Is social projection based on simulation or theory? Why new methods are needed for differentiating.

The literature on social cognition reports many instances of a phenomenon titled 'social projection' or 'egocentric bias'. These terms indicate egocentric predictions, i.e., an over-reliance on the self when predicting the cognition, emotion, or behavior of other people. The classic method to diagnose egocentric prediction is to establish high correlations between our own and other people's cognition, emotion, or behavior. We argue that this method is incorrect because there is a different way to come to a correlation between own and predicted states, namely, through the use of theoretical knowledge. Thus, the use of correlational measures is not sufficient to identify the source of social predictions. Based on the distinction between simulation theory and theory theory, we propose the following alternative methods for inferring prediction strategies: independent vs. juxtaposed predictions, the use of 'hot' mental processes, and the use of participants' self-reports.

Introduction

How do we come to know about the mental states (e.g., beliefs, or desires) that are existing in ourselves? One influential philosophical view suggests that we know about our own current mental states through introspection. Descartes argued that we are in doubt about everything except what is happening in our own mind. This view came under attack by Ryle (1949) and Wittgenstein (1958), who proposed a radically different view. These philosophers maintained that there is nothing like beliefs or desires in the mind and that mental terms refer to dispositions to behave in certain ways rather than to internal mental states. Similarly, Sellars (1956) argued against introspection and proposed, as an alternative, that lay people have a theory of mind that posits mental states. The present paper deals with a similar question for the topic of other minds: How do we come to know about the mental states that exist in other people? We will discuss the merits of a distinction between simulation theory and theory theory for understanding and predicting the behavior of other people. Specifically, we will show that this distinction can be used in research on social cognition in order to better explain the bases of understanding and predicting other minds.

Imagine the following experiment: You ask people to rate T-shirts of different colors on a scale from 1 to 10 according to their liking. Then you ask them to make predictions for another person, using the same scale. How can people accomplish this task? They can either project their own judgment onto others, i.e., they use their own opinion as a basis for the prediction (the strategy of using ‘self as proxy’), or they can base their predictions on folk psychological theories about other people (e.g., what colors are popular among people in general). Consider the result of the hypothetical experiment to be a mean correlation of r = +0.40 between own and predicted liking. What does this result mean? Did participants make their predictions using a theory or the self as proxy?

Inspection of the literature shows a clear tendency to interpret such findings as indicating social projection, i.e., using the self as a proxy. The logic for this interpretation is that social projection will result in a correlation between own and predicted preference. Although true in general, this interpretation faces two problems: First, the use of a theory can also lead to significant correlations, if that theory happens to produce a prediction similar to one's own judgment. Second, social projection is tested against the null hypothesis of no correlation; a proper test of social projection should be against a correlation of 1, however, given that complete social projection implies a perfect correlation. We argue that correlations by themselves do not show that predictions are based on the self, i.e., they do not show social projection. Instead of relying on correlations, alternative methods are needed to determine whether people use the self or theoretical knowledge for predictions.

Predicting others

Predicting other people's behavior is one of the essential requirements of our daily lives. Knowing what other people think or feel and what they therefore might do, makes our social lives easier and allows us to maneuver smoothly through the world that we share with others. Although we are generally successful in our predictions of others (which is evident since we are not constantly surprised by the behavior of people around us), sometimes predictions go wrong. One particular error that is attributed to people is the tendency to find others more similar to themselves than they actually are (e.g., Dawes, 1989; Gilovich, Jennings, & Jennings, 1983). People tend to believe that other people will act, think and feel as they do. This assumption is true on a very basic level, since we are all human, hardwired in a similar way, and, unless suffering from a severe mental disorder, our way of thinking and feeling is therefore similar. However, it is certainly not wise to over-apply this idea. The erroneous assumption that we can automatically derive other people's experience from our own experience has been discussed by philosophers (e.g., Gregory, 2004). Turing reminded us that, although we may perceive someone as human and therefore similar to ourselves, he could also be a computer trying to deceive us (as explicated in the famous ‘Turing Test’). The basic philosophical issue of the idea of likeness notwithstanding, social psychology has also collected evidence of overgeneralization. Indeed, people are psychologically different in many aspects—our opinions, beliefs, ideals, morals, and preferences need not be the same. At times, people realize this and do not assume that other people will behave similarly to themselves (e.g., Johnson, 1987; Karniol, Eylon, & Rish, 1997). However, research has documented many instances of over-application of similarity (e.g., the false consensus effect; Ross, Greene, & House, 1977). This phenomenon is called ‘social projection’, where the basis for predicting others (a social task) is our own judgment projected onto others—a strategy that can be described as using oneself as proxy. However, there is a real alternative to projection: The use of purely theoretical information that can also be the basis for predicting other people. For instance, one could use specific knowledge about a good friend, or knowledge about how other people in general would react. These two poles—projection and theory—contrast in the amount to which the self is involved: High in the case of projection, not necessarily in the case of a theory. In philosophy of mind, these two concepts are known as simulation theory (Gordon, 1986; Heal, 1986) and theory theory (Churchland, 1981; Fodor, 1987). Even though concepts related to simulation and theory are used in social psychology (Ames, 2004a, 2004b; Epley, Keysar, Van Boven, & Gilovich, 2004), we argue that current theorizing does not adequately distinguish between theory and simulation. Before we go into detail about the methodological problems, we will introduce simulation theory and theory theory.

Simulation theory and theory theory

Simulation theory was introduced by philosophers (Gordon, 1986; Heal, 1986) as an explanation for our ability to understand and predict other people. It holds that we use ourselves as a model for understanding and predicting other people, in a process that does not require any knowledge about the mental makeup of other people. Simulation theory claims that the same sequence of mental processes that can be found in another person also takes place in us, triggered by the imagined situation. This replication of mental processes (Heal, 1986) is the core assumption of simulation theory. Perner and Brandl (2009, p. 438) distinguish three stages of simulation:

Identification: We pretend to be in the other person’s situation.

Replication: Our own mental processes work in the same way as those of the other person.

Interpretation: The mental states that were involved in the replication stage are extracted and are projected onto the other person.

Simulation is a simple method for using the self as a tool for predicting other people, and it has been shown that it is used as a strategy for predictions (Kühberger, Kogler, Hug, & Mösl, 2006; Perner & Kühberger, 2003).

The counterpart to the simulation view is theory theory (Churchland, 1981; Fodor, 1987), which states that we possess folk psychological theories about mental processes. Other people's behavior is accordingly explained and predicted by ascribing beliefs, desires and other mental states to them. Explanation and prediction of other people's behavior exploit an internally represented knowledge structure of human behavior, i.e., a ‘folk psychology’ (Stich & Nichols, 1992). Folk psychology, in its broadest sense, contains information about how the human mind works and how that makes people act in different situations. We possess concepts about different mental states, their relationships among each other and about the relationship between mental states and behavior (e.g., that people act to satisfy their desires according to their beliefs; Apperly, 2008). Folk psychology helps us to reconstruct what goes on in other minds and allows us to describe, explain and predict other people’s experiences and behavior (Saxe, 2006). In contrast to simulation theory, we do not relive the other person's mental states, but rather exploit knowledge about other people and their mental workings. To arrive at social predictions, we use our theories of how people's mental states interact with each other to generate behavior, and we combine these theories with any additional initial theoretical knowledge that we may have about the target person (Apperly, 2008). Although most theory theorists consider our ability to understand and predict other people as realized through commonsense psychological theories, there are different versions of theory theory. For example, there is debate about what exactly constitutes these theories—while some see a ‘theory’ in a network of different beliefs, assumptions, and platitudes about the mental domain (Stich & Nichols, 1998), others claim that folk psychology should show the characteristics of a real scientific theory (Gopnik & Meltzoff, 1997). Further, it has been suggested that there are innate modules in our brain that are specifically designed for the understanding of other minds (e.g., Leslie, 1987). Nevertheless, the basic idea of all versions of theory theory is that we employ theoretical knowledge to arrive at social predictions.

It is also reasonable to assume that we sometimes use both simulation and theory, as hybrid models suggest (e.g., Epley et al., 2004). This integration allows adaption to different target people and situations. Theoretical information can be added before (upstream) or after (downstream) the simulation process (Stich & Nichols, 1997).

Simulation, theory, and social psychology

The basic distinction between simulation and theory has made its way into social psychology. Epley et al. (2004) integrated the concepts of simulation and theory in an anchoring and adjustment model. The model proposes that we rely heavily on the self when predicting others, but try to adjust from our perspective with theoretical information. This model can be viewed as a hybrid of simulation and theory: “Our results suggest that by adulthood, individuals' attempts at perspective taking are often something of an integration of theory and simulation” (Epley et al., 2004, p. 338). Similarly, Ames (2005) distinguishes between ‘stereotyping’ and ‘projecting’ as the major strategies that people rely on when predicting others—processes that are similar to theory and simulation. Even though those ideas are analogous to the dichotomy of theory and simulation, they do not make the original concepts redundant, since the philosophical concepts are more elaborated theoretical positions. For example, Ames (2004a, 2004b) considers ‘stereotyping’ as one of two prediction strategies, equating stereotyping with theory theory. However, stereotypes are only one way to use theoretical information. Another possibility is using specific information about other, often well-known people, which has nothing to do with stereotypes. In a social-cognitive variant of theory theory, the self-as-distinct (SAD) model (Karniol, 2003) proposes that we do not use the self as the basis for predicting other people, as self-knowledge has no special status, but is merely defined in terms of its distinctiveness from prototypical social knowledge. Rather, social predictions are based on generic representations of prototypical others. Only if the target person deviates from this default, idiosyncratic representations are used instead. Thus, there are various similar models that combine theory and simulation, which differ with respect to what is the primary source of information.

How do we determine whether theory or simulation was used? One possible method is to consider the factors that lead to either one of the two strategies. Features such as familiarity (Krienen, Tu, & Buckner, 2010; Nickerson, 1999), similarity (Ames, 2004b; Ames, Weber, & Zou, 2011; Jones, 2004b; Pollmann, Finkenauer, & van Djik, 2008) or the sense of being unique (Karniol, 2003) have been discussed. As we will show, there is a problem with most of the research investigating the theory-simulation issue, since most of the studies rely on correlations. The use of correlations, however, does not allow for unambiguous interpretation of these findings.

Ways of social projection

Although projection has different meanings in psychology (see Holmes, 1968, for a review), for the purposes of the current paper, social projection is defined as “(…) assigning a state of one's own to someone else“ (Goldman, 2006, p. 40). The assumption that other people will act, feel or decide the same way as we do is the basic idea of social projection. Katz and Allport (1931) coined this term after noticing that the more students admitted having cheated on exams, the more they expected other students having cheated as well. Different labels for this process are attributive projection (Holmes, 1968), egocentric attribution (Heider, 1958), egocentric bias (Epley et al., 2004), assumed similarity (Cronbach, 1955), and, most famously, false consensus effect (Ross et al., 1977).

Measurement of social projection

Social projection seems to occur in various kinds of situations, including predictions about personality traits (Krueger & Clement, 1994; Krueger & Zeiger, 1993), attitudes (Biernat, Manis, & Kobrynowicz, 1997), feelings (Van Boven & Loewenstein, 2003), and decisions (Ross et al., 1977; Van Boven, Loewenstein, & Dunning, 2005). The people whose reactions are being predicted also vary greatly: In some studies, specific individuals are the targets of prediction (Bender & Hastorf, 1950; Pollmann et al., 2008). These individuals vary in familiarity—sometimes, the prediction is made for a well-known other, like a spouse (Davis, Hoch, & Easton Ragsdale, 1986); in many studies it is a stranger whose reaction has to be predicted (Van Boven et al., 2005). Sometimes, a prediction about people in general is required (Clement & Krueger, 2000; Krueger & Clement, 1994; Krueger & Zeiger, 1993). Participants also have to estimate reactions of groups of people (de la Haye, 2000; Krueger & Stanke, 2001; Monin & Norton, 2003), with the predictor being part of the group that has to be predicted (Biernat et al., 1997; Dawes & Mulford, 1996), or not (Clement & Krueger, 2002; Monin & Norton, 2003).

When investigating social projection, there are two basic methodological approaches. One approach is the false consensus paradigm (e.g., Ross et al., 1977), in which participants indicate their own answer to a dichotomous question (yes/no; agree/disagree) and are then asked what percentage of other people would respond in one way or another. If people who endorse an item think that most other people would do the same, the conclusion is that they have fallen prey to the bias. The other approach is the assumed similarity paradigm (e.g., Cronbach, 1955) which requires participants to indicate their own answer along a particular dimension and then to estimate the position of another (individual) person. Thus, the false consensus paradigm focuses on frequencies (i.e., estimation of percentages) in a group, while the assumed similarity paradigm focuses on central tendencies and the prediction of individual people (Jones, 2004a).

The false consensus effect

False consensus (Ross et al., 1977) means that we overestimate the similarity between ourselves and others and therefore think that our own beliefs, opinions, and decisions are very common—in other words, people tend to project their own subjective experience and action tendencies onto others. In the paradigmatic study (Ross et al., 1977), students were asked whether they would wear a sandwich board around campus that read ‘Repent’. The students who agreed to do this thought that most other students would do the same, but those who refused stated most would not do that. Countless studies on this phenomenon have been conducted since, and while some of them considered it as being due to a false consensus (see also Mullen et al., 1985, for a meta-analysis), others used the more neutral term social projection, underscoring its source, (i.e., that people rely on the similarity between themselves and others) rather than its evaluation (i.e., that they overestimate similarity, which leads to a bias).

Over the years, research has come up with several explanations for the false consensus effect (cf. Marks & Miller, 1987). Most of them were interpreting the phenomenon as some kind of bias, resulting from selective exposure (Ross et al., 1977), cognitive availability (Marks & Miller, 1985, 1987; Ross et al., 1977); causal attribution (Gilovich et al., 1983; Ross, 1977), motivational factors (Marks & Miller, 1987; Ross et al., 1977), or differential construal (Gilovich, 1990; see also Marks & Miller, 1985, for a detailed overview). However, even within the bias view, it has been acknowledged that sometimes the only information available about people's reactions or behavior is the information that the individual holds about himself, therefore “many times one's own reactions may be the best basis for predicting others” (Davis et al., 1986, p. 35). Hence, the strategy of using the self to make predictions about others is reasonable (Hoch, 1987), but adjustments are needed to avoid egocentric errors (Epley et al., 2004).

Taken together, studies about social projection vary greatly by investigating different target behavior (e.g., personality, opinions, feelings, decisions), different measures of projection (consensus vs. similarity), and different targets of prediction, either groups (participant as member or non-member) or individuals (strangers or well-known others). For illustrative purposes, a sample of the results is presented in Table 1, organized by target of prediction, topic of prediction, and paradigm.

Table 1

Sample of social projection studies with main characteristics and findings.

Study	Target of prediction	Topic of prediction	Paradigm	Resulta
Bender and Hastorf (1950)	Class members	Personality characteristics	AS	r = 0.55, r = 0.71
Clement and Krueger (2000)	Majority of people	Trait adjectives	AS	r = 0.24
Dawes and Mulford (1996)	Students	CPI	AS	r = 0.59
de la Haye (2000)	Students	Personal statements	FC	r = 0.30, r = 0.34
Hoch (1987)	Average consumer, peer, spouse	Attitude statements	FC/AS	r = 0.26, r = 0.47, r = 0.52
Krueger and Clement (1994)	General population	MMPI	FC	r = 0.35
Krueger and Zeiger (1993)	General population (men & women)	MMPI	FC	r = 0.45, r = 0.47
Monin and Norton (2003)	Students	Questionnaire	AS	r = 0.28, r = 0.34
Pollmann et al. (2008)	Similar person (average student, average Dutch person)	Scenarios	AS	r = 0.74, r = 0.94

Note. AS = Assumed similarity method, FC = False consensus method, CPI = California Psychological Inventory, MMPI = Minnesota Multiphasic Personality Inventory.

Correlations between own judgment and the prediction for others. For multiple results, projection was measured on various occasions, or for different target people.

The original notion of social projection probably has implied a process similar to simulation in that the individual's own states are projected onto the other person, as opposed to just ascribing knowledge (which has nothing to do with the individual's own position). Current methods in the field of social projection research, however, fail to identify whether social projection stems from simulation or from theory. If the phenomenon of ‘social projection’ is the result of theory use, then it is not a bias people fall prey to because they overestimate the similarity between the self and others, but use of knowledge about humans in general that does not necessarily have anything to do with the self. Therefore, proper understanding of this phenomenon is impossible without clarification of the basis for predictions.

The interpretation of correlation: measurement issues

The assumption that other people's actions, opinions, or thoughts are related to one's own translates into a significant positive correlation between these two measures. Thus, the royal road for showing social projection is to measure the correlation between one's own position and the prediction of other people's position (Dawes, 1989; Hoch, 1987; Krueger, 1998; Krueger & Clement, 1994). This seems reasonable, but has several problems: First, how strong should a correlation be to indicate social projection? Indeed, as can be seen in Table 1, the size of the reported relationship varies greatly, and any significant positive correlation between a predictor's opinion and prediction, no matter how small, can be interpreted as social projection. Moreover, although the range of correlations is wide, the interpretation is always the same: Social projection. However, unless we can account for the error, a significant difference between correlations calls for an explanation. As it stands now, any significant correlation means the same thing—if it is only different from zero. What is needed is an error theory, telling us why correlations are high or not, or at least an idea on how big a correlation should be in order to indicate social projection.

In an attempt to get an idea about the size of error in a typical prediction task, we had participants predict themselves, since predicting oneself is a relatively error-free projection task. Thus, in order to gauge the consistency of self-projection, we asked 19 undergraduates to rate the pleasantness of stimuli (colors and surfaces) on a scale from −5 ‘very unpleasant’ to +5 ‘very pleasant’. Then, we asked one group of participants to rate the same stimuli assuming that the experiment would take place on the next day, and the other group to rate stimuli assuming that the experiment would have taken place the day before. This scenario—‘prediction’ of self in a very similar situation—is a setting in which the current self is the basis for prediction and therefore, this task constitutes projection in its purest form. The mean correlations between judgments and predictions are presented in Table 2. These correlations are clearly higher than the correlations found in most studies on social projection (see Table 1). Taking these correlations as benchmarks, we would expect that social projection is indicated by correlation coefficients around r = 0.80. What, then, is the justification for an r of 0.40 as indicating the same process, given that, statistically speaking, there is a significant difference between these correlations? It is unclear whether this variation in correlation size is only a variation around a fixed effect indicating the same process, or a variation that indicates different processes.

Table 2

Correlations between own actual judgment and ‘prediction’ for similar situation.

	Futurea	Pastb
Stimulus	r (SDr)	r (SDr)
Color	0.85 (0.60)	0.91 (0.40)
Surface	0.83 (0.50)	0.87 (0.52)

n = 9.

n = 10.

A second concern is that social projection studies test against an utterly unrealistic null hypothesis, namely, that people predict the behavior of others randomly. This translates into a zero correlation. However, it would be an utter catastrophe if people were that bad at social predictions. This is surely not the correct alternative to social projection. Rather, the alternative is that there must be another source for successful prediction. Indeed, there is such a source: That people predict on the basis of rules about behavior, i.e., on the basis of folk psychology. This has nothing to do with social projection, but it can lead to results similar to those described in studies. A correct theory would also lead to a high correlation, even if no projection was going on. Take the following example: Imagine being asked what you would like to eat: Carrion or an apple? Surely, your answer would be that you would like to have the apple. Then you are asked what another person would prefer. Your correct prediction is that the other person would also prefer the apple. If we ran an experiment, we would end up with a perfect positive correlation between own and predicted choice. So does this imply that this is a case of social projection? No, since the predictions are based on general knowledge, namely that humans normally do not eat carrion. Thus, one general rule is applied to both the individual's own behavior and the behavior of others. No social projection is necessary at all. If the majority of people prefers apples over carrion and a majority also predicts that people prefer apples over carrion, this is simply a correct prediction which translates into a positive correlation between own and predicted behavior. This example shows that a correlation between a judgment and a prediction is not enough to constitute a case of social projection. A theory can lead to the exact same result. The use of correlational values alone—or related measures like discrepancy between our own judgment and prediction (Tamir & Mitchell, 2010)—is not informative, because an alternative to social projection is always possible, namely, theoretical knowledge. Thus, a decision between these alternatives cannot be based on a correlation, but instead, we need to rely on other methods that are based on the difference between simulation and theory.

Methods for deciding between simulation and theory

A new approach is needed to decide between simulation and theory use. As described above, correlational values are not sufficient, since theory as well as simulation can lead to very high and also to low correlations. However, alternative approaches are possible; some of these approaches have been already established, others need to be explored.

Juxtaposed vs. independent predictions

Perner, Gschaider, Kühberger, and Schrofner (1999) have developed a method that focuses on contrasting a pair of situations that lead to a target effect, for example, the framing effect: This effect describes the tendency for risk-aversion if a problem is framed in terms of gains, and the tendency for risk-seeking if the same problem is framed in terms of losses, for example, if 600 people are at stake, saving 200 (a gain of 200) is equivalent to having 400 die (a loss of 400) (Tversky & Kahneman, 1981). Participants are asked to make predictions under two different conditions. One group of participants predicts for both situations (juxtaposed predictions), and two other groups predict for only one of the two situations (independent predictions). In the context of the framing effect, this means that under juxtaposed prediction, participants make a prediction for both the positively and the negatively framed situation, but under independent prediction, participants make a prediction only for either the positively or the negatively framed condition. The idea is that prediction by simulation is disturbed by juxtaposing the two situations, as one has to set aside relevant information from the other situation, which is hard to do. If predictions are made by simulation, then the predictions made under independent presentation of the situations should be more accurate than the predictions made under juxtaposed presentation. In contrast, juxtaposition of the situations highlights the relevant difference between the two situations and makes application of a theory easier. In short, juxtaposition makes theory application easy and simulation difficult, and independent presentation is optimal for simulation, but is intransparent with respect to an adequate theory. Hence, if predictions made under juxtaposed presentation are more accurate than predictions made under independent presentation, this indicates use of theory, and when predictions are more accurate under independent presentation, this indicates use of simulation. With this method, a series of consistent findings has been reported indicating that participants do use simulation for some predictions and a theory for others (Perner et al., 1999; Perner & Kühberger, 2003). For example, the framing effect is best predicted if situations are presented independently, indicating simulation (Perner & Kühberger, 2002).

This method is useful, but not foolproof. First, a reliable target effect is necessary. Second, in order to decide whether simulation or theory is used, the juxtaposed and independent predictions need to lead to different predictions. Third, since the differences in presentation mode also have an influence on other features beyond simulation or theory, the findings could have an alternative interpretation. Thus, only a series of studies resulting in a consistent pattern of findings can be meaningfully interpreted.

Simulation is hot

If I am predicting the feelings of another person in a specific situation on the basis of simulation, I actually have to feel the same feelings as the target person—the process is the same in me as it is in the other person (Goldman, 1989). This quality of simulation—that it is ‘hot’—can be used for deciding whether people use simulation. Hot simulation can be assessed by asking people about their emotions or by taking physiological measures that indicate their current emotions: blushing, sweating palms, etc. (LeBar, 2001). For example, consider the ‘hot-cold empathy gap’ (Loewenstein, 1996). This is the finding of an interference of people's current (visceral) states with their predictions; we misjudge how much we are (or will be) influenced by our visceral states. Thus, when predicting other people who are in different visceral states, we have a hard time ignoring our current situation and arousal; rather, we tend to project those states onto others, such that our own emotions (Van Boven et al., 2005), bodily drives (Van Boven & Loewenstein, 2003), or valuation of objects (Van Boven, Dunning, & Loewenstein, 2000) contaminate our predictions of others. This influence of one’s own condition on the prediction indicates simulation, as this contamination of the social prediction by one’s own states cannot be easily explained by theory. Therefore, measuring whether current states have been projected onto social predictions is a useful method for deciding between theory and simulation.

Explicit verbal report

One obvious way to determine whether people use simulation or theory is by directly asking them. However, Nisbett and Wilson (1977) argued in their influential article that people cannot properly report their own mental processes. If this was true, participants' answers and their actual behavior may differ. However, Nisbett and Wilson's claim has been criticized for being too strong, as it seems that people are able to report on mental processes in many situations (Ericsson & Simon, 1980; Smith & Miller, 1978). So despite that criticism, participants' self-reports can be a valid source of evidence, especially if they are combined with corroborating methods. For instance, Krienen et al. (2010) asked participants whether they based social predictions on specific information or on the self, and showed that participants reported different strategies depending on the familiarity of the target person.

In another study (Bazinger & Kühberger, submitted for publication), participants made a simple social prediction (color or food preference). After performing the prediction, participants were asked how they had arrived at this prediction, using items that probed for subprocesses of theory and simulation. Specifically, they were asked about use of general knowledge about people, that is, a ‘folk psychology’, which contains (naïve) knowledge about how psychological processes typically work (Gordon & Cruz, 2002), as well as about use of specific knowledge about the target person (i.e., idiosnycratic information about another person, which might be applicable to this specific person only and that cannot be generalized to other people). The latter information is a form of evidence since it stems from directly interacting with one specific individual, from observation, or from being told by other people (Frith & Frith, 2006). In addition, participants indicated whether they had used what according to Perner and Brandl's (2009) model constitutes simulation: Imagining the situation of the other person (‘perspective taking’), and simulating the other person with one's own mind (replication, or using ‘self as proxy’), and ascribing the outcome of one’s own mental process to the other person.

The results showed that participants had a pretty good intuition about the strategies they had used for arriving at their predictions. Thus, the verbal report can be a valuable method for distinguishing whether simulation or theory was used (Bazinger & Kühberger, submitted for publication). Verbal self-reports can provide information about the prediction process that would be overlooked with conventional methods (e.g., correlations). For example, participants reported using both simulation and theory in their predictions, which would not have been obvious when just looking at correlations. Further, different aspects of simulation and theory were used, depending on characteristics of the other person (e.g., general knowledge was used for unknown others, but specific knowledge for well-known others), which is also not evident when focusing on correlations alone. In addition, participants' self-reports were sound, for example, the more they reported using ‘self as proxy’ as prediction strategy, the higher was the likelihood that the prediction would match their own choice. Therefore, this method can differentiate between simulation and theory, despite the obvious drawbacks of self-reports, and can provide useful information about the prediction process that would otherwise be overlooked.

Conclusion

The strategy of assuming that other people would do the same as oneself seems to be a good idea in many situations (Ames et al., 2011), and therefore, using the self as a proxy (i.e., social projection) may frequently be adopted as a prediction strategy. It seems, however, that there are cases that include theoretical information. Thus, the term ‘projection’ should be used carefully, as it may not always be the case that projection onto others really takes place. Calculating correlations between own position and predicted position has been used as the royal road to infer social projection. This is a problem, however, since correlations varying widely in size have been interpreted as indicating the same process. A strong correlation between people's own answers and their predictions for others is probably necessary for establishing social projection, but not sufficient, since theory can also lead to strong correlations. On the other hand, a weak correlation between these two measures raises serious doubts about whether simulation is really existent in the case at hand, or whether it is mainly the use of different theories that leads to this low correlation (as theory can account for both high and low correlations).

We suggest that future research should concentrate on alternative approaches for deciding between simulation and theory. The method by Perner et al. (1999) is a good start and has been successfully applied. Another alternative approach is to exploit one of the essential characteristics of simulation theory: That the same process takes place in the target person as well as in the predicting person. Stages of this process can be investigated using self-reports—first attempts have already shown that this method is indeed valid.