The Finnish Line: Students’ ICT Use, Reading Attitudes and Reading Performance in Three Nordic Countries

The Finnish Line: Students’ ICT Use, Reading Attitudes and Reading

Performance in Three Nordic Countries

Using PISA 2018 Data with

Structural Equation Modelling Approach

Mudar Saied Alhusien

M.Sc. in Assessment, Measurement and Evaluation 120 Credits

CEMO, University of Oslo

November 2021

The Finnish Line: Students’ ICT Use, Reading Attitudes and Reading

Performance in Three Nordic Countries

Using PISA 2018 Data with

Structural Equation Modelling Approach

Mudar Saied Alhusien

CEMO, University of Oslo

November 2021

We've got mountains of content Some better, some worse

If none of its of interest to you, you'd be the first…

It was always the plan to put the world in your hand”

Bo Burnham 

Welcome to the Internet 

The expanding access and use of digital devices among young students led to ongoing debates about the advantages and disadvantages of such use. Educational

organizations in particular were concerned about the impact of ICT use on students’

academic performance. This study sheds light on this complex relationship by investigating the impact of ICT use on a) students reading literacy achievement and b) students; attitudes towards reading in Denmark, Finland, and Sweden. Employing Structural Equation Modeling on data from PISA 2018 cycle, this study found that a) the relationship between ICT use and reading performance follow non-linear relationship; b) attitudes towards reading can partially mediate this relationship and c) the three countries showed different results with Finland being distinct from the other two in implementing what seems to be the optimal ICT integration in education system.

Key words: ICT use, reading literacy, reading attitudes, PISA, structural equation modeling, mediation.

Acknowledgments

I am profoundly grateful to my supervisor Prof. Ronny Scherer for providing generous support in every stage of this research. His excellent judgment and faithful guidance had made this paper possible. I also would like to extend my sincere thanks to my co-supervisor Prof. Ove E. Hatlevik who kindly assisted me through this work.

Indeed, it is hard to find words to thank you both.

I am also thankful to the teaching and administrative staff of CEMO that went above and beyond to facilitate this master program through challenging times of the Covid-19 pandemic. I extend my special heartfelt gratitude to Tony Tan whose motivating words and brilliant educational skills inspired me to pursue this degree, 严师出高徒.

I owe a debt of gratitude to Norway for being a safe haven for me and my family, and its education system that made the dream of pursuing my studies come true. This work is dedicated to my parents to whom I am eternally indebted for all the prayers and blessings, my wife and our lovely children without their support and sacrifice this paper would have been impossible.

Introduction

The world is connected like never before. Globalization and digitalization have brought great opportunities for individuals and communities, at the same time this momentum has made life more elusive. Students in the past can find information in books and encyclopedias that are always reliable, today the internet is unlimited source for information, however, their reliability must always be questioned (OECD, 2021). As the digital era continues to expand, the boundaries between information, and media has become increasingly blurred (Buckingham, 2006).

In this digital age, students need not only traditional teaching but also appropriate tools and competencies to navigate and investigate (OECD, 2021). In 2018, data revealed by PISA found that the average 15 year-old increased by 66%, spending time equivalent to an adult workweek across OECD countries (OECD, 2021). Such increase in access and use of ICT has contributed to an ongoing worldwide societal debate about integrating of ICT in educational systems. The growing importance of students’ ICT skills has exceeded the mere inclusion of these skills as a separate independent subject. ICT competencies are now included across schools’ curricula so that other subjects require ICT tasks (European Commission et al., 2013). Accordingly, ICT skills and competencies are imperative for students to succeed in this digital age (Lorenceau et al., 2019).

Digital Literacy

Although the definition of “digital literacy” is debated among researchers, it has become the recognized term to describe individual use of technological platforms for information (Buckingham, 2006). By using the word “literacy” the definition of digital literacy expands past function (Buckingham, 2006). In fact, in the 21^st century literacy is all about constructing and validating information (OECD, 2021). Lennon et al., (2003)

introduced more detailed definition as “The interest, attitude and ability of individuals to appropriately use digital technology and communication tools to access, manage, integrate and evaluate information, construct new knowledge and communicate with others in order to participate effectively in society”. Many researchers and organizations specified five main competencies that form the digital literacy construct. First competency is students’ ability to access and evaluate information; second is their ability to communicate information; third is problem solving in digital context as well as computational thinking; last competency is students’ awareness about digital threats (Lorenceau et al., 2019).

To better understand the depth and complexity of digital literacy, Buckingham (2006) outlines a basic conceptual framework as a useful way to map the avenues. The framework includes; representation, language, production, and audience. Representation dives into the notion that digital media doesn’t reflect the world but represents it. Language is much broader than the use of words, but how particular forms reflect alternating

messages. The meaning of a sentence may vary based on genre and culture. This means an individual needs analytical skill to be able to decipher the language of text more in depth.

Thus, digital literacy needs an individual to be aware of how digital media is being constructed and the unique rhetoric. Production involves understanding who is

communicating to whom and why. Lastly, Audience entails one’s own understanding of how media is targeting audience and how different audience relate and use the information and media.

To emphasize the digital literacy role, the European Commission launched 470 digital literacy initiatives in Europe in its i2010 strategy. Even further, it labeled digital literacy as an essential competence for life (Bennett, 2008). Later in 2011, the European Commission’s Digital Agenda for Europe called for prioritizing digital literacy to act for (European

Commission et al., 2013). In the Nordic region, the focus region for this study, digital literacy found to be a vital cultural factor that can largely contribute to the quality of education, according to the E-learning Nordic report (Fraillon et al., 2014a). Therefore, technology goes beyond computer games, and mobile phones, but provides a new way of representing and mediating in the world. In the new generation, technology also influences culture (Buckingham, 2006).

Digital Reading

Digital literacy is not restricted to the mechanical use of technology, but also the ability of one to navigate and understand digital text (Buckingham, 2006). Texts are the main source of information that students can access through digital devices (Gubbels et al., 2020). The increasing availability of digital texts begs the question of is it better than the paper reading, especially for young students. Therefore, digital reading as a distinct type from the traditional print reading has been included in PISA cycles since 2009 (Fraillon et al., 2014a) and the focus of many studies as well. While some studies concluded that digital reading has advantages over print reading other studies found contradictory results. For example, Bando et al., (2016) praised the cost-effectiveness of digital books which contribute to reducing knowledge gap. Additionally, digital reading allows for interactive environment where readers can communicate and interact with others (National Academies of Sciences &

Medicine, 2018). Furthermore, it is more accessible to read texts from various digital devices specially that these devices can store large number of reading materials. Despite these advantages, digital reading seems not as pleasant and engaging as paper reading (Mangen &

Kuiken, 2014). Even more, the emotional experience of reading from paper seems to be superior to that from digital devices (Kaakinen et al., 2018). Even though digital screens have advanced to cause less eyestrain, it is still a concern that is not raised in paper reading

(Rosenfield et al., 2015). Lastly, according to Daniel & Woody (2013) readers take more time to read from digital devices than paper without any extra benefit. In fact, some researchers heavily advocated paper over digital assessments, stating that weaker performance is correlated with digital assessments (Ackerman & Lauterman, 2012; Clinton, 2019). The aforementioned contradictory results suggests that digital devices come with opportunities and challenges for readers. Indeed, digital reading can be both engaging and distracting, thus readers are demanded to adapt metacognitive skills to navigate digital texts (Wylie et al., 2018)

Theoretical Framework ICT Users Profiles

Students’ increasing use of ICT has led to extensive research aiming at analyzing students’ patterns of ICT use and categorizing these patterns into profiles. This

differentiated view on students’ use of ICT will help teachers as well as parents identifying the potential benefits and harms of ICT use based on the different students’ profiles (Scherer et al., 2017). For example Scherer et al., (2017) indicated the existence of two profiles after conducting latent profile analysis with a sample from Norwegian students. The first profile described students who use ICT frequently for different purposes inside and outside school, while the second profile identified students who use ICT more frequently for study purposes.

In line with Scherer findings, Xiao & Sun (2021) also districted two user profiles from a sample of American students, the first subgroup, 20.50%, were students who excessively and actively used ICT for leisure but not sufficiently for education purposes; the other subgroup, 79.40%, are students who moderately used ICT for both purposes, in other words not excessive nor insufficient.

OECD further proposed a 9-users-profiles based on two purposes, leisure, and education; as well as three frequencies (rare, monthly, and frequent) this design results in 9 categories Out of the 9 profiles, however, OECD highlighted six distinct profiles based on the frequency of ICT use for leisure and educational purposes. The Digi-wired group are students who use ICT frequently for leisure; the Digi-educational students who low ICT use for leisure and frequent for education; the Digi-zappers students who frequently use ICT for both purposes; the Analogous group who, contrary to the former group, who do not frequently use ICT for neither purposes; the Digi-sporadic students who reports a monthly use for both leisure and education; the last group is the Digi-casuals students who reported a once a

month use for leisure and more frequent use for education up to once a day (OECD, 2010).

The above-mentioned profiles are largely associated with background variables such as gender (OECD, 2010), self-efficacy in ICT (Tømte & Hatlevik, 2011), Immigration status and motivation towards ICT (Scherer et al., 2017).

Students’ Use of ICT

The OECD (2005) has defined students' use of information and communication technology (ICT) as follows:

“ICT is the use of any equipment or software for processing or transmitting digital information that performs diverse general functions, whose options can be specified or programmed by its users”

This definition entails that students may use ICT via various devices and systems, at various places, and for various purposes. For all that, the use of ICT involves one core component, that is, communication. Hence, students are further expected –when using ICT— to socialize, seek and share information, and informally learn (European Commission et al., 2014).

Despite the fact that students are increasingly accessing and using ICT devices, there is growing gaps in the modalities of this use among students (Lorenceau et al., 2019). Prior research has indeed highlighted that the sheer availability and access to ICT devices do not guarantee better achieving students but ensuring effective use of ICT do (Bulman & Fairlie, 2016). Therefore, it is essential to inspect how each different types of ICT use may predict academic achievement. In its ICT questionnaire, PISA has decomposed students ICT use into the three following variables. 

ICT Use Outside of School for Leisure

The vast majority of students appear to use ICT outside school more frequently than at scho9ol (OECD, 2010). With increasing access to digital devices, smartphones in

particular, students in general spend more time, an average of 40 minutes, on using ICT outside school (OECD, 2017). Similarly, the IEA international Computer and Information Literacy Study revealed in its international 2013 report that, on average, 75% of students used ICT for social communication, 82% listened to music, 68% watched recreational videos at least once a week (Fraillon et al., 2014b). Three years later ICIL has reported further increase in the same use that among all the types of ICT, students used ICT outside school for leisure purposes the most (IEA, 2018). According to the same report, on average 70% of students reported this type of use on a daily basis. Such findings are further supported by the European Commission (2013) report which concluded that most of this time spent on ICT devices outside school is dedicated to leisure purposes. This dramatic increase of students’ use of ICT for leisure has put higher demand on parental guidance, since their awareness and attitudes towards ICT can impact their children use of ICT at home (Adrien, 2021). For example parents can encourage their children to play educational games, or to regulate their use. Based on the mentioned above, ICT use outside school for leisure has its own opportunities and risks with cognitive development and cyber bullying being examples respectively. Van Deursen & Helsper, (2015) have pointed that these opportunities and risks are associated with the frequency and the nature of leisure activity students spend time on.

ICT Use Outside of School for Schoolwork

If not for leisure, students are likely to use ICT devices outside school for schoolwork, in this context, schoolwork doesn’t include attending school classes online. It is more and more common, especially during covid-19 pandemic, that students can learn and do schoolwork regardless of being at or outside school. As a matter of fact, earlier OECD research has revealed that the relationship between students’ overall academic performance and their ICT use outside school is actually stronger than their use in school (OECD, 2010).

In 2013 ICIL (Fraillon et al., 2014b)results showed that only about 50% of the students used ICT at least once a week to search for information related to schoolwork when they are outside school. The frequency of this ICT use continued to be less common than using ICT for leisure in the following years. According to the 2018 ICIL report(Fraillon et al., 2020) only 20% of the students reported using ICT devices outside school for schoolwork on a daily basis.

Althoughexacerbated by the pandemic, there has been already strong initiatives to provide schools with ICT infrastructure for education, this has expanded to use of ICT for homework as well so that teachers can assign homework to students through ICT and monitor their progress remotely (Magalhães et al., 2020). Accordingly, students have unprecedented opportunity to learn after-school. OECD has recognized the leading role of ICT in seizing this opportunity, through its impact on students’ learning activities and providing them with a variety of options e.g. educational games, tutorials, podcasts…etc.

(Lorenceau et al., 2019).

ICT Use at School

Being a matter of concern, decision makers have been interested in how to efficiently integrate ICT in schools and school systems. Most education systems have implemented polices and plans to successfully integrate ICT while avoiding its undesirable side-effects (European Commission, 2017). The process of introducing ICT use to schools, however, has been slow, as it is largely dependent on teachers and school budget (Leino, 2014).Given the high costs and scarce evidence of effectiveness, there are still many schools and teachers that hesitate to include and implement educational technology. This has made the topic of ICT use in schools a critical debate topic (Petko et al., 2017).

In spite of the above, students access and use of ICT devices in schools have sharply increased in the last decade (Lorenceau et al., 2019). In its 2013 international survey ICIL reported that at, on average, least 39% of the participating students used ICT inside schools for schoolwork at least once a week. ICIL has realized that students might use ICT at schools for non-school related purposes as well. Therefore in its 2018 survey, ICIL has

divided students’ use of ICT inside schools into two types: for schoolwork and for non-school work (Fraillon et al., 2020). According to the same survey, student’s ICT use inside school for leisure was the second most frequent type of use among students. On average, about 30%

of students reported using ICT at school for leisure at least once a day, compared to only 18% who reported using ICT inside school for schoolwork purposes.

Finally, it is noteworthy to recognize that there are factors that largely influence students’ use of ICT, some are internal such as gender, immigration status (Luu & Freeman, 2011), other are external such as country and school-level ICT policies. These factors do not share the same impact on all types of ICT use, for instance the school policies largely impact students’ ICT use inside the school (European Commission, 2019), while outside school. The same cannot be said for outside schools, in that students chiefly self-select and decide

patterns of using ICT devices (Fariña et al., 2015).

Reading Literacy

Reading literacy has been defined in many ways; the OECD (2019a) has adopted the following definition:

“Reading literacy is an individual’s capacity to understand, use, evaluate, reflect on, and engage with texts in order to achieve one’s goals, develop one’s knowledge and potential, and participate in society.” (OECD, 2019a, p. 14)

The OECD’s assessment approach focuses not only on students’ ability to retain and reproduce knowledge but also on their ability to extrapolate and apply such knowledge in novel situations (Lorenceau et al., 2019), that’s why PISA emphasize the term literacy. This definition implies that reading literacy is a multifaceted construct that greatly influence individuals as well as communities as a whole. What makes reading literacy exceptional is that its role in other cognitive domains and it is a prerequisite for any participation as an adult (Cunningham & Stanovich, 1997).

In a rapidly changing world, the concept of reading literacy has been and will be evolving. Reading literacy assessments used to be a tool to merely evaluate student’s ability to understand and reflect on single texts, however, upon the arrival of ICT, the concept of reading literacy has expanded to include newer skills such as searching and accessing text via search engines and digital navigation (OECD, 2019a). The fact that more people can access texts via digital devices nowadays, these newer skills are strongly related to digital reading.

Therefore, reading literacy is beyond the skills of the sheer comprehension of texts, further, it incorporates processing, analyzing, synthesizing, interpreting, organizing, and integrating multiple texts from multiple sources in multiple contexts (Leu et al., 2015; OECD, 2019a).

Accordingly, PISA has – when assessing reading literacy— accounted for students’ ability to find, select, interpret, and evaluate information from multiple sources (OECD, 2019a).

ICT Use and Reading Literacy

The latest expansion of ICT has certainly impacted how students read and

communicate information whether in or outside school. This led to a great deal of studies to investigate the association between ICT use and reading performance. While consensus about the existence of the relationship between ICT use and reading performance seems high

(Punie et al., 2006) (Cox et al., 2003), the same cannot be said about the nature of such

relationship. On the one hand, researchers like (Kirkpatrick & Cuban, 1998; Oppenheimer, 1997; Wainer et al., 2008) advocate a negative or a null relationship between the two variables; on the other hand researchers like (Kubiatko & Vlckova, 2010; Luu & Freeman, 2011) advocate the opposite, i.e. the existence of positive relationship.

In particular, many researchers have advocated the positivity of this relationship, and results from studies by –among others— Hu et al., (2018); Skryabin et al., (2015) and Biagi

& Loi, (2013) confirmed there is a positive association between ICT use outside school for leisure and reading performance, on the other side several studies showed opposite results that justifies a negative relationship between ICT use for leisure and reading (Luu &

Freeman, 2011; Petko et al., 2017). Similarly, when exploring the relationship between ICT use outside school for schoolwork and reading performance, many studies found that this relationship is positive (Lee & Wu, 2012; Woessmann & Fuchs, 2004); in contrast, other scholars found the same relationship to be negative such as (Agasisti et al., 2020; Gumus &

Atalmis, 2011). Then as well, studies on the relationship between ICT use at school and reading performance also revealed contradictory results, where studies like (Skryabin et al., 2015a) presented positive relationship between the two variables, other studies like (X. Hu et al., 2018; Leino, 2014; Petko et al., 2017) then again found a negative relationship. Apart from all the above, other studies like (Wittwer & Senkbeil, 2008) found a non-significant relationship between ICT use and reading performance.

These mixed results provoked many scholars to further examine these relationships, and to introduce different rationales. For example Song & Kang, (2012) suggested that these relationships can be mediated. Also Falck et al. (2018) claimed that the combination of positive and negative effects might result in a null relationship. Others like Fariña et al.

(2015) and Ponzo (2011) raised the issue of endogeneity as a reason of bias since students

decide their ICT use at home. Furthermore, other researchers pointed to the non-linearity issue as an underlying factor why these relationships are complicated. Papanastasiou et al., (2005) found out that students who use ICT devices for programming tend to have higher scores in science as long as this use is not excessive. Additionally, Gubbels et al., (2020) studied the relationship between ICT-related variables and reading achievement. One major conclusion of the study is that there is an optimal frequency of ICT use that covary with the highest reading achievement. Excessive use beyond this optimal frequency covary with lower reading performance.

Attitudes toward Reading

Reading literacy is associated with student background variables such as gender and parents’ education and students’ attitudes towards reading (OECD, 2019b). Attitudes towards reading were defined by several scholars. One of the first definitions appeared by Alexander and Filler (1976) as a “system of feelings” towards reading that impact ones inclination to seek or avoid reading. More recent definition came from Mathewson (1994) who identified such attitudes as a multidimensional construct referring to dominant feelings, readiness, and beliefs towards reading. PISA’s previous cycles have measured these attitudes through reading engagement which is a construct that comprise interest in and enjoyment of reading, in 2018 PISA has added two constructs, self-efficacy which is the student’s

“perceived capacity of performing specific tasks”, and self-concept, student’s “own perceived abilities in a more general domain.” (OECD, 2021)

Data from earlier PISA cycles, showed that reading attitudes can mediate the relationship between students’ background variables for instance ESCS and their reading performance (OECD, 2021). At this point of this study, it is vital to investigate whether these reading attitudes can also mediate the relationship between ICT use and reading

performance, by exploring the relationship between ICT and reading attitudes from one side;

and the relationship between reading attitudes and reading performance from the other.

ICT Use and Attitudes toward Reading

Several research findings suggested a relationship between ICT use and attitudes towards reading for instance, according to Leino (2014), the use of computers do increase students motivation to learn. Furthermore, Patricia Alexander and Emily Fox (2010) suggested that reading from alternative sources other than traditional books outside school can be a key factor to increase reading motivation. Other research concluded that a rich ICT environment can encourage students to engage in reading, as well as increase their

motivation and sense of competency (Coiro, 2007; Kamil et al., 2000). Moreover, the use of ICT found to boost self-esteem and attitudes towards learning (Solomon, 2002). Finally, (Kaakinen et al., 2018) discussed profoundly how the advancement of technology has impact readers’ immersion and accordingly their enjoyment of reading.

Reading Attitudes and Reading Performance

Among all the students’ background variables, students’ own motivation and

engagement towards reading seem to be one of the strongest predictors of reading proficiency (Becker et al., 2010; Guthrie et al., 1999; Klauda & Guthrie, 2015). In fact, in PISA 2000 motivational attributes towards reading were stronger predictor than ESCS (Kirsch et al., 2003). Not only that but Wigfield & Guthrie (2000) found out that reading performance can be explained by motivation more than any other variable except previous reading

achievement.

Indeed, two-thirds of students reported that they read for the sake of reading enjoyment on a daily basis and this type of reading is positive predictor for reading performance (OECD, 2011). Perception of competence in reading that is confidence is, as

well, correlated with reading performance. Simply put, students tend to engage in tasks they feel confident in and avoid those they do not (Pajares, 2002). This relationship is further supported by Leino (2014)who indicated that students’ confidence as a learner is a strong predictor of good reading performance.

The OECD employed self-concept scales to measure students’ perception of

competence and perception of difficulty in reading. The concept of self-concept is regarded as a multidimensional construct in itself by many researchers. Not only in academic domain, but self-concept is also applied to other domains like social and physical (Chapman &

Tunmer, 1995). An interesting and early definition of self-concept introduced by

Coopersmith, and Feldman as set of “beliefs, hypothesis, and assumptions that the individual has about himself. It is the person’s view of himself as conceived and organized from his inner vantage” (Pajares & Schunk, 2001). Combs (1962) yet summarized the self-concept into

“What an individual believes he is”. When associated with reading, self-concept include three components, namely, perception of competence; perception of difficulty and attitudes

towards reading (Marsh, 1986). Well-documented findings about the distinction between the two perceptions is highlighted in Nicholls & Miller (1984) where they concluded that when one perception is positive or negative, the other one does not necessarily concur. The relationship between self-concept in reading and reading achievement is documented in abundant studies (Petscher, 2010). For example, Cloer and Ross (1996) found that self- concept strongly predicts students’ reading performance. Moreover, Jensen et al., (2019) found that self-concept has a direct significant relationship with students’ reading

achievement, as well as significant mediating effect between teacher emotional support and reading achievement. In fact, Wolff & Heath (1993) concluded that when parents do not support their children self-concept, by lowering their expectations, the children actually see

themselves as not good learners. This strong relationship is further supported by well-

recognized behavioral theories such as the self-determination theory that supports this strong correlation between self-concept and reading achievement as J De Naeghel (2012) concluded.

The above findings invoke testing a mediation that reading attitudes can play between ICT use and reading achievement which will be further investigated in this study.

Research Questions and Hypotheses

Based on the theoretical perspectives and literature presented above, this study will implement a structural equation model to investigate the relationship between students’ ICT use and their reading performance as well as testing whether this relationship is mediated by attitudes toward reading. Since prior research concluded that the three groups of variables are correlated, I pose the following research questions

RQ1: To what extent the mode and the frequency of ICT use impacts reading performance?

RQ2: To what extent students’ attitudes toward reading mediate the relationship between ICT use and reading performance?

RQ3: Do the three countries differ in these relationships?

Given the above review of the evidence on the relations among the constructs relevant to address these RQs, I hypothesize:

H1: Reading attitudes mediate the relationship between ICT use and reading performance.

H2: There is a non-linear relationship between ICT use and reading performance.

H3: Finland shows significant difference than Denmark and Sweden in these relationships.

Methods Data and Sample

The international dataset used in this analysis is secondary data from the student level OECD’s PISA 2018 cycle provided for public at OECD official website. In addition to data from the main cognitive test, data from both the ICT Familiarity Questionnaire and the Students Background Questionnaire (OECD, 2018a, pt. ICT familiarity questionnaire) were also included in the analysis to obtain contextual knowledge about students. In its latest cycle, 79 countries participated in PISA assessment, as an optional questionnaire, only 52 countries have administered the ICT questionnaire to students (OECD, 2019c, Chapter 1)

Using the 4.0.3 version of R software (R Core Team, 2020) and intsvy package (Caro

& Biecek, 2017), the original data file from OECD was treated to produce a final dataset including the three Scandinavian countries with total sample size of N = 18810 observations

—by observations we refer to the 15-year-old students who participated in the assessment—

distributed as 7657 Denmark, 5649 Finland, and 5504 students from Sweden. Participants were chosen randomly within their own cluster that is school. With 348 participating schools from Denmark, 214 from Finland and 223 from Sweden, the total number of schools was 785.

It is worth to mention that the items used for the cognitive tests, including reading, are confidential except for small numbers that were released to demonstrate the scaling process (OECD, 2018c, sec. Reading items).

Country Differences Within the Three Nordic Countries

The Nordic countries, even though scoring above PISA 2018 average in the three domains, demonstrate a notable difference, Finland is doing better that the other Nordic sisters. Of course, PISA results demonstrated larger gaps among participating countries,

these gaps are associated with country-, school- and student- level factors. Some of these factors are attributed to cultural, social, and economical grounds. Of course drawing a comparison between high and low performing countries would widen the scope of the comparison, however such comparisons might instigate complex and invalid conclusions due to the sharp socioeconomic differences. Consequently this study will target three Nordic countries namely, Denmark, Finland, and Sweden for several considerations.

First, all of them are top performing countries in reading as per PISA 2018 results.

The Nordic countries are known to have the strongest welfare and stable economies with focus on education (Social Progress Index, 2020), equity is essential in education that

everyone has equal opportunity and access to education (Tømte & Hatlevik, 2011), as well as the highest rates of individuals using the internet(World Bank, 2020). On top on that, these countries are among those with the narrowest gender gaps (World Economic Forum, 2020) all the above factors refer to common grounds, yet these countries perform differently in international assessments which further motivates this study, specially that the variables in this study, namely, ICT, attitudes towards reading, gender, and social-economic status, ought to show no polar differences.

Measures

Table 1 exhibits an overview of the instruments (variables) used in this study; these items are further explained below.

Table 1

Summary of Instruments

Predictors (X)

ICT use Items Mediators (M) Attitudes

towards reading Items Control (C) Student background

Outcome (Y) Reading performance

Outside school for

leisure (ENTUSE) 12 Reading enjoyment

(READJOY) 5 Economic, social and cultural status (ESCS)

Reading test score (READ)

Outside school for schoolwork

(HOMESCH) 11 Perception of competence

(READCOMP) 3 Gender

(FEMALE)

Inside school in

general (USESCH) 10 Perception of ease

(READEASE) 3 Note. The names do not necessarily match PISA naming.

ICT Usage: The Predictors (X)

ICT use variables are drawn from the ICT Familiarity Questionnaire, for this study, three indices were used to capture the entire scope of students’ usage of ICT devices. These continuous indices are derived—by means of IRT scaling technique—from items that investigates the students’ place, purpose, and frequency of this usage, i.e. where, what for and how often students use the ICT. such items, rather activities, build the indexes in question, namely, (a) USESCH the students general use of ICT at school e.g. browse the Internet for schoolwork and doing homework on a school computer (10 items), (b) ENTUSE the use of ICT outside the school for leisure e.g. playing one-player games and chatting online, (12 items), and (c) HOMESCH use of ICT outside the school for school-work e.g.

browsing the internet to follow up lessons and using social networks for communication with teachers (11 items).

All the items are self-reported and correspond to a 5-point Likert scale about the frequency of each activity (item), ranging from 1 (never or hardly ever) to 5 (everyday), (see Appendix B for the detailed items). OECD standardized these indices to a mean of 0 and SD of 1, therefore higher values of theses indices indicate more frequency of the activity (OECD, 2018b, Chapter 16).

Reading Attitudes: The Mediators (M)

Three reading related-attitudes indices from PISA student background questionnaire were implemented. The three indices, or indicators, are derived variables from numbers of items. The READJOY variable measures the enjoyment of reading—corresponds to the attitude component— by using 5 self-reporting items e.g. “I read only if I have to” and

“reading is one of my favorite hobbies” with four Likert-scale response categories ranging from “strongly disagree” to “strongly agree”. The items of this variable were mixed (negatively and positively) worded, therefore the negatively worded items were reverse- scored to implement IRT scaling so that higher values indicate greater enjoyment of reading (OECD, 2018b, Chapter 16).

The second variable, READCOMP—corresponds to the Subjective Norm

component—is a 3-items derived indictor that measures the student’s self-concept of reading competence, in other words their perception of competence in reading, through 4 Likert scale categories, strongly disagree to strongly agree, all the 3 items are positively worded such as “I am a good reader” therefore higher values indicate greater self-concept of competence.

Finally to represent the third component of the TPB, the Perceived Behavioral Control, the variable measuring student’s self-concept of reading difficulty was used. The variable is constructed from 3 items that are negatively worded e.g. “I have to read a text several times before completely understanding it”. Since the other two indicators were

positively worded, and in order to avoid mixed signals, this variable was reverse-coded so that higher values indicate greater self-conception of ease, not difficulty in reading. Table 4 shows the scale reliabilities for all the independent variables range between acceptable and excellent, as reported by OECD.

Reading Performance: The Outcome (Y)

Reading, the dependent variable in this study, was the main domain of PISA 2018 assessment, therefore more test items focused on reading abilities (245 items) than the other two domains i.e. science (115 items) and mathematics (83 items), this new inclusion of larger number of reading items improved measurement. To further increase measurement validity and accuracy, computer-based multistage adaptive method was used for reading test, in that students receive the next items based on their performance in the ones before (OECD, 2019a, Chapter 2). The 2018 test lasted for two hours and was mainly computer-based. The reading items varied in difficulty level and type including multiple-choice and short-constructed.

According to OECD, its reading framework involved several cognitive processes (Table 2) and the distribution of tasks, or marks, in the reading test corresponded to these cognitive processes (Table 3).

Table 2

Mapping Reading Framework

Category of scaling Cognitive processes

Reported on PISA scale Reading fluently

Locating information Accessing and retrieving information within a text

Searching for and selecting relevant text

Understanding Representing literal meaning

Integrating and generating inferences

Evaluating and reflecting Assessing quality and credibility

Reflecting on content and form Detecting and handling conflict

Since PISA uses rotated-booklet design, a number of 15 clusters (units) of items was developed. Students were not administered the complete set of reading items, instead they responded to 7 units—between 33 and 40 items— depending on which testlet was taken at each stage (OECD, 2019a). This “missing by design” is the reason why PISA is not able to provide one single measure for cognitive domains such as reading, rather, PISA estimates students cognitive abilities using 10 plausible values (PVs) calculated by multiple imputation from both reading test and background questionnaire (Odell et al., 2020; OECD, 2019a, Chapter 16; von Davier et al., 2018), such PVs represent the range of abilities of a student if they had completed the whole test (Willms & Smith, 2005). As per prior recommendations from (OECD, 2009; Rutkowski et al., 2010) all the ten PVs of reading (PV1READ – PV10READ) were included. One dataset for each PV, then combining all parameters following Rubins combination to better measure parameters of the mediation model

(Laukaityte & Wiberg, 2017). Finally, as the case for the other cognitive scales, reading PVs were set with a mean of 500 and a standard deviation of 100.

Table 3

Distribution of Tasks by Targeted Process and Text Type

Multiple Text Single Text

Searching and selecting relevant text 10%

15%

Scanning and locating

15%

Multiple-text inferential comprehension 15%

Literal comprehension

10%

Corroborating/handling conflict 15%

Inferential comprehension

20%

Assessing quality and credibility

Reflecting on content and form

Student Background: The Control Variables

In addition to the ICT, reading attitudes, and reading performance, two student demographic variables were included in the analysis to serve as control variables and to draw groups comparison. The economic, social and cultural status (ESCS) and student’s gender.

ESCS is an index derived from a) parents’ education level, b) parents’ occupation and c) home possessions, to indicate social and economic status of students (OECD, 2018b, Chapter 16). When it comes to SEM, scholars argue whether to measure ESCS in a reflective or formative measurement model (Howell et al., 2007). This study, however, used the OECD scale, i.e. formative measurement.

Data Analysis

This study used full information maximum likelihood (FIML) method under the assumption that they occurred randomly (Enders, 2010) , a method well suited to handle data missing at random (Savalei & Rhemtulla, 2012). All the missing values were recoded to (-99) with the purpose that Mplus software treats them with FIML method (Muthén &

Muthén, 2017). An SPSS (IBM Corp, 2019) diagnose for variance inflation factor (VIF) showed that no variable poses multicollinearity threat, as all VIF values were <5, see Appendix D.1, (Marcoulides & Raykov, 2019). Since PISA uses stratified random sampling in all the participating countries (Westat, 2016), in which schools were sampled at the first stage, and students within schools were sampled in the second stage, such design will lead to different probabilities for students to be chosen as participants in the assessment

(Asparouhov, 2005). To ensure data is nationally and internationally representative, OECD employs weighting where the final student weight include both “school weight (the inverse of the school’s probability of selection) and the within-school student weight (the inverse of the students probability of selection)” (Westat, 2016).

Using these replicate weights in analyses would reduce the sampling bias (Gómez- Fernández & Mediavilla, 2021) as this study does by the virtue of the WEIGHT command in Mplus (Muthén & Muthén, 2017). Finally, by default Mplus implements a robust maximum likelihood estimator (MLR), a widely used estimator to deal with the type of data this study used (Marsh et al., 2013). The command CLUSTER in Mplus was used to account for the nesting i.e. students nested in schools. After testing for two-level structure and the lack of the evidence thereof, the COMPLEX type of analysis was chosen. This type allows to obtain corrected standard errors and chi square test of model fit while considering stratification and non-independence of observations (Muthén & Muthén, 2017; Scherer et al., 2018).

Structural Equation Modelling (SEM)

This study employed Structural Equation Modelling due to the mediation assumption to account for the mediation effect of students’ attitudes towards reading in the relationship between their ICT use and reading achievement. SEM allows a set of regressions equations to be analyzed simultaneously while controlling for measurement error (Jensen et al., 2019).

SEM also provides fit indices that allows evaluation and comparison among models and the extent to which the data body forth the hypotheses (Brown, 2006).

Path Analysis

After assessing the measurement part, it is rational to analyze the structural part. To specify structural model it is constructive to describe the model using path diagrams first introduced by Wright (1934). Path coefficients, manifested in Figure 1, specify and assess the relationships between a) predictors (X) and mediators (M), b) mediators and outcome

variable (Y), c) predictors (X) and outcome variable (Y).

Figure 1

Path Diagram of the Hypothesized Relationship between ICT and Reading with Mediators

In the diagram, path coefficients β₄₄, β₅₄, and β₆₄ represent direct effect on the outcome READ, while all the other path coefficients have indirect effect on the outcome.

The indirect effect is the product of both coefficients of the path i.e. (a*b) as illustrated in (Baron & Kenny, 1986).

By using MODEL INDIRECT command, Mplus will by default implement

multivariate delta method (Wang & Wang, 2012) suggested by (Sobel, 1982) which produce more accurate results for large samples (MacKinnon et al., 2002). In correspondence with the suggested mediation effect this study hypothesizes, this analysis adopted mediation model (MacKinnon, 2008) following the equations in Figure 2 for the path analysis in the model.

Figure 2 SEM Equation

READJOY =  τ

 +  ^β

.ENTUSE +  ^β

.HOMESCH +  β

.uUSESCH + 

.FEMALE + 

.ESCS +  ε

  READCOMP =  τ

 +  ^β

.ENTUSE +  ^β

.HOMESCH +  β

.USESCH + 

.FEMALE + 

.ESCS +  ε

READEASE =  τ

 +  ^β

.ENTUSE +  ^β

.HOMESCH +  β

.USESCH + 

.FEMALE + 

.ESCS +  ε

  READ      =  τ

 +  ^β

.READJOY +  ^β

. READCOMP +  ^β

. READEASE + 

+  ^β

.ENTUSE +  ^β

.HOMESCH +  β

.USESCH +  ^β

.FEMALE +  ^β

.ESCS +  ε

Assuming the following for residual distribution:

𝜀 𝜀 𝜀 𝜀

~  ^MVN  

⎝

⎜⎛ 0 0 0 0

,

⎣⎢

⎢⎢

⎡ 𝜎 𝜎 𝜎 0

𝜎 𝜎 𝜎 0

𝜎 𝜎 𝜎 0

0 0 0 𝜎 ⎦⎥⎥⎥⎤

⎠

⎟⎞

Note. MVN = multivariate normal distribution, N = Normal distribution,

σ² =Standard deviation, Note.

τ  

,   β  

,  ε  

to elaborate the equations Figure 3 shows the matrix of the model.

Figure 3 SEM Matrix

⎣ ⎢

⎢ ⎢

⎡ 𝑅𝐸𝐴𝐷𝐽𝑂𝑌 𝑅𝐸𝐴𝐷𝐶𝑂𝑀 𝑅𝐸𝐴𝐷𝐸𝐴𝑆𝐸 𝑅𝐸𝐴𝐷 ⎦ ⎥ ⎥ ⎥ ⎤

  =  

⎝

⎜ ⎛ τ τ τ τ ⎠

⎟ ⎞    +   

⎝

⎜ ⎜

⎛

0 0 0 0

0 0 0 0

0 0 0 0

β β β 0 ⎠

⎟ ⎟

⎞  

⎣ ⎢

⎢ ⎢

⎡ 𝑅𝐸𝐴𝐷𝐽𝑂𝑌 𝑅𝐸𝐴𝐷𝐶𝑂𝑀 𝑅𝐸𝐴𝐷𝐸𝐴𝑆𝐸

𝑅𝐸𝐴𝐷 ⎦ ⎥ ⎥ ⎥ ⎤    +

⎝

⎜ ⎜

⎛

β β β 0 0

β β β 0 0

β β β 0 0

β β β β β ⎠

⎟ ⎟

⎞

⎣ ⎢

⎢ ⎢

⎡ 𝐸𝑁𝑇𝑈𝑆𝐸 𝐻𝑂𝑀𝐸𝑆𝐶𝐻

𝑈𝑆𝐸𝑆𝐶𝐻 𝐹𝐸𝑀𝐴𝐿𝐸

𝐸𝑆𝐶𝑆 ⎦ ⎥ ⎥ ⎥ ⎤ +

⎝

⎜ ⎛

𝜀

𝜀

𝜀

𝜀

⎠

⎟ ⎞

Note. MVN = multivariate normal distribution, N = Normal distribution,

σ² =Standard deviation, Note.

τ  

,   β  

,  ε  

Model Modification and Evaluation

Considering the clustering nature of the PISA data, a null model with no covariates is required in order to investigate how much variance in the dependent variable i.e. reading achievement can be attributed to within and between clusters (Willms & Smith, 2005). To address the hypotheses mentioned earlier, the analysis introduces a model with mediators to test mediation effect (H1). Then, the same model is extended by including quadratic

predictors i.e. ICT variables to test the potential non-linear relationships between ICT use and reading performance (H2). The quadratic terms are generated by squaring the ICT variables so that X becomes X² (Little, 2013). To avoid a potential bias in estimation caused by the strong intercorrelations among the predictors and their quadratic variables, the variables were centered to their means, the mean-centering will reduce this correlation and therefore the issue of multicollinearity is no longer a concern (Hayes, 2017).

This study invokes an analysis of explorative nature, therefore the analysis adopted the “top-down” approach introduced by Hox et al., (2017) to build the model, this approach comprise two steps as described by West et al., (2006): first, including all the effects in the model then removing insignificant effect(s), afterwards, including all effects then removing significant ones.

To improve the model, modification indices (MIs) were used to diagnose the model fit and to identify its misspecification. MIs reveal the decrease in the χ² statistics of the model with 1 df if a given parameter is freed from model constraints (Sörbom, 1989). Although there is no certain strict cut-off for how large the MIs should be to be considered, a 3.84 decrease in χ² with 1 df indicates a P = 0.05 fit improvement in the model (Wang & Wang, 2012), a significant fit this study adopted. Parameters with high MIs were freed one at a time, in descending order, as each freeing will results in change in other parameters (MacCallum et al., 1992). Further, the expected parameter change (EPC), which are measures that estimates the expected change in the value of a certain parameter if that parameter was freed (Saris et al., 1987). Chou & Bentler (1993) developed fully—in contrast to Kaplan 1989 partially— standardized EPC which is, unlike Kaplan’s, invariant under different scalings of observed and latent variables, hence favored by this study. Mplus provides both MIs and fully standardized EPC for all fixed and constrained parameters.

Results with unrealistic values, was not considered. Last, worth mentioning that the modification process was not blindly decided upon statistics only, but thoughtfully justified through theoretical basis. When evaluating the SEM goodness-of-fit, this study referred to the widely used (L. Hu & Bentler, 1999), cut-offs that is ≤.06 for (RMSEA); ≤.08 for (SRMR); ≥.95 for (CFI) and Tucker-Lewis index (TLI). These cut-offs can be less strict when evaluating the models with WLSMV estimator (DiStefano & Morgan, 2014).

Results Descriptive Statistics and Correlations

Table 4 shows all the variables’ univariate parameters of the population. Glancing the table, the sample sizes are relatively proximate, also stands out that ICT variables are more homogeneous between Denmark and Sweden than Finland. All the three countries score above the international average of reading performance (453.4), Finland comes first amongst the three, and second in Europe. As for the predictors, students in the three

countries reported notably higher than the international average (0.02) use of ICT devices at school, likewise, in Denmark and Sweden the mean score of the ICT use outside school – both for leisure and schoolwork— is higher than international average (0.09 for schoolwork, 0.02 for leisure), interestingly the Finnish students reported lower averages than the

international ones for both variables. For ICT use over all view see Appendix D 6.

ESCS averages of the three countries put them within the international upper half of the third quartile (ESCS international average is -0.28). With regard to data normality, following both Hair et al., (2006) and Byrne (2011) suggested cut offs for normality when the skewness is between ±2 and kurtosis between ±7, the statics shows that normality is

maintained for all variables except for minor immoderate kurtosis of ENTUSE. For broader view, Figure 6 visualizes the distribution differences in ICT use variables between the countries. However, Mplus estimator is prepared for possible non-normality when implementing the model (Muthén & Muthén, 2017).

*Note: Reading was averaged from 10 plausible values.

Denmark

FEMALE 7657 0.50 0.50 0 0 1 0.01 -2

IMMI1GEN 7409 0.04 0.19 0 0 1 4.96 22.57

IMMI2GEN 7409 0.17 0.38 0 0 1 1.73 0.99

ESCS 7431 0.39 0.84 0.56 -6.73 3.89 -0.89 1.31

HOMESCH 6038 0.17 0.83 0.13 -2.7 3.32 0.82 5.51 0.873

USESCH 5939 0.62 0.68 0.56 -1.79 3.35 1.29 6.13 0.788

ENTUSE 6358 0.05 0.87 -0.06 -3.59 4.24 1.62 9.53 0.803

JOYREAD 7341 -0.34 1.02 -0.33 -2.73 2.66 0.07 0.66 0.855

READCOMP 6781 0.31 0.94 0.12 -2.44 1.88 0.13 -0.22 0.856

READEASE 6773 0.13 0.93 0.07 -2.78 1.89 -0.01 0.28 0.691

Reading 7657 501.13 91.18 493.13 188.7 745.94 -0.18 -0.37 0.93

Finland

FEMALE 5649 0.49 0.5 0 0 1 0.04 -2

IMMI1GEN 5524 0.03 0.18 0 0 1 5.28 25.88

IMMI2GEN 5524 0.02 0.15 0 0 1 6.18 36.23

ESCS 5557 0.30 0.79 0.42 -4.24 3.28 -0.51 0.25

HOMESCH 4888 -0.28 0.89 -0.23 -2.3 3.31 0.46 3.10 0.931

USESCH 4873 0.18 0.75 0.13 -1.72 3.3 1.13 4.92 0.892

ENTUSE 5050 -0.04 0.81 -0.1 -3.59 4.24 1.02 9.70 0.794

JOYREAD 5489 -0.25 1.11 -0.24 -2.71 2.66 0.13 0.29 0.872

READCOMP 5396 0.09 1 0.12 -2.44 1.88 0.11 -0.12 0.864

READEASE 5407 0.11 1.05 0.07 -2.78 1.89 -0.2 0.06 0.798

Reading 5649 520.33 96.13 528.27 150.22 791.43 -0.39 -0.09 0.94 Sweden

FEMALE 5504 0.50 0.50 1 0 1 -0.01 -2

IMMI1GEN 5338 0.09 0.29 0 0 1 2.79 5.80

IMMI2GEN 5338 0.10 0.31 0 0 1 2.59 4.71

ESCS 5348 0.36 0.89 0.54 -5.48 3.90 -0.84 1.29

HOMESCH 4603 0.15 0.94 0.07 -2.30 3.31 0.99 3.94 0.920

USESCH 4576 0.40 0.88 0.36 -2.54 3.27 0.37 3.75 0.877

ENTUSE 4819 0.04 0.95 -0.07 -3.59 4.24 1.69 9.28 0.805

JOYREAD 5296 -0.32 1.10 -0.27 -2.73 2.66 0.03 0.40 0.852

READCOMP 5220 0.32 1.04 0.12 -2.44 1.88 -0.12 -0.36 0.872

READEASE 5215 0.05 1.07 0.07 -2.78 1.89 -0.15 -0.03 0.769

Reading 5504 505.53 103.7 512.42 157.07 774.58 -0.34 -0.29 0.94

between ICT variables and reading attitudes are significant yet not strong (r ≤ 0.09). In

contrast, students who more often use ICT devices at any given place (inside or outside school) for a given purpose (for leisure or schoolwork) would more often use them at the other place for the other purpose (r ≥ 0.43). Background variables correlate differently with other variables, while higher ESCS is strongly associated with positive attitudes towards reading and reading achievement as well, being an immigrant, first or second generation, is negatively associated with reading achievement. Lastly, girls seem to enjoy reading more than boys who tend to spend more time using ICT devices for leisure. For the big picture and to capture the relationships between ICT use and reading achievement a simple plot was created for the overall sample and for each country to see the difference among them in vision. Figure X shows that for the overall sample and each country separately there is an inverted bell-shape that indicate a nonlinear relationship.

Testing for Multilevel Structure

PISA uses clusters design in its assessments, where students are nested in schools and schools are nested in countries, therefore it is a advised to consider hierarchical linear modelling (HLM) as it accounts for this clustering and consequently avoid statistical issues (Goldstein, 2011). To test for HLM a null-model with no covariates was constructed to investigate the intraclass correlation coefficients (ICC), which indicates the proportion of variance that is accounted for by the group level, the result showed only 16% of the variation is due to clusters, which is below the 20% cut-off to consider HLM, this is explained by the shared educational, social, and economical systems of the three countries. For this reason single (student) level modelling was favored over multilevel modelling (Snijders & Bosker, 2011).

Correlations with Confidence Intervals for the Total Sample (for each country see Appendix D 2)

Variable 1 2 3 4 5 6 7 8 9 10

1. FEMALE

2. IMMI1GEN –0.01

[–.02, .00]

3. IMMI2GEN 0.02* –0.08**

[.00, .03] [–.10, .07]

4. ESCS –.00 –0.17** –0.23**

[–.02, .01] [–.19, .16] [–.24, –21]

5. HOMESCH –0.05** 0.07** 0.11** 0.09**

[–.06, .03] [.05, .09] [.10, .13] [.07, .10]

6. USESCH –0.08** 0.05** 0.07** 0.07** 0.62**

[–.10, .07] [.03, .06] [.06, .09] [.06, .09] [.61, .63]

7. ENTUSE –0.23** 0.01 0.01 0.07** 0.46** 0.43**

[–.24, .21] [–.00, .03] [–.01, .02] [.06, .09] [.45, .47] [.42, .44]

8. JOYREAD 0.25** 0.06** 0.05** 0.14** 0.02** –0.04** –0.09**

[.23, .26] [.05, .08] [.04, .07] [.13, .16] [.01, .04] [–.05, .02] [–.11, .08]

9. READCOMP 0.03** –0.03** 0.03** 0.17** 0.05** 0.04** 0.07** 0.32**

[.01, .04] [–.04, .01] [.01, .04] [.16, .19] [.03, .07] [.02, .05] [.05, .08] [.31, .34]

10. READEASE –0.04** –0.05** –0.03** 0.14** –0.05** –0.05** 0 0.20** 0.50**

[–.06, .03] [–.07, .04] [–.05, .02] [.13, .16] [–.07, .03] [–.06, .03] [–.01, .02] [.18, .21] [.49, .51]

11. Reading 0.18** –0.22** –0.21** 0.32** –0.18** –0.19** –0.09** 0.31** 0.36** 0.34**

[.17, .20] [–.23, .20] [–.22, .20] [.31, .34] [–.19, .16] [–.21, .18] [–.10, .07] [.30, .33] [.35, .38] [.33, .35]

Note. Values in square brackets indicate the 95% confidence interval for each correlation. The confidence interval is a plausible range

.

Note. Red = Outside school for leisure. Blue = Outside school for schoolwork. Green = At school in general

Path Models

The model without mediation i.e. X to Y showed less that desired fit indices CFI = .85, TLI = .70, RMSEA = .077, SRMR = .064, but a significant relationship between the ICT use and reading. The introduction of mediators Joy, Confidence, and Ease has further improved the model fit as well as significantly impacted the paths between X and Y.

Figure 5 represents the direct relations between X and Y. All the standardized path coefficients for overall sample (Nordic) and for each country are presented in Table 6, most of which deem to be significant. The model fit indices showed good fit as per Hu and Bentler (1999) for the overall sample and each country with Finland having the fittest model (CFI = .987, TLI = .923, RMSEA = .057, SRMR = .025).

ICT Use to Reading Achievement (Direct Effect)

The ICT use in general do associate – both negatively and positively— with reading achievement directly. This association is interesting, that it negatively predicts reading performance when the ICT devices are used at school (USESCH) or for schoolwork purposes outside of school (HOMESCH). Conversely, it positively predicts reading performance when students use ICT devices for leisure (ENTUSE). On a closer look, this association revealed a sharp divide between the three countries. First, Finnish students reading performance is, and to a greater extent, negatively associated with HOMESCH. The same association is meager in Sweden and even positive in Denmark. Second, both Danish and Swedish students READ is negatively predicted by USESCH, while the same path is insignificant for Finnish

students. Lastly, using ICT for leisure purposes seems to be a greater positive predictor for Danish and Swedish than Finnish students.

Figure 5

Path Diagram Results

Reading attitudes as Mediators (Indirect Effect)

The four conditions for a mediation model assumed by Baron & Kenny (1986) are satisfied. First the effect of ICT variables on reading performance is statistically significant.

Second, the regression of mediators on ICT variables are statistically significant (β ≥ .148, SE ≤ .016, P = 0) for further details (see appendix D.3). Third, the regression of reading on both ICT variables and reading attitudes is statistically significant albeit the small size effect of the regression. Lastly, the direct effect of ICT variables on reading performance is reduced when mediators are introduced. Further exploring the mediation paths amongst countries makes clear the divide. While enjoying reading negatively mediates the influence of ICT use on reading performance in Finland; this mediation is positive or slight in Denmark and Sweden. Also, Perception of reading competence showed substantial mediation effect for all the ICT variables in Finland and Sweden in comparison to slight effect in Denmark.

Perception of reading ease substantially and positively mediated the effect of all predictors in Finland, but negatively mediated the effect in both Sweden and Demark.

Without Mediation With Mediation

Path Nordic Nordic Denmark Finland Sweden

β SE β SE β SE β SE β SE

Intercept 487.153 3.866*** 495.8 3.870*** 448.5 7.390*** 511.0 4.936*** 480.1 8.355***

ENTUSE (total) 0.157 0.012*** 0.225 0.019*** 0.157 0.017*** 0.172 0.020***

direct 0.158 0.01*** 0.202 0.012*** 0.247 0.017*** 0.168 0.017*** 0.204 0.020***

Total indirect –0.046 0.004*** –0.022 0.008** –0.011 0.006 –0.032 0.006***

via Joy –0.008 0.003* –0.008 0.006 –0.008 0.004 0.001 0.004 via Confidence –0.032 0.004*** –0.003 0.007 –0.021 0.004*** –0.026 0.005***

via Ease 0.202 0.012*** –0.011 0.003*** 0.018 0.005*** –0.007 0.003*

HOMESCH (total) –0.109 0.014*** –0.008 0.021 –0.261 0.022*** –0.061 0.021**

direct –0.112 0.014*** –0.062 0.014*** 0.002 0.021 –0.219 0.023*** –0.020 0.021*

Total indirect –0.047 0.004*** –0.009 0.004 –0.042 0.009*** –0.040 0.006***

via Joy –0.006 0.003** –0.003 0.002 –0.011 0.006 0.001 0.005 via Confidence –0.034 0.004*** –0.001 0.003 –0.051 0.008*** –0.029 0.005***

via Ease –0.006 0.002** –0.005 0.002* 0.019 0.005*** –0.012 0.005*

USESCH (total) –0.212 0.012*** –0.244 0.019*** –0.101 0.023*** –0.236 0.019***

direct –0.212 0.012*** –0.156 0.012*** –0.216 0.020*** –0.045 0.025 –0.162 0.019 Total indirect –0.056 0.005*** –0.029 0.006*** –0.055 0.011*** –0.075 0.008 via Joy –0.008 0.004* 0.019 0.004*** –0.014 0.008 0.002 0.007 via Confidence –0.037 0.004*** –0.001 0.003 –0.066 0.011*** –0.059 0.009***

via Ease –0.011 0.004** –0.022 0.005*** 0.024 0.006*** –0.018 0.008*

Path Without Mediation Nordic Denmark Finland Sweden

β SE β SE β SE β SE β SE

ESCS (total) 0.305 0.010*** 0.307 0.01*** 0.287 0.015*** 0.289 0.015*** 0.320 0.016***

FEMALE (total) 0.136 0.008*** 0.124 0.008*** 0.103 0.013*** 0.172 0.014*** 0.098 0.015***

READ R² 0.202 0.008 0.777 0.008 0.221 0.013 0.255 0.011 0.237 0.013

Est SD Est SD Est SD Est SD Est SD

AIC 360740 124 487192 116 196776 58 135558 38 141164 46

BIC 360943 124 487622 116 197156 58 135922 38 141526 46

χ² 1409.44 14.93*** 507.95 12.27*** 114.26 3.037*** 149.51 6.22*** 254.85 8.53***

RMSEA 0.077 0.000 0.059 0.001 0.042 0.001 0.057 0.001 0.077 0.001 CFI 0.856 0.001 0.973 0.001 0.980 0.001 0.987 0.001 0.969 0.001 TLI 0.701 0.003 0.837 0.004 0.880 0.005 0.923 0.003 0.808 0.006 SRMR 0.064 0.000 0.030 0.000 0.016 0.000 0.025 0.000 0.032 0.000 Note. *** p < .0001, ** p < .01, * p < .05; df = 8.

Path Model with Non-linear Relations

Introducing quadratic terms (see Figure 5) showed further improved model fit indices than the previous models (see Table 6), CFI = .979, TLI = .925, RMSEA = .043, SRMR = .027. The quadratic ICT variables showed significant quadratic direct relationship with reading performance, ENTUSESQ: β = – .048, P = 0; HOESCHSQ: β = .062, P = 0;

USESCHSQ: β = –.044, P = .012. Such results suggest curvilinear relationship aligning with findings from Gubbles (2020). The negative effects indicate a diminishing marginal return, that at a certain threshold after which the opposite effect takes place. The quadratic variables, however, seem to have non-significant indirect relationships with reading. The only one significant indirect quadratic effect is USESCHSQ through confidence in Finland, β

= –.026, P = 0.

Figure 6

SEM with Quadratic Variables

Note. For ease of presentation 1. non-significant paths were deleted, 2. Correlation paths between predictors deleted.

Path Nordic Denmark Finland Sweden β SE β SE β SE β SE Intercept 495.81 3.870*** 448.52 7.390*** 511.03 4.936*** 480.18 8.355***

ENTUSE (total) 0.131 0.014*** 0.208 0.022*** 0.145 0.017*** 0.150 0.024***

direct 0.177 0.013*** 0.226 0.021*** 0.153 0.018*** 0.184 0.023***

Total indirect –0.045 0.005*** –0.018 0.008* –0.009 0.006 –0.033 0.007***

via Joy –0.007 0.003* –0.006 0.006 –0.007 0.004 0.001 0.004 via Confidence –0.033 0.004*** –0.002 0.006 –0.018 0.004*** –0.027 0.006***

via Ease –0.005 0.002*** –0.011 0.003*** 0.016 0.004*** –0.007 0.003*

ENTUSESQ (total) –0.048 0.011*** –0.051 0.022* –0.092 0.016*** –0.031 0.020 direct –0.044 0.011*** –0.051 0.022* –0.089 0.016*** –0.026 0.020 Total indirect –0.004 0.002* –0.001 0.002 –0.003 0.003 –0.005 0.003

via Joy 0.000 0.000 0.000 0.002 0.000 0.001 0.001 0.000

via Confidence –0.004 0.002** 0.000 0.000 –0.002 0.003 –0.004 0.003 via Ease 0.000 0.000 0.000 0.000 –0.001 0.001 –0.001 0.001 HOMESCH (total) –0.130 0.015*** –0.005 0.021 –0.310 0.030*** –0.064 0.021**

direct –0.082 0.015*** 0.003 0.022 –0.271 0.031*** –0.024 0.022

Total indirect –0.048 0.004*** –0.008 0.004* –0.039 0.008*** –0.040 0.006***

via Joy –0.006 0.003** –0.002 0.002 –0.010 0.006 0.001 0.005 via Confidence –0.035 0.004*** –0.001 0.003 –0.043 0.007*** –0.028 0.005***

via Ease –0.006 0.002** –0.005 0.002* 0.014 0.004*** –0.012 0.005*