-normal distribution (with scale = 1) for the standard deviation of the person random-intercepts.
In order to find evidence for the null hypothesis, it is important that the sample size is sufficiently large to result in narrow CIs. To ensure that this is the case, and to test our analysis strategy in general, we performed preliminary analyses using only two covariates (the zero-order effects of which are well-established and have been shown for the present data set; thus, they are of no interest for the present investigation): (a) Honesty-Humility, which is known to have a strong negative association with dishonesty, and (b) a simulated covariate with a true association of OR = 1 resembling the null hypothesis. The results showed that even with our smallest sample size (N = 1,567), the expected width of the 95% CIs of the OR is sufficiently small to indicate when there is no association.

6) Describe exactly how outliers will be defined and handled, and your precise rule(s) for excluding observations.
Data stem from the Prosocial Personality Project (PPP), a large-scale project measuring various aspects of personality and related (social) behaviors (see https://osf.io/m2abp/) in multiple waves. Exclusion criteria were pre-defined and the same for all waves considered here; they included failing instructed attention checks, extremely low variance in item responses on the trait scales (only for measures containing at least 25% reverse-keyed items), low response times for the trait scales, inconsistencies in demographic variables across waves, and multiple participation in one and the same wave. Detailed information on sampling, exclusion criteria, payment, etc. is available on the OSF (see https://osf.io/m2abp/).

7) How many observations will be collected or what will determine sample size?
No need to justify decision, but be precise about exactly how the number will be determined.

Data have already been collected as part of the Prosocial Personality Project (PPP, see https://osf.io/m2abp/) and comprise 1,567 ≤ N ≤ 1,916 participants with complete data (per trait and both dependent measures, see 3).

8) Anything else you would like to pre-register?
(e.g., secondary analyses, variables collected for exploratory purposes, unusual analyses planned?)

Whereas the data have already been collected (see above), this Registration is completed prior to analysis of the data for the current purpose (i.e., the current hypotheses). To further guard against HARKing, the co-author responsible for data analysis has not had access to a dataset comprising both the independent and dependent variables at the time of pre-registration. For the preliminary analyses, only the scores for Honesty-Humility and the Dark Factor as well as the responses in the two cheating paradigms were available.

We will explore whether Hypotheses 1 and 2 also hold for Big 5 Agreeableness as measured by the IPIP-NEO (Maples et al., 2015) and AG+ (Crowe et al., 2018). The reasons why we consider these analyses exploratory and did not include corresponding hypotheses with confirmatory tests are
- that (unlike for other Big 5 Agreeableness scales) there is little to no prior data to derive any confident hypotheses
- that these scales overlap more with HEXACO Honesty-Humility and the Dark Factor than other Big 5 Agreeableness scales and may thus actually predict dishonesty
- that (only) these scales were assessed in later follow-up waves of the PPP (after the cheating tasks), thus involving notably smaller sample sizes (N < 500) which may simply not allow for confident conclusions in case of relatively small effects.

Version of AsPredicted Questions: 2.00