Iconic realism or representational disregard? How young children and adults reason about pictures and objects

Abstract

When reasoning about a representation (e.g., a toy lion), children often engage in “iconic realism,” whereby representations are reported to have properties of their real-life referents. The present studies examined an inverse difficulty that we dub “representational blindness”: overlooking (i.e., being ‘blind’ to) a representation’s objective, non-symbolic features. In three experiments (N = 302), children (3–6 years) and adults saw a series of representations (pictures and toys) and were tested on how often they endorsed a property that was true of the real-world referent (e.g., reporting that a toy lion is dangerous; iconic realism) or rejected a property that was true of the representation (e.g., denying that a toy elephant can be lifted with one hand; representational blindness). We found that representational blindness and realism were separable tendencies. Children (and to a lesser extent, adults) displayed both, but at different rates for pictures than for toys. We conclude that children’s reasoning about representations includes a bias to overlook the features of the representation itself. Further, although pictures and toys are both representations, they provoke ontologically distinct interpretations. We discuss the implications of these results for a variety of important conceptual tasks, including learning to read, draw, or objectively evaluate scientific evidence.

In 1929, Belgian surrealist René Magritte composed a painting of a single pipe above the words “Ceci n’est pas une pipe” [“This is not a pipe”]. Although this painting elicited much confusion from the public, Magritte explained that his painting was clearly not a pipe—it was simply a representation of a pipe. He noted that you could not “stuff” this picture of a pipe in the way that you could stuff a real pipe, and thus he felt that his statement “This is not a pipe” was completely earnest. He titled this painting “The Treachery of Images” (Magritte, 1929).

Magritte’s painting sheds light on the dual nature of representations, which both “stand in” for reality (depicting real things in the world) and are objects in their own right (e.g., the canvas has a certain size, weight, and substance). Viewers of Magritte’s famous painting often fall prey to a dual-representation problem, as they interpret the written statement on the painting as referring to the item depicted in the painting, rather than to the painting itself.

The dual nature of representations is particularly challenging for young children, who at times seem to confuse a representation or symbol with its referent.¹ This is sometimes referred to as iconic realism (Beilin & Pearlman, 1991), symbol-referent confusion (Claxton, 2011), or a dual-representation problem (Uttal, O’Doherty, Newland, Hand, & DeLoache, 2009). For example, when 9-month-olds see photos in a book, they try to grasp and manipulate them as if they were the real three-dimensional objects (DeLoache, Pierroutsakos, Uttal, Rosengren, & Gottlieb, 1998). Similarly, toddlers attempt to use toy objects as if they were full-size equivalents (e.g., trying to sit in a toy chair; DeLoache, Uttal, & Rosengren, 2004).

This difficulty can also be seen in judgments made by preschool-aged children. For example, children 3.5 years of age tend to say that if a TV were turned upside-down or shaken, this would affect the televised image (e.g., milk would spill) (Claxton, 2011; Flavell, Flavell, Green, & Korfmacher, 1990). Furthermore, children 3–5 years of age predict that a picture will change if its referent changes (e.g., a sticker placed on a doll after it is photographed will then appear in the photo; Robinson, Nye, & Thomas, 1994; Zaitchik, 1990). In other words, young children seem to assume that if an item in the real world undergoes some change then this will be reflected in a representation of the object, and that if a representation undergoes change then the same change will be reflected in the real-world object it represents (Donnelly, Gjersoe, & Hood, 2013; Robinson et al., 1994; Wimmer, Robinson, Koenig, & Corder, 2014).

A similar confusion is found in children’s reasoning about linguistic representations (words). Children make nominal realism errors, treating category labels as interchangeable with their referents (e.g., judging that the word ‘nickel’ can buy a piece of candy; Markman, 1976), or judging the form of a word as non-arbitrary (e.g., cats cannot be called ‘dogs’, even if all speakers agree; Rosenblum & Pinker, 1983). Similarly, they report that a written name has changed if its referent has changed (Thomas, Jolley, Robinson, & Champion, 1999).

Representational blindness: A hypothesis

Although the errors summarized above are well-documented, an open question is how they are to be explained. Prior research has focused primarily on how children over-attribute properties of the referent to the representation (e.g., saying that a picture of ice cream is cold)—a ‘realism’ error. We propose that children may face an additional, separate challenge when reasoning about representations, namely, they may also under-appreciate the objective, non-symbolic properties of the representation itself (e.g., rejecting that a picture of ice cream is flat). A full appreciation of a representation entails not only understanding what it stands for, but also being able to step back from its symbolic meaning and objectively focus on the physical attributes of the representation itself (e.g., a picture’s size, texture, or inertness). Although such properties are perceptually evident, we hypothesize that they may at times be overlooked in the face of a representation’s meaning. We call this tendency ‘representational blindness’. In effect, knowing what a picture represents may lead children to overlook (i.e., be ‘blind’ to) the objective features that they can plainly see. The distinction between realism and representational blindness is analogous to the distinction between intellectual realism and phenomenism in the appearance/reality literature (Flavell, Flavell, & Green, 1983). To date, however, little is known regarding children’s ability to overlook a representation’s meaning and consider its physical properties.

In positing representational blindness, we do not suggest that people literally confuse representations with reality. Viewers of The Treachery of Images certainly understood that they were not viewing a real pipe, for example. Even infants are capable of grasping the symbolic status of representations (Callaghan & Corbit, 2015; Ganea, Pickard, & DeLoache, 2008; Geraghty, Waxman, & Gelman, 2014). Research has also shown that, under favorable testing conditions, young preschoolers can focus on representations as distinct from reality (Preissler & Bloom, 2007; Robinson et al., 1994). For example, even those 3-year-olds who treated the televised image as if it were real (e.g., that milk would spill if the TV set were turned upside down) successfully reported that the TV image was not real, and when given a comparison between a real object and a televised image, endorsed the real object as the one that would spill or make noise (Claxton, 2011).

Rather, we suggest that when considering representations, there can be slippage between levels of analysis, which may manifest as overlooking the vehicle for expressing meaning (the representation itself) and focusing directly on the symbolic interpretation (the referent). A slippage between a representation’s symbolic content and its physical manifestation is in fact built into the meaning of words such as “book” (“The book is exciting” vs. “The book is heavy”) (Srinivasan & Rabagliati, 2015), and a slippage between reference to a specific individual vs. reference to an indefinite member of the kind is built into de re/de dicto confusions (e.g., Plantinga, 1969).

The present studies: Three goals

The present studies have three primary goals: (1) to examine whether children display representational blindness in addition to iconic realism, (2) if so, to determine whether these tendencies differ when reasoning about pictures versus toy objects, and (3) to examine the developmental course of these tendencies.

It is first important to operationalize iconic realism and representational blindness as two distinct constructs. Iconic realism entails endorsing a feature that is true of the real-world referent but not the representation (e.g., reporting that a toy pig rolls in the mud). In contrast, representational blindness entails rejecting a feature that is true of the representation but not of the real-world referent (e.g., reporting that a small toy pig cannot fit in one’s hand).

From an informational standpoint, iconic realism and representational blindness may be considered two sides of the same coin: thinking about a representation in terms of what it stands for (iconic realism) would appear to be equivalent to failing to think about a representation in terms of its non-representational physical qualities (representational blindness). In other words, if we assume that a person has a consistent construal of a representation, then someone who displays one tendency should display the other tendency as well. Indeed, prior research has not separated these two phenomena, instead treating either tendency as evidence of a unified difficulty (iconic realism, symbol-referent confusion, or a dual-representation problem).

From a psychological standpoint, however, these tendencies need not be equivalent, but instead can be examined as distinct construals. For example, a person could overly weight what a representation stands for (displaying realism) but nonetheless still be capable of reasoning about the properties of the representation itself (displaying representational awareness) (e.g., reporting that a toy pig rolls in the mud but also that it fits in one’s hand). The possibility that a person may at times show one tendency but not the other would fit with a growing literature indicating that people’s concepts are not always internally consistent or coherent (Shtulman & Lombrozo, 2016). It is thus an open question as to whether we see evidence of representational blindness in children’s (or adults’) reasoning, and how this compares to realism.

In addition to assessing representational blindness vis-à-vis iconic realism, we assessed whether these tendencies differ for different types of representation (pictures vs. toy objects). Iconic realism and representational blindness may be domain-general tendencies, whether representations are pictures or objects. Yet there is reason to suspect that representational blindness, if obtained, may be more pronounced when reasoning about pictures than objects. Pictures have less physicality (they are flat, with fewer affordances) and serve only to represent. In contrast, toys are fully 3-dimensional, and are interacted with as objects in their own right (e.g., their substance, size, weight, are relevant to how we interact with them).

Indeed, there is an intriguing contrast in the literature in the symbolic reasoning errors that pictures and objects elicit. As noted above, when reasoning about pictures, children tend to focus on what they represent, even when doing so is in error. In contrast, when reasoning about three-dimensional objects such as scale models or manipulables, children tend to focus on them as objects in their own right (Uttal et al., 2009). For example, below about 3 years of age, children fail to appreciate that a scale model of a room can represent a full-size room (DeLoache, 1987). Moreover, children have greater difficulty appreciating representations as symbolic when those representations are objects versus pictures (DeLoache, 1991; DeLoache & Marzolf, 1992).

Given the salience of objects as entities in their own right, we hypothesized that toys may elicit less representational blindness than pictures. To our knowledge, however, no prior work has included both pictures and objects, while simultaneously measuring both iconic realism and representational blindness.

The present studies: Overview

We conducted three studies in which participants were presented with a series of items, and asked questions that focused either on a feature of the representation (picture or toy) or on the referent (what the picture or toy represents). For example, on one item participants saw either a toy pig or a realistic color drawing of a pig matched in appearance to the toy, and heard a question about a feature true only of the referent (e.g., “Does this roll around in the mud?”) or a question about a feature true only of the representation (e.g., “Can this fit in your hand?”). Clearly, the representation does not roll around in the mud (neither the picture nor the toy); rolling around in the mud is a feature linked to the referent. Conversely, real-life pigs cannot fit in someone’s hand, but both the toy and the picture can. Therefore, these questions assess whether children construe the representation in terms of its symbolic meaning (saying ‘yes’ to the referent question), in terms of its non-symbolic physical properties (saying ‘yes’ to the representation question), both, or neither. All studies also included control questions about properties true of both the representation and the referent (e.g., “Is this pink?”) and false of both the representation and the referent (e.g., “Does this go meow?”), to ensure that any patterns obtained cannot be due to a tendency simply to say ‘yes’ or ‘no’.

In Studies 1 and 2, the questions were worded ambiguously (e.g., “Does this roll around in the mud?”), to examine construals when language was minimally directive (analogous to Magritte’s painting). Referential underspecification is a useful tool for determining the conceptual inferences that children bring to a situation (Jara-Ettinger, Floyd, Huey, Tenenbaum, & Schulz, 2019). In contrast, the questions in Study 3 focused explicitly on the representation (e.g., “Does this toy roll around in the mud?” or “Does this picture roll around in the mud?”), to provide a stronger test.

To obtain data on development, participants included young children as well as adults. We were particularly interested in whether realism and representational blindness would show different developmental patterns. Child participants ranged in age from 3–6 years, as these ages overlap with those in prior studies of children’s understanding of representations (e.g., Claxton, 2011; Donnelly et al., 2013; Robinson et al., 1994; Wimmer et al., 2014), and the task was not appropriate for children younger than 3. By including adult participants, we could test whether realism errors or representational blindness are overcome with age, or whether instead they reflect something more foundational about how people reason about representations, with traces into adulthood. For example, adults as well as children often erroneously assume that words have a good ‘fit’ to their meaning (Sutherland & Cimpian, 2015).

STUDY 1

Method

Participants

Thirty-two children (19 female and 13 male) and thirty-two adults (29 female and 3 male) participated in the main study. Children ranged from 4.13–5.39 years of age (M = 4.61, SD = .28) and adults ranged from 17.89–22.59 years of age (M = 18.47, SD = .96). Four additional children were excluded from the data analysis for responding incorrectly to more than one of the control questions in the main study (see Procedure). In addition, 36 adults and 19 children participated in a pretest of the materials (see below).

Children were recruited from preschools in a midwest college town, and adults were undergraduate introductory psychology students at the University in the same town, receiving credit for their participation. Children were tested individually at their local preschool, either in a separate room or at an isolated table in their classroom. Adults were tested individually in a laboratory space at the University. Children sampled were predominantly White; adult demographics were: 16 White, 10 Asian, 3 African American/Black, and 3 Hispanic/Latino.

Materials

There were 16 item sets (8 foods, 8 animals), previously used in Ware, Gelman, and Kleinberg (2013). Each item set consisted of a toy object and a corresponding 2-dimensional color picture depicting the toy (see Figure 1). The toys were plastic; the pictures were 3.5×2.5 inches and laminated in clear plastic. Additional toy shapes, pictures of shapes, and Duplo blocks were used for a warm-up and post-test (see Procedure).

Figure 1.

Study 1, sample picture and toy object.

We created four test questions for each item set, illustrated here for the pig item: true of the referent only (“Does this roll around in the mud?”), true of the representation only (“Can this fit in your hand?”), true of both the referent and the representation (“Is this pink?”), and true of neither (“Does this go ‘meow’?”). There were four versions of the task, each including only one question about each item set (e.g., a particular version included only one of the four questions about the pig), but across task versions, every item set appeared with all four types of questions. Each version included 16 questions – four questions for each of the four types, split equally between pictures and toys. Specifically, each version included two of each of the following questions: representation-only (picture), representation-only (toy), referent-only (picture), referent-only (toy), both (picture), both (toy), neither (picture), neither (toy). For each of the question types listed above, one item was an animal and one item was a piece of food.

Adult pretest.

A separate set of 36 adults participated in a pretest to validate the test questions (at least 12 adults per test question). The pretest questions explicitly referred to either the referent (e.g., “Is a real watermelon juicy?”) or the representation (e.g., “Is this picture juicy?”) (modeled on Beilin & Pearman, 1991). Each test question was asked twice, once for the referent and once for the representation. Half the participants were asked about the toys and half were asked about the pictures; all were asked about the referents. On average, adults answered questions about the items selected for use in the two studies correctly 97.63% of the time.

Child pretest.

Nineteen children (age range 3.30–5.07; M age = 4.29) participated in a pretest to ensure that children of this age understood that the representations were not real. Each child was shown each of the 16 representations (either pictures only or toys only), as well as 4 real control items (e.g., a pencil), and were asked whether each item was real (e.g., “Is this a real lion?” or “Is this a real pencil?”). Children answered ‘yes’ 9% of the time for toys, 13% of the time for pictures, and 74% of the time for the control items. This indicates that the children understood that the toys and pictures used in these studies were not the real-world referents.

Design

The study had a 2 (representation type: pictures, objects) × 2 (question type: true of representation only, true of referent only) × 2 (age group: children, adults) design. (Although participants received four types of questions, two types (true of both representation and referent, true of neither representation or referent) were intended as controls and not included in the analyses.) Representation type and question type were within-subjects factors and age group was a between-subjects factors.

Procedure

Participants were randomly assigned to one of four task versions (see Materials, above), and one of two block orders: pictures-first or objects-first. At the start of each block, participants received a warm-up to give them experience answering both “yes” and “no” to questions regarding that type of representation (object or picture). For example, prior to the object block, participants saw a blue ball and were asked “Does this have blue on it?”, and a yellow triangle block and were asked “Is this a star?”, in counterbalanced order. If participants responded incorrectly to one of the warm-up questions, they were given a third warm-up question.

Following the warm-up, participants were asked 16 questions total, in two blocks of eight questions each (pictures and objects). Each block consisted of two instances each of four different types of questions: True Only of the Representation, True Only of the Referent, True of Both, and True of Neither (see Materials for more detail). The latter two types of questions were used as controls; participants (n = 4) who responded incorrectly to more than one of these questions were excluded. After the experimenter placed an item in front of the participant, she withdrew her hand before asking the test question. If a participant asked the experimenter what the question was referring to, the experimenter explained that she could not clarify, and repeated the question. Participants engaged in a two-minute break between blocks, in which children played with a set of Duplo blocks, and adults conversed with the experimenter.

Post-test.

After the last block of questions, participants were asked to label the colors red, blue, green, and yellow, to test for color-blindness. Only one participant responded incorrectly to any of these items; they also failed more than one control question and so were excluded from the data set.

Results

We first examined performance on the control questions, separately for pictures and objects and separately for children and adults. On True of Both questions, “yes” responses ranged from 94–97%; on True of Neither questions, “yes” responses ranged from 0–3%. These results indicate that participants appropriately attended to the task, and any patterns obtained cannot simply be due to a tendency to say ‘yes’ or ‘no’.

Our primary analyses focused on the yes/no responses to the two key question types of interest: True of Representation only, and True of Referent only. We submitted adults’ and children’s responses to separate logistic regression models using the glmer command in the lme4 package in R (Bates, 2007; see top of Table 1 for the coefficients of the models). The model for adults included representation type (picture = 0, object = 1; within subject), question type (true of representation = 0, true of referent = 1; within subject), and their interactions as predictors. The model for children additionally included age (continuous, mean-centered). We also included random effects for participant and item in both models.

Table 1.

Logistic mixed-effects regression models predicting responses in Studies 1, 2, and 3

Study 1 - Adults	Fixed Effects	Estimate	SE	p-value	95% CI
	Intercept	−1.43	0.34	< .001	−2.15	−0.80
	Representation Type	2.11	0.42	< .001	1.30	2.98
	Question Type	3.19	0.49	< .001	2.28	4.20
	Representation Type * Question Type	−3.04	0.62	< .001	−4.29	−1.85
	Random Effects		SD
	Participant	Intercept	0.00
	Item	Intercept	0.46
Study 1 - Children	Fixed Effects	Estimate	SE	p-value	95% CI
	Intercept	−2.44	0.55	< .001	−3.69	−1.49
	Representation Type	2.02	0.54	< .001	1.02	3.18
	Question Type	6.43	1.12	< .001	4.57	9.10
	Age	0.74	1.54	.63	−2.37	3.89
	Representation Type * Question Type	−2.87	1.15	.01	−5.41	−0.73
	Representation Type * Age	−0.62	1.78	.73	−4.16	2.95
	Question Type * Age	−4.42	2.93	.13	−10.55	1.21
	Representation Type * Question Type * Age	−1.18	3.67	.75	−8.75	5.98
	Random Effects		SD
	Participant	Intercept	0.98
	Item	Intercept	0.71
Study 2 - Adults	Fixed Effects	Estimate	SE	p-value	95% CI
	Intercept	−0.60	0.29	.04	−1.20	−0.05
	Representation Type	0.21	0.37	.58	−0.52	0.94
	Question Type	2.62	0.47	< .001	1.74	3.61
	Representation Type * Question Type	−1.08	0.61	.08	−2.30	0.10
	Random Effects		SD
	Participant	Intercept	0.00
	Item	Intercept	0.43
Study 2 - Children	Fixed Effects	Estimate	SE	p-value	95% CI
	Intercept	−1.67	0.30	< .001	−2.31	−1.10
	Representation Type	0.81	0.32	.01	0.18	1.46
	Question Type	3.85	0.42	< .001	3.08	4.72
	Age	−0.32	0.24	.19	−0.80	0.15
	Representation Type * Question Type	−0.91	0.52	.08	−1.93	0.11
	Representation Type * Age	−0.27	0.32	.39	−0.89	0.35
	Question Type * Age	0.16	0.36	.67	−0.56	0.87
	Representation Type * Question Type * Age	1.00	0.51	.047	0.02	2.02
	Random Effects		SD
	Participant	Intercept	0.34
	Item	Intercept	0.61
Study 3 - Adults	Fixed Effects	Estimate	SE	p-value	95% CI
	Intercept	0.82	0.24	<.001	0.36	1.31
	Representation Type	1.90	0.40	<.001	1.15	2.75
	Question Type	−0.83	0.27	.002	−1.37	−0.30
	Representation Type * Question Type	−2.58	0.49	<.001	−3.58	−1.65
	Random Effects		SD
	Participant	Intercept	0.31
	Item	Intercept	0.49
Study 3 - Children	Fixed Effects	Estimate	SE	p-value	95% CI
	Intercept	0.17	0.27	.51	−0.36	0.72
	Representation Type	0.46	0.26	.08	−0.06	0.98
	Question Type	1.41	0.29	< .001	0.86	2.00
	Age	0.51	0.21	.01	0.11	0.93
	Representation Type * Question Type	−0.81	0.40	.04	−1.62	−0.03
	Representation Type * Age	−0.13	0.27	.62	−0.67	0.40
	Question Type * Age	−0.44	0.30	.14	−1.03	0.14
	Representation Type * Question Type * Age	0.10	0.41	.81	−0.71	0.91
	Random Effects		SD
	Participant	Intercept	0.64
	Item	Intercept	0.72

For adults, we observed a main effect of question type, with more “yes” responses to questions about properties true of the referent than to questions about properties true of the representation. This pattern is consistent with a tendency to engage in both realism (endorsing properties true of the referent) and representational blindness (rejecting properties true of the representation). We also observed a main effect of representation type, indicating that participants were more likely to say “yes” when the questions were about objects than about pictures (see Figure 2).

Figure 2.

Proportion of “yes” responses for adults and children in Study 1. Error bars are 1 SE.

However, these main effects need to be interpreted within the context of a significant interaction between representation type and question type. For questions about properties of the representation (e.g., Can this fit in your hand?), adults provided more “yes” responses for objects than for pictures (Average Marginal Effect (AME) = 0.40, SE = 0.04, p < .001, 95% CI = 0.31, 0.48). In contrast, for questions about properties of the referent (e.g., Does this roll in the mud?), they provided more “yes” responses for pictures than for objects (AME = −0.16, SE = 0.07, p = .03, 95% CI = −0.30, −0.01). Together, these findings indicate a greater tendency toward representational blindness and toward realism for pictures than objects. It should be noted, however, that the picture-object difference was substantially greater when reasoning about features of the representations than when reasoning about features of the referents. What this means, in other words, is that representational blindness was substantially lower in the context of objects, as predicted, whereas realism persisted at high rates for both objects and pictures.

We observed the same overall pattern in children. There were main effects of question type and representation type. We additionally observed an interaction between representation type and question type. For questions about properties of the representation (e.g., Can this fit in your hand?), children provided more “yes” responses for objects than for pictures (AME = 0.30, SE = 0.07, p < .001, 95% CI = 0.17, 0.44). In contrast, for questions about properties of the referent (e.g., Does this roll in the mud?), they provided numerically more “yes” responses for pictures than for objects (AME = −0.07, SE = 0.04, p = .12, 95% CI = −0.15, 0.02), though this comparison was not significant.

Together, these findings indicate that children, like adults, show tendencies toward both representational blindness and realism, and more so for pictures than objects. At the same time, picture-object differences were greater for questions about representations than for questions about referents. Indeed, for children, the effect of representation type was found only for representational blindness and not for realism.

Pattern analyses.

We also examined participants’ individual response patterns to determine the extent to which realism and representational blindness coincide (see Table 2). To do so, we examined the responses for each participant, separately for pictures vs. objects. They were classified as endorsing the referent only (e.g., “yes” to “Does this roll in the mud?”), the representation only (e.g., “yes” to “Can this fit in your hand?”), or both. (No participant in this study endorsed neither.) A participant was credited with endorsement if they said ‘yes’ at least once for a given question type. A “referent only” pattern is consistent with the participant displaying both realism and representational blindness. A “representation only” pattern is consistent with a participant showing neither realism nor representational blindness. A “both” pattern is consistent with a participant displaying realism, but not representational blindness.

Table 2.

Response patterns in Study 1

	Pictures			Objects
	Referent Only	Representation Only	Both	Referent Only	Representation Only	Both
Children	75%	0%	25%	41%	3%	56%
Adults	69%	6%	25%	16%	22%	63%

Note. Percentage of participants in each age group who endorsed referent properties only, representation properties only, or both, separately for pictures and objects.

For adults, we observed a difference in patterns of responses across representation types (pictures vs. objects), McNemar-Bowker χ² (3) = 19.00, p < .001. Post-hoc tests revealed that the primary pattern for pictures was to treat them strictly in terms of the referent, thus displaying both realism and representational blindness. In contrast, the primary pattern for objects was to endorse questions about both the referent and representation, thus displaying realism while also being aware of the representation’s properties, McNemar test, p < .001 (with Bonferroni correction for multiple comparisons). We observed the same pattern of responses for children: They treated pictures strictly in terms of the referent and objects in terms of both the referent and the representation, McNemar test, p = .006, supporting the findings from the regression analyses by showing that participants’ tendency towards representational blindness was greater for pictures than objects, whereas realism was common for both objects and pictures.

Discussion

Study 1 presented participants with a series of representations (e.g., a toy pig), and assessed their tendency to interpret the items in terms of their dual status: as standing in for a referent (e.g., representing a real-life pig) and as physical objects (e.g., a small, plastic toy). Each item was paired with a simple yes/no question that included an underspecified pronoun, ‘this’ (e.g., “Can this fit in your hand?”). Results indicated evidence for both realism (saying ‘yes’ to questions about the referent—the real pig) and representational blindness (saying ‘no’ to questions about the physical object—the plastic toy). This held for both children and adults, though children were more susceptible to both of these tendencies than adults.

Although both children and adults showed evidence for both realism and representational blindness, the individual response patterns also indicate that these tendencies are not simply two sides of the same coin, but rather are separable and emerge in different contexts. Most notably, when reasoning about toy objects, the majority of participants engaged in realism while also being aware of the properties of the representation itself.

Furthermore, as predicted, participants responded differently depending on the representation type (picture vs. toy object), according to both the regression and the pattern analyses. With pictures, participants showed greater tendencies toward both representational blindness (saying ‘no’ to questions about the representation) and realism (saying ‘yes’ to questions about the referent), than with toy objects. This is consistent with prior work demonstrating that children more readily construe pictures than objects as representations (DeLoache, 1991). However, the contrast between pictures and objects differed by question type. Whereas the representation type yielded a relatively small effect on realism, it yielded a substantially larger effect on representational blindness. This is interesting in its own right, but once again supports the conclusion that the two phenomena reflect different psychological constructs (i.e., representational blindness is not simply equivalent to realism).

STUDY 2

One question that Study 1 raises is why pictures and objects elicited different patterns of response. Study 2 was designed to disentangle effects of representation type (picture vs. object) from the affordances of the item (two- vs. three-dimensional). To do so, we converted all the stimuli into two-dimensional images by creating photographs of the items in Study 1. As such, all the stimuli were two-dimensional photographs, though they differed in what was depicted: photos of color drawings versus photos of toy objects. If the effects of Study 1 were strictly due to differential affordances of the stimuli, then in Study 2, the representation type x question type interaction should disappear entirely. That is, in both cases, participants should respond as they did to pictures in Study 1. However, if pictures and toys are understood as different sorts of representations, then we should replicate the results of Study 1. In Study 2, we also included children from 3–6 years of age to examine a wider age range of children, and thus to provide a test of developmental change in the understanding of representations within the pre-school period.

Method

Participants

Sixty-four children (34 female and 30 male) and 32 adults (19 female and 13 male) participated in the study. Children ranged from 3.02–6.98 years of age (M = 5.03, SD = 1.05) and adults ranged from 18.2–21.6 years of age (M = 19.32, SD = 0.84). Twenty-one additional children were excluded from the data analysis for not completing the task (n = 5), not being a fluent speaker of English (n = 1), failing the warm-up task (i.e., answering more than one warmup question incorrectly in a block; n = 4), failing the post-test (n = 3), or responding incorrectly to more than one of the control questions in the main study (n = 8). The higher rate of participant exclusion in this study compared to Study 1 appears to be due to widening the age range to include 3-year-olds, who were disproportionately represented in the number of children dropped (n = 10/21). This suggests that the current task was challenging for the youngest age group.

Children were recruited from preschools in a midwest college town, and adults were undergraduate introductory psychology students at the University in the same town, receiving credit for their participation. Children were tested individually at their local preschool, either in a separate room or at an isolated table in their classroom. Adults were tested individually in a laboratory space at the University. Children sampled were predominantly White; adults were: 22 White, 7 Asian, 2 African American/Black, and 1 Hispanic/Latino.

Materials and Procedure

Materials were identical to those from Study 1, except that instead of the original pictures and toys, we presented color photographs of the original pictures and toys; these photos were also laminated in clear plastic (see Figure 3). The procedure was identical to that of Study 1, except that children below 4 years of age received a one-minute distraction task rather than a two-minute distraction task, and received shorter versions of two of the questions. In addition, given the younger age of some of the participants, we excluded from the data analysis those who failed the warm-up task by answering more than one of the warm-up questions in a block incorrectly.

Figure 3.

Study 2, sample picture of a picture and picture of a toy object.

Results

We first examined performance on the control questions, separately for pictures of pictures and pictures of objects and separately for children and adults (i.e., 2 scores per question type). On True of Both questions, “yes” responses ranged from 91–95%; on True of Neither questions, “yes” responses ranged from 0–2%. These results indicate that participants appropriately attended to the task.

We submitted adults’ and children’s True of Representation and True of Referent responses to the same logistic regression models as in Study 1 (see middle of Table 1 for the coefficients)². Representation type was coded as follows: picture of a picture = 0, picture of an object = 1; within subject.

Adults provided more “yes” responses regarding properties of the referent than of the representation (see Figure 4). As in Study 1, this indicates a tendency to engage in both realism (endorsing properties true of the referent) and representational blindness (rejecting properties true of the representation). Unlike Study 1, however, there was no main effect of representation type or interaction between representation type and question type. Making all items two-dimensional resulted in adults treating pictures--whether of objects or of pictures--equivalently.

Figure 4.

Proportion of “yes” responses for adults and children in Study 2. Error bars are 1 SE.

For children, there were main effects of question type and representation type, as in Study 1. However, unlike in Study 1, there was a significant three-way interaction among representation type, question type, and age. For questions about properties of the representation (e.g., Can this fit in your hand?), we observed different patterns across ages. Four- and 5-year-olds, like the children in Study 1, provided significantly more “yes” responses for pictures of objects than for pictures of pictures (4-year-olds (−1SD) AME = 0.21, SE = 0.08, p = .005, 95% CI = 0.07, 0.36; 5-year-olds (0SD) AME = 0.14, SE = 0.05, p = .01, 95% CI = 0.03, 0.24). In contrast, 6-year-olds did not (+1SD) (AME = 0.07, SE = 0.07, p = .27, 95% CI = −0.06, 0.20). In contrast, for questions about properties of the referent (e.g., Does this roll in the mud?), children of all ages provided equal rates of “yes” responses for pictures of pictures and pictures of objects (4-year-olds (−1SD): AME = −0.10, SE = 0.06, p = .12, 95% CI = −0.22, 0.03; 5-year-olds (0SD): AME = −0.001, SE = 0.04, p = .81, 95% CI = −0.09, 0.07; 6-year-olds (+1SD): AME = 0.06, SE = 0.05, p = .30, 95% CI = −0.05, 0.16). Altogether then, younger but not older children were influenced by the modality of the representation, even when the functional affordances of the stimuli at test were identical – specifically, they showed a greater tendency toward representational blindness for pictures of pictures than for pictures of objects.

Pattern analyses.

We classified participants’ patterns of responses as in Study 1, again to determine the degree to which representational blindness and realism coincide (see Table 3). One child was excluded from the analysis due to not endorsing any trials regarding either a referent or a representation for the pictures of objects.

Table 3.

Response patterns in Study 2.

	Pictures of pictures			Pictures of objects
	Referent Only	Representation Only	Both	Referent Only	Representation Only	Both
Children	67%	0%	33%	43%	3%	54%
Adults	47%	9%	44%	38%	19%	44%

Note. Percentage of participants per age group endorsing referent properties only, representation properties only, or both, separately for pictures of pictures and pictures of toy objects.

Adult responses did not differ across representation types, McNemar-Bowker χ² (3) = 2.27, p = .52, showing equal rates of referent-only and both response patterns for both, indicating that roughly half showed both realism and representational blindness and half displayed realism only. When all the items were two-dimensional, adults were more likely to treat the pictures – whether of objects or of pictures – equivalently. Children, in contrast, showed the same pattern as in Study 1: pictures of pictures were typically construed strictly in terms of the referent, indicating both realism and representational blindness. In contrast, pictures of objects were typically construed in terms of both the referent and the representation, McNemar test, p = .03, indicating realism while also being aware of the representation’s properties. These data further suggest that even when the functional affordances of the stimuli were identical, children showed more representational blindness for pictures of pictures than for pictures of objects.

Discussion

In Study 2, when participants viewed photographs of drawings and toys, and thus both types of representations were rendered two-dimensional, we again found representational blindness in both children and adults. At both ages, participants typically denied that these representations had their objective physical properties (e.g., they denied that a photo of a toy elephant could be lifted with one hand). In some respects, this result was to be expected, since the items in Study 2 were two-dimensional, as were the pictures in Study 1 that evoked the highest levels of representational blindness.

At the same time, however, this pattern was notable in that it meant that participants denied properties that were doubly true, as they were true of both the photographs that were immediately present, and the representations displayed in the photographs (picture or toy). Accordingly, representational blindness does not just involve ignoring the most immediate layer of representation (in this case, properties of the photograph), but rather extended to ignoring an additional layer of representation (in this case, properties of the photograph as well as properties of the picture or object displayed in the photograph).

Despite the overarching similarity between Studies 1 and 2, there were differences as well. Study 1 obtained a stark difference between pictures and objects, whereas Study 2 revealed a smaller difference, and only in the younger children. These results suggest that affordances of the representation contribute to representational blindness. Reducing picture/object differences by making items less manipulable increases their status as representations (DeLoache, 2000; Gelman, Chesnick, & Waxman, 2005)—though does not necessarily eliminate representation type differences altogether, as the younger children still displayed a small but consistent picture/object difference, consistent with prior work (Callaghan & Corbit, 2015).

Finally, as in Study 1, the data also indicate that realism and representational blindness are not simply two sides of the same coin. First, object/picture differences were obtained for representational blindness but not for realism for children. And second, individual response patterns indicated that nearly half the participants showed a pattern in which realism was stronger than representational blindness.

STUDY 3

Study 3 was designed to provide a stronger test of both realism and representational blindness than in Studies 1 and 2. As noted previously, the wording in the first two studies was deliberately ambiguous, so as not to cue participants toward a focus on either the referent or the representation. Thus, given the question “Does this roll in the mud?”, the pronoun “this” could refer to either the representation (i.e., the toy or the picture) or to the referent (i.e., a pig). In this sense, all responses were accurate, as the question did not specify a given interpretation.

The use of ambiguous questions may have created pragmatic considerations that guided people’s responses, even tipping participants toward displaying both realism and representational blindness. Given that both pictures and toy objects function as representations of real-world items, a participant may have inferred that the experimenter’s ambiguous wording was asking about such referents. Moreover, pragmatic considerations may have operated differently for pictures versus objects, and differently for children versus adults. It is thus important to assess how participants would respond when presented with questions that explicitly disambiguate by asking about the representation per se. To address these issues, in Study 3, we asked participants to answer questions explicitly about the representations (e.g., “Does this toy roll around in the mud?” or “Does this picture roll around in the mud?”).

Method

Participants

Seventy-eight children (37 female and 41 male) and sixty-four adults (47 female and 17 male) participated. Children ranged from 3.01–6.85 years (M = 5.13, SD = 0.97) and adults ranged between 18 and 21 years. An additional 27 children were excluded for: responding incorrectly to more than one of the control questions in the main study (n = 17), failing the warm-up task (n = 3), failing the post-test (n = 3), not completing the study (n = 2), or parental interference (n = 2). As in Study 2, most children who were excluded were 3-year-olds (n = 16/27), suggesting that the current task was challenging for this young age group. Ten additional adults were excluded due to experimenter error (n = 7), technological issues (n = 1), or responding incorrectly to more than one of the control questions in the main study (n = 2).

Children were recruited at local museums and childcare centers in a midwestern U.S. college town and at a children’s science museum in Western Canada. Adults were undergraduate introductory psychology students at a large university in a midwestern town and received course credit for participating. Children and adults were tested individually at local museums or University laboratory spaces, respectively. Children sampled were predominantly White. We did not collect adult demographic data in this study.

Materials

Materials were similar to those in Study 1; however, the exact toys and pictures differed because we could not locate the original materials.

The primary difference between Studies 1 and 3 was in the wording of the questions. In contrast to the earlier studies, Study 3 explicitly named the representation type in the question (e.g., “Does this toy fit in your hand?” or “Does this picture fit in your hand?”). We also altered some of the questions to ensure that the True Only of the Representation and True Only of the Referent questions were not simply the inverse of one another (e.g., in Study 1 for the lion set, the questions were: “Is it safe to touch this?” and “Is this dangerous?”, respectively; in Study 3, the first question was replaced with, “Can this toy/picture stay on a bookshelf?”).

Procedure

The procedure was identical to that of Study 1, except that we now directly specified the representation in the question (e.g., “Does this toy…?” or “Does this picture…?”).

Results

We first examined performance on the control questions, separately for pictures versus objects, and separately for children versus adults. On True of Both questions, “yes” responses ranged from 93–99%; on True of Neither questions, “yes” responses ranged from 0–5%. These results again indicate that participants appropriately attended to the task, as in the prior studies.

Adults’ and children’s True of Representation and True of Referent responses in Study 3 were submitted to the same logistic regression models as in Study 1 (see bottom of Table 1 for the coefficients of the models). For adults, as in Study 1, we observed a main effect of question type, but unlike in Study 1, participants provided more “yes” responses to questions about properties true of the representation than to questions about properties true of the referent. This difference was expected, given that the questions in Study 3 explicitly named the representation (picture or toy). It is nonetheless important to note that responses to questions about the representation were not at ceiling, nor were responses to questions about the referent at floor, indicating that the adults showed a persistent (albeit smaller) tendency toward both realism and representational blindness. We also observed a main effect of representation type, indicating that participants were more likely to say “yes” when the questions were about objects than about pictures (see Figure 5).

Figure 5.

Proportion of “yes” responses for adults and children in Study 3. Error bars are 1 SE.

These main effects need to be interpreted within the context of a significant interaction between representation type and question type. As in Study 1, for questions about properties of the representation (e.g., “Can this toy/picture fit in your hand?”), adults provided more “yes” responses for objects than pictures (AME = 0.26, SE = 0.05, p < .001, 95% CI = 0.15, 0.36). In contrast, for questions about properties of the referent (e.g., “Does this toy/picture roll in the mud?”), they provided more “yes” responses for pictures than objects (AME = −0.16, SE = 0.06, p = .01, 95% CI = −0.27, −0.04). These findings indicate a greater tendency toward representational blindness and realism for pictures than objects, as we observed in Study 1.

For children, there was a significant main effect of question type, wherein participants provided more “yes” responses to questions about properties true of the referent than to questions about properties true of the representation. This is the same pattern as observed in Studies 1 and 2, but the opposite of what we saw in adults in Study 3. Thus, although the questions explicitly referred to the representation (picture or object), children maintained patterns of both realism and (to a lesser extent) representational blindness. There was also a main effect of age, with older children being more likely to say “yes” than younger children.

We also observed an interaction between representation type and question type, indicating that the question type difference was larger for pictures than for objects. Pairwise comparisons indicated no differences between objects and pictures, however. For questions about properties of the representation (e.g., “Can this toy/picture fit in your hand?”), children provided numerically more “yes” responses for objects than for pictures (AME = 0.10, SE = 0.05, p = .07, 95% CI = −0.007, 0.20), whereas for questions about properties of the referent (e.g., “Does this toy/picture roll in the mud?”), they provided an equal number of “yes” responses for pictures and objects (AME = −0.06, SE = 0.05, p = .24, 95% CI = −0.15, 0.04). These findings indicate that children showed more representational blindness for pictures than objects (as in Studies 1 and 2), though the effect of representation type was not as strong for children as it is for adults.

Pattern analyses.

We classified participants’ patterns of responses as in Studies 1 and 2, to determine the degree to which representational blindness and realism coincide (Table 4)³. One child was excluded from the analysis due to not endorsing any trials regarding either a referent or a representation for pictures or for objects. One adult was excluded from the analysis due to not endorsing any trials regarding either a referent or a representation for the pictures.

Table 4.

Response patterns in Study 3

	Pictures			Objects
	Referent Only	Representation Only	Both	Referent Only	Representation Only	Both
Children	26%	9%	65%	16%	5%	79%
Adults	14%	38%	48%	3%	57%	40%

Note. Percentage of participants in each age group who endorsed referent properties only, representation properties only, or both, separately for pictures and toy objects.

Adults showed different response patterns across representation types (pictures vs. objects), McNemar-Bowker χ² (3) = 13.11, p = .004. However, these differed from Study 1. Post-hoc tests revealed that the primary pattern for pictures was to endorse questions about both the referent and representation, thus displaying realism while also being aware of the representation’s properties. In contrast, the primary pattern for objects was to endorse questions about the representation only, thus showing neither realism nor representational blindness, McNemar test, p = .002 (with Bonferroni correction). When asked explicitly about the representation in Study 3, adults’ tendencies towards realism and representational blindness were decreased, especially for toys. In contrast, children did not display different patterns of responses across representation types (pictures vs. objects), McNemar-Bowker χ² (3) = 5.19, p = .16. For both pictures and objects, the primary pattern was to endorse questions about both the referent and representation, thus displaying realism while also being aware of the representation’s properties. Thus, when explicitly asked representations, children were less likely to display representational blindness, but still likely to make realism errors.

Discussion

Study 3 provided a strong test of realism and representational blindness by explicitly asking about the toy or picture presented. In contrast to Studies 1 and 2, for which the test questions were ambiguous (e.g., “Does this roll in the mud?”), and left open whether the pronoun “this” referred to the referent (real pig) or the representation (picture or toy), in Study 3 the test questions included the word “toy” or “picture” (e.g., “Does this picture roll in the mud?”).

Both children and adults in this study showed less realism and representational blindness, compared to Studies 1 and 2. Specifically, they were less likely to endorse properties true of the referent (e.g., to say that a toy pig rolled in the mud), and more likely to endorse properties true of the representation (e.g., to say that a toy pig fits in their hand). Thus, disambiguating the language reduced these tendencies, indicating that results in prior studies were partly the result of the pragmatic inferences due to the wording of the questions.

Nonetheless, even though the representation was explicitly named in the test questions (“this picture”, “this toy”; emphases added), realism and representational blindness did not disappear altogether, for either adults or children. Nearly half the adults endorsed a property true of the referent at least once, as shown in their individual response patterns, and children did so the majority of the time, thereby most often answering in a manner inconsistent with the question wording. The developmental differences in Study 3 were striking, as children and adults displayed qualitatively distinct responses: adults most often endorsing properties true of the representation, and children most often endorsing properties true of the referent.

Finally, Study 3 again revealed differences between pictures and toy objects, at both ages. Adults’ tendency to focus on the representation was more pronounced when reasoning about the toys than the pictures, whereas children’s tendency to focus on the referent was more pronounced when reasoning about the pictures than the toys, as seen in the individual response patterns. Yet again, these results are consistent with the notion that pictures lead more readily and directly to iconic realism and representational blindness, compared to toy objects.

GENERAL DISCUSSION

The present paper was designed to address three questions: (1) Do children display representational blindness, in addition to realism? (2) Do these tendencies differ for pictures versus objects? (3) How do these tendencies change over development? In this section, we discuss what we have learned about each of the questions, based on the three studies reported. We conclude by briefly discussing broader implications for representational reasoning.

Do children display representational blindness, in addition to realism?

In three studies, we have found that when provided with questions about features of representations, children and even adults evidenced representational blindness (denying properties true of a representation) as well as realism (endorsing properties true of the referent). Thus, for example, participants reported that a toy elephant cannot be lifted with one hand (representational blindness) and that a picture of a lion is dangerous (realism). It is intriguing that participants were prone to representational blindness, which entails overlooking immediately perceptible attributes, and supports the notion that children’s concepts cannot be fully understood as a collation of outward perceptual features (e.g., Gelman, 2003).

The responses in Studies 1 and 2 could not be considered errors, as the test questions were framed with an underspecified pronoun, this (e.g., “Can this fit in your hand?”; emphasis added). Accordingly, neither answer--regarding either the referent or the representation--was ‘right’ or ‘wrong’. However, pragmatic considerations cannot fully account for our findings, as we saw both tendencies even with unambiguous questions that explicitly asked about the representation (e.g., “Can this picture fit in your hand?” or “Can this toy fit in your hand?”) in Study 3. These results demonstrate the power of representational blindness as well as realism, especially for children.

One potential alternative account that needs to be addressed is whether representational blindness is simply another way of measuring realism, rather than a distinct phenomenon. In other words, is asking about properties of the representation just the inverse of asking about properties of the referent? Our data demonstrate that these are in fact independent questions, not inverses. As indicated in the pattern analyses, many participants showed a pattern in which they endorsed realism but not representational blindness, by agreeing with both kinds of features (both those true of the representation and those true of the referent). This would not be predicted if the two questions were merely inverses, in which saying ‘yes’ to referent-only questions (e.g., rolling in the mud) would be equivalent to saying ‘no’ to representation-only questions (e.g., lifted with one hand). Furthermore, as noted earlier, the contrast between pictures and objects differed by question type. Whereas representation type yielded a relatively small effect on realism, it yielded a substantially larger effect on representational blindness. If these questions were just inverses, then the effects of representation-type on realism should be equal to those on representational blindness.

Do these tendencies differ for pictures vs. objects?

As predicted, representation type (pictures vs. objects) had a strong effect on responses for children and adults: although both pictures and toys elicited representational blindness and realism, pictures led to more realism and more representational blindness than toys. This is consistent with our observation that prior work showing realism errors typically used pictures as stimuli, whereas prior work showing symbolic-understanding errors typically used objects as stimuli (e.g., DeLoache, 1987, 1991, 2000; DeLoache & Marzolf, 1992; Uttal et al., 2009). This picture/object difference seems to be due in part to affordances of the objects, since reducing object affordances in Study 2 by presenting photos of the stimuli, led to more representational blindness as well as more realism—especially for adults and older children.

An open question for future research is whether picture/object differences reflect different kinds of talk that children hear about pictures and objects. Whereas pictures more often elicit parental conversation about categories, including generic phrases (“What do you make with lemons?”) and ostensive labeling phrases (“That’s a watch”), objects more often elicit talk about individuals, including proper names (“Mr. Frog”) and talking to or for the item (Gelman et al., 2005; Gelman, Waxman, & Kleinberg, 2008; Ware et al., 2013). Although the distinction between individuals and kinds differs from that between representations and items represented, in both cases pictures seem more readily to call to mind a concept that is abstracted away from the reality at hand. Children may have learned, from those around them, that this is the appropriate way to talk about pictures versus objects—thus raising the chicken-and-egg question of how to account for the origins of this pattern of talk and the differences between the way that children respond to questions about pictures versus questions about objects.

What is the developmental course of these tendencies?

Both representational blindness and realism were stronger among young children than adults. Study 3 was especially revealing in this regard. The inclusion of questions explicitly focused on the representation resulted in lower rates of representational blindness for children (as seen in the high rates of “both” patterns for pictures and objects, leading them to endorse features true of both the representation and the referent). In contrast, such questions did not seem to have the same effect on realism (as the vast majority of children still fell prey to this error). For adults, in contrast, it appears that questions explicitly focused on the representation resulted more clearly in a reduction in both tendencies (as seen in the higher rates of “representation only” patterns, especially in the case of objects). Together, these results suggest an intriguing developmental pattern: Early in childhood (i.e., 3–6 years of age), realism errors seem to persist, even when children are given strong cues that they should focus on the representation. In contrast, adults more readily overcome this tendency when given unambiguous linguistic cues.

At the same time, neither tendency was fully absent among adults, even with unambiguous wording. This pattern of improvement with age coupled with persisting conceptual challenges even for adults is analogous to the pattern found in perspective-taking difficulties across development, and suggests that certain reasoning challenges may linger throughout the lifespan (e.g., Epley, Morewedge, & Keysar, 2004; Van Boven & Loewenstein, 2003).

Conclusions

Ultimately it is important for children to be able to consider both the symbolic meaning of a representation and the representation itself, depending on the goal at hand. Often it is important to focus children on the meaning of representations, as when using manipulatives to teach abstract mathematical concepts, or when using dioramas or three-dimensional models to teach generic facts about animals or chemical structures. Yet at other times it may be more beneficial to focus children on the representations per se, as when teaching children to decipher an alphabetic writing system by analyzing the sounds of words, not just their meaning, or when teaching children how to draw, by considering elements of the representation itself, such as perspective and foreshadowing. More generally, the ability to overlook meaning to focus on a representation—the capacity that children have difficulty with in representational blindness—is an intellectual capacity that lies at the heart of a breadth of human achievements, including analyzing literature (reflecting on a poet’s language, or a translator’s choices) and conducting science (objectively analyzing an actor’s behaviors and not just seeing through to their goals or intentions). By studying children’s basic capacity to evaluate both representations and their referents, we come one step closer to understanding how these foundational human skills emerge.