Causal Models Drive Preference between Drugs that Treat a Focal versus Multiple Symptoms Causal Models Drive Preference between Drugs that Treat a Focal versus Multiple Symptoms KELLY SAPORTA SOROZON,[.]
Journal of Behavioral Decision Making, J Behav Dec Making (2017) Published online in Wiley Online Library (wileyonlinelibrary.com) DOI: 10.1002/bdm.1999 Causal Models Drive Preference between Drugs that Treat a Focal versus Multiple Symptoms KELLY SAPORTA-SOROZON,1* SHAI DANZIGER2 AND STEVEN SLOMAN3 Department of Education and Psychology, Open University of Israel, Ra’anana, Israel Coller School of Management, Tel Aviv University, Tel Aviv, Israel Cognitive, Linguistic, & Psychological Sciences, Brown University, Providence, RI, USA ABSTRACT This research examines the effects of causal beliefs on drug preference In three studies, 374 undergraduate participants imagined that they suffered from a focal symptom and then indicated their preference between a drug claiming to treat only the focal symptom (single treatment) and a drug claiming to treat the focal symptom and a nonfocal symptom (dual treatment) they thought resulted from a common-cause or from a different cause Participants who thought that the symptoms resulted from different causes significantly preferred the single treatment drug more and the dual treatment drug less than participants who thought the symptoms resulted from a common-cause Process analysis yielded a significant mediation effect Specifically, an individual’s causal model determines preference by affecting the estimates of the potential gain and loss associated with using each drug Participants who held a common-cause model concerning the source of the symptoms thought they would be more likely to need the nonfocal treatment provided by the dual treatment drug and less likely to suffer from side effects when taking this drug, than those with a different-cause model The results demonstrate the influence of causal structure on judgment and choice © 2017 The Authors Journal of Behavioral Decision Making Published by John Wiley & Sons Ltd key words causal model theory; causal structure; preference; judgment; consumer reasoning Imagine you suffer from insomnia during stressful periods Fearing sleepless nights, you decide to purchase an overthe-counter (OTC) drug you hope will help you sleep Drug A claims to prevent insomnia and drug B claims to prevent insomnia as well as a second symptom that you not have, nausea If the drugs are similarly priced, would you follow the adage “the more the merrier” and purchase the drug promising treatment of both symptoms or would you follow the adage “grab more, get less” and purchase the drug promising treatment of only the focal symptom, insomnia? Such dilemmas are becoming increasingly common for people According to a report by Visiongain, a British research company, the worldwide market share of OTC drugs may exceed $127bn in 2016 (10 May 2016) Further, the US Food and Drug Administration is continuously considering making more commonly used prescription drugs (like for birth control, asthma, and diabetes) available OTC The upside of such actions is that people will have increased access to many drugs But a potential downside is that people will have more responsibility; they will have to choose the right drug to treat their ailments In the present research, we examine one type of choice people may encounter, that of choosing between a drug that claims to treat only an expressed focal symptom and a drug that claims to treat both the expressed symptom and a second symptom that is absent or may be latent Inspired by causal model theory (CMT; e.g., Sloman, 2009; Waldmann & Holyoak, 1992), we propose that decisions of this sort are determined by peoples’ causal beliefs about the source of the symptoms When a person believes that the symptoms result from a common-cause, they will prefer the dual-treatment option more than when they believe the symptoms result from different-causes This is a reasonable inference: If two symptoms have a common-cause, the second symptom is more likely to develop than if they arise from different-causes So the gain of treating the second symptom, even if it has not occurred yet, by taking the dual-treatment drug, may outweigh the side effects of that drug (loss) Furthermore, under a common-cause model, people may expect side effects to be weak because the drug must treat only a single cause and therefore is expected to contain only one active ingredient In contrast, when the person believes the symptoms result from different-causes, the second symptom is less likely to arise, so the gain derived from treating it is less likely to outweigh the side effects that may result from taking the drug (loss) Moreover, under a different-cause model people may expect side effects to be strong because the drug must treat two different-causes and therefore may contain more than one active ingredient In sum, CMT predicts that people are more likely to choose the dual-treatment over the single-treatment drug when they believe the symptoms are due to a common-cause relative to when they believe they are due to different-causes (for a fuller explication of CMT’s application to decision making, see Sloman & Hagmayer, 2006) JUDGMENT AND EVALUATION OF SINGLE AND MULTIPLE BENEFIT PRODUCTS *Correspondence to: Kelly Saporta-Sorozon, Open University of Israel, Education and Psychology, University Road Ra’anana Israel, Ra’anana, Israel, 808 E-mail: kelisa@openu.ac.il We thank the Open University and the Henry Crown Institute of Business Research in Israel for funding this research Few studies have addressed the general question of the effects of adding product features on peoples’ product evaluation Bertini, Ofek, and Ariely (2009), for example, © 2017 The Authors Journal of Behavioral Decision Making Published by John Wiley & Sons Ltd This is an open access article under the terms of the Creative Commons Attribution-NonCommercial-NoDerivs License, which permits use and distribution in any medium, provided the original work is properly cited, the use is non-commercial and no modifications or adaptations are made Journal of Behavioral Decision Making examined the effects of add-on features on product evaluations They found that people evaluate products with nonalignable add-on features more favorably than products with alignable add-on features (e.g., 20 MB memory vs 32 MB memory) Thompson, Hamilton, and Rust (2005) show that adding features to a product may increase the products perceived capability but decrease its perceived usability Because consumers care more about product capability than usability during purchase (before using the product), they end up purchasing products that are too complex for them and therefore not maximize their satisfaction when using the products they purchase Even fewer studies have directly compared how consumers evaluate single and multiple benefit offerings Han, Chung, and Sohn (2009) investigated peoples’ preference between “convergent technological products” (e.g., camera phones) and “dedicated products” They found that for products with a relatively low level of technological performance (e.g., a digital camera with a low resolution – megapixel), participants preferred convergent products (e.g., a mobile phone with a digital camera), while at relatively high levels of technological performance (e.g., a digital camera with a high resolution – megapixels) participants preferred the dedicated option (a digital camera) Chernev (2007) examined participants’ perceptions of performance for a specialized option on its focal benefit when they considered it in isolation relative to when they considered it jointly with an all-in-one option (i.e., a dual-benefit product) Participants perceived the product offering a single-benefit to be superior on the focal benefit relative to the all-in-one option Chernev (2007) concluded that people believe options in a choice set are balanced such that advantages on one product dimension are compensated for by disadvantages on other dimensions These studies contribute to our understanding of adding product features and benefits on consumer product evaluations Yet, none of the studies ascribed importance to the causal relations between the product’s features/benefits In this paper, we propose that the perceived causal relations participants hold regarding the benefits significantly impact their preference between single and multiple benefits products In a paper that did ascribe importance to benefits and the relation between them, Zhang, Fishbach, and Kurglanski (2007) studied the effect of adding more goals (“effects”) to the same means (“product”) on the belief that the means will attain each goal Zhang et al (2007) provide support for a dilution effect They find that when a means (e.g., eating oranges) is thought to address a single goal (e.g., satisfy the need for fiber), the associative strength between the means and the goal, and as a result the belief that the means will attain the goal, is higher than when the means is thought to satisfy an additional goal (e.g., satisfy the need for fiber and for acquiring vitamins) Zhang et al (2007) conclude that participants are less likely to choose a means that is associated with multiple (vs single) goals when only one of the goals is activated Although the dilution effect pertains to causal claims, Zhang et al (2007) did not analyze it using the prism of causal reasoning Doing so suggests that peoples’ causal model acts as a moderator in determining the dilution effect Specifically, Zhang et al contend that the perceived distinctiveness (similarity) between the goals (effects) influences the extent of dilution, with highly distinctive (less similar) goals amplifying it, and less distinctive (more similar) goals attenuating it In our context, when symptoms are similar (e.g., vomiting and nausea), there should only be a small reduction in the belief that the dual-treatment product will treat the focal symptom (vomiting) as compared with the single-treatment product When the symptoms are dissimilar (e.g., insomnia and nausea), there should be a large reduction in belief that the dual-treatment product will treat the focal symptom (insomnia) as compared with the single-treatment product While we also make this prediction, we contend that it is causal model (CM) and not goal similarity (S) that affects the degree of dilution (D) In fact, we contend that the causal model affects both similarity and the degree of dilution (S ← CM → D) Notice that Zhang et al (2007) did not control for the degree of distinctiveness between goals in their studies, and they found evidence for dilution in each study An examination of their materials reveals that they chose goals that people most probably believe result from different-causes (e.g., prevents heart disease and helps to maintain healthy bones; prevents heart cancer and prevents degenerative disease of the eye; and increases the blood oxygen levels and strengthens the immune system) We contend that the causal model acts as a moderator in determining the degree of dilution, such that when participants perceive the symptoms as resulting from different-causes, there will be more dilution than when participants perceive the symptoms as resulting from a common-cause The dilution model also distinguishes between the reduction in the belief that the means will provide the benefits (dilution effect) and preference According to the dilution model, the addition of goals to the same means will reduce the belief that the means can obtain each goal but will not necessarily result in choosers preferring the single benefit option Only when the chooser is interested in satisfying the original goal will he or she prefer a means that satisfies only this goal Building on CMT, we propose that people’s causal model dictates the number of goals that will become relevant in a given context and that under a common-cause model, choosers will want to satisfy all of the goals (effects) that are associated with the same means (cause) We show that peoples’ causal model influences their preference even when they have a single original goal (treating the focal symptom) and that under a common-cause model, their preference for the dual-treatment option increases In sum, we contend that the causal model acts as a moderator in determining the degree of dilution, such that when the symptoms are perceived as resulting from different-causes (as in all the studies in Zhang et al.), there will be more dilution than when they are perceived as resulting from a common-cause Moreover, we assert that the causal model drives dilution and preference and not symptoms’ similarity (distinctiveness) And finally, we claim that the causal model affects preferences via the perceived gains and losses associated with adding the second treatment and not the degree of dilution In Study 3, we provide results supporting these assertions © 2017 The Authors Journal of Behavioral Decision Making Published by John Wiley & Sons Ltd J Behav Dec Making (2017) DOI: 10.1002/bdm K Saporta-Sorozon et al CAUSAL MODEL THEORY Over the last decade, there has been a paradigm shift in the study of human causal inference (Rottman & Hastie, 2013; Sloman & Lagnado, 2015) Instead of focusing on ‘cues to causality’ and causal beliefs (Ahn, Kalish, Medin, & Gelman, 1995; Einhorn & Hogarth, 1986) and their effect on judgment and inference (e.g., Cheng & Novick, 1992; Fernbach, Sloman, St Louis, & Shube, 2013), the new line of research focuses on peoples’ mental models of causal relations This research has led to the formulation of CMT (e.g., Sloman, 2009; Waldmann, 1996; Waldmann & Holyoak, 1992), a psychological understanding of human causal reasoning that is based on normative probabilistic models of causal inference (Pearl, 2000; Spirtes, Glymour, & Scheines, 1993) These probabilistic models are generalizations of Bayesian networks that involve inference over directed, acyclic graphs that represent causal structure The psychological version, CMT, posits that people represent causal knowledge in mental models that specify relevant variables and the causal structure relating them without requiring that people are probabilistically coherent To illustrate, the knowledge that malnutrition (M) causes physical disease (P) is mentally represented as M → P; the knowledge that M causes P by weakening the immune system (I; a mediating variable) is represented by a causal-chain M → I → P; and the knowledge that malnutrition also impairs cognitive skills development (C) is represented as a common-cause-model P ← M → C The theory posits that the structure in which the causal knowledge is represented (the causal model) constrains the nature of causal inferences and the ordering of expected covariations between the variables in the model (Lagnado, Waldmann, Hagmayer, & Sloman, 2007) For example, if we believe malnutrition causes physical and cognitive problems (P ← M → C), we realize that treating physical problems by training the cognitive system is absurd Also, if we believe two symptoms result from a common-cause, we expect the covariation between them to be higher than if we believe they result from different-causes CMT has motivated many studies that focus on how different causal structures (such as common-cause, common-effect, and several kinds of causal chains) affect causal judgments For example, de Kwaadsteniet, Hagmayer, Krol, and Witteman (2010) found that clinical child psychologists’ causal models of the factors causing clients’ problems influence their perceptions of the effectiveness of various interventions Other studies have shown that the causal structure of events influences, for example, judgments of conditional probability (Bes, Sloman, Lucas, & Raufaste, 2012), the expected covariation between events (Perales, Catena, & Maldonado, 2004), and how people make decisions (Hagmayer & Sloman, 2009) APPLYING CAUSAL MODELS TO INFERENCE ABOUT PRODUCTS Most artifacts are designed to address functional needs (Keil, 1992) In other words, the purpose of most artifacts is to deliver benefits In causal terminology, products (which are Causal Models Drives Preference artifacts) are causes that, together with human action and the right environment, deliver benefits (Fernbach et al., 2013) Thus, it is likely that people use causal models to represent causal knowledge for products just as they represent causal knowledge in the physical, biological, and psychological domains (Waldmann, Hagmayer, & Blaisdell, 2006) People base this mental representation on their knowledge concerning the causal relations among a product’s features For example, if we believe chest congestion (S1) and nasal stuffiness (S2) have a common-cause, we represent a drug that claims to treat both with a common-cause model: S1 ← C → -S2 (hyphen “-” means that the cause reduces the likelihood of the effect); and if we believe laundry static (S1) and stains (S2) have different-causes, we represent a laundry detergent that claims to prevent both with independent causes (C1 → -S1; C2 → - S2) Thus, how we represent the causal relations in products influences our inferences about them THE CURRENT RESEARCH Although the causal representation of products in the mind may influence product evaluation, research has yet to show this We fill this gap in the literature by showing that when people hold a different-causes structure, they prefer a single-treatment option more and a dual-treatment option less than when they hold a common-cause structure Our results indicate this pattern occurs because people evaluate the gain and the loss associated with the nonfocal benefit differently depending on their causal model HYPOTHESES Assume a person intends to purchase an OTC drug to treat a Symptom (S1) and at the point of purchase he or she is exposed to two drugs: drug A, which claims to treat S1 and drug B that claims to treat S1 as well as a second symptom S2 that the person does not have We propose that the person’s causal structure concerning the source of the symptoms drives their decision whether to purchase drug A or drug B (H1); and it drives their inferences concerning the gains (H2a) and the loss (H2b) associated with the decision to also treat the second symptom (S2) Moreover, we propose that the effect of the causal model on preference is mediated by the gains and the loss associated with the person’s decision to treat the second symptom (S2) (H3) Specifically, we hypothesize that: People will prefer the single-treatment drug more and the dual-treatment drug less when they believe the symptoms result from different-causes than when they believe the symptoms result from a common-cause (H1) We contend that two parameters determine the attractiveness of the drug that treats also the second symptom: the perceived probability of S2 occurring given that S1 occurred, P(Os2/Os1), and the probability of a side effect resulting from treating S2 (side effect addition, SEA), P(SEA) The first parameter [P(Os2/Os1)] reflects the gain element The higher © 2017 The Authors Journal of Behavioral Decision Making Published by John Wiley & Sons Ltd J Behav Dec Making (2017) DOI: 10.1002/bdm Journal of Behavioral Decision Making P(Os2/Os1), the greater the gain of choosing the drug that treats both symptoms because S2 may occur in the future or may be latent and therefore must be treated The second parameter [P(SEA)] reflects the loss element: the greater the probability of a side effect that may result from treating S2, the greater the potential loss With regard to gains, we predict P(Os2/Os1) will be larger for participants holding a common-cause structure than for participants holding a different-causes structure (H2a) This prediction is based on a fundamental tenet of CMT that the structure in which the causal knowledge is represented constrains the expected covariation between the variables in the model (Waldmann & Holyoak, 1992) If we believe that two effects result from a common-cause, we expect their covariation to be higher than if we believe they result from different-causes Specifically, as proposed by CMT, when the symptoms result from a common-cause, the occurrence of the first symptom (e.g., vomiting), makes the occurrence of the second symptom (e.g., nausea) – latently or in the future – more probable This prediction resonates with findings showing that the causal structure of events influences judgments of conditional probability (Bes et al., 2012), how covariation is assessed (Waldmann & Hagmayer 2001), and the expected covariation between events (Perales et al., 2004) In regards to losses, we propose P(SEA) will be smaller for participants that hold a common-cause structure than for participants that hold a different-causes structure (H2b) When people believe the symptoms result from a commoncause, they believe the same drug (i.e., same active agents) can treat both symptoms Because people believe that the dual-treatment drug uses the same active agent they will not expect additional side effects that may result from the addition of new active agents or hazards arising from drug interactions (which can happen when people take several drugs) However, when the symptoms result from different- causes, people will not believe the same drug can treat both symptoms and hence treating the second symptom requires an additional active agent; consequently, they may expect extra side effects and hazards resulting from drug interactions Finally, a prediction that CMT has in common with all of behavioral decision theory (Edwards, 1961) is that the subjective expected utilities of gains and costs will mediate the effect of the causal model on preference (H3) That is, preference should be proportional to the product of the subjective probability that certain outcomes will follow the act multiplied by the respective subjective values (or importance) attached to these outcomes (Ajzen & Fishbein, 1972) In line with this definition, we use the term subjective expected utility of gain to specify the product of expected probability of S2 occurrence conditioned on S1 occurrence [P(Os2/Os1)] and the importance of curing S2 Similarly, the subjective expected utility of loss is the product of the probability of additional side effects resulting from using the dual-treatment option [P(SEA)] and the importance of avoiding potential loss Figure 1a presents our conceptual model, and flowchart in Figure 1b depicts the reasoning and inferences we propose that underlie participants’ decisions OVERVIEW OF THE STUDIES In each study, we manipulated the causal model, we measured drug preference, and we measured the subjective expected utility of gain and loss associated with the second treatment offered (S2 treatment) To manipulate participants’ causal model, we changed S1 across common-causes and different-causes conditions while holding S2 constant For example, in Study 1, the single-treatment drug claimed to Figure a) The proposed conceptual model of how the causal model influences choice between the single and the dual-treatment drugs (H1); the subjective expected utility of gain (H2a) and loss (H2b); choice between the single and the dual-treatment drugs through subjective expected utility of gain and loss (H3) b) Flowchart of the reasoner inferences and preference between the single and dual-treatment drug, depending on their beliefs regarding the source of the symptoms (CC = common cause; DC = different causes) © 2017 The Authors Journal of Behavioral Decision Making Published by John Wiley & Sons Ltd J Behav Dec Making (2017) DOI: 10.1002/bdm K Saporta-Sorozon et al treat the person’s focal symptom (S1; insomnia in the different-causes condition and vomiting in the commoncause condition), while the dual-treatment drug in both conditions claimed to treat not only the person’s focal symptom (S1) but also to treat nausea (S2) We manipulated the causal model by changing S1 and holding constant S2 because our main goal was not to determine whether participants would want to treat S1 (this was given), but rather whether they would prefer a drug that also treats an additional symptom, S2 Thus, to examine the effect of the causal model on inferences and preferences concerning treating S2, we had to keep S2 constant across conditions so that any effects we observe are due to the causal model and not differences in S2 To remind the reader that S2 is kept constant while S1 changes depending on the causal model, we label S1 as S1common under the common-cause condition and S1diff under the different-causes condition To assess the expected subjective utility of gain, we measured the perceived probability that S2 occurred conditioned on the occurrence of S1 [P(Os2/Os1)] and the importance of treating S2 (if present) To assess the expected subjective utility of loss, we measured the perceived probability that each product (A – the single-treatment drug and B – the dualtreatment drug) would produce side effects and the importance of avoiding them To probe for a dilution effect, we used the same measure used by Zhang et al (2007) We measured the belief that each drug will treat each symptom (“to what extent you believe that drug A (B) will effectively treat S1” on a scale from (not at all) to 10 (completely believe) In all studies, we focused on drug treatments for human symptoms The purpose of Study was to provide support for the idea that beliefs concerning the source of the symptoms, that is, the causal structure of the disease, drive inferences and preferences To rule out an alternative account whereby beliefs concerning the causal structure of the medicine (how the medicine works) and not the causal structure of the disease drive the effects, we manipulated only the causal model concerning the disease and held constant the mechanism by which the medicine works The purpose of Study was to rule out an alternative explanation whereby symptom covariation drives preferences We addressed this question by orthogonally manipulating the causal model and symptoms’ covariation The aim of Study was to show that the causal model drives the degree of dilution and preferences and not symptoms’ similarity; and that the causal model’s effect on preference is mediated by participants’ assessment of the potential gains and losses associated with treating S2 and not by the degree of dilution We achieved this goal by orthogonally manipulating the causal model and symptoms’ similarity In a pilot study we identified symptoms people believe result from common-cause or different-causes For each of 10 symptom pairs, 29 participants (17% males, Mage = 29, SD = 9.6) rated whether they believe the two symptoms result from different-causes ( = completely sure, = sure, = pretty sure) or a common-cause (1 = pretty sure, = sure, = completely sure) We also included this measure in each of our studies Causal Models Drives Preference Study 1: The causal model regarding the source of symptoms drives preference The primary aim of Study was to provide evidence that people’s beliefs regarding the source of the symptoms (the disease’s causal structure) drives inferences and preferences between the single versus dual-treatment drugs Based on the pilot study, the symptoms in the common-cause condition were vomiting (S1common) and nausea (S2), and in the different-causes condition were insomnia (S1diff) and nausea (S2) Note that there is a difference between what people believe regarding the source of the symptoms (the disease’s causal structure) and what they believe concerning the medicine’s mode of operation (the drug’s causal structure) People may believe that although insomnia and nausea result from different-causes, the same medicine can treat both, because the mechanism by which the medicine operates suits both To rule out the possibility that people’s belief concerning the drug’s causal structure drives the effect rather than their belief concerning the causal structure of the disease, we held constant the medicines’ mode of operation Method Participants We randomly allocated participants to either a different-causes (insomnia and nausea) or common-cause (vomits and nausea) condition Seventy-three undergraduates from a distance-learning university (men = 22.9%, Mage = 30.06, SD = 8.69) were presented with a web-based questionnaire and performed the task in return for course credit Materials and procedure Based on the pilot study findings, we chose nausea as the second symptom (S2), insomnia (S1diff) as the symptom the person has in the differentcauses condition, and vomiting (S1common) as the symptom the person has in the common-cause condition We asked participants to imagine they were students facing a month of final exams In the different-causes condition, they were told that they suffer from insomnia (S1diff) during the exam period, and therefore, they intend to take sleeping pills for the entire month In the common-cause condition, they were told that during the exam period they suffer from vomiting (S1common), and therefore, they intend to take sickness pills for the entire month We then informed participants that while browsing the internet, they came across two OTC pills In the different-causes condition, pill A’s vial includes 30 pills that prevent insomnia (S1diff), while pill B’s vial includes 30 pills that prevent insomnia (S1diff) and they prevent nausea (S2) In the common-cause condition, pill A’s vial includes 30 pills that prevent vomiting (S1common), while pill B’s vial includes 30 pills that prevent vomiting (S1common) and nausea (S2) As mentioned earlier, we held the drugs mode of operation constant across conditions In both common-cause and different-causes conditions, we informed participants that “each tablet contains 100 mg of Diphenhydramine, a substance which blocks the histamine levels which trigger hyperactivity of the various bodily systems.” First, we measured participants’ relative preference for a drug that treats only S1 versus a drug that treats both S1 © 2017 The Authors Journal of Behavioral Decision Making Published by John Wiley & Sons Ltd J Behav Dec Making (2017) DOI: 10.1002/bdm Journal of Behavioral Decision Making and S2 Depending on condition, participants read “You looked for a sleeping pill (S1diff)/ a pill that prevents vomiting (S1common).” Assuming the medications cost the same, which you prefer? (1 = definitely pill A, = to a large extent pill A, = to some extent pill A, = indifferent, = to some extent pill B, = to a large extent pill B, = definitely pill B) Then, we measured participants’ perceived probability of S2 occurrence conditioned on S1 occurrence [P(Os2/Os1)] We asked participants to assume a sample of 20 people who suffer from S1 and to indicate how many of them they thought would suffer from nausea on a scale ranging from (none) to 10 (all) We made clear that each scale point indicated an additional 10% of the sample To probe whether treating S2 would result in additional side effects, we asked participants to indicate the likelihood of side effects (1 = nearly zero to 10 = very high) resulting from using the single-treatment drug (A) and from using the dual-treatment drug (B) Finally, we measured the significance of gain (the importance of curing S2) and of loss (fear of side effects) (1 = not at all to 10 = very much) To form the measure of subjective utility of loss that we used to examine H3, we first subtracted the expected side effects resulting from using the dual-treatment drug (option B) from the expected side effects resulting from using the single-treatment drug (option A) This score could range from [the participant assessed the likelihood of side effect occurrence using option B as nearly zero (score 1), and using option A as very high (score 10)] to [the participant assessed the likelihood of side effect occurrence using option A as nearly zero (score 1) and using option B as very high (score 10)] Thus, the higher the score, the higher the expected side effect resulting from also treating S2 To probe for dilution, as measured by Zhang et al (2007), we asked participants to rate their belief that each pill effectively treats each symptom (1 = not at all to 10 = very much) Results Manipulation check Most participants rated insomnia and nausea as resulting from different-causes (73%), and vomiting and nausea as resulting from a common-cause (92%) The effect of the causal model on preference (H1) As predicted, the causal model had a significant effect on preference (F(1, 71) = 32.10; p < 000; η2p = 31; Table 1) Participants preferred the single-treatment drug more in the different-causes condition (M = 3.31, SD = 1.97), than in the common-cause condition (M = 5.71, SD = 1.64) To more directly test participants’ preferences, we divided participants into those who chose the single-treatment drug (values 1–3), those who were indifferent between the two drugs (value 4), and those who chose the dual-treatment drug (values 5–7) This classification corresponds to our response scale labels because the scale explicitly indicated that the values 1, 2, and reflect preference for the single-treatment option, while the values 5, 6, and 7, reflect preference for the dual-treatment option The association between the causal model and this preference score was significant (Phi = 56, p < 0001, chi square (2) = 23.01, p < 0001) Table The effects of the causal model (Study 1), the causal model and the degree of covariation between the symptoms (no data, low, high) (Study 2), the causal model and the degree of similarity between the symptoms (similar, dissimilar) (Study 3), on preference between the single and dual-treatment drugs (H1) Source of variance F df1,df2 η2p Causal model Causal model Sym correlations Causal model * Sym Corr Causal model Sym similarity Causal model * Sym similarity 32.10 23.30