We use causal reasoning when deciding to take medication (to alleviate symptoms), save money for retirement (to ensure a comfortable old age), or try a diet (to modify our body weight). We select actions because we believe they can bring about desired effects (Kleinberg, 2013). Our beliefs may be correct, such as recognizing a medication's ineffectiveness for ourselves, or faulty, such as believing “lucky” socks determine the outcome of a game. Beliefs inform our everyday decisions, influence whether we seek out new information, and affect how we use it.

Yet little is known about how beliefs influence our receptivity to new information and ability to use it to make decisions. Information seeking has been extensively studied for health (Anker et al., 2011), particularly regarding where people obtain health information (Korshakova et al., 2022). People often have incorrect health beliefs, such as thinking preventable illnesses cannot be prevented (Smith et al., 1999) or believing in conspiracy theories about health, as almost half of Americans do (Oliver and Wood, 2014). Such misconceptions are not limited to health. Lay beliefs also influence how we use new information. (Kleinberg et al. 2023) found that, even when incorrect, beliefs affected the perceived reasonableness of options, suggesting that evidence may not correct entrenched beliefs. Thus, understanding how beliefs interact with causal information is important.

In causal cognition, actions are a form of information gathering. After we take an action, we see its result and may behave differently next time. The types and roles of interventions chosen have been examined in the context of how people learn causal models (Bramley et al., 2017; Coenen et al., 2015), but in artificial domains where people cannot use prior knowledge, such as learning how a made-up machine works. We face a different problem in everyday choices as experience can influence perceptions of new information. For example, omitting information people expected to see in a causal model reduced their trust in it (Kleinberg et al., 2022). This finding is only possible because the questions were about topics such as how gray hair develops. In a novel domain in which participants are dependent on the information provided by researchers, they do not know what other information could be included and might be missing. This highlights a core difference: In familiar domains, we bring prior beliefs that may be hard to change.

Research testing decision-making with causal models in familiar domains, such as managing health and finances, revealed a puzzle: Causal information can lead to worse choices on familiar topics while aiding people in novel domains (Zheng et al., 2020). Before this, much work on causal cognition found that people are adept at learning and using causal models (Bramley et al., 2017; Lagnado and Sloman, 2004, 2006; Rottman, 2017) but focused on novel domains in which participants relied on the information provided and could not use their prior experiences, preferences, or knowledge. Recent work on domains in which people have varying experience (e.g., life choices such as buying a house) found that narrowly targeted information can improve decisions, while extra information led to worse choices (Kleinberg and Marsh, 2023). Highlighting the relevant paths within a complex model led to similar results as when only that information was presented, suggesting that people may have difficulty determining when which parts of a model are most relevant.

Engaging with a causal model may also lead to different types of reasoning. (Liefgreen and Lagnado 2023) found that participants who drew and updated causal models of legal evidence after hearing each side cited different reasons for judgments than participants who were asked only to describe the evidence. In particular, participants who drew causal models preferred simpler explanations. This work focused on the legal domain, so whether, in general, people use different types of reasoning depending on whether they use causal models is an open question.

Causal information is pervasive and should lead to better choices as it provides effective strategies for intervention. Prior work, however, suggests that this may not be true in familiar domains, but as it only examined decision accuracy, it could not fully answer this question. We addressed this open question in an exploratory study using a mixed-methods approach, eliciting qualitative information about how people make decisions with and without causal models in two domains (finance and health), along with quantitative information on decision accuracy. We examined (1) whether different types of reasoning are used with causal information compared to when people make decisions using their existing knowledge, (2) whether specific types of reasoning are associated with the correctness of decisions, and (3) whether consistency between prior beliefs and causal information was associated with decision accuracy.

Prior work revealed a conflict: People are adept at learning causal models and using them for decision-making in novel domains (Rottman, 2017), yet causal models can lead to worse choices in familiar domains (Zheng et al., 2020). Our results help reconcile why causal models aid decisions in novel domains but hinder them in familiar ones by showing that (1) people use different types of reasoning when making decisions with causal information, (2) different reasoning types predict accuracy with and without causal information, and (3) perceived conflict between a causal model and beliefs may yield worse choices. These results have implications for understanding cognition and decision-making and methodology.

(Zheng et al. 2020) first showed that causal models could yield worse decisions compared to no information. (Kleinberg and Marsh 2023) found that providing subsets of a model (i.e., only the relevant causal paths) improved accuracy, while including extra information led to worse results. Our results offer a potential mechanism and explanation why this effect is absent in novel contexts: In familiar domains, people do not use only the information provided. In both information conditions (diagram and text), participants incorporated their prior knowledge, their beliefs about feasibility of actions, and their own experiences, while in novel setups, participants cannot bring in experience and knowledge. While participants were more likely to reason logically when given no information compared to causal diagrams, logic significantly predicted accuracy in all three conditions. Reliance on outside information has been examined for causal attribution (determining who and/or what caused or is responsible for an event). Faced with a sparsely described vignette, people introduce additional assumptions and inferences to assign causality and responsibility (Alicke, 2000; Hilton, 2017; Kirfel and Lagnado, 2021). Taken together, this suggests a need for future work to better understand how causal reasoning differs in familiar domains in which people are able to add in details beyond the vignette.

We further found that belief conflict may contribute to how well people use causal information. A partial match between beliefs and information was negatively associated with accuracy in the text condition and may be especially challenging due to increased cognitive load or dissonance as people attempt to reconcile new information with conflicting beliefs (Yang et al., 2024; Harmon-Jones et al., 2009). With a full match, people do not need to resolve differences between mental and causal models, while with no match, people may ignore the information and use prior knowledge. In contrast, partial overlap may increase difficulty in making decisions. Conversely, when participants reported using only the information provided, the caveat is that with realistic causal models, some degree of interpretation is still required (e.g., translating exercise recommendations into specific activities and frequencies). Prior work has shown that causal evidence can be persuasive in changing beliefs (Slusher and Anderson, 1996; Hornikx, 2005), yet many other works show how difficult changing beliefs is (Ecker and Antonio, 2021; Jelalian and Miller, 1984) and that new information does not change our rating of decision options (Kleinberg et al., 2023). However, we did not assess accuracy of participants' prior knowledge. Potentially, the individuals who most need information may ignore it or struggle to reconcile it with their own beliefs. More work is needed to understand how to help people assimilate new information into their decision-making.

Our findings have significant methodological implications for studying causal reasoning. Prior work often used novel setups (e.g., blicket detectors), creating a controlled setting to fully manipulate a participant's information. While these studies provide insight into cognition, such as how people use observation and intervention to learn causal models (Steyvers et al., 2003), our study suggests that results may not be representative of those in familiar domains. For example, the degree to which participants relied on prior knowledge in our study and how experience influenced their choices (e.g., considerations about which interventions are easier) could not be predicted by studies with novel stimuli. Thus, examining the role of knowledge and experience in how people learn about and use causal information more fully is needed.

Finally, our research has practical implications for how to present causal information to maximize its benefit to a decision-maker. While (Kleinberg and Marsh 2023) showed that the simplest information leads to the best decisions, we now see that the match between a decision-maker's knowledge and the information presented also plays a role. When giving health advice to patients, understanding the relationship between their beliefs and information in decision aids is crucial. A complex diagram that does not capture their beliefs could do more harm than no information, whereas an overly simplistic diagram could be overridden by an individual's prior and possibly faulty knowledge.

Our results have some limitations. First, because decisions were hypothetical, we cannot say whether the same reasoning would be used for real-world decisions. However, because the decisions were realistic and not high stakes, we expect similar patterns in real-world contexts. Second, we did not assess participants' knowledge, so we could not distinguish between perceived vs. actual knowledge in our models. We expect that the key factor is perceived knowledge because people may be incorrect in their self-assessment. We are currently testing this in other work. Furthermore, our study was exploratory in nature, as we aimed to discover themes from the qualitative data. Now that we have identified key features related to decisions with causal information, future work can now test these hypotheses with preregistered studies (with predefined codes and criteria for assigning them) to replicate results. Finally, we studied two common areas of decision-making (health and finance) to keep the length of our study manageable for participants, but it is possible that results may differ for other domains. Future work is needed to examine the role of domain.

We use causal knowledge to navigate the world, yet little is known about how our prior knowledge and experiences shape how we interpret new information. Through qualitative and quantitative analysis of decision-making in everyday domains, we found that people engage in different reasoning with causal information compared to using existing knowledge, that different reasoning processes are associated with decision accuracy in each setting, and, finally, that the degree of concordance between prior beliefs and causal information is related to decision accuracy. Our work highlights the need to study decision-making in familiar domains and examine the role of prior causal beliefs in decision-making. Our findings may ultimately be used to improve communications for the lay public for decision-making.