The show must go on(line): the impact of content and system quality on the usage of television streaming content libraries

ABSTRACT

This study explores the impact of strategic product characteristics (content and system quality), price value, and habit on attitudinal (word of mouth and brand perception) and behavioral (actual usage) outcomes in the context of television companies’ online streaming services (content libraries). The study utilizes structural equation modeling to analyze data from a survey of 1,038 content library users in Germany. The research conceptualizes and validates content quality as a second-order construct comprising four dimensions (actuality, range, relevance, and originality). The study validates system quality as a second-order construct encompassing usability, reliability, and effort expectancy. The results highlight the importance of habit formation in driving actual usage, emphasizing the need for consistent content delivery and personalized content offerings. The results also underscore not only the importance of content quality (particularly actuality, range, and relevance) as a driving factor for usage, brand perception, and word of mouth recommendation, but also that streaming service providers should pay attention to system quality. Overall, the study enriches our understanding of strategies for content libraries, contributing to both theory and practice in the streaming industry.

Introduction

Video streaming is on the rise globally: In 2022, 83% of American households subscribed to a Subscription Video on Demand (SVoD) service, and 80% of European Internet users watched streaming content (Abrams von, 2022; Leichtman Research Group, 2022). However, while SVoD services like Netflix and Amazon Prime Video have been extensively studied, the content libraries of television companies, a significant component in the streaming landscape, have received limited academic attention. These libraries, often comprising repurposed content from linear television programming, are increasingly important for television broadcasters in retaining viewers, especially younger demographics, and competing with SVoD services (Doyle, 2016; Lobato, 2019; Lotz, 2017; Telkmann, 2021a).

Strategically, content libraries serve a set of three interwoven goals, which can be broadly attributed to an attitudinal and behavioral dimension (J. Lee et al., 2012; Vallerand, 1997): enhancing brand image and attractiveness, encouraging user interaction and generating word of mouth, and driving content usage as basis for direct or indirect revenue generation (Chakravarty et al., 2010; Doyle, 2015; J. Lee et al., 2012; Rahe et al., 2021; Vallerand, 1997).

Despite their growing importance, there is a lack of in-depth academic study of content libraries, particularly in contrast to international streaming services. This paper addresses this gap by analyzing the impact of content libraries’ product characteristics on attitudinal and behavioral outcomes, focusing on content quality, user experience, pricing, and habitual viewing patterns (Cha & Chan-Olmsted, 2012; Guo & Chan-Olmsted, 2015; Kotler & Keller, 2016; Madanaguli et al., 2021). To do so, we conduct a quantitative survey of content library users in Germany, which can be considered as one of the largest SVoD markets in the world (Rahe et al., 2021).

The findings of our research indicate that habit formation is a key driver in user behavior, particularly in usage and word of mouth recommendations. Content quality emerges as more influential than system quality and price value in determining attitudinal and behavioral outcomes. This study advances theoretical understanding by conceptualizing content and system quality as multi-dimensional constructs, offering a more nuanced perspective than previous unidimensional approaches (e.g., Madanaguli et al., 2021). It also sheds light on the extent to which consumers perceive indirect financing, such as advertising or license fees, as a price.

Our research not only contributes to the academic understanding of content libraries but also offers practical implications for media managers. Emphasizing habit-forming measures, personalized content, and system quality is vital for content libraries to remain competitive. Furthermore, the insights from the German market can be extrapolated to other countries, enhancing the strategic approaches of content library providers globally. Given the wide-spread distribution of content libraries, insights from our study of the German market are therefore also valuable when looking at other countries and other forms of streaming services.

In the following, we first present the conceptual framework with hypotheses on the effects of independent variables (i.e., content quality, system quality, price value, and habit) on dependent variables (i.e., usage, word of mouth, and brand perception). In the method section we demonstrate how the variables were operationalized and how the data was collected and analyzed. The results section presents the measured effects and the tests of our proposed hypotheses. In the discussion section, we identify implications for theory as well as management practice before pointing out limitations and further research options.

Conceptual framework and hypotheses

The main aim of this paper is to analyze the influencing factors of attitudinal and behavioral outcomes of content library usage. Content libraries (sometimes also called media libraries, Gutzeit et al., 2021) can be defined as closed television station portals structured like libraries with professionally produced (long-format) content; the majority of content is repurposed from the linear television programming schedule (Lobato, 2019; Lotz, 2017). Such content libraries have become a global phenomenon. In countries such as the UK, Germany, Sweden, Norway, Argentina, or Japan, television broadcasters complement their linear offerings via Internet Protocol-distributed, advertising-financed and/or subscription-based content libraries (see examples in Table A1 in the Appendix). These offerings have become a relevant part of the streaming video competition landscape. Hulu, which started by curating the catch-up offerings of U.S. television channels, counts among the most popular SVoD providers, with 28% of U.S. adults tuning in at least every week (Morning Consult, 2022).

In the European context, several television groups have complemented their catch-up content libraries with original content, establishing those offers as part of the streaming consumption mix. Our research focuses on the German media market, one of the largest markets for SVoD in the world (Rahe et al., 2021). The German media system is characterized by strong competition between private and public service broadcasters, which extends to the SVoD market, with both groups attracting substantial usage of their content library offerings (ARD/ZDF Forschungskommission, 2023). Also, private and public groups both invest substantial sums in content production and content libraries (ARD, 2023; Meier, 2023), making these well-developed offerings a suitable object of investigation. As a result, content library usage is well established: 39% of Germans use content libraries on a daily or weekly basis (ARD/ZDF Forschungskommission, 2023; Schauerte et al., 2021).

Because our aim is to analyze attitudinal and behavioral outcomes, actual usage and word of mouth recommendation (behavioral outcomes) as well as brand perception (attitudinal outcome) serve as our dependent variables. We include actual usage as a measure of success instead of behavioral intention or intention to continue using/subscribing, which to date have been mostly used in recent studies of streaming services (e.g., Camilleri & Falzon, 2021; Menon, 2022; Pan et al., 2022; Sulaiman & Tjhin, 2023). As Wu and Du (2012) stated in their examination of intention to use and system usage, behavioral intention “is not a good surrogate for usage, […] actual usage is the least highly correlated with behavioral intention, and researchers should examine […] actual usage” (Wu & Du, 2012, p. 680). For this reason, we employ actual usage as the dependent variable instead of behavioral intention. In online environments, “customers are faced with various information stimuli at the same time to determine what and where to buy […] and what and where to say (WOM) [word of mouth]” (Bilgihan et al., 2016, p. 111f.). In this setting, differentiated media brands can have a competitive advantage in terms of competing for subscribers and establishing customer loyalty (Doyle, 2015; Rahe et al., 2021). Here, brand perception and word of mouth are critical for Video on Demand (VoD) streaming services to attract and maintain consumer interest (Sabrina et al., 2022).

We hypothesize the relationships between the different influencing factors on these dependent variables, summarizing them in a framework. We consider that content and system quality, habit, and price value are factors with potential effects on actual usage, word of mouth recommendation, and perception of content libraries’ brands. Since content libraries are a global phenomenon, we develop a general and not a country-specific conceptualization. Although the framework is tested for German content library users, it should be transferable to other markets as well. In addition, we control for sociodemographic variables and consider usage of SVoD as the opposing streaming content offering. Furthermore, since streaming is a relatively new technology compared to television, for example, we include lead usership and media innovativeness to verify that our model is valid for a broader user base than lead users or individuals with a particular affinity for media innovation alone (Camilleri & Falzon, 2021; Mütterlein et al., 2019; Ozer, 2009). Figure 1 depicts the research model and the constructs (see Table A1 in the Appendix for operationalizations).

Figure 1. Research model.

Price value

Price value is defined as the consumers’ cognitive tradeoff between the value they get from using an application and the monetary cost of using it (Pratama et al., 2022; Venkatesh et al., 2012). Included in the extended model of the Unified Theory of Acceptance and Use of Technology (UTAUT2), price value has been shown to be directly related to behavioral intention, which in turn is a proxy for future usage of information technology (Davis, 1989; Venkatesh et al., 2012). Price value has also been demonstrated to significantly affect actual usage of other information technologies (e.g., mHealth services; Alam & Khanam, 2022). Furthermore, research on service bundles has shown the direct effects of price value on word of mouth recommendation as well as customer attitudes toward a service provider (Ranaweera & Karjaluoto, 2017). Regarding SVoD services, price value has been shown to significantly impact the intention to use Netflix (Pratama et al., 2022). Therefore, as price value seems to be a relevant factor in streaming services, we will consider its relevance for customer use and perception of content libraries’ offerings. We thus formulate the following hypothesis for price value:

H1:

Price value has a positive direct effect on (a) actual usage, (b) word of mouth, and (c) brand perception of content libraries.

Content quality

Arguably, the quality of the content offered can be considered a central dimension of television and SVoD usage alike (e.g., Dasgupta & Grover, 2019; Jung et al., 2009; Noh, 2021). Content quality has often been measured as a one-dimensional construct (e.g., Jung et al., 2009; Pritania & Mulia, 2023; Sulaiman & Tjhin, 2023). In the context of mobile television, content quality has been defined as “a consumer’s assessment that programs are applicable (relevance), up-to-date (timeliness), and sufficient (sufficiency)” (Jung et al., 2009, p. 125). Pritania and Mulia (2023) focused on the variety and perceived quality of content, and Sulaiman and Tjhin (2023) operationalized content quality of VoD services through three items: providing complete movie content, updated movies, and movie quality. Some studies added a second dimension by capturing content range as well (e.g., Kunz, Zabel, et al., 2022; Torres et al., 2014).

Given the importance of content for consumers’ streaming platform usage (Kübler et al., 2021), we propose to conceptualize the quality of the content offering as a multi-dimensional construct. In addition to actuality (timeliness; Jung et al., 2009; D. H. Shin, 2009) and content range (Torres et al., 2014), which are based on past conceptualizations, we consider the dimensions of relevance and originality. Relevance can be understood as the way users perceive the relationship between the information/content and their need for that information/content at that moment (Niininen, 2023). In the context of the uses and gratifications approach, relevance thus refers to the ability of content to meet the needs of users (Katz et al., 1973). In this case, it explicitly refers to the degree to which the content offered matches the personal interests of the consumers (Jung et al., 2009; Kunz, Notbohm, et al., 2022). Public service broadcasters in particular (albeit not exclusively) are obliged to offer relevant and varied content (Telkmann, 2021b), which can be presumed to be relevant for the usage and perception of content libraries. Originality, understood as unique content offered exclusively, represents another key quality distinction between television channels and streaming services, and between one streaming service and its competitors (Palomba, 2020, 2022). So, we measure content quality as a second-order construct comprising four dimensions (actuality, range, relevance, and originality) and propose the following hypothesis:

H2:

Content quality has a positive direct effect on (a) actual usage, (b) word of mouth, and (c) brand perception of content libraries.

System quality

The system that presents the content is also a key feature of media services (S. Shin & Park, 2021). System quality describes the “extent to which the presentation of the content facilitates its reception” (Berger & Matt, 2016; adapted from, p. 5; Nelson et al., 2005). System quality thus pertains to the structural characteristics and performance dynamics of a system (Xu et al., 2013).

System quality has been conceptualized as a multi-dimensional construct (e.g., Berger & Matt, 2016; Gupta & Bhatt, 2021; Nelson et al., 2005). For example, in the context of data warehousing, Nelson et al. (2005) differentiated between system-related aspects (e.g., accessibility and reliability) and task-related aspects (e.g., response time) of system quality. In the context of mobile banking apps, Gupta and Bhatt (2021) identified the dimensions reliability, ease of use, user interface, response time, security, and functionality.

However, the system quality of streaming services has to date been measured mainly one-dimensionally (e.g., Sulaiman & Tjhin, 2023). Some studies assessed quality of presentation in terms of content organization and layout (e.g., Berger & Matt, 2016; Nelson et al., 2005). Other studies included ease of use as a proxy for system quality (e.g., Peukert et al., 2019) or built an overarching usability construct incorporating system performance, ease of use, and design elements (e.g., Palomba, 2022; Sulaiman & Tjhin, 2023). S. Shin and Park (2021) organized the system’s properties into quality and ease of use, where quality referred to system reliability and ease of use referred to convenience (M.-K. Lee et al., 2019; Li, 2017). Since user experience can affect word of mouth recommendation (Huang et al., 2017), brand perception (Walter et al., 2018), and actual usage (Topolewski et al., 2019), we conceptualize system quality as a multi-dimensional construct using three dimensions: technical system reliability refers to the “degree to which users can use the system reliably while using the OTT [over-the-top] service” (S. Shin & Park, 2021, p. 5), usability and UX design include the design and ease of finding content within the system (Palomba, 2022), and effort expectancy (Peukert et al., 2019). While effort expectancy is originally a construct of the UTAUT2 model, we include it as a facet of system quality since it seems to be important for the transition from television to streaming usage, especially for new users (Elsafty & Boghdady, 2022). In this way, we combined constructs already used in other contexts in a new composition against the background of content libraries as media services. We thus propose the following hypothesis:

H3:

System quality has a positive direct effect on (a) actual usage, (b) word of mouth, and (c) brand perception of content libraries.

Habit

Habit has been defined as “the extent to which people tend to perform behaviors (use IS [information systems]) automatically because of learning” (H. Limayem et al., 2007, p. 705). Prior studies using the UTAUT model or uses and gratifications approach have identified habit to be a major driver of television viewing (e.g., Guo & Chan-Olmsted, 2015; Rubin, 1983, 2009). Since habit-forming has also been identified as a predictor of behavioral intention to use streaming services (Madanaguli et al., 2021), it is reasonable to assume that habit influences the actual use of content libraries, too. Furthermore, a recent study by Soren and Chakraborty (2023) indicated that the habit of using OTT platforms is a significant predictor of word of mouth recommendation for such platforms. This influence can therefore also be assumed to apply in the case of content libraries. An important task of (media) brands is to reduce the complexity of users’ decisions, a task that is becoming more important because of the increasingly complex media ecosystem. According to Berkler (2008), habit-forming is a key factor in simplifying the use of linear television content. Habit, as an inherently longer-term construct, is therefore particularly relevant to the strategic long-term goals of content library providers (Albarran, 2019). Thus, we propose the following hypothesis:

H4:

Habit has a positive direct effect on (a) actual usage, (b) word of mouth, and (c) brand perception of content libraries.

Method

Operationalization and scale selection

The following section provides an overview of the operationalization and selection of the scales used. We relied on established items and scales identified in the literature to operationalize the dependent and independent variables. Derived from previous studies (i.e., on mobile television, streaming services, or information technology) we have aimed for a general conceptualization that is applied here to content libraries. A complete list of items for all constructs can be found in Table A2 in the Appendix. In contrast to other studies (e.g., Jung et al., 2009; Pan et al., 2022; Pritania & Mulia, 2023), our independent variables and dependent variables brand perception and word of mouth refer to a specific provider, namely the content library most familiar to the respondent. However, our dependent variable actual usage refers to content libraries in general. This contributes to high discriminant validity and helps to counter common-method bias (Podsakoff et al., 2012). Nevertheless, the effects of the independent variables on actual usage and the overall explanatory power (regarding actual usage) may be smaller when compared with studies that measure all items in relation to the same object.

The first dependent variable is actual usage, which comprises the actual frequency and intensity of use of content libraries as well as the extent of this use. Thus, we used three different measures for actual usage of content libraries. First, we adapted the scale from Venkatesh et al. (2012) to evaluate frequency (several times per day to less than once a month). Frequency was inverted, that is, decreasing from left to right. Second, we measured intensity of using content libraries (“I use content libraries intensively”) on a 7-point Likert scale (1 = Strongly disagree, 7 = Strongly agree; modified from Kunz, Notbohm, et al., 2022; Venkatesh et al., 2012). Third, we asked respondents how many hours per week they use content libraries (e.g., Zhang et al., 2006).

The two other dependent variables are brand perception and word of mouth, as critical factors for attracting and retaining users of content libraries. Brand perception was measured by four items (e.g., “The image of the brand … is unique.”) selected from Rahe et al. (2021) more comprehensive item battery. Based on Turel et al. (2010) and Eelen et al. (2017), the third dependent variable word of mouth was also measured by four items (e.g., “I say positive things to other people about … ;” Turel et al., 2010).

We furthermore operationalized the two independent variables perceived content and system quality as multi-dimensional second-order constructs, which were validated in a pre-study with 702 university students (see data analysis section). Perceived content quality – as content can fulfill different purposes such as information, entertainment, or education (Berger & Matt, 2016) – was operationalized by actuality, relevance, and content range. We added the dimension of originality to reflect the recent development that some SVoD providers now offer original content. All four content dimensions were measured by four items each – content actuality (e.g., “ … provides up-to-date content.;” Jung et al., 2009; D. H. Shin, 2009); content relevance (e.g., “ … offers content that is of interest to me.;” Jung et al., 2009; Kunz, Zabel, et al., 2022); content range (e.g., “ … has a large amount of content.;” Torres et al., 2014); and content originality (using four items created by the authors, e.g., “ … offers its own productions (originals).”) – based on Palomba’s (2020) study of SVoD product attribute trade-offs.

Perceived system quality, which represents the structural characteristics of a technical media system (Xu et al., 2013), was operationalized using three constructs: technical system reliability, usability, and effort expectancy. In terms of technical system reliability (e.g., “I think that … provides a very reliable service.;” D. H. Shin, 2009), we considered reliability, security, and freedom from defects in line with D. H. Shin (2009) and Palomba (2022). We defined usability, first, by how easily users can navigate the system (e.g., “Navigation within … is easy to understand.;” M.-K. Lee et al., 2019) and, second, by perceived quality of presentation (Berger & Matt, 2016) using UX design items (e.g., “ … is well thought-out and stylish.;” Palomba, 2022). We based effort expectancy on Davis (1989) (e.g., “I find … easy to use.;” Peukert et al., 2019).

To operationalize the two additional, established independent variables, we adapted the wording for habit (e.g., “I must use … ”) and price value (e.g., “ … is good value for money.”) from the operationalization established by Venkatesh et al. (2012). Habit as learned behavior was measured using three items to reflect addiction, compulsion, and automatism to using content libraries (M. Limayem & Hirt, 2003). Price value describes not only the price-performance ratio but also that price is perceived as good value for money (Venkatesh et al., 2012). Additionally, we used life satisfaction as a marker variable to detect common-method variance (Diener et al., 1999; Podsakoff et al., 2012).

The two second-order constructs content quality and system quality were specified formatively, since they encompass a range of content- and system-related facets. In contrast, the established constructs actual usage, word of mouth, brand perception, habit, and price value were specified reflectively, as they consist of indicators that each represent a (varied) manifestation of the theoretical concept and are closely correlated with one another within their respective construct (Hair et al., 2020).

Study design and data collection

Based on the operationalization described above, a fully structured questionnaire was developed and implemented using Tivian XI GmbH’s EFS survey. The online survey was conducted by acquiring participants with the help of certified panel provider Bilendi. The survey was conducted in two waves in February and March 2023, four weeks apart. We invited subjects who had completed the survey in the first wave to participate in the second wave. The two datasets were linked and merged through individual test IDs.

A total of 1,264 respondents completed both survey waves. During the subsequent data cleansing, we excluded all participants who had completed the survey in less than 5 minutes (average duration of response about 14 minutes). Next, we checked for inconsistent response behavior by controlling for identical answers in mutually exclusive item batteries. One-sided clicking behavior (e.g., respondents always chose a value of 1, 4, or 7) was checked for the preferred genres by calculating the standard deviation for each of the item batteries and manually checking the cases with a standard deviation (SD) of zero to see whether this one-sided clicking behavior was present throughout the questionnaire. We also checked for outliers in free response fields, such as the specification of hours per week in which streaming content was used. In total, we excluded 226 cases in this way. After completing the data cleansing, n = 1,038 cases remained. Individual missing values were replaced in IBM SPSS Statistics 27 by the mean values of the respective data series.

The final data sample consists of 49% female and 51% male participants. This resembles the proportion of men and women using paid streaming services according to a study by the consumer offices in Germany (Schmidt & Zaborowski, 2017). The age of the participants ranged between 18 and 89 years with a mean value of 46.73 years (SD: 14.71). The age distribution of the sample covers the most relevant age groups of German content library users (Beisch et al., 2021): The 30–49 age group (52.3% in this study) is the most prolific user group of content libraries in Germany, followed by the 50–69 age group (28.6% in this sample) and the 14–29 age group (10.9% in this sample). The over-70 age group is the least likely to use content libraries (8.2% in this study).

Data analysis

The survey data was analyzed using IBM SPSS Statistics 27 and SmartPLS 3.3.5 for factor analyses and variance-based structural equation modeling (SEM). Exploratory factor analysis (EFA; Netemeyer et al., 2003) using principal component analysis with varimax rotation was conducted to examine the reliability and validity of the adapted constructs and measures and to determine the dimensionality of the second-order constructs. In a second step, we evaluated the measurement and structural models using SEM and tested the hypotheses. To analyze the regressions within the model and test for significance via bootstrapping, we used partial least squares (PLS) path modeling (Hair et al., 2019). In our study, we applied PLS-SEM to extend the theory by testing a new model that we created based on the hypotheses presented above. We evaluated the quality of the SEM analysis by using established threshold values (e.g., Sarstedt et al., 2017). In addition, we tested for significant differences between public and private streaming services by applying multi-group analysis in SmartPLS.

To create reliable and valid measures for the main study, we used a preliminary study with German university students for item trimming and scale validation. The survey was conducted in two weeks in January and February 2023. After elimination of 23.5% of the responses, because of incomplete surveys for example, n = 702 datasets could be used for the analysis. Based on the EFA, the factor loadings and correlations of the surveyed constructs of the main study were examined, too. We decided to eliminate the items Convenience 1 and 2, Exclusivity 1, Brand 3, Personalization 4, and Word of Mouth 1 due to low factor loadings (<0.5 and 0.7). Second-order constructs were formed for content quality based on the constructs timeliness, relevance, exclusivity, and content range, and for system quality based on the constructs effort expectancy, technical system quality, and usability. Finally, we tested the second-order constructs using confirmatory factor analysis.

Results

The results of our analysis demonstrate three main findings. First, the modeling of content quality as well as system quality as multi-dimensional second-order constructs show reliable and valid results. Because our constructs were based on general (albeit less complex) conceptualizations, they may be fruitfully applied in other video/streaming settings as well. Second, the quality and robustness of the overall model is satisfactory including the absence of a potential common method bias. Third, this allows the meaningful testing of the proposed hypotheses. The independent variables explain an essential part of the dependent variables’ variance, underscoring the explanatory power of our model. Also, we find that except for two hypotheses (price value and system quality on actual usage) most hypotheses are supported in our study. Anyhow, the effect size varies considerably, pointing to the fact that our studied antecedents affect actual usage, word of mouth and brand perception differently. This could also be used to warrant a replication of the study in other market settings.

Content and system quality as multi-dimensional second-order constructs

We modeled content and system quality as second-order constructs to ensure construct validity (Bagozzi et al., 1991). In our EFA, we identified that the content quality construct consists of four highly significant dimensions (p ≤ 0.001): content actuality (α = 0.909, loadings: 0.844–0.916), content exclusivity (α = 0.854, loadings: 0.855–0.898), content relevance (α = 0.938, loadings: 0.890–0.939), and content range (α = 0.943, loadings: 0.920–0.930). After scale trimming and item selection, all identified factors still consist of at least three items. To assess factor reliability, we use Cronbach’s α and the factor loadings to assure item reliability (Fabrigar et al., 1999; Nunnally, 1978). All indicators reached the Cronbach’s α threshold of 0.7 and can thus be considered satisfactory. Regarding the second-order construct content quality, the indicators’ factor loadings across all four dimensions of content quality range from 0.844 to 0.939 (see Figure 2).

Figure 2. Dimensionality of content quality as a second-order construct.

The system quality construct includes three highly significant dimensions (p ≤ 0.001): effort expectancy (α = 0.921, loadings: 0.852–0.922), technical system reliability (α = 0.917, loadings: 0.888–0.907), and usability (α = 0.917, loadings: 0.870–0.909). With factor loadings between 0.852 and 0.922 and Cronbach’s α above 0.7, both factor and item reliability can be considered satisfactory, too (see Figure 3).

Figure 3. Dimensionality of system quality as a second-order construct.

In addition, the thresholds for assessing composite reliability (CR > 0.7), convergence validity (AVE >0.5), and discriminant validity (AVE > MSV) were met (see Table A2 in the Appendix) (Fabrigar et al., 1999; Hair et al., 2010).

Quality and robustness assessment

Overall, the quality and robustness of all the constructs in our research model are satisfactory (the constructs and results of our EFAs are shown in Table A2 in the Appendix). Regarding the first-order constructs, the independent variables habit (α = 0.877, loadings: 0.855–0.935) and price value (α = 0.959, loadings: 0.953–0.968) were measured by three reliable items each. Also, the dependent variables actual usage (α = 0.819, loadings: 0.776–0.903), word of mouth (α = 0.903, loadings: 0.911–0.918), and brand perception (α = 0.885, loadings: 0.879–0.915) were measured by three reliable items each.

With values of Cronbach’s α above the threshold of 0.7 for all factors, the model exhibits good factor reliability. The thresholds for composite reliability (CR > 0.7), convergence validity (AVE >0.5), and discriminant validity (AVE > MSV) were also met (see Table A3 in the Appendix). Construct validity was checked by an analysis of the Fornell-Larcker criterion and the heterotrait-monotrait (HTMT) ratio of correlations (<0.9; Henseler et al., 2015; see Tables A4, A5, and A6 in the Appendix). The constructs showed sufficient discriminant validity for all factors apart from the two second-order constructs, because these were formed of correlating factors (Koufteros et al., 2009). In the structural model, the constructs correlated on low to medium levels, also underscoring good discriminant validity (see Table A6 in the Appendix).

Since the variance inflation factors (VIFs) showed values between 1.220 and 2.846 (conservative threshold: < 3; Hair et al., 2019) for inner constructs, the EFA-based scale-trimming and item-selection process was effective and we do not consider multicollinearity as a problem (Hair et al., 2010). With a standardized root mean squared residual (SRMR) value of 0.076 (<0.08; Hu & Bentler, 1999), the overall goodness of fit of the model is acceptable.

Additionally, we evaluated the model’s predictive power according to Hair et al. (2020) and Shmueli et al. (2019) using the PLSpredict procedure. The blindfolding-based and cross-validates redundancy measurement Q² shows values above zero for all indicators and constructs, indicating the model has good predictive validity (Shmueli et al., 2019).

As this study is based on self-reported survey data, we used different measurement scales as well as inverted scales to avoid common-method bias. Low VIF values are already an indicator that common-method bias is not a central problem (Kock, 2015). We further tested our model post hoc for common-method bias by using Harman’s single-factor test. Some 44.43% of the variance was explained by building one common factor (<50%; Harman, 1967). To further control for this bias, we examined the correlations between the dependent variables, the independent variables, and the marker variable. The marker variable and the research model variables display only minor correlation (0.073–0.136, see Table A6). Thus, both tests confirm that common-method bias was not an issue for this study.

Hypotheses testing

Looking at the main effects of the model, 37.1% of the total variance of the actual usage of content libraries, 50.3% of the total variance of word of mouth, and 46.2% of the total variance of brand perception could be explained (R² = 0.371; 0.503; 0.462; see Figure 4). This indicates moderate to substantial explanatory power, according to Sarstedt et al. (2017).

Figure 4. Main effects of the research model.

Except for the effects of price value (p > 0.05; γ = 0.027) and system quality (p > 0.05; γ = 0.058) on actual usage, all independent variables proved to have a significant direct influence on the actual usage of content libraries, as well as users’ word of mouth behavior and brand perception (γ = 0.098–0.422). Habit has the strongest influence on actual usage (p ≤ 0.001, γ = 0.402), followed by content quality (p ≤ 0.001, γ = 0.263). In addition, habit has the strongest influence on word of mouth (p ≤ 0.001, γ = 0.422), followed by content quality (p ≤ 0.001, γ = 0.253), price value (p ≤ 0.001, γ = 0.134), and system quality (p ≤ 0.05, γ = 0.098). Brand perception is most strongly influenced by content quality (p ≤ 0.001, γ = 0.282), habit (p ≤ 0.001, γ = 0.276), system quality (p ≤ 0.001, γ = 0.166), and price value (p ≤ 0.001, γ = 0.151). Table 1 summarizes the results of the hypotheses testing.

Table 1. Hypotheses testing.

H1a: Price Value ➔ Actual Usage	not supported
H1b: Price Value ➔ Word of Mouth	supported
H1c: Price Value ➔ Brand Perception	supported
H2a: Content Quality ➔ Actual Usage	supported
H2b: Content Quality ➔ Word of Mouth	supported
H2c: Content Quality ➔ Brand Perception	supported
H3a: System Quality ➔ Actual Usage	not supported
H3b: System Quality ➔ Word of Mouth	supported
H3c: System Quality ➔ Brand Perception	supported
H4a: Habit ➔ Actual Usage	supported
H4b: Habit ➔ Word of Mouth	supported
H4c: Habit ➔ Brand Perception	supported

Control for sociodemographic factors (age, gender, education, and income) showed no significant influences on the dependent variables actual usage, word of mouth, or brand perception. We also used media innovativeness and lead usership as well as actual usage of SVoD services as control variables. Lead usership has only a small significant effect on word of mouth (p < 0.05; γ = 0.093). SVoD usage has a slight significant effect on actual usage of content libraries (p < 0.01; γ = 0.096) and a slightly negative effect on word of mouth (p < 0.05; γ = −0.065). When considering the significant effects of the control variables, the explanatory power increases slightly: 38.1% of the total variance of actual usage (R² = 0.381), 52.2% of the total variance of word of mouth (R² = 0.522), and 47.2% of the total variance of brand perception (R² = 0.472). Since most of the control variables have either a non-significant or only a very weak influence on the dependent constructs, we did not include these variables in the final model. We also examined interaction effects between the dependent variables: While actual usage as a behavioral outcome shows no significant interactions with the attitudinal outcomes (p > 0.05), word of mouth and brand perception show significant interaction effects (brand perception ➔ WOM: p < 0.01; γ = 0.294), increasing the explained variance of word of mouth to 55.0% (R² = 0.550).

Finally, a multi-group analysis was also carried out in SmartPLS to check for differences arising from the provider type of the content libraries. For this purpose, we compared responses about public service broadcasters’ content libraries with responses about private broadcasters’ content libraries. The comparison shows that the established relationships studied in our model do not differ significantly between these two groups. The only significant difference between the two datasets is the influence of habit on word of mouth (p ≤ 0.05). This effect is more pronounced for the private providers than for the public content libraries (public broadcasters: p ≤ 0.001, γ = 0.389; private broadcasters: p ≤ 0.001, γ = 0.539). The fact that there are otherwise no significant differences indicates that this is a stable model that is valid for both the public and private content libraries.

Our empirically tested research model explains a large part of the dependent variables (i.e., actual usage, word of mouth and brand perception), underscoring its relevance. Also, with the exception of two relationships (among price value and system quality and actual usage) most hypotheses are supported. The study demonstrates that the effect sizes vary considerably, with habit and content quality having the strongest effects and thus being the most relevant antecedents, followed by weaker effects of price value and system quality on the dependent variables.

Discussion

Implications for theory

Our study yields several theoretical implications for research on streaming services in general and content libraries in particular. The conceptualization developed in this paper is not specific to the German market and thus can be applied to other countries with similar content library offerings (e.g., relatively strong local players with a high degree of local productions). Also, their general nature, basis on earlier general approaches and good concept fits indicate that these conceptualizations may be fruitfully applied in related settings, for example, other streaming offerings such as SVoD services. More specifically, we derive four key implications for theory:

First, our study underscores the multi-dimensional nature of the content quality construct, which, arguably, can be seen as the most relevant product-related feature of content libraries or streaming services in general. We identify four independent dimensions of content quality, namely content actuality, content range, content relevance, and originality. The construct also reflects the exclusivity of content, which helps to conceptualize content quality as a competition-related construct. Originality has gained popularity and importance in the streaming market since differentiation has been shown to be a key factor leading to competitive advantage (Chiang & Jhang‐Li, 2020; Palomba, 2020). This complements previous research that has conceptualized content quality in one dimension (e.g., Jung et al., 2009; Pritania & Mulia, 2023; Sulaiman & Tjhin, 2023) or at most in two dimensions (e.g., Kunz, Zabel, et al., 2022; Torres et al., 2014). In addition to previous research, we thus provide a holistic conceptualization that unifies the insights of investigations in various empirical settings (such as mobile television or virtual reality) (Jung et al., 2009; Kunz & Santomier, 2019). In addition, the approach does not rely on content classification schemes based on industry conventions (such as genres, e.g., Noh, 2021), avoiding misperceptions from respondents. We also validate the concept empirically for the relevant use case of content libraries. The analysis shows that content range, relevance, and actuality can be considered roughly equal in importance, with content originality considered less important. The media libraries’ success thus depends on the programming of the associated (linear) television stations, which provides the bulk of their content offering, but original production contributes to market differentiation to a lesser extent.

Second, regarding system quality, we develop and empirically test a unifying second-order construct that allows us to assess the effects of the “technical quality” of a service. This approach builds on prior studies that have conceptualized system quality multi-dimensionally (e.g., Berger & Matt, 2016; Gupta & Bhatt, 2021; Nelson et al., 2005), albeit in other settings. Our approach is tailored to hedonic audiovisual services, such as SVoD offerings or content libraries. It thus extends prior literature in this sector, which operationalized only singular aspects (e.g., Sulaiman & Tjhin, 2023) or multiple aspects with single items (e.g., Li, 2017), increasing construct validity. Our results show that respondents considered the independent dimensions usability and system reliability relatively more important than effort expectancy. This implies that customers perceive hardware- and design/UX-related “hard” product characteristics more important than “soft” or abstract expectations about ease of use when assessing technical system quality. Indeed, in contrast to previous studies on the transition from television to streaming usage (Elsafty & Boghdady, 2022), our study suggests that ease of use is no longer a relevant factor. This might be because a large part of the population has by now “learned” streaming as a technology and can use it without much effort.

Third, our model proposes a comprehensive set of (competition-related) attitudinal and behavioral outcomes. An analysis of interaction effects between the attitudinal and the behavioral outcomes did not yield significant results, underscoring their discreteness as sets of goals. Our study thus helps to broaden previous research that considered either behavioral (usage; Wu & Du, 2012) or attitudinal outcomes (brand image, e-WOM; Sabrina et al., 2022). Therefore, our approach reflects the nature of the competitive space that content library providers must navigate, and where they must pursue different agnostic or even competing goals. The results on word of mouth – which, in line with the literature (e.g., Donthu et al., 2021; Rageh Ismail & Spinelli, 2012), is also affected by the perception of the underlying television brand – highlight the significance of online offerings for television networks, which may well go beyond attracting and monetizing actual usage. Moreover, the importance of this effect will grow when streaming service providers pursue subscription strategies, since attracting and retaining subscribers is more relevant for subscription-based business models than actual viewing numbers (Lotz, 2022).

Finally, our results underscore the role of price value for word of mouth and brand perception of content library services. Where price value has been demonstrated to significantly affect actual usage in other settings (e.g., mHealth services; Alam & Khanam, 2022) and has been shown to influence intention to use streaming services (Pratama et al., 2022), this could not be replicated in our study. Perhaps the particularities of German content libraries play a role here, where public service providers offer all content for free and even the private competitors pursue a freemium model with attractive content available free of charge. While price value does not significantly influence actual usage when (most) options are completely free (i.e., demanding a price of zero), the significant effects on word of mouth and brand perception persist, in line with other studies (Ranaweera & Karjaluoto, 2017). This implies that perception of price value (even for a prize of zero) could still play an important role in forming attitudes and attracting new customers.

Implications for management

Our study yields meaningful implications for managers of content libraries or competing services. First, the high impact of habit on actual usage (and on brand perception and word of mouth) emphasizes the relevance of a strategic approach to attract and retain users, creating established patterns of consumption. This hints at the need to produce or provide a constant stream of content rather than focusing on headline- and attention-grabbing high-profile productions (Kunz, Notbohm, et al., 2022). This may be the case at least for content libraries, which are linked to the linear television networks, but may play out differently for other SVoD provider types. At the same time, habit-forming measures appear worthwhile managerial strategies, even if they do not directly translate into actual usage. This might include the repeated (if selective) activation of the already acquired user base, seen for example in serial program brands. Such an approach – similar to “stripping” strategies in linear program planning – targets users’ learning and habituation effects (Telkmann, 2021a). In addition, cross-promotion with linear television helps to activate television consumers. Since habit has the strongest effect on word of mouth recommendation, reminding customers of relevant, specific content appears particularly appealing. This might imply the use of differentiated and targeted communication campaigns that leverage specific interest content, which may not attract the broadest customer groups but provide a strong trigger for some groups. Since word of mouth recommendation is particularly driven by user engagement (Anastasiei & Dospinescu, 2019), targeted communication might disproportionately increase word of mouth, which, due to the importance of word of mouth for brand perception, would also pay off for the brand.

Our analysis of content quality – which plays a central role in usage, word of mouth, and brand perception – reinforces this notion. The most relevant factor here is personal relevance, which means catering to audiences’ tastes with specific offerings. Here, overall content range (reflecting the time-sensitive property of these offerings) and actuality of the productions are particularly relevant and might also warrant providing an up-to-date program schedule. While Seifert et al. (2023) highlighted the implementation of exclusive digital purchase strategies from a movie distributor’s point of view, the originality of production, while increasing content quality, has a significantly smaller effect in our study. Therefore, managers should be aware that securing exclusivity (which comes at a price) yields diminishing returns for networks offering content libraries, once again.

Our study also differentiates provider types. In general, the group analysis of public and private content libraries identifies similarities between public service and private broadcasters. This highlights the strategic connectedness of both player groups, in that they effectively inhabit the same competitive space. However, several differences also emerge. For instance, relevance is of greater importance for public service media offerings, but the identified factors have more impact on word of mouth for private broadcasters. Therefore, it can be argued that private content libraries have better adapted their content mix to increase word of mouth (or provide content that is more appropriate, such as entertainment content in the form of reality shows or daily soaps).

The data also shows that system quality is a differentiating factor, particularly when it comes to the perception of the service and word of mouth recommendations for it. For managers, system quality can be seen as a basic prerequisite. It can have negative attitudinal repercussions if not very well developed (e.g., customers complaining on social media about bad usability), but it can help to differentiate a brand and increase word of mouth recommendations for the service. Netflix, which is seen as a benchmark in the streaming sector (Hennig-Thurau et al., 2019), invests significant resources in this area, raising the expectations of customers for other services, too. Our analysis shows that users value purely technical properties such as transmission quality and UX-based aspects to the same degree. However, where users are already familiar with the former from linear television distribution, the latter represent a challenge that demand new competencies and learning routines.

Limitations and further research

While this study offers significant theoretical and managerial implications, it is important to acknowledge several limitations. Although this study establishes the effects of content quality, system quality, habit, and price value on brand perception, word of mouth, and actual usage of content media libraries, we are still at the beginning of theory building. Several empirical studies upon which our research is based are context-specific (e.g., with regards to use cases or industries), resulting in differing concepts and operationalizations, often validated through only a few or no replication studies. We tried to advance theory by developing and validating multi-faceted second-order constructs for the crucial concepts of content quality and system quality. These were contextualized and tested for content libraries; the wider application of these concepts in further research on other streaming services such as SVoD as well as adjacent media technologies has the potential to link different, previously isolated research streams.

From a methodological perspective, the use of a convenience sample from Germany may limit the generalizability of the findings to the entire content library population. Nevertheless, it is worth noting that the demographic composition of our sample closely aligns with the broader streaming consumer market.

Furthermore, the coefficient of determination, which assesses the extent to which our model explains the variability of dependent variables, suggests the potential for extending our model. Unexplored factors, such as substitutability with other (competing) streaming services (Kunz, Notbohm, et al., 2022), personal topical interest (Autenrieth et al., 2021), or antecedents like media usage motives (e.g., fun, relaxation, or escapism) or personality traits (e.g., the Big Five) of our research model’s influencing factors (i.e., habit, content quality, system quality) have the potential to indirectly affect actual usage, brand perception, and word of mouth (i.e., mediation effects).

To mitigate common-method bias, we implemented various measures, including the use of formative constructs, reversely labeled items, marker variables, and diverse scales (MacKenzie & Podsakoff, 2012; Podsakoff et al., 2003). We also used data generated at two different time points to reduce common-method bias. Nevertheless, it is important to highlight that our study relies exclusively on data collected through standardized questionnaires from respondents. Consequently, correlations between variables in the research model may, to some extent, be influenced by self-reporting tendencies among participants. Although this research approach is well established in media economics and management, as well as fields like marketing and information systems, future behavioral research endeavors should strive to overcome this limitation. For instance, incorporating diverse data sources, such as peer evaluations, could enhance the exploration of causality between SEM variables (Bullock et al., 1994; Iacobucci, 2009). Regarding further research, it would be interesting to replicate the study with other SVoD services as a whole: The importance of different content and system quality dimensions could change, depending on the service type and competitive environment. Such studies could verify how the dimensions of content quality are perceived in other streaming and SVoD settings.

In addition, a closer examination of the price value concept in general and the perception of indirect prices (e.g., resulting from offerings financed through license fees or advertising) in particular would yield interesting insights, linking this research to the wider literature discussing willingness to pay for digital journalism or entertainment. Another interesting aspect for subsequent research could be the analysis of possible moderating or mediating effects of price value on dependent variables. It may be assumed here that content quality, system quality, and habit have a significant influence on consumers’ perception of the price value of content libraries. At the same time, the possible substitution effects between SVoD and content library usage merit further investigation. This would complement the model introduced in this paper and could also indicate the degree to which content library and SVoD offerings compete in the same space. Finally, the model could better reflect newer developments in the streaming sector, most notably the use of AI for data analytics and recommendations, content personalization, etc. Understanding these implications would yield additional practical insights not only for the whole content streaming sector, but also for national economies that switch content or service provision to an over-the-top, on-demand delivery model in their education, training, or software development sectors, for example.

Acknowledgement

The authors would like to thank the co-editors and the anonymous reviewer. Their support during the review process and constructive suggestions contributed significantly to improving the quality of this manuscript.

Disclosure statement

No potential conflict of interest was reported by the author(s).

Additional information

Notes on contributors

Christian Zabel

Christian Zabel is Full Professor for Innovation and Corporate Management at TH Köln - University of Applied Sciences. His research focuses on production and distribution in digital media, digital ecosystems, emergent media technologies and the digital transformation of (media) companies. Before, he headed the product management of t-online.de, Germany’s largest online publisher and was executive assistant to Deutsche Telekom’s CEO René Obermann, overseeing strategic cooperations with the media industry. Christian Zabel studied journalism in Dortmund and Brussels and political science at Sciences-Po Paris (IEP).

Reinhard E. Kunz

Reinhard Kunz is a full professor in the department of media management at Bauhaus-Universität Weimar in Germany and holds the chair of „Innovation Management and Media. Before, he was an Associate Professor at the University of Cologne, a Professor for Marketing and Sales at the Management Center Innsbruck in Austria and a Junior Professor for Media Management, Sport Media at the University of Bayreuth in Germany. He focuses on various aspects of media research and innovation management. Primarily, he conducts research on the media and tech industries, organizational and individual behavior, as well as the management of digital innovations and transformations and their impact on customers, enterprises, and the society. His research emphases are in business model dynamics, proactive decision making, entertainment science, and media and technology user behavior and engagement. At the intersection of entertainment science and user behavior, he studies, for instance, the sports, esports, gaming, and motion picture industries in terms of media technology acceptance, streaming media usage, service ecosystems, or success factors.

Verena Telkmann

Verena Telkmann is a research assistant in the field of production structures for digital media and a Ph.D. candidate at Johannes Gutenberg-University Mainz and TH Köln – University of Applied Sciences. In her research, she focuses on digital media production and media economics. She completed her bachelor’s degree in business administration with a focus on marketing in 2014 at the institute for dual study programs at Osnabrück University of Applied Sciences and her master’s degree in communication, multimedia and market management in 2017 at Dusseldorf University of Applied Sciences. She gained practical experience in online marketing of CHECK24 Mobilfunk GmbH.

Daniel O’Brien

Daniel O’Brien is a research assistant at the Technische Hochschule Köln. He has been supporting Professor Zabel in his research there since 2022. Previously, he was a research assistant at the University of Cologne at the Chair of Media and Technology Management. His Ph.D focussed on the digital transformation of entrepreneurial journalism. In addition, he works as a freelancer in research and teaching.

Appendix

Table A1. Exemplary countries and their television stations’ content libraries.

Exemplary countries	Examples of television stations‘ content libraries
Germany	ARD Mediathek (https://www.ardmediathek.de) RTL+ (https://plus.rtl.de)
UK	BBC iPlayer (https://www.bbc.co.uk/iplayer) ITV Player (https://www.itv.com)
US	ABC Player (https://abc.com) NBC Player (https://www.nbc.com)
France	France TV (https://www.france.tv)
Netherlands	NPO Start (https://npo.nl/start)
Italy	RaiPlay (https://www.raiplay.it)
Scandinavia	SVT Player Sweden (https://www.svtplay.se) TV4 Player Sweden (https://www.tv4play.se) NRK TV Norway (https://tv.nrk.no)
Japan	NHK Player (https://www3.nhk.or.jp/nhkworld/en/live/)
Argentina	Argentinisima Satelital (https://www.argentinisimatv.com.ar)

Table A2. Constructs and results of exploratory factor analyses.

Construct	Item	Factor Loading	Cronbach’s Alpha
Actual Usage
AU_Inten	I use media libraries intensively.	0.889	0.819
AU_Freq	Overall, how often do you use media libraries (e.g., ARD or ZDF Mediathek)? (Less than once a month – several times a day)	0.903
AU_hours	Without thinking about it too much, roughly how many hours a week do you spend using media libraries (e.g., ARD or ZDF Mediathek)?	0.776
Brand Perception
Brand1	The … brand is different from other competing video streaming providers.	0.909	0.885
Brand2	The image of the brand … is unique.	0.915
Brand4	The … brand is quite different and more individual than competing video streaming providers.	0.879
Word of Mouth
WOM2	I’m trying to get others interested in …	0.911	0.903
WOM3	I actively defend … against its critics.	0.915
WOM4	I say positive things to other people about …	0.918
Content Actuality
Content_Act1	… provides up-to-date content.	0.879	0.909
Content_Act2	… offers contemporary content.	0.905
Content_Act3	… offers fresh content.	0.916
Content_Act4	… offers many new releases.	0.844
Content Relevance
Content_Rele1	… offers content that is of interest to me.	0.918	0.938
Content_Rele2	… offers content that is relevant to me.	0.939
Content_Rele3	… offers content that I need.	0.890
Content_Rele4	… offers content related to my concerns.	0.928
Content Range
Content_Ran1	… has a large amount of content.	0.920	0.943
Content_Ran2	… offers a wide range of different productions.	0.930
Content_Ran3	The content offered by … is varied.	0.926
Content_Ran4	On … I can access a broad range of content.	0.923
Content Originality
Content_Orig2	… offers its own productions (originals).	0.886	0.854
Content_Orig3	… offers content that I can’t find with other providers.	0.855
Content_Orig4	… offers content produced by this provider.	0.898
Effort Expectancy
EE1	Learning how to use … is easy for me.	0.908	0.921
EE2	My interaction with … is clear and understandable.	0.922
EE3	I find … easy to use.	0.915
EE4	I find it easy to get … to do what I want it to do.	0.852
Usability
UX1	… is well thought-out and stylish.	0.909	0.917
UX2	… has a visually appealing design.	0.898
UX3	Navigation within … is easy to understand.	0.870
UX4	The app/website of … is well designed.	0.902
Technical System Reliability
System_Rel1	I think that … provides a very reliable service.	0.907	0.917
System_Rel2	I think that … loads quickly as an application.	0.891
System_Rel3	I think that … is safe to use.	0.888
System_Rel4	… is free of defects and functions as advertised.	0.895
Habit
Habit1	I am addicted to using …	0.897	0.877
Habit2	I must use …	0.935
Habit3	I automatically use … without thinking much about it.	0.855
Price Value
PP1	… is good value for money.	0.953	0.959
PP2	For the current price, … provides good value.	0.968
PP3	… is reasonably priced.	0.962

Table A3. Composite reliability and average variance extracted.

Construct	Composite Reliability	Average Variance Extracted
Content Actuality	0.936	0.785
Content Originality	0.911	0.774
Content Quality	0.960	0.618
Content Relevance	0.956	0.844
Content Range	0.959	0.855
Effort Expectancy	0.944	0.809
Habit	0.924	0.803
Brand Perception	0.928	0.812
Price Value	0.973	0.924
System Quality	0.953	0.628
Actual Usage	0.893	0.735
System Reliability	0.942	0.802
Usability	0.941	0.801
Word of Mouth	0.939	0.837

Table A4. Discriminant validity through the square root of average variance extracted (AVE on diagonal).

Construct	(1)	(2)	(3)	(4)	(5)	(6)	(7)	(8)	(9)	(10)	(11)	(12)
Content Actuality (1)	0.886
Content Exclusivity (2)	0.622	0.880
Content Relevance (3)	0.717	0.571	0.919
Content Range (4)	0.742	0.607	0.702	0.925
Effort Expectancy (5)	0.440	0.399	0.441	0.413	0.900
Habit (6)	0.305	0.271	0.449	0.347	0.146	0.896
Brand (7)	0.506	0.454	0.531	0.554	0.311	0.485	0.901
Price Value (8)	0.514	0.415	0.52	0.525	0.404	0.322	0.494	0.961
Actual Usage (9)	0.386	0.286	0.535	0.405	0.289	0.534	0.388	0.340	0.858
Technical System Reliability (10)	0.608	0.539	0.524	0.617	0.627	0.230	0.495	0.546	0.328	0.895
Usability (11)	0.619	0.522	0.569	0.628	0.618	0.326	0.546	0.510	0.344	0.772	0.895
Word of Mouth (12)	0.472	0.385	0.590	0.481	0.298	0.595	0.629	0.470	0.454	0.405	0.502	0.915

Table A5. Heterotrait-monotrait (HTMT) ratio of correlations.

Construct	(1)	(2)	(3)	(4)	(5)	(6)	(7)	(8)	(9)	(10)	(11)	(12)	(13)
Content Actuality (1)
Content Exclusivity (2)	0.706
Content Quality (3)	0.963	0.864
Content Relevance (4)	0.774	0.636	0.925
Content Range (5)	0.801	0.677	0.941	0.744
Effort Expectancy (6)	0.48	0.451	0.521	0.472	0.443
Habit (7)	0.339	0.313	0.438	0.497	0.379	0.162
Brand (8)	0.562	0.52	0.645	0.582	0.605	0.342	0.549
Price Value (9)	0.549	0.457	0.599	0.545	0.55	0.429	0.349	0.533
System Quality (10)	0.677	0.613	0.725	0.611	0.661	0.907	0.289	0.555	0.577
Actual Usage (11)	0.444	0.339	0.532	0.603	0.457	0.33	0.628	0.451	0.377	0.407
Technical System Reliability (12)	0.665	0.609	0.705	0.561	0.662	0.682	0.254	0.547	0.581	0.971	0.372
Usability (13)	0.678	0.591	0.725	0.612	0.675	0.672	0.361	0.604	0.543	0.967	0.393	0.841
Word of Mouth (14)	0.519	0.437	0.606	0.642	0.519	0.325	0.668	0.704	0.501	0.489	0.522	0.442	0.55

Table A6. Correlations on the level of the constructs.

Construct	(1)	(2)	(3)	(4)	(5)	(6)	(7)	(8)
Content Quality (1)	1
System Quality (2)	0.658	1
Habit (3)	0.409	0.272	1
Price Value (4)	0.577	0.555	0.322	1
Actual Usage (5)	0.483	0.534	0.534	0.340	1
Brand (6)	0.597	0.520	0.485	0.494	0.388	1
Word of Mouth (7)	0.570	0.463	0.595	0.470	0.454	0.629	1
Marker Variable (8)	0.123	0.129	0.073	0.136	0.081	0.123	0.100	1