Touch or Watch to Learn? Toddlers’ Object Retrieval Using Contingent and Noncontingent Video

Abstract

The experiment reported here was designed to examine the effect of contingent interaction with touch-screen devices on toddlers’ use of symbolic media (video) during an object-retrieval task. Toddlers (24–36 months old; N = 75) were randomly assigned to watch an animated character hiding on screen either in a no-contingency video (requiring no action), a general-contingency video (accepting touch input anywhere on screen), or a specific-contingency video (requiring touch input on a particular area of interest). After the hiding event, toddlers searched for the character on a corresponding felt board. Across all trials, younger toddlers were more likely to search correctly after a specific-contingency video than after a no-contingency video, which suggests that contingent interaction designed to emphasize specific information on screen may promote learning. However, this effect was reversed for older toddlers. We interpret our findings with respect to the selective encoding of target features during hiding events and the relative strength of memory traces during search.

Keywords

toddlers learning video deficit transfer deficit contingency interactive media object retrieval

The presence of interactive mobile devices in children’s daily lives has increased in recent years, stoking public interest in its potential for enhancing early learning (Rideout, 2013). Learning with symbolic media, such as video and pictures, is a critical part of early cognitive development because of its significance for various problem-solving activities in children’s everyday lives (DeLoache, 2002). However, researchers understand little about how young children process interactive media. The purpose of this study was to examine whether and how toddlers use interactive and noninteractive video to solve real-world problems.

With respect to noninteractive media, 2-year-olds exhibit a transfer deficit; that is, they have difficulty transferring information from video demonstrations to real-world problems (Anderson & Kirkorian, 2015; Barr, 2013; Troseth, 2010). For example, when 24-month-olds search for an object after watching an experimenter hide it in the real world (e.g., looking through a window), they display near-perfect performance in finding the object; however, if they view the hiding event using symbolic media (e.g., closed-circuit video), they often perform at chance levels. Performance with video improves with age and matches performance in in-person conditions by the age of 36 months (Schmitt & Anderson, 2002; Troseth & DeLoache, 1998). This transfer deficit appears to be domain general, having also been found for language (Krcmar, Grela, & Lin, 2007; Kuhl, Tsao, & Liu, 2003; Roseberry, Hirsh-Pasek, Parish-Morris, & Golinkoff, 2009) and imitation (Hayne, Herbert, & Simcock, 2003; Strouse & Troseth, 2008).

Why do toddlers experience difficulty learning from noninteractive video? Observation of children’s performance over multiple trials suggests that mental representations that are derived from on-screen experiences may be more fragile than those arising from in-person experiences with real-world objects (Troseth, 2003, 2010), perhaps because of decreased arousal and engagement relative to in-person interactions (Kuhl, 2007). For example, in object-retrieval tasks using video, 24-month-olds typically perform well on the first trial but poorly on subsequent trials because of perseveration; that is, they make a “smart error” on the basis of their direct experience finding the object on the prior trial rather than updating their memory to the current trial using information from the video (Kirkorian, Lavigne, et al., 2016; Schmidt, Crawley-Davis, & Anderson, 2007; Schmitt & Anderson, 2002; Troseth, 2003). Therefore, performance drops after the first trial, when the relatively strong representation of the previous real-world search event competes with the relatively weak representation of the subsequent video hiding event.

Despite toddlers’ difficulty in generalizing from symbolic media to real-world objects, they can learn from media that afford contingent interactions. For example, 24-month-olds are better at object retrieval after having socially contingent interactions with an adaptive partner via video chat than after watching prerecorded, noninteractive video (Troseth, Saylor, & Archer, 2006). This benefit of social contingency has been replicated using imitation tasks (Nielsen, Simcock, & Jenkins, 2008) and word-learning tasks (Roseberry, Hirsh-Pasek, & Golinkoff, 2014). Moreover, this finding has been extended to contingency in the absence of an adaptive, reciprocal social exchange: Lauricella, Pempek, Barr, and Calvert (2010) found that 30- and 36-month-olds performed better during object retrieval when using interactive computer games than when watching noninteractive video. Likewise, we reported elsewhere that 24-month-olds learned words from an interactive application that required them to touch specific parts of the screen highlighting the location of target objects (Kirkorian, Choi, & Pempek, 2016). Together, these findings suggest that contingency supports children’s symbol-mediated learning.

Does contingency strengthen mental representations derived from video through increased arousal and engagement? This might make little difference on the first search trial: In the absence of competition between mental representations (of the current trial versus previous ones), 24-month-olds succeed on object-retrieval tasks even when they rely on weak memory traces remaining after video hiding events (e.g., Schmitt & Anderson, 2002). Conversely, on subsequent search trials, children must choose between relatively weak representations of video hiding events and more salient representations of real-world search events on previous trials (Troseth, 2010), resulting in increased perseveration. Thus, strengthening mental representations of video hiding events should increase selection of these representations over outdated ones. In other words, if contingency strengthens mental representations of on-screen hiding events as a result of increased arousal and engagement, we would expect not only that toddlers would perform better at object retrieval after interacting with a computer than after viewing noninteractive video (Lauricella et al., 2010) but also that this effect would be more pronounced after the first trial because of decreased perseveration.

Forming a strong mental representation that is salient in memory is necessary but insufficient for successful transfer. Children must also have encoded the relevant features from the event (e.g., the current location of the object rather than locations of distractors). Thus, an alternative hypothesis is that contingency supports learning by increasing selective encoding of target information rather than irrelevant information (Kuhl, 2007). Such attentional support may be particularly useful for younger viewers, given that infants deploy visual attention less systematically than do adults during video viewing; infants’ attention is captured largely by irrelevant but salient features (Franchak, Heeger, Hasson, & Adolph, 2015; Frank, Vul, & Johnson, 2009; Kirkorian, Anderson, & Keen, 2012). In results consistent with this hypothesis, we reported elsewhere that 24-month-olds were able to transfer learning of a novel word from video to real objects only when they used an interactive program that required viewers to touch the location of target objects as they were labeled; word-learning rates did not differ from chance after viewing noninteractive video or using a general interactive program that allowed children to touch anywhere on the screen (Kirkorian, Choi, & Pempek, 2016).

Thus, 24-month-olds may benefit from a contingency that draws attention to target information, facilitating generalization from video to real-world objects. In this case, contingency would improve selective encoding of target features (i.e., content of mental representations) rather than the relative salience of memory traces (i.e., strength of mental representations). As a result, specific contingency would improve object-retrieval performance to the same extent on all search trials because of increased encoding of target information during hiding events. In other words, contingency that guides attention to target information would increase overall performance even on the first trial, without necessarily leading to an even larger effect on subsequent trials as a result of decreased perseveration.

Overview of the Current Study

The purpose of the current study was to evaluate the extent to which contingency helps to (a) increase arousal and engagement that strengthens overall mental representations of video events or (b) increase selective encoding of target features within video events (Kuhl, 2007). We used an object-retrieval task in which 2-year-olds viewed hiding events on video and then retrieved the object from a real-world felt board. This task enabled a microgenetic approach to assess the role of competition between multiple mental representations (on the screen and on the corresponding felt board) across several trials. Children participated in one of three conditions with hiding events presented via video: no contingency (advancing automatically), general contingency (accepting touch input anywhere on screen), or specific contingency (requiring touch input on a particular area of interest).

We predicted that, consistent with previous findings (Kirkorian, Lavigne, et al., 2016; Schmidt et al., 2007; Schmitt & Anderson, 2002; Troseth, 2003), object retrieval in the no-contingency condition would decrease after the first trial because of competition between mental representations, as evidenced by perseveration. Moreover, we expected contingency to increase object-retrieval performance (Kirkorian, Choi, & Pempek, 2016; Lauricella et al., 2010). Of particular interest was whether this benefit would be found for both general- and specific-contingency conditions and whether this benefit would be manifested equally on the first trial and on subsequent trials (because of overall improved performance) or would increase across search trials (because of a reduction in perseveration in particular).

If contingency helps toddlers by increasing arousal and engagement during the hiding event, leading to a stronger memory trace of the hiding location, we would expect better overall performance in both contingent conditions compared with the no-contingency condition. In addition, we would expect the magnitude of this effect to be greater on the second through fourth trials than on the first trial. This effect would be demonstrated by reduced perseveration as a result of decreased competition between multiple mental representations. On the other hand, if contingency helps toddlers learn from symbolic media by increasing selective attention to and encoding of target information during the hiding event, we would expect a larger contingency effect for specific-contingency video (rather than general-contingency video), and we would expect the magnitude of this effect to be the same on the first trial as on subsequent trials because of an overall increase in the likelihood of an errorless retrieval.

Method

Participants and design

The participants were 75 toddlers between 23.5 and 35.8 months of age (42 females, 33 males). Two-year-olds were selected on the basis of the age range used in previous video-based object-retrieval studies, and we used the number of participants in these studies to determine our target sample size (Kirkorian, Lavigne, et al., 2016; Lauricella et al., 2010; Schmidt et al., 2007; Schmitt & Anderson, 2002; Suddendorf, 2003; Troseth & DeLoache, 1998). Participants were randomly assigned to one of three conditions: no contingency (n = 24, 15 females; mean age = 30.82 months), general contingency (n = 25, 15 females; mean age = 29.43 months), and specific contingency (n = 26, 12 females; mean age = 29.91 months). Two additional children were recruited but were dropped from the sample because of inattentiveness to the task.

Stimuli and apparatus

We used video stimuli (for hiding events) and a corresponding felt board (for search events) that were adapted from those used in prior studies (Kirkorian, Lavigne, et al., 2016; Schmidt et al., 2007). The video stimuli were displayed on a touch-screen tablet computer (10.1-in. Galaxy Tab; Samsung America, San Jose, CA) using a mobile application that was developed for this project. The video included images of four hiding locations (a blue gift box, a brown cake, a green balloon, and a red gift box; 4 × 4 cm) and a cartoon teddy bear named Cece (2.5 × 2.7 cm). The felt board had the same dimensions and appearance as the video hiding space (Fig. 1). To create the four hiding locations, we printed screenshots of video images and glued them to white felt, which attached loosely to the surface of the blue felt mounted on a black foam board. The cartoon character was printed on sticker paper and was used as the object for which children searched on the felt board.

Fig. 1.

Screenshot from a hiding event shown on the tablet (left) and the initial set-up of the felt board for a search event (right). The pictures on the felt board were printed screenshots of the video stimuli.

In this study, instead of an experimenter demonstrating a hiding event (Kirkorian, Lavigne, et al., 2016; Schmidt et al., 2007), the bear on the screen was animated so that the character could talk, move, and hide behind one of the four hiding locations. Thus each child’s ability to recall the object’s location was based only on video information, rather than on observation of a live experimenter. See Figure 2 for a full script and procedure during training and test.

Fig. 2.

Script and protocol for training and testing. The phrases in square brackets (e.g., “[Watch the screen/touch the screen/touch me]”) are what the on-screen character said in the no-contingency, general-contingency, and specific-contingency conditions, respectively. See the Procedure section for more information.

The stimuli for the three experimental conditions differed in only two ways: (a) the specific verbal instruction given by the animated character and (b) the type of response required to advance the video. In the no-contingency condition, Cece the bear told the child to “watch the screen,” and the video continued automatically after a 1-s delay. In the general-contingency condition, Cece told the child to “touch the screen,” and the video paused, resuming when the child touched any part of the screen. In the specific-contingency condition, Cece told the child to “touch me,” and the video resumed only when the child touched the character; touching anywhere else on the screen resulted in no change.

Procedure

Two researchers conducted experimental sessions in an empty classroom at preschools during normal school hours. Before each session, the experimenter met with the child in his or her classroom and then ushered him or her into an adjacent room. The experimenter explained that they would play a hide-and-seek game together. During the game, an assistant noted the number of tries needed to find the sticker and the order in which the child searched the locations. The child completed two tasks in the experimental session: correspondence training with the tablet and felt board and the object-retrieval test.

Training

The experimenter placed the tablet computer and the felt board on two side-by-side stands (Fig. 1). The experimenter opened the training application corresponding to the child’s experimental condition. This training was designed to familiarize the child with the instruction and the device, as well as the correspondence between the symbol (the screen) and the referent (the felt board). The character moved to each hiding location on the screen while the experimenter called attention to the corresponding location on the felt board. (See Fig. 2 for the script and procedure for training.) If needed, the experimenter encouraged the child to follow directions (e.g., “What did she say? She said to touch the screen!”).

Testing

There were four object-retrieval trials, and the order of locations was randomly assigned for each child. In each trial, the child watched the hiding event on the touch-screen device. During the hiding event, the animated character moved behind one of the objects. At the end of the hiding event, the experimenter occluded the touch screen, turned the felt board away from the child, and hid the sticker on the felt board. Then the experimenter turned the felt board toward the child and asked the child to find the sticker of Cece. If the child pointed to a location rather than searching behind it, the experimenter encouraged the child to lift it up. If the child was incorrect, the experimenter encouraged the child to search another place until the child found the sticker. Thus each trial ended with the child finding Cece. (See Fig. 2 for more detail.) The same procedure was followed for each of the four trials.

Coding

Search events were coded to denote whether each trial resulted in an errorless search (i.e., the child searched behind the correct location on the first try). If a child made an error during Trials 2 through 4, the error was further classified as either a perseverative error (i.e., the child searched behind the location that was correct on the previous trial on the first try) or a nonperseverative error (i.e., the child searched behind one of the other two incorrect locations on the first try). Independent observers coded 20% of participants with 99% agreement.

Parent survey

Parents completed an online questionnaire that requested demographic information (e.g., parent’s education, child’s race and ethnicity), assessed the child’s vocabulary, and requested information about the child’s media use at home. The MacArthur Communicative Development Inventories (CDI) checklist was used to measure each child’s productive vocabulary (Level II for children younger than 30 months of age, and Level III for children 30 months of age and older; Fenson et al., 2000). For comparison across the age range, raw vocabulary scores were converted to standardized percentile ranks on the basis of the child’s age and gender. For media use at home, parents reported whether they allow their child to use interactive touch-screen devices and how much time their child spent using different types of media (e.g., television, computer, mobile device) on the previous day.

Results

Parent survey and preliminary analyses

Among the 79% of participating families who completed the survey, most (83%) identified their child as White. On average, parents had completed 19.56 years of education (SD = 2.78, range = 13–25 years). The children’s average productive vocabulary, measured as percentile rank on the CDI checklist, was at the 53rd percentile (SD = 25.21, range = 6th–92nd percentiles). Seventy-eight percent of respondents reported that they allowed their child to use a touch-screen device at home. The children’s total amount of screen time on the previous day averaged 39.74 min (SD = 36.49, range = 0–215 min), during which children spent 31.47 min (SD = 35.08, range = 0–200 min) watching television and 6.33 min (SD = 10.85, range = 0–45 min) using a touch-screen device.

Preliminary analyses indicated that child gender, child race and ethnicity, parent education, child exposure to television and touch-screen devices, child vocabulary, and the order of hiding locations did not differ significantly across experimental conditions. Moreover, these variables were not related to search performance. Thus, they were not considered further.

Probability of errorless search by trial and condition

Prior research demonstrated that toddlers’ object retrieval using video changes over trials, such that performance is often higher on the first trial than on subsequent trials (Kirkorian, Lavigne, et al., 2016; Schmidt et al., 2007; Schmitt & Anderson, 2002). Thus search performance on each trial (rather than total correct searches across all four trials) was considered as an outcome measure. A two-level linear mixed-effects model with trials (Level 1) nested within participants (Level 2) was used to assess the effect of condition on errorless search. Given that each trial had a binary outcome—successful (coded as 1) or unsuccessful (coded as 0)—a generalized linear mixed-effects model was specified with binomial error structure and logit link function. Models were estimated using the function glmer from the package lme4 (Version 7.0.1; Bates, Maechler, Bolker, & Walker, 2014) in the R software environment (Version 3.2.0; R Development Core Team, 2015). Two measures of model comparison were used to evaluate fitted models: Akaike’s information criterion (AIC) and log-likelihood ratio (LLR) tests derived from the maximum log-likelihood by taking into account the number of parameters.

The model structure was as follows:

\begin{array}{l} Level 1 model : \\ Probability ({ES}_{t i} = 1) = ϕ_{t i} \\ \log [ϕ_{t i} / (1 - ϕ_{t i})] = η_{t i} \\ η_{t i} = π_{0 i} + π_{1 i} \times {trial}_{t i} + π_{2 i} \times {trial}_{t i}^{2} + ε_{t i} \end{array}

\begin{array}{l} Level 2 model : \\ π_{0 i} = β_{00} + β_{01} \times {age}_{i} + β_{02} \times {GC}_{i} + β_{03} \times {SC}_{i} + \\ β_{04} \times {GC}_{i} \times {age}_{i} + β_{05} \times {SC}_{i} \times {age}_{i} + γ_{0 i} \\ π_{1 i} = β_{10} \\ π_{2 i} = β_{20} \end{array}

\begin{array}{l} Combined model : \\ η_{t i} = β_{00} + β_{01} \times {age}_{i} + β_{02} \times {GC}_{i} + β_{03} \times {SC}_{i} + \\ β_{04} \times {GC}_{i} \times {age}_{i} + β_{05} \times {SC}_{i} \times {age}_{i} + \\ β_{10} \times {trial}_{t i} + β_{20} \times {trial}_{t i}^{2} + γ_{0 i} + ε_{t i} \end{array}

In these models, φ_ti and η_ti represent the probability of errorless search (ES) and the log of the odds of ES, respectively, at trial t for participant i. π_pi represents the trajectory parameter for participant i associated with the polynomial of degree p (p = 0, 1, 2). β₀₀ represents the overall intercept, and β_0q represents the effect of variable q (q = 1, 2, 3, 4, 5) on π_0i. β₁₀ and β₂₀ represent average linear and quadratic regression slopes, respectively. The age and trial variables were treated as continuous predictors and centered at the first point (i.e., a centered age of 0 represents 24 months, and a centered trial of 0 represents Trial 1). The square of the centered trial variable was also included to assess the quadratic effect of the trial variable. The three experimental conditions were represented by two dummy variables, each of which represented the contrast between the coded condition (i.e., general contingency, or GC, and specific contingency, or SC) and the reference condition (no contingency, or NC). All of the predictors were modeled as fixed effects except γ_0i, which represents the random intercept. The residual in the log of the probability of ES is represented by ε_ti. We refer to the combined model as Model 1.

All fixed effects from the final model are reported in Table 1, and the modeled probability of errorless retrieval is plotted in Figure 3 as a function of age and condition. The model revealed significant changes in object-retrieval performance across the four trials, including a significant linear decrease across trials, β₁₀ = −1.80, SE = 0.44, Wald z = −4.10, p < .001, odds ratio (OR) = 0.17, and a positive quadratic component, β₂₀ = 0.46, SE = 0.14, Wald z = 3.38, p < .001, OR = 1.59. Together, the linear and quadratic effects reflected a U-shaped pattern such that performance initially decreased across the first few trials and then increased at the final trial (Fig. 3). Age was also a significant predictor of errorless retrieval, β₀₁ = 0.32, SE = 0.08, Wald z = 3.94, p < .001, OR = 1.38. As the age of children increased, so did their performance during the search task; on average, each additional month in age was associated with a 38% increase in the odds of an errorless retrieval.

Table 1.

Fixed Effects From the Final Mixed Logit Model Predicting the Probability of Errorless Search

Predictor	β	SE	Wald z	OR	95% CI (OR)
Intercept (β₀₀)	−1.31	0.66	−1.98*	0.27	[0.07, 0.94]
Trial (β₁₀)	−1.80	0.44	−4.10**	0.17	[0.07, 0.38]
Trial² (β₂₀)	0.46	0.14	3.38**	1.59	[1.22, 2.09]
Age (β₀₁)	0.32	0.08	3.94**	1.38	[1.22, 2.09]
GC condition (β₀₂)	1.01	0.80	1.26	2.74	[0.57, 14.44]
SC condition (β₀₃)	1.67	0.79	2.10***	5.29	[1.16, 28.25]
GC Condition × Age (β₀₄)	−0.10	0.11	−0.96	0.90	[0.72, 1.12]
SC Condition × Age (β₀₅)	−0.26	0.10	−2.50*	0.77	[0.62, 0.94]

Note: The Level 1 variable was trial number (both linear and quadratic components); this predictor was centered at the first trial, such that the intercept reflects data at Trial 1. Condition and age were included as Level 2 variables. Two dummy-coded condition variables were included: general contingency (GC) and specific contingency (SC); the no-contingency condition was the reference category for each. Age was centered at 24 months. Interactions that are not listed did not significantly improve the fit of the model. OR = odds ratio; CI = confidence interval.

p < .05. **p < .001.

Fig. 3.

Predicted probability of errorless search as a function of trial for the no-contingency, general-contingency, and specific-contingency conditions. Age was a continuous predictor; however, for illustrative purposes, the predicted probability is presented separately for children at the mean age (i.e., 30.04 months), at 1 SD above the mean (33.94 months), and at 1 SD below the mean (25.15 months).

The main effect of condition reflects performance on Trial 1 (centered trial = 0). The specific-contingency condition was a significant predictor such that the probability of an errorless search was higher for the youngest toddlers (centered age = 0) in this condition than for those in the no-contingency condition, β₀₃ = 1.67, SE = 0.79, Wald z = 2.10, p = .035, OR = 5.29. In terms of probability, the youngest children in the specific-contingency condition were about twice as likely to search correctly on Trial 1 as their peers in the no-contingency condition. Thus, this condition effect existed even on the very first trial, before children faced competition between multiple mental representations (i.e., of the current hiding event and previous search event).

We also examined whether the linear and quadratic slopes for trial were modified by experimental condition at Level 2 (denoted as Model 2). The model comparison was not significant, AIC for Model 1 = 372.37, AIC for Model 2 = 387.63, LLR χ²(10) = 4.74, p > .250, indicating that the decrease in performance over trials was of the same magnitude for all of the experimental conditions. In other words, there was no evidence of an interaction between trial and condition, suggesting that the effect of condition had the same magnitude on Trial 1 as on subsequent trials.

The main effect of specific-contingency condition was modified by a significant interaction with age, such that the size of the condition effect decreased as age increased, eventually reversing among the oldest children in the sample, β₀₅ = −0.26, SE = 0.10, Wald z = −2.50, p = .012, OR = 0.77. In other words, specific-contingency video appears to have improved object retrieval among the younger toddlers in our sample, to have had little effect for children at the mean age, and to have actually hindered object search for the older toddlers (Fig. 3).

Neither the main effect of general-contingency condition, β₀₂ = 1.01, SE = 0.80, Wald z = 1.26, p = .207, OR = 2.74, nor the interaction between general-contingency condition and age, β₀₄ = −0.10, SE = 0.11, Wald z = −0.96, p > .250, OR = 0.90, was a significant predictor of errorless retrieval. Thus, overall performance did not differ significantly between children in the no-contingency condition and those in the general-contingency condition. Follow-up analyses comparing the general-contingency and specific-contingency conditions directly indicated that the probability of errorless search in the general-contingency condition was intermediate and not significantly different from performance in either of the other two conditions.

We also examined, in addition to the fitted random-intercept model (defined as Model 1), whether the linear and the quadratic slopes of trial varied among individual participants (defined as Model 3). Model 1 was compared with Model 3 to determine whether adding random slopes would increase the fit to the data. The assessment of relative differences in model comparisons suggested that Model 3 did not provide a better fit than Model 1, AIC for Model 1 = 372.37, AIC for Model 3 = 376.94, LLR χ²(5) = 5.42, p > .250. That is, the linear and the quadratic slopes did not vary significantly across individuals, which suggests a similar rate of change over trials for all participants.

Probability of perseverative errors by trial and condition

Prior research suggested that performance in the no-contingency condition would decrease across trials because of a relatively high frequency of perseverative errors in Trials 2 to 4. We obtained results consistent with the prior findings: The majority of errors (55.29%) were perseverative across Trials 2 to 4 and across all children in the study. However, researchers have yet to report whether the probability of making a perseverative error differs across these trials. Thus, we investigated whether there was any difference between frequencies of perseverative errors on Trials 2, 3, and 4. We also investigated the influence of age and condition on the likelihood that, for any given trial, an error would be a perseverative error. A two-level (trials within subjects) generalized linear mixed-effects model was used to predict the probability of making a perseverative error after having made an error on the first search attempt in a particular trial. Only the instances in which errors were made on Trials 2 to 4 were included in this analysis. The outcome measure was a dichotomous variable indicating whether the error was perseverative in nature. The model structure was as follows:

\begin{array}{l} Level 1 Model : \\ Probability ({PE}_{t i} = 1) = ϕ_{t i} \\ \log [ϕ_{t i} / (1 - ϕ_{t i})] = η_{t i} \\ η_{t i} = π_{0 i} + π_{1 i} \times {trial}_{t i} + ε_{t i} \end{array}

\begin{array}{l} Level 2 Model : \\ π_{0 i} = β_{00} + β_{01} \times {age}_{i} + β_{02} \times {GC}_{i} + β_{03} \times {SC}_{i} + γ_{0 i} \\ π_{1 i} = β_{10} \end{array}

\begin{array}{l} Combined Model : \\ η_{t i} = β_{00} + β_{01} \times {age}_{i} + β_{02} \times {GC}_{i} + \\ β_{03} \times {SC}_{i} + β_{10} \times {trial}_{t i} + γ_{0 i} + ε_{t i} \end{array}

In these models, φ_ti and η_ti represent the probability that an error would be perseverative (PE) and the log of the odds of PE, respectively, at trial t for participant i; ε_ti represents the residual in the log of the probability of PE. All other terms have the same meanings as in the prior set of equations. A full report of fixed effects can be found in Table 2.

Table 2.

Fixed Effects From the Final Mixed Logit Model Predicting the Probability That an Error Would Be Perseverative

Predictor	β	SE	Wald z	OR	95% CI (OR)
Intercept (β₀₀)	1.09	0.47	2.31*	5.47	[1.20, 7.78]
Trial (β₁₀)	−0.71	0.22	−3.17**	0.48	[0.31, 0.76]
Age (β₀₁)	0.02	0.05	0.36	0.91	[0.93, 1.12]
GC condition (β₀₂)	−0.48	0.45	−1.08	0.19	[0.25, 1.48]
SC condition (β₀₃)	−0.01	0.44	−0.03	0.57	[0.42, 2.34]

Note: The Level 1 variable was trial number; this predictor was centered at the first trial, such that the intercept reflects data at Trial 1. Condition and age were included as Level 2 variables. Two dummy-coded condition variables were included: general contingency (GC) and specific contingency (SC); the no-contingency condition was the reference category for each. Age was centered at 24 months. Interactions that are not listed did not significantly improve the fit of the model. OR = odds ratio; CI = confidence interval.

p < .05. **p < .001.

The probability that an error would be perseverative decreased significantly across trials: 75.51% (37 of 49 errors) in Trial 2, 53.19% (25 of 47 errors) in Trial 3, and 42.22% (19 of 45 errors) in Trial 4. However, even though there were age and condition effects on the overall probability of making any kind of error (as described in the previous section), neither age nor condition affected the likelihood that an error was perseverative. Thus, the rate of perseveration was the same across the entire age range and in all three conditions. Rates of errorless searches, perseverative errors, and nonperseverative errors are graphed in Figure 4 as a function of trial.

Fig. 4.

Search performance as a function of trial for errorless searches, perseverative errors, and nonperseverative errors.

Discussion

The mental representations that toddlers derive from symbol-mediated experiences (e.g., video) may be weaker than those based on real-world experiences: During object-retrieval tasks, the relatively salient memory of the previous in-person search event is selected over the weaker memory of the video hiding event, resulting in perseveration (Kirkorian, Lavigne, et al., 2016; Schmidt et al., 2007; Schmitt & Anderson, 2002; Troseth, 2003). However, interactivity can improve video-mediated learning (Lauricella et al., 2010; Troseth et al., 2006). The purpose of the current study was to determine whether contingency improves toddlers’ learning and, if so, what the likely cognitive mechanisms underlying this effect are.

If contingent interactions with video enhance learning by increasing arousal and engagement, thereby strengthening mental representations of hiding events, then any interaction should decrease perseveration. However, our current findings do not support this hypothesis. First, among younger children, the facilitative effect of contingency was greatest for specific-contingency video, suggesting that contingency benefits learning by guiding selective attention to target information rather than by increasing overall attention and engagement. Moreover, the children in this condition made fewer errors overall, not just fewer perseverative errors, and the magnitude of the effect was the same on Trial 1 as on subsequent trials. Therefore, the advantage of specific-contingency media cannot be explained by decreased competition between multiple mental representations. Instead, by directing attention appropriately, interactive media may give toddlers essential cues about target information just as social cues (e.g., eye gaze) may guide attention during reciprocal interactions with real people (Kuhl, 2007). This may be particularly useful for younger viewers who have difficulty selecting target information within complex scenes (Franchak et al., 2015; Frank et al., 2009; Kirkorian et al., 2012).

Although specific-contingency media supported learning among younger toddlers in our sample, the effect decreased with age and reversed among older children. We reported elsewhere that specific-contingency media benefited 24-month-olds but disrupted learning by older children (Kirkorian, Choi, & Pempek, 2016). It is noteworthy that the same pattern has been found in two studies using different tasks, suggesting that these effects reflect domain-general processes related to symbol use. Specific-contingency video might provide the greatest benefit to the youngest children, who are not capable of learning from noncontingent video and therefore require attentional support to ignore irrelevant information (Kirkorian, Choi, & Pempek, 2016). Older children, on the other hand, are capable of learning from noncontingent video without attentional support. Therefore, a moderate amount of information may be optimal, and encoding too many features (e.g., via specific-contingency video) may overly contextualize representations in memory, thereby impeding transfer in the face of contextual changes (see Vlach, 2014).

Taken together, findings from the current study suggest that certain types of contingency facilitate toddlers’ use of symbolic media by enabling them to selectively encode target information. However, this same contingent experience disrupts learning by older children. These findings extend prior research by demonstrating that the impact of children’s contingent interactions with media depends on the age of the children and the type of contingency. Moreover, our findings illuminate potential underlying mechanisms by investigating the type of contingency that improves learning and the timing of effects. This line of research contributes to the growing body of literature on the use of symbolic media during early childhood and has practical implications for the design of educational media for young children. In light of the pervasiveness of screen-based technologies in young children’s lives, research in this area is greatly needed.

Footnotes

Acknowledgements

We thank research assistants Danielle Joyce, Alexandra Nicholas, Laura Pierpont, Stacy Schloesser, and Kristin Shafer.

Action Editor

Brian P. Ackerman served as action editor for this article.

Declaration of Conflicting Interests

The authors declared that they had no conflicts of interest with respect to their authorship or the publication of this article.

Funding

This work was supported by National Science Foundation Grant BCS-1226550.

References

Anderson

D. R.

Kirkorian

H. L.

(2015). Cognitive development through media. In Lerner

R. M.

(Ed.), Handbook of child psychology and developmental science: Vol. 2. Cognitive processes (7th ed., pp. 949–994). Hoboken, NJ: Wiley.

Barr

(2013). Memory constraints on infant learning from picture books, television, and touchscreens. Child Development Perspectives, 7, 205–210. doi:10.1111/cdep.12041

Bates

Maechler

Bolker

Walker

(2014). lme4: Linear mixed-effects models using Eigen and S4 (Version 1.1-7) [Computer software]. Retrieved from http://cran.r-project.org/package=lme4

DeLoache

J. S.

(2002). Early development of the understanding and use of symbolic artifacts. In Goswami

(Ed.), Handbook of childhood cognitive development (pp. 206–226). Oxford, England: Blackwell.

Fenson

Pethick

Renda

Cox

J. L.

Dale

P. S.

Reznick

J. S.

(2000). Short-form versions of the MacArthur Communicative Development Inventories. Applied Psycholinguistics, 21, 95–115.

Franchak

J. M.

Heeger

D. J.

Hasson

Adolph

K. E.

(2015). Free viewing gaze behavior in infants and adults. Infancy. Advance online publication. doi:10.1111/infa.12119

Frank

M. C.

Vul

Johnson

S. P.

(2009). Development of infants’ attention to faces during the first year. Cognition, 110, 160–170. doi:10.1016/j.cognition.2008.11.010

Hayne

Herbert

Simcock

(2003). Imitation from television by 24- and 30-month-olds. Developmental Science, 6, 254–261. doi:10.1111/1467-7687.00281

Kirkorian

H. L.

Anderson

D. R.

Keen

(2012). Age differences in online processing of video: An eye movement study. Child Development, 83, 497–507. doi:10.1111/j.1467-8624.2011.01719.x

10.

Kirkorian

H. L.

Choi

Pempek

T. A.

(2016). Toddlers’ word learning from contingent and noncontingent video on touch screens. Child Development, 87, 405–413. doi:10.1111/cdev.12508

11.

Kirkorian

H. L.

Lavigne

H. J.

Hanson

K. G.

Troseth

Demers

L. B.

Anderson

D. R.

(2016). Video deficit in toddlers’ object retrieval: What eye movements reveal about online cognition. Infancy, 21, 37–64. doi:10.1111/infa.12102

12.

Krcmar

Grela

Lin

(2007). Can toddlers learn vocabulary from television? An experimental approach. Media Psychology, 10, 41–63. doi:10.1080/15213260701300931

13.

Kuhl

P. K.

(2007). Is speech learning ‘gated’ by the social brain? Developmental Science, 10, 110–120. doi:10.1111/j.1467-7687.2007.00572.x

14.

Kuhl

P. K.

Tsao

F. M.

Liu

H. M.

(2003). Foreign-language experience in infancy: Effects of short-term exposure and social interaction on phonetic learning. Proceedings of the National Academy of Sciences, USA, 100, 9096–9101. doi:10.1073/pnas.1532872100

15.

Lauricella

A. R.

Pempek

T. A.

Barr

Calvert

S. L.

(2010). Contingent computer interactions for young children’s object retrieval success. Journal of Applied Developmental Psychology, 31, 362–369. doi:10.1016/j.appdev.2010.06.002

16.

Nielsen

Simcock

Jenkins

(2008). The effect of social engagement on 24-month-olds’ imitation from live and televised models. Developmental Science, 11, 722–731. doi:10.1111/j.1467-7687.2008.00722.x

17.

R Development Core Team. (2015). R: A language and environment for statistical computing (Version 3.2.0) [Computer software]. Retrieved from https://www.r-project.org/index.html

18.

Rideout

(2013). Zero to eight: Children’s media use in America 2013. Retrieved from the Common Sense Media Web site: http://www.commonsensemedia.org

19.

Roseberry

Hirsh-Pasek

Golinkoff

R. M.

(2014). Skype me! Socially contingent interactions help toddlers learn language. Child Development, 85, 956–970. doi:10.1111/cdev.12166

20.

Roseberry

Hirsh-Pasek

Parish-Morris

Golinkoff

R. M.

(2009). Live action: Can young children learn verbs from video? Child Development, 80, 1360–1375. doi:10.1111/j.1467-8624.2009.01338.x

21.

Schmidt

M. E.

Crawley-Davis

A. M.

Anderson

D. R.

(2007). Two-year-olds’ object retrieval based on television: Testing a perceptual account. Media Psychology, 9, 389–410. doi:10.1080/15213260701291346

22.

Schmitt

K. L.

Anderson

D. R.

(2002). Television and reality: Toddlers’ use of visual information from video to guide behavior. Media Psychology, 4, 51–76. doi:10.1207/S1532785XMEP0401_03

23.

Strouse

G. A.

Troseth

G. L.

(2008). ‘‘Don’t try this at home”: Toddlers’ imitation of new skills from people on video. Journal of Experimental Child Psychology, 101, 262–280. doi:10.1016/j.jecp.2008.05.010

24.

Suddendorf

(2003). Early representational insight: Twenty-four-month-olds can use a photo to find an object in the world. Child Development, 74, 896–904. doi:10.1111/1467-8624.00574

25.

Troseth

G. L.

(2003). TV guide: Two-year-old children learn to use video as a source of information. Developmental Psychology, 39, 140–150. doi:10.1037/0012-1649.39.1.140

26.

Troseth

G. L.

(2010). Is it life or is it Memorex? Video as a representation of reality. Developmental Review, 30, 155–175. doi:10.1016/j.dr.2010.03.007

27.

Troseth

G. L.

DeLoache

J. S.

(1998). The medium can obscure the message: Young children’s understanding of video. Child Development, 69, 950–965. doi:10.1111/j.1467-8624.1998.tb06153.x

28.

Troseth

G. L.

Saylor

M. M.

Archer

A. H.

(2006). Young children’s use of video as a source of socially relevant information. Child Development, 77, 786–799. doi:10.1111/j.1467-8624.2006.00903.x

29.

Vlach

H. A.

(2014). The spacing effect in children’s generalization of knowledge: Allowing children time to forget promotes their ability to learn. Child Development Perspectives, 8, 163–168. doi:10.1111/cdep.12079