Effectiveness of branded article content
The persuasion knowledge mechanism in online native advertising. Which role is played by the age of the audience and its familiarity with the brand?
by Tommaso Pettenazzo 1
Master Thesis Marketing Management: EBM867B20.2019-2020 Supervisor: dr. J.C. Hoekstra
Co-assessor: prof. dr. J. van Doorn University of Groningen
Faculty of Economics and Business Nettelbosje 2
9747 AE, Groningen, The Netherlands, Date of submission: Jan uary 13 , 2019
Abstract
Despite the growing popularity of native advertising, applied in the news consumption market and, despite the academic relevance given to this type of advertising, literature has shown a gap in the analysis of which type of audience characteristics influence its effectiveness.
A consistent amount of literature, indeed, has mainly analyzed the sponsorship disclosures efficacy and the consumer’s reactions to native ads based on specific age groups. On the contrary, the present study strives to challenge the persuasion knowledge model (PKM), which states how persuasion expertise is gained by consumers over time (hence getting older). The path we are embarking here, aims to support, on one side, how persuasion in the online environment is higher when it comes to younger readers, as literature recently disclosed. On the other side, the study addresses how age can strengthen the persuasion knowledge activation when the sponsoring brand is more prominent, and weaken the negative attitude toward the advertised article raised by the persuasion knowledge.
Results of the experimental study (N=199) showed, indeed, how the more prominent brand was effective in stimulating a higher persuasion knowledge. Nonetheless, the role of age as moderator was not significant. Finally, the familiarity of the brand, expected to mitigate as well the negative attitude toward the advertised article, didn’t show any relevant consumer reaction, in opposite to brand attitude, taken as well into account and demonstrating, indeed, significa nt influence.
Keywords: audience age, brand familiarity, brand prominence, online native advertising, persuasion knowledge
Acknowledgements
Table of Contents
1. INTRODUCTION……….………. 1
2. THEORETICAL FRAMEWORK ………. 5
2.1 Conceptual model………...………. 5
2.2 Literature background and hypothesis……… 6
2.2.1 Brand prominence and persuasion knowledge……… 6
2.2.2 Attitude toward the advertising content……… 8
2.2.3The moderation of audience age on brand prominence……… 9
2.2.4 The moderation of educational level……….. 10
2.2.5 The moderation of audience age on persuasion knowledge activation…... 11
2.2.6 The moderation of Brand familiarity………. 12
3. METHODOLOGY ………..………... 15
3.1 Type of research and study design………... 15
3.1.1 Manipulation of brand prominence……… 15
3.1.2 Manipulation of Brand familiarity………. 17
3.2 Data collection………. 17 3.3 Procedures……… 18 3.3.1 Pretest procedure………... 18 3.3.2 Study procedure………. 19 3.4 Data measurement……… 20 3.4.1 Control variables……… 22
3.5 Factor and reliability analysis……….. 23
3.6 Plan of analysis……… 24
3.7 Data adjustment and sample descriptives……… 25
3.7.1 Data adjustment………. 25
3.7.2 Sample descriptives……… 26
3.8 Manipulations and confounding checks……….. 26
3.8.1 Manipulation checks………. 27 3.8.2 Confounds checks……….. 27 4. EMPIRICAL ANALYSIS ………. 29 4.1 Correlation analysis………. 29 4.2 Hypothesis testing……… 30 4.2.1 Analysis of Covariance H1……….. 30 4.2.2 Regression analysis H2……… 31 4.2.3 Moderation analysis H3, H4, H5……… 33
4.2.4 Moderation analysis H6 and H7………. 34
1 INTRODUCTION
Advertising is ubiquitous and omnifarious in its way to persuade and seduce consumers into buying a particular product or service, donating for charitable causes, voting on elections, and so on. Advertising stimuli literally bombard us. An American consumer is estimated to be exposed on average to more than 1,000 commercial stimuli every day (Fennis,and Stroebe, 2015). In this era of advertising clutter where companies fight to capture the attention of the audience, it's becoming crucial to employ new salient techniques to achieve that goal. In 2017 online advertising became the world's most diffused advertising medium, accounting for more than 37% of the expenditures and it is expected to consistently grow above this value in the future (Harms, Bijmolt, and Hoekstra, 2019). However, online advertising alone doesn't circumvent the consumers' tendency to prevent their exposure to such stimuli through, for instance, ad-blocking technologies or much more naturally through the scrolling of the content to avoid banner advertising. In particular, younger generations show familiarity with these technologies: an analysis reported by Miller et al. (2017) showed that 63% of Millennials block ads when viewing digital content. It's for this reason that forms of subtle advertising are emerging, driving the companies to invest increasing amounts of their budget in online native advertising (Vranica, 2016). The peculiarity of native advertising is that "it takes the form and the appearance of the platform where it is embedded" (Wojdynski, and Evans, 2016). In this way, it intends to gain the consumers' engagement through the fruition of content in many different forms and platforms (social media, search engines, blogs, editorial content called article-style native advertising) (Harms et al. 2019).
consumers will capture the persuasive aim of its communication, and the more effective it will result (McCarty, 2004). Nevertheless, the application of such a technique increasingly captures the attention of policymakers as well as of publishers. Through their websites, they spread the branded article contents with the risk of jeopardizing their credibility.
Even if the data collected by eMarketer in 2015 documented a relevant 10% increase in budget spending on brand-content advertising as well as its effectiveness is well documented in literature, the circumstances under which it operates and the mix of elements that can make it more or less effective are still underresearched (e.g. Harms et al., 2017). In particular a consistent part of literature is limited to consumer's reaction to native ads and the efficacy of the sponsorship disclosures (Campbell and Marks, 2015; Lee, Kim, and Ham 2016; Sweetser et al. 2016; Wojdynski, 2016; Wojdynski and Evans 2016, An et al., 2019). However, little is known about how the different characteristics (for instant demographics) of the consumers that form the target audience of the native advertisements, shape how they react to them.
An interesting study, conducted by Amazen and Wojdynski (2019), demonstrated how age was a direct antecedent of the persuasion knowledge activation. Hence younger generations could differently react to this gentle type of advertising at least in the online environment. That is, the persuasion knowledge model (Friestad and Wright, 1994) would be at least somewhat reverted. In fact, contrary to what was previously established by the PKM, in which Friestad and Wright (1994) sustained how consumers acquire knowledge about persuasion tactics over time and therefore with age, younger audience, because of their familiarity with internet and its fruit io n of content, would acknowledge better the marketers' tactics. This actual interpretation, if confirmed in the future literature, is an important contribution for policymakers and consumers' organizations heavily involved in denoting the deceptive advertisement and in touting out the role of its sponsorship disclosures (Campbell et al., 2012; An, Kang and Koo, 2019). Among these entities, the federal trade commission (FTC) enacted a body of guidelines (FTC 2015), overstating the importance of the sponsorship disclosure (among which the visibility of the brand paying the content advertised).
consumers as 'persuaders'. Equal attention, for similar reasons, is given by the publishers, who profit from the advertisements income gained, although risking to jeopardize their credibility. A relevant extent of literature, which examines the different reactions to the type of online advertising, found it easier to recognize its persuasive intent when it comes to banners rather than native advertisements. They are generally considered more prominent communicated (e.g. Harms et al., 2019; Tutaj and Van Reijmersdal, 2012). Wei, Fisher and Main (2008), embedding a brand name inside a non-profit radio program, found out as it enhanced a higher recognition of persuasive intents, leading to a more negative evaluation of the stimulus. At the same time, the authors showed how the familiarity of the brand employed, had a moderating effect on this mechanism, attenuating the negative assessment of the article. Other studies showed how consumers exposed to unfamiliar brands are more likely to respond negatively to persuasion attempts. Consumers' little knowledge about unknown brands can in fact foster the activation of persuasion knowledge leading to a critical evaluation of the validity of their messages (Campbell and Keller, 2003; MacKenzie and Spreng, 1992). In general, the activat io n of persuasion knowledge, which stimulates scepticism with a consequent critical evaluat io n and less-favourable attitude toward the message advertised, is a largely corroborated path in the literature (Friestad and Wright 1994, 1995; Oates, Blades, and Gunter 2002).
The present study investigates whether the brand located in a higher prominence inside the article foments similar persuasion activation and negative attitude toward the message as in the case of banners advertising. And acknowledging the results obtained in the audio branded context examines if the brand familiarity moderating role is confirmed even in the editoria l context.
Gilbert et al. 1990, 1993). Hence the younger consumers, heavy internet users, are more prompt to accept the claim with automatic processing instead of elaborating deeper and counterargui ng it.
2 THEORETICAL FRAMEWORK
2.1 Conceptual model
Figure 1 presents the conceptual framework. It is argued that the level of brand prominenc e should negatively influence the attitude towards the advertising content because of the enhancement of persuasion knowledge activation, which will act as a mediator in the mentioned relationship.
Figure 1: Conceptual model
2.2 Literature background and hypothesis
2.2.1 Brand prominence and persuasion knowledge
The article-style native advertising, as formally is defined in the field of branded content, relies its success on different content and context factors, for example, the publisher’s credibilit y source (Wojdynsky and Golan, 2016). The extent to which a brand is present in it, and then how much awareness customers have about reading a paid content is likely the most influe nt driver of its effectiveness (e.g. Wojdynski and Evans, 2016). The online native advertising, in fact, for its main characteristic of being embedded in the editorial source, with which sometime s share even the style and the font used, is hardly recognizable by the readers (An, Kang, and Koo, 2019). That constitutes, at the same time, one of the most significant factors of its increasing success (Goss, 2015). The typical persuasive goal of the advertising message is therefore concealed, making it difficult for consumers to cope with, as Friestad and Wright (1994) genuinely explained in the persuasion knowledge model (PKM). The persuasion knowledge theory among scholars (e.g. Friestad and Wright, 1994; Campbell and Kirmani, 2000) is believed to be the principal source to which consumers rely on in replying to persuasion techniques. It will represent even in this study the most meaningful literatur e contribution, on which we will rely.
consumer to understand the camouflaged persuasive nature of the content. Hence without the disclosure, it would not be feasible for readers to have a clear recognition of ads and non-ads content, making it difficult to critically evaluate the message and its persuasive attempts (An et al., 2019). The disclosure mechanism that we wish to employ in the present study is ‘brand prominence’. This term was first introduced by Han, Nunes, and Dreze (2010) to identify a construct that could represent the conspicuousness of a brand on a product or in our case the article content. In the qualitative research conducted by Harms et al. (2017), 22 brand managers and online marketing practitioners from media and publishing companies were asked to evaluate the importance of this concept in native digital advertising. The results underlined how this variable was declared almost unanimously central in the native advertising field, but the optimal level of brand prominence is still unclear. In fact, on the one hand, the higher the visibility of the brand, the more likely the persuasion knowledge of readers should be activated, leading to the defence mechanism activation which in turn, in research, is observed in consumers’ resistance to the advertising claim (Fransen et al., 2015; Jacks and Cameron, 2003). On the other hand, low prominence, if the real intent of the content has been noted by the readers, can shed light on the misleading technique and can be evaluated in an even more harmful way (Thota et al. 2012).
In literature it was shown how respondents found 'less irritating and more amusing' the branded content in contrast to the banner ad (Tutaj and Van Reijmersdal, 2012; Harms et al. 2019). In fact, in the banner advertising is thought to activate more easily the persuasion knowledge, because its persuasive intent and its source are more accessible (Nebenzahl and Jaffe, 1998). We can then suppose that a similar result will be obtained in the comparison between two different forms of native advertising articles with a different extent of brand prominenc e. In other words, the more prominent the brand sponsoring the content, the easier it will result for the consumer to recognize the editorial content as an advertisement, hence activat ing persuasion knowledge.
H1: Brand prominence positively influences persuasion knowledge activation
2.2.2 Attitude toward the advertising content
grasp the persuasive intent of native advertising. For this reason, several consumer protection associations and policy companies have expressed their worrying, among which the Federal Trade Commission (FTC 2015) which stated very clearly that the media industry should carry clear and prominent disclosures informing audiences that the content is an advertisement. Many academic research articles though indicate that in the vast majority of news, consumers are deceived by this new form of advertisement (Amazeen and Muddiman, 2018; Amazeen and Wojdynski, 2018; Wojdynski and Evans, 2016). If consumers acknowledge the persuasive intent of the editorial content, a sort of change of meaning will occur (Friestad and Wright, 1994), which alters the manner they will respond to that stimulus and subsequent ones. In other words, a defence mechanism will be activated in the consumer mind, and this will happen faster the more the paid nature of the content will be perceived. In fact, in the same study, Friestad and Wright (1994) number how consumers’ recognition of the persuasive direction of the attempt (in this case editorial content) can stimulate scepticism or critical attitudes toward the message. This critical evaluation leads to less favourable attitudes toward the message and its credibility (Moore and Rodger 2005; Oates, Blades, and Gunter 2002). Different studies have shown how recognition of hidden advertising resulted in negative attitudes toward the brand and the sponsored content (e.g. Boerman, Van Reijmersdal, and Neijens 2014, 2015; Wojdynski, 2016).
It has been shown that consumers who detect persuasion attempts in a native ad will react with several coping behaviours to respond to the episode (Campbell andKirmani, 2008; Friestadand Wright, 1994; Jacks and Cameron, 2003; Nelson and Ham, 2012). That happens because consumers, at first sight, consider the content as impartial information rather than biased on the advertiser side. Although they ‘cope’ (i.e. to contend or strive in Friestad and Wright vocabulary) with the real aim of the communication, it evokes a sort of defence biased systematic processing such as “encountering” or negative perceptions (Harms et al. 2019). This mechanism leads to negative judgments of the advertisement, which undermines its credibilit y and evaluation (Dark and Ritchie, 2007).
Cameron, 2003; McGuire, 1985) or challenging the accuracy or validity of a position (Mille r and Baron, 1973). In the present context, counterarguing may imply claims in the news article , resisting the article format, or complaining about a publisher or the brand involved in sponsored content. In the cognitive-response tradition, how individuals elaborate on a message predicts whether persuasion or resistance occurs. Cognitive elaborations that are positive generate persuasion while those that are negatively valenced generate resistance (Petty and Cacioppo, 1977; Petty, Cacioppo, and Goldman 1981). For example, in the context of sponsored blogs, audience members who were aware of the sponsored content in the blog they were reading had higher self-reported levels of cognitive resistance (Van Reijmersdal et al., 2016). Van Reijmersdal and colleagues (2016) speculate that recognition of advertising via a disclosure may both activate persuasion knowledge as well as foster counterarguing. This is consistent with inoculation research that indicates how the forewarning process itself motivates the generation of counterarguments (Ivanov et al., 2015; Petty &Cacioppo, 1977), which negatively affect the evaluation of the stimulus. Nonetheless, it is possible that following the recognition of native advertising, news consumers may respond with a cognitive elaboration involving counterarguing or in general a negative attitude against the format, the publisher the brand sponsor and in particular the article content (Amazeen and Wojdynski, 2019).
Based on the preceding, we hypothesize that:
H2: Persuasion knowledge negatively influences attitude towards the advertising content.
2.2.3 The moderation of audience age on brand prominence
which the closest antecedent to the article-style native advertising was the ‘advertorial’ (indeed defined in Van Reijmersdal et al. (2005) as a mixture of editorial content with advertising). In the actual contest of online native advertising, marked by the online social environme nt, influencers, blogs, multimedia contents, little is known about the dynamics of the aforesaid process of knowledge acquisition. The Net generation or better known as Millennials (1976-1994) and even more generation Z (the individuals from 1995) have been around computers since before they could speak (Tapscott, 2009). For them, “technology is like air,” necessary but invisible. These generations are more familiar with the usage of computers and novel media communication in which the persuasive mechanism is regularly cloaked (Conill, 2016). They have been targeted with massive advertising and commercials since they were born; as a result, this generation is quite suspicious of all marketing campaigns (Qader and Omar, 2013). Therefore, since not so many consumers acknowledge the native digital advertising, and the younger audiences consume more of their news on the web (Mitchell et al., 2016, eMarketer 2017), we can expect that at least somewhat of the established mechanism of persuasion knowledge acquisition will be reverted in terms of age, as it was previously found in the study of Amazen and Wojdynski (2019). However, instead of investigating just the direct effect of the audience age on the persuasion knowledge activation, which we will acknowledge from the previous literature, the actual study intends to capture the potential moderator effect of the audience age. Hence, the more highlighted is the sponsorship indication, in this case, the presence of the brand inside the article, the more the respondents will capture the persuasive intent (as established in hypothesis 1), but we add that the younger they are, the easier it will be the disclaimer. Based on afore discussed argumentation, we hypothesize:
H3: Persuasion knowledge activation will be more likely for people who are younger.
H4: Audience age will weaken the positive effect of brand prominence on persuasion knowledge
activation
2.2.4 The moderation of educational level
spend, on average, 70 minutes per day with the news (Mitchell et al., 2016). However, it is proportional to the education level obtained, passing from 58 minutes of high school education level to 96 minutes of postgraduate level respondents (Miller et al., 2017). Therefore, we expect that respondents with a higher educational level will present a higher ability to analyze the article and capture its persuasive intent even if older in age.
H5 A higher educational level will weaken the audience age effect (hypothesis 4) on the
relationship between brand prominence and persuasion knowledge activation.
2.2.5 The moderation of audience age on persuasion knowledge activation
In the development of the hypotheses mentioned above about the role of the age (i.e. H3 and H4), a new recently observed path was argued. In fact, at least concerning the capability to recognize the advertising nature of a message, the age is thought to have an inverse correlated relationship with it. Hence the younger a reader it is, the easier the disclaimer would be. Nevertheless, once the persuasion knowledge is stimulated a considerable number of studies sustain as younger consumers have a better attitude toward advertising than older adults. Moschis and Churchill (1979), for instance, reported the rise and progression of resistance attitudes in early adolescent, the same concept appears in Boush, Friestad and Rose (1994), where is defined as a sceptical schema of attitudes that increase with age. The scepticism of advertising relation with age is also reported in Obermiller and Spangenberg (1998), where a positive and significant moderate correlation was found. It has been shown that younger consumers reported a more positive attitude toward advertising messages compared with older generations, for instance, they communicated being less offended by advertising, 40% (aged between 18 and 34), compared with 57-58% of older adults (Shavitt et al., 1998).
effortless and unconscious processing is, therefore, more likely to happen when it comes to navigating the internet (Guadagno et al., 2013).
Acknowledging that the youngs have major usage of internet with respect to the older adults as encountered in Statista report of 2019 (more than seven and half hours a day between mobile and computer for 18-24 aged people compared with four and half hours for 55-64 years old adults), we can expect that this automatic heuristical processing, is more pronounced in the younger audience. Part of the research has synthesized this web heuristics in a unique construct, the ‘if it comes from the computer it must be true’ heuristic (Guadagno et al., 2013). In synthesis, in situations of chronic cognitive overload, frequent with multitasking and permanently online young users, it is harder to counterarguing and disbelieving the claim instead of accepting it (Gilbert et al., 1990, 1993). It found a theoretical explanation and reinforcement is another online consumer psychological framework: the Google effect (Sparrow et al., 2011, 2013) A phenomenon that explains how new generations are restructuring the declarative memory system ('what' is knowledge) in favour of a more procedural one ('how' to knowledge). This transactive memory system results in a lack of prior memory base in order to elaborate and then accept the information processed, and therefore it is likely to enhance the online truth effect previously described.
In conclusion, acknowledging the aforementioned literature, the following is hypothesized:
H6: Audience age will strengthethe negative effect of persuasion knowledge on attitude toward
the content of advertising.
2.2.6 The moderation of Brand familiarity
Numerous studies on persuasion knowledge have taken place without acknowledging the effect that brand familiarity can lead. That is, therefore, one of the aims of the current study to investigate such an effect.
Hutchinson's study (1987), ‘‘reflects the extent of a consumer’s direct and indirect experience with a brand’’. This statement founds different foundations in the literature, one of which that well explained the mechanism behind is the subjective ‘brand knowledge structures’. It numbers how consumers are characterized by a determined knowledge about the differe nt brands, which make it easier to recognize and process the brands that are familiar (i.e. hence well known). It indicates a more limited, weaker and less accessible knowledge in consumers’ minds (Campbell and Keller, 2003). Lee and Labroo (2004, 2006) examined in their studies the application of the ‘hedonic fluency model’ in the processing of the brands and its implicatio n on awareness, affect judgment, and choice. Hedonic fluency model states that the subjective ease with which a stimulus is perceived and processed. can be used as an information base to evaluate the stimuli themselves (Winkielman et al., 2003).. In fact, the fluency signal is hedonically marked and manifests itself as a positive affective reaction whereas the fluency is high rather than low, leading to evaluate in a qualitative way stimuli that at first are presented as neutral. Hence stimuli that are encountered before are processed and encoded more easily, resulting in positive emotion, a function of the previous stimuli, but which is misattributed to the focal stimulus, as the ad or brand (Fennis and Stroebe, 2015). Therefore familiar brands, evolving a pleasant feeling in being processed, enjoy more cognitive and affective advantages (Lee and Labroo, 2004; 2006) and they are usually better liked (Kent and Allen, 1994; Dahle´n and Lange, 2004).
Moreover, Campbell and Keller (2003) showed how consumers are more likely to respond negatively to persuasion attempts of unfamiliar brands because they have little knowledge about them. It can temper the activation of persuasion knowledge. In fact, for brands with low familiarity, consumers are more likely to be concerned about the validity of their statements and then more likely to express critical evaluation (MacKenzie and Spreng, 1992).
evaluation (Machleit and Wilson, 1988), in the advertising interference (Kent and Allen, 1994), in delaying the advertising ‘wear-out’ phenomenon (Campbell and Keller, 2003).
Based on the literature mentioned above, the following is expected:
H7: Brand familiarity weakens the effect of persuasion knowledge on attitude towards ad
3. METHODOLOGY
This chapter describes the methodology applied to test the hypothesis. A description of the design and sample will be firstly presented respectively in sections 3.1 and 3.2, followed by the procedures (the pretest and main analysis) in section 3.3.The measurement methods and the factor and reliability analysis will be eventually defined in sections 3.4 and 3.5, followed by the plan of analysis (section 3.6).
3.1 Type of research and study design
In order to study the hypothesis described before, an experimental study was conducted as it is considered the best option to test causality (Malhotra, 2006). The experimental study with the random assignment of participants, more than the other types of researches, allowed us to have enough control over the situation: the manipulations (the independent variable brand prominence and the moderator brand familiarity) while keeping everything else constant (Aronson et al. 1998).
The design consists of a factorial 2 (brand prominence: high/low) x 2 (brand un/familiar ) between-subject experiment. The structure, therefore, resulted in four different conditions that were randomly assigned to the 199 respondents, as stated in Table 3.1 below (stimuli in Appendix 1):
Table 3.1: Experimental conditions (in parenthesis the number of respondents)
2x2 between-subject design High brand prominence Low brand prominence
Familiar brand CONDITION 1(N= 55) CONDITION 2 (N= 44)
Fictitious brand CONDITION 3 (N= 49) CONDITION 4 (N= 51)
3.1.1 Manipulation of brand prominence
terms of which kind of indicators, if language or sponsored content indication, are more effective and then should be adopted. To enhance reality, in this study, we took into consideration the recommendation given by the FTC’s 2015, "Disclosures guidelines" , because it is more likely that it will be applied by the practitioners. It states that the indication, in the form for instance of the logo or the brand name, sufficiently visible, should be positioned on the top of the content. Therefore the different stimuli were all provided with a disclosure at the top of the article stating the name of the brand in the subsequent indication: ‘Sponsored by Brand X’. Nevertheless, in the high prominent conditions this brand name was provided in the form of a colourful logo, instead of the simple brand name (low prominent condition).
On top of that first differentiation, the brand prominence manipulation of high versus low was assured in other two different ways:
- Firstly in the articles characterized by high brand prominence, the picture of the article was accompanied by the visible brand logo (just under the picture). This disclosure was at the centre of the article content inasmuch Wojndynski and Evans (2016) in their study about different types of disclosure, using eye-tracking technology, found out as the middle centre position of the statement was more effective in attracts the greatest visua l attention.
- Secondly, the brand name was repeated 3 times inside the text in the high prominenc e conditions; instead in the low ones, the brand name was replaced by means of general words as ‘a collaboration of clothing producers’ (row 4 of the article), ‘collaboration of these producers’ (row 10) and ‘many companies’ (row 17).
The other parts of the article were exactly the same for all the conditions, indeed identical to the original article written by Adam Sass; it was just reduced in length to facilitate the reading. A resume of the manipulation is represented in Table 3.2 but a direct view of the stimuli is showed in Appendix 1.
The effectiveness of the manipulation was firstly assured through the usage of a pretest that has proved if it effectively works. Here participants were randomly provided with two differe nt articles, but the same content, one characterized by high prominence and the other one by low prominence and they were asked to answer a single question (yes/no): ‘Did you encounter any
brand inside the text?’. The same question was then repeated in the primary survey to ensure
Table 3.2: Manipulations adopted for prominence conditions
High prominence conditions Low prominence conditions
Indication “Sponsored by” Yes Yes
Colourful logo Yes No (just name of the brand)
Repetition of the brand name inside the article
Yes Row 4 Row 10 Row 17
No. Replaced by the words: “a collaboration of clothing producers”
“collaboration of these producers” “many companies”
3.1.2 Manipulation of Brand familiarity
The second manipulation regarding the familiarity of the brand is obtained through the usage of one real and one fictitious brand. H&M, a well-known brand in the fashion industry, which different rankings evidence its popularity: Forbes, Financial Times, Interbrand and ‘Festalia’ which logo was created thanks to Logomaster.ai (figures 2 and 3). As it was done for the brand prominence, a pretest will be performed to ensure that the familiarity is significantly different between the two brands. They will be asked to evaluate two statements: ‘I know this brand very well" and "This brand is very familiar to me’ on a 7 point Likert scale (Churchill, 1979).
Figure 3.1: Real familiar brand Figure 3.2: Fictitious brand
3.2 Data collection
et al. 1998). Moreover, it enhanced external validity, counting that in 2015 thirty-eight of the forty digital- native news publishers received more unique visitors via mobile devices than from desktops (eMarketer 2017).
The interviewees were provided with no additional information besides the fact that they were contributing to the development of a marketing management master thesis. Nevertheless, they were told that they had the chance to win a 20€ voucher expendable on Amazon.com as an incentive alignment (mechanism widely applied in conjoint analysis experiments) to prompt people to participate in the survey and reveal their real preferences (Ding et al. 2005, Ding, 2007).
3.3 Procedures
3.3.1 Pretest procedure
Before starting with the primary analysis, a pre-test was performed to verify if the manipulations implemented worked and hence they could be used in the main survey. The manipulations described in section 3.1 were organized in a 2x2 between-subjects pre-test experiment, same as the research design. To develop the pre-test mentioned above a sample of 44 respondents were asked through the usage of a link shared in Whatsapp.com to participate in a minimal survey. In fact, besides the reading of the same article presented in the main questionnaire, participants were asked to answer just the questions about the manipulatio ns checks, before stated (section 3.1) and some demographics. In the data collection, two requirements have complied with the selection of respondents: Firstly, to gather the most heterogeneous sample possible in line with the one that would represent the main analysis base. This requirement was achieved: the frequency table of demographics showed a heterogeneous sample (age between 20 and 63 years old, 24 were males and from 10 different nationalities ). Secondly, the data were collected in a different way with respect to the primary study data collection in order to minimize the risk that participants were randomly selected again. Therefore, the researcher made sure that the survey link was shared with different people, hence the usage of personal messages in Whatsapp.
means for high vs low prominence conditions were significantly different as it could be presumed by their absolute value of 1 and 1.8 respectively (the responses were coded 1-2 if the brand was noticed or not). Therefore it was assured that the prominence manipulation described in section 3.1 could be employed in the main questionnaire. The same implication was obtained after the analysis of the t-test brand familiarity. Before proceeding with it, a single variable measuring the familiarity with the brand was calculated on the average of the two items adopted for the brand familiarity check (items presented in section 3.1), because a significant correlatio n between them was observed (r = .971, p < .01). The results showed in this case as well how the F calculated was higher than the critical value, and therefore null hypothesis about equal variances was rejected (F = 27,53, p < 0,05). The t-value was also significant, stating how the means (absolute value of 6.5 and 2.64 for familiar and unfamiliar conditions respective ly, measured on a 7 Likert scale) were significantly different from each other (t = 6.928, df = 24,825, p < 0.05). Hence the familiarity manipulation worked, and it was implemented in the primary survey as well.
3.3.2 Study procedure
At the beginning of the survey, the respondents were presented with a page of instructio ns where it was indicated as the data collected would remain completely anonymous and they would be used only for the academic purposes of developing the master thesis. Mention of native advertising wasn’t made to not affect the advertising recognition and consequently the potential distortion of the dependent variable measure (attitude towards ad content) and the mediator (persuasion knowledge activation): hence we believe that differences in the responses given should be genuine. Moreover, it was mentioned that they could withdraw from it at any time by simply closing the tab in their web browser.
Thirdly, respondents were asked to evaluate three statements about the persuasive intent of the advertising content and their attitude toward the article content according to the measuring scales presented in the subsequent section (section 3.4).
In the fourth part, it was asked first to investigate whether the participant had paid attention in reading the article (attentional check): ‘Do you remember what the article was about?’: The respondents had to choose between three options, among which one only was correct (Recycling for a better and sustainable future of fashion business). The respondents, who failed this check, were filtered out from the database used for analysis (section 3.7.1). Afterwards, manipulation checks were performed. The brand prominence was checked through the following question (yes/no): ‘Did you encounter any brand inside the article you have
read?’. Then the statements concerning the brand familiarity measures: "I know this brand very well" and "This brand is very familiar to me" were proposed in order to obtain a brand
familiarity dimension and control if the manipulation worked even in the main survey. Along with these, the respondents’ brand attitude was secured through the evaluation of five bipolar adjectives as a mean to control it (see table 3.3, section 3.4).
In the fifth and last part, some demographics were collected: age, gender, nationality, education level attained, digital literacy (web experience) and English proficiency. In particular age and education level were used to measure the respective main variables presented in the conceptual model.
3.4 Data measurement
Table 3.3: Scale measurements with respective factor loadings and Cronbach’s alpha
Variable Source Items Factor
loading
Cronbach’s alpha Activation persuasion
knowledge
Harms et al. (2019) 3 items, 1-7 Likert scale
-‘The aim of this text is to sell products or services’ (selling)
- ‘The aim of this text is to influence your opinion’ (persuading)
-‘The aim of this text is to give information about products or services’ (informing)
.749 .779 .694 .600 Attitude towards ad content Amazeen and
Wojdynski (2019) 7-point semantic differential scale: good/bad favourable/unfavourable unappealing/appealing unlik able/likable unpleasant/pleasant .797 .833 .871 .870 .838 .915
Brand familiarity Churchill (1979) 2 items, 1-5 Likert scale: "I k now this brand very well" "This brand is very familiar to me"
.964 .960
.939
Age How old are you?
Educational level attended (highest level attended)
Measured on 8 levels:
- Compulsory education (primary and secondary school)
- Secondary vocational education (MBO) - Bachelor (HBO/University of applied science)
- Bachelor (WO/ Research University) - Pre-Master - Master - PhD - Other Brand attitude (control variable)
Spears and Singh (2004)
7 point semantic differential scale: unlik eable/likable unfavourable/favourable unpleasant/pleasant bad/good appealing/unappealing .892 .893 .929 .905 .859 .953 Internet literacy (control variable)
Tutaj and Van Reijmersdal (2012)
Average hours/week, calculated in this way: ‘How many days per week on average do you use the Internet?’
‘How many hours per day on a week day (Monday–Friday) do you use the Internet?’ ‘How many hours per day on a weekend day (Saturday and Sunday) do you use the Internet?’
3.4.1 Control variables
Several variables were measured to control for alternative explanations.
Internet literacy (web experience)
The literacy competencies needed to scrutinize an article content are a crucial mean in order to disclose the persuasive intent of the native advertisements as it was aforementioned in the literature review. In the digital context, the literacy relates to more than general confidence in handling screen texts, the “study of written or symbolic representation that is mediated by new technology” (Merchant, 2007). One would then expect persuasion knowledge to be positive ly correlated with web experience (Mallinckrodt and Mizerski, 2007). However, a fair amount of literature shows how the web experience of an individual is correlated with its age (Van Deursen and Van Dijk, 2009; Noh, 2017), hence the younger, the greater knowledge gained about navigating the web, because of frequent usage amongst the younger population (van den Eijnden et al., 2008). Therefore the researcher decided to control internet literacy (WEB_LTCY) to reduce the risk of confounding elements. To measure said literacy, we employed the same items used in Tutaj and Van Reijmersdal (2012), who controlled it in their study as well (reported in Table 3.3 above). An average number of hours per week was then calculated to represent the participant’s internet literacy (M = 23.68, SD = 11.56).
Brand attitude
Brand attitude (B_ATT) was controlled because of the presence of the brand familiar it y variable. Although the literature is controversial about the mentioned relationship depending on whether familiar rather than unfamiliar brands are employed (Machleit and Wilson, 1988), there is strong evidence in a considerable amount of studies of the relationship between brand attitude and attitude towards advertising (Machleit and Wilson, 1988; Batra and Ray, 1985; Gresham and Shimp, 1985; Edell and Burke, 1986; Messmer, 1979). The brand attitude (α= 0,953; M= 4.57, SD= 1.17) was estimated using Spears and Singh’s (2004) measurement scale consisting of five bipolar adjectives (unlikeable-likeable, unfavourable-favourable,
unpleasant-pleasant, bad-good, appealing-unappealing) assessed on a 7 Likert scale. English proficiency
manipulations, the level of English proficiency was taken into account. It could potentiall y affect the dependent variable (attitude towards the advertising content). In fact, relating to the hedonic fluency model, the ease of processing the article for an individual with a high knowledge of English, could be transferred to a positive evaluation of the stimulus itself (i.e. affect-based explanation), compared with an individual with a low level of English
proficiency (ENG_KNOW) (Winkielman et al., 2003). The English knowledge (M = 72.83, SD
= 19.90) was self-reported measured with an interval scale of 1-100.
Demographics (gender and nationality)
On top of the previous control variables, respondents were also asked a couple of questions pertaining to specific demographics described here below.
Firstly, findings of gender differences on beliefs and attitudes towards web advertising are somewhat confounding. Wolin and Korgaonkar (2003), for instance, showed how males exhibit more positive beliefs and attitudes towards web advertising, and they surf the web more frequently for functional and entertainment purposes. Other literature contributions pointed out how males are more selective information processors than females, and they are more likely to use heuristics processing and miss subtle cues, for instance, sponsorship disclosures (Darley and Smith, 1995). Therefore the researcher decided to take into account gender (GENDER) as a potential confounding element.
Secondly, the nationality (COUNTRY_ORIGIN) was controlled to ensure a better understanding of potential cultural differences in respondents’ answers.
3.5 Factor and reliability analysis
the amount of variance a variable shares with the others of the construct, and it should be higher than 0.40 (Malhotra et al., 2006), which ensures the goodness of the factor analysis applicatio n. Accordingly, to the procedures so far described, the KMO obtained from the items used in attitude towards advertising content, persuasion knowledge activation, brand familiarity, and brand attitude showed positive values (KMO = 0.810; Barlett’s test significant (p<0.01); and communalities all higher than 0.4). Furthermore, the engine values (total variance explained by each factor) of the principal components method analysis indicate as four components were extracted (engine value > 1.0, Guttman-Kaiser rule). However, since IBM SPSS Statistics bases the extraction only on engine values, other criteria were taken into account. The inspection of the screen plot (presence of the main elbow in the subsequent component of the ones extracted) and the variance explained by each component (it needs to be higher than 5%) supported the engine values assumption2. Although the cumulat ive percentage variance (selection of the component where 60% is met) states the extraction of 3 components, three out of four of the criteria examined confirmed the four components extraction. In fact, looking at the factor loadings obtained through the orthogonal rotation of varimax procedure, no cross-loadings were found, as it can be observed in Table 3.3. Each item fits well with just one component, acknowledging that just values over 0.3 are considered high and indeed acceptable (Malhotra et al., 2006).
Finally, the reliability analysis (Cronbach’s alpha) was considered to ensure the presence of the internal consistency in the measurement scales. A factor is considered sufficiently reliable for alpha higher than 0.60 (Malhotra et al., 2006). Almost all the constructs highlighted excellent values of Cronbach’s alpha: attitude towards advertising content (0.915), brand
familiarity (0.939), brand attitude (0.953), apart from persuasion knowledge activation that
showed a value on the edge (0.60) but still sufficient (see Table 3.3).
3.6 Plan for analysis
The table below (table 3.4) will give the readers a brief summary of which analysis will be conducted in the following chapter for the hypothesis explained in section 2.2. The programs SPSS Statistics 26 and Hayes Process 3.4 were adopted.
Table 3.4 Type of Analysis and estimated models for each hypothesis
3.7 Data adjustment and sample descriptives
3.7.1 Data adjustment
First, only respondents that finished the survey (N= 199) were moved in the dataset in IMB SPSS Statistics 26 for the analysis, so the further exclusion of uncompleted responses was not necessary. In addition, no missing values were founded, thanks to the 'force responses' setting of the survey in Qualtrics. Second, to prepare the collected dataset for analysis, the researcher excluded the respondents who failed the attentional check question, presented in section 3.3.2 (N=7). Third, people under 18 years old (N=1), were excluded because not wholly competent to express judgments (Cauffman and Steinberg, 2000). Fourth, the researcher probed the dataset for outliers. Outliers employing box plot method in the relevant variables of the model were not founded. However, the researcher examined even the time spent to complete the survey and the self-reported English proficiency because of possible accounts of inattention and misunderstanding of the asked questions (for instance respondents who entered the questionnaire and finished it later on). For these variables was applied the interquartile range of 2.2, as found in Hoaglin and Iglewicz (1987) the more accurate way to detect outliers in normal distributions, which were assumed given the large sample size (Khan and Rayner, 2003). Therefore people who completed the survey in more than 18 minutes and reported an English knowledge under level 14 (range 1-100, table 3.3) were excluded (N=28). As a consequence, a binary filter variable was applied (FILTER), whereas people who met the
H1 Analysis of Covariance AAC = Attitude
toward advertising content BP = Brand prominence PKA = Persuasion knowledge activation AGE = Audience age EL = Educational level attended BF = Brand familiarity
ε = error term
H2 Regression analysis AAC = β0 +β1PKA + ε H3, H4 and
H5
Regression analysis (model 3 of Hayes
Process)
PKA = β0 +β1BP + β2AGE + Β3EL + β4AGE*BP + β5EL*AGE + β6EL*BP + β7EL*AGE*BP + ε
H6 and H7 Regression analysis (model 2 of Hayes
Process)
conditions mentioned above were coded as '1' and the rest with '0'. The final sample size consists of 163 (199-7-1-28) individuals who were used for further analyses.
3.7.2 Sample descriptives
Regarding the respondents' demographics, the sample was made up of slightly more males than females (50.9%, 49.9%), age included between 19 and 70 with an average of 33 years (SD=13.19) among which 58.3% under 30 years old; 18.4% between 30 and 40 years, and the rest (27.6%) over 40 years old. In terms of nationality, 123 (75.5%) were Italians, the others from 20 different countries with no significant distribution percentages, hence were categorized as 'other' (24.5%).
Concerning the highest educational level attained 95 individuals indicated a master's degree (58.3%), which already states alone as the sample is not heterogeneously distributed on this variable and it would likely lead to bias in its analysis. Therefore, the educational level attended was recategorized in three groups: high educational level, coded as '1' (master, pre-master, 'PhD and other specializations', 115 respondents); medium educational level, coded as '2' (Bachelo r MBO and WO, 29 respondents), low educational level coded with '3' (compulsory or secondary vocational education, 19 respondents).
Moving to the Web experience, the sample was very heterogeneously distributed, passing from 1 hour per week on average spent on internet to 49 hours, with a mean of 23.68 hours and a standard deviation of 11.56.
In terms of English proficiency, 110 (67.5%) people reported a level of English knowledge greater or equal to 70, the other 40 individua ls (24,5%) a level between 40 and 70 and the remained 8% of the respondents under 40. Thus, with the vast majority of the sample displaying a high-moderated level of English, it is assumed how they probably understood well what was presented in the survey.
3.8 Manipulations and confounding checks
1982). Indeed, to see whether the random assignment was successful an ANOVA analysis was conducted for each control variable as the dependent variable and the four conditions of the manipulation as between-subject factors (Tutaj and van Reijmersdal, 2012) (see section 3.8.2). The categorical control variable gender (coded '1' male, '2' female) and the nationality ('1' Italian, '2' other) were investigated through the non-parametric test Chi-square, given the impossibility to use ANOVA for nominal or ordinal variables as dependent measures. Finally, to find out whether the control variables could be considered covariates of the dependent variable and then been included in the primary analysis, a linear regression per each control variable on the 'conceptual' dependent variable was investigated.
3.8.1 Manipulation checks
The manipulations checks through the application of independent sample t-test in case of brand familiarity conditions, and Chi-square test in case of brand prominence (indeed brand prominence was measure in a dichotomous variable: yes/no; see section 3.3.2) showed how both manipulations worked. The Leven’s test for equality of variance presented a non-significant F value (F = .138, p > 0,05); hence the null hypothesis of equal variances assumed was met. The t-value showed a significant result (t = 6.928, df = 24,825, p < 0.05), stating how the means (absolute values of 6.5 and 2.64 for familiar and unfamiliar conditions respective ly, measured on a 7 Likert scale) were significantly different from each other.
The same implication was obtained for the prominence conditions. In the high prominent case 64 of the 84 respondents noted the brand advertising the article, in the low prominent one, as expected, only 29% of the respondents noted it. Finally, the Chi-square statistic gave us confirmation that the proportions were statistically significantly different from each other (ϰ2 = 36.255, p < .01).
With the aforementioned results, it was ascertained what was already found in the pretest; that is, the manipulations were effective.
3.8.2 Confounds checks
As stated before a 2-way ANOVA (2 factors with two levels: brand prominence high- low and brand familiarity high-low) was performed with each continuous control variables (brand
attitude, internet literacy, English proficiency) as dependent measures and the conditions brand
= 0.26 for brand prominence and F = 1.003, p = 0.318 for brand familiarity). Still, in the case of Internet literacy (F = 6.194, p = 0.14 for brand prominence and F = 2.929, p = 0.089 for brand familiarity) and English proficiency (F = 5.161, p = 0.024 for brand prominence and F = 0.650, p = 0.421 for brand familiarity) a significant effect on brand prominence was found. However, since the assumption of homogeneity of variances was violated for Brand
attitude and English proficiency, the independent samples t-test was run per each factor since
it gives the possibility to interpret the t-value for equal variances not/assumed. Brand
attitude showed comparable non-significant results for both manipulations, but English proficiency showed an ‘equal variances assumed’ t-value significant for brand prominence (t =
2.373, p = 0.019). The overall conclusion of the results above, suggest to include the control variables English proficiency and Internet literacy in further analysis.
Gender and nationality (recoded in section 3.7.2), because of nominal dichotomous variables, were analyzed employing the Chi-square test of homogeneity. It was run four times, given the four associations between gender and nationality with brand prominence and brand familiarit y. The results demonstrated no significant difference in proportions between levels of conditions and levels of the dependent variables (gender and nationality): brand prominence - gender (ϰ2 = .059, p = .809); brand prominence - nationality (ϰ2 = 4.180, p = .051); brand familiarity - gender (ϰ2 = .745, p = .388); brand familiarity -nationality (ϰ2 = .102, p = .749).
The effect of control variables on the dependent variable was also investigated. With this purpose, a linear regression was performed with the continuous control variables as independent variables and attitude towards advertising content as the dependent one. In the case of gender and nationality, because nominal scales, an ANOVA was performed. The results showed as only brand attitude had a significant outcome on the dependent variable (β = 0.413, p = 0.000) and therefore, it was included as covariant in the main analysis. The other control variables: Internet literacy ( β = -0.057, p > 0.05), English proficiency (β = 0.028, p > 0.05), Gender (F = 1967, p > 0.05), Nationality (F = 0.530, p > 0.05) were not significant and hence not taken into account.
4 EMPIRICAL ANALYSIS
4.1 Correlation analysis
A Pearson correlation analysis was first conducted to get familiar with the relations between the main variables of the conceptual model. However, educational level and brand prominenc e were not included inasmuch nominal variables. The control variables’ correlations were also included.
The only significant correlation between the main variables was found between age and persuasion knowledge activation (r = -.177, p = .024) that was negative. Even if moderately low, according to Cohen (1988), it allows the researcher to make some inference s about the developed hypothesis (i.e. H3). The negative correlation found in fact could imply a potential predictive relation, although it’s good to remember that correlation doesn’t mean causation. The sign of the association implies that the relationship is inversely proportiona l, indicating a decrease in the age would express an increase of persuasion understanding, precisely what was sustained in the literature review (chapter 2).
The other significant correlations found regarded the control variables. Brand attitude, for instance, had a moderately strong positive significant correlation with advertising content attitude (r = .412, p = .000), English proficiency a negative moderate correlation with age (r = -.371, p = .000) and a positive one with the other confound variable Internet literacy (r = .316, p = .000), which in turn had a negative moderate correlation with age (r = -.297, p = .000), as it was found in literature (section 2.2).
Table 4.1: Correlation matrix between the main variables of the model. * Correlations significant at p< .05 Persuasion
knowledge
Advertising
4.2 Hypothesis testing
To test the hypothesis described in the literature review (chapter 2), IBM SPSS Statistics 26 was employed. As we mentioned in the plan for analyses (section 3.7), the primary analyses accomplished were Analysis of Covariance, Moderated regression, and simple regression, and moderated Mediation.
4.2.1 Analysis of Covariance H1
Figure 4.1: Synthetic model representation and finding for hypothesis 1.
The effect of the brand prominence manipulation (low vs high) on the mediator persuasion
knowledge activation was carried out through the implementation of one-way ANCOVA,
proficiency and internet literacy with brand prominence (F(1,10.138) = 6,922. p = 0.009,
F(1,7.248) = 4,949, p = 0.028). Therefore we need to be careful with the results found.
Acknowledging the aforesaid assumptions and after controlling for the covariates English
proficiency, internet literacy, and brand attitude, it was revealed a significant difference in the
means of persuasion knowledge between the two brand prominence categories (F(1,9.488) = 6.234, p = 0.014). In particular the respective mean values of persuasion knowledge activat io n for high prominence condition was: M = 4.393, SD = 1.261; and for the low brand prominenc e one was: M = 3.194, SD = 1.251.
Overall we can conclude that the first hypothesis was validated and respondents who were assigned to the condition of high prominence reported a higher perception of persuasive intent and therefore persuasion knowledge activation.
4.2.2 Regression analysis H2
A multiple regression analysis was conducted to test whether persuasion knowledge activation leads to a lower attitude toward the content presented. The covariates English proficiency, internet literacy, and brand attitude were also included (subsequently named independent variables as well).
Figure 4.2: Synthetic model representation and finding for hypothesis 2.
Qq plot of the studentized residuals showed values not always close to the line, the robustness of linear regression to normality and large sample employed gave enough assurance to continue with the analysis. The last important assumption regarded the multicollinearity analysis. Multicollinearity is the potential distortionary correlation between the independent variables. The VIF values, in the coefficients table (Table 4.2), all lower than 4, assured low levels of multicollinearity (Malhotra et al., 2006).
The overall model fit, observed through the R Square estimate (0.174 with an adjusted R Square of 0.153) reported the variance explained by the model, which was relatively low (Cohen, 1988). However, the F-test ANOVA, which indicates the comparison of the variance explained with the unexplained one, was still significant (F(4,158)=8.331, p=0.000).
From the table of coefficients below reported (table 4.2), it’s shown as no significant effect (β = .031, t=.488, p=.626) of the persuasion knowledge activation on the attitude towards advertising was found. Therefore the H2 stating that a certain level of persuasion knowledge activation would have led to a higher / lower attitude toward the content presented was not supported. However, the control variable brand attitude showed a significant predictive value on attitude towards advertising content (β =.377, t=.502, p<.001).
Table 4.2: Regression Results (unstandardized coefficients) for Attitude toward advertising content
Coefficientβ Standard
error p-value VIF
4.2.3 Moderation analysis H3, H4, H5
To test the moderation effects of the continuous variable Audience age and its interaction with the polytomous educational level attended (high, medium and low) on the relationship between brand prominence and persuasion knowledge activation, model 3 of Process Hayes was implemented. In addition to the aforenamed variables, covariates English proficiency, brand attitude, and internet literacy were included in the analysis.
Thanks to bootstrap methods integrated into Hayes Process, the normal distribution of the variables was not required (Haukoos and Lewis, 2005).
Model 3 is composed of a single independent variable (brand prominence), a single dependent variable, in this case, persuasion knowledge activation, and two hierarchical moderators: Age and, the moderator of the moderation, educational level, that represents the third- way interaction indeed. The sub-conceptual model is here represented to give a clearer image of the analysis.
Figure 4.3:Synthetic model representation and finding for hypothesis 3,4,and 5.
was confuted even with a higher degree of freedom left, the analysis was repeated without including educational level. In this case, model 1 of the Hayes Process 3.4 was applied. Nevertheless, besides the confirmed effectiveness of the main effects of age and brand prominence, the results showed again as the interaction effect stated in H3 was not significa nt (β = .024, t= 1.486, p= .139).
Table 4.3:Moderation analysis results (unstandardized coefficients) for Persuasion knowledge activation Coefficient β Standard error p-value VIF Constant 2,850 .557 .000 Main Variables Brand prominence (BP) .564 .201 .006 1.122
Audience age (Age) -.021 .008 .013 1.355
Educational level attended (Ed. Level)
-.016 .155 .920 1.277
BP * Age .016 .016 .332 1.217
Ed. Level * Age -.009 .010 .388 1.532
Ed. Level * Age * BP -.002 .021 .906 1.571
Control variables Brand attitude .119 .084 .157 1.052 English proficiency .008 .006 .137 1.343 Internet literacy .006 .009 .528 1.190 R2 .147 F-value 2.623 .006
4.2.4Moderation analysis H6 and H7
This part of the analysis dealt with testing the moderation effects of Age (hypothesis 6) and brand familiarity (hypothesis 7) on the relation between persuasion knowledge
activation and advertising content attitude, as the graph below represents (figure 4.4). To this
intent, model 2 of Hayes process 3.4 was employed. In addition to the afore named variables of the hypotheses, the covariates’ English proficiency, internet literacy, and brand
Figure 4.4:Synthetic model representation and finding for hypothesis 6, and 7.
The model summary showed how it was explained almost 18% of the overall variance (R^2= .179, F(8,154)= 4.187, p= .000). There were not found any kind of significant interact io n effects between the variables: age*persuasion_knowledge (β= .001, t= .345, p= .731), brand_familiarity*persuasion_knowledge (β = -.023, t= -.554, p= .580). Looking at the main effects persuasion knowledge as already investigated in the correlation matrix and H2 didn’t show any significant effect on the dependent variable (β = .022, t=.317, p= .752). The same insignificant main effects were encountered for age and brand familiarity (β = .000, t= .028, p= .978; β = -.028, t= .587, p= .558 respectively). The only significant coefficient was disclosed in the analysis regarded the predictive effect of the control variable brand attitude, that shows a highly significant positive value, in line with what was previously observed in the correlatio n analysis (β = .388, t = 5.484, p < .001).
Table 4.4: Moderation analysis results (unstandardized coefficients) for attitude toward advertising Coefficient
β
Standard
error p-value VIF
Constant 3,920 .463 .000
Main Variables
Persuasion knowledge activation
(PK) .022 .069 .752 1.197
Audience age (Age) .000 .007 .978 1.312
Brand familiarity (BF) -.029 .049 .558 1.064
PK * Age .001 .004 .731 1.159
PK * BF -.023 .042 .580 1.138
