The influence of extrinsic and intrinsic variables on children’s reading frequency and attitudes: An exploration using an artificial neural network

Abstract

Children’s attitudes towards, and frequency of recreational reading influence their reading skill level. The aim of this study was to determine the relative influence of research-supported intrinsic and extrinsic variables that can shape this attitude and practice, and to investigate the use of artificial neural network as an adjunct approach in such analysis. Data from 997 Australian students in primary Years 4 and 6 were analysed to examine the influence of 10 variables on reading frequency and attitudes. Data were subject to analysis using artificial neural network, binomial logistic regression and linear regression. The results from the three methods of analysis indicate that library visitation, perception of importance and early literacy are the top three ranked variables in artificial neural network, with all three also significant (p < .001) in binomial logistic regression and linear regression. Providing students regular opportunity for library visitation may be a valuable intervention for educators and parents seeking to enhance children’s reading frequency and attitudes.

Keywords

Reading attitudes reading habits reading enjoyment quantitative research regression (statistics)statistics

Introduction

With reading skill acquisition and maintenance increasingly associated with academic, vocational, social, health and economic outcomes at an individual and societal level (e.g. Daggett & Hasselbring, 2007; Keslair, 2017; Kirsch et al., 2002; McIntosh & Vignoles, 2001), research into how reading can be enhanced in young people is an ongoing interest across nations. While numerous factors may influence reading skill development and progression, reading engagement may exert a noteworthy impact on student achievement (Guthrie, Klauda, & Ho, 2013; Merga, 2019a). Reading engagement has been conceptualised in diverse ways (e.g. Afflerbach & Harrison, 2017; Guthrie, Wigfield, & You, 2012; Kennedy, 2018), and such reading engagement constructs typically involve consideration of young people’s attitudes towards reading and their reciprocally related frequency of performance of the practice (e.g. Merga & Gardiner, 2019). Student reading behaviours and attitudes are positioned with bidirectional association with student reading achievement (e.g. Mullis, Martin, Kennedy, Trong, & Sainsbury, 2009), and research suggests that a positive relationship exists between reading engagement, and reading skill acquisition or performance (e.g. Guthrie, Wigfield, et al., 2012; Ho & Lau, 2018; Organisation for Economic Co-operation and Development [OECD], 2011a). For instance, Tse et al. (2016) note that ‘favourable reading attitudes are likely to involve children in regular reading activities and, in time, may benefit the development of reading skills as they may stimulate reading interest and intrinsic motivation in reading’ (p. 385). As such, reading engagement can be conceptualised as follows.

(from Merga & Gardiner, 2019, p. 39)

Reading engagement may also play a powerful role in mitigating socioeconomic disadvantage (Wrigley, 2017). Kirsch et al. (2002) note that

adolescents whose parents have the lowest occupational status but who are highly engaged in reading obtain higher average reading scores in PISA than students whose parents have high or medium occupational status but who report to be poorly engaged in reading. (p. 3)

As such, increasing children’s independent reading frequency and fostering positive attitudes towards reading is an educational priority from both a skill and equity perspective.

Reading of books in particular may be most strongly associated with literacy benefit. At this stage, book reading appears to offer greater literacy benefits than the reading of other text types, as reading of text types such as emails, social networking pages, text messages, newspapers and magazines is not associated with equal literacy skill advantages, and fiction books are particularly strongly associated with benefit (OECD, 2011b; Pfost, Dörfler, & Artelt, 2013; Spear-Swerling, Brucker, & Alfano, 2010; Zebroff & Kaufman, 2016). However, young people are reading far fewer books than in previous years (e.g. Twenge, Martin, & Spitzberg, 2018), so it cannot be assumed that they are highly engaged book readers.

The relationships between reading skill, reading frequency and reading attitudes are increasingly well-understood in the research, and diverse theoretical constructs account for the impact of intrinsic and extrinsic variables on reading engagement, such as expectancy value theory (e.g. Wigfield & Eccles, 2000) and the Matthewson and McKenna models (as explicated in McKenna, Kear, & Ellsworth, 1995). While these prior models are acknowledged, this study does not seek to confine its inquiry to the parameters defined by any of these models, instead choosing to select variables emerging from more recent research inclusive of Australian populations. More needs to be understood about the relative influence of these variables to ensure that young peoples’ educational opportunity is optimised.

Variables that may influence reading engagement

Research suggests that reading frequency and attitudes towards reading may be strongly shaped by a variety of extrinsic and intrinsic variables. Many variables have been previously implicated as potentially influential in this regard, including, but not limited to external stimulus such as friends’ attitudes and parental influence, school-related factors and students’ traits such as gender, age and reading skills. Research in this field is extensively reviewed in recent research books (Merga, 2019a, 2019c).

We confine our research variables to age, gender, year level, encouragement to read, early literacy experience, genre preference, perception of the importance of reading, shared reading opportunity, library visitation and family modelling, as these have been explored, or earmarked, for urgent exploration in both international and recent Australian research (Merga, 2019a, 2019c), and therefore could be expected to have currency and contextual applicability. These 10 selected variables are very briefly detailed in relation to the supporting research, as their relative influence will be further investigated in this paper.

Age and year level

Age and year level could be influential related intrinsic variables, with research finding a tendency towards declining attitudes towards, and frequency of reading as children age and move through the years of schooling, with children in the early years of schooling showing more positive attitudes towards reading, and greater reading frequency, than those in the later years (e.g. Merga, 2014b; Parsons et al., 2018; Scholastic, 2015, 2016a, 2016b).

Gender

Gender could also be a notable shaping intrinsic variable; in addition to the gap in literacy attainment between girls and boys, there is a gap in reading frequency and attitudes towards reading, with girls more engaged and outperforming boys (e.g. Merga, 2014b; OECD, 2015).

Encouragement to read

As noted by McKenna et al. (1995), children’s attitudes towards reading are shaped by ‘normative beliefs, beliefs about the outcomes of reading, and specific reading experiences’ (p. 939), and encouragement to read can counter expired expectations, whereby older children can deem reading no longer of value for them once encouragement declines, leading to reading infrequency (Merga, 2019a). It could thus be an influential extrinsic variable.

Early literacy experience

Where the early years of learning to read were positive and enjoyable, this can exert an ongoing effect on student engagement in later years, and therefore it has been included as an extrinsic variable. Children’s early literacy experiences can influence their reading skill development, and children who learn to read in an enjoyable and supportive context are more likely to read frequently and widely in subsequent years, value reading, and have a positive view of themselves as readers (e.g. Baker, 2003; Baker, Scher, & Mackler, 1997).

Genre preference

Research with adults suggests that readers who typically choose fiction may read more frequently than those who typically choose non-fiction (Merga & Mat Roni, 2018a), though the reading of fiction may be more strongly associated with literacy benefit for children than the reading of non-fiction (Jerrim & Moss, 2019). As an extrinsic variable, the reading of fiction may also offer a more deeply immersive experience (Mar, Oatley, & Peterson, 2009; Tamir, Bricker, Dodell-Feder, & Mitchell, 2015), which could influence student reading engagement, though there is limited research in this area that focusses on young people.

Perception of the importance of reading

Research typically drawing on expectancy value theory has found that young people’s perception of the importance of reading as an extrinsic variable can influence their engagement in the practice. In addition to other factors such as expectancies for success and ability beliefs, children’s subjective task values can shape their ‘choice, persistence and performance’ of activities, as well as influence ‘the extent to which they value the activity’ (Wigfield & Eccles, 2000, p. 68), and research suggests that children who see reading as important are nearly twice as likely to read very day (Merga & Mat Roni, 2018b).

Shared reading opportunity

While shared reading offers a range of benefits for young and older children across diverse literacy skills such as reading comprehension and vocabulary acquisition (Mol & Bus, 2011; Westbrook, Sutherland, Oakhill, & Sullivan, 2018), shared reading opportunities often dwindle early, while they are still enjoyed and valued by students (Merga, 2017). Shared reading opportunity as an extrinsic variable has been linked to fostering positive attitudes towards reading (Ivey & Broaddus, 2001; Merga, 2015a; Mol & Bus, 2011).

Library visitation

While there is a paucity of research on the impact of library visitation frequency on reading attitudes and frequency, this variable warrants inclusion, as while libraries are typically underutilised in schools (Merga, 2019a), research suggests that library visitation as an extrinsic variable offers benefits for student literacy performance.

Elementary schools with libraries visited by students this often or more averaged better CSAP reading performance than those visited less often. More students earned proficient or advanced reading scores and fewer students earned unsatisfactory scores where they visited their school libraries more often. (Francis, Lance, & Lietzau, 2010, p. iii)

Similarly, others have found that test scores are higher with higher school library usage (Lonsdale, 2003), and student achievement has been found to be boosted by greater student use of library, as measured by library visits and circulation rates (Hughes, Bozorgian, & Allan, 2014).

Family modelling

Indirect socialisation towards reading, such as having parents who model reading, can exert a positive influence on young people’s reading engagement as an extrinsic variable (Mancini & Pasqua, 2011; Mullan, 2010; Wollscheid, 2013). Recent research suggests that parent’s reading engagement may be the most robust predictor of reading engagement among selected parent involvement indices (Ho & Lau, 2018), and there may be ‘a significant and sustained relationship between parents’ reading-related knowledge and their children’s reading outcomes’ (Segal & Martin-Chang, 2018, p. 14).

Introduction to artificial neural network

While as illustrated above, intrinsic and extrinsic variables influencing reading frequency and attitudes are diverse, there is a paucity of research that explores the relative importance of these 10 variables in relation to each other, such as those aforementioned. This research is essential to inform appropriately targeted interventions and resourcing allocation to increase children’s reading frequency.

While this paper seeks to contribute to the research around variables supporting reading frequency and positive attitudes towards reading, perhaps its greater contribution is that it seeks to make visible the limitations of binomial logistic regression (BLR) and linear regression (LR) and explore use of an alternative, but far less frequently employed analytical tool: artificial neural network (ANN). While ANN is not yet frequently employed in the literacy education research space (e.g. Chang, Plaut, & Perfetti, 2016), it has been used in some contexts, such as for predictive modelling of self-regulated learning (e.g. Sabourin, Shores, Mott, & Lester, 2013). A study assessing student achievement prediction (Lykourentzou, Giannoukos, Mpardis, Nikolopoulos, & Loumos, 2009) found a neural network reached ‘higher correlation at all prediction stages, while their error was approximately half the error of linear regression’ (p. 380). It is also increasingly being used to find complex patterns in data in the sciences (e.g. Ali, Alharbi, Alothman, Badjah, & Alwarthan, 2018) and accounting and finance spaces for bankruptcy and financial fraud prediction (e.g. Ahmadpour Kasgari, Divsalar, Javid, & Ebrahimian, 2013; Omar, Johari, & Smith, 2017), where it is has been found to deliver more accurate results than other tools and methods (Alaka et al., 2018).

ANN is a computing system inspired by the networks in the brain. As noted by Esfe, Saedodin, Sina, Afrand, and Rostami (2015)

in complicated systems with several effective input parameters, artificial neural network (ANN) can predict output data. ANN is inspired from the human brain in order to process data and information. It includes integrated process units called neurons that can process input data. The multi-layer perceptron neural network includes input layer, hidden layer, and output layer. (p. 51)

Therefore, ‘by applying algorithms that mimic the processes of real neurons, we can make the network “learn” to solve many types of problems’ (Krogh, 2008, p. 195). For those unfamiliar with the tool, descriptive details are provided in the methods herein. The introduction of this new tool is timely, with increasing suggestions from expert statisticians that researchers should not overstate the meaning of their statistical findings, and that researchers should become increasingly open and thoughtful in their uses of statistical tools, and not inflexibly adhere to accepted ways of exploring data (Wasserstein, Schirm, & Lazar, 2019). In this space, new ideas and approaches derived from sound but novel reasoning could be expected to begin to thrive, and this paper invites a consideration of the place of ANN in the future of educational data analysis. Drawing on data from children in upper primary school in Western Australia, the paper explicates this method, while exploring the relative importance of variables that could influence reading frequency and attitudes towards reading.

Method

Participants and procedures

The Western Australian Study in Children’s Book Reading (WASCBR) explored children’s book reading practices in Years 4 and 6 at 24 Western Australian primary schools. It explored a range of research questions through a mixed-methods approach, with a particular focus on social influences on fostering recreational book reading, and children’s book reading preferences. It should be noted that use of the term reading in the methods, results and discussion, refers to a specific type of reading. In this instance, the focus is on the independent reading of books, due to the aforementioned differences in benefits conferred by reading of diverse text types.

As per Table 1, students in upper primary Years 4 and 6 were recruited from schools that were sampled using a purposeful sampling technique, to secure representative diversity in geographic location and socioeconomic context. Characteristics of participating schools are detailed in Table 1.

Table 1.

Characteristics of the 24 participating schools.

	N
School type
Public/government	19
Private	5
Coeducational	24
School location
Metropolitan area^a	21
Rural^b	3
Index of Community Socio-Educational Advantage^c
Highest	1132
Lowest	918
Range	214
Mean	1040.9

^aSchools within the Perth Statistical Division as defined by the Australian Bureau of Statistics, with Perth being the capital city of Western Australia.

^bSchools outside the metropolitan area.

^cAs per 2015 (Australian Curriculum, Assessment and Reporting Authority [ACARA], 2016).

While both survey and interview data were collected for analysis in the WASCBR, this paper focusses on the quantitative data from the survey tool, and this yielded N = 997 survey responses. As per Table 1, at 1040.9, the Index of Community Socio-Educational Advantage (ICSEA) of participating schools was just above average (1000). The ICSEA is calculated in relation to students’ parental education and occupation, as well as the proportion of Indigenous students and geographic location of Australian schools (ACARA, 2015).

All consenting students took part in the surveys, with descriptive detail on these participants in Table 2.

Table 2.

Characteristics of N = 997 survey participants by percentage.

Answer choices	Responses (%)	Responses (N)
Year group
Year 4	54.5	542
Year 6	45.5	453
Age
8	6.3	63
9	46.0	458
10	7.0	70
11	39.8	396
12	0.9	9
Mean 9.83
Gender
Girl	56.9	566
Boy	43.1	429

Institutional ethics approvals and Department of Education permissions to conduct research onsite were sought and granted, and the research tools were subsequently piloted in a robust sample at a local school (N = 100). During the pilot, children were also asked about their interpretation of the interview and survey items, ease of use and comprehension after using the tools, so direct student and teacher feedback helped to shape the final tools. As explored previously, this piloting phase was essential as the children showed unpredicted tendencies towards misrepresentation of their age and they had issues with understanding some of the original terminology (Merga & Mat Roni, 2017). Pilot data were not used in the final data set. Data collection occurred at participating schools, with the Chief Investigator, Merga, acting as the survey invigilator on site. Data were collected between 23 March and 21 June 2016 from students who had provided both individual and parental consent.

Research questions and analysis

As previously noted, the WASCBR explored a range of diverse research questions. In this paper, the focus is on the following research questions.

What is the relative importance of variables affecting reading frequency in children?

What is the relative importance of variables affecting attitudes towards reading in children?

How does ANN compare with the more traditionally applied BLR and LR for both reading frequency and reading attitudes?

As explored in the introduction, reading frequency and attitudes towards reading may be influenced by a variety of variables, including (but not limited to) age, gender, year level, encouragement to read, early literacy experience, genre preference, perception of the importance of reading, shared reading opportunity, library visitation and family modelling. The data on reading frequency, reading attitudes and these variables were collected on the following survey items (Table 3).

Table 3.

Variables and survey items.

Answer choices	Code	Survey question	Response options
Reading frequency (Outcome variable/output layer)	Q4. ReadFreq	Q4. How often do you read books in your free time?	Never/Sometimes/Often/Every day
Reading attitude (Outcome variable/output layer)	Q23. Attitude	Q23. Do you like reading books in your free time?	Yes/No
Predictor variables/input layer
1. Gender	Q1. Gender	Q1. Are you a:	Girl/Boy
2. Age	Q2. Age	Q2. How old are you right now?	8/9/10/11/12
3. Year level	Q3. Year	Q3. What is your school year?	Year 4/Year 6
4. Encouragement to read	Q6. Encourage	Q6. Does someone you know encourage you to read?	Yes/No
5. Shared reading opportunity	Q8. SharedRead	Q8. Does anyone read books to you at home?	Yes/No
6. Library visitation	Q16. LibVisit	Q16. How often do you visit a school or public library to choose books to read?	Every day/Often/Sometimes/Never
7. Family model	Q19. FmlyModel	Q19. Does anyone in your family like reading books?	Yes/No
8. Genre	Q21. Genre	Q21. If you have to choose a book to read, what type of book do you prefer?	Story books/Books about information and facts/Either story books or books about information and facts – you don’t mind
9. Perception of importance	Q28. Importance	Q28. Do you think that reading books is important?	Yes/No
10. Early literacy experience	Q29. EarlyLiteracy	Q29. Was learning to read fun for you?	Yes/No

We ran ANN on these data, using multilevel perceptron (MLP) to identify the relative importance of the 10 variables that research suggests could influence reading frequency and attitudes in young people. As previously noted, ANN is an artificial intelligence technique which is modelled on a biological neural network in a human brain. The neurons in the human brain act as processing units (nodes) which are connected by synapses creating a large network to process inputs. An output from one neuron or node is an input for the other.

The way ANN calculates the weight is called a network architecture. We chose the MLP as the network architecture, as it is extensively used to map nonlinear relationships between predictor and output variables in function fitting, pattern recognition and prediction (Omar et al., 2017). As noted elsewhere, the MLP ‘is a type of ANN that is trained using supervised learning procedures’ (Hodo et al., 2016, p. 3). MLP essentially is a feedforward architecture where an output from previous layer is fed into the next layer for processing; the synapses or connection weights are adjusted to reduce output errors by comparing the produced and target outputs. The errors at the output layer propagate back to the hidden layer until they reach the input layer in a process called backpropagation iterative. This iterative process happens in the ANN training sample, where it stops when the average error between the intended and the actual outputs is less than a predetermined threshold (Park, El-Sharkawi, Marks, Atlas, & Damborg, 1991). The dataset is split into two parts – the training set and the test set. A typical dataset split is 70/30, where 70% of the data are used to train the network to find an optimal solution, and the 30% of the data are used to test the solution.

In order to enhance understandings of this tool, we illustrate a typical topology of an ANN in Figure 1, which consists of three layers – input, hidden, and output. Each layer has multiple nodes represented by the circle and each node is connected by multiple synapses which are represented by the arrows. The arrows grow thicker as more statistical weight is assigned when the connection is valid. This process will trigger the other nodes in subsequent layer to fire up.

Figure 1.

A typical artificial neural network.

The ANN model for this study comprises of one input layer of 11 nodes, one hidden layer with seven nodes and an output layer with three nodes (one scale variable and one two-level dichotomous variable). We summarise the network architecture in Table 4 with details of the parameters in a form of syntax in Appendix A.

Table 4.

ANN network architecture.

	Number of units	Rescaling for covariates/scale dependents	Activation function
Input layer	Seven factorsThree covariates	Standardised
Hidden layer	One layerSix units		Hyperbolic tangent
Output layer	Three units	Standardised	IdentityError function: Sum of squares

ANN: artificial neural network.

Unlike conventional statistical methods where relationships between input and output variables are determined through predefined algorithms, ANN processes data in an unstructured manner, often through learning from data, and refining the solutions during the training session. ANN enhances the correlations between the predictor and output variables through this iterative process. Once ANN finds an optimal algorithm, it uses this algorithm onto a test dataset in a test phase to produce the statistical estimates. The training session for the current study uses batch type with scaled conjugate gradient as the optimisation algorithm. The stopping rule for the training phase to find the optimum algorithm is included in Table 6.

From a statistical point of view, ANN does not have an emphasis on data distribution assumption (i.e. whether or not the data are normally distributed), and the method has been found to be more accurate in identifying solutions in complex situations. For example, Alaka et al. (2018) argued that ANN is less susceptible to multicollinearity among predictor (input) variables when compared to (linear) regression method. A severe multicollinearity in a regression model can result in unstable parameter estimates, making the output less accurate. The authors also found that ANN is suitable when a relationship among variables is not predefined to fit into certain pattern (e.g. linear or quadratic). Because of these attributes, the inner workings of ANN in processing the data are less clear. This less transparent feature makes the technique particularly useful in situations where interactions between input and output variables are complex and undefined with no clear algorithmic solutions (Ahmadpour Kasgari et al., 2013).

We subsequently compared the results from ANN against an output from conventional statistical approaches – BLR for dichotomous output variables (i.e. attitude) and LR for scale-dependent variables (i.e. reading frequency). We ran BLR and LR as a robustness check, and to compare the performance of these three methods in identifying main factor(s) affecting the reading frequency and attitude towards reading among the children. BLR is commonly used in studies when there are two possible outcomes (i.e. binary). In our study, the respondents were divided into two groups for reading attitude. These are keen readers, who claim to enjoy reading books in their free time, and reluctant readers, who do not enjoy reading books in their free time. BLR is therefore suitable to model the predictor–outcome relationship with attitude as the binary response variable. We also use LR to explain the reading frequency by factoring 10 predictor variables. This is because reading frequency is a four-level scale-dependent variable. We present the results of these two methods in the subsequent section.

Results

There are four scale variables in our study. These are reading frequency (Q4.ReadFreq), age (Q2.Age), year (Q3.Year) and library visitations (Q16.LibVisit). We present the descriptive statistics for these and other variables in Table 5.

Table 5.

Descriptive statistics.

	Minimum	Maximum	Mean	Std. Deviation
Q4.ReadFreq	1.00	4.00	3.03	.88
Q23.Attitude	1.00	2.00	1.17	.38
Q1.Gender	1.00	2.00	1.43	.50
Q2.Age	1.00	5.00	2.83	1.06
Q3.Year	1.00	2.00	1.46	.50
Q6.Encourage	1.00	2.00	1.31	.46
Q8.SharedRead	1.00	2.00	1.59	.49
Q16.LibVisit	1.00	4.00	2.50	.71
Q19.FmlyModel	1.00	2.00	1.07	.25
Q21.Genre	1.00	3.00	1.97	.96
Q28.Importance	1.00	2.00	1.05	.22
Q29.EarlyLiteracy	1.00	2.00	1.16	.37

Artificial neural network

ANN splits the dataset, N = 997 (valid: n = 958, missing: n = 39) into two subsamples. These are the training sample (n = 658, 68%) and test sample (n = 308, 32%). Table 6 provides a model summary of the analysis for both training and testing samples. In the training sample, the model correctly categorises 86% of cases with 97% accuracy in predicting positive attitude (i.e. Yes group) and 24% accuracy in predicting negative attitude (i.e. No group). In the test sample, the overall accuracy is 81% (prediction accuracy for positive attitude: 98%; negative attitude: 19%). In both training and test samples, ANN consistently performs well in predicting positive attitude, but indicates low accuracy (i.e. less than 30% accurate) when categorising negative attitude with 10 predictor variables in the model.

Table 6.

Classification and model summary.

Sample	Observed	Predicted		Percent correct/relative error/sum of square
		Yes	No	Percent correct/relative error/sum of square
Training	Yes	533	15	97.3%
	No	78	24	23.5%
	Overall percent	94.0%	6.0%	85.7%
	Stopping rule	1 consecutive step(s) with no decrease in error^a
	Sum of squares			327.981
	Relative error			0.793
Testing	Yes	236	5	97.9%
	No	54	13	19.4%
	Overall Percent	94.2%	5.8%	80.8%
	Sum of squares			180.84
	Relative error			0.769
	Accuracy	.808
	Precision	.814
	Recall	.979
	Specificity	.194
	F1-score	.889
Categorical DV: Q23.Attitude
Scale DV: Q4.ReadFreq

^aError computations are based on the testing sample.

Note: Accuracy, precision, recall, specificity and F1-Score are based on the testing sample.

Despite the low accuracy prediction for the No group, the receiver operating characteristic (ROC) curve for the binary dependent variable (Q23.Attitude) suggests that the ANN performance is relatively good. The ROC curve is plotted with accuracy (i.e. true positive rate) on the y-axis against a false positive rate (1-specificity) on the x-axis, which is illustrated in Figure 2. In this diagram, the areas under the curve (AUC) for Yes and No predictions are similar (Yes = .762, No = .762). By comparing Cohen’s (1988) d effect size of small (.20), medium (.50) and large (.80), the AUC of .762 not only demonstrates the predictability of the ANN model, it also implies that the model has a large effect size¹ (see Rice & Harris, 2005).

Figure 2.

ROC curve.

The ANN results also indicate that library visitation (Q16.LibVisit) is the most influential factor in the model, with attitude and reading frequency as the outcome variables. This is illustrated in Figure 3. This is followed by perception of importance (Q28.Importance) and early literacy experience (Q29.EarlyLiteracy), both of which have more than 60% normalised importance.

Figure 3.

Relative importance of predicators ranked by ANN.

Linear regression

The 10 predictor variables in the LR model account for 21% of the variance in students’ reading frequency, R² = .21, adj. R² = .20. The model also indicates that 5 of 10 predictors are statistically significant. Perception of importance (Q28.Importance) is identified as the most influential factor shaping the reading frequency (β = .52, p < .001), followed by early literacy (Q29.EarlyLiteracy: β = .48, p < .001), library visitations (Q16.LibVisit: β = .30, p < .001), encouragement (Q6.Encourage: β = .29, p < .001) and genre (Q21.Genre: β = .06, p < .05).

Age and year were both highly correlated with the variance inflation factor (VIF), which is in excess of 7.0. This is expected as these variables relate to the age of the young respondents. A more conservative estimate for VIF is 3.3 or less (Djajadikerta, Mat Roni, & Trireksani, 2015), although some scholars tolerate VIF <10.0.

Binomial logistic regression

The BLR model is statistically significant against a constant only model, χ² (11) = 158.06, p < .001. Similar to the LR in the magnitude of explained variance in reading frequency, the BLR model explains a small 25.1% of the variance in children’s reading attitudes, Nagelkerke R² = .25; and correctly classifies 84% of all cases. Out of 10 predictors, BLR identifies library visitations (Q16.LibVisit) as both a significant and the most dominant predictor of the children’s attitude towards reading, Exp(B) = .51, p < .001. The BLR model also indicates early literacy experience (Q29.EarlyLiteracy) and perception of importance of reading (Q28.Importance) as other significant factors.

Results comparison

The top three factors affecting attitude and reading frequency among children identified by ANN are consistent with the results from BLR and LR model. These are library visitation, perception of importance and early literacy. We summarise the results of these statistical procedures in Table 7.

Table 7.

Comparison of the results from ANN, BLR and LR models.

Factors (predictors)	ANN (relative importance)	Binomial logistic regression (DV: attitude)	Linear regression (DV: reading frequency)
Library visitation	Ranked 1	Significant, p < .001	Significant, p < .001
Perception of importance	Ranked 2	Significant, p < .001	Significant, p < .001
Early literacy experience	Ranked 3	Significant, p < .001	Significant, p < .001
Family model	Ranked 4	Not significant, p = .15	Not Significant, p = .06
Encouragement to read	Ranked 5	Not significant, p = .11	Significant, p < .001
Age	Ranked 6	Not significant, p = .17	Not significant, p = .70
Year level	Ranked 7	Not significant, p = .79	Not significant, p = .28
Genre	Ranked 8	Significant, p < .05^a	Significant, p < .05
Shared reading opportunity	Ranked 9	Not significant, p = .67	Not significant, p = .16
Gender	Ranked 10	Not significant, p = .22	Not significant, p = .07

^aGenre as a whole is not statistically significant. However, preference on story book genre is statistically significant.

DV: dependent variable; ANN: artificial neural network; BLR: binomial logistic regression; LR: linear regression.

Despite the similarities of the results across all three tests (as indicated in Table 7), we find that there is a slight difference in the importance of predictors ranked by ANN and LR. ANN shows that library visitation is the most influential factor affecting reading frequency, while LR suggests it is perception of importance with the largest beta value. It is a common practice in LR to gauge the strength of predictor variables by looking at the regression coefficient – the beta value. In our study, perception of importance’s beta coefficient is .52 compared to .30 for library visitation. However, after a closer inspection of semi-partial correlation of these variables, unique variance in reading frequency relating to library visitation (sr² = .05) is larger than perception of importance (sr² = .02).

Discussion

The ANN, BLR and LR can be used to identify factors and predict a binary output – an output with two possible outcomes. In our study, we used 10 predictor variables to predict children reading frequency and attitude. We fed these variables into ANN, LR and BLR models and compared the results (Table 5). Essentially, this three-method approach allows us to identify variables that can inform which variables should be targeted in interventions to improve the children’s reading frequency. All three methods identify that library visitation strongly correlates with the reading frequency and positive attitude towards reading. Although we initially used BLR and LR as robustness checks for the ANN procedure, the pattern of the results produced by these methods also warrants closer attention. The conventional statistical methods, BLR and LR identify 4 of 10 factors as having statistical significance affecting children’s attitude on reading and reading frequency. For example, one variable, family model, which is ranked fourth by ANN in relative importance, is not identified as a significant contributor to attitude on reading in LR. Similarly, LR does not indicate the family model as an important factor affecting reading frequency. Without the inclusion of ANN, scholars could draw an incorrect conclusion based on overreliance on p-values (Wasserstein et al., 2019). While inclusion of ANN to the method does not mitigate the pertinent issues raised by Wasserstein et al. (2019), it does enable for higher confidence in the findings, though limitations still apply.

As summarised in Table 6, the differences in the results indicate potential misallocations of personal or organisational resources to support children’s reading engagement if there is a sole reliance on regression models. From a methodological perspective, we opine that scholars may inadvertently disregard the importance of certain variables in the model if only BLR and LR outputs are considered. This is simply because the BLR and LR models in our study explain merely 25% of variations in children’s attitude towards reading and 21% on reading frequency, respectively. This level of variation can be interpreted as having relatively small effect, or perhaps the model does not accurately capture contributing factors. Use of the three analytic approaches enables us to contend that library visitation, children’s perception of the importance of reading and early literacy experience may be the most important variables to account for in interventions seeking to strengthen student reading frequency and attitudes, and related literacy performance. This finding is particularly promising, as whereas intrinsic variables (such as age) cannot be shaped by external intervention, library visitation, children’s perception of the importance of reading and early literacy experience can all be promoted by influential social agents such as teachers and parents (Merga, 2019a). In addition, as both educators and parents may play a valuable role in influencing these variables, this highlights the importance of ongoing efforts to strengthen home and school partnerships to promote library visitation, children’s perceptions of the value of reading beyond skill acquisition and positive early experiences of learning to read.

The finding that library visitation had the strongest positive association with reading frequency and attitudes was unanticipated and intriguing, as this is an area that has been neglected in education research, though it has noteworthy educational implications. Library research with educational implications is infrequently reported beyond library research space (e.g. Hartzell, 2002), raising ‘a need within the school library and educational research community for more systematic, rigorous and collaborative research on factors related to school libraries and their effect on student achievement’ (Stefl-Mabry, Radlick, Armbruster, & Keller, 2016, p. 1). School libraries are underfunded and devalued (Burns, 2016; Kachel, 2015), and library staff face notable barriers to supporting students’ literacy and literature learning (Merga, 2019b). Therefore, it is of high importance that this research be promoted in the education space, to increase understandings of the power of access to the library in an effort to help shape student reading frequency and attitudes towards reading.

As previously noted, the influence of children’s perception of the importance of reading on reading engagement aligns with the construct of expectancy value theory. As the second highest ranked variable, this finding suggests that greater attention needs to be paid to recent research suggesting that not all children understand the continuing value of reading beyond independent reading skill acquisition (Merga & Mat Roni, 2018b). As this perception relates to reading frequency, more must be done by social influences such as teachers and parents to foreground the early and ongoing importance of reading. It has been noted that for children to see reading as important, they must understand the benefits conferred by the practice, and research also suggests that children may know little of these benefits (Merga & Mat Roni, 2018b). Educators should also ensure that parents are familiar with these benefits and committed to communicating the ongoing value of reading beyond the early years. As reviewed elsewhere, these include benefits for the development of a diverse range of literacy skills (Merga, 2019a); however, attention should also be given to the likelihood that literacy skills do not remain static. US research on Summer Literacy Decline suggests a cumulative deficit impact over periods of limited exposure to reading, that influences spelling (Cooper, Nye, Charlton, Lindsay, & Greathouse, 1996), vocabulary (Lawrence, Hinga, Mahoney, & Vandell, 2015) and reading comprehension (Guryan, 2015). Students need to read in order to maintain their literacy skills.

The third ranked variable, enjoyment of the early literacy experience, may be obscured in cases of excessive focus on reading purely for skill acquisition, without attention to the related affective experience. Reading skill acquisition and related testing emphasis may entirely eclipse enjoyment when building reading skills both at home and at school, with recent research finding that by upper primary, children may view reading as something primarily done for the purposes of assessment and learning, rather than something that can confer enjoyment (Merga, 2016). Ensuring both teachers and parents include fostering enjoyment in their consideration of promotion of reading skill acquisition and subsequent development is essential, Baker (2003) proposes a partnership approach to optimise engagement, suggesting that ‘teachers should provide advice to parents to help them assist their children at home, and parents should provide advice to teachers to help them motivate children at school’ (p. 92).

Finally, it can be contended that use of ANN as an adjunct approach in analysis of variables influencing children’s attitudes towards, and frequency of engagement in reading can enhance the quality of analysis. As these attitudes and practices influence reading skill, robust analysis is particularly important in this area, as improving children’s literacy attainment is linked to individual opportunity and societal benefit. Such findings can also be used to direct resource and time allocation when seeking to enhance reading engagement in school and home contexts. ANN could also be used to ensure high quality findings across a broader range of areas of inquiry within the reading research space, together with the conventional statistical approaches such as LR and BLR.

Limitations

Data are limited by the constraint of students’ self-report. Analysis was applied to an extant data set, thus survey items could not be shaped (in retrospect) to precisely align with the analytical goals of this study, though the extant data set still met the requirements of the analysis. In terms of the analytic method used, multi-item questions are preferred to increase the accuracy of the measurement. Not all of the included variables met this criterion – attitude towards reading and family model, for example, are single-item variables. However, given the age of the respondents, even if the data were instead run on data from a purpose-designed survey, the requirement for the young respondents to accurately differentiate different facets of the same latent construct with multi-item questions could potentially be too great, leading to unreliable data. In future studies with an older target group, multi-item questions may enhance the quality of the findings. Additionally, the ANN outputs are based on a single architecture. As such, it is suggested that future studies compare ANN results of different network architectures, e.g. using different activation functions for hidden and output layers.

Conclusion

Researchers, educators, parents and other educational stakeholders may seek to enhance students’ reading frequency and their attitudes towards reading, as reading engagement is associated with reading skill development (Merga, 2019a). The findings presented in this paper indicate that increasing students’ opportunities to access a library can have a strong positive influence on their reading engagement. While previous research has supported a positive relationship between school library visitation and children’s literacy performance (Francis et al., 2010; Hughes et al., 2014; Lonsdale, 2003), this is the first study that we are aware of that shows how potentially powerful library access can be in relation to other, perhaps more closely considered variables, such as gender. Interventions that enhance student access to their school libraries could be the most efficacious for enhancing student reading engagement. However, while many Australian schools have libraries on site, even where this resource is freely available, libraries may be underutilised, with most students visiting a school or community library to select books for recreational reading less than once a month (Merga, 2015b, 2019c). In addition, the school library may be precariously positioned in Australian schools, subject to funding cuts, and staffing may also be particularly vulnerable. A 2011 Australian government-initiated inquiry into school libraries found that ‘it is indisputable that the value of teacher librarians’ work has been eroded over the years and undervalued by many in the community, be it by colleagues, principals, parents or those in the wider school community’ (House of Representatives [HOR], 2011, p. 117). Before Australian libraries face further cuts to resourcing and staffing, the role of libraries in supporting students’ reading engagement and the close relationship between reading engagement and literacy skills need to be better understood.

Footnotes

Declaration of Conflicting Interests

The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding

The author(s) disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: The authors are grateful for financial support from the Ian Potter Foundation (grant 20160013), which funded this research.

Note

Appendix A

References

Afflerbach

Harrison

(2017). What is engagement, how is it different from motivation, and how can I promote it? Journal of Adolescent & Adult Literacy, 61, 217–220.

Ahmadpour Kasgari

Divsalar

Javid

M. R.

Ebrahimian

S. J.

(2013). Prediction of bankruptcy Iranian corporations through artificial neural network and probit-based analyses. Neural Computing and Applications, 23, 927–936.

Alaka

H. A.

Oyedele

L. O.

Owolabi

H. A.

Kumar

Ajayi

S. O.

Akinade

O. O.

Bilal

(2018). Systematic review of bankruptcy prediction models: Towards a framework for tool selection. Expert Systems with Applications, 94, 164–184.

Ali

Alharbi

O. M.

Alothman

Z. A.

Badjah

A. Y.

Alwarthan

(2018). Artificial neural network modelling of amido black dye sorption on iron composite nano material: Kinetics and thermodynamics studies. Journal of Molecular Liquids, 250, 1–8.

Australian Curriculum, Assessment and Reporting Authority (ACARA) (2015). What does the ICSEA value mean? Retrieved from http://docs.acara.edu.au/resources/About_icsea_2014.pdf

Australian Curriculum, Assessment and Reporting Authority (ACARA). (2016). My school. Retrieved from https://www.myschool.edu.au/

Baker

(2003). The role of parents in motivating struggling readers. Reading & Writing Quarterly, 19, 87–106.

Baker

Scher

Mackler

(1997). Home and family influences on motivations for reading. Educational Psychologist, 32, 69–82.

Burns

(7 April 2016). Too many school libraries ‘face cuts or closure’. BBC News. Retrieved from https://www.bbc.co.uk/news/education-35940002

10.

Chang

L. Y.

Plaut

D. C.

Perfetti

C. A.

(2016). Visual complexity in orthographic learning: Modeling learning across writing system variations. Scientific Studies of Reading, 20, 64–85.

11.

Cohen

(1988). Statistical power analysis for the behavioural sciences. (2nd ed.). Hillsdale, NJ: Lawrence Erlbaum Associate.

12.

Cooper

Nye

Charlton

Lindsay

Greathouse

(1996). The effects of summer vacation on achievement test scores: A narrative and meta-analytic review. Review of Educational Research, 66, 227–268.

13.

Daggett

Hasselbring

(2007). What we know about adolescent reading. Rexford, NY: International Centre for Leadership in Education.

14.

Djajadikerta

H. G.

Mat Roni

Trireksani

(2015). Dysfunctional information system behaviors are not all created the same: Challenges to the generalizability of security-based research. Information & Management, 52, 1012–1024.

15.

Esfe

M. H.

Saedodin

Sina

Afrand

Rostami

(2015). Designing an artificial neural network to predict thermal conductivity and dynamic viscosity of ferromagnetic nanofluid. International Communications in Heat and Mass Transfer, 68, 50–57.

16.

Francis

B. H.

Lance

K. C.

Lietzau

(2010). School librarians continue to help students achieve standards: The third Colorado study (2010). Closer Look Report. Denver, CO: Colorado State Library, Library Research Service.

17.

Guthrie

J. T.

Klauda

S. L.

A. N.

(2013). Modeling the relationships among reading instruction, motivation, engagement and achievement for adolescents. Reading Research Quarterly, 48, 9–26.

18.

Guthrie

J. T.

Wigfield

You

(2012). Instructional contexts for engagement and achievement in reading. In S. L. Christenson, A. L. Reschley, & C. Wylie (Eds), Handbook of research on student engagement (pp. 601–634). New York, NY: Springer.

19.

Guryan

(2015). Can a scaffolded summer reading intervention reduce socioeconomic gaps in children’s reading comprehension ability and home book access? Results from a randomized experiment. Doctoral dissertation, University of Texas, Austin.

20.

Hartzell

(2002). The principal’s perceptions of school libraries and teacher-librarians. School Libraries Worldwide, 8, 92–110.

21.

E. S. C.

Lau

K. L.

(2018). Reading engagement and reading literacy performance: Effective policy and practices at home and in school. Journal of Research in Reading, 41, 657–679.

22.

Hodo

Bellekens

Hamilton

Dubouilh

P. L.

Iorkyase

Tachtatzis

Atkinson

(2016 May). Threat analysis of IoT networks using artificial neural network intrusion detection system. In 2016 International Symposium on Networks, Computers and Communications (ISNCC) (pp. 1–6). Piscataway, NJ: IEEE.

23.

House of Representatives. (2011). School libraries and teacher librarians in 21st century Australia. Canberra: Commonwealth of Australia.

24.

Hughes

Bozorgian

Allan

(2014). School libraries, teacher-librarians and student outcomes: Presenting and using the evidence. School Libraries Worldwide, 20, 29–50.

25.

Ivey

Broaddus

(2001). Just plain reading: A survey of what makes students want to read in middle school classrooms. Reading Research Quarterly, 36, 350–377.

26.

Jerrim

Moss

(2019). The link between fiction and teenagers’ reading skills: International evidence from the OECD PISA study. British Educational Research Journal, 45, 181–200.

27.

Kachel

(2015). The calamity of the disappearing school libraries. The Conversation. Retrieved from https://theconversation.com/the-calamity-of-the-disappearing-school-libraries-44498

28.

Kennedy

(2018). Engaging children as readers and writers in high‐poverty contexts. Journal of Research in Reading, 41, 716–731.

29.

Keslair

(2017). How much will the literacy level of the working-age population change from now to 2022? Paris, France: Organisation for Economic Co-operation and Development (OECD).

30.

Kirsch

de Jong

Lafontaine

McQueen

Mendelovits

Monsuer

(2002). Reading for change: Performance and engagement across countries. Washington, DC: Organisation for Economic Co-operation and Development (OECD).

31.

Krogh

(2008). What are artificial neural networks? Nature Biotechnology, 26, 195–197.

32.

Lawrence

J. F.

Hinga

B. M.

Mahoney

J. L.

Vandell

D. L.

(2015). Summer activities and vocabulary development: Relationships across middle childhood and adolescence. IJREE – International Journal for Research on Extended Education, 3, 71–93.

33.

Lonsdale

(2003). Impact of school libraries on student achievement: A review of the research. Camberwell, VIC: Australian Council of Educational Research.

34.

Lykourentzou

Giannoukos

Mpardis

Nikolopoulos

Loumos

(2009). Early and dynamic student achievement prediction in e‐learning courses using neural networks. Journal of the American Society for Information Science and Technology, 60, 372–380.

35.

Mancini

A. L.

Pasqua

(2011). On the intergenerational transmission of time use patterns: Is a good example the best sermon? ZA Discussion Paper No. 6038, October 2011.

36.

Mar

R. A.

Oatley

Peterson

J. B.

(2009). Exploring the link between reading fiction and empathy: Ruling out individual differences and examining outcomes. Communications, 34, 407–428.

37.

McIntosh

Vignoles

(2001). Measuring and assessing the impact of basic skills on labour market outcomes. Oxford Economic Papers, 53, 453–481.

38.

McKenna

M. C.

Kear

D. J.

Ellsworth

R. A.

(1995). Children’s attitudes toward reading: A national survey. Reading Research Quarterly, 30, 934–956.

39.

Merga

M. K.

(2014b). Are Western Australian adolescents keen book readers? Australian Journal of Language and Literacy, 37, 161–170.

40.

Merga

M. K.

(2015a). ‘ She knows what I like’: Student generated best-practice statements for encouraging recreational reading in adolescents. Australian Journal of Education, 59, 35–50.

41.

Merga

M. K.

(2015b). Access to books in the home and adolescent engagement in recreational book reading: Considerations for secondary school educators. English in Education, 49, 197–214.

42.

Merga

M. K.

(2016). “ I don’t know if she likes reading”: Are teachers perceived to be keen readers, and how is this determined? English in Education, 50, 255–269.

43.

Merga

M. K.

(2017). Interactive reading opportunities beyond the early years: What educators need to consider. Australian Journal of Education, 61, 328–343.

44.

Merga

M. K.

(2019a). Reading engagement for tweens and teens: What would make them read more? Santa Barbara, CA: ABC-CLIO/Libraries Unlimited.

45.

Merga

M. K.

(2019b). School libraries fostering children’s literacy and literature learning: mitigating the barriers. Literacy. Retrieved from https://onlinelibrary.wiley.com/doi/pdf/10.1111/lit.12189

46.

Merga

M. K.

(2019c). Librarians as literacy educators in schools. Basingstoke, UK: Palgrave Macmillan.

47.

Merga

Gardiner

(2019). The role of whole-school literacy policies supporting reading engagement in Australian schools. English in Australia, 53, 37–50.

48.

Merga

M. K.

Mat Roni

(2017). The influence of access to eReaders, computers and mobile phones on children’s book reading frequency. Computers and Education, 109, 187–196.

49.

Merga

M. K.

Mat Roni

(2018a). Characteristics, preferences and motivation of avid non-fiction readers. Collection and Curation, 37, 50–59.

50.

Merga

M. K.

Mat Roni

(2018b). Children’s perceptions of the importance and value of reading. Australian Journal of Education, 62, 135–153.

51.

Mol

S. E.

Bus

A. G.

(2011). To read or not to read: A meta-analysis of print exposure from infancy to early adulthood. Psychological Bulletin, 137, 267–296.

52.

Mullan

(2010). Families that read: A time-diary analysis of young people’s and parents’ reading. Journal of Research in Reading, 33, 414–430.

53.

Mullis

I. V. S.

Martin

M. O.

Kennedy

A. M.

Trong

Sainsbury

(2009). PIRLS 2011 assessment framework. Chestnut Hill, PA: TIMSS & PIRLS International Study Center, Boston College.

54.

Omar

Johari

Z. A.

Smith

(2017). Predicting fraudulent financial reporting using artificial neural network. Journal of Financial Crime, 24, 362–387.

55.

Organisation for Economic Co-operation and Development (OECD). (2011a). Do students today read for pleasure? PISA in focus 8. Paris, France: OECD Publishing.

56.

Organization for Economic Co-operation and Development (OECD). (2011b). Education at a glance. Washington, DC: OECD Publishing.

57.

Organisation for Economic Co-operation and Development (OECD). (2015). Trends shaping education 2015 spotlight 7. Washington, DC: OECD.

58.

Park

D. C.

El-Sharkawi

M. A.

Marks

R. J.

Atlas

L. E.

Damborg

M. J.

(1991). Electric load forecasting using an artificial neural network. IEEE Transactions on Power Systems, 6, 442–449.

59.

Parsons

A. W.

Parsons

S. A.

Malloy

J. A.

Marinak

B. A.

Reutzel

D. R.

Applegate

M. D.

, … Gambrell

L. B.

(2018). Upper elementary students’ motivation to read fiction and nonfiction. The Elementary School Journal, 118, 505–523.

60.

Pfost

DöRfler

Artelt

(2013). Students’ extracurricular reading behavior and the development of vocabulary and reading comprehension. Learning and Individual Differences, 26, 89–102.

61.

Rice

M. E.

Harris

G. T.

(2005). Comparing effect sizes in follow-up studies: ROC area, Cohen’s d, and r. Law and Human Behavior, 29, 615–620.

62.

Sabourin

J. L.

Shores

L. R.

Mott

B. W.

Lester

J. C.

(2013). Understanding and predicting student self-regulated learning strategies in game-based learning environments. International Journal of Artificial Intelligence in Education, 23, 94–114.

63.

Scholastic. (2015). Kids and family reading report, 5th edition. Retrieved from http://www.scholastic.com/readingreport

64.

Scholastic. (2016a). Kids and family reading report Australia. Retrieved from http://www.scholastic.com.au/readingreport

65.

Scholastic. (2016b). Kids and family reading report United Kingdom. Retrieved from http://www.scholastic.co.uk/readingreport

66.

Segal

Martin-Chang

(2018). The apple doesn’t fall from the tree: Parents’ reading-related knowledge and children’s reading outcomes. Reading and Writing, 31, 1231–1247.

67.

Spear-Swerling

Brucker

P. O.

Alfano

M. P.

(2010). Relationships between sixth-graders’ reading comprehension and two different measures of print exposure. Reading and Writing, 23, 73–96.

68.

Stefl-Mabry

Radlick

Armbruster

Keller

(2016). Breaking down Information silos: Sharing decades of school library research with educational researchers. Paper presented at the 2016 Annual Meeting “Public Scholarship to Educate Diverse Democracies” April 8–12, Washington, DC.

69.

Tamir

D. I.

Bricker

A. B.

Dodell-Feder

Mitchell

J. P.

(2015). Reading fiction and reading minds: The role of simulation in the default network. Social Cognitive and Affective Neuroscience, 11, 215–224.

70.

Tse

S. K.

Xiao

X. Y.

H. W.

Lam

J. W. I.

Hui

S. Y.

H. W.

(2016). Do reading practices make a difference? Evidence from PIRLS data for Hong Kong and Taiwan primary school Grade 4 students. Compare: A Journal of Comparative and International Education, 46, 369–393.

71.

Twenge

J. M.

Martin

G. N.

Spitzberg

B. H.

(2018). Trends in US Adolescents’ media use, 1976–2016: The rise of digital media, the decline of TV, and the (near) demise of print. Psychology of Popular Media Culture. http://doi.org/10.1037/ppm0000203.

72.

Wasserstein

R. L.

Schirm

A. L.

Lazar

N. A.

(2019). Moving to a world beyond “p < 0.05”. The American Statistician, 73, 1–19.

73.

Westbrook

Sutherland

Oakhill

Sullivan

(2018). “Just reading”: The impact of a faster pace of reading narratives on the comprehension of poorer adolescent readers in English classrooms. Literacy. Retrieved from https:// onlinelibrary.wiley.com/doi/full/10.1111/lit.12141

74.

Wigfield

Eccles

J. S.

(2000). Expectancy–value theory of achievement motivation. Contemporary Educational Psychology, 25, 68–81.

75.

Wollscheid

(2013). Parents’ cultural resources, gender and young people’s reading habits—Findings from a secondary analysis with time-survey data in two-parent families. International Journal about Parents in Education, 7, 69–83.

76.

Wrigley

(2017). Synthetic phonics and the phonics check: The hidden politics of early literacy. In M. M. Clark (Ed.), Reading the evidence: Synthetic phonics and literacy learning (pp. 98–114). Birmingham, UK: Glendale Education.

77.

Zebroff

Kaufman

(2016). Texting, reading, and other daily habits associated with adolescents’ literacy levels. Education and Information Technologies, 22, 2197–2216.