Attitude to ambiguity as a predictor of analytic thinking

Abstract

Thinking dispositions are considered important predictors of analytic thinking. While several thinking dispositions have been found to predict responses on a range of analytic thinking tasks, this field is arguably underdeveloped. There are likely many relevant dispositional variables associated with analytic thinking that remains to be explored. This study examines one such dispositional variable, namely, attitude to ambiguity. The disposition is implied in the literature given that internal conflict – likely with associated ambiguity – is typically experienced in cognitive tasks used to study thinking and reasoning. In this article, the association between attitude to ambiguity and analytic thinking is empirically examined using Bayesian methods. A total of 313 adults (mean age = 29.31, SD = 12.19) completed the Multidimensional Attitude Toward Ambiguity (MAAS) scale, along with the Cognitive Reflection Test and a syllogism-based measure of belief bias. Results found one component of the MAAS scale, Moral Absolutism, to be a robust predictor of scores on both the Cognitive Reflection Test and the measure of belief bias.

Keywords

Analytic thinking attitude to ambiguity belief bias Cognitive Reflection Test thinking disposition

Decades of research in judgement and decision-making has shown that miserly information processing is a prominent characteristic of human cognition (Evans, 2008; Kahneman, 2011; Stanovich, 2009, 2011) ‘Miserly information processing’ (hereafter ‘miserly thinking’) is a commonly used term to refer, for example, to instances where a person fails to engage analytic thinking in situations where it is required (i.e., Toplak et al., 2014). The literature shows how this feature of mental processing can culminate in poor performance on a diverse range of analytic thinking tasks. Not only has this effect been established in research labs (e.g., Baron, 2008; Evans, 2007; Johnson-Laird, 2006), there is also ample evidence of its impact in real-world settings (e.g., Kahneman & Tversky, 2000; Stanovich, 2011; Thaler & Sunstein, 2008).

Despite ongoing research regarding different cognitive architectures (Stanovich, 2018), dual process theory of cognition has arguably provided the most comprehensive theoretical framework to date with which to understand this feature of human thinking – failing to engage analytic thinking in relevant situations (De Neys, 2018; Evans & Stanovich, 2013). In essence, dual process theory posits two qualitatively distinct types of cognitive processing (Evans & Stanovich, 2013), commonly referred to as Type I and Type II processing. Accordingly, some features of Type I include fast, autonomous, non-conscious, and parallel processing that is relatively independent of working memory. Typical features of Type II include slow, serial, associative, and conscious processing that is dependent on working memory. Thinking dispositions represent a special component of Type II processing (Stanovich, 2011; Stanovich et al., 2016).

An early error in dual process theory was the idea that Type I processes mostly resulted in incorrect responses on analytic reasoning tasks, while Type II was assumed to be the path to accurate responding via analytic thinking, or Type II processing (Evans, 2018). However, the dynamic between Type I and II processes, such as how and when Type I and II thinking culminates in correct responses, is complex (De Neys, 2018). For example, overlearned associations can become Type I responses, which will lead to correct responses in activities measuring such cognitive tasks. However, the focus in this article is on cognitive tasks where an incorrect Type I response needs to be inhibited, and Type II analytic thinking engaged to produce a correct response. As an example, consider the following item: ‘A bat and a ball cost $1.10. The bat costs one dollar more that the ball. How much does the ball cost?’ The intuitive answer is 10c, although it can’t be because if the bat costs $1 more, it’s price is $1.10, plus the 10c of the ball, equals $1.20. The right answer therefore has to be 5c. These and many other tasks are commonly employed in the judgement and decision-making and thinking and reasoning literatures to evoke intuitive or heuristic responses, where they have to be inhibited and analytic thinking engaged to avoid making an error (Stanovich, 2011).

Despite remarkable progress in recent years, dual process theory, the dominant framework in which these psychological characteristics are examined, has faced significant challenges regarding the key mechanism(s) driving analytic thought. In fact, Pennycook (2018) considers the fact that ‘we don’t yet know what makes us think’ as the most salient challenge currently facing dual process theory (p. 2).

Stanovich and his collaborators have long argued that thinking dispositions are an underappreciated determinant in this process, and they have provided much evidence in support of this view (Macpherson & Stanovich, 2007; Stanovich & West, 1998; Toplak et al., 2011). Examples of thinking dispositions shown to be predictive of analytic thought include; Need for Cognition (Cacioppo et al., 1996), Actively Open-minded Thinking (Baron, 1985, 1993; Stanovich & West, 1998), Need for Closure (Webster & Kruglanski, 1994), and the Master Rationality Motive (Stanovich et al., 2016).

What is clear, however, from Stanovich and his collaborators’ work is that good thinking, as measured on analytic thinking tasks, requires both the ability and willingness to do so (Pennycook, 2018). Yet the dispositional structure underlying this willingness to engage analytic thought is not yet well understood. Despite the fact that several dispositional variables have been linked to performance on analytic reasoning tasks, much work remains to develop a comprehensive understanding of this general thinking disposition that seems to foreclose analytic thought in settings where it is required.

There are likely many as yet unexplored dispositional variables related to analytic thinking. In addition to identifying them, we further need to examine the degree to which these dispositional variables are related to different thinking tasks. It is unlikely that they will be equally associated – positively or negatively – to all thinking tasks. The dispositional structure underlying analytic thought is further likely to be nuanced, as the cognitive tasks are themselves complex and not pure measures of analytic thinking (Stanovich, 2018). Dispositional variables are therefore likely to be differentially related to different thinking tasks. Against this background, the present study seeks to explore an as yet unexplored dispositional variable conjectured to predict the propensity towards analytic thinking. While the trait is not new to psychology, its role as a dispositional variable potentially predictive of analytic thought remains unexamined.

Attitude to ambiguity

While implied in previous research, but not yet investigated explicitly, is how individuals tolerate ambiguity, and how this relates to analytic thinking. For instance, the fact that conflict detection is a key component of the thinking and reasoning literature suggests that ambiguity is an inherent element of the situation that needs resolving. For example, De Neys and colleagues have provided ample evidence of unconscious conflict detection in many classic judgement and heuristic tasks (De Neys, 2014; De Neys et al., 2013; De Neys & Glumicic, 2008). Moreover, recent research has shown that individuals who experience conscious conflict between rational knowledge and intuitive feelings frequently acquiesce to the latter while having information that the former is in fact correct. Thus, an explicit choice is made to override rational knowledge known to be superior in a given context, in favour of a powerful, but incorrect intuition (Walco & Risen, 2017). An important question, therefore, is why people manage such internal conflict so badly? One possibility not previously considered is whether dispositions related to the management of ambiguity might be involved with information processing. Specifically, whether individual differences in the attitude to ambiguity is associated with the degree to which people engage in analytic thought.

A salient individual difference variable in this regard is tolerance to ambiguity (Frenkel-Brunswik, 1948), which has been extensively researched over the last 70 years in several domains including clinical, organizational, personality, and cross-cultural psychology (for a review, see Furnham & Marks, 2013). Attitudes to ambiguity have been found to be related to a wide array of behaviours (Furnham & Marks, 2013). Some of these findings imply a mechanism through which a dispositional approach to ambiguity might impact the way individuals respond to the conflict experienced in typical tasks studied in the judgement and decision-making literature. For instance, intolerance to ambiguity has been found to be related to anxiety (Bardi et al., 2009) and worry (Buhr & Dugas, 2006; Dugas et al., 1997). This association with worry was found with a slightly different construct, intolerance to uncertainty. However, it shares considerable conceptual overlap with intolerance with ambiguity, and there is little evidence that they are substantively different (Grenier et al., 2005).

Since ambiguity is part of life and cannot be completely avoided, there are arguably better and worse ways to manage it. A non-optimal strategy might be to minimize exposure to sources of ambiguity, or the period of time one is exposed to it, if it cannot be avoided entirely. Theoretically, such an attitude to ambiguity is likely to influence judgement and decision-making negatively, especially if one’s dispositional default is to shy away from ambiguity that needs further processing to resolve it effectively.

In this study, the aim is to empirically test the hypothesis that an individual’s attitude to ambiguity is a predictor of analytic thinking outcomes. The expectation is that individuals averse to ambiguity will engage in less analytic information processing. Specifically, that individuals scoring higher on the Multidimensional Attitude toward Ambiguity Scale (MAAS) will score lower on the Cognitive Reflection Test (CRT) and on a syllogistic-based measure of belief bias. Both the CRT and belief bias measures are commonly used reasoning tasks with which to index failures of analytic thinking, as both require overriding an intuitive response and little prior knowledge to get to the correct response (Stanovich et al., 2016).

Method

Participants

A total of 313 adults participated in the study, ranging between 18 and 77 years of age (M = 29.31 years, SD = 12.19 years), with women comprising 64.5% (n = 202) and men 35.5% (n = 111) of the sample. Levels of education were represented as follows: Grade 12 = 24.37%; Certificate = 8.86%, Diploma = 10.44%, = B-degree = 33.86%, Honours degree = 16.14%, Master’s degree = 3.80%, Doctorate degree = 0.63%, Unspecified = 1.90%.

Instruments

The Multidimensional Attitude toward Ambiguity Scale

This is a multidimensional scale comprising three facets or subscales; Need for Complexity and Novelty, Discomfort with Ambiguity and Moral Absolutism (Lauriola et al., 2016). The scale is the result of factor analytic efforts with seven measures to comprehensively operationalize attitude towards ambiguity. Each facet consists of seven items. Participants responded on a 5-point Likert-type scale (‘Strongly Disagree’ to ‘Strongly Agree’) rather than the original 7-point Likert-type scale, which could affect measurement; however, good reliability was observed for all scales in this study as reported in the ‘Results’ section. Lauriola et al. (2016) offer the following descriptions of the facet scales:

Discomfort with ambiguity

This facet scale measures unpleasant experiences of ambiguity in interpersonal relationships, social and job settings. Example items include: ‘If I am uncertain about the responsibilities of a job, I get very anxious’; ‘I get pretty anxious when I’m in a social situation involving me which I have little control of’. Higher scores reflect more discomfort with ambiguity.

Need for complexity and novelty

This measure reflects an approach-orientation to new and complex situations. An example item is ‘I’m drawn to situations which can be interpreted in more than one way’. Higher scores reflect a higher need for complexity and novelty.

Moral absolutism

This facet scale measures aspects of thinking characterized by a rigid black-and-white approach to life, including having difficulty with the idea that the same thing can have positive and negative characteristics. An example item is ‘There’s a right way and a wrong way to do almost everything’. Higher scores reflect a higher degree of rigid or absolutist thinking.

Cognitive Reflection Test

The well-known three item CRT (Frederick, 2005) is considered a measure of analytic thinking (Baron et al., 2015; Toplak et al., 2011). An example item is: A bat and a ball cost $1.10. The bat costs one dollar more that the ball. How much does the ball cost? Higher scores indicate more cognitive reflection.

Belief bias syllogisms

Participants completed seven items adapted from Markovits and Nantel’s (1989) article, containing a mix of items in which the validity of the argument and its believability conflict, or not. Such problems reflect the critical thinking skill of setting aside one’s prior knowledge to reason effectively with present content (Toplak et al., 2014). Included were logically valid items with unbelievable conclusions (two), logically invalid items with believable conclusions (three), and items where believability and logical validity were congruent (two). An example item that is logically valid but has an unbelievable conclusion is (1) ‘All things that contain sugar is good for your health, (2) Chocolate contains sugar (3) Chocolate is good for your health’. Higher scores reflect less belief bias.

Procedure

Students taking a course in research methodology were invited to participate in the study. To improve generalizability of results, students were encouraged to invite peers and family members older than 18 years to participate in the study as well. The time frame for this was 2 weeks. Individuals with mental health/handicap challenges were not eligible to participate. No incentives were offered for participation. The protocol was a non-timed, self-administered booklet containing information about the study, eligibility criteria, contact details of the principle investigator should participants have any questions, and information explaining rights (i.e., respondents take part voluntarily and can withdraw anytime, that informed consent is required to take part in the study, and that no identifying data is collected to ensure anonymity).

Ethical considerations

The research was approved by the University of Johannesburg’s department of psychology, who provided ethical approval to conduct the study in accordance with university and national guidelines for research ethics.

Data analysis

This study made use of Bayesian rather than conventional (so-called frequentist) statistics, given its inferential advantages. For example, stronger inferences can be made regarding evidence for the null and alternative hypotheses using Bayes factors (Wagenmakers, Marsman, et al., 2018). Bayes factors are interpreted following Wagenmakers, Love, et al. (2018). Bayes factors larger than three support the alternative hypothesis (moderate evidence), while Bayes factors larger than 10 are considered strong evidence. Less than one supports the null hypothesis and between one and three is considered anecdotal with no clear evidence either way. For regression analysis, full Bayesian estimation provided point estimates of effect size along with comprehensive indicators of uncertainty (Nicenboim & Vasishth, 2016). This allowed evaluation of the magnitude of effects, and not just whether the association is different from zero. Bayesian credible intervals allow direct interpretations about all plausible parameter values, in contrast to more abstract frequentist confidence intervals (Morey et al., 2016). In addition, multiple models can be tested without having to correct for multiple testing and finally, the expected ability of the models to predict in out-of-sample data can be evaluated using Bayesian information criteria (Vehtari et al., 2017). Specifically, the widely applicable information criterion (WAIC; also known as Watanabe-Akaike information criterion), and leave-one-out cross validation (LOO-CV) (Vehtari et al., 2017). Both provide an indication of a model’s expected predictive accuracy in new data. In other words, it provides an indication of a model’s expected replicability on out-of-sample data, with lower values indicating better models.

McDonald’s (1999) omega reliability estimates were computed for each measure. This method takes a latent variable approach to reliability estimation and is therefore based on the actual structure of the test (Revelle & Condon, 2019). This stands in contrast to other well-known, but more problematic indicators of reliability such as Cronbach’s alpha (Revelle & Condon, 2019). McDonald’s omega (total) provides an estimate of the total reliability present on a measure. Missing values ranged between 1.6% and 3.5%. Multiple imputation was used to impute missing values using the mice (van Buuren & Groothuis-Oudshoorn, 2011) package in R.

Results

Table 1 presents descriptive statistics and bivariate Pearson correlations among the variables of the study. For inferential purposes, both frequentist significance tests and Bayes factors are reported, although only the latter can be used to draw conclusions for and against the null and alternative hypothesis (Wagenmakers, Marsman, et al., 2018; van Zyl, 2018). Bayesian credibility intervals (95%) are also provided for each correlation. There was good convergence between the frequentist and Bayesian analysis, as all statistically significant correlations also had strong evidence against the null hypothesis.

Table 1.

Descriptives, reliability estimates and zero-order correlations among the variables of the study.

Measure	MA	DA	NFCN	CRT	BB
Moral absolutism (MA)
Pearson’s r	–
BF10	–
Upper 95% CI	–
Lower 95% CI	–
Discomfort with ambiguity (DA)
Pearson’s r	0.468***
BF10	9.191e +14
Upper 95% CI	0.550
Lower 95% CI	0.373
Need for complexity and novelty (NFCN)
Pearson’s r	0.128	0.185*
BF10	0.829	11.800
Upper 95% CI	0.237	0.291
Lower 95% CI	0.015	0.072
Cognitive Reflection Test (CRT)
Pearson’s r	−0.293***	−0.130	−0.027
BF10	55498.520	0.924	0.080
Upper 95% CI	−0.186	−0.018	0.086
Lower 95% CI	−0.391	−0.238	−0.139
Belief-bias syllogisms (BB)
Pearson’s r	−0.361***	−0.089	0.004	0.360***
BF10	97840000	0.233	0.073	8.997e + 7
Upper 95% CI	−0.258	0.024	0.117	0.453
Lower 95% CI	−0.453	−0.199	−0.109	0.257
M	2.93	3.37	3.23	0.71	3.74
SD	0.90	0.68	0.66	1.04	1.55
McDonald’s omega	.87	.76	.81	.75	.76

MA: moral absolutism; DA: discomfort with ambiguity; NFCN: need for complexity and novelty; CRT: Cognitive Reflection Test; BB: belief-bias syllogisms; BF₁₀: Bayes factor in favour of the alternative hypothesis; CI: credible interval; SD: standard deviation. *p < .05; ***p < .001

McDonald’s omega (total) estimates are reported at the bottom of Table 1 for each variable in the study. The reliability estimates were satisfactory with coefficients ranging between .75 and .87.

To examine whether the three facets of the MAAS scale are predictive of analytic thinking, they were included as predictors in a Bayesian regression analysis: first as predictors of the CRT, and second, predicting scores on a measure of Belief Bias. For the prediction of both CRT and Belief Bias, several models were computed, including a full model with all three facet scales, an intercept only model, and models containing one, two, and three predictors, including models in which the order of entry varies. Table 2 details the content and order of each model computed.

Table 2.

Models predicting the CRT and belief bias.

Outcome: CRT			Outcome: belief bias
Model	Predictors	Bayes R²	Model	Predictors	Bayes R²
Model 1.1	DA + NFCN + MA	.0962	Model 2.1	DA + NFCN + MA	.1423
Model 1.2	intercept only	.000	Model 2.2	intercept only	.000
Model 1.3	MA	.0935	Model 2.3	MA	.1293
Model 1.4	MA + DA	.0956	Model 2.4	MA + DA	.1392
Model 1.5	MA + NFCN	.0956	Model 2.5	MA + NFCN	.1341
Model 1.6	DA	.0237	Model 2.6	DA	.0103
Model 1.7	DA + MA	.0957	Model 2.7	DA + MA	.1394
Model 1.8	DA + NFCN	.0265	Model 2.8	DA + NFCN	.0137
Model 1.9	NFCN	.0036	Model 2.9	NFCN	.0032
Model 1.10	NFCN + DA	.0267	Model 2.10	NFCN + DA	.0138
Model 1.11	NFCN + MA	.0960	Model 2.11	NFCN + MA	.1338

CRT: Cognitive Reflection Test; DA: discomfort with ambiguity; NFCN: need for complexity and novelty; MA: moral absolutism.

Weakly informative priors were used in each model (Gelman, 2018). A normal(0, 10) prior was used for the intercept and normal(0, 1) prior was used for beta parameters. No previous research of this type could be found as a basis for constructing informative priors. The models were estimated using the brms wrapper package (Bürkner, 2017) for conducting Bayesian analysis using Stan (Stan Development Team, 2018) within the R statistical programming language (R Core Team, 2014). For each model, the posterior distribution was estimated with four chains that included 1000 warmup samples and 3000 iterations. These results are presented graphically in Figure 1. The WAIC and LOO-CV estimates provide almost identical results of model performance, supporting confidence in the out-of-sample accuracy of the models. Inspection of the results show that all models that excluded Moral Absolutism performed much weaker. This was true for the prediction of both the CRT (WAIC and LOO ≈ 920) and Belief Bias (WAIC and LOO ≈1175), noting again that smaller values indicate better fit. Models that included Discomfort with Ambiguity and the Need for Complexity and Novelty did not perform much better than the intercept model. This was again true for both the CRT and Belief Bias. The results suggest that Moral Absolutism is the only meaningful predictor on both analytic tasks.

Figure 1.

Information criteria for model comparison.

This view corresponds with the Bayesian R-squared results presented in Table 2. Models containing all three facets accounted for 9.6% of the variance on the CRT (Model 1.1), and 14.2% of Belief Bias (Model 2.1), which only decreased to 9.4% (Model 1.3) and 12.9% (Model 2.3) respectively, when dropping Discomfort with Ambiguity and the Need for Complexity and Novelty.

We can examine the posterior distributions and 95% credible intervals of models containing all three predictors (Model 1.1 and Model 2.1) in Figure 2. The median point estimates and exact credible interval values for the graphs are reported in Table 3, along with relevant chain convergence values. While the Rhat chain convergence indicator was above the expected value of one for each beta coefficient, this is to be expected when working with multiple imputed data, as parameter values are computed on several slightly different datasets (Bürkner, 2018). Combined, the results presented in Figure 2 and Table 3 further support Moral Absolutism as the only meaningful predictor of both analytic tasks. It shows that when predicting the CRT, null, and values close to null fell towards the middle of the 95% credible interval for Discomfort with Ambiguity (−0.18, 0.19) and Need for Complexity and Novelty (−0.15, 0.19), suggesting these are values most compatible with the data. For Belief Bias, small effects were most compatible for these two facet scales but there remains considerable uncertainty surrounding their parameter estimates. Indeed, their credible intervals suggest that negative, positive, as well as null values were all compatible with the data (95% CI for DA = −0.05, 0.48; and 95% CI for NFCN = −0.17, 0.33). In contrast, none of the plausible parameter values for Moral Absolutism included null under the specified models, for either the CRT (−0.49, −0.21) or Belief Bias (−0.90, −0.48).

Figure 2.

Posterior parameter distributions with 95% credible intervals.

Table 3.

Regression coefficients for the prediction of the CRT and belief bias.

Model	Estimate	SE	95%lb CI	95%ub CI	Rhat
CRT (Model 1.1)
DA	0.01	0.10	−0.18	0.19	1.01
NFCN	0.02	0.09	−0.15	0.19	1.03
MA	−0.35	0.07	−0.49	−0.21	1.01
Bayes R² = .0962
Belief bias (Model 2.1)
DA	0.21	0.14	−0.05	0.48	1.01
NFCN	0.08	0.13	− 0.17	0.33	1.01
MA	−0.69	0.11	− 0.90	− 0.48	1.04
Bayes R² = .1423

CRT: Cognitive Reflection Test; SE: standard error; 95%lb CI: lower bound of the 95% credible interval; 95%up CI: upper bound of the 95% credible interval; DA: Discomfort with Ambiguity; NFCN: Need for Complexity and Novelty; MA: Moral Absolutism.

Discussion

The purpose of this study was to examine the possibility that an individual’s dispositional attitude towards ambiguity predicts performance on analytic thinking tasks. The results are interesting seeing as the hypothesis, considered broadly, was partially supported, given that two of the three facet scales, Need for Complexity and Novelty, and Discomfort with Ambiguity were not predictive of the reasoning tasks. There is, however, some uncertainty in these parameter estimates, although it is probably reasonable to conclude that any true population effects, should they exist, are likely to be quite small. On the other hand, Moral Absolutism emerged as a meaningful predictor of both the CRT and Belief Bias. This appeared to be a particularly robust finding, as it emerged consistently in all models computed, that those without Moral Absolutism had much weaker out-of-sample accuracy, irrespective of the order of entry or whether the model included one or two covariates.

Most surprising, however, was that Moral Absolutism as the sole predictor accounted for a considerable amount of variance in both thinking tasks – 9.4% predicting the CRT, and 12.9% predicting Belief Bias, constituting large effect sizes in individual differences research (Funder & Ozer, 2019; Gignac & Szodorai, 2016). Indeed, including the other two facets to the model containing only Moral Absolutism only marginally increased the variance to 9.6% and 14.23%, respectively.

The strong effect for Moral Absolutism combined with the fact that Discomfort with Ambiguity and the Need for Complexity and Novelty were not particularly predictive of the thinking tasks, necessitate a nuanced interpretation when drawing conclusions regarding the hypothesis investigated. First, while the MAAS is a validated measure, it is not surprising that the Need for Complexity and Novelty was not predictive of the thinking tasks, as this construct is not consistent with ‘ambiguity’ in the way it was conjectured to play a role in analytic thinking, as described in the introduction. The Discomfort with Ambiguity facet was more in line with this perspective. However, results did not support this view. Only Moral Absolutism was a meaningful predictor. High scorers on this facet would typically approach life with an absolute way of dealing with people, issues, and moral questions, as opposed to a nuanced approach based on a realization that things are, in fact, more complicated than one would like to believe.

This finding suggests that a disposition to view the world in categorical terms can become a mechanism rendering one unlikely to, or ‘unwilling’ to engage the thought processes required to generate the correct response in these analytic thinking tasks. Whether this unwillingness is done consciously or non-consciously cannot be determined here and will require further research. It is considered an ‘unwillingness’ rather than inability, because the thinking tasks do not typically require high levels of cognitive ability to find the correct answers, although it has been reported that Belief Bias items are somewhat more cognitively demanding than those of the CRT (Stanovich, 2011, 2018).

It is important to note that the conjecture in this study includes a theorized pathway, flowing from the disposition to performance on the thinking tasks. Thus, that an individual’s dispositional attitude towards ambiguity influences the degree to which the individual will engage in analytic thinking. The finding of this study supports the theoretical framework of Stanovich (2009, 2011), in which thinking dispositions represent a special component of Type II processing and constitutes what he calls the reflective mind. In this framework, thinking dispositions determine the need to override an intuitive Type I response (as primed in the thinking tasks). Moral Absolutism fits well in this framework as another thinking disposition that gives shape to the reflective mind, along with others such as Actively Open-Minded Thinking and Need for Cognition. In this case, a ‘moral absolutist’ disposition makes it less likely that a strong, intuitive Type I response will be overturned, as there will be automatic steering away of perceived mental conflict and ambiguity, when faced with a decision of having to override the intuitive response and to engage an analytic thinking process. However, while the present findings support this theoretical framework, and expands it by identifying another dispositional variable that might influence analytic thinking, the causal pathway cannot be confirmed in this study. The possibility also exists that the causal pathway could be reversed, so that weak analytic thinking results in ‘moral absolutist’ ways of reasoning.

This study has limitations that should be noted. An important limitation is the issue of possible confounding given that demographic characteristics such as age, gender and educational level were not statistically controlled for in this study. Although it is quite reasonable to imagine confounding effects, there is an important reason for this methodological decision. The main point is that confounding is per definition a causal question that does not apply to associations (Hernán, 2018). This study examines the degree to which attitude towards ambiguity is predictive of scores on analytic reasoning tasks. Thus, it is an association study from which no causal inferences can be drawn. Further research is required to determine whether the association found here could be causal. This would require a causal identification strategy, making use, for example, of directed acyclic graphs (DAG; see Hernán & Robin, 2019; Pearl, 2009; Pearl & MacKenzie, 2018; see also the potential outcomes framework: Foster, 2010; Holland, 1986; Rubin, 2005). These causal frameworks are more established in fields like epidemiology and economics, but is gaining traction in psychology, as appreciation of its importance is increasing (Foster, 2010; Grosz et al., 2020; Rohrer, 2018). A critical aspect in such an approach is to determine proper covariate control, to prevent mistakenly controlling for collider and (or) mediator variables in the analysis, and thereby introducing bias (Grosz et al., 2020). While this article sought to explore predictive associations, a next step might therefore be to explicitly examine the causal nature of the associations observed in this article in a causal inference framework where the conditional independence structures among relevant variables can be examined.

Another limitation is that the sample is diverse. This tacitly assumes that the measures employed in the study function similarly across different cultural, language, and gender groups. However, measurement invariance across demographic subgroups was not examined in this study. A further limitation is the use of convenience and snowball sampling, as this may have violated the independence assumption of observations.

Finally, the findings reported should be considered exploratory rather confirmatory (Wagenmakers et al., 2012), since there was no preregistration plan detailing each step of the study in advance. Thus, the generalizability of the findings should be done cautiously given the possible influence of so-called researcher degrees of freedom (Simmons et al., 2011; Wicherts et al., 2016). Future research might be required to confirm the robustness of the present results.

Conclusion

The findings of this study suggest that one’s attitude to ambiguity, as measured by the MAAS scale, is partially predictive of analytic thinking. While two of the three facets were not predictive of the outcome variables, a particular sub-component of the larger construct, namely Moral Absolutism, had meaningful associations with both analytic thinking tasks. It predicted noteworthy amounts of variance on the Cognitive Reflection Test (CRT) and a measure of Belief Bias, and seems to hold promise as a robust predictor in out-of-sample data.

Footnotes

Declaration of conflicting interests

The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding

The author(s) received no financial support for the research, authorship, and/or publication of this article.

ORCID iD

Casper JJ van Zyl

References

Bardi

Guerra

V. M.

Sharadeh

Ramdeny

(2009). Openness and ambiguity intolerance: Their differential relations to well-being in the context of an academic life transition. Personality and Individual Differences, 47, 219–223. https://doi.org/10.1016/j.paid.2009.03.003

Baron

(1985). Rationality and intelligence. Cambridge University Press.

Baron

(1993). Why teach thinking? An essay. (Target article with commentary.). Applied Psychology: An International Review, 42, 191–237.

Baron

(2008). Thinking and deciding (4th ed.). Cambridge University Press.

Baron

Scott

Fincher

Metz

S. E.

(2015). Why does the Cognitive Reflection Test (sometimes) predict utilitarian moral judgment (and other things)? Journal of Applied Research in Memory and Cognition, 4, 265–284. https://doi.org/10.1016/j.jarmac.2014.09.003

Buhr

Dugas

M. J.

(2006). Investigating the construct validity of intolerance of uncertainty and its unique relationship with worry. Journal of Anxiety Disorders, 20, 222–236. https://doi.org/10.1016/j.janxdis.2004.12.004

Bürkner

P. C.

(2017). brms: An R package for Bayesian multilevel models using Stan. Journal of Statistical Software, 80(1), 1–28. https://doi.org/10.18637/jss.v080.i01

Bürkner

P. C.

(2018). Handle missing values with brms. https://cran.r-project.org/web/packages/brms/vignettes/brms_missings.html

Cacioppo

J. T.

Petty

R. E.

Feinstein

Jarvis

(1996). Dispositional differences in cognitive motivation: The life and times of individuals varying in need for cognition. Psychological Bulletin, 119, 197–253. https://doi.org/10.1037/0033-2909.119.2.197

10.

De Neys

. (2014). Conflict detection, dual processes, and logical intuitions: Some clarifications. Thinking & Reasoning, 20, 333–351. https://doi.org/10.1080/13546783.2013.854725

11.

De Neys

. (Eds.). (2018). Dual process theory 2.0: Current issues in thinking and reasoning. Routledge.

12.

De Neys

Glumicic

. (2008). Conflict monitoring in dual process theories of thinking. Cognition, 106, 1248–1299. https://doi.org/10.1016/j.cognition.2007.06.002

13.

De Neys

Rossi

Houde

. (2013). Bats, balls, and substitution sensitivity: Cognitive misers are no happy fools. Psychonomic Bulletin & Review, 20, 269–273. https://doi.org/10.3758/s13423-013-0384-5

14.

Dugas

M. J.

Freeston

M. H.

Ladouceur

(1997). Intolerance of uncertainty and problem orientation in worry. Cognitive Therapy and Research, 21, 593–606. https://doi.org/10.1023/A:1021890322153

15.

Evans

J. St. B. T

. (2007). Hypothetical thinking: Dual processes in reasoning and judgement. Psychology Press.

16.

Evans

J. St. B. T

. (2008). Dual process accounts of reasoning, judgement and social cognition. Annual Review of Psychology, 59, 255–278. https://doi.org/10.1146/annurev.psych.59.103006.093629

17.

Evans

J. St. B. T

. (2018). Dual process theory: Perspectives and problems. In de Neys

(Ed.), Dual process theory 2.0 (pp. 145–164). Routledge.

18.

Evans

J. St. B. T.

Stanovich

K. E.

(2013). Dual process theories of higher cognition: Advancing the debate. Perspectives on Psychological Science, 8, 223–241. https://doi.org/10.1177/1745691612460685

19.

Foster

E. M.

(2010). Causal inference and developmental psychology. Developmental Psychology, 46(6), 1454–1480. https://doi.org/10.1037/a0020204

20.

Frederick

(2005). Cognitive reflection and decision making. Journal of Economic Perspectives, 19, 25–42. https://doi.org/10.1257/089533005775196732

21.

Frenkel-Brunswik

(1948). Intolerance of ambiguity as an emotional and perceptual personality variable. Journal of Personality, 18, 108–123. https://doi.org/10.1111/j.1467-6494.1949.tb01236

22.

Funder

D. C.

Ozer

D. J.

(2019). Evaluating effect size in psychological research: Sense and nonsense. Advances in Methods and Practices in Psychological Science, 2(2), 156–168.

23.

Furnham

Marks

(2013). Tolerance of ambiguity: A review of recent literature. Psychology, 4(9), 717–728. https://doi.org/10.4236/psych.2013.49102

24.

Gelman

(2018). Prior choice recommendations. https://github.com/stan-dev/stan/wiki/Prior-Choice-Recommendations

25.

Gignac

G. E.

Szodorai

E. T.

(2016). Effect size guidelines for individual differences researchers. Personality and Individual Differences, 102, 74–78.

26.

Grenier

Barrette

A. M.

Ladouceur

(2005). Intolerance of uncertainty and intolerance of ambiguity: Similarities and differences. Personality and Individual Differences, 39(3), 593–600. https://doi.org/10.1016/j.paid.2005.02.014

27.

Grosz

Rohrer

J. M.

Thoemmes

(2020). The taboo against explicit causal inference in nonexperimental psychology. Perspectives on Psychological Science, 15(5), 1243–1255. https://doi.org/10.31234/osf.io/8hr7n

28.

Hernán

M. A.

(2018). The C-word: Scientific euphemisms do not improve causal inference from observational data. American Journal of Public Health, 108(5), 616–619.

29.

Hernán

M. A.

Robin

J. M.

(2019). Causal inference book. https://www.hsph.harvard.edu/miguel-hernan/causal-inference-book/

30.

Holland

P. W.

(1986). Statistics and causal inference. Journal of the American Statistical Association, 81, 945–960. https://doi.org/10.1080/01621459.1986.10478354

31.

Johnson-Laird

P. N.

(2006). How we reason. Oxford University Press.

32.

Kahneman

(2011). Thinking, fast and slow. Farrar, Straus & Giroux.

33.

Kahneman

Tversky

(Eds.). (2000). Choices, values and frames. Cambridge University Press.

34.

Lauriola

Foschi

Mosca

Weller

(2016). Attitude toward ambiguity: Empirically robust factors in self-report personality scales. Assessment, 23(3), 353–373. https://doi.org/10.1177/1073191115577188

35.

Macpherson

Stanovich

K. E.

(2007). Cognitive ability, thinking dispositions, and instructional set as predictors of critical thinking. Learning and Individual Differences, 17, 115–127. https://doi.org/10.1016/j.lindif.2007.05.003

36.

Markovits

Nantel

(1989). The belief-bias effect in the production and evaluation of logical conclusions. Memory & Cognition, 17, 11–17. https://doi.org/10.3758/BF03199552

37.

McDonald

R. P.

(1999). Test theory: A unified treatment. Lawrence Erlbaum.

38.

Morey

R. D.

Hoekstra

Rouder

J. N.

Lee

M. D.

Wagenmakers

E. J.

(2016). The fallacy of placing confidence in confidence intervals. Psychonomic Bulletin & Review, 23(1), 103–123.

39.

Nicenboim

Vasishth

(2016). Statistical methods for linguistic research: Foundational ideas – Part II. Language and Linguistics Compass, 10(11), 591–613.

40.

Pearl

(2009). Causality: Models, reasoning and inference (2nd ed.). Cambridge University Press.

41.

Pearl

MacKenzie

(2018). The book of why: The new science of cause and effect. Basic Books.

42.

Pennycook

(2018). A perspective on the theoretical foundation of dual process models. In de Neys

(Ed.), Dual process theory 2.0 (pp. 13–35). Routledge.

43.

R Core Team. (2014). R: A language and environment for statistical computing. R Foundation for Statistical Computing. http://www.R-project.org/

44.

Revelle

Condon

D. M.

(2019). Reliability from α to ω: A tutorial. Psychological Assessment, 31(12), 1395–1411.

45.

Rohrer

J. M.

(2018). Thinking clearly about correlations and causation: Graphical causal models for observational data. Advances in Methods and Practices in Psychological Science, 1, 27–42. https://doi.org/10.1177/2515245917745629

46.

Rubin

D. B.

(2005). Causal inference using potential outcomes: Design, modeling, decisions. Journal of the American Statistical Association, 100, 322–331. https://doi.org/10.1198/016214504000001880

47.

Simmons

J. P.

Nelson

L. D.

Simonsohn

(2011). False-positive psychology: Undisclosed flexibility in data collection and analysis allows presenting anything as significant. Psychological Science, 22(11), 1359–1366.

48.

Stan Development Team. (2018). The Stan core library (Version 2.18.0). http://mc-stan.org

49.

Stanovich

K. E.

(2009). What intelligence tests miss: The psychology of rational thought. Yale University Press.

50.

Stanovich

K. E.

(2011). Rationality and the reflective mind. Oxford University Press.

51.

Stanovich

K. E.

(2018). Miserliness in human cognition: The interaction of detection, override and mindware. Thinking & Reasoning, 24, 423–444. https://doi.org/10.1080/13546783.2018.1459314

52.

Stanovich

K. E.

West

R. F.

(1998). Individual differences in rational thought. Journal of Experimental Psychology: General, 127, 161–188. https://doi.org/10.1037/0096-3445.127.2.161

53.

Stanovich

K. E.

West

R. F.

Toplak

M. E.

(2016). The rationality quotient: Toward a test of rational thinking. Massachusetts Institute of Technology Press.

54.

Thaler

R. H.

Sunstein

C. R.

(2008). Improving decisions and health, wealth, and happiness. Yale University Press.

55.

Toplak

M. E.

West

R. F.

Stanovich

K. E.

(2011). The Cognitive Reflection Test as a predictor of performance on heuristics-and-biases tasks. Memory & Cognition, 39, 1275–1289. https://doi.org/10.3758/s13421-011-0104-1

56.

Toplak

M. E.

West

R. F.

Stanovich

K. E.

(2014). Assessing miserly information processing: An expansion of the Cognitive Reflection Test. Thinking & Reasoning, 20(2), 147–168. https://doi.org/10.1080/13546783.2013.844729

57.

van Buuren

Groothuis-Oudshoorn

. (2011). mice: Multivariate imputation by chained equations in R. Journal of Statistical Software, 45(3), 1–67. https://doi.org/10.18637/jss.v045.i03

58.

van Zyl

C. J. J

. (2018). Frequentist and Bayesian inference: A conceptual primer. New Ideas in Psychology, 51, 44–49. https://doi.org/10.1016/j.newideapsych.2018.06.004

59.

Vehtari

Gelman

Gabry

(2017). Practical Bayesian model evaluation using leave-one-out cross-validation and WAIC. Statistics and Computing, 27(5), 1413–1432. https://doi.org/10.1007/s11222-016-9696-4

60.

Wagenmakers

E. J.

Love

Marsman

Jamil

Verhagen

. . . Morey

R. D.

(2018). Bayesian inference for psychology. Part II: Example applications with JASP. Psychonomic Bulletin & Review, 25, 58–76. https://doi.org/10.3758/s13423-017-1323-7

61.

Wagenmakers

E. J.

Marsman

Jamil

Verhagen

Love

. . . Morey

R. D.

(2018). Bayesian inference for psychology. Part I: Theoretical advantages and practical ramifications. Psychonomic Bulletin & Review, 25, 35–57. https://doi.org/10.3758/s13423-017-1343-3

62.

Wagenmakers

E. J.

Wetzels

Borsboom

van der Maas

H. L. J.

Kievit

R. A.

(2012). An agenda for purely confirmatory research. Perspectives on Psychological Research, 7, 632–638. https://doi.org/10.1177/1745691612463078

63.

Walco

D. K.

Risen

J. L.

(2017). The empirical case for acquiescing to intuition. Psychological Science, 28(12), 1807–1820. https://doi.org/10.1177/0956797617723377

64.

Webster

Kruglanski

(1994). Individual differences in need for cognitive closure. Journal of Personality and Social Psychology, 67, 1049–1062. https://doi.org/10.1037/0022-3514.67.6.1049

65.

Wicherts

J. M.

Veldkamp

C. L.

Augusteijn

H. E.

Bakker

van Aert

R. C.

van Assen

M. A.

(2016). Degrees of freedom in planning, running, analyzing, and reporting psychological studies: A checklist to avoid p-hacking. Frontiers in Psychology, 7, Article 1832. https://doi.org/10.3389/fpsyg.2016.01832