Authors: Jeffrey C. Zemla
Categories: Article, mild cognitive impairment, reasoning, cognitive style, heuristics, judgment
Source: Neuropsychology, development, and cognition. Section B, Aging, neuropsychology and cognition
Authors: Jeffrey C. Zemla
Reasoning can be fast, automatic, and intuitive or slow, deliberate, and analytical. Use of one cognitive reasoning style over the other has broad implications for beliefs, but differences in cognitive style have not previously been reported in those with mild cognitive impairment (MCI). In one experiment, the cognitive reflection test is used to measure cognitive style in healthy older adults and those with MCI. Those with MCI performed significantly worse than cognitively healthy older adults on the cognitive reflection test, indicating that they are more likely to engage in intuitive thinking than age-matched adults. This association is reliable after controlling for additional cognitive, self-report, and demographic factors. Across all measures, subjective cognitive decline was the best single predictor of cognitive status in the data. A difference in cognitive style represents a novel behavioral marker of MCI, and future work should explore whether this explains a broader pattern of reasoning errors in those with MCI, such as increased susceptibility to scams or impaired financial reasoning.
Mild cognitive impairment (MCI) can significantly disrupt a person’s life, causing changes to both cognition and daily routine. The most prevalent symptom of MCI is impaired memory (Petersen et al., 1999), though changes in reasoning and decision-making have also been documented. Individuals with MCI show declines in financial reasoning (Martin et al., 2019) and medical decision-making (Okonkwo et al., 2007), and may be more vulnerable to scams (Han, Boyle, James, Yu, & Bennett, 2016) and misinformation (Marin et al., 2021). Despite this, little research has explored the cognitive basis for impaired reasoning in MCI. Here, I report a novel association between MCI and cognitive reasoning style (Kahneman, 2011; Sloman, 1996; Stanovich & West, 2000), measured with the cognitive reflection test (Frederick, 2005). I find those with MCI are more likely than age-matched cognitively healthy adults to adopt an automatic, associative, intuitive thinking style, as opposed to a deliberate, rule-based, analytical thinking style that is characteristic of normative reasoning. The present work contributes to our understanding of how reasoning is affected in individuals with MCI. Among healthy adults, cognitive style has been linked to impairments in applied reasoning (e.g., poor financial reasoning; Bucher-Koenen & Ziegelmeyer, 2011), and future work should test whether a shift from analytical to intuitive thinking explains these impairments in MCI.
People routinely engage in complex mental reasoning yet are also susceptible to a surprising number of systematic cognitive biases. Many reasoning biases can be attributed to using the wrong style of dual-process theories of reasoning propose that people engage in both intuitive and analytical thinking^1^ (Kahneman, 2011; Sloman, 1996; Stanovich & West, 2000), and failure to use the correct style of reasoning can result in error.
Intuitive thinking is fast, automatic, and associative. It does not place high demands on cognitive resources. In contrast, analytical thinking is slow, controlled, and requires deliberate rule-based manipulation of information. People frequently engage in both styles of thinking depending on the context. For example, when solving the anagram involnutray, the correct answer involuntary is likely to automatically pop into one’s head without effort, an example of intuitive thinking. When solving the anagram uersoippv, it is unlikely that the anagram will be solved quickly and without deliberation. Instead, one is likely to re-arrange the letters in a deliberate and methodical manner before arriving at the solution purposive, an example of analytical thinking (Sloman, 1996).
Intuitive thinking is characterized by the use of heuristics that simplify decision-making. For example, Kahneman & Tversky (1973) found that when making a judgment about someone’s group membership, people often rely on subjective similarity rather than statistical beliefs because judging similarity is easy and intuitive, while reasoning about probabilities is hard and unnatural. Many cognitive heuristics have been documented to explain biases in decision-making, but together these heuristics all serve a common reduce the cognitive effort required (Shah & Oppenheimer, 2008) when a normative approach is too difficult or effortful. As a result, intuitive thinking is often a default reasoning strategy that emerges with little effort.
In contrast, analytical thinking is characterized by the use of formal, rule-based strategies. For example, analytical thinking may make use of logical, probabilistic, causal, or numerical reasoning. The misapplication of intuitive thinking to situations that require analytical thinking has been identified as the source of many cognitive biases, including logical reasoning failures in the Wason selection task (Thompson, Evans, & Campbell, 2013), belief bias in syllogisms (Toplak, West, & Stanovich, 2014), base-rate neglect (De Neys & Glumicic, 2008), and poor conflict detection on the cognitive reflection test (Frederick, 2005). In each of these cases a cognitive bias emerges because an intuitive response comes to mind first, automatically. It is necessary to suppress this response and engage in analytical thinking to obtain the normatively correct response. That is, one must not only have the knowledge and ability of how to solve a problem analytically, but must also recognize that the default intuitive strategy is incorrect and override that strategy (Evans & Stanovich, 2013). This often leads to errors even among mathematical psychologists make intuitive judgments about the representativeness of small sample sizes that are mathematically unfounded (Tversky & Kahneman, 1971), and physicians make systematic errors in diagnosis that are inconsistent with standard medical practice (Saposnik, Redelmeier, Ruff, Tobler, 2016).
An example of the belief bias effect clarifies the distinction between the two cognitive styles. Consider the following syllogism (from Markovits & Nantel, 1989):
When asked whether the conclusion logically follows from the premises, most people report that it does–though in fact, it does not. This bias stems from the fact that most people believe the conclusion to be true in the real world. Conversely, people often reject valid syllogisms when they find the conclusion unbelievable. The automatic and intuitive response is to judge the validity of the syllogism based on the believability of the conclusion (Evans, 2007). When explained in detail, most people grasp that the conclusion does not logically follow; the bias does not emerge because people are incapable of understanding syllogisms, but rather because many people fail to recognize that analytical thinking is required (Pennycook, Fugelsang, & Koehler, 2015). As a result, they do not override their initial intuitive response.
In the example above, there is an objectively correct answer (the conclusion does not follow from the premises) that is not intuitive to most people but can be derived from analytical thinking. However it is worth noting that intuitive thinking is not always the “wrong” style of thinking. Rather, the most appropriate reasoning style depends on context. For example, an expert chess player might rely on intuition to save time when playing speed chess, but plan their moves methodically and consider multiple options when time is not a factor. In other cases, intuitive and analytical thinking can produce the same response; it is easy to design syllogisms like the one above where the conclusion both logically follows from the premises and is also true in reality. Experimental stimuli like the belief bias syllogism above are designed to put the two systems in conflict in order to test whether a participant detects and resolves this conflict (Pennycook, Fugelsang, & Koehler, 2015). While these stimuli may give the impression that analytical thinking is inherently superior, it is untenable to rely solely on analytical thinking for everyday reasoning because it is “much too slow and inefficient to serve as a substitute for [intuitive thinking] in making routine decisions” (Kahneman, 2011).
Though simple (and perhaps contrived) problems such as belief bias syllogisms are often used to measure cognitive style, the impact of cognitive style is pervasive. Cognitive style is strongly associated with all sorts of fundamental beliefs about the world, including political values (Deppe et al., 2015), moral values (Pennycook, Cheyne, Barr, & Koehler, 2014), religiosity (Gervais & Norenzayan, 2012; Zemla, Steiner, & Sloman, 2016), conspiratorial beliefs (Swami, Voracek, Stieger, Tran, Furnham, 2014), and much more (Pennycook, 2023). As a result, the predisposition to engage in analytical thinking or intuitive thinking can have a big impact on one’s life and how they view the world.
While older adults show declines in many areas of reasoning and decision-making, few studies have directly explored whether cognitive style is associated with healthy aging. In one study, Hertzog, Smith, & Ariel (2018) found that older adults were more likely to engage in intuitive thinking on the cognitive reflection test. Consistent with this, Ding et al. (2020) found that older adults were also more susceptible to belief bias than younger adults. However, other studies have failed to find an association between age and cognitive style. Pehlivanoglu et al. (2022) was unable to replicate the age-related difference in cognitive reflection test scores found by Hertzog et al. (2018). Pollack, Overton, Rosenfeld, & Rosenfeld (1995) found no evidence for age-related differences in the Wason selection task.
In addition to empirical evidence, it is also useful to consider theoretical arguments for why analytical thinking might decline with age or impairment. Why are people susceptible to heuristic biases at all? Some authors (e.g., Ariely, 2008; Kahneman, 2003) have framed these biases as irrational and a product of limited cognitive resources (i.e., bounded rationality; Simon, 1955). Another view (Gigerenzer, 1991) is that cognition is rationally adapted to the constraints of the environment, and that many of the results purported to demonstrate bias are stripped of the ecological context from which cognitive heuristics evolved. For example, Gigerenzer and Hoffrage (1995) found that base-rate neglect is more pronounced when quantities are described as probabilities and less pronounced when using frequencies. They argue that mind has evolved to reason about frequencies because for most of human history this is the format that we consumed information, whereas reasoning about probabilities is less natural and cognitive heuristics did not evolve to think about base rates as probabilities.
Analogously, one can speculate on why older adults would show a decline in analytical thinking. One possibility is that this shift in older adults represents an impairment—a deviation from rational thinking that stems from declining cognitive and brain resources. Alternatively, one might view it as a rational cognitive heuristics allow us to maintain relatively high levels of reasoning in spite of cognitive declines. The latter interpretation aligns with theories of aging that contend older adults are more selective of when to engage in cognitively effortful tasks as a result of declines in cognitive resources (Hess, 2014; Touron, 2015). Heuristic strategies require less cognitive effort (Shah & Oppenheimer, 2008), which may explain why older adults may be more prone to intuitive thinking than analytical thinking.
To date, there are no studies that have directly tested whether MCI or dementia is associated with a shift in cognitive style. The most commonly used dementia screeners, including the Mini-Mental State Exam (Folstein, Folstein, & McHugh, 1975) and Montreal Cognitive Assessment (MoCA; Nasreddine, 2005), do not measure cognitive style. These screeners focus primarily on measuring basic cognitive processes such as memory, attention, visuospatial skills, executive functioning, and language, but do not include any direct measures of reasoning, judgment, or decision-making.
In this experiment, cognitively healthy older adults and older adults with MCI were tested on the cognitive reflection test, one of the most widely used measures of cognitive style. I hypothesized that individuals with MCI would score lower on the cognitive reflection test than healthy adults, indicating increased use of intuitive cognitive style. Participants also completed a letter and animal fluency task measuring executive and semantic memory functioning, respectively, as well as a subjective cognitive decline scale. These tasks were included in order to see whether an association between cognitive status (healthy or MCI) and cognitive reflection persists after controlling for these strong indicators of MCI. The study was approved by the Institutional Review Board of the University of Wisconsin-Madison 2020–0408.
105 participants were recruited from the Alzheimer’s Prevention Registry (Langbaum et al., 2020; endALZnow.org). Participants were eligible if they were at least 55 years old and were either cognitively healthy or were previously diagnosed with MCI. 12 participants were excluded from analysis after reporting at least one additional diagnosis that might impair cognition (such as Lewy Body Dementia, Alzheimer’s disease, Huntington’s disease, Parkinson’s disease, medication-induced delirium, or other self-reported diagnoses). Demographics for the remaining 93 participants are reported in Table 1 (one cognitive healthy participant was missing demographics). There were no significant differences in age, gender, or education between cognitively healthy and MCI groups, though the proportion of females was slightly higher for MCI participants (75% female) compared to cognitively healthy participants (56% female), p = .08 by chi-square test.
Participants completed the experiment remotely in a web browser. They began the experiment by completing an unrelated task, reading short vignettes about scientific discoveries and making judgments about them (Zemla, Chambers, Austerweil, & Cimpian, 2021). Afterwards, all participants completed a semantic fluency task and letter fluency task; the cognitive reflection test; and a subjective cognitive decline scale (Miebach et al., 2019).
In the semantic fluency task, participants listed as many animals as they could in two minutes. The letter fluency task was identical except that participants listed words that begin with the letter A instead of animals. The data were visually inspected to ensure participants followed instructions, but intrusions (non-category items) and perseverations (repeated items) were not removed from the data set. Cognitively healthy older adults generate more responses than those with MCI in letter fluency and semantic fluency because the tasks measure executive and semantic memory functioning (Nutter-Upham et al., 2008). The tasks are used in the current experiment in order to see whether an association between cognitive style and cognitive status persists after controlling for variance explained by declines in semantic memory and executive functioning.
The cognitive reflection test consisted of three fill-in-the-blank
Each of the questions has an intuitive response that comes to mind automatically, and an analytical (correct) answer.^2^ The responses were manually coded as either analytical, intuitive, or other (any incorrect response that does not match the intuitive response).
The scale (Miebach et al., 2019) consisted of four yes-or-no
The number of ‘yes’ responses for each participant was summed and used as a measure of subjective cognitive decline (ranging from 0 to 4). Subjective cognitive decline is known to be one of the earliest predictors of dementia (Jessen et al., 2014), and the scale was included in this experiment to determine whether an association between cognitive style and cognitive status persists after controlling for subjective decline.
Results indicate that cognitively healthy participants engaged in analytical thinking more than those with MCI. Cognitively healthy participants answered twice as many questions correctly on the cognitive reflection test than those with MCI, Mhealthy = .91, MMCI = .44, t(79.4) = 2.75, p = .007, Cohen’s d = .58. Scored differently, individuals with MCI provided the intuitive lure response more than cognitively healthy participants, Mhealthy = 1.73, MMCI = 2.17, t(87.2) = 2.31, p = .023, Cohen’s d = .48.^3^
Figure 1 shows these results separately for each question in the cognitive reflection test. Incorrect answers that did not match the intuitive lure were much less common, and their prevalence did not differ between groups (Mhealthy = .36, MMCI = .40, p = .74).
Scores on other cognitive measures also significantly differed (or nearly so) between cognitively healthy and MCI groups. Those who were cognitively healthy listed more animals, Mhealthy = 22.4, MMCI = 17.7, t(86.1) = 3.11, p = .003, Cohen’s d = .65, more words beginning with the letter A, Mhealthy = 29.6, MMCI = 16.8, t(89.6) = 1.95, p = .054, and reported less subjective cognitive decline, Mhealthy = 1.98, MMCI = 3.06, t(80.7) = 4.12, p < .001, Cohen’s d = .86. Despite this, the association between cognitive style and MCI appeared to be independent of these other cognitive predictors. Multiple logistic regression was used to predict cognitive status (healthy or MCI) from the number of correct responses on the cognitive reflection test, animal fluency scores, letter fluency scores, and subjective cognitive decline. The number of correct responses on the cognitive reflection test remained a significant predictor of cognitive status, odds ratio = 0.45:1, p = .017^4^. Subjective cognitive decline was the best predictor (odds ratio = 1, p < .001), while animal fluency (p = .1) and letter fluency (p =.78) were not significant predictors in the model. Given the slight gender imbalance of the samples, an additional logistic regression model was run that included gender as a factor. Gender did not significantly predict cognitive status (p = .12), but cognitive style remained a significant predictor (odds ratio .51:1, p = .047). However, if age and education are added to this model (i.e., including all possible covariates), cognitive style is no longer significant (p = .077).
Pairwise comparisons between dependent measures and demographics are presented as a correlation matrix in Figure 2. In addition to the association reported above, the following significant correlations were letter and animal fluency, letter fluency and cognitive reflection (correct and intuitive scorings), animal fluency and cognitive reflection (correct and intuitive scorings), animal fluency and subjective cognitive decline, gender and cognitive reflection (correct scoring), as well as correlations among all three methods of scoring the cognitive reflection test. Notably, subjective cognitive decline was uncorrelated with cognitive reflection. Age was not correlated with any variable (though marginally correlated with animal fluency, p = .059).
Individuals with MCI were less likely to adopt an analytical style of thought compared to cognitively healthy adults of the same age, as evidenced by the number of correct responses to the cognitive reflection test. An alternative explanation might be that those with MCI are attempting to engage in analytical thinking, but are simply unable to answer the questions correctly (perhaps because of poor numeracy). However, this explanation seems unlikely or at best incomplete, because those with MCI generated significantly more intuitive lure responses than healthy controls (Figure 1B) associated with intuitive thinking. If those with MCI were attempting to use analytical thinking, one would expect an increase in incorrect but non-intuitive responses, but these responses were much less common and their prevalence did not differ between cognitively healthy and MCI participants (Figure 1C).
Some researchers have questioned whether the cognitive reflection test is nothing more than a numeracy test (e.g., Sinayev & Peters, 2015). On its surface, it is easy to see why this might be the the test requires participants to use math in order to answer the questions correctly. However, previous work has also found that while the cognitive reflection test requires numeracy, it is not merely numeracy. Performance on the cognitive reflection test is moderately correlated with measures of numeracy (Cokely & Kelley, 2009; Obrecht, Chapman, & Gelman, 2009), but factor analyses have shown that the two constructs are separable (Liberali, Reyna, Furlan, Stein, & Pardo, 2012). Moreover, the cognitive reflection test appears to predict a wide range of beliefs that cannot be statistically explained by numeracy alone (see Pennycook & Ross, 2016 for a discussion). Still, numerical ability is undoubtedly necessary to complete the cognitive reflection test. Future experiments may disambiguate these components (numerical ability and disposition to engage in analytical thinking) by controlling for numeracy.
The association between cognitive style and MCI was also distinct from other cognitive measures used in the experiment. Namely, the animal and letter fluency tasks require semantic memory and executive functioning, but controlling for performance on these tasks did little to attenuate the association between cognitive style and MCI. Similarly, the association was unaffected after controlling for subjective cognitive decline (an early indicator of MCI). The association between MCI and verbal fluency (both letter and semantic) has been explored extensively (e.g., Chasles et al., 2020; Clark et al. 2014; Nutter-Upham et al., 2008; Rinehardt et al., 2014). In contrast, an association between cognitive style and MCI has not been previously reported, even though the pairwise correlation suggests that the association is equally strong (|r| = .28 vs. |r| = .31 and .2 for semantic and letter fluency, see Figure 2). Because the cognitive reflection test is predictive of MCI but theoretically distinct from cognitive abilities commonly assessed (e.g., memory and executive functioning), future work should test whether it improves diagnosticity when combined with a dementia screener like the MoCA. However, the MoCA was not administered in this experiment so it is premature to make a definitive claim.
If the cognitive reflection test proves useful in identifying individuals with MCI, a concern is whether the test is susceptible to practice effects through familiarity with the items (Haigh, 2016; Stieger & Reips, 2016). Many people will take a dementia screener more than once (at different timepoints), and so practice effects could diminish its utility. However, previous work has found that when no feedback is provided to the participant, performance on the cognitive reflection test is relatively stable across time (Stagnaro, Pennycook, & Rand, 2018). In addition, a number of alternate form cognitive reflection tests have been developed (e.g., Toplak, West, & Stanovich, 2014; Thomson & Oppenheimer, 2016) that could be used to attenuate concerns of repeated testing of the same individual. Some of these alternate forms include more than three items (e.g., Toplak, West, & Stanovich, 2014 use a seven item version of the cognitive reflection test that includes the original three items plus four new ones), which may attenuate concerns about restricted range in the original test. Concerns about practice effects are not unique to the cognitive reflection test, and these concerns have motivated the development of alternate forms for the MoCA as well (Bruijnen et al., 2020; Chertkow, Nasreddine, Johns, Phillips, & McHenry, 2011). While the current experiment provides some evidence for an association between cognitive style and MCI, more work is needed to determine whether the cognitive reflection test is the most suitable scale for measuring intuitive thinking as a trait in MCI, or whether other scales are more appropriate.
Given the results, it is also possible that a scale related to subjective reasoning may augment subjective cognitive decline scales (e.g., Miebach et al., 2019). In the experiment, subjective cognitive decline was completely uncorrelated with cognitive reflection (|r| < .01) despite being correlated with cognitive status (r = .4). In one sense, this is the subjective decline scale used here asks about memory, word finding, planning, and attention, but it does not ask about reasoning. However, it also demonstrates that the scale does not solicit self-reported decline in all aspects of cognition (in this case, reasoning) that are associated with cognitive status.
The increased use of intuitive thinking in those with MCI could possibly explain the mixed results in healthy older adults. While Hertzog et al. (2018) found that older adults score lower on the cognitive reflection test, this effect was not replicated by Pehlivanoglu et al. (2022). However, the overall accuracy of older participants in Hertzog et al. (2018) was much lower than Pehlivanoglu et al. (2022), and the performance of older adults in Hertzog et al. (2018) was similar to data from this current experiment.^5^ One possibility is that a subset of the older participants in Hertzog et al. (2018) have undiagnosed or unreported MCI. Neither work reports whether participants have a cognitive disorder or tests participants on a dementia screener.
Much of the research on reasoning and decision-making in those with MCI has focused on applied domains such as financial reasoning (Martin et al., 2019) and detecting scams (Han, et al., 2016). These domains have rightfully received attention because of their ecological relevance and impact on the lives of those with MCI. Nonetheless, judgments and decisions in these domains may be shaped by traits like cognitive style. Here, I discuss two areas of research that may benefit from further exploring the connection between cognitive style and MCI.
Older adults navigate many complex financial decisions, including how to save for and budget during retirement and how to plan for unexpected health care costs. Exercising sound financial judgment can help individuals better manage their financial risk. While many older adults prepare well in advance for these circumstances, cognitive decline can impede one’s ability to carry out those plans. A common tool used to measure financial reasoning is the Financial Capacity Instrument (Marson et al., 2000), which measures one’s ability to understand financial concepts and engage in financial tasks such as bill payment and bank statement management. Using this scale, Griffith et al. (2003) found that those with MCI have significantly worse financial capacity than cognitively healthy adults. Longitudinal studies have found that changes in financial capacity precede conversion to dementia (Triebel et al., 2009), and that capacity declines in those with MCI but remains relatively stable in cognitively healthy adults (Martin et al., 2019). These results suggest that a decline in financial capacity is a behavioral marker of MCI.
In addition to scales such as the Financial Capacity Instrument, those with MCI are impaired at laboratory tasks that involve numeracy or risk. For example, in the Iowa Gambling Task cognitively healthy adults reliably learn to select the best gambles over time when provided feedback, while those with MCI have difficulty doing so (Zamarian, Weiss, & Delazer, 2011). Those with MCI also have difficulty with numerical tasks in other domains, such as medical decision-making with numerical information (Delazer, Kemmler,, & Benke, 2013; Pertl et al., 2014), and are impaired at more fundamental numerical abilities like ratio processing (Pertl, Benke, Zamarian, & Delazer, 2015).
Intuitive reasoning has also been associated with impaired financial reasoning. Among healthy adults, intuitive thinkers are less likely to choose monetary gambles that maximize expected value, less willing to select delayed rewards in intertemporal choices, and more risk-seeking in loss-frame gambles (Frederick, 2005)—all of which can translate to poor financial decisions. Intuitive thinking is also correlated with poor numeracy in general (e.g., Zemla, Steiner, & Sloman, 2016). Stevens (2017) found that, after controlling for numeracy, participants who score high on the cognitive reflection test scored higher on a financial literacy scale (see also Bucher-Koenen & Ziegelmeyer, 2011).
MCI is also associated with poor ability to detect scams. Han et al. (2016) measured scam awareness with a five-item subjective questionnaire that probed willingness to engage with sales pitches, credulity towards grand claims, and beliefs about older adults’ vulnerability to scams. Participants with MCI were more likely to show poor scam awareness, and this increased with the severity of the cognitive impairment. Boyle, Yu, Schneider, Wilson, & Bennett (2019) used the same questionnaire and found that participants with low scam awareness were more likely to develop MCI and Alzheimer’s disease, and had a higher burden of Alzheimer’s pathology. Experiments with simulated e-mail phishing tasks in healthy older adults have found that the ability to detect phishing e-mails declines with age (Grilli et al., 2021) and is further impaired in those with a genetic predisposition to Alzheimer’s disease (Spreng, Ebner, Levin, & Turner, 2021).
Intuitive thinking is prima facie associated with a failure to detect scams that are designed to be appealing and persuasive. This is backed by studies that find intuitive thinkers are less adept at detecting phishing scams (Jones, Towse, Race, & Harrison, 2019) and are more willing to share fake news (Pennycook & Rand, 2019).
In summary, research has found evidence for (a) an association between MCI and poor financial reasoning and scam detection, and (b) an association between intuitive cognitive style and poor financial reasoning and scam detection in cognitively healthy adults. The current experiment adds to this by establishing a link between intuitive cognitive style and MCI. These connections suggest a novel hypothesis, that cognitive style mediates the relationship between MCI and poor decision-making in the financial and scam domains. Future experiments should test this hypothesis in order to better understand the cognitive underpinnings of declines in reasoning that accompany MCI. Similarly, a longitudinal study could explore whether intuitive thinking coincides with the incidence of MCI, and whether analytical or intuitive priming affect reasoning in those with MCI. Nonetheless, the current study identifies cognitive style as a novel behavioral marker of MCI.