Exploring the role of “in the moment” and global caregiver and child factors in caregiver questioning during shared book viewing

Abstract

Questions of high (vs. low) cognitive demand (CD), which encourage children to engage in abstract or critical thinking (e.g., problem solve, reason about cause-and-effect relations, make inferences), may drive relations between children’s language exposure and early skills. The present study adopted a micro-analytic approach to examine caregivers’ high-CD questioning with their preschool-aged children while viewing a wordless picture book (n = 121) and “in the moment” (e.g., interaction time, child responses) and global factors (e.g., caregiver education). The probability of caregivers’ high-CD questioning increased with interaction time and caregiver education. Post-hoc exploratory analyses revealed that the relation between children’s responses and caregivers’ high-CD questioning depended on caregivers’ perceptions of children’s vocabulary skills. Specifically, the probability of caregivers’ subsequent high-CD questioning was greater if their child did not respond previously and if caregivers perceived them to have high vocabulary skills. In contrast, caregivers’ questioning remained relatively constant for responsive children across different vocabulary skills. Thus, caregivers may employ certain types of input during brief, informal learning interactions with their children by considering their own and their child’s propensities and micro-level changes that occur during their conversations.

Introduction

Caregiver language input during informal home learning activities makes important contributions to the development of children’s academic skills, including language (see Anderson et al., 2021 for a recent meta-analysis), vocabulary (Rowe, 2013), and mathematics skills (Reynolds et al., 2019; Ribner et al., 2020). Questioning is one type of input that caregivers use in daily conversations with their children (Yu, Bonawitz, et al., 2019) that may be especially important to children’s developing cognitive skills. Compared to directive forms of input (e.g., statement prompts, pedagogical instructions, “controlling” speech), questions can be more effective at drawing children’s attention to and providing cues about the importance of the concepts or topics mentioned. Additionally, questions may suggest to children that certain topics are worth pursuing or thinking about and thus encourage children’s exploration (Strouse et al., 2013; Willard et al., 2019; Yu, Landrum, et al., 2018; Yu, Bonawitz, et al., 2019). Further, caregivers’ questioning encourages children to form their own thoughts about the content and often elicit verbal responses. Thus, questions may help reinforce children’s existing early competencies and encourage them to practice a diversity of emerging skills related to language and literacy (Reynolds et al., 2019; Strouse et al., 2013), math (Reynolds et al., 2019), scientific reasoning (Spruijt et al., 2020), causal learning (Daubert et al., 2020), and emotion knowledge (Bailey et al., 2013).

Considerable research with teachers and caregivers has shown that exposure to more cognitively demanding questions relates to more advanced math and language skills in children (Duong et al., 2021; Massey et al., 2008; Scull et al., 2013; Tompkins, Zucker, et al., 2013; Tompkins, Bengochea, et al., 2017; Uscianowski et al., 2020). Although researchers vary in the terminology that they use to define question types, there is a consensus that questions can be of low or high order, consistent with Bloom’s taxonomy (1956). High- (versus low-) cognitive demand (CD) questions are comprised of abstract, conceptual, elaborative, inferential, decontextualized, and some open-ended input (see Walsh & Hodge, 2018 for a review of questioning styles). These types of questions encourage children to adopt relatively higher levels of abstract thinking about the content beyond perceptually present or immediate information to respond. Thus, children may engage in problem solving, combining various pieces of information, reasoning about cause-and-effect relations, providing explanations, and/or making inferences, predictions, or generalizations (Benjamin et al., 2010; Callanan et al., 2017; van Kleeck, Gillam, et al., 1997; Lee & Kinzie, 2012; Scull et al., 2013; van Kleeck, Vander Woude, et al., 2006; Zucker et al., 2010). In line with Vygotsky’s (1978) ‘Zone of Proximal Development’ (ZPD), caregivers can scaffold children’s abilities by engaging them in conversations that push (slightly) beyond their actual cognitive abilities with tools such as high-CD questions. Thus, questions of high-CD may drive the links between children’s exposure to questions and their cognitive skills. These findings underscore the importance of considering this aspect of caregiver language input beyond the quantity of questions within a broader category, e.g., wh-questions which comprise of both low-CD input such as “what” and “where” questions and high-CD input such as “why” and “how” questions.

More broadly, the aforementioned mechanisms that may underlie the contribution of high-CD input to children’s early cognitive abilities are consistent with sociocultural theories of cognitive development which view learning and development as culturally- and socially-mediated processes. Specifically, young children’s knowledge and skills emerge and grow through interactions with more-skilled individuals, such as their caregivers, during informal learning activities and everyday routines. These interactions involve collaboration as well as guidance and scaffolding from caregivers, and they increase in complexity as children develop relevant skills that can be leveraged in these interactions (Gauvain et al., 2011; Rogoff, 2003; Vygotskij & Cole, 1981). For instance, as children engage in more conversation with their caregivers in cultural contexts such as a shared book reading, they may develop the necessary language skills to respond to caregiver prompts. Over time, caregivers may begin to employ more cognitively demanding talk during conversations with their children which may further facilitate their cognitive development. This cyclical and adaptive process also occurs within social interactions on a moment-to-moment basis, which is aligned with a dynamic systems perspective on dyadic interactions and learning. This view posits that dyadic conversations are adaptive, iterative, non-stationary, and sensitive to feedback. In particular, speakers tend to adapt their language input to the other’s previous patterns of input, and this “tuning” relates to child factors like their language skills (Cox & van Dijk, 2013; Denby & Yurovsky, 2019; van Dijk et al., 2013). In the context of adult questioning, this perspective would suggest that caregivers may adjust their input by introducing more or less low- and high-CD input throughout a conversation, which is aligned with past work on caregivers’ use of both low- and high-demand language input to scaffold children’s developing language skills (e.g., van Kleeck, Gillam, et al., 1997). Sociocultural theories of cognitive development and dynamic systems theory are not mutually exclusive and should be considered together in studies of children’s exposure to high-CD input in the context of cognitive development.

Given the contribution of adults’ high-CD questioning during (in)formal learning activities in the classroom or home to children’s developing cognitive skills, it is important to understand the factors that relate to caregivers’ use of cognitively demanding questions. Extant research has uncovered associations between caregiver input, including questioning, and child factors, such as responsiveness (e.g., Danis et al., 2000; Kuchirko et al., 2016) and language skills (e.g., Duong et al., 2021; Reynolds et al., 2019), as well as caregiver factors like education (Rowe, 2018), suggesting that factors related to interactions “in the moment” and global characteristics of caregivers and children play a role in children’s language exposure.

However, many studies model “in the moment” factors of caregiver-child interactions with aggregate statistics, commonly summing or averaging the quantity of input or using a rating scale, which cannot tell us much about the sequential, adaptive nature of caregiver and child conversations (although see Callanan et al., 2020, for an exception). Also, it is less well-known how other types of caregiver input, such as caregiver scaffolding, relates to their tendency to provide certain types of language input later on during the same interaction. Micro-level, sequential analyses of adult questioning and child responsiveness (Justice et al., 2002; Luo & Tamis-LeMonda, 2017; Tompkins, Bengochea, et al., 2017; Tompkins, Duffy, et al., 2019) and conversation analysis of discourse in pedagogical settings (e.g., between students and teachers) provide a useful framework for organizing and analyzing observational data in sequences. Specifically, conversations can be organized into exchange sequences comprised of initiations, responses, and feedback (sometimes referred to as follow-ups or evaluations), which predominate classroom discourse (known as ‘initiation-response-feedback’ or ‘question-response-evaluation’ sequences; Campuzano, 2018; Howe & Abedin, 2013; Mehan, 2014; Sinclair & Coulthard, 1975). Even caregiver-child conversations during play or shared book reading can be organized in such a way (Eason et al., 2021; Kurkul & Corriveau, 2018). Below is an example of an exchange between a caregiver and child that is initiated by a high-CD question in the context of sharing a wordless picture book. This exchange was captured by a coding scheme intended to uncover initiation-response-and follow-up sequences as they naturally occurred:

Parent (P): There are candles on the cake now. What is going to happen next? ➔ initiation

Child (C): They will leave the party. ➔ response

P: Are you sure? What do we usually do after we put candles on a birthday cake? ➔ follow-up

With micro-level, sequential analysis of dyadic conversations, researchers have examined how prior, current, and subsequent linguistic input relate to each other (Bottema-Beutel et al., 2020; Hellermann, 2003; Justice et al., 2002; Kurkul & Corriveau, 2018; Li, 2018; Schegloff, 2007; Tompkins, Bengochea, et al., 2017; Tompkins, Duffy, et al., 2019). Additionally, this method may illuminate children’s ability to respond to different types of caregiver input, caregivers’ scaffolding tendencies, and more generally, what components of semi-structured interactions may lend themselves to better learning.

The present study examines caregiver-child interactions in the context of a shared picture book task involving questions that are sequenced as follows: caregiver exchange-initiating questions, child responses, and caregiver follow-ups. We use a hierarchical linear modeling (HLM) approach to investigate how child and caregiver factors relate to caregivers’ high-CD questioning. Our data comprises of features of exchange sequences, including interaction time, whether the caregiver asked a high-CD question and provided support during the conversation, and whether the child responded in an exchange (level-1), nested in dyads and global factors, i.e., child cognitive skills and caregiver education (level-2). Specifically, we ask whether caregivers’ cognitively demanding questioning during dyadic interactions is driven by what happens during the conversation (i.e., “in the moment” factors) and/or what dyads bring to the conversation (i.e., global characteristics). In the sections below, we review literature on the associations between the aforementioned caregiver and child factors and caregivers’ questioning.

Adults’ questioning influences children’s responses

Exposure to the type of input, such as questions vs. statements or high- CD input vs. low-CD input, relates to how children respond (Chandler-Campbell et al., 2020; Danis et al., 2000; Eason et al., 2021; Justice et al., 2002). Specifically, adults’ low- and high-CD language input differentially relates to children’s responses (Danis et al., 2000; Justice et al., 2002; Kuchirko et al., 2016; Lee & Kinzie, 2012; Luo & Tamis-LeMonda, 2017; Tompkins, Bengochea, et al., 2017; Tompkins, Duffy, et al., 2019; Zucker et al., 2010), such that adults’ input tends to elicit child responses within the same level of abstraction or cognitive demand and high-CD input tends to elicit more conversational contributions from children. For instance, Kuchirko et al. (2016) found in a sample of mothers and their 3-year-old children followed over 2 years that mothers’ open-ended questions predicted children’s narrative contribution during shared book reading. The researchers’ definition of open-ended questions included high-CD prompts, such as questions that asked children to make inferences, predictions, judgements, or explanations about story elements beyond what could be seen on a page. Children contributed more if mothers asked more questions, and this association was stronger for open-ended (including high-CD) vs. referential (low-CD) questions. Overall, caregivers’ questioning predicted children’s own contributions during book reading concurrently and over time (Kuchirko et al., 2016).

Additionally, Luo and Tamis-LeMonda (2017) used lag-sequential analysis to examine caregiver questions and four-year-old children’s responses during shared book reading, and they found that caregiver questions elicited children’s narrative contributions and vice versa. Moreover, caregiver questions were likely to elicit child responses of the same cognitive demand or level. In general, high-CD questions may effectively “push” children to engage in cognitively demanding tasks or thinking to respond (Danis et al., 2000; Luo & Tamis-LeMonda, 2017).

Adults adjust their questioning based on “in the moment” factors

Children’s responses are an important “in the moment” factor that relates to adults’ language input, including how they provide subsequent prompts. Qualitative studies of adult-child interactions during formal learning settings illustrate that adults often adjust their language input or questioning based on children’s responses. In the context of exchanges, these inputs can constitute new initiations. When children give an appropriate or correct response, adults may provide prompts or questions at the same level of CD immediately after, as they consider children’s correct responses to reflect their knowledge state. They may also provide a question at a higher CD as they consider children’s correct responses to be a signal that they are ready to engage in higher order thinking. In contrast, when children give an inappropriate or incorrect response, particularly to high-CD input, adults may subsequently provide low-CD input (Lee & Kinzie, 2012).

Continuing with the same exchange represents another way that adults adjust their input based on children’s responses. Adults provide follow-ups or feedback that vary in directiveness and depend on whether they view children’s responses as appropriate or correct, which may provide children with information about their performance (e.g., Cullen, 2002). Specifically, adults may follow-up to children’s responses by echoing what children have said, affirming their response (Cullen, 2002; Reigel, 2008), expressing praise (Cullen, 2002; Gunderson, Gripshover et al., 2013; Gunderson, Sorhagen, et al., 2018; Morris & Zentall, 2014; Pomerantz & Kempner, 2013), sharing disagreements or corrections (Kamins & Dweck, 1999), or providing scaffolding, i.e., guidance in the form of questions, suggestions, strategies, explanations, or modeling to help the child understand and/or improve on their response (Bozkurt & Polat, 2017; Cullen, 2002; Dieterich et al., 2006; Hammond et al., 2012; Hyde et al., 2006; Lee & Kinzie, 2012; Lowe et al., 2013). Such feedback also functions to sustain the interaction, support children’s autonomy, and contribute to their engagement in and their motivation to continue the activity. However, little is known about how following-up affects caregivers’ subsequent language input.

It is less well-known how other types of caregiver input, such as the extent to which they provide feedback or scaffolding, relates to their subsequent questioning. For example, it is possible that caregivers view the need to scaffold as a signal that their child cannot answer challenging questions, so they are less likely to ask high-CD questions later. Alternatively, they may observe that their child successfully answered a question that required scaffolding and this motivates them to ask high-CD questions to provide their child with more opportunities to practice and strengthen their knowledge. The current study investigates child responses and caregiver follow-ups as two “in the moment” factors that may influence caregivers’ subsequent questioning.

Adults adjust their questioning based on global factors

Relations between adult questioning and children’s cognitive skills and behaviors

Next, we shift to considering how global factors, such as caregiver and child characteristics relate to adults’ questioning. Specifically, caregivers’ and teachers’ subsequent language input may be driven by children’s social and cognitive skills and behaviors, as adults may be aware that high-CD questions are more or less effective for children at different ages, cognitive ability, or social skill levels. Correlational analyses of the frequency of low- and high-CD adult language input, including questions, and children’s early academic skills suggest the possibility that adults are sensitive to their child(ren)’s competencies and adjusting their input based on them. Caregivers and teachers may perceive an association between children’s cognitive skills and their capacity to respond to or familiarity with high-CD questions and thus, they may ask more high-CD questions to children with greater language or math abilities (Duong et al., 2021; Huttenlocher et al., 2010; van Kleeck & Beckley-McCall, 2002; Kuchirko et al., 2016; Leech et al., 2013; Ribner et al., 2020; Rowe, 2013; Zucker et al., 2010). Additionally, adults may question based on children’s attentional or executive function (EF) skills or use questions as a general method to capture and sustain attention or interest (Kuchirko et al., 2016; Spruijt et al., 2018). We aim to replicate previous work on the domain-specific association between caregivers’ questioning and children’s language skills and explore whether caregivers’ perception of or sensitivity to their child’s (in)attention, which affects their behavior in social settings including interactions with their caregiver, drives what questions they ask their child.

Associations between caregiver education and questioning

Caregiver education has been strongly related to the types of language input that they expose their children to, including high-CD questions (Uscianowski et al., 2020), and this association may be driven by caregivers’ educational beliefs, goals, and experiences (Luo et al., 2021; Rowe et al., 2005; Silander et al., 2018; Uscianowski et al., 2020) (see Rowe, 2018 for a review) as well as socio-environmental factors that give rise to differences in parenting behaviors. Caregiver education is also associated with their cultural capital, which refers to various knowledge, abilities, and behaviors that signify an individual’s social status, enable social mobility, and can perpetuate socioeconomic inequities in education (Bourdieu, 2002; Breinholt & Jæger, 2020; Huang & Liang, 2016; Tan, 2017). Cultural capital may facilitate children’s early cognitive and academic skill development by equipping them with abilities that benefit learning, e.g., vocabulary skill and an interest in reading (Breinholt & Jæger, 2020). Overall, educational attainment may relate to a range of learning beliefs, experiences, and goals, as well as socio-environmental factors, which are in turn linked with caregivers’ (high-CD) questioning during dyadic interactions with their children. These links are further supported by work demonstrating that increases in caregivers’ educational attainment have been associated with improvements in children’s language skills (Magnuson et al., 2009). We hope to extend previous work on the associations between caregiver education and children’s exposure to different types of language input by simultaneously considering the effect of more proximal factors, such as child responses during these interactions.

An analytic approach for simultaneously considering “in the moment” and global factors

Overall, much of the research reviewed has examined “in the moment”, time-variant (e.g., child responses) and global, time-invariant influences (e.g., caregiver education) on caregiver-child conversations separately, despite the possibility that both of these factors are involved in shaping the nature of dyadic interactions. Specifically, “in the moment” factors represent temporally proximal influences on interactions while global factors may reflect what each partner brings to new conversations, including past interaction experiences and knowledge (e.g., how children responded to high-CD questions in previous conversations). Although lag sequential analysis is a valuable analytic approach to modeling “moment-to-moment” events within caregiver-child interactions that has been used in previous work (e.g., Luo & Tamis-LeMonda, 2017), it is limited in its capacity to (easily) account for global, time-invariant variables. Instead, methods like hierarchical linear models (HLMs) can be used to examine simultaneously how both “in the moment” and global factors relate to behaviors that occur during dyadic conversations, such as caregivers’ high-CD questioning.

The utility of using HLMs to engage in more nuanced explorations of caregiver-child interactions is demonstrated in a study by Callanan et al. (2020) in which caregivers and children engaged with a gear exhibit in a science museum. Using HLM as a tool to examine “moment-to-moment” talk, global caregiver-child interaction style (i.e., caregiver-, child-, or jointly-directed), and demographic factors (e.g., child age and caregiver income), the researchers found that caregivers’ prior causal language was positively related to the probability that children would engage in subsequent exploration of the gears and that caregiver-child interaction style and child age were associated with children’s overall tendency to explore. Although these findings may have been uncovered using several different approaches in tandem, e.g., lag sequential analysis for the “in the moment” variables and multiple linear regression for the global factors, HLMs allow us to statistically control for the effect of a variable at one level or unit of analysis while examining other variables at a different unit of analysis. Thus, we view the present study as an extension of past work by considering “in the moment” and global factors concurrently to predict caregiver questioning.

The present study

The current study investigates caregiver-child interactions during shared book viewing in the context of exchanges initiated by questions, which are comprised of three elements: caregiver exchange-initiating questions, child responses, and caregiver follow-ups. We use hierarchical linear models (HLMs) to examine how “in the moment” child and caregiver factors, namely interaction time, child responses, and caregiver follow-ups during these interactions, and global factors, such as children’s cognitive skills and caregivers’ education level, relate to caregivers’ high-CD questioning. This approach aims to address a gap in the literature on the potential influence of moment-to-moment changes in dyadic conversations, independently and in the context of global factors, on caregivers’ provision of high-CD input during informal learning activities with their children. Our research questions are as follows:

Research question (RQ) 1: Does the probability that caregivers ask high-CD questions change as a function of “in the moment” factors derived from dyadic interactions, including interaction time, the previous child response, and the caregiver follow-up in the last exchange?

We hypothesize that caregivers’ tendency to ask a high-CD question during shared book viewing increases over time within the interaction, given that caregivers and children will have more opportunities to discuss past events or summarize and compare information from various pages, which can be prompted with high-CD questions. Also, since adults sometimes adjust the CD of their questions based on children’s responses (e.g., Lee & Kinzie, 2012), we hypothesize that the likelihood that caregivers ask a high-CD question is greater if their child previously responded appropriately or directly, e.g., they made an attempt to address the question. Further, given that caregivers’ provision of feedback may serve as a similar signal about their child’s ability to respond to high-CD input, we expect caregivers’ high-CD questioning to be related to whether they previously provided a scaffolded follow-up. Finally, we statistically control for caregivers’ total verbal input because we anticipate that caregivers who talk more during dyadic interactions may ask more questions.

RQ2: Does caregivers’ overall tendency to ask high-CD questions relate to global child-level factors, such as child age and caregivers’ perceptions of children’s vocabulary and attentional skills, and caregiver factors, namely education?

Given the consistent relations between caregivers’ high-CD questioning and children’s language skills (e.g., Duong et al., 2021; Kuchirko et al., 2016), and the possibility that caregivers may use questions as a way to capture and sustain children’s attention in an activity (Spruijt et al., 2018), we hypothesize that caregivers’ general tendency to ask high-CD questions is positively associated with these child-level factors. Considering that caregivers’ education levels are associated with differences in the quality and quantity of question-asking during conversations with their children (e.g., Luo et al., 2021; Uscianowski et al., 2020), we hypothesize that caregivers’ general tendency to ask high-CD questions positively relates to their educational attainment.

RQ3: Exploratory analyses were conducted post-hoc to examine whether child responses related to caregivers’ probability of asking high-CD questions in certain situations for specific subgroups of the sample. Specifically, is the relation between the probability of caregivers’ high-CD questioning and the previous child response moderated by other “in the moment” factors (i.e., interaction time), or global factors such as children’s age, attention, and vocabulary, or caregivers’ total utterances and education?

Method

Participants

Data were derived from a study examining how caregivers promote learning in the home environment with preschool-aged children in a large, mid-Atlantic metropolitan area. Our original sample consisted of 128 families and data from 7 families were excluded from the current analyses due missing or insufficient (i.e., the caregiver only asked one question) video data (n=5), the presence of a non-target individual (e.g., sibling or another caregiver) in the video (n=1), or other examiner error (n=1). Missing survey data from 22 dyads were imputed for analyses (see the Statistical Plan section for more details) and thus, the final sample in this report consisted of 121 caregiver-child dyads (M child age=4;5, SD= 3.6 months, Range=4-4;11). We planned to collect a sample of 200 caregiver-child dyads, but this was interrupted by the COVID-19 pandemic. Our final sample of 121 dyads (and 2,992 exchanges) is comparable to the range of sample sizes seen in past work using a similar analytical method, namely lag-sequential analysis (e.g., Tompkins et al., 2017, n=49 dyads and n=903 utterances). Half of the children in the sample were male, and the majority of caregivers were mothers (n=106). Caregivers reported their race as White (77%), Black (12%), Asian or Pacific Islander (3%), or did not report this information (8%). Caregivers’ educational attainment ranged from 11 (high school diploma or G.E.D.) to 18 years (some graduate work or more; M=16.07 years, SD=2.15) and their average yearly household income was $107,018.19 (SD=$69,972.66, Median=$98,500, Range=$0-$350,000).

Measures and Procedures

The study procedures involved observations that were completed at participants’ homes during one of two home visits and an online survey, which asked parents to report on their children’s vocabulary skills, attention skills, and demographic information. Families were recruited through community flyer distributions and in-person tabling at preschools and daycare centers. These procedures were approved by the local institutional review board and caregivers gave written informed consent to participate in the study prior to engaging in any research activities.

Observations and transcription coding.

Caregivers and children were video-taped interacting with a wordless picture book, titled Fox’s Fun Day! (Figure 1), created by the research team with the primary goal of providing caregivers and children opportunities to discuss math- and spatial-related concepts. A variety of animals and objects were depicted in the book, including raccoons, birds, a birthday cake, and candles. The story followed a fox who was having a birthday party at a park; with each page, different animals arrived to the party and performed various actions including presenting the fox with presents and moving around the park. Toward the end of the book, a cake was brought to the party, candles were placed on the cake, and all of the animals gathered to celebrate. The numbers of animals and objects progressively increased with each page to elicit math talk about counting, labeling sets, and arithmetic, and birds were shown changing locations on each page to elicit spatial talk about locations and directions. Additionally, the birthday banner, which appeared on all pages, had alternating colors to elicit talk about patterns. Dyads were instructed to look at the book however they wanted for five minutes and were given additional time (up to three minutes) if they did not view all the pages to complete the activity. Thus, interactions could range from five to eight minutes and on average lasted 5 minutes and 48 seconds (SD = 1 minute and 53 seconds).

Figure 1:

Example pages from Fox’s Fun Day!

The video-recorded interactions were transcribed verbatim at the utterance level (Gunderson, Gripshover, et al., 2013; Gunderson, Sorhagen, et al., 2018; Pomerantz & Kempner, 2013) and coded by trained research assistants in the context of exchanges. Exchanges were defined as conversation segments about new topics, which occurred when attention was drawn to new objects, entities, events, actions, and/or concepts. They comprised of an exchange-initiating question, a response, and a follow-up or feedback (based on Kurkul & Corriveau, 2018). Caregivers’ questions were coded as either low- or high-cognitive demand (CD) based on previous work (Duong et al., 2021; Uscianowski et al., 2020). Low-CD questions related to perceptually present or immediate information and required a response low in cognitive demand, including labeling, locating, identifying, recalling information, counting, and completing sentences (e.g., “What color is that?”). High-CD questions required the respondent to think beyond perceptually present information and a response higher in cognitive demand such as predicting, summarizing, comparing, unifying a sequence of events, problem solving, or explaining (e.g., “Why do you think that happened?”). In cases where caregivers asked several consecutive exchange-initiating questions of low- and high-CD regarding the same topic, the high-CD code was prioritized. This typically occurred in the form of a low-CD question followed by a high-CD question (e.g., “What’s on the cake? What do you think will happen next?”). Few questions occurring in low frequency and unrelated to the task such as, “Can I go to the bathroom?” were not included in our analyses.

Children’s responses were coded as either responding to the question or not, and responses were further categorized into (1) direct/on-topic or (2) indirect/off-topic for exploratory analyses. Direct/on-topic answers included any attempts to address caregiver questions, regardless of whether the attempt was correct. Indirect/off-topic responses included turning the question back or saying something related to the topic of the question but not addressing it (e.g., a caregiver asks, “What kind of cake is that?” and the child responds, “I like cake” or “There’s a raccoon there”). Lastly, caregiver follow-ups referred to how caregivers reacted to children’s responses to their questions. These were coded for whether they included scaffolded content, such as providing an explanation or additional prompts to guide the child to an appropriate or desired response. Non-scaffolded follow-ups included confirming a child response, praising the child, and making a related comment about the question asked or child response (e.g., caregiver says, “I like cake too” in the previous example). Sometimes, caregivers followed up with questions during scaffolding. These questions were strictly coded as follow-ups and the use of caregiver questions in our analyses refer to exchange-initiating questions.

For the purposes of this study, analyses were carried out at the level of exchange-initiating questions. The total number of observations in this dataset is equal to the number of exchanges that were initiated by caregiver questions (n=2,504). See Table 1 for an example of exchanges and how the data were structured for analyses.

Table 1.

Example segment of caregiver-child exchange data (bolded observations’ utterances shown below)

Dyad ID	Time t from start (seconds)	Caregiver exchange-initiating question (CD at t)	Did the child respond directly in the previous exchange (at t-1)?	Did the caregiver provide a scaffolded follow-up in the previous exchange (at t-1)?
1	198.46	Low	Yes	Yes
1*	213.93	Low	Yes	No
1	240.31	Low	Yes	No
1	285.40	High	Yes	Yes
1	288.18	High	Yes	No
1 Λ	298.59	High	No	Yes
1	310.25	High	No	Yes
1	329.39	Low	Yes	Yes
1	331.70	High	Yes	No
1	340.22	Low	Yes	No
* Exchange example #1
Parent (P): What would you like to say about this page? ➔ Low-CD question
Child (C): The owl brings cake. Fox/ the foxes like cake. ➔ Direct, on-topic response
P: Yay. ➔ Non-scaffolded follow-up; end of the exchange
ΛExchange example #2
P: Why is it [the birthday party] a mess? ➔ High-CD question
C: Because. ➔ Indirect or off-topic response
P: Because why? ➔ Scaffolded follow-up
C: So-
P: What made it a mess?
C: Because of the birthday cake. ➔ End of the exchange

Twenty-five percent of all transcripts were double-coded and kappa statistics were computed to assess the degree of agreement between each pair of coders in identifying caregiver exchange-initiating questions, child responses, and caregiver follow-ups. See Table 2 for a detailed description of exchanges, summary statistics on the frequency of these codes within dyads, and reliability statistics.

Table 2.

Description and summary statistics of caregiver-child exchange codes

Code (κ)	Description	Total number		Proportion of exchanges
		M (SD)	Range	M (SD)	Range
Caregiver exchange-initiating questions		25.73 (13.23)	2-60	--	--
Low-CD (.70)	Relate to perceptually or immediately present information and/or requires a response of low cognitive demand (e.g., label, locate, identify, count)	20.27 (11.21)	1-53	.69 (.26)	.00-.98
High-CD (.85)	Require the respondent to think beyond perceptually present information and/or requires a response of higher cognitive demand (e.g., predict, explain, summarize, compare, problem solve)	5.45 (4.56)	0-21	.21 (.15)	.00-.61
Child responses		23.19 (12.60)	1-57	.88 (.10)	.44-.98
Direct, on-topic (.81)	Addressing the question with a low- and/or high-CD response, regardless of whether the response was correct	20.43 (11.44)	1-51	.78 (.12)	.39-.97
Indirect or off-topic (.86)	Adempts to get the caregiver to respond to their own question or change the topic	2.76 (2.36)	0-15	.10 (.08)	.00-.29
No response (.78)	A lack of response	1.55 (2.17)	0-9	.06 (.08)	.00-.50
Caregiver follow-ups		18.93 (10.54)	0-53	.72 (.15)	.00-.94
Scaffolded (.85)	Responses to children’s answers including a response to their own question, an explanation, or additional utterances in the form of questions or statements to guide the child to an appropriate answer	6.44 (4.20)	0-25	.25 (.11)	.00-.57
Non-scaffolded	Follow-ups that confirmed a child response, praised the child, and/or made a related comment about the question asked or child response	12.50 (7.72)	0-37	.47 (.15)	.00-.74
No follow-up	A lack of follow-up	5.80 (4.33)	0-20	.23 (.13)	.00-.62

Developmental vocabulary assessment for parents (DVAP).

Caregivers’ perception of children’s vocabulary skills was measured using the DVAP (Libertus et al., 2015), in which they were presented with 212 words ranging in difficulty (e.g., “girl”, “jumping”, and “parallelogram”) and instructed to indicate which words they had heard their child say. The words were derived from Form A of the Peabody Picture Vocabulary Test, 4^th edition (PPVT-4; Dunn & Dunn, 2007) and previous work has shown that the DVAP is a valid indicator of children’s expressive vocabulary, as DVAP scores highly correlate with children’s PPVT-4 performance (Libertus et al., 2015). The dependent measure was the total number of words indicated with a higher number indexing higher vocabulary skills, which was standardized.

Social Skills Improvement System (SSIS) Hyperactivity/Inattention Subscale.

Caregivers completed the Hyperactivity/Inattention subscale of the SSIS (Gresham & Elliott, 2008) as a measure of their perception of their child’s attentional skills. They were presented with 7 items describing a range of child behaviors (e.g., “My child has difficulty waiting their turn” and “My child gets distracted easily”) and instructed to indicate how often they observed those behaviors on a scale from 1 (almost always) to 4 (never). The dependent measure was the total summed score ranging from 4 to 28, with higher scores representing higher attentional skills; this value was z-scored for analyses (α = .79).

Caregiver and child demographics.

Caregivers completed questionnaires in Qualtrics after the home visit. They provided their child’s birth date to allow us to calculate children’s age in months at the first visit. Additionally, caregivers reported on background characteristics including educational attainment, which was converted to a numeric variable representing years of completed education (less than a high school diploma or GED=11 years, high school diploma=12 years, some college but no degree=13 years, Associate’s degree=14 years, Bachelor’s degree=16 years, Graduate degree=18 years). Child age and caregiver education were both z-scored for analyses.

Statistical Plan

Patterns of missing survey data were examined, revealing that child vocabulary (DVAP) scores had the greatest number of missing cases (n=14 or 10.94%), followed by child attention (SSIS) scores (n=13 or 10.16%). These data were missing at random; the caregivers in our sample did not start or finish the surveys that collected these and thus, they were imputed using the ‘Amelia’ package in R (Honaker, King, & Blackwell, 2011). Specifically, multiple imputation using the bootstrap expectation-maximization algorithm was applied to derive 10 imputed datasets and model estimates across all datasets were pooled by averaging.

Logistic HLMs, with exchanges nested in caregiver-child dyad, were estimated using the ‘Zelig’ package in R (Choirat et al., 2017; Imai et al., 2008) to predict caregivers’ high-CD questioning using language input factors derived from the interactions (e.g., the child’s response to the previous exchange), child factors (e.g., age), and caregiver factors (i.e., education). The outcome variable was whether each exchange-initiating caregiver question was low- or high-CD; it was treated as binary with 0 = low-CD and 1 = high-CD. Random intercepts for dyad ID were included to account for the non-independence of observations within dyads, as conversation patterns within dyads are likely to be more similar than conversations between dyads. The HLM covariance structure is an autoregressive process of order 1, which considers how immediately adjacent observations in time (i.e., times t and t−1) are likely to be more strongly correlated than distanced observations. A description of each model and their formulas are presented below.

To answer our first research question about the relations between proximal interaction factors and caregivers’ high-CD questioning, interaction time, whether children responded in the previous exchange, and the type of caregiver follow-up in the last exchange were included as level-1 predictors and total caregiver utterances were included as a level-2 control variable in Model 1. For each exchange-initiating question asked, time (in seconds) from the start of the interaction was calculated and standardized. This was included in the model to examine how caregivers’ questioning changed over time. The previous child response was included as a categorical variable, coded as 0 = no response, 1 = provided a response. Similarly, the last caregiver-follow-up was included as a binary variable (0 = no, 1 = yes) and reflected whether caregivers provided scaffolding. Lastly, the total number of parental utterances were derived from each dyad’s interaction and included in this model.

To address our second research question about the influence of child- and caregiver-level factors, such as child age, caregivers’ perception of children’s cognitive skills, and caregiver education, on caregivers’ tendency to ask high-CD questions, we included child age, attentional skills (SSIS scores), vocabulary (DVAP scores), and caregivers’ years of education, as level-2 predictors of level-1 intercept in Model 2. The formula for Model 2 which encompasses Model 1 is:

$$\begin{gathered} \mathit{High}-\mathit{CD}{Q}_{ij}\sim {\beta }_0+{\beta }_1{\mathit{Time}}_{ij}+{\beta }_2{\mathit{Child\; Response}}_{ij}+{\beta }_3\mathit{Caregiver\; Follow}-U{p}_{ij} \\ +{\beta }_4{\mathit{Caregiver\; Utterances}}_j+{\beta }_5{\mathit{Child\; Age}}_j+{\beta }_6{\mathit{Child\; Attention}}_j \\ +{\beta }_7{\mathit{Child\; Vocabulary}}_j+{\beta }_8{\mathit{Caregiver\; Education}}_j+b_{0i}+{\epsilon }_{ij} \end{gathered}$$

Lastly, exploratory analyses were conducted post-hoc to examine whether child responses related to caregivers’ probability of asking high-CD questions in certain situations or for particular subgroups of the sample. Specifically, three separate models were run with interactions between whether children responded to the previous exchange and interaction time (Model 3), child factors (Model 4), and caregiver factors (Model 5). The interaction between time and child responses was included in Model 3 to investigate whether caregivers were more or less likely to ask high-CD questions based on child responses throughout the interaction. The formula for Model 3 is below:

$$\begin{gathered} \mathit{High}-\mathit{CD}{Q}_{ij}\sim {\beta }_0+{\beta }_1{\mathit{Time}}_{ij}+{\beta }_2{\mathit{Child\; Response}}_{ij}+{\beta }_3\mathit{Caregiver\; Follow}-U{p}_{ij} \\ +{\beta }_4{\mathit{Caregiver\; Utterances}}_j+{\beta }_5{\mathit{Child\; Age}}_j+{\beta }_6{\mathit{Child\; Attention}}_j \\ +{\beta }_7{\mathit{Child\; Vocabulary}}_j+{\beta }_8{\mathit{Caregiver\; Education}}_j+{\beta }_9{\mathit{Child\; Response}}_{ij} \\ \ast {\mathit{Time}}_{ij}+b_{0i}+{\epsilon }_{ij} \end{gathered}$$

One possibility is that caregivers are less likely to ask high-CD questions toward the end of the interaction if their child did not respond to the previous exchange. A lack of response likely means the child does not know the answer or requires support, but caregivers may feel that they do not have enough time to provide scaffolding given the limited duration of this semi-structured activity.

Also, interactions between child factors, i.e., age, vocabulary, and attention, and child responses were included in Model 4 to examine whether caregivers chose to ask high-CD questions based on child responses for some children differing in those domains. The formula for Model 4 is:

$$\begin{gathered} \mathit{High}-CD{Q}_{ij}\sim {\beta }_0+{\beta }_1{\mathit{Time}}_{ij}+{\beta }_2{\mathit{Child\; Response}}_{ij}+{\beta }_3\mathit{Caregiver\; Follow}-U{p}_{ij} \\ +{\beta }_4{\mathit{Caregiver\; Utterances}}_j+{\beta }_5{\mathit{Child\; Age}}_j+{\beta }_6{\mathit{Child\; Attention}}_j \\ +{\beta }_7{\mathit{Child\; Vocabulary}}_j+{\beta }_8{\mathit{Caregiver\; Education}}_j \\ +{\beta }_9{\mathit{Child\; Response}}_{ij}\ast {\mathit{Child\; Age}}_j \\ +{\beta }_{10}{\mathit{Child\; Response}}_{ij}\ast {\mathit{Child\; Attention}}_j+{\beta }_{11}{\mathit{Child\; Response}}_{ij} \\ \ast {\mathit{Child\; Vocabulary}}_j+b_{0i}+{\epsilon }_{ij} \end{gathered}$$

It is possible that caregivers are less likely to ask high-CD questions to their child if they did not reply in the previous exchange and they are younger in age or have lower attentional or vocabulary skills. Caregivers may interpret a lack of response from these children to mean that they are unable to respond to “advanced” prompts (e.g., rather than being disinterested).

Lastly, interactions between caregiver characteristics and child responses, including education and total utterances, were examined in Model 5 to determine whether caregivers of different education levels and caregivers who spoke more or less during the interactions moderated the relation between children’s previous responses and caregivers’ high-CD question asking. The formula for Model 5 is:

$$\begin{gathered} \mathit{High}-CD{Q}_{ij}\sim {\beta }_0+{\beta }_1{\mathit{Time}}_{ij}+{\beta }_2{\mathit{Child\; Response}}_{\mathit{ij}}+{\beta }_3\mathit{Caregiver\; Follow}-U{p}_{ij} \\ +{\beta }_4{\mathit{Caregiver\; Utterances}}_j+{\beta }_5{\mathit{Child\; Age}}_j+{\beta }_6{\mathit{Child\; Attention}}_j \\ +{\beta }_7{\mathit{Child\; Vocabulary}}_j+{\beta }_8{\mathit{Caregiver\; Education}}_j+{\beta }_9{\mathit{Child\; Response}}_{ij} \\ \ast {\mathit{Caregiver\; Utterances}}_j+{\beta }_{10}{\mathit{Child\; Response}}_{ij} \\ \ast {\mathit{Caregiver\; Education}}_j+b_{0i}+{\epsilon }_{ij} \end{gathered}$$

Results

Descriptive statistics

Caregivers’ language input varied considerably; they produced an average of 118.99 utterances (SD=50.84, Range=15-259), 20.27 low-CD exchange-initiating questions (SD=11.21, Range=1-53), and 5.45 high-CD questions (SD=4.56, Range=0-21). Children primarily provided verbal responses to caregiver questions (Mean=23.19, SD=12.60, Range=1-57) and non-responses occurred infrequently (Mean=1.55, SD=2.17, Range=0-9). Caregivers provided an average of 6.44 supportive scaffolded follow-ups (SD=4.20, Range=0-25). Summary statistics of caregiver-child conversation variables are provided in Table 2.

Additionally, caregivers’ reports of children’s vocabulary and attentional skills varied widely. Specifically, caregivers indicated that their children knew an average of 105.08 words (SD=29.92, Range=11-179) and children’s mean attentional skills score was 20.63 (SD=3.50, Range=10-27). Zero-order correlations between caregiver and child language input and characteristics are presented in Tables 3 and 4.

Table 3.

Zero-order correlations between caregiver and child language input, non-imputed data

	1	2	3	4	5	6	7	8	9	10
1. Total caregiver Qs	--
2. Caregiver low-CD Qs	.94***	--
3. Caregiver high-CD Qs	.58***	.28**	--
4. Total child responses (resp)	.99***	.93***	.58***	--
5. Child direct resp	.97***	.93***	.53***	.99***	--
6. Child indirect resp	.57***	.47***	.50***	.56***	.41***	--
7. Child non-resp	.37***	.35***	.19*	.21*	.19*	.23*	--
8. Total caregiver follow-ups	.96***	.90***	.56***	.96***	.94***	.57***	.27**	--
9. Caregiver scaffolded follow-ups	.74***	.66***	.51***	.70***	.66***	.54***	.41***	.78***	--
10. Caregiver non-scaffolded follow-ups	.91***	.87***	.49***	.93***	.92***	.48***	.15	.94***	.52***	--
11. Caregiver non-follow-ups	.72***	.69***	.41***	.68***	.67***«	.36***	.46***	.49***	.35***	.49***

Note. Significance values include:

^***p<.001.

Table 4.

Zero-order correlations between caregiver and child language input and characteristics, non-imputed data

	1	2	3	4	5	6
1. Child age	--
2. Caregiver education	−.01	--
3. Yearly income	−.07	46***	--
4. Caregiver utterances	.02	.04	−.06	--
5. Child vocabulary	.21*	.45***	.23*	.00	--
6. Child attention	.02	.07	.20*	−.05	.17	--
7. Total caregiver Qs	.05	.09	.18	.59***	.07	.14
8. Caregiver low-CD Qs	.04	−.01	.13	60***	−.01	.08
9. Caregiver high-CD Qs	.06	.28**	.20*	.22*	.20*	.20*
10. Total child responses	.05	.10	.20*	.57***	.07	.14
11. Child direct resp	.08	.06	.18	.56***	.06	.13
12. Child indirect resp	−.10	.21*	.22*	.35**	.07	.17
13. Child non-resp	.03	−.03	−.07	.26**	.02	.01
14. Total caregiver follow-ups	.09	.13	.18	.62***	.12	.12
15. Caregiver scaffolded follow-ups	.16	.08	.13	.57***	.05	.11
16. Caregiver non-scaffolded follow-ups	.04	.14	.18	.54***	.14	.10
17. Caresiver non-follow-ups	−.06	−.05	.10	.27**	−.09	.14

Note. Significance values include:

^*p<.05

^**p<.01

^***p<.001.

The language input variables, i.e., caregiver questions, child responses, and caregiver follow-ups, are sums.

RQ1: Dyadic interaction factors

Model results are presented in Table 5. For ease of interpretation, standardized estimates of odds ratios (ORs) are presented. Overall, caregivers’ questions were more likely to be low-than high-CD. For every 3 high-level questions asked, there were about 10 low-level questions asked (intercept OR=.30, p<.001). Model 1 showed that interaction time was positively related to the probability that an exchange-initiating caregiver question was high-CD while controlling for total caregiver utterances. For every one standard deviation increase in time from start (57 seconds) caregiver questions were 1.23 times more likely to be high- than low-CD (p<.001). Also, caregivers who spoke less during the interactions generally asked more high-CD questions overall (OR=.80, p<.01). Whether children responded to or caregivers provided scaffolding in the previous exchange were not significant predictors of caregivers’ subsequent high-CD questioning.

Table 5.

Results of logistic hierarchical linear models relating caregiver questioning (1=high-CD question), “in the moment” (level-1 predictors) and global factors (level-2 predictors), imputed data

	Model 1	Model 2
Fixed Effects	OR [CI]	OR [CI]
Proximal, interaction factors
Time from start	1.23*** [1.11, 1.36]	1.22*** [1.10, 1.35]
Child response	.90 [.63, 1.27]	.89 [.62, 1.28]
Caregiver scaffolded follow-ups	.96 [.78, 1.18]	.96 [78, 1.18]
Caregiver utterances	.80** [.68, .95]	.79 [.68, .92]
Global child factors
Age		.95 [.82, 1.11]
Vocabulary		1.11 [.95, 1.31]
Attentional skills		1.11 [.96, 1.27]
Global parent factor
Education		1.28** [1.09, 1.50]
Random Effect (Intercept)	.30*** [.21, 42]	.29*** [.20, .42]

Note. Table values include odds ratios (OR) and confidence intervals (CI). Significance values include:

^*p<.05

^**p<.01

^***p<.001.

RQ2: Global caregiver and child factors

Model results are presented in Table 5. Model 2 shows that caregiver education related to caregivers’ overall tendency to ask high-CD questions (OR=1.28, p<.01) such that the proportion of high-CD questions (out of total questions) generally increased as caregivers’ years of education increased. Time and caregiver utterances remained significantly associated with caregivers’ high-CD question asking (OR=1.22, p<.001 and OR=.79, p<.01 respectively). In contrast, caregivers’ perception of children’s vocabulary skills and attentional skills and child age were not significantly related to caregivers’ tendency to ask high-CD questions.

RQ3: Post-hoc analyses on interactions with child responses

Model results are presented in Table 6. Consistent with past models, Models 3 and 4 show that parent utterances (OR=.79 in models, p<.01) and all models show that parent education (ORs=1.29-1.48, p<.05) related to parents’ high-CD questioning. Further, Model 4 shows that the interaction between parents’ perception of children’s vocabulary skills and whether children responded in the previous exchange related to parents’ high-CD questioning (OR=.62, p<.01). As shown in Figure 2, if children did not respond in the previous exchange, they were more likely to be asked a high-CD question in the subsequent exchange if their parents perceived them to have high vocabulary skills. In contrast, if children responded in the previous exchange, they were relatively equally likely to receive a subsequent high-CD question across vocabulary ability (Figure 2). No other interactions tested in Models 3 and 5 were significantly associated with parents’ high-CD questioning.

Table 6.

Results of post-hoc logistic hierarchical linear models relating caregiver questioning (1=high-CD question and interactions between “in the moment” and global factors, imputed data

	Model 3		Model 4		Model 5
Fixed Effects	OR	[CI]	OR	[CI]	OR	[CI]
Proximal, interaction factors
Time from start	1.05	[.80, 1.37]	1.22***	[1.10, 1.36]	1.22***	[1.09, 1.35]
Child response	.87	[.60, 1.25]	.97	[67, 1.39]	.92	[.64, 1.32]
Caregiver scaffolded follow-ups	.96	[.78, 1.18]	.96	[.79, 1.18]	.96	[.78, 1.18]
Caregiver utterances	.79**	[.68, .92]	.79**	[.68, .92]	.68	[.46, 1.00]
Global child factors
Age	1.05	[.91, 1.22]	.76	[.51, 1.13]	1.05	[.90, 1.22]
Vocabulary	1.12	[.95, 1.31]	1.75**	[1.23, 2.49]	1.12	[.96, 1.31]
Attentional skills	1.11	[.96, 1.27]	.85	[.62, 1.16]	1.11	[.96, 1.27]
Global caregiver factor
Education	1.28**	[1.10, 1.50]	1.29**	[1.10, 1.51]	1.48*	[1.04, 2.12]
Interactions
Child response*Time	1.18	[.90, 1.53]
Child response*Age			1.41	[.93, 2.12]
Child response*Vocab			.62**	[.44, .89]
Child response*Attention			1.33	[.95, 1.86]
Child response*Caregiver education					.85	[.59, 1.23]
Child response*Caregiver utterances					1.18	[.80, 1.75]
Random Effect (Intercept)	.30***	[.21, .42]	.27***	[.19, .38]	.28***	[.20, .40]

Note. Table values include odds ratios (OR), standard errors, and confidence intervals (CI). Significance values include:

^*p<.05

^**p<.01

^***p<001.

Figure 2. Predicted probability plot of caregivers’ high-CD questioning.

Note. Predicted probabilities are plotted as a function of caregivers’ perception of children’s vocabulary skills at the mean and −/+1 SD mean (low and high, respectively) and grouped by whether children responded to the previous exchange-initiating question

Additionally, the same model as Model 4 was estimated with child responses as a categorical variable consisting of 3 groups (i.e., responses were coded as 2=direct/on-topic, 1=indirect/off-topic, and 0=non-responses) and non-responses as the reference group. The only term that reached conventional significance was the interaction between direct/on-topic responses and vocabulary skills (OR=.61, CI [.43, .87], p<.05), suggesting that if children who were perceived to have high vocabulary skills responded directly in the previous exchange, they were less likely to be asked a high-CD question in the subsequent exchange than if they did not respond. As shown in Figure 3, the probability that caregivers asked a high-CD question in the next exchange for children with greater perceived vocabulary skills was greatest when children did not respond in the previous exchange, followed by when children responded off-topic or indirectly, and then when they responded directly. Moreover, the probability of receiving a high-CD question after responding on-topic or directly showed relatively little difference for children across the three levels of vocabulary skills. These probabilities showed more separation for children who responded off-topic or indirectly and were the most different for children who did not respond immediately in the previous exchange (Figure 3).

Figure 3. Predicted probability plot of caregivers’ high-CD questioning.

Note. Predicted probabilities plotted as a function of caregivers’ perception of children’s vocabulary skills at the mean and −/+1 SD mean (low and high, respectively) and grouped by whether children responded to the previous exchange-initiating question.

Discussion

The current study examined the likelihood of caregivers’ high-CD questioning during shared book viewing with their preschool-aged children based on “in the moment” factors (i.e., interaction time, child responses, and caregiver follow-ups) and global factors (i.e., caregiver education and children’s cognitive skills). HLMs revealed that overall, caregivers tended to ask more low-CD questions during conversations with their children, and interaction time and caregiver education were associated with caregivers’ high-CD questioning after controlling for total caregiver utterances. Additionally, post-hoc analyses revealed that children’s previous responses related to caregivers’ subsequent high-CD questioning, depending on caregivers’ perception of children’s vocabulary skills. Overall, both proximal factors (e.g., time) and global factors (e.g., caregiver education) are associated with caregivers’ use of high-CD questions, which may facilitate the development of children’s cognitive skills. Our study highlights the utility of taking a micro-analytic approach to studying caregivers’ provision of cognitively challenging input during dyadic interactions with their young children.

Dyadic interaction factors and caregiver questioning

Consistent with our hypotheses, interaction time related to caregivers’ high-CD questioning during shared book viewing while controlling for caregivers’ overall number of utterances. Specifically, the relation between interaction time and caregivers’ questioning, i.e., the probability of high-CD questions increased with time, may be attributable to the design of the task. The wordless picture book that dyads viewed increased in complexity with each page, such that animals, objects, and events continued to populate the pages. Thus, there was more opportunity to engage in high-CD conversations as dyads progressed through the book. For example, with time, caregivers had more information to ask about the increasing number of items on each page, comparisons of event sequences, predictions about what would happen next, or inferences about why something already occurred. This suggests that the structure of storybooks can partially drive children’s differential exposure to high-CD questions, which is consistent with work showing that the design of informal learning materials can elicit specific types of talk (e.g., Gibson et al., 2020). For instance, Gibson and colleagues (2020) manipulated caregiver talk about numbers and math during storybook interactions with their 2- to 4-year-old children in the home over a span of four weeks. Specifically, caregiver-child dyads were randomly assigned to receive storybooks that focused on small numbers (one to three), large numbers (four to six), or non-numerical vocabulary words. The storybooks were successful in eliciting caregivers’ talk about the targeted numbers; caregivers spoke a greater quantity of number words with the large (versus small) number books, demonstrating that the design and content of informal learning activities can influence caregivers’ provision of opportunities for children to engage in certain types of talk (Gibson et al., 2020).

Alternatively, the link between interaction time and the probability of caregivers’ high-CD questioning may be related to their beliefs about when it is appropriate to introduce talk about complex concepts or ideas. This may be reflected in how they tend to organize informal learning activities with their children, particularly with materials that they have not seen before (e.g., the book in this study). For instance, caregivers may prefer to scaffold interactions with their children while they view an unfamiliar book by initially introducing relatively simple or familiar concepts to “orient” their child to the activity before moving onto more cognitively demanding conversation topics. This would be consistent with work showing that caregivers tend to employ “task-organizing” talk at the beginning of activities with their children to provide them with a structure, role, and expectations, or “task-orienting” talk to guide children through the process or required steps of the activity, in an attempt to increase or maintain their self-regulated behaviors and attention (e.g., Son & Hur, 2020). Thus, the likelihood of caregivers asking high-CD questions is greater toward the end of informal learning activities, as they may prefer to spend the beginning of the activity orienting their child with low-CD questions.

These findings also demonstrate that caregivers do not employ high-CD input consistently throughout dyadic interactions with their children. Caregivers may ask low-CD questions throughout structured observation as a general communication strategy that serves to engage children in an activity. Consistent with this notion, in our sample, the frequency of caregivers’ low-CD questions positively correlated with caregivers’ total utterances (r(119) = .60, p < .001) and total questions (r(119) = .94, p < .001). In contrast, caregivers may employ high-CD questions at a lower frequency as these may be more likely to require scaffolding or lead to child disengagement or confusion. Thus, caregivers’ balance of low- and high-CD questioning may reflect their attempt to make the most out of brief, informal learning interactions with their child. Indeed, past work examining shared book reading interactions between middle-class caregivers and their 3- to 4-year-old children found that caregivers varied the cognitive demands of their input at four levels of abstraction or complexity ranging from low to high. Moreover, caregivers with children who had the greatest gains in language skills used a combination of low- and high-CD input; low-CD input helped create a sense of mastery for children during their conversations and high-CD input helped create opportunities for growth and learning (van Kleeck, Gillam, et al., 1997). Overall, these findings suggest that caregivers may vary their language input based on the type of activity they are engaged in with their child and the time that they have to spend on the activity.

Additionally, the association between caregivers’ total utterances and their overall tendency to ask high-CD questions (i.e., fewer utterances are associated with proportionally more exchange-initiating high-CD questions) may be driven by the types of caregiver and child responses that follow high-CD input. Caregivers who asked more high-CD questions may speak less overall because these types of questions typically require longer, more detailed responses from children. This is consistent with work showing that high-CD questions tend to elicit responses from children that are relatively longer (than those that follow low-CD questions), take on more complex sentence structures, and include more diverse vocabulary (e.g., Lee & Kinzie, 2012). High-CD questions are also more likely to require caregiver support or scaffolding, which can result in longer exchanges than those that do not require additional caregiver prompts. These exchanges may contain a mixture of low- and high-CD prompts or explanations that guide the child to the desired response.

In contrast, the probability of caregivers’ high-CD questioning did not relate to children’s responses or caregivers’ follow-ups to the previous question. One possibility is that caregivers may use more than the immediate or adjacent response or follow-up to guide their next question. Thus, the “signal” for whether it is appropriate to ask a question may come from a culmination of previous child responses or caregiver follow-ups. Alternatively, it is possible that our coding scheme did not capture the types of responses that caregivers actually used to determine when and how they ask questions. We took a conservative approach to coding child responses, indicating whether children attempted to directly and immediately address caregivers’ questions or not. In other words, whether children’s responses were “correct” or occurred during caregiver follow-ups were not considered. However, previous research has found that caregivers or instructors adjust their language input based on the (in)correctness of children’s responses, typically in form of scaffolding or corrections (Bozkurt & Polat, 2017), and children may have provided a direct or correct response after scaffolding. It is possible that caregivers may have used this information or type of response to determine how to question as well. We did not code whether children’s responses were correct to avoid having coders make subjective judgements in an activity where caregivers’ prompts did not always have a correct response. However, we recognize that caregivers likely evaluate their child’s responses this way and/or consider child responses that occur when they provide support, so it may be fruitful to analyze dyads’ conversations while they engage in a task with objectively correct behaviors or responses (e.g., a puzzle activity) and apply a more liberal coding scheme to capturing child responses immediately following a caregiver’s question as well as after the caregiver provides scaffolding.

Another possibility is that child responses matter to caregivers in certain situations or depending on their perception of their child’s language or attentional skills. We explored this possibility in post-hoc analyses and found that caregivers were more likely to ask a high-CD question in the subsequent exchange if their child did not respond in the previous exchange and if caregivers perceived them to have high vocabulary skills. In contrast, if children did respond in the previous exchange, caregivers’ likelihood of asking high-CD questions remained relatively constant for children of different vocabulary skills. In other words, caregivers’ perception of children’s vocabulary skills may be a factor that influences caregivers’ high-CD questioning in situations where the child is not responding. This behavior may reflect caregivers’ differing interpretations of a lack of response. For instance, caregivers who perceive their children to have low vocabulary skills may take non-responses to mean that their child does not have the capacity to answer their question, e.g., because their child does not know the definition of a keyword in the question, or their question is too advanced. Hence, caregivers may tend to move on and subsequently introduce low-CD input that their child is more comfortable with and capable of responding to. In contrast, caregivers who perceive their children to have high vocabulary skills may interpret a non-response to mean that their child is capable of responding but they may not want to (e.g., their attention is waning, or the activity is not engaging). Thus, caregivers persist in asking high-CD questions.

Caregiver education and questioning

Consistent with our hypothesis, caregivers with a relatively higher education level tended to ask more high-CD questions overall. This parallels past work on the association between caregiver language input, including high-CD questions, and caregiver educational attainment (e.g., Uscianowksi et al., 2020). Specifically, caregivers with higher educational attainment, a component of socioeconomic status, may have greater access or opportunities to acquire the knowledge, skills, and behaviors to provide high-CD questioning during interactions with their children, while caregivers with lower educational attainment have less exposure to and/or experience with this type of caregiver support of children’s early academic skills.

Limitations and future directions

Although the current study provides insights about “in the moment” and global factors that may drive caregivers’ questioning while engaging in shared book viewing with their children, there are several limitations. Given the possibility that caregivers may use a culmination of their past experiences to drive their subsequent questioning, future research should consider ways to capture and incorporate measures of children’s and caregivers’ cumulative responses and follow-ups and examine their relation to caregivers’ likelihood of asking high-CD questions. Additionally, our coding of child response types was relatively conservative in that it only captured whether children attempted to directly respond to caregivers’ questions. However, it is possible that specific types of direct responses, such as whether they are low- or high-CD, relate to caregivers’ subsequent high-CD questioning. For instance, if caregivers’ high-CD questions are met with high-CD responses from their child, caregivers may be more likely to subsequently provide high-CD prompts. The matching of CD in caregiver and child language input may serve as a signal that children are ready to or capable of engaging in conversations that require higher-level thinking, such as summarizing a sequence of events, making inferences, or explaining a process. This is consistent with past literature showing that caregivers’ questions and children’s responses reciprocally elicit one another across time (Danis et al., 2000; Luo & Tamis-LeMonda, 2017). Future work should explore specific types of child responses and the contexts in which they matter most for caregivers’ subsequent question input.

Further, our analysis of caregiver-child conversations was limited to exchanges that were initiated with questions. Thus, we did not consider children’s responses to exchanges that were started by (non-question) prompts or statements from the caregivers, which may have been equally meaningful to or relevant for caregivers’ decision to initiate the subsequent exchange with a low- or high-CD question. Some dyadic conversations contained a relatively low quantity of exchange-initiating questions even though each interaction lasted for approximately 5 minutes; 11 caregivers used only a total of two exchange-initiating questions and 11 other caregivers used three. Caregivers may have initiated exchanges with instructions or directions (e.g., “Look at those raccoons!” or “Point to the owl.”) to keep their child engaged in the activity or after asking a low- or high-CD question and receiving an unsatisfactory response. These initiations are less cognitively demanding (than questions) for children, as they require them to produce a specific action or speech (e.g., “Count the candles”), rather than allowing children to determine how to respond (e.g., “How many candles are there?” which can be answered by simply looking at the image in the picture book or counting the candles using their fingers). Children’s adequate responses to these non-question initiations may act as a signal to caregivers that it is appropriate to increase the complexity or demand of their prompts by presenting a question. Thus, future work should examine the sequential nature of the entirety of caregiver-child interactions, given that caregivers’ subsequent question-asking can be influenced by children’s reactions to different types or functions of caregiver input.

Additionally, we replicated past work on the sources of individual variability in caregiver language input by demonstrating a relation between caregivers’ educational attainment and high-CD questioning. We recognize that our sample was fairly highly educated and that the dyads we excluded from our sample had slightly lower years of educational attainment than the included group. This limitation should be addressed in future work by recruiting a more socioeconomically diverse sample. Relatedly, past work demonstrates that caregivers’ knowledge of children’s development across domains, e.g., language, is related to the complexity of caregivers’ language input during dyadic interactions with their infants. Specifically, caregivers’ developmental knowledge mediated the association between caregivers’ education level and their mean length of utterances during a free play session with toys (Vernon-Feagans et al., 2008). Thus, caregivers with greater knowledge about children’s language and vocabulary development may have age-appropriate expectations about their child’s ability to verbally contribute to informal learning conversations, which may affect how often and when they provide high-CD input.

Relatedly, we did not find a significant association between child age and caregivers’ high-CD questioning, which may seem surprising given that preschool-aged children experience age-related improvements in executive functioning and theory of mind (Marcovitch et al., 2015). Recent work has found bidirectional relations between children’s theory of mind and vocabulary (De Mulder et al., 2019), suggesting that children’s development of this social-cognitive skill may be partially shaped by and aid in children’s responses to high-CD input. Our study examined one conversation between dyads with a single, novel book and included only four-year-old children, so it is possible that any age-related differences were small and/or dominated by caregiver influences and may only be detectable if we observed other types of activities.

Also, our results may not generalize across cultures, as research has demonstrated cross-cultural differences in patterns of caregiver-child interactions during informal learning activities in the home, including storybook reading (Rochanavibhata & Marian, 2021; Wang, 2001; Winskel, 2010). For instance, Rochanavibhata and Marian (2021) examined dyadic conversations during book sharing between American and Thai mothers and their preschool-aged children and found that American mothers tended to employ more elaborative story-building language, such as using descriptions, extensions, action requests, recasting (e.g., the child says “doggy bed” and the caregiver says, “is the doggy under the bed?”), than Thai mothers, who used more attention directives (e.g., “Look there!”) and expansions (e.g., child says “frogs” and the caregiver says “they are frogs”). These differences in caregiver language input may be driven by the degree of individualism and collectivism within cultures, which relate to how much value is placed on certain communication styles. Elaborative language can support the development of children’s autonomy and agency, which individualistic cultures promote, through active participation in story-building, while directives can be used to teach children to respect and listen to their caregivers, which collectivistic cultures emphasize (Rochanavibhata & Marian, 2021). While the authors did not find cross-cultural differences in caregivers’ use of questions, it is possible that caregivers’ use of high-CD questions may vary, as this type of question can be more effective than low-CD questions at promoting children’s autonomy and agency. Specifically, high-CD questions provide children with the opportunity to “own” their ideas, thinking, and learning by requiring explanations, arguments, justifications, or evaluations (Rogat et al., 2014). Thus, future work should examine cross-cultural differences in caregivers’ high-CD questioning, as this research has the potential to highlight and inform resources that promote extant, culture-specific, and educationally beneficial interactions in the home between caregivers and children from different cultural backgrounds.

Conclusion

Caregivers’ high-CD questioning during informal learning activities with their children has the potential to support children’s developing cognitive skills and a critical next step was to understand what factors drive caregivers to ask these types of questions. We found that caregivers’ educational attainment was positively associated with their high-CD questioning during shared book viewing with their preschool-aged children. Further, in post-hoc analyses, we found that children’s responses to the previous caregiver exchange-initiating question related to the probability that caregivers asked a subsequent high-CD question. Children’s lack of response may matter to caregivers, serving as different indicators about their ability to respond to caregivers’ “in the moment” questions depending on children’s vocabulary skills. Our results provide nuanced insight into the contribution of “in the moment” and global caregiver and child factors to caregivers’ high-CD language input. Events that occur throughout the conversation and what partners bring to it (e.g., caregivers’ beliefs about children’s competencies) together shape the nature of interactions in a book sharing context where opportunities to engage in high-CD exchanges may increase as dyads progress through the pages. This is consistent with sociocultural theories of cognitive development (e.g., Gauvain et al., 2011) and a dynamic systems approach to understanding development (e.g., Cox & van Dijk, 2013), which emphasize the role of reciprocal and adaptive interactions between adults and children. Specifically, our results extend our understanding of how caregiver-child conversations, and potentially children’s cognition, develops by uncovering some factors that may impact when and how speakers reciprocate and adapt to each other in the context of shared book viewing and caregiver questioning. Last, this work has practical implications for the design of engaging, at-home activities such as books that leverage caregivers’ and children’s existing conversational tendencies and promote high-CD talk in ways that benefit children’s cognitive development. Overall, findings from the present study motivate future work to take a micro-analytic approach to studying caregiver-child interactions in the home during informal learning activities such as shared book viewing.

Highlights.

• Hierarchical linear models are used to study caregivers’ high-cognitive demand input

• Caregivers’ high-cognitive demand questioning relates to task time and education level

• Child responses and vocabulary combined relate to high-cognitive demand questioning