Do eye have your attention?
A large eye tracking study of comparing the effects of print and online
advertising on brand recall through increased attention
by
DEMI TESS WIJNGAARDEN
Do eye have your attention?
A large eye tracking study of comparing the effects of print and online
advertising on brand recall through increased attention
by
DEMI TESS WIJNGAARDEN
University of Groningen
Faculty of Economics and Business
MSc Marketing Intelligence and MSc Marketing Management
Master Thesis
13-01-2020
Nieuwe Kijk in ‘t Jatstraat 7 9712 SB Groningen
+316 36309399
d.t.wijngaarden@student.rug.nl S3522075
1st Supervisor 2nd Supervisor External Supervisor
Dr. Ir. M.J. Gijsenberg Prof. Dr. T.H.A. Bijmolt D. Hoogervorst
m.j.gijsenberg@rug.nl t.h.a.bijmolt@rug.nl dennis.hoogervorst@sanoma.com
Nettelbosje 2 Nettelbosje 2 Capellalaan 65
DUI332 DUI329 Sanoma Media Netherlands B.V.
9747 AE, Groningen 9747 AE, Groningen 2132 JL, Hoofddorp
Summary
This research investigates the effectiveness of display advertising and branded content in the print and online advertising medium, in terms of their influence on attention and brand recall. This study is based on a largely performed eye tracking study, on behalf of the media company Sanoma Media Netherlands B.V., which I will refer to as ‘Sanoma’ in this paper. This led to a unique dataset of 1,252 panelists with a total of 15,452 observations. After a strict validation, this study uses a sample of 5,655 observations of 605 panelists.
Currently, the digitalization impacts the magazine industry: advertisers under-consider print advertising, because of the increasing popularity of the online medium and the high price perception of print. In addition, many marketers have a hard time justifying the expenditures of advertising with branding related goals in both the print and online medium. This paper sheds a new light on four different advertising instruments, which are print display advertising, print branded content, online display advertising, and online branded content. The goal of this study is to assess the effectiveness of these four advertising instruments in order to move from a pricing strategy based on reach and views towards a focus on the advertising effectiveness. This is an important development in media planning for both media companies and advertisers. In this study, an advertisement is effective if it leads to brand recall, which is a core branding KPI.
The results show that the influence of the advertising media on unaided recall are fully indirect through attention for all advertising instruments except for print display advertising, which also shows a direct influence. In contrast to prior research, the advertising media show a remaining direct effect on aided recall, which means that display advertising and branded content in the print and online medium influence aided recall both directly and indirectly through attention. This effect is larger in the print medium, which is likely a result of the tactile feel of print which leads to better information processing. Since the effect of the advertising media on unaided recall is fully indirect through attention, attention plays a larger role in unaided recall compared to aided recall.
In the online medium, almost half of the respondents did not devote attention to a display advertisement while it was viewable on screen. When excluding those respondents, the results show that the effect size of online display advertising on aided recall almost doubles. Thus, online display advertising can potentially lead to brand recall, however, consumers tend to scroll past digital banners, multitask, and are distracted when browsing online, which tremendously lowers its effectiveness. This stresses the importance of a new pricing strategy which focuses on advertising effectiveness rather than the advertisement’s reach and views.
Preface
In front of you lies my master thesis ‘Do eye have your attention?’. I have been working on this thesis from September 2019 to January 2020 in order to finish my masters Marketing Intelligence and Marketing Management at the University of Groningen. During this thesis, I combined my analytical skills with academic writing and strong managerial implications in order to assess the current advertising pricing strategy at Sanoma. I really enjoyed working on this thesis combined with my internship at Sanoma. I am happy that I could support Dennis Hoogervorst with his research into effective attention, which will be used by the sales department in the upcoming years. I am very satisfied with the final result of my thesis.
First of all, I would like to thank my supervisor Maarten Gijsenberg for the good feedback, collaboration, and for challenging me throughout the entire process of writing this master thesis. Second, I would like to thank my external supervisor from Sanoma, Dennis Hoogervorst, for providing me with the right business and market information and for supervising me during my internship. Third, thanks to Sergio Romijn from Braynz B.V. for the additional information about the eye tracking study. Fourth, special thanks to David Bassett from Lumen Research Ltd. for his help and the explanation of the raw eye tracking data set. Finally, I would like to thank Tammo Bijmolt in advance for reading and evaluating my thesis.
I hope you will enjoy reading my thesis.
Demi Tess Wijngaarden January 2020
Table of Contents
Summary ... 3 Preface ... 5 Table of Contents ... 6 1. Introduction ... 8 2. Theory ... 11 2.1 Magazine advertising ... 112.2 The under-consideration of print magazine advertising... 11
2.3 The indirect effect of the advertising medium on brand recall... 13
2.3.1 Capturing attention ... 13
2.3.2 The role of the advertising medium in capturing attention ... 13
2.3.3 The role of attention in brand recall ... 15
2.3.4 Brand recall in advertising ... 17
2.4 The direct effect of the advertising medium on brand recall ... 18
2.5 The role of the different types of advertising formats in the relationship between the advertising medium and brand recall ... 19
2.5.1 Display advertising ... 19
2.5.2 Branded content ... 20
2.6 Conceptual model ... 22
3. Data and measurements ... 23
3.1 Data collection ... 23
3.1.1 Setting ... 23
3.1.2 Eye tracking data ... 23
5. Results ... 37
5.1 Preliminary model checks ... 37
5.2 Model-free results ... 38
5.3 Mediation analyses ... 39
5.3.1 Branded content ... 41
5.3.2 Display advertising ... 42
5.3.3 Comparison of branded content versus display advertising ... 44
6. Business simulation ... 45
6.1 Advertising effectiveness ... 45
6.2 Practical example ... 46
6.2.1 Current pricing strategy ... 46
6.2.2 Effective reach... 47
6.2.3 Price for effective reach ... 48
6.2.4 New pricing strategy ... 49
7. Discussion ... 52
7.1 Scientific contributions ... 55
7.2 Managerial implications ... 56
7.3 Limitations and recommendations for future research ... 58
References... 60
Appendix... 69
Appendix I: Questionnaire ... 69
Appendix II: Correlation matrices ... 72
Appendix III: Multicollinearity statistics ... 73
Appendix IV: ANOVA-like tables for random effects ... 74
Appendix V: ANOVA tables for fixed effects ... 76
Appendix VI: Bootstrapped confidence intervals ... 78
Appendix VII: Bootstrapped confidence intervals: testing the significance of difference ... 83
Appendix VIII: R-code ... 88
1. Introduction
Print magazines have been an important advertising medium for a long time now. While several studies show that the total print magazine circulation is declining (Breitenbach, 2018; Stuart-Turner, 2018), Zwinkels and Tielman (2019) show that although TV and radio channels are having a hard time, print magazines appear to be still popular. This is supported by the NOM (2019b), who shows that 75% of the Dutch population still reads magazines. Besides, consumers stayed loyal to magazines in 2018 (Magazine Media Associatie, 2019b) and the range of print consumer magazines were still stable in 2019 (Magazine Media Associatie, 2019a). This suggests not a decline, but a stable circulation of magazines in the past two years.
The trend towards digitalization is growing and impacts the magazine industry: one third of the Dutch consumers consume magazine content online (Detrixhe, 2019; NOM, 2018), 35% of the 20-34 years old read magazine content via their smartphone (NOM, 2018), and 15% of the Dutch consumers use a magazine app (GfK, 2019). This trend affects magazine advertising, because the composition of consumer magazine advertising will change. Online consumer magazine advertising is expected to grow and surpass print consumer magazine advertising by 2022 (Marketing Charts, 2019). The magazine industry is able to react to the changing environment by offering electronic print magazines, a magazine app, a website for each brand, presence on social media, and media and e-commerce events next to the print magazines. So, good news for the magazine industry: it is not dead, it is just changing. While first the advertising budget had to be split across TV, radio, out-of-home and print, it now has to be split over way more channels, including many digital channels (Carey, 2017). This provides new challenges for the magazine market.
Currently, firms are under increasing pressure to justify their marketing expenditures (Van Heerde, Gijsenberg, Dekimpe, & Steenkamp, 2013). Compared to online advertising, it is harder for print advertising to directly measure the performance in terms of sales. Therefore, advertisers often perceive print advertising as rather expensive (Verdurmen, Hannema, & Hoogervorst, 2019), and under-consider the use of print advertisements (Myers, 2019). However, the overall sales effects of advertising are low, therefore, advertising is more useful for branding related goals (Rossiter, Percy, & Bergkvist, 2018). Branding effects are not only hard to measure in the print medium, but in the online medium as well. This is the case, because one cannot directly determine branding effects based on data, thus, researchers need to test these effects by questioning consumers who are exposed to the advertisement (Verdurmen et al., 2019). This study provides more insights into the effectiveness of advertising in the print and online medium in terms of attention and brand recall, which are two core branding key performance indicators (KPIs).
This study focuses on print consumer magazines from Sanoma Media Netherlands B.V., which I will refer to as ‘Sanoma’ in this paper, which consumers can also read as an electronic print version. This study considers an electronic print version of a magazine as print, since it is a close replica and the advertisements do not provide the same opportunities as online advertisements do (Consterdine, 2012). This paper relates to several lines of research, in which the effect of the advertising medium (print versus online) on the ability of the consumer to recall the brand is investigated. One can expect this relationship to be fully indirect through the amount of attention consumers devote to the advertisement. This study analyses this effect for two types of advertising formats, namely display advertising and branded content.
This research is based on a largely performed eye tracking study of several magazines and websites from Sanoma. The dataset consists of 605 valid eye tracking sessions with in total 5,655 observations and provides insights into the amount of attention that consumers devote to several types of advertisements. Although attention is a required first step in advertising effectiveness, it not sufficient. Advertisements needs to leave a durable effect in memory in order for consumers to recall the advertised brand (Wedel & Pieters, 2000). Therefore, additional recall tests assess the branding effects of advertising on the consumer’s memory.
recently been performed into this branch. This study provides new insights into the differential effect of display advertising and branded content in the print and online medium on the attention for an advertisement and the ability to recall the brand. Media companies can use these findings to base their advertising prices on the effectiveness of the advertisement rather than its reach and views. This makes it better possible for marketers to estimate the effect of their marketing campaign and justify their advertising expenditures.
2. Theory
2.1 Magazine advertising
Advertising is one of the oldest and most important instruments of the marketing communication mix (De Pelsmacker, Geuens, & Van den Bergh, 2010) and can be defined as ‘any paid message that a firm delivers to consumers in order to make the offer more attractive to them’ (Tellis, 2004, p. 9). The focus of this study is on magazines that target the consumer market. Nowadays, consumers can also read their magazine on screen, as an electronic version of the print magazine (De Pelsmacker et al., 2010). This study refers to the electronic print magazine as print magazine, since it is a close replica of the printed magazine in content, with also relatively limited interactive features (Consterdine, 2012). Moreover, print and electronic print magazine advertisements receive the same amount of attention and result in similar recall rates (Consterdine, 2014; Galin, Baim, Slaten, & Becker, 2011).
The main goal of many magazine advertisers is branding, also called brand-building advertising. Rossiter, Percy and Bergkvist (2018) refer to this as advertisements that are placed in mass-media, corporate image advertising, or sponsorships, in which a branded company is advertised rather than specific brand-items. One can best use magazine advertising to maintain brand awareness and not to introduce a new brand into the market (Rossiter & Percy, 1997; Rossiter et al., 2018). Brand awareness is defined as ‘the buyer’s ability to identify (recognize and recall) the brand, within the category, in sufficient detail to make a purchase’ (Rossiter et al., 2018, p. 101). An important benefit of advertising in print magazines is that these have a longer shelf life than online website advertising. For example, a magazine may stay on a coffee table for months and consumers often pass their magazines on to friends (Sherman, 2019).
2.2 The under-consideration of print magazine advertising
therefore, one can view print as a niche platform (GroupM, 2019). In addition, Chandra and Kaiser (2014) show that online advertising does not lower the revenues of advertising in print magazines with a more homogeneous audience and that the effect of print and online advertising increases when targeting consumers through both media. Thus, despite the benefits of online media, print has not vanished (Carey, 2017; Holzinger et al., 2011).
Second, the online media makes it possible to better target consumers individually based on their interests. Each magazine reader gets to see the same advertisement, causing magazines as an offline media channel to have a lower target ability than online media channels (Bergemann & Bonatti, 2010). Therefore, print cannot target consumers the way online marketing campaigns can (Sherman, 2019). However, print magazine advertising maintains its value and overcomes this lower targeting ability, by reaching a carefully selected variety of demographic groups, lifestyles and interests (Kim et al., 2015). In addition, magazines provide the opportunity to advertisers to cater a niche audience (Sherman, 2019). Magazines offer relatively easy and affordable media access for advertisers; therefore, advertisers can easily reach their target audience by advertising in magazines (Kim et al., 2015). Besides, Rossiter, Percy, and Bergkvist (2018) mention that print magazine advertising is the recommended direct-response medium for products or services of which the target audience is well-defined by an occupational or other psychographic or demographic readership. Therefore, it seems that the lower ability to target consumers in the print medium does not apply to print magazine advertising.
Third, firms are under increasing pressure to justify their marketing expenditures (Van Heerde et al., 2013). Advertisers can justify their expenditures by maximizing key performance indicators such as reach and frequency of the advertisement (Paulson, Luo, & James, 2018). For online advertising, one can measure performance with click-through-rates, time on page, and purchases, but for print advertising this is not digitally registered and thus harder to measure. While sales effects are harder to measure in print, branding effects are difficult to measure in both the print and online advertising medium. Still, advertisers often perceive print advertising as more expensive (Verdurmen et al., 2019).
of print compared to online advertising on brand recall, which measures the effectiveness of the advertisement. This study analyses this effect for two different types of advertising formats: branded content and display advertising. I expect this relationship to be largely indirect through the effect of attention. Therefore, this study first explains the indirect relationship of advertising medium on brand recall through attention, followed by a review of the direct relationship and the advertising formats.
2.3 The indirect effect of the advertising medium on brand recall
2.3.1 Capturing attention
The importance of attention in advertising effectiveness already became clear more than 50 years ago. Diamond (1968) mentions that an advertisement can only affect a consumer if it attracts the consumer’s attention to the content of the advertisement. Attention is a big barrier to advertising effectiveness, because not all consumers exposed to an advertisement will actually pay attention to it (Rossiter et al., 2018). Besides, Lamme (2004) shows that attention can take several different forms, implying that consumers can process advertisements in different ways. One can see attention as a complex process linked with perception, memory and action (Diamond, 1968). Subconscious processes play a large role in this process, since consumers have a limited capacity for processing information (Styles, 2006). Szmigin and Piacentini (2018, p. 186) define attention as: ‘the process by which we devote mental activity to a stimulus’, in which the main characteristics are that attention is limited, selective, and can be divided. Styles (2006) explains that consumers analyze information pre-attentively for meaning, but do not process everything: only the most important signals break through their ‘selection filter’ and receive conscious attention.
2.3.2 The role of the advertising medium in capturing attention
In describing the role of the advertising medium in capturing attention, this study assesses two different types of advertising media: (electronic) print advertising and online (website) advertising. Print used to be the main medium for advertising in the past, however, one should acknowledge that online advertising also plays a large role nowadays. This creates a new challenge for publishers of print media (Cherian, 2015).
attention. Baron et al. (2017) shows that on-screen reading creates more convenience, but also creates more distraction and eyestrain. Besides, several studies show that consumers prefer print documents (Ackerman & Goldsmith, 2011; Holzinger et al., 2011; Kong, Seo, & Zhai, 2018; Kretzschmar et al., 2013), as they learn more (Green, Perera, Dance, & Myers, 2010; Ji, Michaels, & Waterman, 2014) and comprehend information better (Kong et al., 2018; Mangen, Walgermo, & Brønnick, 2013) when reading in print compared to the same information on-screen. However, these studies compare print and on-screen reading for academic textbooks. One cannot directly translate these findings to (electronic) magazine reading, because consumers read magazines for leisure purposes and this motivation stays the same, independent of the medium (Consterdine, 2014). Magazines provide consumers with an offline moment without stimuli and distraction (Daniel & Woody, 2013; Van Leeuwen & Detrixhe, 2019), suggesting that when reading a magazine, consumers are less influenced by external factors. Consterdine (2014) describes this ‘magazine bubble’ as attentive, thorough and characterized by repeated reading. Consumers tend to feel more relaxed when reading a magazine (Payne, Severn, & Dozier, 1988), causing them to be in a more ‘lean back’ mood (Baron et al., 2017; Bassett, 2017). In this lean back mood, consumers browse more casually for their own leisure, making them more receptive to see and receive brand messages (Bassett, 2017). Moreover, print magazines allow consumers to process information at their own sequence and pace, which enhances the opportunity to process information (Jacoby, Hoyer, & Zimmer, 1983). Research shows that the lean-back mood in magazine reading does not change when reading an electronic print instead of a hard copy (Green et al., 2010; Holzinger et al., 2011). Therefore, one can assume that reading a print or electronic print magazine has an identical influence on attention.
preference for print. In line with this finding, Myers (2019) argues that the ‘power of touch’ is the reason that print gets more attention, reaches higher levels of emotional connection and creates an unique experience for consumers. Accordingly, magazine reading is characterized by high primary attention and low multi-tasking (Consterdine, 2014). Therefore, it seems that the print medium has several advantages in capturing attention over the online medium.
To further highlight the difference in attention in the print and online advertising medium, several studies show that consumers spend approximately 40% (United States Postal Service, 2015), 70% (Simmons, 2012), and even 80% (Bassett, 2017) more time on viewing a print advertisement compared to an online advertisement. Besides, Consterdine (2014) found that advertising awareness is higher for magazine websites compared to non-magazine websites.
Although many organizations mainly focus on digital advertising nowadays, a big player in the market, P&G, recently cut $200 million of its digital advertising spending and reinvested this in other channels. This resulted in a 10% increase in their reach and an elimination of ineffective marketing of 20% (Johnson, 2018). Thus, it appears that advertising in the online medium is not always as successful as advertisers expect.
To sum up, print advertising seems to be better able to capture attention than digital advertising. Therefore, this study presumes that print advertising leads to more attention
than online advertising.
2.3.3 The role of attention in brand recall
In order for an advertisement to foster brand awareness, it must induce the consumer’s attention to the advertisement. This suggests that attention is an important first step for advertising to impact consumer behavior (Fennis & Stroebe, 2010; Rossiter et al., 2018). Consumers need repeated attentional responses in order to fully process an advertising message (Rossiter et al., 2018). However, attention is not sufficient: advertisements needs to leave a durable effect of brands in memory (Wedel & Pieters, 2000).
2005), because consumers do not need conscious attention to memorize an advertisement: it can influence judgment without the consumer remembering it (Fennis & Stroebe, 2010). This (unconscious) feature-analysis only influences future behavior and judgment if the advertisement’s features are available during judgment (Shapiro, 1999). Moreover, Tellis and Ambler (2007) mention that in order to encode information in memory, the consumer has to pay attention to the information. Aravindakshan and Naik (2011, p. 316) define the impact of consumer memory as ‘delaying the forgetting of advertisements’, which means that brand awareness as a result of advertising does not immediately disappear, but stays in the consumer’s memory for a while after advertising exposure. In addition, Weilbacher (2003) argues that advertising is an addition to everything someone has previously learned about the brand. Thus, after exposure to advertising, the information adds to the current brand associations. This is in line with Szmigin and Piacentini (2018), who mention that attention is influenced by mental concepts, like consumers’ goals, motivation, memory and expectations.
Visual attention is important to increase memory for objects (Pieters, Warlop, & Wedel, 2002; Simmonds, Bellman, Kennedy, Nenycz-Thiel, & Bogomolova, 2019), and determines what consumers perceive and process (Myers, Deitz, Huhmann, Jha, & Tatara, 2019). Therefore, the role of attention in recall is that it serves as a form of memory storage (Bunting & Cowan, 2005). Attention is especially needed for non-users and light users, since high brand users already have accessible favorable attitudes towards the brand (Simmonds et al., 2019). Myers et al. (2019) show that the features that consumers pay attention to are important indicators of what they remember, because attention for brand-identifying elements improves recall. Therefore, in order to increase brand recall, making brand elements attention-getting seems to be equally important as making them interesting. In addition, Leding (2019) shows that if consumers divide their attention, such as while multi-tasking, it reduces the overall recall rates.
likely that they can recognize parts of it. When an advertisement is consciously perceived, consumers can discriminate it as familiar (Holender, 1986), suggesting that especially conscious perception can be recalled. In addition, Claycamp and Liddy (1969) show that correct recall is a relevant measure of consumer responses. One of the reasons that advertising exposure can lead to brand recall, is the consumer’s affection towards the advertisement (Baumann, Hamin, & Chong, 2015). Consumers develop emotional responses during the processing of the advertisement (Rossiter et al., 2018), and can be developed quickly: consumers can determine in 0.3 seconds if they like an advertisement or not (Detrixhe, 2014).
To sum up, it appears that if consumers pay more attention to an advertisement, they are better able to recall the advertisement. Therefore, this study presumes that attention has a
positive influence on brand recall.
2.3.4 Brand recall in advertising
Actual buying behavior is the ultimate metric to measure advertising effectiveness. However, this is hard to measure for advertising with branding purposes, for both print and online advertising. Therefore, brand recall, which is a key branding KPI, is an interesting measure for the advertising effectiveness (Consterdine, 2014). Brand recall can result from brand building activities and has a large and positive effect on future sales growth (Binet & Field, 2017). Therefore, one can see brand recall as an indicator of buying behavior (Keller, 2013). In addition, Beerli and Santana (1999) show that recall is a strict measure of memory and therefore influences the consumer’s attitude (Rossiter & Percy, 2017) and decision making (Macdonald & Sharp, 2000). Consumers store the recall of a brand in their top-of-mind awareness (Leong, Ang, & Tham, 1996; Shamsollahi, Amirshahi, & Ghaffari, 2017). The top-of-mind awareness of a brand influences the presence or absence of that brand in the consumer’s consideration set. Hence, brand recall is very important in today’s competitive market (Shamsollahi et al., 2017).
matching the stored representation with the stimulus, therefore the question acts as a cue to retrieval from memory (Waller et al., 2004). Aided recall, also called prompted recall, is based on asking consumers to indicate which brands out of a list of brands they have seen (Zinkhan, 1982). This means that the researcher gives the subject a hint about the initial stimulus object (Claycamp & Liddy, 1969). An aided recall task compares new input with representations stored in the memory.
Advertising in magazines is especially useful for campaigns with brand recognition as goal and is a little less useful for free brand recall, since this has a high frequency requirement (Puto & Wells, 1984). However, the line between recall and recognition is not clear-cut: some recognition tasks involve recall and some recall tasks may be as automatic as recognition (Waller et al., 2004). Because of the similarities in the measures, Waller, McCaffery, and Wardle (2004) see close similarities between aided recall and recognition. The baseline performance-memory dimension of aided recall is characterized by brand awareness and is related to positive affective reactions (Zinkhan et al., 1986). In their study, Reinares-Lara and Martín-Santana (2019) found that aided recall shows higher recall levels than unaided recall, due to the greater difficulty of the unaided recall task. This is the case, because an unaided recall task requires stronger brand memory than an aided recall task. Moreover, aided recall has a greater effect on memory, for both high and low involved respondents (Beerli & Santana, 1999). Still, the two recall measures may be highly correlated (Beerli & Santana, 1999). Recall of well-known facts is not much improved by aided (vs unaided) recall, since recall of very familiar facts is so automatic as to be equivalent to a recognition task (Waller et al., 2004).
To sum up, it appears that consumers are better able to recall brands aided versus unaided. Therefore, this study presumes that attention has a stronger influence on aided recall than
on unaided recall.
2.4 The direct effect of the advertising medium on brand recall
advertisement is processed and recalled. Sundar, Narayan, Obregon, and Uppal (1998) show that print advertising performs better than online advertising, because print allows readers to process an advertisement entirely, whereas consumers have to scroll down to see and process and advertisement in the online medium. Besides, Dijkstra et al. (2005) mention that readers of print magazines are forced to notice an advertisement when they are turning over the pages, causing them to have higher cognitive responses to print advertisements compared to online advertisements.
To sum up, attention seems to be the underlying reason for the influence of the advertising medium on brand recall. Therefore, this study presumes that the relationship between the
advertising medium and brand recall is fully indirect through attention.
2.5 The role of the different types of advertising formats in the relationship between the advertising medium and brand recall
There are different advertising formats which advertisers can use, such as display advertising and branded content. One can use both formats in both the print and the online medium.
2.5.1 Display advertising
A display advertisement, also called press advertisement (print display) or banner advertisement (online display), is the most easily recognizable advertisement in a magazine (Macromark, 2019). One can place a display advertisement both in the print and online advertising medium. Tellis and Ambler (2007, p. 242) define a (online) banner advertisement as ‘a small, typically rectangular, graphic image that is linked to a target communication’. This advertisement is similar to a traditional print advertisement, since it also consists of text and an image. The biggest difference between print and online display advertising is that the online medium has the ability to navigate consumers directly to the advertiser’s website to convert (Tellis & Ambler, 2007).
editorial content in 0.1 seconds, which is 1/3rd of an eye fixation (Pieters & Wedel, 2012), and
they can judge an advertisement in 0.3 seconds (Detrixhe, 2014). Therefore, this dwell time is sufficient to have an effect. In a print display advertisement, the brand’s name and logo should be large, in color, and preferably accompanied by other visual content to hold the reader’s attention for at least 2 seconds (Rossiter et al., 2018). In his study, Bassett (2017) found that the attention (average dwell time) for print displays is much higher than for online displays. The main reason for this is that print display advertisements are about the same content of the magazine and are often bigger than online display advertisements.
Sanoma Media and Media Test (2012) show that 33% of the consumers can recall the message of a print display advertisement in a magazine, while 23% can do so for an online display advertisement on the website of the magazine. This suggests that display advertisements are better recalled in print versus online.
2.5.2 Branded content
Branded content, also called branded entertainment or advertorials (Consterdine, 2012; Macromark, 2019), is defined as ‘the integration of advertising into entertainment content, whereby brands are embedded into storylines of a firm, or another entertainment medium’. This involves co-creation and collaboration between entertainment, media and brands (Hudson & Hudson, 2006, p. 492). Therefore, one can see branded content as an atypical advertisement in comparison to a display advertisement. With branded content, the advertising content is written in the style of the magazine or website, with a small note on top that it is an advertisement to avoid misleading the readers (Consterdine, 2012). Hackley and Tiwsakul (2006) mention that this type of entertainment marketing inserts brands and brand references in entertainment vehicles, for example into the context and storyline of a magazine. Therefore, consumers often view this as part of the magazine content, causing them to not expect commercial messages (Um & Kim, 2014). When an advertisement is aligned with the editorial content, it can attract more attention (Bassett, 2017; Moorman, Neijens, & Smit, 2002), is easier to process and remember, generates greater recognition (Hermie et al., 2005; Moorman et al., 2002), creates more favorable attitudes towards the advertisement (Yoo, 2009), and positively influences advertising engagement (Wang, 2006).
branded content from editorial content (Pieters & Wedel, 2012). Consumers are able to recall more brands when they process the information content of an advertisement instead of simply processing the appearance (Shamsollahi et al., 2017). Consumers are more likely to process the complete information content of branded content, since they might think that it is editorial content. Hackley and Tiwsakul (2006) argue that branded entertainment unconsciously influences consumers’ associations with the advertised brand. Therefore, branded content can circumvent consumers’ persuasion knowledge (Um & Kim, 2014), which makes consumers more sensitive to be influenced by the brand message. Moreover, branded content can create assimilation effects, in which consumers judge the advertisement more positively because of the assimilation between the advertisement and the magazine content (Fennis & Stroebe, 2010).
NRC media assumes that if consumers read an entire article, which is often the case if consumers visit an article for at least 15 seconds, it establishes positive brand associations (Hafkamp, 2017). This suggests that branded content needs more attention than a display advertisement to reach the same level of brand awareness.
Sanoma Media and Media Test (2012) show that 44% of the consumers can recall the message of branded content in a print magazine, while 88% can do so for online branded content on the website of the magazine. This suggests that branded content is highly appreciated and recalled in both media, but even more on the magazine publishers’ website.
To sum up, consumers seem to devote more attention to and are better able to recall the brand in branded content compared to display advertisements. This study presumes that the
advertising format influences several relationships in the conceptual model, in which this study assesses the strength and direction of these effects as an empirical question without a
2.6 Conceptual model
The literature review leads to the following conceptual model:
Note: The a-, b-, and c’-path in the model are the traditional names of these paths in a mediation analysis, which shows the direction of the effect between the two variables that the path connects.
3. Data and measurements
This chapter discusses the data and measurements of this study. The first paragraph describes the data collection methods, which includes a description of the setting, the population, and the research design. In the second paragraph, I prepare the data for further analyses and give a description of the sample that this study uses. The final paragraph provides an overview of the measures and focal variables in this study.
3.1 Data collection
3.1.1 Setting
The goal of this study is to provide insights into the amount of attention that consumers devote to display advertising and branded content in the print and online advertising medium, in order to assess the effectiveness of the different advertising instruments. Currently, advertising prices are often based on the reach and views of an advertisement. However, media companies argue that it is more important to measure the effectiveness of an advertisement, rather than its reach and views (Binet & Field, 2017). In this study, an advertisement is effective if consumers can recall the brand, which I expect to be a result of the advertising medium and attention. Eye tracking studies have been performed to measure the amount of attention that consumers devote to display advertisements and branded content in the print and online advertising medium. The eye tracking study uses a descriptive and quantitative research design, which has the objective to describe characteristics and the role of attention in the relationship between the advertising medium and the ability to recall the advertised brand (Malhotra, 2010).
3.1.2 Eye tracking data
individuals pay attention to print advertisements, in an unobtrusive way (Fox, Krugman, Fletcher, & Fischer, 1998), which makes researchers certain of the time that consumers look at an advertisement, instead of only the time they spend at a page.
3.1.3 Procedure
This study uses several general and specific interest magazines and websites from Sanoma’s portfolio of brands. This study uses several webpages from ‘Autoweek’, ‘Libelle’, and ‘NU.nl’ to measure the visual attention for online advertising and uses the magazines ‘Libelle’, ‘Ouders van Nu’, and ‘vtwonen’ to measure the visual attention for print advertising. These print magazines and online websites are representative for Sanoma’s portfolio of brands (Hoogervorst, 2019). The eye tracking sessions in this study are controlled experiments and take approximately 10 minutes, which is too short to read a complete magazine. Asking respondents to read a whole magazine in this short period of time will result in unrealistic results. Besides, a longer eye tracking session will lead to too few observations on page level. Therefore, in order to make sure that all participants can read the same number of pages, this eye tracking study uses smaller editions of the magazines and a carefully selected collection of online webpages, which both include a combination of editorial content, branded content and display advertising. This study uses webcam eye tracking to measure eye movements, therefore, it uses electronic print magazines to measure the attention for print magazines.
3.2 Data sample
3.2.1 Data preparation
The raw dataset of this study consists of 15,452 observations of 1,252 panelists. In order to be able to use this data for further analyses, this study prepares the data. First of all, 40 participants did a complementary eye tracking session with Tobii eye tracking glasses instead of webcam eye tracking. This method tracks the eye movements of the participants for 10 minutes while they read several pages of a magazine. However, not all participants were able to read the same number of pages. Therefore, I exclude these 40 respondents and their observations from the dataset. This results in a sample of 1,212 panelists with a total of 14,892 observations.
One can regard an eye tracking session to be valid if the accuracy (distance to the truth in pixels) and precision (deviation around the mean in pixels) of the eye tracking sessions are smaller than or equal to 300 pixels, since this is about the size of a typical advertisement (Bassett, 2019). Furthermore, the measure of the gaze direction should be smaller than or equal to 0.1 seconds, which is the shortest possible fixation duration (Wass, Smith, & Johnson, 2013). A gaze smaller than or equal to 0.1 seconds means that the webcam is able to register all eye fixations (Bassett, 2019). Therefore, I exclude all observations with an accuracy and precision larger than 300 pixels and a gaze direction larger than 0.1 seconds from the dataset. This results in a sample of 617 panelists and 6,522 observations.
Missing values
The analysis of the missing values in the dataset shows a total of 591 missing values in the variables of interest. First of all, the dataset contains 9 missing values in the variable ‘brand’. These missing values are all in online medium and are a result of no visible and viewable seconds of the advertisement. This means that the panelist did visit the page, but the advertisement was not visible on screen. Possibly, the panelist did not scroll down on the page and thus did not see the advertisement. Therefore, I exclude these 9 observations from the dataset. This results in a dataset of 617 panelists and 6,513 observations.
2002). In this case, these 257 missing values are 3.95% of the data, therefore it does not lose statistical power. This exclusion results in a dataset of 607 panelists and 6,256 observations.
Third, the dataset contains 63 missing values in the variable ‘ad_dwell’, which shows the time in seconds that a panelist devotes attention to the advertisement. For these observations there is also no page dwell time available (time that the panelist looks at the page), but there is information about time on page present. By creating a dummy variable which shows if the value is missing (1) or not missing (0), this study can analyze if there is a pattern in the missing values. A linear regression with this dummy variable as independent variable and the recall measures as dependent variable determines if the missing values are Missing Not at Random (MNAR), Missing at Random (MAR) or Missing Completely at Random (MCAR). This analysis shows a significant result for both aided (β = 0.098, p = 0.004) and unaided recall (β = 0.135, p = 0.011), which suggests that there is a pattern in the missing values. Therefore, the missing values are at least not MCAR. In line with this, the Chi-Square tests between the dummy variable and all other relevant variables in the dataset show significant results (p < 0.001), which means that the missing values in the variable ad_dwell are dependent on the other variables in the dataset. To ensure the validity of this study, I exclude these 63 observations from the dataset, which results in a dataset of 6,193 observations from 606 panelists.
Finally, the dataset contains 4 missing values in the variables ‘gender’, ‘age’, and ‘income’. These four missing values are all missing in the observations of one panelist. This panelist did fill out a questionnaire, but did not write down any personal information. In order to efficiently run the models, I delete the observations of this panelist from the dataset. This results in a dataset of 6,189 observations and 605 panelists.
Observations of non-viewable advertisements
from the dataset, because the participants had the opportunity to see these advertisements, but they did not see or did not devote attention to the advertisement. This contains interesting information. This results in a dataset of 5,655 observations from in total 605 panelists. Table 1 shows the distribution of the panelists over the different conditions in this study.
Table 1: Participants, number of pages, and number of advertisements for each print and online medium condition
Print Online
Libelle Ouders van Nu
Vtwonen Libelle Autoweek Nu.nl
Total sampled population
194 206 249 180 196 187
Valid eye tracking sample
114 101 127 78 84 101
Pages in study 17 17 17 6 6 6
Advertisements in study 6 6 6 18 18 18
In their study, Orquin and Holmqvist (2018) show that a low number of eye tracking trials threatens the external validity of the study. This is not the case in this study, since the sample is, after a strict validation, still very large compared to other eye tracking studies.
3.2.2 Data description
Participants in the print medium condition immediately see an entire page. Therefore, the variables time_on_page, viewable_seconds, and visible_seconds are the same for these observations (M = 34.283, sd = 64.096). This is not the case in the online medium, since participants have to scroll down on the website to be able to see the entire page. In the online medium, the time_on_page (M = 40.080, sd = 273.82) is a lot higher than the advertisement’s visible (M = 20.026, sd = 19.18) and viewable seconds (M = 15.823, sd = 17.26). This means that in the online medium, consumers open a page for a longer time than that the advertisements are visible or viewable on screen. Participants in the print medium devote on average 7.780 seconds of attention to an advertisement, while this is on average 2.253 seconds for the online medium. For both advertising media, this is an average of branded content and display advertising. Table 2 shows the differences in ad dwell time for branded content and display advertising in the print and online medium. All values are the average number of seconds that consumers devote to an advertisement that was viewable on screen. This means that the consumer had the opportunity to see the ad in both advertising media. This is especially important for online advertising, because consumers may have to scroll down on the page to be able to see all advertisements.
Table 2: Average ad dwell time (in seconds) consumers look at the advertisements among different media and formats
Medium Print Online Online viewed ads
Format Mean (SD) Mean (SD) Mean (SD)
Branded Content 11.043 (14.723) 14.195 (15.093) 14.721 (15.116)
Display advertising 4.513 (4.607) 0.507 (1.227) 0.983 (1.566)
3.3 Measures
The ultimate outcome variable in this study is brand recall. This study measures brand recall based on two different recall measures, namely unaided and aided recall. The participants answer the unaided recall question first, followed by the aided recall question. In order to measure unaided recall, the participants answer the question ‘Which brands, if any, do you remember seeing advertising for on the pages you were just shown?’ This is an open question in which the participants can write down as many brands as they want. In this study, the variable ‘spont’ represents unaided recall. In order to measure aided recall, the participants answer the question: ‘Which of the following brands, if any, do you remember seeing advertising for on the pages you were just shown?’. This is a multiple-answer question, in which the participants see a list of the brands that were shown in the eye tracking study, a ‘none-of-these’ option and a maximum of seven red herrings, which are brands that were not shown in the study. The variable ‘prompt’ represents aided recall in this study. A binary variable indicates for both aided and unaided recall if the answer is 0 (FALSE, brand not recalled) or 1 (TRUE, brand recalled).
This study includes several variables of interest. The first focal variable is the advertising medium, which is either print or online. The variable ‘media’ represents the different advertising media, which indicates the print medium as ‘magazine’ and the online medium as ‘desktop’. This variable is a dummy variable, in which 0 is online and 1 is print.
screen, which indicates that this variable is a pure measure of the seconds that a consumer actually looks at the advertisement. One can interpret a longer dwell time as a result of more refixations (United States Postal Service, 2015), high design complexity, low feature complexity (Pieters et al., 2010), interestingness of the stimulus (Pieters & Wedel, 2012), or as a result of top-down or bottom-up processes (Lee & Ahn, 2012). Analyzing multiple eye movement metrics can threaten the statistical validity, since they tend to be highly correlated (Orquin & Holmqvist, 2018). Therefore, this study only uses the ad dwell time to measure the attention that a participant devotes to an advertisement. This variable is a continuous numeric (ratio scaled) variable.
The third focal variable in this study is the advertising format, which is either branded content or display advertising. The variable ‘ad_type’ represents the different advertising formats. There are three different page types in this dataset, which are branded content, Full Page (FP) ad, and editorial. In the print medium, branded content is always placed on a branded content page type and a display advertisement is always placed on a FP ad page type. In the online medium, branded content is always placed on a branded content page type, and a display advertisement is either placed on a branded content page or an editorial page. This means that a branded content page in the online medium contains both branded content and display advertisements. This is possible on an online website, because there is more space and consumers can scroll down on the page. This is not the case in the print medium, because in a magazine the space on a page is limited. This variable for the advertising format is a binary variable with two levels, in which 0 is branded content and 1 is display advertisement.
4. Methodology
This chapter discusses the methodology of this study, by explaining the model building steps, which are specification, estimation, validation, and use (Leeflang, Wieringa, Bijmolt, & Pauwels, 2015). First of all, I explain the specification of the model, followed by a description of the estimation of the model. This includes an explanation of the models and econometric methods used in this study. Next, I describe the validation criteria of the model. This study elaborates on the use of the model in the business simulation and discussion.
4.1 Specification
The analysis of this study consists of a mediation analysis, which is a decomposition of the total effect of the advertising medium on brand recall into an indirect effect that is transferred via the mediator variable attention and a conditional direct effect (Pieters, 2017, p. 693). In order for a mediation analysis to permit plausible causal inferences, it must meet six conditions. First of all, the analysis is directional, since the path from attention to recall is according to literature plausible. Second, the analysis is reliable, since the data preparation ensures a good treatment of the data and missing values to ensure validity. Besides, this study applies validation criteria to the data, in order to account for eye tracking measurement errors. Third, the analysis is unconfounded, because this study includes the relevant variables for this analysis, including covariates to account for possible variation. Fourth, the constructs of the focal variables in this study are conceptually distinguishable. Fifth, the large sample size accounts for sufficient statistical power to identify true non-null effects. Finally, a significant mediation effect must exist, which this study assesses when performing the analyses. This study meets the six conditions and therefore meets the validity requirements of mediation analysis (Pieters, 2017).
This study performs several analyses to estimate each path of the conceptual model, based on the characteristics of the variables. This study estimates each path of the model (a-path, b-path, and c’-path) for the different samples (full sample versus non-zero subsample), the different advertising formats (branded content versus display advertising), and the different brand recall measures (aided recall versus unaided recall). Therefore, this study estimates eight different mediation models, which all include two regressions: one for the a-path of the model and one for the b- and c’-path of the model. Table 3 shows the composition of the eight models. Table 3: Different models used in this study
Sample Advertising format Attention Brand recall Model 1 Full sample Branded content Ad dwell time Aided recall
Model 2 Full sample Branded content Ad dwell time Unaided recall
Model 3 Non-zero subsample Branded content Ad dwell time Aided recall
Model 4 Non-zero subsample Branded content Ad dwell time Unaided recall
Model 5 Full sample Display advertising Ad dwell time Aided recall
Model 6 Full sample Display advertising Ad dwell time Unaided recall
Model 7 Non-zero subsample Display advertising Ad dwell time Aided recall
Model 8 Non-zero subsample Display advertising Ad dwell time Unaided recall
brands and publishers. One can determine if the model should include the variables as a fixed or random effect by calculating correlations between these variables and the focal variables in this study. The correlation matrices in appendix II show these correlations. The variable ‘panelist’ shows very low correlations with the focal variables, of which most effects are not significant in both the full sample and the non-zero subsample. Without correlations, a fixed effect cannot adequately control for inter-unit variation in the focal variables (Fan, 2012, p. 499). Therefore, this study includes the variable ‘panelist’ as a random effect in the model, which accounts for heterogeneity within (within the observations of one panelist) and between (across panelists) observations (Fan, 2012, p. 495). This provides an unconditional inference about a larger set of studies from which one assumes this study to be a random sample (Hedges & Vevea, 1998). Therefore, it addresses the average true effect of a larger population, which makes the results broader generalizable (Viechtbauer, 2010). This study includes the random effect in the intercept, in which it fits each level of the random variable.
The variables ‘publisher’ and ‘brand’ show highly significant correlations (p < 0.001) with the focal variables in the study in both the full sample and the non-zero subsample. This means that the levels of these variables have a differential effect on the focal variables (Fan, 2012, p. 497). One uses fixed effects when these variables correlate with the focal variables in the study. Therefore, this study includes the variables ‘publisher’ and ‘brand’ as fixed effects in the model by adding an extra parameter for each level of the variable. These effects only account for variation within the variable, while ignoring the variation between levels (Fan, 2012).
This study focuses on the a-, b-, and c’-path of the mediation model. First of all, the a-path in this study represents the relationship between advertising medium and attention. This means that the amount of attention (Attentioni,j,k) that panelist i devotes to the advertisement of brand
j in publisher k depends on the advertising media, the fixed effects, the control variables, and the random effect. Therefore, the mathematical representation of Attentioni,j,k is as following:
𝐴𝑡𝑡𝑒𝑛𝑡𝑖𝑜𝑛𝑖,𝑗,𝑘 = 𝛽1𝐴𝑡𝑡𝑂𝑛𝑙𝑖𝑛𝑒𝑗,𝑘 + 𝛽2𝐴𝑡𝑡𝑃𝑟𝑖𝑛𝑡𝑗,𝑘+ 𝛽3𝐴𝑡𝑡𝐵𝑟𝑎𝑛𝑑𝐹𝐸𝑗 + 𝛽4𝐴𝑡𝑡𝑃𝑢𝑏𝑙𝑖𝑠ℎ𝑒𝑟𝐹𝐸𝑘 + 𝛽5𝐴𝑡𝑡𝐺𝑒𝑛𝑑𝑒𝑟𝑖+ 𝛽6𝐴𝑡𝑡𝐴𝑔𝑒𝑖 + 𝛽7𝐴𝑡𝑡𝐼𝑛𝑐𝑜𝑚𝑒𝑖+ 𝛾𝑖𝐴𝑡𝑡+ 𝜀𝑖,𝑗,𝑘𝐴𝑡𝑡
model. The ability of panelist i to recall brand j in publisher k (Aided/Unaided recalli,j,k) depends
on the attention one devotes to the advertisement, the advertising medium, the fixed effects, the control variables, and the random effect. If panelist i recalls brand j in publisher k, Aided/Unaided recalli,j,k equals 1; otherwise, it equals 0. Therefore, the mathematical
representation of Aided/Unaided recalli,j,k is as following:
𝐴𝑖𝑑𝑒𝑑 𝑟𝑒𝑐𝑎𝑙𝑙𝑖,𝑗,𝑘 = 𝛽1𝐴𝑅𝐴𝑡𝑡𝑒𝑛𝑡𝑖𝑜𝑛𝑖,𝑗,𝑘+ 𝛽2𝐴𝑅𝑂𝑛𝑙𝑖𝑛𝑒𝑗,𝑘 + 𝛽3𝐴𝑅𝑃𝑟𝑖𝑛𝑡𝑗,𝑘 + 𝛽4𝐴𝑅𝐵𝑟𝑎𝑛𝑑𝐹𝐸𝑗 + 𝛽5𝐴𝑅𝑃𝑢𝑏𝑙𝑖𝑠ℎ𝑒𝑟𝐹𝐸 𝑘+ 𝛽6𝐴𝑅𝐺𝑒𝑛𝑑𝑒𝑟𝑖+ 𝛽7𝐴𝑅𝐴𝑔𝑒𝑖 + 𝛽8𝐴𝑅𝐼𝑛𝑐𝑜𝑚𝑒𝑖+ 𝛾𝑖𝐴𝑅+ 𝜀𝑖,𝑗,𝑘𝐴𝑅 𝑈𝑛𝑎𝑖𝑑𝑒𝑑 𝑟𝑒𝑐𝑎𝑙𝑙𝑖,𝑗,𝑘 = 𝛽1𝑈𝑅𝐴𝑡𝑡𝑒𝑛𝑡𝑖𝑜𝑛 𝑖,𝑗,𝑘+ 𝛽2𝑈𝑅𝑂𝑛𝑙𝑖𝑛𝑒𝑗,𝑘 + 𝛽3𝑈𝑅𝑃𝑟𝑖𝑛𝑡𝑗,𝑘+ 𝛽4𝑈𝑅𝐵𝑟𝑎𝑛𝑑𝐹𝐸𝑗 + 𝛽5𝑈𝑅𝑃𝑢𝑏𝑙𝑖𝑠ℎ𝑒𝑟𝐹𝐸𝑘+ 𝛽6𝑈𝑅𝐺𝑒𝑛𝑑𝑒𝑟𝑖 + 𝛽7𝑈𝑅𝐴𝑔𝑒𝑖 + 𝛽8𝑈𝑅𝐼𝑛𝑐𝑜𝑚𝑒𝑖+ 𝛾𝑖𝑈𝑅+ 𝜀𝑖,𝑗,𝑘𝑈𝑅
The indirect effect of this model is the effect of the advertising medium on brand recall through attention. Therefore, the mathematical representation of the indirect effect is as following:
𝐼𝑛𝑑𝑖𝑟𝑒𝑐𝑡 𝑒𝑓𝑓𝑒𝑐𝑡 𝑜𝑛𝑙𝑖𝑛𝑒 = 𝛽1𝐴𝑡𝑡∗ 𝛽1 𝐴𝑅/𝑈𝑅
𝐼𝑛𝑑𝑖𝑟𝑒𝑐𝑡 𝑒𝑓𝑓𝑒𝑐𝑡 𝑝𝑟𝑖𝑛𝑡 = 𝛽2𝐴𝑡𝑡∗ 𝛽1𝐴𝑅/𝑈𝑅
Finally, the total effect of the mediation model represents the total effect of the advertising medium on brand recall, including the indirect effect through attention and the remaining direct effect. Therefore, the mathematical representation of the total effect is as following:
𝑇𝑜𝑡𝑎𝑙 𝑒𝑓𝑓𝑒𝑐𝑡 𝑓𝑜𝑟 𝑜𝑛𝑙𝑖𝑛𝑒 𝑎𝑖𝑑𝑒𝑑 𝑟𝑒𝑐𝑎𝑙𝑙 = (𝛽1𝐴𝑡𝑡 ∗ 𝛽 1𝐴𝑅) + 𝛽2𝐴𝑅 𝑇𝑜𝑡𝑎𝑙 𝑒𝑓𝑓𝑒𝑐𝑡 𝑓𝑜𝑟 𝑝𝑟𝑖𝑛𝑡 𝑎𝑖𝑑𝑒𝑑 𝑟𝑒𝑐𝑎𝑙𝑙 = (𝛽2𝐴𝑡𝑡∗ 𝛽 1𝐴𝑅) + 𝛽3𝐴𝑅 𝑇𝑜𝑡𝑎𝑙 𝑒𝑓𝑓𝑒𝑐𝑡 𝑓𝑜𝑟 𝑜𝑛𝑙𝑖𝑛𝑒 𝑢𝑛𝑎𝑖𝑑𝑒𝑑 𝑟𝑒𝑐𝑎𝑙𝑙 = (𝛽1𝐴𝑡𝑡 ∗ 𝛽1𝑈𝑅) + 𝛽2𝑈𝑅 𝑇𝑜𝑡𝑎𝑙 𝑒𝑓𝑓𝑒𝑐𝑡 𝑓𝑜𝑟 𝑝𝑟𝑖𝑛𝑡 𝑢𝑛𝑎𝑖𝑑𝑒𝑑 𝑟𝑒𝑐𝑎𝑙𝑙 = (𝛽2𝐴𝑡𝑡∗ 𝛽 1𝑈𝑅) + 𝛽3𝑈𝑅 Where:
𝛽1𝐴𝑡𝑡 / 𝐴𝑅 / 𝑈𝑅, …, 𝛽8𝐴𝑡𝑡 / 𝐴𝑅 / 𝑈𝑅 = the slope parameter for the independent variables, control variables, and fixed effects for either attention (Att), aided recall (AR), or unaided recall (UR). Aided recalli,j,k = aided brand recall of panelist i for brand j in publisher k.
Unaided recalli,j,k = unaided brand recall of panelist i for brand j in publisher k.
Onlinej,k = online advertising medium for brand j in publisher k.
Printj,k = print advertising medium for brand j in publisher k.
Attentioni,j,k = attention in seconds of panelist i for brand j in publisher k.
BrandFEj = fixed entity for brand j. The model includes a dummy for each j (n-1).
PublisherFEk = fixed entity for publisher k. The model includes a dummy for each k (n-1).
Agei = control variable for the age of panelist i.
Incomei = control variable for the annual combined gross household income of panelist i.
γi = random effect for each panelist i.
𝛾𝑖𝐴𝑡𝑡/𝐴𝑅/𝑈𝑅~ N(0, 𝜔2) εi,j,k 𝐴𝑡𝑡/𝐴𝑅/𝑈𝑅 ~ N(0, 𝜎2)
4.2 Estimation
The estimation of the model is the determination of the parameter estimates by using several techniques (Leeflang et al., 2015). This study uses linear mixed-effects regression models in order to estimate each path of the mediation analysis, in which it fits the model by using maximum likelihood estimation. This study performs these analyses by estimating linear relationships between a criterion variable and one or more predictor variables (Leeflang et al., 2015, p. 96). Besides, this study estimates each path using a bootstrap analysis to accommodate for the non-normality in the disturbance term and to create robustness in the standard errors. Bootstrapping is a method for data-based simulations in order to make statistical inferences (Efron & Tibshirani, 1994). It is a nonparametric approach to estimate effect-sizes without making assumptions about the distribution of the variables. By running the regression several times, it results in a normal distribution of the error terms. I simulate 1000 datasets by using resampling with replacement, which is also the default number of simulations in the R function I use (Tingley, Yamamoto, Hirose, Keele, & Imai, 2014). The bootstrapping procedure considers the sample in this study as only one possible sample out of a distribution of several samples. Below, I explain the estimation procedure for each mediation path.
The b-path has a continuous independent variable and a binary dependent variable, while the c’-path has a binary independent and dependent variable. The disturbance term of the outcome variable in these paths are not normally distributed, because the outcome variable is binary. Therefore, one cannot directly apply a linear regression. However, linear probability models with robustness in the standard errors yield consistent and unbiased estimates (Ameri, Honka, & Xie, 2019; Horrace & Oaxaca, 2006; Sun, Zhang, & Zhu, 2019). Besides, Ameri et al. (2019) chose a linear probability model in order to accommodate the large number of fixed effect in their model specification. This model has robustness in the standard errors as a result of the bootstrapping procedure. Therefore, this study uses the same linear model as for the estimation of the a-path to estimate the b- and c’-path. This linear mixed-effects model accounts for the random and fixed effects in the models, makes the calculation of the indirect and total effect better possible, and results in a more meaningful interpretation. I estimate this model using maximum likelihood, which estimates the beta value that is most likely to match the data.
The indirect effect represents the effect of the a- and b-path together (Zhao, Lynch, & Chen, 2010). This study estimates this effect by multiplying the estimate of the print and online advertising medium for attention with the slope parameter of attention for aided/unaided recall.
The total effect represents the indirect effect plus the remaining direct effect (Zhao et al., 2010). This study estimates this effect by summing the estimate of the indirect effect with the estimate of the advertising medium in the aided/unaided recall equation.
4.3 Validation
Validation of the regression models
5. Results
This chapter reports the results of the analyses in this study. First of all, this study performs preliminary model checks in order to test the basic assumptions of a General Linear Model. Second, I present model-free evidence which provides insights about potential relationships between variables. Finally, this study shows the results of the linear mixed-effects regression analyses, which presents the relationships between the focal variables.
5.1 Preliminary model checks
Basic assumptions of the General Linear Model regression
Significance of the fixed and random effects
In order to determine the significance of the fixed and random effects in the model, I perform several analyses. First of all, I perform an ANOVA-like table for random-effects in order to test the significance of the random effect in this study. This table reduces each random-effect term and performs likelihood ratio tests based on the model reductions (Kuznetsova, Brockhoff, Christensen, & Jensen, 2019). Appendix IV shows these results, which show a highly significant result for all paths for both the full sample and the non-zero subsample (p < 0.001). This means that the random effect significantly influences the model. In order to test the significance of the fixed effects for the publishers and brands, I perform a type III ANOVA with Satterthwaite’s method for each path of the model, which is a specific method of calculating the degrees of freedom (Kuznetsova et al., 2019). Appendix V presents these results, which show a highly significant effect for ‘brand’ for all paths in both the full sample and the non-zero subsample (p < 0.001). For the variable ‘publisher’ the results are highly significant for the paths with aided recall as outcome variable (p < 0.001) and significant for the paths with unaided recall as outcome variable (p < 0.05) for both samples. This means that both the random effect and the fixed effects contribute to fit of the models.
5.2 Model-free results
This paragraph shows model-free evidence of potential relationships between the focal variables in this study. First of all, I compose correlation matrices in order to find the strength of associations between the variables. Appendix II displays the correlation matrices for the full sample and the non-zero subsample. The matrices show an insignificant correlation between advertising format and unaided recall in the full sample, and highly significant results for all other variables (p < 0.001) in both samples in this study. There is a moderate high correlation between advertising format and advertising medium, advertising format and attention, and aided and unaided recall. This indicates that the levels of those variables likely move together. All other variables show low correlations, which means that there is little, if any, linear correlation (Malhotra, 2010).
significant results for both the full sample (t = -19.44, p < 0.001) and non-zero subsample (t = 11.82, p < 0.001). As expected, the ANOVA results also show highly significant effects for both the full sample (F = 498.3.6, p < 0.001) and the non-zero subsample (F = 142.8, p < 0.001). Thus, the advertising media have a differential effect on attention.
The b-path consists of a continuous independent variable and a binary dependent variable. I obtain model-free evidence for this relationship by assessing the correlations between these variables. Attention shows a highly significant correlation with both aided recall (r = 0.17, p < 0.001) and unaided recall (r = 0.09, p < 0.001) in the full sample and with both aided (r = 0.11, p < 0.001) and unaided recall (r = 0.05, p < 0.001) in the non-zero subsample. The correlation coefficients are higher for aided compared to unaided recall in both samples.
The c-path consists of a binary independent variable and a binary dependent variable. I use chi-square tests to investigate if a potential systematic association exists between the advertising medium and brand recall (Malhotra, 2010, p. 499). The results show highly significant results in the full sample for both aided recall (χ2 = 594.21, p < 0.001) and unaided recall (χ2 = 274.24,
p < 0.001). Besides, in the non-zero subsample, the tests also show highly significant results for both aided recall (χ2 = 274.26, p < 0.001) and unaided recall (χ2 = 128.56, p < 0.001). This
means that there is a significant association between the variables. However, this does not directly imply a remaining direct effect of the advertising medium on brand recall, because this analysis does not consider the effect of attention.
5.3 Mediation analyses
