Explorer The Student Expectations of Learning Analytics Questionnaire (SELAQ)

Student engagement within the development of learning analytics services in Higher Education is an important challenge to address. Despite calls for greater inclusion of stakeholders, there still remains only a small number of investigations into students ’ beliefs and expectations towards learning analytics services. Therefore, this paper presents a descriptive instrument to measure student expectations (ideal and predicted) of learning analytics services. The scales used in the instrument are grounded in a theoretical framework of expectations, specifically ideal and predicted expectations. Items were then generated on the basis of four identified themes (Ethical and Privacy Expectations, Agency Expectations, Intervention Expectations, and Meaningfulness Expectations), which emerged after a review of the learning analytics literature. The results of an exploratory factor analysis and the results from both an exploratory structural equation model and confirmatory factor analysis supported a two ‐ factor structure best accounted for the data pertaining to ideal and predicted expectations. Factor one refers to Ethical and Privacy Expectations, whilst factor two covers Service Feature Expectations. The 12 ‐ item Student Expectations of Learning Analytics Questionnaire (SELAQ) provides researchers and practitioners with a means of measuring of students ’ expectations of learning analytics services.


| INTRODUCTION
Learning analytics (LA) is commonly defined as "the measurement, collection, analysis and reporting of data about learners and their contexts, for purposes of understanding and optimising learning and the environments in which it occurs" (Siemens & Gašević, 2012). As we have previously stated (Whitelock-Wainwright, Gašević, and Tejeiro, 2017), the implementations of LA into higher education institutions can be viewed as a service offered to optimize learning and learning environments. For example, the Open University has implemented initiatives that aim to improve retention rates (Calvert, 2014). Put differently, this higher education institution implemented LA as a service with the aim of optimizing student learning, specifically with a specific view of increasing retention rates. Thus, although LA refers to the general field, including the research undertaken, LA services relate to eventual functionalities that are implemented within an educational setting. In terms of actual LA service implementations, its higher education institutes continue to remain within the exploratory stages of such pursuits (Ferguson et al., 2016;Tsai et al., 2018;Tsai & Gašević, 2016), with most institutes being at the fringes of developing institution-wide LA systems. This parallels what has been referred to as a definition stage in information system development, where focus is placed on making decisions as to what data are collected and fed back and what the system will do (Ginzberg, 1981). At this stage, successful implementation of information systems rests on the inclusion of stakeholders early on their development so that designers can identify and assimilate various expectations to reduce the likelihood of service dissatisfaction in the future (Brown, Venkatesh, & Goyal, 2014;Ginzberg, 1975).
Although the need for the early engagement of stakeholders has been specifically highlighted for LA (Drachsler & Greller, 2016;Ferguson et al., 2014), there are limited instances where this is actually happening (Tsai & Gašević, 2017a). Without stakeholder engagement, it is likely that the multitude of LA policies available (Sclater, 2016) are driven primarily by the institutional managers' expectations and beliefs.
In those cases, even if the key driver for the intention to adopt LA is to improve learning performance (Tsai & Gašević, 2017b) and to provide additional support to learners (Siemens & Gašević, 2012), that intention is still shaped by the managers' preconceived beliefs and ideas-not necessarily reflective of what other stakeholders (e.g., students) would expect. This may perpetuate an ideological gap (Ng & Forbes, 2009) whereby services reflect a difference between what institutions believe students should receive and what students expect to receive.
LA, by definition, is student centred (Siemens & Gašević, 2012), but relatively few attempts have been made to explore students' beliefs towards the use of LA (Arnold & Sclater, 2017;Ifenthaler & Schumacher, 2016;Roberts, Howell, Seaman, & Gibson, 2016;Schumacher & Ifenthaler, 2018;Slade & Prinsloo, 2014). As shown in the LA dashboard evaluation work of Park and Jo (2015), students expressed negative opinions towards being provided with visualizations of login frequency metrics, particularly on the basis of them not being pedagogically meaningful. This is concerning, particularly with the attention placed on relaying resource usage statistics (75% of 93 student-facing LA dashboard articles, according to Bodily & Verbert, 2017), as it exemplifies how LA has largely overlooked student expectations. Adding to this is the finding that only 6% of 93 articles that have detailed LA dashboard implementations have explored student expectations of such services (Bodily & Verbert, 2017). Given the importance of actively exploring and gauging stakeholder expectations, particularly with regard to future service satisfaction and usage (Brown et al., 2014;Brown, Venkatesh, & Goyal, 2012), student engagement cannot continue to be at a nominal level. Instead, it is necessary for research to address this gap through the provision of tools that enable higher education institutions to open dialogues with students to understand the LA service they expect.
From those limited investigations with students, findings have shown that although students have strong expectations towards the institution's handling of educational data (Roberts et al., 2016;Slade & Prinsloo, 2014) and the LA service features offered (Roberts, Howell, & Seaman, 2017;Schumacher & Ifenthaler, 2018), despite largely being unaware of LA practices (Roberts et al., 2016). In light of such findings, it can be argued that despite student exposure to LA services being limited, they are able to form expectations towards the procedures undertaken and the services offered. Moreover, given the relatively small proportion of LA implementations readily assessing what students expect of such services, there is a need to address this limitation.
As a means to gauge stakeholder expectations of a possible service, Szajna and Scamell (1993) have encouraged the use of psychometric instruments during different stages of implementations.
Within the context of LA, a measure is available to assess an institute's readiness for LA (Oster, Lonn, Pistilli, & Brown, 2016), but no preexisting scale is available to gauge student expectations of LA services.
Even though Arnold and Sclater (2017) used a survey to understand student perceptions of data handling, their reported findings can be questioned on the basis of using an on the fly scale. Schumacher and Ifenthaler (2018) do, however, present an exploration of expected LA dashboard features from the perspective of students. Although these authors ground this work in expectations, the distinction between expectations and perceptions is not completely conceptualized. As a great majority of the student population is unlikely to have experienced institutional LA services, measures of experience (perceptions; Parasuraman, Zeithaml, & Berry, 1988) are not always appropriate, particularly given that majority of students are not acquainted with LA services (Roberts et al., 2016). Expectations, however, can be measured prior to implementations and are an important determinant in the acceptance of systems (Davis & Venkatesh, 2004).
As indicated above, although the importance of systematically gathering university students' expectations about LA is of paramount importance for the success of the service, little has been done in this regard and no adequate tool is still available. In the present research, we have attempted to close this gap by developing and validating a descriptive questionnaire to collect students' expectations of LA services. Throughout the development of this instrument, the accessibility and understanding of the items from the student perspective were always considered. Put differently, although students are largely unaware of LA services, the phrasing of each item had to be balanced between providing an institution with an informative understanding of what students expect and also general enough for all students to understand. In doing so, the university can identify particular areas of focus for their LA implementation, which can then inform direct engagement strategies with their students.

| Expectations as beliefs
A widely utilized definition of belief presents it as "the subjective probability of a relation between the object of the belief and some other object, value, concept, or attribute" (Fishbein & Ajzen, 1975, p. 131). For example, a student may hold a belief that they themselves have the knowledge and skills required to attain a good grade. An expectation, on the other hand, can be defined as "the perceived likelihood that a product possesses a certain characteristic or attribute, or will lead to a particular event or outcome" (Olson & Dover, 1976, p. 169). An example of this would be a judgement of whether a future LA service will enable users to receive a full breakdown of their learning progress. Taking both aforementioned terms into consideration, the only discernible difference is the point in time at which the judgement relates to; that is, expectations are framed as beliefs about the future (Olson & Dover, 1976).
Expectations are an important feature of human cognition (Roese & Sherman, 2007). From the behaviours an individual enacts to the motivation they exert, there is an underlying influence of how they expect to manage within a particular setting (Bandura, 1977(Bandura, , 1982Elliot & Church, 1997). In relation to the judgements we form, our expectations are an anchor to which we compare our actual experiences (Christiaens, Verhaeghe, & Bracke, 2008;Festinger, 1957). As a term, however, an expectation is quite ambiguous, particularly in light of the decomposition presented by Thompson and Suñol (1995). For these authors, expectations can broke down into four subtypes: ideal, predicted, normative, and unformed (Thompson & Suñol, 1995). An ideal expectation refers to a desired outcome or what an individual hopes for in a service (Leung, Silvius, Pimlott, Dalziel, & Drummond, 2009), whereas a predicted expectation is a realistic belief, an individual's view of the service they believe is the most likely to receive. Evidence does support the view that predicted and ideal expectations are two different subtypes (Askari, Liss, Erchull, Staebell, & Axelson, 2010;David, Montgomery, Stan, DiLorenzo, & Erblich, 2004;Dowling & Rickwood, 2016). The two remaining expectation subtypes relate to what service users believe they deserve from a service (normative expectation) and the circumstances where they are unable to form a set of expectations (unformed expectations).
The importance of focusing on service user expectations has been demonstrated in both health services (Bowling et al., 2012;Thompson & Suñol, 1995) and technology adoption research (Brown et al., 2012;Brown et al., 2014;Davis & Venkatesh, 2004). In the case of Bowling et al., these researchers explored patients' ideal and predicted expectations as it allowed for both an upper and lower reference point with regard to knowing what service elements to focus on. Put differently, the responses present an idealized perspective of a service and also a realistic profile of what users believe is most likely. This approach would be advantageous for LA service implementation decisions as it can differentiate between what features students would like and what should be a priority (i.e., what is realistically expected). In addition to providing a deeper understanding of stakeholder perspectives, both research streams have shown that failure to gauge user expectations can lead to dissatisfaction and low adoption of the implemented service (Bowling et al., 2012;Brown et al., 2012;Brown et al., 2014;Davis & Venkatesh, 2004). Thus, by measuring stakeholder expectations towards a service early on the service implementation process, the provider can proactively identify main areas of focus and manage expectations.
Together, these abovementioned theoretical concepts and considerations outlined constitute our reference framework. For the present work, an expectation is defined as a belief about the likelihood that future implementation and running of LA services will possess certain features. Also, our approach is based on the need to consider separately the desired outcomes (ideal expectations) and the realistic beliefs (predicted expectations).

| Research aim
Measuring student expectations of LA services is a fundamental step to the success of future implementations. Although others have offered solutions (Arnold & Sclater, 2017;Schumacher & Ifenthaler, 2018), the use of inconsistent terminology, limited scope, and methodological limitations does leave a lot to be desired. Using the identified expectation themes (ethics and privacy, agency, intervention, and meaningfulness) and expectation types (ideal and predicted), we aim to develop and validate a descriptive questionnaire that offers a robust and methodologically sound solution to measuring student expectations of LA services. An overview of the steps taken in the current work is presented in Figure 1. This figure provides a breakdown of each of the three studies undertaken, a description of how the items were generated or how the data were analysed, the number of items retained or dropped, and how many responses were collected at each stage. Furthermore, to illustrate the utility of the instrument in measuring students' expectations of LA services, we will present a brief overview of how beliefs towards certain features vary in accordance to the two expectation types (ideal and predicted). It is anticipated that being able to gauge and measure student expectations of potential LA services will promote further engagement with these stakeholders in the implementation process, with a view of understanding the specific requirements of the student population.

| Scale development
Items for the questionnaire were created on the basis that students are largely unaware of LA services (Roberts et al., 2016) and adoption rates of LA services at an institutional level being low (Tsai & Gašević, 2017b). Thus, the aim was to phrase items so they would be accessible to all students and to provide institutions with a general understanding of what their student population expect of LA services.
Underlying this was the view that by having a general measure of student expectations, a higher education institution can begin to open dialogues with students during the implementation process, as is recommended in the technology adoption literature (Brown et al., 2012;Brown et al., 2014).
Given that there is no model of student expectations towards LA services to draw upon, the review of the literature was guided by an overarching aim of identifying themes raised in by students in qualitative interviews or by research streams in LA. It is important to remain cognizant of the limitations of the adopted approach to item generation, particularly as it may become skewed towards a particular viewpoint (Streiner et al., 2015). Nevertheless, the process tried to identify key areas of LA services that could be applicable to the student perspective.
From the literature review and expert feedback, we identified four general themes characterizing LA services (Whitelock-Wainwright et al., 2017): ethical and privacy expectations, agency expectations, intervention expectations, and meaningfulness expectations. It is important to acknowledge that these themes represent categories that embody different research streams and discussions within LA. At no point did we hypothesize that the final model would be composed of these constructs nor did we assume that these themes were orthogonal from one another. Put differently, the themes pertaining to agency, intervention, and meaningfulness are likely to be closely linked, but we discuss them here as separate components for clarity purposes. Each theme is discussed in turn, with an emphasis on how it links to the student perspective.

| Ethical and privacy expectations
The LA literature is replete with discussions over the provision of a service that is ethical in the collection, handling, and analysis of student data (Arnold & Sclater, 2017;Drachsler & Greller, 2016;Prinsloo & Slade, 2015;Sclater, 2016;Slade & Prinsloo, 2014). Here, authors tend to highlight the importance of transparency and consent in LA services (Prinsloo & Slade, 2015;Sclater, 2016). The importance of engaging with students within the data-handling decision process (e.g., what data are used and how it will be interpreted) has been stressed by Prinsloo and Slade (2015), who believe it to be key to the progression of LA services.
From those studies exploring student perspective of ethical issues surrounding LA services, they have been shown to hold strong expectations towards data-handling processes. In their interviews with students, Slade and Prinsloo (2014) found a clear expectation that the institution should seek informed consent, or at least permit opting out, when it comes to an LA process. Similar remarks were also expressed in the work of Roberts et al. (2016), who found students to expect the university to respect privacy, to seek informed consent, and to be transparent at all times. Finally, the work of Ifenthaler and Schumacher (2016) showed that although students were against the processing of identifiable data, they were open to data pertaining to their studies being used.
From each of these aforementioned studies, it is clear that students have strong expectations regarding their privacy and being able to make independent decisions about how their data are used (Ifenthaler & Schumacher, 2016;Roberts et al., 2016;Slade & Prinsloo, 2014). More importantly, each of these authors stress the importance of the university actively engaging students in LA service FIGURE 1 Diagrammatic overview of the Student Expectations of Learning Analytics Questionnaire development and validation steps implementation decisions. Thus, based on these two points, the theme of ethical and privacy expectations was decided upon, which was considered to cover elements of data security and consent.

| Agency expectations
When asked about their expectations towards LA services as a form of additional support, students do not expect it to undermine their ability to be self-determined learners (Roberts et al., 2016).
For those students in the samples used by Roberts et al., they consider being an independent learner a fundamental requirement for university; thus, LA services should not foster a dependency on metrics.
These student views resonate with the concerns towards the obligation to act raised by Prinsloo and Slade (2017). Within their discussions on this topic, Prinsloo and Slade do state that the analysis of student data should be guided by a view of providing improved support but at no point should it undermine their (the students') responsibility to learn. This view has further been captured in the concerns raised by Kruse and Pongsajapan (2012), who view interventioncentric LA services as creating a culture of passivity. Put in a different way, LA services that are designed to intervene when students are struggling ignore their ability to be self-directed learners who continually evaluate their progress to set goals (Kruse & Pongsajapan, 2012). The importance of viewing students as active agent in their own learning should be a central tenant to LA services (Gašević, Dawson, & Siemens, 2015;Winne & Hadwin, 2012). Therefore, institutions should be considerate of this and not implement LA services that remove the ability for students to make their own decisions on the data received (Slade & Prinsloo, 2013;Wise, Vytasek, Hausknecht, & Zhao, 2016).
Taken together, students hold an expectation of wanting to remain as independent learners if any LA service were to be implemented, which is also advocated by some researchers. Nevertheless, examples of LA services such as Course Signals are focused upon early alerts (Arnold & Pistilli, 2012). This establishes the importance of the theme of agency expectations, which we consider as introducing a much needed student perspective on who bears the main responsibility for learning under LA services (the student or institution). In doing so, it will add to the previous discussions raised by students and researchers (Prinsloo & Slade, 2017;Roberts et al., 2016).

| Intervention expectations
The anticipated output following the collection and analysis of student data is the introduction of a service designed to optimize both the student learning and the learning environment (Siemens & Gašević, 2012). Despite this aim to support students, there have been few attempts to know what LA services features students want (e.g., 6% of LA dashboard research undertook a needs assessment; Bodily & Verbert, 2017). As stressed in the work of Schumacher and Ifenthaler (2018), student expectations of LA service features should be considered prior to any implementation.
Thus, as with any technology implementation (Brown et al., 2012;Brown et al., 2014;Davis & Venkatesh, 2004), steps should be taken to understand what is expected from the main stakeholders to ensure future acceptance.
Types of LA services offered in the literature vary with respect to the educational problem they seek to resolve. A common service implementation has been the identification of students who are underperforming or at risk (Campbell, DeBlois, & Oblinger, 2007). In undertaking this pursuit, there is a belief that interventions can be actioned to mitigate the possibility of the student dropping out (Gašević, Dawson, Rogers, & Gašević, 2016), although this may not always be the case (Dawson, Jovanovic, Gašević, & Pardo, 2017).
Other approaches have moved away from building predictive models to identify at-risk students; instead, focusing on the development of systems aimed at improving the student-teacher relationship (Liu, Bartimote-Aufflick, Pardo, & Bridgeman, 2017) or presenting graphical overviews of learner behaviour (Verbert, Duval, Klerkx, Govaerts, & Santos, 2013). In all cases, the services are designed to with a view to improve education for students, but there is a prevailing absence of researchers gauging what students expect of these services.
Of those studies seeking to understand what students expect of LA services, the findings have presented an important perspective that institutions cannot overlook. For Roberts et al. (2016), some students did not desire a service that allowed for peer comparisons, stating that they were unnecessary. When asked about their views towards receiving information on progress (e.g., underperforming), students did not expect such services on account of the unnecessary anxiety it would create (Roberts et al., 2016). From the work carried out by Schumacher and Ifenthaler (2018), students expected to receive LA service features that facilitated self-regulated learning, which included real-time feedback and updates on how progress compares to a set goal. Similarly, Roberts et al. (2017) found students to expect services such as dashboards to be customizable and contain features to set goals and track progress.
With regard to the LA service features being developed, it appears that researchers are aiming to improve both the learning experience and the learning environment. The issue, however, is that these developments are primarily guided by the views of the researchers, not the students, which may lead to features that are not expected (e.g., the provision of login metrics in Park & Jo, 2015). Student perspectives, on the other hand, show them to expect features that support them being self-directed learners, as opposed to making them passive recipients of a service. Thus, the theme of intervention expectations was proposed, which entails the various types of service features commonly offered in the LA literature and those raised in the student perspective work.

| Meaningfulness expectations
Closely entwined with both agency and intervention expectations is the theme of meaningfulness expectations. Whereas agency expectations captures the importance of students being independent learners and intervention expectations refer to the LA service features, meaningfulness expectations relates to the utility of information fed back to students. More specifically, meaningfulness expectations are associated with the student perspectives towards the information conveyed in LA service features and whether this has any meaning for their learning.
Introducing new forms of feedback as a result of implementing LA services should, theoretically, promote positive changes in student behaviour such as motivating learning (Park & Jo, 2015;Verbert et al., 2013). However, if meaningful inferences about learning progress cannot be drawn from the information received through LA services (i.e., how visual representations of performance relates to personal learning goals), then it is unlikely to be incorporated into any decisions made (Wise et al., 2016). An example of information that was found to not be meaningful for students was the provision of login metrics in Park and Jo's (2015) LA dashboard, which was perceived as being unhelpful for the purposes of reflecting upon their learning. In other words, although resource use metrics continue to be used in LA service implementations (e.g., 75% of LA dashboards; Bodily & Verbert, 2017), their utility, from the perspective of students, can be questioned.
It has been shown that usefulness expectations are an important determinant in the future success of a technology (Brown et al., 2014). This is also true of LA services, where beliefs towards the utility of certain features (e.g., visualizations and the level of detail provided) affect adoption rates (Ali, Asadi, Gašević, Jovanović, & Hatala, 2013).
Together, this does reinforce the importance of gauging what stakeholders in a service want, with a focus on the type of information and its relevance to learning.
The challenge for LA to provide information that is pedagogically meaningful is not a recent concern (Gašević et al., 2015;Macfadyen & Dawson, 2010;Tsai & Gašević, 2017a). In particular, Gašević et al. (2015) warn against the use of trivial measures in LA service implementations on the basis that it will not promote effective learning. Taking what is known in relation to self-regulated learning theory, students do utilize various information that are fed back to understand how their learning is progressing towards set goals (Winne & Hadwin, 2012). Having simple performance metrics are unlikely to meet the necessary conditions to facilitate self-regulatory behaviour (Ali, Hatala, Gašević, & Jovanović, 2012;Gašević et al., 2015), which are to be constructive, promote higher order thinking, and allow students to bridge the gap between the current and desired level of performance (Nicol & Macfarlane-Dick, 2006). Therefore, for the information presented through LA services to become more informative, there is a need to both ground the approach within necessary educational frameworks and understand what information stakeholders need (Gašević et al., 2015). The meaningfulness expectations attempts to meet these recommendations by exploring what forms of information are expected from one of the main stakeholders.
With these four themes in mind, we generated 79 items capturing the various aspects of LA services identified in the literature (Appendix B). Each item was phrased as an expectation (e.g., the university will or the LA service will). Responses were made on both ideal (Ideally, I would like that happen) and predicted (In reality, I would expect that to happen) expectation Likert scale ranging from 1 (strongly disagree) to 7 (strongly agree), which were adapted from the work of Bowling et al. (2012). These preliminary items were subject to peer review by two experts in LA, both of whom are well known in the field of LA and co-founders of the Society for Learning Analytics Research. Items were then removed or reworded based on repetition, clarity, and relevance. As noted in Appendix B, the LA experts suggested the addition of one item "The feedback from analytics will be presented as a visualization (e.g., in the form of a dashboard)" (Item 37; Appendix C). This peer review process undertaken by LA experts led to 37 items being retained (Appendix C).
As students were unlikely to be aware of LA and what it entails, an introduction to the survey was created (Appendix A). The contents of this introduction outline to students the various sources of educational data used in LA services such as that extracted from the virtual learning environment. In addition, examples of possible LA service implementations are provided (e.g., the creation of early alert systems).
This information provided was peer reviewed by LA experts in order to assess whether the scope of LA services was suitable and whether the concept of LA services can be easily understood. Moreover, the information contained in this introduction was influenced by both the LA definition (Siemens & Gašević, 2012) and the commonly used data types in LA studies . Ethics approval was obtained for data collection at the University of Edinburgh and the University of Liverpool.

| Sample
A total of 210 student respondents (females = 131; M age = 24 years, SD = 6.12) out of a possible 448 students (47% response rate) from the University of Edinburgh completed the 37-item pilot survey (Appendix C), which was distributed through an online survey system. This was a self-selecting sample of students from across the university who have agreed to be contacted for research purposes in return for monetary reward on a task-by-task basis. This sample is broadly representative of the student population (undergraduate/postgraduate taught, U.K. vs. non-U.K., and age/gender).

| Statistical analysis
All raw data were analysed using R version 3.4 and the psych package (R Core Team, 2017;Revelle, 2017). The predicted and ideal expectation scales were analysed separately. If items were removed from one scale (e.g., the predicted expectation scale), the corresponding item was removed from the other scale (i.e., the ideal expectation scale).
The analysis steps were to first run Bartlett's (1951) test to assess whether a factor analysis was appropriate. The Kaiser-Meyer-Olkin (KMO) index (Kaiser, 1974) was then calculated to further check whether the data are adequate for a factor analysis. The determinant of the correlation matrix was also calculated to assess for any multicollinearity problems (Field, Miles, & Field, 2012). Following these scale purification steps, an exploratory factor analysis using oblimin rotation was ran on the raw data using the results of a parallel analysis to determine the sufficient number of factors to extract. Finally, a reliability analysis was run on the items of each factor.
Each item in the instrument also contained an open textbox to allow respondents to provide qualitative comments on each item.
Respondents were prompted to leave feedback about the clarity and understanding of each item. Thus, by obtaining both quantitative and qualitative data from the instrument, it allowed the researchers to refine items using the scale purification techniques and to reword certain items on the basis of student feedback.

| Ideal expectations scale
Eighteen items were dropped from the analysis based on the identification of multicollinearity issues, having loadings lower than 0.40, or whether dropping the item could improve the Cronbach's α value.
An exploratory factor analysis using the minimum residual factor extraction method and oblimin rotation was run on the remaining 19 items. The overall KMO was found to be 0.88 (great according to Kaiser, 1974), with individual item values being greater than or equal to 0.75, which was above the acceptable limit of 0.50. Bartlett's test of sphericity, χ 2 (190) = 1,613, p < .001, suggested that the correlation matrix did not resemble an identity matrix, so factor analysis was appropriate. The parallel analysis suggested to retain two or three factors; in order to align with the predicted expectations scale, a twofactor solution was selected. The two-factor solution was deemed sufficient; it accounted for 42% of the variance in the data, and the correlation between the two factors was r = .30. Factor 1 represented service feature expectations (items: 1,9,13,18,20,22,26,30,31,and 33;Appendix D), whereas Factor 2 relates to ethical and privacy expectations (items: 5,6,10,11,14,15,17,19,and 21;Appendix D). Both subscales had high reliabilities; for service feature expectations, Cronbach's α = .88, whereas for ethical and privacy expectations, Cronbach's α = .82.

| Predicted expectations scale
Eighteen items were dropped from the analysis based on the identification of multicollinearity issues, having loadings lower than 0.40, or whether dropping the item could improve the Cronbach's α value.
An exploratory factor analysis using the minimum residual factor extraction method and oblimin rotation was ran on the remaining 19 items. The overall KMO was found to be 0.91 (superb according to Kaiser, 1974), with individual item values being greater than or equal to 0.86, which was above the acceptable limit of 0.50. Bartlett's test of sphericity, χ 2 (171) = 1,631, p < .001, suggested that the correlation matrix did not resemble an identity matrix, so factor analysis was appropriate. The parallel analysis suggested to retain two factors.

| Discussion
The results of the pilot study led to the identification of a two-factor solution (ethical and privacy expectations and service feature expectations) that explain student expectations of LA services. For both the ideal and predicted expectation scales, the same items loaded onto the identified factors. This is important for future research directions as it will enable researchers to segment expectations across end-users.
In other words, desired and realistic beliefs regarding LA services may show differences based on demographic information (e.g., level of study).
Even though four expectation themes were identified from the literature, they are captured by this two-factor solution. The service feature expectation factor covers items relating to whether students believe they should responsibility to make sense of their own data (Item 18; Appendix C) and whether teaching staff are obliged to act when students are at risk or underperforming (Item 31; Appendix C).
Together, these items reflect the agency expectations theme identified in the literature. Items 26 and 33 (Appendix C) refer to beliefs about students receiving profiles of their learning following the analysis of their data and LA services being used to offer support directed at academic skill development. It is indicative from these items that there is overlap with the theme of intervention expectations. The theme of meaningfulness expectations is well captured by Item 20 (Appendix C), which is concerned with LA services connecting feedback to learning goals. The ethical and privacy expectations factor relates to the identified ethics and privacy expectations theme. As exemplified by Items 6 and 11 (Appendix C), these cover topics relating to the provision of consent for both universities utilizing personal information and prior to giving data to any third-party company, respectively.

| Sample
A total of 674 student respondents (females = 429; M age = 24.51 years, SD = 7.94) from the University of Edinburgh (n = 6,664; 10.11% response rate) completed the 19-item survey (Appendix F), which was distributed through an online system. N = 6,664 corresponds to one third of the whole university undergraduate and postgraduate taught student population based on a random selection; thus, students from the pilot could have also participated in Study 2. This sample was then checked against college, school, student type (i.e., students being from Scotland, the United Kingdom, a European (EU) country, or a non-European country), and other demographic information to ensure that the sample was representative of the university as a whole. All respondents consented to taking part in the online survey and were offered the chance to be included in a prize draw. Of these respondents, 396 (59%) were undergraduate students, 62 (9%) were masters students, and 216 were PhD students (32%). A total of 475 (70%) respondents identified themselves as "home/EU students," and 199 (30%) identified themselves as "overseas students."

| Questionnaire
Following the pilot study, the 37-item questionnaire was reduced to 19 items (Appendix F). The comments left by respondents in the pilot study were used to modify items in order to make them clearer (details of how item wordings were changed are presented in Appendix F).
The remaining 19 items (Appendix F) were also reviewed by an LA expert in order to identify any wording issues. As in the pilot study, each item contained two scales corresponding to ideal (Ideally, I would like that happen) and predicted (In reality, I would expect that to happen) expectations. Responses again were made on a 7-point Likerttype scale, ranging from 1 = strongly disagree to 7 = strongly agree.

| Statistical analysis
Qualitative comments from the pilot study were used in conjunction with a further peer review of the 19 items to clarify and rewrite particular items (Appendix F). An example of this was Item 1 from the 19-item questionnaire (The university will provide me with guidance on how to access the analysis of my educational data). Within the 37-item questionnaire, this item (Item 1) referred to whether the university is expected to instruct students on how frequently they should access educational data (The university will provide me with guidance on when and how often I should consult the analysis of my educational data). Feedback on this question showed that it would not be for an institution to decide how frequently educational data analyses should be consulted. A more appropriate alternative, which aligns with LA services being transparent (Sclater, 2016), would be an item on universities clearly telling students how to find any analyses of their educational data.
Similarly, for Item 2 of the 19-item questionnaire (The university will explain all the LA service processes as clearly as possible; e.g., how my educational data are collected, analysed, and used), this was a slight amendment of Item 5 from the 37-item questionnaire (The university will explain all analytic processes as clearly as possible; e.g., how my educational data are collected, analysed, and used).
Within the 37-item version, this item was not connected well with the overall aim of the questionnaire, which was to explore expectations of LA services, which go beyond analytics. Therefore, to make this a more inclusive item that refers to any possible processes involved, the item now refers to LA services in general.
Due to the various amendments to the questionnaire items, it was decided that exploratory factor analysis would again be used in a follow-up sample. This is because subtle changes in the item wordings could lead to different interpretations or model outcomes. What is more, the pilot study only had 210 respondents, which falls short of what has been recommended as a good sample size (300 according to Comrey & Lee, 1992). Therefore, for the main study, the recommended sample sizes proposed by Comrey and Lee (1992), which suggests at least 500 respondents should be used whenever possible.
Given the high number of low communalities (below 0.50) found with the pilot study exploratory factor analysis, it further reinforced the need to rerun the exploratory factor analysis with the data obtained from the larger sample of students in Study 2 (MacCallum, Widaman, Zhang, & Hong, 1999).
As with the pilot study, the same scale purification steps were undertaken here with an assessment of multicollinearity problems, item KMO inspection, and an assessment of whether factor analysis is appropriate using Bartlett's test of sphericity. Any item removed from one scale (ideal or predicted expectation) was removed from the corresponding scale. After these steps, an exploratory factor analysis using the minimum residual factor extraction method and oblimin rotation was ran on the raw data using the results of a parallel analysis to determine the sufficient number of factors to extract.
Finally, a reliability analysis was run on the items of each factor.

| Ideal expectations scale
Seven items (1, 2, 4, 9, 12, 14, and 15; Appendix F) were dropped from the analysis based on the identification of multicollinearity issues, having loadings lower than 0.40, or whether dropping the item could improve the Cronbach's α value.
An exploratory factor analysis using the minimum residual factor extraction method and oblimin rotation was ran on the remaining 12 items (3, 5, 6, 7, 8, 10, 11, 13, 16, 17, 18, and 19; Appendix F). The determinant of the correlation matrix exceeded 0.00001, so there was no issue with multicollinearity (Field et al., 2012). The overall KMO was found to be 0.90 (superb according to Kaiser, 1974), with individual item values being greater than or equal to 0.86, which was above the acceptable limit of 0.50. Bartlett's test of sphericity, χ 2 (66) = 4,093, p < .001, suggested that the correlation matrix does not resemble an identity matrix, so factor analysis was appropriate.
The parallel analysis suggested to retain two factors. The two-factor solution was deemed sufficient; it accounted for 56% of the variance in the data, the correlation between factors was r = .37, all loadings exceeded 0.40 (Table 1), and communalities were in an acceptable range (Table 1). Factor 1 represents service feature expectations (items: 7, 11, 13, 16, 17, 18, and 19; Appendix F), whereas Factor 2 relates to ethical and privacy expectations (items: 3, 5, 6, 8, and 10; Appendix F). Both subscales had high reliabilities; for service feature expectations, the Cronbach's α = .90, whereas for ethical and privacy expectations, Cronbach's α = .85.

| Predicted expectations scale
Seven items (1, 2, 4, 9, 12, 14, and 15; Appendix F) were dropped from the analysis based on the identification of multicollinearity issues, having loadings lower than 0.40, or whether dropping the item could improve the Cronbach's α value.
An exploratory factor analysis using the minimum residual factor extraction method and oblimin rotation was ran on the remaining 12 items (3, 5, 6, 7, 8, 10, 11, 13, 16, 17, 18, and 19; Appendix F). The overall KMO was found to be 0.93 (superb according to Kaiser, 1974), with individual item values being greater than or equal to 0.89, which was above the acceptable limit of 0.50. Bartlett's test of sphericity, χ 2 (66) = 4,476, p < .001, suggested that the correlation matrix does not resemble an identity matrix, so the factor analysis was appropriate. The parallel analysis suggested to retain two factors.

| Descriptive statistics
The descriptive statistics of the final 12 items are presented in Table 3.
Across each item, it is clear that average responses for ideal expectations are higher than predicted expectations. Within each expectation type (ideal and predicted), the items relating to the ethical and privacy expectations factors (E1-E5) were higher than service feature expectations (S1-S7). For the ideal expectations scale, the mean responses for the ethical and privacy expectations factor ranged from 6.12 to 6.58, whereas for the service feature expectations, the range was between 5.56 and 5.74. Whereas for the predicted expectations scale, the average responses for the ethical and privacy expectations factor

| Discussion
The results of the factor analysis again identified a two-factor solution The descriptive statistics provide a general insight into student expectations of LA services (  are supportive of two levels of belief. In addition to providing descriptive statistics for each item, the mean and standard deviations for each item by gender (Table 4) and level of study (Table 5) are also provided.
Comparing the ethical and privacy expectations and service feature expectations factor responses on both the ideal and predicted scales does suggest that beliefs towards the ethical procedures involved in LA service implementations are of greater importance. This is based on the range of average responses across ideal and predicted expectation scales being greater for ethical and privacy expectation items than service feature expectation items (Table 3). A tentative conclusion that can be drawn from this is that students do hold stronger beliefs about ethical procedures involved in LA service implementations. Thus, in line with the findings of Slade and Prinsloo (2014), it appears that students do place considerable importance on how a university handles their educational data, particularly with regard to controlling who access to any data and whether consent is required. Although in the case of service feature expectations, students may desire such features (e.g., being able to compare current progress to learning goals), but the importance of such services are not comparable with the ethical procedures of LA services.
For the ethical and privacy expectations factor, the item with both the highest mean response across ideal (M = 6.58, SD = 0.86; Table 3) and predicted (M = 6.05, SD = 1.28; Table 3) expectations was Item 5 (The university will ensure that all my educational data will be kept securely; Appendix F). Slade and Prinsloo (2014)   Abbreviations: E1-E5, ethical and privacy expectation items; S1-S7, service feature expectation items.

| Sample
The 12-item SELAQ (Appendix G) was distributed to students at the University of Liverpool through an online survey system. The 12 items were identified as per the results of the exploratory factor analysis in Study 2. Some 191 responses were collected (females = 129). Students were aged between 18 and 50 (M = 20.41, SD = 3). The majority of students were undergraduates (n = 188, 98%), whereas the remaining sample was composed of masters students (n = 3, 0.02%). Of the sample, 19% were taking a science subject (n = 36), 13% were studying engineering (n = 24), 21% were studying a social science subject (n = 41), 24% were taking an arts and humanities subject (n = 45), and 24% were studying a medicine and health care subject (n = 45).
Eighty per cent (n = 153) of the sample was home/EU students, with the remaining being international students (20%, n = 38).

| Instrument
The 12-item SELAQ was used for this study (Appendix G). Responses to the items are made on two 7-point Likert scales (1 = strongly disagree; 7 = strongly agree) corresponding to ideal (Ideally, I would like that to happen) and predicted (In reality, I would expect that to happen) expectations. As with the survey distributions for the pilot study and Study 2, respondents were given the same introduction to the survey (Appendix A).

| Data analysis
Exploratory structural equation modelling using geomin rotation and confirmatory factor analysis was carried out on the raw data using Mplus 8 (Muthén & Muthén, 2017) in order to test the suitability of the two-factor solution (ethical and privacy expectations and service feature expectations). It is important to note that the exploratory structural equation modelling was used as a confirmatory tool (Marsh, Morin, Parker, & Kaur, 2014). As recommended by Marsh et al. (2014), the model fit indices obtained from both confirmatory factor analysis and exploratory structural equation modelling will be compared. If the fit indices from both models are marginally different, then the confirmatory factor analysis model will be discussed on the basis of parsimony (Marsh et al., 2014). Table 6 presents the descriptive statistics for the 12 items of the SELAQ, along with the factor key that shows the items to either correspond to the ethical and privacy expectation factor (E1-E5) or the service feature expectation factor (S1-S7). The ideal expectations scale responses were negatively skewed (Table 6). This ceiling effect was anticipated as the ideal expectation scale corresponds to what an individual hopes for, so individuals are likely to respond positively.
The predicted expectation scale also showed negatively skewed responses (Table 6). Due to the responses being categorical and skewed, along with the small sample size (n = 191), the scale-shifted approach to the unweighted least squares estimation (ULSMV) was used (Muthén, Muthén, & Asparouhov, 2015).
To assess the suitability of the two-factor model for both scales, the χ 2 test is presented along with the following alternative fit indexes: the comparative fit index (CFI), Tucker-Lewis index (TLI), and root mean square error of approximation (RMSEA), with 90% confidence intervals. In terms of cut-offs, an RMSEA value within the range of 0.08 and 0.10 is indicative of a mediocre fit (MacCallum, Browne, & Sugawara, 1996), whereas values close to or below 0.06 would support a good fit (Hu & Bentler, 1999). As for both TLI and CFI, Hu and Bentler (1999) recommend values close to or above 0.95. These proposed cut-offs, however, were based on continuous data being analysed with the maximum likelihood estimator. In the case of ULSMV, Xia (2016) found that the cut-offs proposed by Hu and Bentler (1999) to not be applicable as they are influenced by thresh-  (Raykov, 1997).
With regard to the χ 2 test of exact fit, Ropovik (2015) does note that it is unrealistic for many applications, but it should not be universally dismissed. If the χ 2 test is found to be significant, this may then point to possible model misspecifications, which can be examined through an assessment of local fit (Ropovik, 2015). Of the various approaches to assessing local fit, the current study will explore modi-   (Kline, 2015). It is important to remain mindful that engaging in datadriven model modifications could be entirely based on chance (MacCallum, Roznowski, & Necowitz, 1992). To address the issue of capitalizing on chance, MacCallum et al. (1992) recommend that any modifications to a model be cross-validated in a second sample. Given that the current sample is small (n = 191), the splitting of the sample for the purposes of model cross-validation is not advisable. Therefore, if problems in the model are identified, we recommend that future research is conducted in order to assess whether these issues are found in independent samples and also whether any modifications can be cross-validated.

| Ideal expectation scale
The purported two-factor model led to an acceptable fitting model using the confirmatory factor analysis approach, χ 2 (53,

| Predicted expectation scale
Compared with the ideal expectation scale, the two-factor model was found to have an acceptable fit using the confirmatory factor analysis    Table 6 presents descriptive statistics for each item across both expectation scales (ideal and predicted); item means and standard deviations are also presented by gender (Table 9) and level of study (Table 10). As with Study 2, the average responses are higher on the ideal than the predicted expectation scale. In general, the mean values on the ethical and privacy expectation items are higher (ranging from 5.77 to 6.53 for ideal expectations and ranging from 5.19 to 6.27 for predicted expectations; Table 6) than those relating to service feature expectation items (ranging from 5.80 to 6.03 for ideal expectations and ranging from 4.96 to 5.35 for predicted expectations; Table 6). This was not the case for

| Descriptive statistics
Item 5 (The university will ask for my consent to collect, use, and analyse any of my educational data; e.g., grades, attendance, and virtual  Abbreviations: E1-E5, ethical and privacy expectation items; S1-S7, service feature expectation items. learning environment accesses) from the ethical and privacy expectation factor, which appeared to not elicit a strong response from students for either ideal (M = 5.77, SD = 1.33; Table 6) or predicted (M = 5.19, SD = 1.62; Table 6) expectations. As with Study 2, the ethical and privacy expectation item with the highest average response for both ideal (M = 6.53, SD = 0.78; Table 6) and predicted (M = 6.27, SD = 1.08; Table 6) expectations was Item 2 (The university will ensure that all my educational data will be kept securely).
As for the service feature expectation items, the highest average response on the ideal expectation scale is for Item 11 (The teaching staff will have an obligation to act, i.e., support me, if the analytics show that I am at risk of failing and underperforming or if I could improve my learning; M = 6.04, SD = 1.31; Table 6). Although for the predicted expectation scale, Item 12 (The feedback from the LA service will be used to promote academic and professional skill development, e.g., essay writing and referencing, for my future employability) received the highest average response (M = 5.35, SD = 1.43; Table 6).

| Discussion
Based on the findings of Study 2, a purported two-factor structure was found to explain student expectations of LA services on both the ideal and predicted expectation scales. In Study 3, the appropriateness of this two-factor structure was assessed through both confirmatory factor analysis and exploratory structural equation modelling. A decision was made to use the confirmatory factor analysis for the basis of further model discussions as the differences in alternative fit indices were marginal and the confirmatory factor analysis model was more parsimonious (Marsh et al., 2014). Put differently, although the RMSEA may not be in line with the cutoff proposed by Hu and Bentler (1999; i.e., RMSEA values close to or below 0.06), its function varies in accordance with measurement quality (McNeish et al., 2018). In addition, these recommended cutoff values are based on continuous data analysed using the maximum likelihood estimator; thus, their applicability to ordinal data analysed using ULSMV can be questioned (Xia, 2016).
Although the measurement quality of both scales (ideal and predicted expectations) was good and the alternative fit indices show the fit to be acceptable, the χ 2 test was found to be significant (p < .05). Following the recommendations set out by Ropovik (2015), the local fit of the model was assessed by examining both MI and SEPC values, along with correlation residuals. This assessment did lead to the identification of possible localized strains within the model, with misfits being found between Item 2 and Item 5 and between Item 11 and Item 12 on both scales (ideal and predicted expectations). For Items 2 and 5, their content relates to the university ensuring all data are kept securely and obtaining consent before engaging in any analysis of data, respectively.
Based on the content of these two items, there is some degree of overlap, as the students consenting to allow the university to collect and analyse collected data will be tied to their beliefs regarding data security. However, this does not provide substantial justification for a respecification of the model that allows the errors between Items 2 and 5 to correlate. As for Items 11 and 12, the content is focused upon beliefs towards the implementation of early intervention systems (Item Abbreviations: E1-E5, ethical and privacy expectation items; S1-S7, service feature expectation items. 11) and using LA services to develop academic/employability skills (Item 12). Thus, from a content perspective, there is no overlap, which again means that the respecification of the model by allowing the errors of Items 11 and 12 cannot be justified.
For the ideal expectation scale, there was a further source of misfit between Items 1 and 2. These items refer to beliefs about the provision of consent towards the collection of identifiable data and ensuring all collected data remain secure, respectively. Whereas for the predicted expectation scale, there was an additional source of misfit between Items 2 and 3. These correspond to beliefs about data security and providing consent before data are outsourced to third-party companies, respectively. Taking both sources of misfit (between Items 1 and 2 for the ideal expectation scale and Items 2 and 3 for the predicted expectation scale) into consideration, it is clear that although they all relate to data security procedures, there is no substantial justification for allowing these errors between these items to correlate.
Even though an assessment of local strains within the model did These aforementioned points, however, do not apply to Item 5 (The university will ask for my consent to collect, use, and analyse any of my educational data; e.g., grades, attendance, and virtual learning environment accesses), which is the lowest ethical and privacy expectation item on both scales (ideal and predicted). The highest average response on the ethical and privacy expectation subscale for Study 3, as found with Study 2, was for Item 2 (The university will ensure that all my educational data will be kept securely) for both ideal and predicted expectations. Thus, student beliefs towards the provision of consent before the university collect educational data may not be as strong as their expectations towards any data collected remaining secure. This resonates with what Roberts et al. (2016) identified as a pertinent concern raised by students, which was the university ensuring that all data remain private. Similarly, Prinsloo and Slade (2016)

| Interpretation of the results
Following a review of the LA literature (Whitelock-Wainwright et al., 2017) and input from experts, four themes were identified: ethical and privacy expectations, agency expectations, intervention expectations, and meaningfulness expectations. These themes were used to guide the generation of items relating to student expectations of LA services. What is more, we grounded these items within the theoretical framework of expectations, drawing mainly from the work achieved in the technology acceptance literature (Brown et al., 2012;Brown et al., 2014;Davis & Venkatesh, 2004) and health service literature (Bowling et al., 2012;Thompson & Suñol, 1995) that has demonstrated the importance of gauging stakeholder expectations. From this, two levels of expectations (ideal and predicted) were identified (David et al., 2004;Dowling & Rickwood, 2016), which are shown to provide a more nuanced understanding of stakeholder beliefs.
Using the above as a framework, we have been able to develop and validate a descriptive12-item (Appendix G) instrument that allows researchers, practitioners, and institutions to obtain a general understanding of students' ideal and predicted expectations towards LA services. The results also show that these 12 expectations can be explained by two first-order factors: ethical and privacy expectations and service feature expectations. The view is that the measurements obtained can then direct more specific engagements with students at different intervals throughout the implementation process, with a view of managing expectations and identifying main areas of focus for the LA service.
The ethical and privacy expectations factor (Items 1, 2, 3, 5, and 6; Appendix G) strongly relates to the identified theme ethical and privacy expectations. Items 1, 3, 5, and 6 refer to expectations towards the provision of consent for universities to use identifiable data (e.g., ethnicity, age, and gender), to outsource data to third-party companies, to collect and use any educational data (e.g., grades, virtual learning environment accesses, or attendance), and to use data for purposes beyond what was originally stated, respectively. Item 2, however, refers to the belief that universities should keep data secure.
These items are well supported by the LA literature, particularly in the work carried out by Slade and Prinsloo (2014) who found students expected universities to require informed consent and to maintain privacy at all times. They also add weight to the work of Ifenthaler and Schumacher (2016), as these items are centred on beliefs towards the control students have over their data.
Expectation items relating to opting out (Item 9; Appendix F) and transparency (Item 2; Appendix F) were not retained in the final 12item instrument. The omission of an opt-out item may be based upon students holding stronger expectations towards their right to decide whether an institution uses their educational data from the outset.
In order to make such a decision, the institution would also have to provide details on their proposed uses of such data. The act of obtaining informed consent can then also be thought of as intrinsically covering the responsibility of being transparent (Sclater, 2016).
With informed consent items being retained for identifiable and educational data usage, it does identify a gap with the opinions offered by experts (Sclater, 2016) who believe consent should only be sought for interventions to offset any likelihood of burdening students with documents. This is an example of an ideological gap, as we have shown that the ethical expectations held by students are concerned with having the right to consent to any processes involved in an LA service. Our findings do not advocate institutions undertaking an approach that overloads the student population with requests for consent; rather, students should be directly involved in policy developments to offset any risks services that are not reflective of student expectations.
In addition, an inspection of the descriptive statistics obtained from Studies 2 and 3 does provide further details regarding students' ethical and privacy expectations. For both samples, it was found that the highest average response across each scale (ideal and predicted) was for the expectation towards the university ensuring all collected data are kept secure (Item 2; Appendix G). Thus, these students expect the university to be responsible for upholding the security of any data collected (Prinsloo & Slade, 2016), which may emanate from concerns about who has access to their data (Roberts et al., 2016). From a policy perspective, these findings together suggest that a university must provide easily accessible information regarding data-handling processes. More specifically, students should be informed as to how the university will securely hold all collected data and prevent disclosure of such information to unauthorized third parties.
The service feature expectations factor (Items 4,7,8,9,10,11,and 12; Appendix G) does overlap with the identified themes of agency, intervention, and meaningfulness expectations. As stipulated in the introduction, these themes were not assumed to be orthogonal from one another; rather, they were presented as a means of collating the various research streams and discussions in LA. Item 8 (Appendix G) refers to the expectation that the LA service should be aimed at updating students on how their progress compares to goals set, which is an example of the meaningfulness expectations theme. Items 7 and 11 (Appendix G) are concerned with students expecting to make their own decisions based on the feedback from LA services and whether teaching staff are obligated to act if students are underperforming or at risk, respectively. Together, these two beliefs address the agency expectations theme. Finally, Items 4, 9, 10, and 12 (Appendix G) correspond to students expecting regular updates on their learning progress, a complete profile of the learning, teaching staff using LA in their feedback, and LA services being designed to improve skill development, respectively. These beliefs all refer to what students expect to receive from LA services, which relates to the intervention expectations theme.
As stated, the meaningfulness expectations theme is captured by Item 8 (Appendix G). This refers to the expectation towards receiving feedback that shows how a student's learning is progressing in relation to a set goal, which has been expressed by students in the work of Schumacher and Ifenthaler (2018). Likewise, Roberts et al. (2017) found students expected LA service features to convey information that is meaningful (e.g., learning opportunities). A possible reason for students expecting LA services to display information such as progress towards a goal does relate to self-regulated learning. As Winne and Hadwin (2012) state, being able to identify discrepancies between performance and goals set enables learners to regulate their own learning (e.g., adopt an alternative learning strategy), whereas feeding information back to students that is not pedagogically meaningful (e.g., number of access times to a virtual learning environment) is unlikely to motivate positive changes in learner behaviour (Gašević et al., 2015;Wise et al., 2016). Thus, although a university may view the provision of more feedback to students as being advantageous, it may not necessarily reflect what students want, which is feedback that is pedagogically meaningful.
The results of the studies presented in the paper closely align with the discussions related to the moral considerations of whether teaching staff are obligated to act (Prinsloo & Slade, 2017). According to Prinsloo and Slade, although institutions should take action when problems are identified, the student still shares a responsibility for their own learning. This acknowledges the fact that students are active agents who metacognitively monitor their progress towards a set goal (Gašević et al., 2015;Winne & Hadwin, 2012), and it is not for LA services to create a culture of passivity (Kruse & Pongsajapan, 2012). These concerns have been voiced by students in the work of Roberts et al. (2016). More specifically, students expressed apprehension towards LA services that would remove the ability to engage in self-directed learning (Roberts et al., 2016). This again illustrates the importance of gauging student expectations towards elements of the LA service. Although institutions may view LA favourably on the basis of instructors being able to provide timely support to students (Pardo & Siemens, 2014), students may consider such systems as a hindrance to independent learning (Roberts et al., 2016). The items of the SELAQ capture this balance between students making their own decisions on the basis of the LA feedback (Item 7; Appendix G) and institutions being obligated to act (Item 11; Appendix G), which together reflect the theme of agency expectations.
The intervention expectations theme centres on the beliefs students hold regarding the LA service they receive in exchange for the disclosure of data. Although there have been advances in introducing new forms of feedback (Verbert et al., 2013), developing ways of improving the student-teacher relationship (Liu et al., 2017), and offering ways to improve retention (Campbell et al., 2007), little has been done to ask what students expect institutions to do with their collected data (Arnold & Sclater, 2017;Schumacher & Ifenthaler, 2018). Put differently, there have been few instances of students being engaged within the development and implementation of LA service features. Of those instances where students have been engaged, it has been found that students want profiles of their learning, updates on their learning progress, and features designed to promote academic skill development (Roberts et al., 2017;Schumacher & Ifenthaler, 2018). These expectations are captured by the retained items of the SELAQ (Items 4, 9, and 12; Appendix G), in addition to an expectation pertaining to teaching staff incorporating LA into their own feedback (Item 10; Appendix G). Together, these items both represent the intervention expectations theme and provide an indication of the LA service features students expect.
From the descriptive statistics obtained in Studies 2 and 3 that refer to the service feature expectations factor, a general understanding of the LA service features students expect does emerge. Moreover, focusing on those items with the highest average responses may be indicative of student expectations of LA services not being homogenous. In Study 2, the highest average response for the desired expectation scale was for teaching staff to incorporate LA into their feedback (Item 10; Appendix G). Although on the predicted expectation scale, the highest average response was for feedback showing how their progress compares to a set goal (Item 8; Appendix G). For these students, although they desire the possibility of teaching staff being able to offer more informative feedback, they expect to receive feedback showing how their learning progresses to a set goal. For Study 3, on the other hand, the highest average response on the ideal expectation scale was for the university having an obligation to act (Item 11; Appendix G), whereas, on the predicted expectation scale, the highest average response was for the use of LA to promote academic or professional skill development (Item 12; Appendix G). Compared with the students in Study 2, those in Study 3 desire the inclusion of early alert systems but realistically expect LA services to be tailored towards promoting academic skill development.
These aforementioned comparisons using items from the service feature expectations factor show that although certain LA service features may be desirable (e.g., the introduction of early alert systems), it may not be the LA service features students expect (e.g., LA services designed to support academic skills such as self-regulated learning).
Thus, although there has been extensive attention paid to the possibility of LA services identifying underperforming or at-risk students (Campbell et al., 2007), students may actually be expecting LA service features aimed at providing them with a way of understanding or improving their learning processes. These beliefs have also been expressed by teaching staff, who viewed LA service features that provide students with insights into their learning more favourably than simple performance metrics (Ali et al., 2012;Gašević et al., 2015).
Taken together, it shows that although the provision of certain LA service features (e.g., early alert systems) may seem advantageous to a higher education institution, it remains necessary to explore what students expect from LA services (Ferguson, 2012).

| Limitations and future research
For the purposes of this work, the scale reduction was based solely upon statistical decisions (e.g., weak factor loadings) set out before analysing the data. Additionally, we wanted the descriptive questionnaire to measure items across two scales (ideal and predicted expectations), which may have accounted for a greater loss in item numbers.
Nevertheless, although adherence to statistical decisions were followed here, item content can also be considered (Flora & Flake, 2017). Future work may then be undertaken to determine whether additional items should be included. It is important to recognize, however, that this descriptive questionnaire only seeks to provide higher education institutions with a general understanding of what students expect of LA services. The anticipated effect of being able to readily gauge such expectations is to open dialogues with students at all stages of LA service implementations. In doing so, the higher education institution can begin to manage expectations and proactively identify main areas to focus upon, which can then utilize more specific instruments and/or a qualitative approach.
On the basis of alternative fit indices, the purported two-factor structure resulted in an acceptable fit for both scales (ideal and predicted expectations). Moreover, an assessment of measurement quality showed the average standardized loadings and reliability to be high. Nevertheless, for both scales, the χ 2 test has found to be significant, which should not be ignored (Ropovik, 2015). Based on the recommendations of Ropovik (2015), an assessment of local misfit was therefore undertaken (i.e., examination of MI and SEPC values, along with an inspection of residual correlations). From this assessment of local fit, local sources of strain were identified in the model, but possible respecifications of the model were not justified on conceptual grounds. In addition, the sample size (n = 191) did not allow for the cross-validation of any model modification (MacCallum et al., 1992). It is important for future researchers to be aware of the local sources of strain identified in Study 3, assess whether these are found using larger samples, and explore whether model improvements can be made.
Even though engaging students in the development of LA services is a critical factor to success (Ferguson et al., 2014), the expectations of teaching staff cannot be ignored. As Ali et al. (2012) show, teaching staff hold beliefs about the type of service they want from LA, particularly with regard to utility of the information that is fed back. Thus, although the needs of students should continue to guide the development of LA services, the expectations teaching staff must also be considered. Future research should therefore seek to develop and validate an instrument designed to explore the beliefs of teaching staff towards LA services. Then together with the SELAQ, institutions can provide and accommodate a greater number of stakeholder perspectives into the implementation of LA services.
An additional consideration that needs to be made is the cultural limitation of the SELAQ, as it has only been developed and validated with U.K. higher education students. It is therefore necessary for researchers to validate this instrument in other contexts. The challenge of unequal stakeholder engagement in LA implementations is not limited to U.K. higher education institutions (Tsai & Gašević, 2017a), and it is necessary for each university that is interested in implementing LA services to actively engage with their stakeholders.
The SELAQ provides a solution to these challenges, but further work is required to assess the reliability and validity of the instrument in cross-cultural contexts including the validation of the instrumentation translated into other languages. Work has been undertaken by the current authors to adapt the SELAQ for use in Dutch, Estonian, and Spanish higher education institutions.
To extend the current work, researchers who use the SELAQ should focus on segmenting students based on their expectations, as it is unlikely that they will hold homogenous beliefs about LA services. It is anticipated that certain groups of students (e.g., undergraduate students) may have higher expectations of the types of feedback they want to receive in comparison with others (e.g., PhD students). Thus, the SELAQ can provide institutions with a means of exploring and understanding the individual differences in student beliefs towards LA services.

| Implications
Research exploring student beliefs towards LA services have provided insightful findings that reinforce the importance of understanding a key stakeholder perspective (Roberts et al., 2016;Roberts et al., 2017;Slade & Prinsloo, 2014). Although these studies have predominately undertaken a qualitative approach to understand student beliefs towards LA services, the SELAQ provides researchers with a tool that enables them to quantitatively measure LA service expectations. The instrument can be used on its own as a way of gauging what large samples of student expect from LA services. The SELAQ can further be combined with scales measuring attitudes, goal orientations, or intentions to use. This can provide a way of understanding how expectations towards LA services form (e.g., based on individual differences in goal orientations) and whether these beliefs are associated with their behaviours or attitude towards the service (e.g., whether students feel positively or negatively about the implemented LA service or whether they intend to use the service). The SELAQ can also be incorporated into mixed methods approaches as it can be used to understand whether the LA service expectations expressed in interviews are reflective of the beliefs in the general student population.
The results of the SELAQ can be used to identify key areas of an LA service that need to be met based on the level of predicted expectations. As this is the level of service that is expected from a student, therefore, it is essential for the institute to meet these expectations effectively, or dissatisfaction is likely to arise (Whitelock-Wainwright et al., 2017). Knowing the importance of ethical issues to students, the university can also create LA service policies that address each of the items contained within the SELAQ. What is more, the results of the SELAQ can be accommodated into interviews with students in order to better understand why certain LA service features elicit higher expectations than others.

| CONCLUSION
Meeting stakeholder expectations is an important determinant in the eventual acceptance of an implemented service (Brown et al., 2012;Brown et al., 2014). Ways to accommodate these expectations into the design and implementation of services are therefore imperative; approaches include, but are not limited to, focus groups and surveys.
In this paper, the authors have discussed how the incorporation of student expectations into the implementation of LA services has been limited, which may increase the risk of future dissatisfaction due to the service not aligning with beliefs. This work builds upon past research that has discussed student expectations as falling into ones referring to ethics and privacy and those associated with service features (Roberts et al., 2016;Schumacher & Ifenthaler, 2018

A.1 | Student expectations of learning analytics
In the forthcoming years, learning analytics will be increasingly prevalent in higher education. Learning analytics involves the collection of educational data, such as grades, lecture attendance, or number of accesses to online resources from various learning environments to better inform how students learn and engage in their studies. The educational data are used to implement support services that are used to aid student learning such as the development of early alert systems for those who may be at risk of failing a course or dropping out, personalized learning environments, and improving student feedback processes. For example, the collection of a student's online learning environment data (e.g., hours spent online) can be used by a learning analytics service to determine whether a student is above or below the average level of engagement for the course/module. If the service detects that the student is below the average level of engagement required for a course, it may alert their personal tutor for providing relevant feedback and support. The learning analytics service provides timely information so that the tutor can contact the student to identify any problems and provide support before these problems jeopardize the student's learning.
As students will be the main beneficiaries from learning analytics, it is important for their opinions and expectations are accommodated into the design and implementation of any developed services. You have been asked to participate in this survey to investigate your expectations towards a learning analytics service and the use of your educational data by the university. These expectation questions have been formatted to understand what you desire from a learning analytics service (e.g., what you ideally hope for) and what is the minimum standard that you expect from the service (e.g., what you expect to happen in reality). By completing this survey, you will be providing critical information on student expectations regarding learning analytics. The findings from the survey will inform how future services are developed to ensure they reflect, and meet, yours and your peers' expectations and needs.
The results of this survey will be used to inform the development of the learning analytics policy at the (university name).

19-ITEM STUDENT EXPECTATIONS OF LEARNING ANALYTICS SERVICES QUESTIONNAIRE USED IN STUDY 2
Changes to item wordings of the 37-item instrument used in Study 1 based on feedback from students and learning analytics experts: Item 1 -In the 37-item instrument, this item was "The University will provide me with guidance on when and how often I should consult the analysis of my educational data"; this was changed to "The University will provide me with guidance on how to access the analysis of my educational data." Item 2 -In the 37-item instrument, this item was "The University will explain all analytic processes as clearly as possible (e.g., how my educational data are collected, analysed, and used)"; this was changed to "The University will explain all the learning analytics Item Retained?
Reason for removal 1. The university will provide me with guidance on how to access the analysis of my educational data No Did not load onto a factor 2. The university will explain all the learning analytics service processes as clearly as possible (e.g., how my educational data are collected, analysed, and used)

No
Did not load onto a factor 3. The university will ask for my consent before using any identifiable data about myself (e.g., ethnicity, age, and gender) Yes 4. The university will provide support (e.g., advice from personal tutors) as soon as possible if the analysis of my educational data suggests I may be having some difficulty or problem (e.g., I am underperforming or at risk of failing) No Item cross-loads 5. The university will ensure that all my educational data will be kept securely Yes 6. The university will ask for my consent before my educational data are outsourced for analysis by third-party companies Yes 7. The university will regularly update me about my learning progress based on the analysis of my educational data Yes 8. The university will ask for my consent to collect, use, and analyse any of my educational data (e.g., grades, attendance, and virtual learning environment accesses) Yes 9. The university will give me the right to opt out of data collection and analysis even if the action reduces the opportunities to provide me with personal support

No
Did not load onto a factor 10. The university will request further consent if my educational data are being used for a purpose different to what was originally stated Yes 11. The learning analytics service will be used to promote student decision making (e.g., encouraging you to adjust your set learning goals based upon the feedback provided to you and draw your own conclusions from the outputs received) Yes 12. The learning analytics service will collect and present data that are accurate (i.e., free from inaccuracies such as incorrect grades)

No
Did not load onto a factor 13. The learning analytics service will show how my learning progress compares to my learning goals/the course objectives Yes 14. The feedback from the learning analytics service will be presented in a format that is both understandable and easy to read No Did not load onto a factor 15. The feedback from the learning analytics service will be used to improve the educational experience in a module/course/programme (e.g., identifying problems in the feedback, assessments, and learning activities) No Item cross-loads 16. The learning analytics service will present me with a complete profile of my learning across every module (e.g., number of accesses to online material and attendance) Yes 17. The teaching staff will be competent in incorporating analytics into the feedback and support they provide to me Yes 18. The teaching staff will have an obligation to act (i.e., support me) if the analytics show that I am at risk of failing and underperforming or if I could improve my learning Yes 19. The feedback from the learning analytics service will be used to promote academic and professional skill development (e.g., essay writing and referencing) for my future employability Yes service processes as clearly as possible (e.g., how my educational data are collected, analysed, and used)." Item 3 -In the 37-item instrument, this item was "The University will ask for my consent before using any sensitive data about myself (e.g., ethnicity and religion)"; this was changed to "The University will ask for my consent before using any identifiable data about myself (e.g., ethnicity, age, and gender)." Item 4 -In the 37-item instrument, this item was "The University will provide real-time support (e.g., advice from tutors) based on the analyses of my educational data"; this was changed to "The university will provide support (e.g., advice from personal tutors) as soon as possible if the analysis of my educational data suggests I may be having some difficulty or problem (e.g., I am underperforming or at risk of failing)." Item 5 -In the 37-item instrument, this item was "The University will reassure me that all my educational data will be kept securely and used properly"; this was changed to "The University will ensure that all my educational data will be kept securely." Item 6 -No changes made to item wording.
Item 7 -In the 37-item instrument, this item was "The University will regularly contact me about my learning progress based on the analysis of my educational data"; this was changed to "The University will regularly update me about my learning progress based on the analysis of my educational data." Item 8 -In the 37-item instrument, this item was "The University will ask for my explicit consent for the collection, use, and analysis of any of my educational data (e.g., grades, attendance, and virtual learning environment accesses)"; this was changed to "The University will ask for my consent to collect, use, and analyse any of my educational data (e.g., grades, attendance, and virtual learning environment accesses)." Item 9 -In the 37-item instrument, this item was "The University will give me the right to opt out of data collection and analysis"; this was changed to "The University will give me the right to opt out of data collection and analysis even if the action reduces the opportunities to provide me with personal support." Item 10 -In the 37-item instrument, this item was "The University will ask for my consent again if my educational data are being used for a different purpose than originally stated"; this was changed to "The University will request further consent if my educational data are being used for a purpose different to what was originally stated." Item 11 -In the 37-item version of the instrument, this item was "The analytics will promote student decision making (e.g., encouraging you to adjust your set learning goals based upon the feedback provided to you and draw your own conclusions from the outputs received)"; this was changed to "The learning analytics service will be used to promote student decision making (e.g., encouraging you to adjust your set learning goals based upon the feedback provided to you and draw your own conclusions from the outputs received)." Item 12 -In the 37-item instrument, this item was "The analytics will collect and present data that are accurate (i.e., free from inaccuracies, such as incorrect grades)"; this was changed to "The learning analytics service will collect and present data that are accurate (i.e., free from inaccuracies such as incorrect grades)." Item 13 -In the 37-item instrument, this item was "The analytics will clearly link my data to my progression towards my learning goals"; this was changed to "The learning analytics service will show how my learning progress compares to my learning goals/the course objectives." Item 14 -In the 37-item instrument, this item was "The analytics will be presented in a format that is both understandable and easy to read"; this was changed to "The feedback from the learning analytics service will be presented in a format that is both understandable and easy to read." Item 15 -In the 37-item instrument, this item was "The analytics will be used to improve the educational experience in a module/course/programme (e.g., identifying problems in the feedback, assessments, and learning activities)"; this was changed to "The feedback from the learning analytics service will be used to improve the educational experience in a module/course/programme (e.g., identifying problems in the feedback, assessments, and learning activities)." Item 16 -In the 37-item instrument, this item was "The analytics will present me with a complete profile of my learning across every module (e.g., number of accesses to online material and attendance)"; this was changed to "The learning analytics service will present me with a complete profile of my learning across every module (e.g., number of accesses to online material and attendance)." Item 17 -In the 37-item instrument, this item was "The teaching staff will be competent in incorporating analytics in the feedback and support they provide to me"; this was changed to "The teaching staff will be competent in incorporating analytics into the feedback and support they provide to me." Item 18 -No changes to item wording.
Item 19 -In the 37-item instrument, this item was "The feedback from analytics will be used to promote skill development (e.g., essay writing and referencing)"; this was changed to "The feedback from the learning analytics service will be used to promote academic and professional skill development (e.g., essay writing and referencing) for my future employability."

12-ITEM STUDENT EXPECTATIONS OF LEARNING ANALYTICS SERVICES QUESTIONNAIRE USED IN STUDY 3 APPENDIX H EXPLORATORY STRUCTURAL EQUATION MODEL FACTOR LOADINGS FOR THE IDEAL EXPECTATION SCALE
Factor key Item E1 1. The university will ask for my consent before using any identifiable data about myself (e.g., ethnicity, age, and gender) E2 2. The university will ensure that all my educational data will be kept securely E3 3. The university will ask for my consent before my educational data are outsourced for analysis by third-party companies S1 4. The university will regularly update me about my learning progress based on the analysis of my educational data E4 5. The university will ask for my consent to collect, use, and analyse any of my educational data (e.g., grades, attendance, and virtual learning environment accesses) E5 6. The university will request further consent if my educational data are being used for a purpose different to what was originally stated S2 7. The learning analytics service will be used to promote student decision making (e.g., encouraging you to adjust your set learning goals based upon the feedback provided to you and draw your own conclusions from the outputs received) S3 8. The learning analytics service will show how my learning progress compares to my learning goals/the course objectives S4 9. The learning analytics service will present me with a complete profile of my learning across every module (e.g., number of accesses to online material and attendance) S5 10. The teaching staff will be competent in incorporating analytics into the feedback and support they provide to me S6 11. The teaching staff will have an obligation to act (i.e., support me) if the analytics show that I am at risk of failing and underperforming or if I could improve my learning S7 12. The feedback from the learning analytics service will be used to promote academic and professional skill development (e.g., essay writing and referencing) for my future employability