Abstract
Researchers have long assumed that complex thinking is determined by both situational factors and stable, trait-based differences. However, although situational influences on complexity have been discussed at length in the literature, there is still no comprehensive integration of evidence regarding the theorized trait component of cognitive complexity. To fill this gap, we evaluate the degree that cognitive complexity is attributable to trait variance. Specifically, we review two domains of evidence pertaining to (a) the generalizability of individuals’ complex thinking across domains and the temporal stability of individuals’ complex thinking and (b) the relationship of complex thinking with conceptually related traits. Cumulatively, the literature suggests that persons’ cognitive complexity at any point in time results partially from a stable and generalizable trait component that accounts for a small-to-moderate amount of variance. It further suggests that cognitively complex persons are characterized by chronic trait-based differences in motivation and ability to think complexly.
The complexity of one’s thoughts—more generally called cognitive complexity—affects an impressive range of outcomes. Cognitive complexity has been linked to outbreaks of war and violence (Suedfeld & Bluck, 1988; Suedfeld et al., 1977; Suedfeld & Jhangiani, 2009; Tetlock, 1985), success in obtaining and keeping political power (Conway et al., 2012; Suedfeld & Rank, 1976; Tetlock, 1981; Thoemmes & Conway, 2007), terrorism (Boyd-MacMillan, 2016; Conway & Conway, 2011; Conway et al., 2011; S. C. Houck et al., 2017; Smith et al., 2008), economics (e.g., Abe, 2011), and health (Conway et al., 2017; Davidson et al., 2007; Suedfeld et al., 1998).
Given that cognitive complexity is linked to such important outcomes, it is vital to more fully understand the origins of cognitive complexity so as to better predict, explain, and potentially intervene in these phenomena. What makes people think complex (vs. simple) thoughts to begin with? In this article, we provide the first comprehensive review of evidence pertaining to one specific root cause of cognitive complexity: differences in trait levels of cognitive complexity that transcend specific situations and time frames. As we will see, despite the fact that much cognitive complexity theory has been built on the assumption that there is a trait component of cognitive complexity (indeed, the construct was first defined as a personality trait), no systematic effort has been undertaken to summarize and interpret the available empirical evidence. In fact, quite the opposite is true: Most discussions of the construct focus on empirical evidence pertinent only to situational influences on complex thinking (for discussions, see Conway et al., 2011; Conway & Woodard, 2019).
To evaluate the potential of a trait component of cognitive complexity, we first define cognitive complexity both conceptually and operationally. Then, guided by the prior personality literature, we discuss how cognitive complexity might be conceptualized as a trait. We subsequently elaborate on the criteria we used for inclusion in our review and proceed to discuss two large-scale categories of evidence relevant to the question of trait cognitive complexity: (a) evidence pertaining to the generalizability and stability of objective behavioral measurements of cognitive complexity and (b) evidence pertaining to cognitive complexity’s empirical relationship with other stable personality traits. Finally, we discuss the limitations of this review, implications for personality theorists, and the future directions it suggests for subsequent research.
The Definition and Measurement of Cognitive Complexity
Cognitive complexity theorists and researchers have provided multiple different conceptual and operational definitions of the cognitive complexity construct (for a review, see Conway et al., 2014). Nonetheless, almost all cognitive complexity theorists and researchers agree that differentiation is a fundamental aspect of cognitive complexity (Conway et al., 2014; C. C. Houck et al., 2014). Differentiation refers to the number of separate concepts evoked by a person on a particular topic. Higher levels of differentiation mean that the person views a particular topic as having more dimensions, and this multidimensional thinking is one of the most pervasive hallmarks of the complex thinker. Across essentially all operations of the construct, a person who identifies a greater number of dimensions is construed as more cognitively complex than a person who identifies fewer dimensions (see Conway et al., 2014, for a review).
In addition, some operationalizations (e.g., integrative complexity; Békés & Suedfeld, 2019; Suedfeld, 2010) have added a second aspect of cognitive complexity: integration. Integration entails drawing hierarchical connections among multiple differentiated dimensions. In these conceptualizations, differentiation is necessary for integration: One cannot integrate multiple dimensions that are not recognized as distinct. Thus, persons who see multiple dimensions are more complex than those that do not; and persons who see that these distinct dimensions overlap are more complex than those who see the dimensions only (but do not recognize their overlap).
Importantly, these operationalizations of cognitive complexity are not interested in the content of the dimensions, but the differentiation (and, in some cases, integration) of the dimensions. That is, it is the structure of thought that produces cognitive complexity, not the content. So, two thoughts of identical structure but differing content (e.g., “I like candy” and “I like politics”) have the same amount of cognitive complexity, whereas two thoughts of similar content but different levels of differentiation or integration (e.g., “I like politics” and “I like politics not only because it makes me think critically, but also because it engages me in my community”) would have different levels of cognitive complexity. See the appendix for a more detailed description of the history of cognitive complexity and its various operationalizations.
Although many self-report measurements of cognitive traits exist (e.g., Need for Cognition; Cacioppo & Petty, 1982) and have validity, our review focuses on direct objective measurements of complex output as the primary conceptual variables. Thus, the primary dependent measures in our review evaluate the complexity of real output (e.g., written or spoken language, quantitative tests of multidimensional thinking) from participants and are not self-report measurements of people’s view of their own complexity-relevant traits. Our reasons for this decision include the following: Although a large percentage of research in personality has historically focused on convenience self-report samples, it is still vital to include more “objective” measurements related to personality (for discussion, see, for example, Rauthmann et al., 2016). These objective measurements focus on observer ratings of situations or traits (e.g., Rauthmann et al., 2016) or on observable or quantifiable behaviors (e.g., Mishra, Barclay, & Sparks, 2017; Mishra, Lalumière, & Williams, 2017; Sherman et al., 2015). In this study, we use objective measurements of complexity that similarly focus either on the scoring of open-ended material by trained observers/coders or on objective summaries of participants’ outcomes for specific complexity tasks. In the case of complexity, it may be doubly important to focus on objective measurements. Research reveals, for example, that people are far from perfect at evaluating both their own and other people’s complexity levels (Conway, Houck, et al., 2016; Suedfeld et al., 1996), and more broadly, individuals have biases that sometimes cause them to overestimate the complexity of valued others (Conway, Houck, et al., 2016). This prior work offers many reasons to think that the path is potentially quite rocky between someone reporting that they, for example, prefer complexity and that same person being chronically likely to produce complex output across time and across domains.
Setting aside the issue of objectivity, we ultimately cannot judge the likelihood that a specific state is the result of a trait without evaluating the generalizability and stability of the specific state itself—that is, evaluating measures of the behavioral output. If we want to judge the degree that “risk-taking” is trait-like, then it is imperative that we include risk-taking behaviors into the equation (see, for example, Mishra, 2014). Similarly, one of the primary reasons that researchers argued for a “contemptuousness” personality trait is that measurable behaviors (e.g., eye-rolling, sarcasm) show rank-order stability over time (Schriber et al., 2017). Furthermore, it is the objective behavioral output and not subjective self-report that is the more direct measurement of the individual’s actual level of cognitive complexity at a given moment (see S. C. Houck & Conway, 2019). 1 For these reasons, it is vital to first evaluate the degree that objective measurements of complexity show the properties we would expect of a trait.
Although self-report measures should not serve as the primary conceptual dependent measure for the present review, they are of course still meaningful. Indeed, as we discuss later in this article, many such self-report measures have been validated as representing traits in their own right (e.g., Need for Cognition). Furthermore, we would expect that these validated self-report measurements of traits would show at least some relationship to cognitive complexity output (a criterion discussed more below), even though they are not direct measurements of cognitive complexity itself. To that end, we provide a review of the literature on that topic in this article.
Trait-Based Component of Cognitive Complexity
Although the definition of trait varies from researcher to researcher, personality traits (sometimes described as units of personality, John & Gosling, 2000) are generally conceptualized as psychological constructs that consistently influence behavior, cognitions, or emotions across time and different situations. Many theorists posit that traits are universal (e.g., Allport, 1929; Lucas et al., 2000; McCrae, 2001; McCrae et al., 2004; McCrae & Terracciano, 2005; see also Church & Katigbak, 2017; Paunonen et al., 2003), such that anyone can be measured and placed on a spectrum ranging from high to low levels of that trait. For example, all individuals have differing levels of trait agreeableness, and an individual with high trait agreeableness tends to act in an agreeable manner across many different time points and situations. The manifestations of a trait in any given instance are most typically referred to as states. States are influenced by both their underlying trait and by the situation (or, more specifically, the psychological aspects of the situation). In fact, an individual is capable of producing most levels of any trait given the right circumstances (Fleeson, 2017). Thus, states are not always perfectly consistent with their underlying trait. Instead, states form a density distribution wherein situational features produce variability around the mean, which is an approximation of the underlying trait (Fleeson, 2017; Mussel, 2013; Rauthmann et al., 2014; Rauthmann & Sherman, 2018; Sherman et al., 2015).
Applied to our current case, this means that whereas cognitive complexity refers to the qualities of produced cognitions at any measured point in time, trait cognitive complexity refers to the theoretical trait component that contributes to produced levels of cognitive complexity occurring across specific measured instances. Following prior researchers (e.g., Rauthmann et al., 2019), we refer to the behavioral outcome measurements at any point in time as state cognitive complexity. These states, taken in aggregate, ought to reflect trait cognitive complexity (Fleeson, 2012, 2017; Mussel, 2013; Rauthmann et al., 2014; Rauthmann & Sherman, 2018; Sherman et al., 2015). In the words of Rauthmann et al. (2019), it means that “states are momentary instantiations of traits and form distributions within persons” (p. 596).
Importantly, however, the degree that the aggregated states in question do in fact result from an underlying trait is contingent on the assumption that there is nomological validity across state and trait measurements, wherein the various associations one would expect to occur are present (see Rauthmann et al., 2019). 2 It is vital, then, to empirically establish the degree that these expected associations are present with respect to cognitive complexity. If these validity associations are present, that provides important evidence of underlying trait influence.
With this in mind, our review follows the examples of researchers in other trait domains (Fleeson, 2012, 2017; Mussel, 2013; Rauthmann et al., 2014, 2019; Schriber et al., 2017; Sherman et al., 2015) using state cognitive complexity as evidence to evaluate the degree that there is a trait component to cognitive complexity. To the degree that cognitive complexity forms a coherent trait-like construct, these repeated aggregate instantiations of state cognitive complexity ought to converge (see Rauthmann et al., 2019). In theory, the more instantiations of cognitive complexity that are stable or generalizable in the aggregate, the more variance could be reasonably attributed to an underlying personality trait or traits (Rauthmann et al., 2019). Relatedly, the more those aspects of the state-trait nomological net overlap, the more likely an underlying personality trait is responsible (for discussion, see Rauthmann et al., 2019).
Is Cognitive Complexity a Likely Candidate for Having a Trait Component?
Is there any reason to expect that cognitive complexity will exhibit properties of a trait? The cognitive complexity literature tells a rather incongruent story on the subject. On one hand, from its earliest inception, cognitive complexity theory suggested that complex thinking is a product of both the situational context and trait-based differences (see Suedfeld, 2009). Indeed, at least two early theoretical perspectives on cognitive complexity—“Interactive Complexity Theory” (Schroder, 1971; Streufert, 1969, 1970, 1972; Streufert & Driver, 1967) and “Systems Theory” (e.g., Harvey et al., 1961)—relied on the assumption that there is a trait component of cognitive complexity (see Streufert & Streufert, 1978). To capture this assumption, Suedfeld (2010) categorized cognitive complexity into trait cognitive complexity (which he called “conceptual complexity”) and state cognitive complexity (which he called “integrative complexity”). This distinction was entirely theoretical—both conceptual and integrative complexity are measured using the same scoring system. In addition, cognitive traits related to cognitive complexity such as Need for Cognition, intellectual engagement, and openness have already been empirically established (Mussel, 2010, 2013), which suggests that complexity might likewise have a trait component.
On the other hand, however, surprisingly little conceptual work has directly evaluated (and no review to our knowledge has comprehensively evaluated) the presumed trait component of cognitive complexity. Indeed, whereas the earliest instantiations of cognitive complexity tended to focus more on its potential as a trait (e.g., Bieri et al., 1966; Schroder, 1971; Streufert & Driver, 1967), the majority of work in the prevailing years has tended instead to focus on situational or contextual influences (for summaries, see, for example, Conway et al., 2011; Conway & Woodard, 2019). Far from examining trait components, some of this work has instead explicitly demonstrated that individual difference variables such as political ideology are subject to domain-specificity with regard to cognitive complexity outcomes (Conway, Gornick, et al., 2016).
Thus, there is a need to return to the roots of the construct and evaluate, based on what evidence we have, the degree to which cognitive complexity is the result of trait differences. However, such a return must be guided by empirical evidence. Baumert and colleagues (2017) recently stated as follows: Personality structures, defined as patterns of covariation in behaviour, including thoughts and feelings, are results of those processes in transaction with situational affordances and regularities. It cannot be presupposed that processes are organized in ways that directly correspond to the observed structure. (p. 503)
Due to the nature of the available evidence for cognitive complexity (discussed in more depth in the following section), there is not enough data currently to provide a comprehensive theory to illuminate the various sets of connections that prior researchers have argued are necessary for a fully integrative personality theory (see, for example, Baumert et al., 2017). However, our review is a necessary and vital starting point for such an integrative endeavor. Specifically, we contribute to the extant literature by evaluating the degree that specific measurements of cognitive complexity at given points in time (“state cognitive complexity”) can be categorized as resulting from a more generalized trait (“trait cognitive complexity”). In the same way that researchers used evidence of state-based behavioral measures of contempt to argue contempt was a trait (Schriber et al., 2017) or that researchers used evidence of state-based risky behavior to argue risk-taking was a trait (Mishra, Barclay, & Sparks, 2017; Mishra, Lalumière, & Williams, 2017), we here evaluate the degree that state-based measurements of cognitive complexity can be used as evidence for trait cognitive complexity.
Criteria for Establishing Trait Components
As previously mentioned, the prevailing opinion has been that, although the sociocultural context has enormous influences on complex thinking, state levels of cognitive complexity are also partially determined by one’s trait cognitive complexity (for discussion, see, for example, Conway & Woodard, 2019; Repke et al., 2018; Suedfeld, 2009, 2010). To evaluate whether or not this assumption is justified, we must first identify the specific criteria necessary for establishing a personality trait. In this section, we outline standard criteria for determining the existence of a trait component of any construct and apply it to our current case.
Previously established criteria for establishing trait components can be broken down at a large level into two main categories for our current case: (a) generalizability and stability of trait cognitive complexity across contexts and times 3 and (b) convergent validity with related traits. These are considered by many to be essential criteria of a personality trait (Bogaert et al., 2008; Caspi et al., 2005; Cronbach & Meehl, 1955; Duckitt & Sibley, 2007; Ekehammar et al., 2004; Johnson et al., 2011; Paunonen & Ashton, 2001; Roberts et al., 2007; Steyer et al., 1992, 2015; Trzesniewski et al., 2001; Van Lange et al., 1997; Wilkowski & Robinson, 2010).
Broadly speaking, the most important of these two sets of criteria is the former: direct measurements of the construct’s generalizability across contexts and stability over time (see Chung et al., 2014; Donnellan et al., 2012; Mishra, Barclay, & Sparks, 2017; Mishra, Lalumière, & Williams, 2017; Sherman et al., 2015; Steyer et al., 1992, 2015; Trzesniewski et al., 2001). If there is a trait component of cognitive complexity, individual differences in cognitive complexity should be relatively consistent across time and differing situations. Someone who is relatively high in trait complexity ought to be high in state complexity (compared with the same sample of persons) on multiple topics and across multiple times. Furthermore, the establishment of a trait component is strengthened by the second set of criteria: establishing its expected relationships with other constructs (Mishra, 2014; Mishra & Lalumière, 2009; Mishra, Lalumière, & Williams, 2017; Schriber et al., 2017; Zuckerman, 2007). If there is indeed a trait component of cognitive complexity, this trait should be related to other stable traits in a consistent manner, and we should be able to describe what kind of person the cognitively complex individual is with theoretical coherence.
Below, each corresponding section will thus evaluate cognitive complexity on (a) its generalizability across domains and stability across time and (b) its relationship with other, conceptually related traits. Although the aim of this review is to investigate whether or not there is a meaningful trait component of cognitive complexity, it is important to note that it does not outline a comprehensive theory of trait cognitive complexity. A comprehensive theory of a trait would address the development of the trait and change across the lifespan (Baumert et al., 2017), detail the situational variables that influence state manifestations of the trait (Rauthmann, 2015; Rauthmann et al., 2014), explain the social and cognitive mechanisms underlying the trait and state manifestations of the trait (Fleeson & Jayawickreme, 2015), and explicitly detail how the trait can be operationalized and measured (Ziegler, 2014). The latter three of these four topics have been at least partially evaluated in the existing cognitive complexity literature. However, as we outline in the section “Limitations and Future Directions,” our review highlights that much more research is necessary to comprehensively investigate each of these questions. Our review instead provides a vital starting point for the construction of any trait theory: establishing the degree that it is reasonable to consider the construct a trait.
Criteria for Inclusion in This Review
To cast as wide a net as possible, all three authors performed independent searches of major databases to find papers that might meet our criteria. Databases searched in this process were PsycInfo, APA PsycInfo, APA PsycArticles, APA PsycBooks, Google Scholar, Ebook Central, ProQuest Central Psychology Database, and ProQuest Dissertations and Theses Global. Furthermore, databases of some specific journals that were especially likely to include major articles were searched (e.g., JPSP, PSPB, Journal of Personality, Political Psychology), as were the Google Scholar research works of major research laboratories with ties to complexity research (e.g., Peter Suedfeld, Philip Tetlock). Database search terms included cognitive complexity, integrative complexity, conceptual complexity, complexity trait, complexity stability, cognitive complexity personality trait, stability over time, temporal stability, rank-order stability, time, longitudinal, context(s), generalizability, stability across contexts, domain(s), and trait(s). In cases where appropriate, we used citation trees to pursue additional citations in existing relevant reviews of the literature in any papers.
Evaluation of papers for inclusion in our review followed from our discussion so far. We included papers that had objective measurements of state cognitive complexity (i.e., measures of cognitive complexity output and not merely self-report). In addition, we evaluated whether the evidence presented in the articles matched the traditional criteria for determining the degree a construct has a trait component. In particular, we searched for articles with evidence pertaining to s, temporal stability, and validity relationships with other trait constructs.
To provide a common metric for comparison across studies, we used estimation procedures to convert effect sizes to the most commonly used metric in our review (Pearson’s r) whenever possible. The specifics of each estimation procedure are detailed in the tables. In the text, we denote the findings that were originally reported with a different metric which we converted to r by referring to them as “estimated r.” 4 In the tables, we further provide separate weighted averages for each section that (a) include both our estimated rs and effect size rs as reported from the original papers and (b) include only effect sizes reported from the original papers (and exclude all converted estimates). As seen there, the resulting weighted averages are generally the same, and thus the narrative below is essentially uninfluenced by these conversion procedures.
Criteria Set 1: Generalizability and Stability of Cognitive Complexity
As discussed above, the most important set of criteria for determining the existence of a trait are a construct’s generalizability across contexts and stability over time (see Chung et al., 2014; Donnellan et al., 2012; Mishra, Barclay, & Sparks, 2017; Mishra, Lalumière, & Williams, 2017; Sherman et al., 2015; Steyer et al., 1992, 2015; Trzesniewski et al., 2001). Even though an individual fluctuates in their state manifestations of a trait, individuals should still consistently differ from each other in their average levels of cognitive complexity across contexts and time. Below, we pursue evidence related to the generalizability across contexts and stability over time of individuals’ cognitive complexity.
Across-Domain Generalizability of Cognitive Complexity
First, we integrate evidence regarding the correlations of individuals’ state cognitive complexity across different domains. This across-domain generalizability is widely used to establish the degree that a particular dimension is trait-like (see, for example, Mishra, Lalumière, & Williams, 2017). For example, researchers have noted that the fact that individuals engage in forms of risk-taking across multiple domains (such as dangerous driving, property crime, and problem gambling) is a key piece of evidence that there is a trait component of risk-taking (see Mishra, 2014). In the same way that other literatures have used the across-domain generalizability of behavioral states (e.g., the decision to gamble) to evaluate trait components, we here use the across-domain generalizability of cognitive complexity to evaluate the amount of complexity that is attributable to a general trait.
Table 1 summarizes the literature presented here on the domain generalizability of cognitive complexity. As Table 1 shows, the evidence generally suggests that a small-to-moderate amount of variance of cognitive complexity is stable across domains (across-domain sample-weighted average r = .20), with higher correlations across domains that are more similar and lower correlations across domains that are less similar. For example, McDaniel and Lawrence (1990) assessed cognitive complexity (operationalized as integrative complexity and scored following guidelines by Schroder et al., 1967) by coding participants’ responses to open-ended questions about a historical event (the Holocaust) and a current issue (problems associated with nuclear waste created during nuclear weapon production). The complexity of participants’ responses on each of these topics was correlated at r(58) = .54, p < .05. In addition, scores from the nuclear waste exercise were highly correlated, r(16) = .65, p < .01, with complexity scores from another exercise in which participants read evidence pertaining to two governmental agencies blamed for the unpreparedness of the United States during the Pearl Harbor attack and then, assuming the role of congresspersons, wrote summaries about which agency was really to blame. Although all are from the same general domain (history and politics), these correlations suggest that measures of cognitive complexity, at the very least, need not be mere artifacts of knowledge of one specific event.
Across-Domain and Time Effects of Cognitive Complexity.
Note. Summarizes the studies presented on the domain generalizability and temporal stability of individuals’ cognitive complexity. Multiple correlations from the same sample were first averaged before being added to the sample-weighted average so as not to overrepresent those samples. Multivariate analyses and studies that do not report either the effect size or sample size are not included in the sample weighted average effect size.
Hand-scored using guidelines described by Schroder et al. (1967). bHand-scored using guidelines updated by Baker-Brown et al. (1992, 1986). cScored using the AutoIC system following guidelines updated by Baker-Brown et al. (1992), Conway et al. (2014), and C. C. Houck et al. (2014). dSignificance level not provided. eMeasured by the Rep test (Bieri et al., 1966; Kelly, 1955). fScored using the automated Linguistic Inquiry and Word Count (LIWC) system (Pennebaker, 2016). gMetric used in the original paper was Cronbach’s alpha. We converted α to r using the formula for computing α from r in Gliem and Gliem (2003). hMetric used in the original paper was ANOVA F value. Converting F to r is fraught with challenges (see, for example, Hullett & Levine, 2003), and here we opted for a simple consistent conversion applied to every case equally—a conversion that assumed equal n across a two-groups, one-way analyses of variance (ANOVAs; adapted from Lipsey & Wilson, 2001), while using the level of analysis to compute the n that was used in the original computation. Although the assumptions of this conversion are violated in most instances in the table, this provides a rough common metric for comparison purposes. However, we acknowledge that this is not a precise conversion method, and thus we also computed summary scores without this metric and found similar results (see Table Summary Averages). iScored following guidelines by Hermann (1983, 1987, 2003a, 2003b, 2003c). jConverted from Cohen’s d. kOriginally reported as an intraclass correlation coefficient in Conway and Woodard (2019); for this analysis, we recomputed this as F and then transformed to r for consistency with other similar cases in the table. lMetric used in the original paper was standardized beta. Because beta and r are identical for simple linear regression, we here estimate r via a simple one-to-one conversion. Furthermore, Pratt et al. (1990) found that the cognitive complexity (also operationalized as integrative complexity and scored following guidelines by Schroder et al., 1967), of middle-age and older individuals’ descriptions of the self, a relationship held by the participant, and relationships in general were positively correlated with each other. Complexity of descriptions of the self and descriptions of a specific relationship were correlated at r(61) = .39, p < .05. Complexity of descriptions of the self and relationships in general were correlated at r(62) = .29, p < .05. Finally, complexity of descriptions of a specific relationship and relationships in general were correlated at r(61) = .48, p < .01. Again, though these three topics were from the same general domain (interpersonal issues), they differed in their specific topic, suggesting at least some degree of domain generalizability. mScored by the Categorical Dynamic Index (Pennebaker et al., 2014).
p ≤ .05. **p ≤ .01. ***p ≤ .001.
Further evidence for domain generalizability in cognitive complexity comes from a study that used several different samples and measurements to correlate cognitive complexity with a wide array of topic domains (Conway & Woodard, 2019). Across all three samples, individual cognitive complexity between the different topic domains was significantly and positively correlated. One sample of participants completed opinion stems regarding either sociopolitical issues (covering a variety of domains within this topic) or a broader set of topics including physical appearance and games. Their completed opinion statements were then hand-coded for integrative complexity (i.e., the amount of both differentiation and integration of separate ideas) following guidelines instantiated by Schroder et al. (1967) and most recently updated by Baker-Brown et al. (1992). Individuals’ integrative complexity scores for the different domains were correlated at r(414) = .17, p < .001. The second sample of participants completed two opinion stems on the current Democratic party leadership and the current Republican party leadership. These completed statements were scored for integrative complexity by the Automated Integrative Complexity scoring system 5 (AutoIC; Conway et al., 2014; C. C. Houck et al., 2014 for updated validity evidence for AutoIC, see Conway et al., 2020). Individuals’ integrative complexity scores on the two different topics were correlated at r(200) = .41, p < .001. Finally, the third sample of participants completed opinion stems regarding political and health issues, which were again scored by the AutoIC scoring system. Individuals’ integrative complexity scores on the two different topics were correlated at r(4765) = .17, p < .001. Overall, these low-to-moderate correlations across topics suggest that a small but significant amount of cognitive complexity variance across different domains is accounted for by trait cognitive complexity.
Additional evidence of generalizability across broad domains comes from research using the Paragraph Completion Test (PCT; see Schroder, 1971). This test involves responding to incomplete sentences designed to provide the potential for both unidimensional and multidimensional responding. Participants’ paragraphs are then coded for their conceptual complexity. 6 The stems for these sentences were modified to represent four conceptually distinct domains: social, nonsocial, perceptual, and executive. The intercorrelations among these different domains typically range from r = .40 to r = .60 (see Streufert & Streufert, 1978), suggesting that cognitive complexity is relatively generalizable across domains. Moreover, using the PCT, de Vries and Walker (1987) found that participants’ cognitive complexity on topics of decision-making (e.g., “When I don’t know what to do . . .”) was significantly and positively correlated with their complexity on the topic of capital punishment, r(70) = .33, p < .002.
Furthermore, Zinkhan and Biswas (1988) found that cognitive complexity assessed using the Repertory Grid Test (Rep test; see Bieri et al., 1966; Kelly, 1955) was consistent across a variety of topic domains. The Rep test (a precursor to both conceptual and integrative complexity; see the appendix for more information) instructs participants to match entities (typically people) with descriptors on a grid and scores the complexity of their completed grids. In their study, Zinkhan and Biswas found that the complexity of participants’ grids for social interactions and cameras had an average correlation of r(124) = .26, social interactions and popular literature had an average correlation of r(124) = .21, and popular literature and cameras had an average correlation of r(124) = .19. Zinkhan and Biswas concluded that there was a generalized component of cognitive complexity accounting for 4% to 16% of the variance. In additional research using the Rep test, Freeman and Barnes (1982) found that the complexity of participants’ grids for a close friend, two political figures (former U.S. presidents Jimmy Carter and Gerald Ford), a news broadcaster, and a social issue were all positively correlated, and six of the eight correlations were statistically significant at the .01 level (see Table 1 for each specific correlation). Additional research by Epting (1967) found that the complexity of grids for interpersonal relationships and social issues were correlated at r(107) = .39, p < .01. Therefore, complexity appears to be consistent across both specific and more general domains.
In closely related research, Jordan et al. (2019) analyzed the domain generalizability of U.S. Presidents’ analytical thinking across a variety of contexts, including State of the Union addresses as well as other presidential speeches, papers, and electoral debates. Analytical thinking is the degree to which one can break down a complex issue into smaller, differentiated components and has been used as a measurement of cognitive complexity. The authors found presidents’ analytic thinking across contexts to be related (estimated r = .40), suggesting consistency in cognitive complexity across many different contexts. The literature presented in this section consistently reports positive correlations of state cognitive complexity across differing domains, supporting the notion that a meaningful amount of state cognitive complexity is attributable to trait cognitive complexity. 7
Across-Time Stability in Cognitive Complexity
In this section, we evaluate evidence related to the degree that individuals’ cognitive complexity is stable over time. Across-time stability is one of the most important criteria for establishing that a particular measurement can be partially attributable to trait variance (see, for example, Chung et al., 2014; Mishra, 2014; Mishra, Lalumière, & Williams, 2017; Schriber et al., 2017; Sherman et al., 2015; Trzesniewski et al., 2001). Furthermore, according to Latent State-Trait (LST) theory, temporal stability provides a fundamental distinction between latent traits and states, with greater degrees of temporal stability indicating greater evidence of a latent trait (Steyer et al., 1992, 2015). Two main methods exist for analyzing trait stability over time. The most common method relies on correlating individuals’ state cognitive complexity across differing time points or contexts using either raw scores of cognitive complexity or rank-ordered scores of participants’ cognitive complexity. Rank-order correlations indicate the degree to which individuals who are relatively high (or low) on a particular measurement at one point in time tend to have the same relative rank (compared with others in the same sample) at another point in time (see, for example, Chung et al., 2014; Donnellan et al., 2012; Schriber et al., 2017; Sherman et al., 2015). The other main method used for analyzing trait stability over time employs Structural Equation Modeling (SEM) on longitudinal data to determine the proportion of variance in the data that can be attributed to the trait, the state (e.g., psychological features of the situation), and error (Alessandri et al., 2020; Cole, 2012; Kenny & Zautra, 1995, 2001; Kuster & Orth, 2013; Orth & Robins, 2019). The higher the variance explained by the trait, the more evidence one has that there is a meaningful trait component of the construct.
Unfortunately, SEM on trait, state, and error variance has yet to be applied to cognitive complexity data, and indeed, generally speaking, the available data would not meet the requirements for running such analyses (as outlined by Kenny & Zautra, 1995). This is undoubtedly in part because measuring cognitive complexity output has historically been a very time-consuming endeavor that requires much more investment per participant than traditional self-report or observational scales (see Conway et al., 2014, 2020). The evidence currently available for the across-time stability of cognitive complexity utilizes the first method described above: correlating individuals’ state cognitive complexity across differing time points or contexts using either raw scores of cognitive complexity or rank-ordered scores of participants’ cognitive complexity. Although less sophisticated, these methods are also very commonly used and scientifically credible (see, for example, Chung et al., 2014; Donnellan et al., 2012; Schriber et al., 2017). Indeed, correlations of state measurements across different time points have been used to establish the temporal stability of trait constructs such as self-esteem (Donnellan et al., 2012) and contempt (Schriber et al., 2017). Nonetheless, future work that utilizes SEM on longitudinal cognitive complexity data would greatly benefit the field. This is further discussed in the section “Limitations and Future Directions.”
Across-time evidence is summarized in Table 1. As can be seen there, quite a bit of accumulated evidence suggests that state measurements of cognitive complexity are stable over time (across-time sample-weighted average r = .43). One example of temporal stability in cognitive complexity comes from a study by Thoemmes and Conway (2007) on the integrative complexity (following updated guidelines by Baker-Brown et al., 1992) of 40 U.S. Presidents’ State of the Union speeches during their first 4 years in office. The authors accounted for several situational factors: the election cycle (which produces lower complexity as elections near), whether or not the nation was at war, economic crises, and whether or not the presidents’ party was in the majority in Congress. After accounting for these factors, a statistically significant effect of the person still remained, estimated r(663) = .06, p < .001. This person effect represents the stability of individual levels of integrative complexity over time (see Thoemmes & Conway, 2007; see also Wasike, 2017). This person effect was recently replicated using AutoIC on U.S. presidents’ State of the Union speeches (Conway et al., 2020), showing a similarly small-but-significant effect, estimated r(18, 848) = .02, p < .001. 8
In similar research, Song (2006) analyzed the complexity of U.S. Presidents’ State of the Union speeches, this time assessing the rank-order stability of presidents’ conceptual complexity over time. Conceptual complexity was scored following guidelines by Hermann (1983, 1987) which identifies terms and phrases that denote differentiation (see the appendix for more information). Song found that the year-to-year correlations in rank-order conceptual complexity were all positively correlated (rs ranging .32–.89), and 26 out of 28 year-to-year correlations were significant (ps ranging from <.05 to <.01), supporting the idea that there are consistent individual differences in cognitive complexity over time.
Dille and Young (2000) focused on the conceptual complexity (following guidelines by Hermann, 1983, 1987) of two presidents, Carter and Clinton, over 4 years in office. A significant effect of the individual was identified—estimated r(342) = .45, p < .001—demonstrating that the two presidents maintained rank-order temporal stability in their complexity, with Carter consistently higher than Clinton. Furthermore, the between-person differences in cognitive complexity were consistently greater than either individual’s variability over time. Additional research by Cuhadar et al. (2017) focused on the conceptual complexity (Hermann, 1983, 1987, 2003a, 2003b, 2003c) of two Turkish prime ministers, Özal and Erdoğan, from 1983 to 1993, and found the average conceptual complexity of the two leaders to be about one standard deviation different from each other (estimated r = .45), again suggesting a trait component of cognitive complexity.
Conway et al. (2012) further tracked the integrative complexity (hand-scored following updated guidelines by Baker-Brown et al., 1992) of Democratic candidates for the 2008 U.S. primaries over the course of 10 debates. Although Conway et al. do not report person-level statistical analyses, we reanalyzed the data reported in Table 1 of their paper to produce comparable across-time person metrics. These results revealed a weak positive tendency for chronic trait variance on integrative complexity, estimated r(654) = .06, p < .05. Conway and colleagues (2020) additionally performed a follow-up study on the Republican primaries in 2012 using AutoIC (Conway et al., 2014; C. C. Houck et al., 2014) and found a stronger person-level effect over the course of 20 primary debates, estimated r(129) = .26, p < .001. 9
As discussed in the previous section on the domain generalizability of cognitive complexity, Jordan et al. (2019) found that presidents’ analytic thinking was reliable across many contexts. In the same study, researchers assessed the reliability of presidents’ cognitive complexity (via their analytic thinking measurement) in State of the Union Addresses and written papers across time, and found that it was highly consistent—State of the Union Addresses estimated r(50) = .77, written papers estimated r(50) = .83. The consistency was so high that the researchers concluded that analytical thinking is likely manifesting an underlying trait in the presidents.
Research by Wallace and Suedfeld (1988) looked beyond U.S. presidents, studying world leaders’ cognitive complexity (operationalized as integrative complexity and scored following updated guidelines by Baker-Brown et al., 1986) before, during, and after seven major crises such as the American intervention in Lebanon in 1958 and the shooting down of the Korean Airlines plane by the Soviet Union in 1983. As expected, world leaders’ cognitive complexity was impacted by these crises, with most leaders’ cognitive complexity dropping during the crises and rebounding afterward. However, Wallace and Suedfeld found a significant effect of the person that predicted cognitive complexity beyond the situational effects, estimated r(84) = .45, p < .001. Even though cognitive complexity was impacted by major events, some variance in cognitive complexity was explained by the person-level of analysis. That is, some degree of cognitive complexity remained stable across time. Wallace and Suedfeld postulated that these individual differences in cognitive complexity are trait differences, perhaps linked to a personality trait such as hardiness.
In additional research, Conway et al. (2003) analyzed the integrative complexity (following updated guidelines of Baker-Brown et al., 1992) of nine Middle Eastern leaders’ nations across five time frames before, during, and after the 9/11 attacks. Again, Conway et al. did not report an effect of the individual (only a marginally significant effect of nation), but we estimated the effect of the individual using the nation-level scores from their Table 1 while only including the six nation groups that had a single individual and computing the average r across each frame-to-frame comparison (with person as the unit of analysis). Note that this method was based on a small number of data points per time frame (in some cases, less than five paragraphs per person per frame) and did not contain data for each person at every frame (meaning that some comparisons only had two persons). Including all available data yielded a very small positive average correlation (average estimated r = .02). Removing the two data points that Conway et al. (2003) identified as having less than standard available paragraphs yielded a somewhat higher average correlation (average estimated r = .15), although this increase was partially driven by one pairing that only included two individuals (without that pairing, average estimated r = .06). Overall, Conway et al.’s (2003) data provide (at best) very weak evidence for across-time stability.
However, a similar but better-powered study by Suedfeld et al. (1993) on Middle Eastern leaders found stronger evidence of across-time stability. Suedfeld et al. (1993) analyzed the integrative complexity (again scored following updated guidelines by Baker-Brown et al., 1992) of more than 800 statements made by eight Middle Eastern leaders around the time of the Gulf Crisis (1990–1991). Although the authors did not report the effect of the individual and instead focused on the effect of different events throughout the Gulf Crisis, we reanalyzed their reported complexity scores to test for the effect of the individual. There was a significant effect of the individual even when controlling for time frame and whether the leaders were aligned with or against Iraq in the Gulf Crisis, estimated r(42) = .72, p < .001. In other words, there were significant individual differences in cognitive complexity among these eight leaders throughout the varied events of the Gulf Crisis. This provides further evidence that there are stable trait differences in individual cognitive complexity.
Across-time stability in cognitive complexity has not only been observed with political figures, but with average citizens as well. One set of studies (Streufert & Streufert, 1978) found test–retest reliabilities for conceptual complexity in laboratory settings (measured on statements produced during the PCT following guidelines by Schroder et al., 1967) are typically near .85. Cognitive complexity as measured by the Rep test (Bieri et al., 1966) also demonstrates strong test–retest reliability, with complexity measurements taken 1 week apart correlating at r(174) = .54, p < .001 (Schneier, 1979).
Additional research conducted by McAdams et al. (2006) and Sengsavang et al. (2017) investigated various metrics, including cognitive complexity (operationalized as integrative complexity and scored following guidelines updated by Baker-Brown et al., 1992) of personal narratives provided by individuals during interviews over 3 and 6 years, respectively. In both of these studies, the rank-order stability of cognitive complexity (i.e., the stability of individual differences in cognitive complexity over time) was measured. In the study by McAdams and colleagues, individual differences in cognitive complexity at the first data collection session and 3 months later were significantly correlated at r(110) = .59, p < .001, and at the first session and 3 years later at r(85) = .53, p < .001 (McAdams et al., 2006). When the first two time points (the initial session and 3 months later) were combined and compared with the third time point (3 years later), the individual differences in cognitive complexity were correlated at r(74) = .60, p < .001. These correlations in complexity over time suggest that at least a portion of the variance in cognitive complexity is due to trait cognitive complexity. The authors further note that cognitive complexity was the most stable metric of participants’ personal narratives—more stable than emotional tone, agency, communion, personal growth, and number of words per sentence. In the study by Sengsavang et al. (2017), individual differences in cognitive complexity (again, operationalized as integrative complexity and scored following guidelines updated by Baker-Brown et al., 1992) at 26 years of age were significantly correlated with those 6 years later at 32 years of age r(70) = .24, p < .05, further suggesting that there is trait component of cognitive complexity.
Another set of studies (Conway & Woodard, 2019) compiled data from multiple samples and measures to investigate across-time stability in individual levels of cognitive complexity. One sample analyzed in this study consisted of university student tobacco smokers who participated in four motivational interviewing counseling sessions. The five longest statements from each participant in each counseling session were hand-coded for integrative complexity following updated guidelines by Baker-Brown et al. (1992). Across the four counseling sessions, individuals’ integrative complexity scores were positively correlated, estimated r(110) = .18, p < .001. A second sample consisted of tobacco smokers recruited via Amazon Mechanical Turk (MTurk) to participate in a 4-day texting invention for smoking cessation. Participants completed open-ended statements regarding their smoking before and after the intervention, which were then coded for integrative complexity by the AutoIC scoring system. Integrative complexity scores of these individuals’ statements before and after the intervention were correlated at r(64) = .39, p < .001. These correlations in individuals’ cognitive complexity over time provide further evidence of the temporal stability of cognitive complexity.
Additional evidence for the temporal stability of cognitive complexity comes from research by Maddux et al. (2014), who sought to investigate the role of multiculturalism and cognitive complexity (operationalized as integrative complexity and scored following guidelines updated by Baker-Brown et al., 1992) over 10 months. Despite the role of cultural engagement in increasing individuals’ integrative complexity, individuals’ cognitive complexity at the beginning of the study predicted their cognitive complexity at the end of the 10 months, r(113) = .40, p = .001, providing evidence for a trait-based component of cognitive complexity that exerts a stable influence on state levels of cognitive complexity over time.
Consistent individual differences in cognitive complexity have also been identified using multivariate analyses. Boyd and Pennebaker (2015) analyzed 55 plays by three playwrights from the late 16th to early 18th centuries (Shakespeare, Fletcher, and Theobald) for their cognitive complexity. Cognitive complexity was measured by the Categorical Dynamic Index (CDI; Pennebaker et al., 2014), which counts the number of function words that signal complexity (e.g., higher rates of nouns are positively associated with cognitive complexity). Across three different kinds of analyses—discriminant analyses, decision trees, and support vector machines—the 55 plays were sorted into distinct clusters based on their cognitive complexity: one cluster for each of the three playwrights. This suggests that the three playwrights maintained consistent individual differences in their cognitive complexity across the decades that these plays were written, lending evidence to a trait component of cognitive complexity.
Taken together, these findings suggest that a significant (albeit modest) portion of individuals’ cognitive complexity remained stable over time. However, it is important to note the limitations of this literature and their findings. Much of the available research was not originally designed to specifically test across-time stability. As a result, it does not allow for more sophisticated methods (e.g., SEM) of testing for stability. Furthermore, the temporal stability of cognitive complexity reported here (sample-weighted average r = .43) is below that of other traits, including the Big Five (Gnambs, 2014). On one hand, this is not entirely surprising and may result from a methodological difference. One might expect the temporal stability of open-ended cognitive complexity measures to be lower than that of traits measured primarily through self-report. Indeed, test–retest correlations are likely to be higher on self-report questionnaires than on behavioral measures scored by multiple coders, as the variability is naturally higher on the latter. And, in fact, stability over time does tend to be lower for implicit tests (Schultheiss et al., 2008) and motives (Goetz et al., 2010). Nonetheless, it is important to note that the temporal stability of cognitive complexity reported here is lower than that of many known self-report measures of traits, and thus we should consider this effect comparatively on the smaller side. Finally, consistent with the fact that research on cognitive complexity as a trait has not been a historical focal point, this literature—while large enough to draw general conclusions—is not overwhelming in its scope. This is further discussed in the section “Limitations and Future Directions.”
Criteria Set 2: How Cognitive Complexity Relates to Other Traits
The studies discussed thus far suggest that there is a consistent, statistically significant degree of cognitive complexity that differs from person-to-person across context and time, with trait effect sizes generally in the small to moderate range. However, none of these studies can be viewed as conclusively “demonstrating” that cognitive complexity has a trait component for several reasons. First, one can never account for all the possible situational factors that might impact complexity, so it is always theoretically possible that one simply did not account for every factor that would have explained the apparent trait variance. Second, it is possible that trait effects really represent situational constraints that were imposed over large blocks of time (e.g., Washington may exhibit higher levels of complexity than Polk overall, not because he is more chronically complex, but because he lived in an immediate post-revolution era whereas Polk did not). Given these limitations, we turn to another set of criteria to evaluate trait variance in cognitive complexity. Although no single set of criteria can conclusively demonstrate that any variable has a meaningful trait component, triangulating evidence from two different types of criteria can nonetheless provide a more compelling case.
The second set of criteria addressed in this review evaluates the relationship between cognitive complexity and other trait constructs. Traits ought to be related to other traits in a consistent way. For example, trait-based risk-taking is consistently correlated with trait-based sensation-seeking, such that levels of one can predict levels of the other (see Mishra, Barclay, & Sparks, 2017, for a review). This empirical evidence has been used to argue that risk-taking is a more generalized trait involving a “taste for risk” (Mishra, 2014; Mishra & Lalumière, 2009; Mishra, Lalumière, & Williams, 2017; Zuckerman, 2007). Indeed, Mishra (2014) noted that because the self-report personality traits in question had themselves demonstrated stability over time, the correspondence of such traits with risky behavior suggests that “personality traits may facilitate, to some degree, stable individual differences in actual risky behavior” (p. 11). Thus, in a similar vein, we pursue two interrelated questions: If there is a trait component to cognitive complexity, what other traits would we expect to be positively correlated with cognitive complexity? And is there empirical evidence that cognitive complexity is consistently associated with these other traits?
As noted by Baumert et al. (2017), different traits can be influenced by very different psychological processes. Baumert and colleagues elaborate on one example comparing anger to agreeableness: Angry and aggressive reactions in daily life appear to be shaped by accessibility of hostile thoughts, whereas the generalized tendency to behave in agreeable ways depends on individual differences in capacity and motivation to control these kinds of thoughts. More generally speaking, each personality trait could have its own information processing signature, and systematic investigations of these signatures across traits are needed. (p. 508)
As a result, it is useful to have a set of guideposts to organize our theoretical expectations of cognitive complexity as a potential trait. Based on past research, we would expect cognitive complexity to function much like research suggests about the operation of agreeableness—that it depends on individual differences in capacity and motivation (Bresin et al., 2012; Meier & Robinson, 2004; Wilkowski & Robinson, 2010; see Baumert et al., 2017, for a discussion). Thus, we expect that the cognitively complex person would have both the chronic, stable motivation and the chronic, stable ability/capacity to recognize and integrate knowledge in a multidimensional manner.
It is important to note that the “motivation” that we are referring to in this section is different from transient motivations that stem from features of the situation. Instead, we are referring to stable motivations that stem from traits, as described by McCabe and Fleeson (2016). For example, persons can be motivated to talk to others both because the situation calls for it and because they have high levels of the trait extraversion. It is the second thing—trait levels of the construct motivating behavior—that we pursue here with respect to cognitive complexity. Similarly, researchers have argued that the fact that individuals who engage in various risk-taking behaviors also tend to possess other risk-based traits is evidence for the generality of risk-taking at a broad level (Mishra, 2014; Mishra & Lalumière, 2009; Mishra, Lalumière, & Williams, 2017; Zuckerman, 2007). Indeed, this kind of evidence is especially important because, unlike research comparing (say) the same self-report measure over time, correspondence between traditional self-report personality trait measures and behavioral measures are not subject to method variance—if anything, they likely underestimate the actual relationship (Mishra, Lalumière, & Williams, 2017).
A great deal of research suggests that people differ in their stable trait motivations to epistemic drives, impulsiveness, social motives, and other traits that might be relevant to complex thinking (e.g., Cacioppo & Petty, 1982; Jost et al., 2003; Neuberg & Newsom, 1993; Webster & Kruglanski, 1994). Thus, while such stable trait motives are not themselves direct evidence that people are more likely to produce complex output (and the relationship between motivation and output can be complicated; see Conway, Gornick, et al., 2016, for a discussion), evidence of overlap between these stable motives and complex states would be important for establishing a trait component of cognitive complexity. Furthermore, research also suggests the existence of intellectual ability-relevant traits (e.g., Gottfredson, 1997; Schermer et al., 2015). As outlined below, many of these traits have direct and indirect conceptual overlap with cognitive complexity. Thus, evidence of overlap between these traits and cognitive complexity output measurements would be useful for establishing a trait component of cognitive complexity.
Our twin motivation and ability guideposts serve as a logical map for what we expect to find when reviewing the literature on cognitive complexity and personality. The degree to which these two domains are linked to cognitive complexity is converging evidence that there is such a thing as a cognitively complex person in the same way that converging evidence for multiple trait constructs predicting risk-taking was viewed as evidence for that construct having a trait component (Mishra, Barclay, & Sparks, 2017). It further helps us begin to understand, if there is a cognitively complex person, what that person is generally like.
Table 2 summarizes the evidence pertaining to cognitive complexity and traits related to stable motivation and ability. In reviewing the evidence presented below, it is important to keep in mind that correlations between measures that are administered at the same time may be driven by a third, occasion-specific variable and not by any direct relationship among the measured variables. This is further discussed in the section “Limitations and Future Directions.”
Relationships Between Cognitive Complexity and Other Traits.
Note. Summarizes the studies presented on the relationships between cognitive complexity and related individual difference constructs. Multiple correlations from the same sample were first averaged before being added to the sample-weighted average so as to not overrepresent those samples. Studies that do not report either the effect size or sample size are not included in the sample-weighted average effect size. TIPI = Ten-Item Personality Inventory. SAT = Scholastic Assessment Test.
Hand-scored using guidelines described by Schroder et al. (1967). bMetric used in the original paper was standardized beta. Because beta and r are identical for simple linear regression, we here estimate r via a simple one-to-one conversion. cHand-scored using guidelines updated by Baker-Brown et al. (1986, 1992). dMeasured using the “H” statistic sorting task (see Linville, 1982; Scott, 1969). eMeasured by the Rep test (Bieri et al., 1966; Kelly, 1955). fMetric used in the original paper was analysis of variance (ANOVA) F value. Converting F to r is fraught with challenges (see, for example, Hullett & Levine, 2003), and here we opted for a simple consistent conversion applied to every case equally—a conversion that assumed equal n across a two-groups, one-way ANOVAs (adapted from Lipsey & Wilson, 2001), while using the level of analysis to compute the n that was used in the original computation. Although the assumptions of this conversion are violated in most instances in the table, this provides a rough common metric for comparison purposes. However, we acknowledge that this is not a precise conversion method, and thus we also computed summary scores without this metric and found similar results (see Table Summary Averages). gMeasured using the Role Category Questionnaire (Crockett, 1965). hMeasured using a 10x10 matrices Rep Test (Tripodi & Bieri, 1963), modified from the original Rep test (Kelly, 1955). iScored using the AutoIC system following guidelines updated by (Baker-Brown et al., 1992; Conway et al., 2014; C. C. Houck et al., 2014). jMeasured using the Learning Environment Preferences (LEP; Moore, 1989). kMeasured using Counselor Cognitions Questionnaire (Welfare & Borders, 2006).
p ≤.05. **p ≤ .01. ***p ≤ .001.
Stable Motivation to Produce Cognitive Complexity
Is state cognitive complexity consistently related to other (and known-to-be-stable) motivational traits? Research suggests that is the case, although the relationships are generally in the low-to-moderate range. The literature discussed here demonstrates a sample-weighted average correlation between cognitive complexity and stable motivational constructs of r = .20. Specifically, cognitive complexity is related to epistemic motives, sensation-seeking, and social motives. The relationship between cognitive complexity and each of these cognitive traits is described in turn below.
Epistemic motives
Some evidence of a stable motivational component to cognitive complexity comes from work on epistemic needs. For example, some persons are chronically more motivated to think effortfully. Those high Need for Cognition persons (Cacioppo & Petty, 1982) produce more complex output as measured by integrative complexity following guidelines by Schroder et al. (1967; r = .41, p < .01; McDaniel & Lawrence, 1990). This relationship makes sense; by definition, people with a high Need for Cognition enjoy effortful thinking. An example item from the Need for Cognition Scale (Cacioppo & Petty, 1982) reads “I really enjoy a task that involves coming up with new solutions to problems.” If one chronically thinks effortfully and deeply about things in general, one is (probabilistically speaking) also more likely to produce complex multidimensional thoughts across different given domains (see Conway et al., 2008). 10
This individual difference in Need for Cognition is also indirectly observed in research on processing difficult versus easy information. One study showed that participants who were more likely to communicate difficult-to-transmit clinical information scored higher in complexity on the Rep Test (pooled r ranging from .15 to .36, p < .05; Tripodi & Bieri, 1964), whereas in another study, higher cognitive complexity (scored following guidelines by Schroder et al., 1967) of participating counseling students predicted higher quality and more clearly expressed clinical hypotheses, r(35) = .52, p < .002, and predicted asking clients more divergent questions, r(35) = .47, p < .004, than those with lower levels of cognitive complexity (Holloway & Wolleat, 1980). The clinical variables assessed in this study were taken at a different occasion than assessments of cognitive complexity, and thus these correlations are less likely to be a result of a situation-specific confounding variable. Despite very different methods of assessing cognitive complexity, both of these studies suggest that highly complex persons are more likely to exert effort in processing “difficult” or “extra” information. This dovetails with research suggesting that, of many different measured constructs, the only construct to influence all forms of measured complexity across several studies was effort (Conway et al., 2008). Specifically, the self-reported effort participants put into writing their paragraphs was positively correlated with participants’ propensity to use different forms of integrative complexity (scored using guidelines updated by Baker-Brown et al., 1992; different forms scored using the guidelines of Conway et al., 2008) of said paragraphs: Study 1a: r(421) = .13, p < .01, Study 1b: r(421) = .22, p < .01. This is further evidenced in research by McDaniel and Lawrence (1990), who report that cognitive complexity (operationalized as integrative complexity and scored following guidelines by Schroder et al., 1967) was positively correlated with a more autonomous, self-directed learning style (measured by Learner Autonomy, McDaniel & Ferreyra, 1989; r(52) = .35, p < .01).
Related research shows that some people are chronically more motivated to seek closure or provide a clear structure, as measured by the related Personal Need for Structure (PNS) scale and the Need for Closure Scale (NFCS). Persons high in Need for Structure and Closure tend to prefer a simple cognitive structure (see Neuberg & Newsom, 1993; Webster & Kruglanski, 1994 for discussions of definitional issues). Evidence for these epistemic needs is particularly important because there is strong evidence that cognitive complexity is negatively correlated with Need for Closure and Structure across different measures and across different domains. These across-domain effects suggest a trait is driving the relationship. First, the PNS scale has been correlated with complexity across four different domains using the same measurement approach, a method called the “H” statistic sorting task (see Linville, 1982; Scott, 1969). This method involves sorting 33 target words into meaningful categories, where each target word can be used in multiple piles. Neuberg and Newsom (1993) gave participants words related to one of four categories: furniture, colors, the elderly, and the self. The cards for the latter two contained the exact same list of trait words (e.g., outgoing, playful, reflective, emotional, and secure), but participants were instructed to make categories about either the elderly or themselves. The results were impressive: For all four domains, cognitive complexity was negatively correlated with PNS, with correlations ranging from r = −.48 to r = −.73 (ps ranging from <.10 to <.005). These correlations are even more impressive considering that the PNS scale had been completed by participants 4 to 6 weeks prior to the card-sorting task and that the domains cover a broad range of topics, both social and nonsocial.
In addition to the across-domain consistency of PNS while using the same measurement tactic, the Need for Structure/Closure construct also provides an example of across-domain consistency using an entirely different measurement approach. The PNS scale is highly correlated with the NFCS with Neuberg, Judice, and West (1997) reporting an average disattenuated correlation of .93 between the two scales. So, it can be safe to assume that these scales tap into virtually the same underlying construct (Kruglanski et al., 1997; Neuberg, West, et al., 1997). NFCS is negatively related to the Rep test measure of cognitive complexity, r(155) = −.30, p < .001 (Webster & Kruglanski, 1994). Notably, the Rep test was different in both domain content and measurement approach from the “H” statistic sorting tasks used by Neuberg and Newsom (1993). Although this relationship was not as strong as the counterpart PNS-complexity relationship, the fact that a similar correlation was found using an entirely different means of operationalizing complexity—and along a different content domain—suggests that the Need for Structure/Closure personality construct may be broadly related to cognitive complexity. Relatedly, in research by Tetlock et al. (1993), individuals who scored higher on cognitive complexity (this time operationalized as integrative complexity and hand-scored following guidelines update by Baker-Brown et al., 1992) also scored higher on cognitive flexibility, r(129) = .22, p < .05: a construct in direct contrast to Need for Structure and Closure. Together, the epistemic needs discussed in this section are correlated to cognitive complexity at sample-weighted r = .23, suggesting a consistent, small-to-moderate positive relationship between complexity and epistemic needs.
Sensation-seeking
In a similar vein, cognitive complexity is positively related to sensation-seeking. For example, positive relationships between sensation-seeking and complexity (operationalized as conceptual complexity, assessed using the PCT and scored using updated guidelines by Baker-Brown et al., 1992) have been observed on three of the four subscales on the Sensation Seeking Scale (Coren & Suedfeld, 1995). Those with high levels of cognitive complexity scored highly on the Thrill, estimated r(275) = .17, p < .001, Experience Seeking, estimated r(275) = .14, p < .05, and Boredom Susceptibility subscales, estimated r(275) = .15, p < .01. 11 The PCT and the Sensation Seeking Scale were administered at different time points, further substantiating their relationship. These relationships may exist because high sensation-seekers chronically search for new, complex, and exciting environments and experiences (Zuckerman, 1979, 1987), and are therefore more likely to have many different and varied cognitive representations of reality because they have experienced reality from many vantage points. Furthermore, persons high in sensation-seeking tend to prefer complex over simple stimuli, r(117) = .36; Looft & Baranowski (1971). Relatedly, individuals high in cognitive complexity (operationalized as integrative complexity and scored following updated guidelines by Baker-Brown et al., 1992) tend to score higher on creativity, r(129) = .27, p < .01 (Tetlock et al., 1993).
Further validation of this relationship comes from the conceptually overlapping Social Non-Conformity subscale of the Psychological Screening Inventory. Persons high in Social Non-Conformity tend to like excitement and nonconventional approaches to life (Lanyon, 1970, 1973), and are thus high sensation-seekers. Persons high in Social Non-Conformity also score higher on conceptual complexity (assessed using the PCT and scored following updated guidelines by Baker-Brown et al., 1992) with estimated r(275) = .15, p < .01 (Coren & Suedfeld, 1995). Importantly, the Social Non-Conformity measure and the conceptual complexity measure were not administered at the same time, but multiple times over the course of 18 weeks. This helps alleviate the possibility that the relationship between these two constructs is driven by a characteristic of the situation. In related research, those who scored higher on integrative complexity (scored using updated guidelines by Baker-Brown et al., 1992) were less likely to use social conformity to achieve their goals, r(129) = −.39, p < .001, and were rated by observers as highly original, r(129) = .24, p < .05 (Tetlock et al., 1993).
In addition, numerous studies that used at least five distinct approaches to assessing cognitive complexity have reported that persons low in complexity tend to prefer consistency whereas persons high in complexity are more comfortable with inconsistency (similar to sensation-seekers who prefer novelty) and indeed tend to be less consistent (rs ranging from .17 to .18; Crano & Schroder, 1967; Press et al., 1969; Scott, 1963; Ware & Harvey, 1967). This work is particularly impressive due to the breadth and scope of the different measures that show the same outcome (see Table 2 for a complete list of effect sizes, p values, and complexity measures).
Further evidence of a link between sensation-seeking and complexity comes from a study by Jennstal (2019) in which cognitive complexity (again operationalized as integrative complexity and hand-scored following guidelines updated by Baker-Brown et al., 1992) increased following participation in a deliberative conversation. The researcher found that individual differences in sensation-seeking-relevant traits (i.e., openness as measured by the Ten-Item Personality Inventory; Gosling et al., 2003) impacted complexity beyond the situational influence of the deliberative conversation—correlation between openness and complexity: r(61) = .29, p < .05—even though the personality inventory and complexity exercise were administered 2 weeks apart. This fits the picture of the high sensation-seeking person being higher in cognitive complexity.
Relatedly, those who are able to maintain high levels of cognitive complexity (measured as integrative complexity) in typically stress-inducing situations (e.g., interpersonal conflict) are also those who have a more positive response to stress (see Fearon & Boyd-MacMillan, 2016, for a review). These individuals do not seem to be averse to the physiological sensation of stress (just as those high in complexity are tolerant of and indeed seek many varied sensations) and instead view it as a signal of a positive and surmountable challenge. This is particularly meaningful research, as cognitive complexity has been shown to help resolve conflicts. Thus, studying those who are able to maintain high levels of cognitive complexity in stressful situations could result in more effective conflict resolution. The sample weighted average correlation between cognitive complexity and sensation-seeking traits discussed in this section is r = .23, again suggesting a consistent, small-to-moderate positive relationship.
Stable social motives
There are many theory-driven reasons to suspect that chronic social motives are related to trait complexity. Relationships are often complex, and thus people who have particularly strong motivations for social relationships (as differentiated from the motivations for social conformity discussed in the prior section) may be especially motivated toward complexity (see Thoemmes & Conway, 2007, for discussion). Furthermore, research in political psychology suggests that peaceful international outcomes are tied to increased cognitive complexity (for summaries, see Conway et al., 2001, 2018; Suedfeld et al., 2005). This is almost certainly because successful long-term relationships at any level require some amount of complex give-and-take (see Suedfeld et al., 2005, for elaboration). This implies that people who are motivated to maintain such relationships may be especially prone to complexity.
Is there evidence that social traits or motives are tied to cognitive complexity? Yes. Coren and Suedfeld (1995) report that conceptual complexity (assessed using the PCT and scored following updated guidelines by Baker-Brown et al., 1992) is related to several scales from the Interpersonal Adjective Scales (Wiggins, 1979; Wiggins & Broughton, 1991). These include a positive relationship to agreeableness, estimated r(274) = .12, p < .05, extroversion, estimated r(274) = .18, p < .001, and dominance, estimated r(274) = .11, p < .05, and a negative relation to introversion, estimated r(274) = −.16, p < .001, and submissiveness, estimated r(274) = −.11, p < .05. This same study found a positive relationship between conceptual complexity and the Expression subscale of the Psychological Screening Inventory (Lanyon, 1970, 1973), estimated r(235) = .15, p < .01. The Expression subscale is a measure of extroversion and introversion which captures both sociability and verbal dominance. These relationships are particularly noteworthy as the personality measures and conceptual complexity measure were not administered at the same time, but at different intervals over the course of 18 weeks.
Interestingly, individuals who are actively engaged in multiple cultures (e.g., immigrants who are actively engaged in both their culture of residence and their culture of origin) tend to have higher levels of cognitive complexity (as measured by integrative complexity) across differing domains. In research by Tadmor et al. (2009), the more that participants identified with two distinct cultures, the higher their cognitive complexity (measured as integrative complexity following guidelines by Baker-Brown et al., 1992) regarding both work (first-, second-, and third-generation East Asian immigrants in the United States: r = .29, p < .05 and r = .34, p = .03; first-generation Israeli immigrants in the United States: r = .31, p = .004 and r = .45, p < .001) and their culture (East Asians: r = .31, p = .04 and r = .52, p = .002; Israelis: r = .40, p < .001 and r = .62, p < .001). This may be because bicultural individuals experience a pluralism of values and a widened attention scope, which then increases cognitive complexity (Tadmor & Tetlock, 2006). Indeed, additional research demonstrates that those with higher levels of integrative complexity are adept at weighing multiple values simultaneously (Tetlock & Tyler, 1996). Finally, this increased complexity appears to mediate the relationship between bicultural individuals and higher creativity and professional success (Tadmor et al., 2012). This set of results suggests that individual differences in sociocultural motivations and experiences are predictive of state cognitive complexity.
By far the single best set of individual difference predictors for 40 U.S. presidents’ integrative complexity (scored using updated guidelines by Baker-Brown et al., 1992) were those related to social variables: affiliation motive, r(39) = .40, p < .05, extraversion, r(39) = .36, p < .05, friendliness, r(39) = .32, p < .06, and wittiness, r(39) = .42, p < .05 (Thoemmes & Conway, 2007). These effects were recently partially replicated using a larger data set scored for AutoIC for State of the Union speeches (Conway et al., 2020): Affiliation motive showed a significant relationship to integrative complexity, r(39) = .35, p < .05, although relationships for friendliness, r(39) = .12, and wittiness, r(39) = .24, were weaker and nonsignificant, and extraversion was actually mildly negatively (though nonsignificantly) related, r(39) = −.08. Wasike (2017) additionally found that U.S. presidents’ integrative complexity (scored using updated guidelines by Baker-Brown et al., 1992) was associated with the social trait charisma (see Table 2 for specifics on this association). 12 Furthermore, in the study by Jennstal (2019) referenced in the section “Sensation-seeking,” higher agreeableness (measured using the Ten-Item Personality Inventory; Gosling et al., 2003) predicted subsequently higher integrative complexity scores (using updated guidelines by Baker-Brown et al., 1992) 2 weeks later, r(61) = .56, p < .01.
There is also a positive relationship between self-esteem (often a marker of social belongingness; see Leary et al., 1998), complexity, and the number of beliefs about the self. Specifically, self-esteem is positively correlated with complexity (measured using a version of the “H” statistic sorting task that asks participants to sort traits related to themselves; Linville & Jones, 1980) at r(65) = .32, p < .01 (Campbell et al., 1991). In another study, self-esteem correlated positively with the number of positive personal qualities, r(99) = .33, p < .001, listed by participants (which is a measure of differentiation; Greenwald et al., 1988). However, it was also negatively correlated with the number of negative personal qualities, r(99) = −.33, p < .001, at an equal level (Greenwald et al., 1988).
In addition, Coren and Suedfeld (1995) found a negative relationship between conceptual complexity (assessed using the PCT and scored using updated guidelines by Baker-Brown et al., 1992) and Machiavellianism, estimated r(259) = .16, p < .001, with these measures being taken at different times. People who score high on Machiavellianism are conceptualized as lacking emotional empathy with others and thus interpersonally cold (Paulhus & Martin, 1987; see also Fehr et al., 1992). This dovetails with research by McDaniel and Lawrence (1990), who report that cognitive complexity (operationalized as integrative complexity and scored following guidelines by Schroder et al., 1967) was positively correlated with lower levels of ego involvement (i.e., evaluating one’s performance positively only if one’s performance is superior to others’ performance), r(52) = −.31, p < .05. Similarly, those with higher cognitive complexity (assessed with the Role Category Questionnaire, Crockett, 1965, which measured the amount of differentiation in participants’ description of peers) tend to be better comforters (MacGeorge & Wilkum, 2012) and discriminate more accurately between different types of supportive messaging (i.e., Low-Person Centered and High-Person Centered), r(326) = .22, p < .001, as long as they are not under high levels of emotional distress (Bodie et al., 2011). Given these findings, it is unsurprising that clinicians with higher levels of cognitive complexity (assessed using a variety of measures) tend to be more experienced and have more positive outcomes with their clients (Granello, 2010; Kindsvatter & Desmond, 2013; Magaletta & McLearen, 2015; Welfare & Borders, 2010; though see Conway et al., 2017, for a qualifying counter-example in smoking domains).
Finally, people high in cognitive complexity (operationalized as integrative complexity and scored following guidelines by Baker-Brown et al., 1992) are generally viewed more positively by their peers, r(53) = .53, p < .01, possibly because high cognitive complexity is perceived as a cue for reliable advice (Williams, 2013). However, this finding is not ubiquitous: in one study of students in a Master of Business Administration program, those with higher integrative complexity (also scored following guidelines by Baker-Brown et al., 1992) were rated by observers as more hostile, r(129) = .24, p = .009, less sympathetic, r(129) = −.29, p = .005, and more narcissistic (rs ranging .25–.27, ps < .01; Tetlock et al., 1993). This is in contrast to correlations found between integrative complexity and the participants’ self-reported personality, which painted a much more prosocial picture of those high in cognitive complexity. For example, integrative complexity was positively related to self-reported empathy, r(129) = .19, p = .04. The lack of agreement between observer and self-ratings in this one context has to be taken against a large amount of research across multiple contexts—research that demonstrates high cognitive complexity is more generally associated with prosocial traits. Overall, the literature reviewed here demonstrates a small, positive correlation between complexity and prosocial traits (sample-weighted average r = .18).
Stable motives: Summary
Although, like most enterprises, all the evidence is not totally uniform, on balance it nonetheless reveals a fairly clear picture. The relatively consistent picture of the cognitively complex person that emerges from this set of findings is one of a person who is motivated toward relationships with others, but nonetheless independent and not prone to mere conformity. They relish new experiences. This person generally appears to be extraverted, expressive, nonconforming, open to experience, charismatic, well-suited for leadership, enjoys effortful thought, and possesses requisite self-esteem. 13 Taken as a whole, the personality correlates mentioned above validate the idea that chronic motivational tendencies are related to one’s cognitive complexity. This suggests that there may be people who are chronically more likely to carry a cognitively complex approach and form cognitively complex structures across multiple domains.
Ability to Produce Cognitive Complexity
No matter how much motivation people have, individuals may differ in either their general capacity for the maintenance of complex thought over time, their upper-end complexity capacity, or both (Suedfeld, 2010). Some evidence provides support for all three of these possibilities. Certain political leaders, for example, are more immune to drops in integrative complexity during stressful situations (i.e., situations in which cognitive resources are rapidly spent; see Suedfeld, 2010, for a summary). Further evidence is explored in the following paragraphs.
Intelligence
What is the mechanism that accounts for potential individual differences in cognitive complexity ability? Perhaps the most intuitively appealing individual difference is intelligence. Highly intelligent people should find engaging in highly complex thinking easier than those lower in intelligence. In fact, general intelligence has—rightly or wrongly—been suggested to be equivalent to the ability to think complexly about things (Gottfredson, 1997, who also suggested in a review that complexity is positively correlated with “everyday intelligence” scores involving map-, form-, and newspaper-reading). Others have argued that cognitive complexity is at the least a component or marker of intelligence (Simonton, 2009). However, although there is some evidence for a relationship between intelligence and cognitive complexity, it is not overwhelming—certainly not as strong as one might expect given the argued conceptual overlap between the two constructs. Indeed, it is the single weakest weighted average correlation in our entire review (as noted in Table 2, the average weighted r for the cognitive complexity–intelligence link is .13). For example, Suedfeld and Coren (1992) found that some intelligence measures, such as measures of divergent thinking and verbal ability, were consistently related to cognitive complexity in the low-to-moderate range, whereas other measures demonstrated mixed results. Specifically, both the Alternate Uses Test (Christensen et al., 1960) and the Comprehensive Ability Battery (Hakstian & Cattell, 1975)—two measures of divergent thinking—were positively correlated with conceptual complexity (assessed using a modified PCT; Coren & Suedfeld, 1995; rs ranging .11–.28, ps ranging from <.05 to <.001). Furthermore, both the Quick Test (Ammons & Ammons, 1962) and the Baddeley Grammatical Transformation Test (Baddeley, 1968)—two measures of verbal ability—were positively (albeit weakly) correlated with complexity (rs ranging .10–.14, ps ranging from .06 to <.05). However, for crystalized intelligence, one measure (the Wonderlic Personnel Test; Wonderlic, 1977) demonstrated a positive correlation with conceptual complexity, r(251) = .19, p < .001, while another (Spelling Component Test; Hakstian & Cattell, 1975) did not, r(251) = .02, ns. Finally, the single measure of fluid intelligence used in this study (Figure Classification; French, 1963) did not correlate with conceptual complexity (see Table 2 for details on this and the previous correlations). Importantly, all of these measures were taken at different data collection sessions spread over the course of 11 weeks, lending further credibility to their relationship for the purposes of trait evaluation.
Two studies of U.S. Presidents’ State of the Union speeches similarly revealed that some archival measurements of intelligence (from the Gough Adjective Checklist; Simonton, 1987) are correlated with U.S. presidents’ integrative complexity. First, Thoemmes and Conway (2007) found that complexity (scored following updated guidelines by Baker-Brown et al., 1992) and intellectual brilliance were positively but not significantly correlated at r(39) = .08. Similarly, Conway and colleagues (2020) found that complexity (measured by AutoIC; Conway et al., 2014) was positively but not significantly correlated with intellectual brilliance at r(39) = .24.
Academic performance
The strongest evidence for an overlap of stable ability with cognitive complexity stems from measures of academic performance (sample-weighted average r = .31). McDaniel and Lawrence (1990) report that cognitive complexity (operationalized as integrative complexity and scored following guidelines by Schroder et al., 1967) was positively correlated at moderate levels with the California Achievement Test, r(151) = .45, p < .01; Scholastic Assessment Test (SAT) verbal scores, r(51) = .39, p < .01; the reading, r(52) = .40, p < .01, language, r(52) = .26, p < .05, and math, r(52) = .29, p < .05, totals of the Indiana State Test for Educational Progress; and high school history (rs ranging .37–.57, ps < .01), science, r(52) = .30, p < .05, English, r(52) = .27, ns, and mathematics, r(52) = .20, ns, grades. Furthermore, those with higher levels of cognitive complexity were found to have deeper and more elaborative processing (measured by Learning Style; Wood & McDaniel, 1990), r(52) = .33, p < .01. On the other hand, SAT math scores and three separate tests of cognitive or thinking skills (Test of Cognitive Skills, Watson–Glaser, and Cornell) were all largely uncorrelated with cognitive complexity (except for the analogies section of the Test of Cognitive Skills), r(50) = .35, p < .01. Although slightly inconsistent, the general pattern favors a modest relationship between academic performance and cognitive complexity. Because academic performance is generally reflective of a large span of time and as such is not particularly prone to transient influences, it seems germane to establish the ability–complexity personality link. This is tempered, however, by the fact that (a) all the reported effects for ability come from the same research project, and (b) the performance measures were not specifically designed to directly tap into trait-level abilities.
Ability: Summary
Taken together, this converging evidence does suggest that the intuitively appealing ability–complexity link exists (sample-weighted average r = .23), though it is still unclear how strong it is. Indeed, some researchers (see Coren & Suedfeld, 1995; Suedfeld & Coren, 1992; Thoemmes & Conway, 2007) have suggested that the available evidence points toward conceptualizing cognitive complexity as a motivational “style” rather than an “ability,” thus proposing that the stable motivational aspects of complexity are more important than ability. This follows research on personality traits in general, which suggests that most individuals can display most levels of a trait construct (i.e., state manifestations of a trait can span most levels of the construct) given the right circumstances (Fleeson, 2017). Along these lines, prior research has combined cognitive traits, including complexity, nonconformity, autonomy, and low degrees of structure, into a particular thinking style that predicts, for example, peer moral and learning environments (Fan & Zhang, 2014). These thinking styles differ from student-to-student even after controlling for gender, grade, major, and socioeconomic status. In sum, despite calls suggesting that integrative complexity is largely unrelated to ability (see Coren & Suedfeld, 1995; Suedfeld & Coren, 1992; Thoemmes & Conway, 2007), the current evidence seems at the very least to support the general notion that stable cognitive and intellectual abilities do have some relationship to stable cognitive complexity. And indeed, the overall weighted effect sizes for motivation (.20) and ability (.23) are very similar—suggesting that the influences that produce trait complexity are roughly equally attributable to stable motivations and stable abilities. Furthermore, the research discussed in this section use different operationalizations of intelligence and academic performance. It is worth noting, however, that conceptualizations of ability differ even more widely across populations of culturally, socially, economically, and educationally diverse populations. Thus, this work would greatly benefit from research that explores the relationship between cognitive complexity and more diverse definitions of ability.
General Discussion
“[I]f individuals have similarly organized mental representations, but their representations differ in content, complexity, or valence, these differences should result in inter-individual covariation of pertinent behaviours” Baumert et al. (2017, p. 510).
As Baumert and colleagues suggest, understanding the potential for cognitive complexity to vary across persons might have wide implications for personality psychology. Indeed, by investigating the extent to which state cognitive complexity is the result of trait cognitive complexity, our review helps fill in a much-needed gap in the literature. Taken together, the evidence presented in this review converges toward this theoretical conclusion: cognitive complexity can be meaningfully discussed as having a trait component, and this component likely accounts for a small-to-moderate amount of variance in state complexity. This effect held across the multiple types of validity tests presented in this review, including generalizability across domains, stability over time, stable motivational correlates, and stable ability correlates. Thus, certain types of people are chronically more likely to form and use cognitively complex representations across many different domains over time. Below, we conclude by discussing implications of our findings, their limitations, and suggestions for future research.
Implications for Cognitive Complexity’s Trait Component
The fact that cognitive complexity has a trait component provides an important theoretical link among hundreds of different empirical studies. Indeed, as mentioned in the introduction, two major early cognitive complexity theories relied on trait differences in cognitive complexity: “Interactive Complexity Theory” (Schroder, 1971; Schroder et al., 1967; Streufert, 1969, 1970, 1972; Streufert & Driver, 1967) and “Systems Theory” (e.g., Harvey et al., 1961). Such theoretical work would be undermined if the link between personality and cognitive complexity does not exist in a meaningful, consistent, empirically verifiable manner.
But why, specifically, might this matter? As mentioned at the beginning of this review, cognitive complexity affects a wide array of important outcomes across a diverse set of areas ranging from politics to economics to health (e.g., Boyd-MacMillan, 2016; Conway & Conway, 2011; Conway et al., 2011; S. C. Houck et al., 2017; Smith et al., 2008; Suedfeld & Bluck, 1988; Suedfeld et al., 1977; Suedfeld & Jhangiani, 2009; Tetlock, 1985). So, what implications does the small-to-moderate trait component of cognitive complexity have on our thinking about these areas? Consider, for example, work revealing correlations between higher levels of terrorism by a group and low levels of cognitive complexity in the group’s leadership (Conway & Conway, 2011; Conway et al., 2011; S. C. Houck et al., 2017; Smith et al., 2008; Suedfeld et al., 2013). How we approach this correlation depends in part on the degree to which we believe that complexity has a meaningful trait component, 14 and what factors we believe contribute to that component. If we believe that complexity does not have a trait component, we would assume that interventions designed to reduce the likelihood of terrorism by increasing complexity must be focused solely on sweeping ecological changes that alter aspects of the situation. If that were the case, interventions designed to produce chronic stable thinkers at the individual level would not be meaningful.
However, if we assume complexity is partially due to trait differences that transcend such situational constraints, we would conclude that at least part of the complexity–terrorism correlations are not caused by situational constraints, but rather are due to self-selection or other processes that may play a role in certain types of individuals taking certain roles, and a very small percentage of these individuals becoming involved in violent extremism. 15 Indeed, some research found that political radicalization—the process of developing extremist ideology—is a result of both trait-based differences (e.g., dispositions to enjoy admiration and adventures) and situational factors (e.g., group-level mechanisms) that are hard to tease apart (McCauley & Moskalenko, 2017). The same researchers later promulgated a Two-Pyramids model to explain radicalization by separating radicalization of opinion from radicalization of actions, which are both influenced by traits and situational factors (McCauley & Moskalenko, 2017).
Of course, the present set of findings does not imply an immutable trait that inevitably leads someone to terrorism. Quite the contrary. Not only does our review reveal that a trait component comprises generally a small-to-moderate amount of variance (thus leaving plenty of room for situational factors), but even stable traits are subject to development and change over time (e.g., Chung et al., 2014). Rather, this work suggests the angle of approach might be different if (as our review suggests) cognitive complexity has a trait component. In that instance, developing more stable complex thinking in individuals across many domains would be a more reasonable target (and not merely attempting to change ecological circumstances). Our review further suggests that this goal could be enhanced by both motivational interventions and capacity interventions. In line with this approach, some have advocated for schools to teach students a balanced mix of opinions on political issues, which demotivates radicalization by increasing the complexity of ideas, as a means of preventing extremism (Davies, 2016). This approach operates on the assumption that these interventions might produce more stable complex thinkers—and our work validates that goal. Other work has similarly shown success in reducing radicalism by specific interventions targeted at increasing cognitive complexity (e.g., Boyd-MacMillan, 2016; Liht & Savage, 2013; Savage et al., 2014), again suggesting the possibility for a more general change in complexity that could be stable over time. Furthermore, given the increasing disparities and polarization during COVID-19, it is increasingly important to focus the influence of cognitive complexity on partisanship in more diverse populations.
Limitations and Future Directions
This review also suggests that there are significant gaps in the literature. First, there is a need for additional literature that investigates the across-domain generalizability and across-time stability of cognitive complexity to provide more specific estimates of the exact amount of variance attributable to the trait (as opposed to a situational) component. Specifically, future theory will especially benefit from studies that measure temporal stability through cognitive complexity scores gathered at multiple time points in a truly longitudinal design, so that SEM can be applied on longitudinal cognitive complexity data to determine the amount of variance that is due to trait cognitive complexity, the situation, and error (see Cole, 2012; Donnellan et al., 2012; Kenny & Zautra, 1995, 2001; Kuster & Orth, 2013).
Furthermore, additional research is necessary to better understand the specific factors contributing to the cognitive complexity trait. Our wide aerial view of the literature suggests that cognitively complex people possess almost as much chronic motivation to complex thinking as they do the ability to produce it. But how might these twin influences on the cognitively complex person work together? Currently, there is not enough focused research to produce a comprehensive theory, the criteria for which have been outlined by other researchers (e.g., Baumert et al., 2017; Fleeson & Jayawickreme, 2015; Rauthmann, 2015; Rauthmann et al., 2014; Ziegler, 2014). However, here we offer some speculative reflections to provide a framework for future investigation. First, it is possible that ability and motivation are independent stable forces that operate orthogonally on the individual. As such, they would in effect be additive, such that persons, for example, low in stable ability but high in stable motivation may be “medium” with respect to cognitive complexity. On the other hand, ability and motivation may interact, such that one of the two might be a necessary precursor for complexity to exist. It is possible, for example, that—as other researchers have suggested (Thoemmes & Conway, 2007)—motivation to complexity is a necessary precursor for its production, and that no matter how much ability one has, no complexity emerges without motivation. If that were true, persons low in motivation but high in ability would be “low” with respect to cognitive complexity. Furthermore, it is also conceivable that, although conceptually orthogonal, ability and motivation may be empirically correlated in nontrivial ways. For example, the person who is gifted with high ability levels for complex thinking may find it more motivating. If true, that could mean that the correlations between ability and complex thinking actually were carried more through motivation than ability (or vice versa). No clear data exists to tease apart these various alternatives. The set of data in our review does, however, provide a clear roadmap for future researchers by illustrating the gaps currently in cognitive complexity research that need to be filled to establish a clearer and more comprehensive trait theory. Specifically, more research that measures ability, motivation, and state complexity in a longitudinal design with large samples would be able to better parse the degree that these components are additive, interact, or mediate one another’s contribution to the trait component of cognitive complexity.
In addition, multivariate analyses could be used in future research to further explore the structure and content of the trait component of cognitive complexity. For example, factor analyses could be conducted on measures of cognitive complexity and other cognitive individual difference constructs, such as Need for Cognition. If these measures load onto separate factors, it would suggest that cognitive complexity and Need for Cognition (or other cognitive variables) represent different latent psychological structures (i.e., different traits). On the other hand, if these measures load onto the same factor, that would suggest that there is a common latent psychological structure (i.e., the same trait) that accounts for both cognitive complexity and the other cognitive variables. Similar work conducted by Mussel (2010, 2013) has demonstrated large overlaps between Need for Cognition, intellectual engagement, and openness of ideas (all three of which are conceptually related to cognitive complexity), suggesting that these three cognitive variables share the same latent trait structure.
Concluding Thoughts
Over 40 years ago, Streufert and Streufert (1978) identified the need for specific programs to fully assess the similarities and differences between widely discrepant measures of cognitive complexity across widely discrepant domains: Which of these interpretations will hold remains to be tested in future research which is specifically designed to determine the communalities and differences among various complexity approaches. It is obvious that research of this kind is needed.
It seems that this advice has not been entirely heeded by psychologists. In the last 40 years, to our knowledge, no serious effort has been made toward theoretically integrating the various programs in cognitive complexity research. Indeed, our sense is that, if we are moving at all, it is in the opposite direction: Although cognitive complexity continues to be a well-researched topic, it is done in isolated theoretical pockets and rarely, if ever, discussed as a broad area of study for its own sake. If we are to understand exactly what cognitive complexity is and develop a full integrative theory of its trait component, psychologists need to take more seriously the call to test the relationships between various different theoretical and methodological programs of research. The present review importantly brings together these various disparate research programs under one umbrella. Our review provides clear evidence for the necessary starting point in this endeavor: Cognitive complexity states are, in fact, partially the result of a trait component, and further offer some clues about what contributes to that component. However, our review also suggests that more work needs to be done to better understand this important construct.
Footnotes
Appendix
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.
