Where the Eye Stops, the Sale Begins:
an Eye-Tracking Study on the Influence of Faces on Attention
on Instagram
Floor van der Meulen Supervisor: Yasin Barış Altaylıgil Student number 11006951 Msc Business Administration: Digital Business Track Faculty of Economics and Business University of Amsterdam Date: 22-06-2018
Statement of Originality
This document is written by Floor van der Meulen who declares to take full responsibility for the contents of this document.
I declare that the text and the work presented in this document is original and that no sources other than those mentioned in the text and its references have been used in creating it.
The Faculty of Economics and Business is responsible solely for the supervision of completion of the work, not for the contents.
Abstract
Instagram has caused a visual revolution for advertisers and marketers and is currently seen as the most valuable advertising tool for marketers. Nevertheless, we still know very little about how firms can optimize these efforts and create engaging Instagram content. As the Instagram platform is becoming more crowded, it becomes more difficult for content to stand out and hold the eye of the user. Thus, insights of what content captures attention are necessary. Previous research has highlighted the effectiveness of human faces in capturing attention, but no such evidence has been provided for Instagram yet. To fill this gap, this research investigated if Instagram images with faces capture more attention than Instagram images without faces. Moreover, it was examined if the gaze direction of the face and the number of people in the image would also lead to higher levels of attention. By using eye-tracking data, this study showed that, indeed, faces in an Instagram image lead to heightened levels of attention. Instagram images with an averted gaze received the most attention and helped viewers to orientate to other elements of the image such as logos and descriptions. Furthermore, this study showed that an increase in the number of people also leads to greater capture of attention on the total image. However, this effect was even more prominent when no faces were presented. This study contributes by providing the first insights on how attention is captured by faces on Instagram. These findings provide new directions for future research on the effect of firm-generated content on Instagram.
Keywords:
Acknowledgements
Hereby, I would like to gratefully acknowledge the support and guidance of my supervisor Yasin Barış Altaylıgil
.
As the use of eye-tracking data was an untraditional method for a thesis, his enthusiasm and trust in this research was a great motivation to see this project to fruition.Furthermore, I would like to thank Adam Cellary from RealEye for providing the eye-tracking software and the continuous support during the collection of the data. Without his support, this research could not have been realized.
Lastly, a warm thank you goes out to my friends and family. Their support and motivational talks made me go the extra mile.
Table of Content
Statement of Originality ... II Abstract ... III Keywords: ... III Acknowledgements ... IV Table of Content ... V List of Tables ... VIII List of Figures ... X1. Introduction... 1
2. Literature Review ... 4
2.2 The capture of attention for different image elements ... 6
2.3 Visual attention ... 7
2.4 Human faces in visual content ... 8
2.4 Gaze cueing ... 9
2.5 Multiple people in an Instagram image ... 11
3. Research Methods ... 13 3. 1 Participants ... 13 3.2 Design ... 13 3.3 Equipment ... 14 3.4 Stimuli development ... 15 3.5 Procedure ... 17
4.1 Sample Characteristics ... 18
4.2 Missing values ... 19
4.3 Normality and Outliers ... 19
4.4 Recoding of the variables ... 19
4.5 Mean Gaze duration of Age and Gender ... 20
4.6 Mean Gaze duration of Brands ... 21
4.7 Mean Gaze Duration of Face Direction on total image ... 23
4.8 Mean Gaze Duration of the number of people on the total image ... 24
4.9 Mean Gaze Duration on the number of faces and Face direction on total image ... 26
4.10 Mean Gaze Duration of Face Direction and Number of People on detailed AOI. ... 28
4.11 Mean Gaze Duration for detailed AOI: Products ... 32
In sum, from all of these results, we can conclude the following regarding the hypotheses: .. 33
5. Discussion ... 34
5.1 The effect of a Face on Instagram ... 35
5.2 The effect of the Gaze of a Face on Instagram ... 36
5.3 The average gaze durations on detailed AOI ... 37
5.4 The effect of the gaze of a face on detailed AOI ... 37
5.5 The effect of multiple people on Instagram ... 38
6. Implications ... 40
7. Concluding Remarks ... 45 8. References ... 45
Appendix A. Example of a heat map of the first timeline with eye-tracking data (Female, 23 year) ... 52 Appendix B. Example of a heat map of the second Instagram timeline with eye-tracking data (Female, 23 year) ... 58 Appendix C. Signed Service and Test Panel agreements with RealEye ... 63 Appendix D. Value labels of all measuring units ... 65
List of Tables
Table 1. T-tests with Mean (M), Standard Deviation (SD), T-score and p-value on total image ... 21 Table 2. Factorial ANOVA with number (n), Mean (M), Standard Deviation (SD) by Brand on total image ... 21 Table 3. Factorial ANOVA with Sum of Squares (SS), degrees of freedom (df), mean squares (MS), F-scores, p-values by Brand on AOI total image ... 22 Table 4. Factorial ANOVA with number (N), mean (M) and standard deviation (Std Dev) of Gaze duration by Brands on AOI detailed ... 22 Table 5. Factorial ANOVA with Sum of Squares (SS), degrees of freedom (df), Mean Squares (MS), Partial Eta Squared (η), F – value and p–value of Gaze duration by Brands on AOI detailed ... 23 Table 6 Factorial ANOVA Factorial ANOVA with number (n), Mean (M) and Standard Deviation (Std Dev) of Gaze duration of FaceDirection on GazeDuration on Total Image .... 24 Table 7. Factorial ANOVA with Sum of Squares (SS), degrees of freedom (df), Mean Squares (MS), Partial Eta Squared (η), F – value and p–value of FaceDirection on total image on GazeDuration ... 24 Table 8. Factorial ANOVA with number (n), Mean (M) and Standard Deviation (Std Dev) of Gaze duration of Number of People on GazeDuration of Total Image ... 25 Table 9. Factorial ANOVA with Sum of Squares (SS), degrees of freedom (df), Mean Squares (MS), Partial Eta Squared (η), F – value and p–value of FaceDirection on Total Image on GazeDuration ... 26 Table 10. Factorial ANOVA with Mean (M), Standard Deviation (SD) and number (N) ... 27
Table 11. Factorial ANOVA with Sum of Squares (SS), degrees of freedom (df), mean squares (MS), F-scores, p-values of the Number of People and the Face Direction on the total image ... 27 Table 12. Factorial ANOVA with Sum of Squares (SS) , degrees of freedom (df), mean squares (MS), F-scores, p-values ... 30 Table 13. Factorial ANOVA with number (n), Mean (M), Standard Deviation (Std Dev) of Nature of Product on GazeDuration ... 33 Table 14. Factorial ANOVA with Sum of Squares (SS), degrees of freedom (df), mean squares (MS), F-scores, p-values for detailed AOI Product on GazeDuration ... 33
List of Figures
Figure 1. Example of Instagram post (Instagram, 2018) ... 6
Figure 2. The conceptual framework of this study ... 13
Figure 3. Example of the detailed Areas Of Interest (AOI) that were used in this study ... 14
Figure 4. Example of image with mutual gaze condition of Starbucks ... 16
Figure 5. Example of image with no-face condition of Starbucks ... 16
Figure 6. Example of image with averted gaze condition of Starbucks... 16
Figure 7. Comparison of group means of Face Direction on Total Image... 23
Figure 8. Comparison of group means of Number of People in Total Image ... 25
Figure 9. The interaction effect NumberFaces and FaceDirection on total image ... 28
Figure 10. The interaction effect between detailed AOI and Facedirection ... 31
Figure 11. The interaction effect between detailed AOI and Brands ... 32
1. Introduction
Unseen is unsold is the advertising adage of the century. Since the rise of new media, scholars have argued that attention has become one of the most valuable resources in our economy (Zulli, 2017). Due to highly crowded media platforms, users are confronted with a vast amount of information. As people can only process a limit amount of information, they will only pay attention to information that stands out and discard information that is irrelevant for them (Drèze & Hussherr, 2003; Sajjacholapunt & Ball, 2014). This means that every piece of information that does not hold the eye of the viewer is likely to be ineffective. Scholars and online marketers have investigated what advertisements are most effective in capturing the attention of the viewer for other media such as print, television, and websites (Aoki & Itoh, 2000; Pieters & Wedel, 2004; Sajjacholapunt & Ball, 2014). However, very little research has specifically focused on Social Networks Sites (SNS) such as Instagram.
Ever since its launch in October 2010, Instagram has been the most fast growing SNS and is seen as the most valuable advertising tool for marketers (Shelly, 2016). Instagram is a mobile application that provides a platform for users to create, modify and share photos and video content. The SNS has seen an explosion in users, reaching over 800 million monthly users in 2017 and is likely to reach a billion in 2019 (Instagram, 2018; Sheldon & Bryant, 2016; Statista, 2017). This enormous user reach has been recognized by many firms, as currently, around 2 million advertisers are active on Instagram every month (Instagram, 2018; Statista, 2017). This results in a highly crowded platform with more than 80 million posts every day of both users and firms (Instagram, 2018). Due to this overload of content, the chances of capturing a user’s attention is increasingly difficult. And although this challenge is likely to be growing due to Instagram’s rising popularity, the extensive literature on what specific content captures attention is still very scarce.
Furthermore, Instagram is seen as a people-centric network, and therefore the majority of content includes photos with human faces (Bakhshi, Shamma, & Gilbert, 2014; Sheldon & Bryant, 2016). For example, the “selfie”, which is a photograph of an individual taken by themselves, is the most popular post (Sheldon & Bryant, 2016). As firms are increasingly active on this platform, one can question if firms should mimic this behavior by creating similar content that includes faces. Studies on other media (i.e., print advertising, websites) have shown that facial images are likely to capture more attention than images without faces. Moreover, previous studies showed that the gaze of a face (i.e. the direction of the eyes), helps the viewer orientate to particular aspects in the image (Adil, Lacoste-badie, & Droulers, 2018; Sajjacholapunt & Ball, 2014). Although many images on Instagram include a face, scholars have not investigated the effectiveness of a face in firm-generated Instagram content yet. This study will examine this by assessing what happens to the attention levels of viewers when firms post facial images. This effect will be assessed for both images with one person, as well as images with multiple persons. Moreover, by manipulating different gaze directions, this study will look at how firms can optimize these images so that the highest brand awareness can be achieved.
Previous studies on Instagram have mainly used quantitative methods. For example, Bakhshi et al. (2014) conducted a large-scale content analysis of likes and comments to assess the influence of facial images on user engagement. However, as this study focuses on the attention level of users, these methods are inadequate to test for the user’s attention. Content analyses neglect the objective side of attention since the capture of attention also occurs unconsciously and does not always translate into actions such as likes, comments or sales (Pieters & Wedel, 2004). Thus, to measure the effect on attention, one should account for this “hidden” type of attention. A popular measure of attention in prior studies concerns
eye-scholars are able to create a more detailed overview and gain a better understanding of what images (or image elements) capture attention. Therefore, this paper will conduct eye-tracking research and investigate the effect of faces and gazes in firm-generated content. This results in the following research question:
RQ: What is the effect of human faces in firm-generated Instagram content on the attention level of the viewer?
By answering this research question, this thesis will contribute to the literature on Visual Attention and Human-Computer Interaction (HCI) as it will provide insights into the attention power of faces and gaze directions. As previous literature has proven the influence of faces for other types of media, this research seeks to extend this understanding for firm-generated content on Instagram.
This paper will be of great value to marketers who seek to optimize their advertising efforts on Instagram. By providing insights into which visual aspects are most effective, marketers can optimize their content so that their efforts are generating the desired results. In addition, as the marketing investments of this SNS is enormously growing, it is crucial to know these guidelines so that the channel is effectively used and the return on investments is maximized (Emily Drewry, 2017).
This thesis will continue by reviewing the current literature on SNS, visual attention and the effect of human faces. At the end of this chapter, a conceptual framework is presented where the hypotheses are visualized. In the following chapter, the research method and analytical strategy will be explained. In the final chapters, the results are presented and discussed and new new insights and suggestions for future research are provided.
2. Literature Review
This chapter will review the existent literature on social media motivations, visual attention selection modes, and the effect of human faces. The findings of this literature will form the basis of the conceptual framework for this thesis. The conceptual framework is presented at the end of this chapter.
2.1 Social Media Services and User Motivation on Instagram
Social media has been described as “a group of Internet-based applications that build on the ideological and technical foundations of Web 2.0 and that allow the creation and exchange of user-generated content” (Kaplan & Haenlein, 2010). One of the most popular SNS is Instagram, which is a mobile application that enables its users to produce, edit, share or simply view images (Bakhshi et al., 2014; Ferwerda, Schedl, & Tkalcic, 2015; Sheldon & Bryant, 2016). Currently, around 32% of all Internet users are on Instagram, with the average user being between the age of 18 and 29 (Statista, 2017). This has been recognized by firms as currently 71% of U.S brands are active on Instagram, which results in around 2 million active advertisers per month (Statista, 2018). The content on Instagram is split between User-Generated Content (UGC) and Firm-User-Generated Content (FGC). Interestingly, 80% of users are following at least one or more brands on Instagram (Drewry, 2017). How traditionally a brand would interrupt consumers with advertisements while reading a magazine or watching television, nowadays brands are integrating into the communication channels of their customers and are trying to reach customers by using the same media channels (Rose, 2013). This means that customers and firms are creating content side by side. Firms are evolving into publishers, where every firm, independent of their product or service offering, is providing content that seeks to satisfy the consumer’s wants and needs and keeps the customer connected to a brand
The early research on SNS states that the main motivations of users for SNS include entertainment, connecting with others, maintaining relationships, information seeking and surveillance and watching of others (Whiting & Williams, 2012). These motivations can be divided between text-based (such as Facebook, Twitter) and visual-based (Pinterest, Instagram) SNS. Whereas text-based platforms are more focused on relational factors (i.e., establishing and maintaining relationships), motivations for visual-based SNS are more related to motivations as self-representation and self-expression (Joinson, n.d.; Sheldon & Bryant, 2016; Whiting & Williams, 2012). More specifically, Sheldon & Bryant (2016) stated that the primary motives of users on Instagram included “surveillance/knowledge about others, documentation, coolness, and creativity” (Sheldon & Bryant, 2016). Therefore, we can conclude that Instagram is mostly focused on lifestyles and self-expression and is a people-centric platform.
As brands are integrating into this channel, guidelines are needed how brands can fit this human-focused theme of content. Instagram users prefer to see content about their lifestyle and experiences instead of product advertisements or promotions (Zhou, 2017). The challenge for brands is therefore to fit into this visual style of users to satisfy its viewers. At the same time, firms should make content that stands out so that the attention of the viewer is captured and the content is recognized. To describe the effect on attention, prior studies often divided the advertisement into certain ad elements (Adil, Lacoste-badie, et al., 2018; Pieters & Wedel, 2004; Sajjacholapunt & Ball, 2014). In respect of these studies, this research will use a similar approach by dividing the Instagram post into different elements.
2.2 The capture of attention for different image elements
When investigating effects on attention, previous literature often divided the advertisement into three elements: brand, pictorial and text (Adil, Lacoste-badie, et al., 2018; Pieters & Wedel, 2004; Pieters, Wedel, & Batra, 2010; Sajjacholapunt & Ball, 2014). In a large-scale eye-tracking experiment by Pieters and Wedel (2004) this division was used to explain how the effect of the elements on the level of attention of the viewer. Their results showed that the pictorial element captured the most baseline attention of the viewer. This means that the pictorial effect has a significant effect on capturing the initial attention of the viewer. These insights are important for a visual communication channel such as
Instagram, as the pictorial aspect is the cornerstone of the platform. However, this division of ad elements is different for Instagram images (Rose, 2013). Whereas traditional advertisements incorporate these three elements into one ad, the division in an Instagram post is bound to the layout of the platform itself, where a clear division exists between the logo, title, image, and description (Instagram, 2018) (Figure 1). There exists very little research that provides insights into where viewers attend to when scrolling through their Instagram feed. Some first insights are provided by Zhou (2017), who found that users not only look at the visual
image but also at the logo and the description (i.e., number of likes) of the post.
Building on these findings, this study will assess which aspects the viewer looks at and measure the attention for each of these elements. To gain a better understanding of how attention is captured, this study will continue by reviewing the existent literature on visual attention selection modes and visual attention in advertising.
Figure 1. Example of Instagram post (Instagram, 2018)
2.3 Visual attention
We are presented with more information than our visual system can simultaneously process. To avoid a processing overload, attentional mechanisms are needed that select only the relevant information for our conscious perception (Vuilleumier & Schwartz, 2001). Previous studies show that these mechanisms exist in two types: bottom-up selection and top-down selection (Greenberg, 2012; Pieters & Wedel, 2004). Top-down selection involves the personal attentional process and resides in the person itself. This selection mode can be directed by the person such as recognizing a familiar brand or selectively searching for a particular product (Pieters & Wedel, 2004). In contrast, bottom-up selection involves processes that are not controlled by the person but is influenced by the stimuli itself. It involves low-level properties of the stimulus such as the size, colors, and objects. Bottom-up selection happens in unrelated contexts such as viewing an unrelated product advertisement when browsing a website. The viewer is drawn to the ad because it stands out due to its form rather than due to the previous engagements with the brand or the intentions of the viewer.
Because of these visual selection mechanisms, we can speak of an “inattentional blindness”, where we unconsciously rule out irrelevant information and process only what our visual system selects. In recent years, scholars have become increasingly interested in these attentional mechanisms. In an age where the online information available greatly exceeds the human capacity to process information, knowing which information will hold the eye of the viewer is very valuable to any firm (Zulli, 2017). Moreover, ads that fail to stand out and hold the attention of the viewer, will be ineffective in guaranteeing ad, product or brand awareness (Keller & Lehmann, 2006). This is why scholars and marketers have become increasingly interested in the bottom-up selection mode. By measuring the impact of specific image elements, one can get insights into which objects and features can drive attention in a
data-driven manner. In this way, the creator of an advertisement can gain control over the influence of the ad on the attention of the viewer.
2.4 Human faces in visual content
One potential attention-capturing element is the human face. Evidence from various eye-tracking studies indicates that people focus on human faces while viewing content (Judd, Ehinger, Torralba, & Durand, 2009; Wedel & Pieters, 2015). This means that the viewer will attend more readily to a face than to other objects or features when they are presented together in a visual scene (Judd et al., 2009; Nummenmaa, Hietanen, Santtila, & Hyönä, 2012). Moreover, behavioral studies show that an ad that includes a human face creates engagement (Bakhshi et al., 2014), requires more processing effort and attention from the user (Walker, Sproull & Subramani, 1994) and is looked at longer when it is perceived as attractive (Collins, 2012). A recent eye-tracking study of (Adil, Lacoste-badie, et al., 2018) investigated the effect on attention for ads that included a face, versus ads without a face. Their results showed that advertisements with a face showed significant longer gaze durations than images without faces. In sum, from these results we can conclude that embedding faces in images is an effective way to increase attention.
In recent years, academics have also started to investigate the effect of faces on Instagram. For example, Bakhshi et al. (2014) show that images with faces receive a higher level of engagement. In their analysis of 1.1 million randomly sampled Instagram posts, they conclude that posts with a face receive 38% more likes and 32% more comments (Bakhshi et al., 2014). However, due to their quantitative method, they could not proprose a causal relationship between a face and the attention of viewers as no objective measures of attention were used. Furthermore, the authors used a random data sample of user-generated content and
Therefore, this thesis will use these findings and extend them by using eye-tracking data to test if the presence of a face in firm-generated Instagram content will result in increased attention to the ad. This results in the following hypothesis:
H1. An Instagram post with a face receives a higher level of attention than an Instagram post without a face
2.4 Gaze cueing
When people look at images that includes a person, they tend to look first at the face before viewing other features of the person (Nummenmaa et al., 2012). Moreover, people are likely to process the internal features of a face better (i.e., the eyes, nose, and mouth) than the external features (i.e., the face contour, hair, and ears) (Althoff & Cohen, 1999; Sajjacholapunt & Ball, 2014). In respect of these findings, the eye gaze has been the interest of many scholars. When testing for the effect of gaze direction, scholars often identify two types of gazes: a face with a mutual gaze and a face with an averted gaze. A mutual gaze is described as a gaze where the dark iris is positioned in the middle of the white sclera, resulting in a straight gaze (i.e., eye-contact). In contrast, when the dark iris is positioned to the left or right of the white sclera, the observer will perceive the gaze as averted (Itier & Batty, 2009; Sajjacholapunt & Ball, 2014).
From previous research, we know that there is a significant difference between these two types of gazes. Evidence of earlier studies demonstrates that observing another person’s gaze direction can shift the attention of the viewer in the corresponding direction (Frischen, Bayliss, & Tipper, 2007; Itier & Batty, 2009; Mansfield, Farroni, & Johnson, 2003; Senju, Tojo, Yaguchi, & Hasegawa, 2005). This is because the processing of a gaze of another person provides us with information about the objects of interest of the other. As this information might be of relevance to us as well, we tend to mirror the gaze towards a similar direction (Itier &
Batty, 2009). A study by Ricciardelli, Bricolo, Aglioti & Chelazzi (2002) demonstrates this effect as they asked their participants to perform a search task by locating a target. They found that when the gaze stimuli included a face of which the gaze was in the direction of the target, the performance of the participants improved, as their time of locating the target was decreased. However, when the gaze of the face was not located towards the target or was replaced by an arrow instead of a face, the performance worsened. This orientation effect of an averted gaze was also tested for print advertisements. With an eye-tracking study, Adil, Lacoste-badie, et al. (2018) provide evidence that a face with an averted gaze leads to higher gaze duration on the advertised product and the surrounding product text compared to a face with mutual gaze. Therefore, we can conclude that a face with averted gaze has the power to direct the viewer to a particular location in the image.
Very little research has investigated these gaze direction effects on social media services. Sajjacholapunt and Ball (2014) show the effect of different gaze direction in banner ads on websites using eye-tracking data. Their findings are in line with previous findings, as banners with mutual gazed models result in a higher average attention level towards the advertising text, product, and product information (measured in mean dwell time) and resulted in an increased gaze duration on these gazed-at elements compared to the mutual gaze or no-face condition. In line with all these previous studies, this thesis will investigate if these findings also apply to the SNS Instagram. Therefore the following hypotheses were stated:
H2a: A face with mutual gaze increases the gaze duration on the face itself compared to a face with averted gaze
H2b: A face with averted gaze increases the gaze duration on the brand-related aspects (i.e., Logo, Description, Product) compared to a face with mutual gaze face
2.5 Multiple people in an Instagram image
Previous research has mainly focused on the effect of a single human face on the level of engagement and attention. However, Instagram posts often include multiple people. Very little research has investigated if increasing the number of people would lead to an increased amount of attention to the image. A content analysis by Bakhshi et al. (2014) suggested that there is no significant difference in levels of engagement (i.e., likes and comments) for photos with multiple faces compared to photos with a single face. However, as this study focuses on attention, we cannot predict if this effect will be similar for attention solely based on this evidence.
One line of literature that provides relevant insights for this issue concerns visual complexity or visual crowding. Pieters et al. (2010) stated that visual complexity is caused in two ways: by features (i.e., variations in colors, luminance, and edges) or by design (i.e., shapes, objects and their arrangement). The higher an image scores on feature or design complexity, the more visually complex the image becomes. Likewise, if an image has a lot of redundancy, the image becomes less complex (Pieters et al., 2010). Pieters et al. (2010) investigated the effect of feature and design complexity on the attention of viewers. They proposed that an increase in the number of objects and shapes (i.e., an increase in design complexity) increases the gaze duration on the total ad. However, the opposite effect was found for feature complexity, of which higher level of complexity resulted in a lower amount of attention.
Thus, from these results, we can conclude that multiple objects in an image would increase the attention to the ad. As this study focuses on the effect of faces, we could argue that an increase in the number is equal to an increase in the number of objects. This increase in people would, therefore, mean that the design complexity would increase, resulting in higher attention levels on the image. Subsequently, this study predicted that an increase in the number
of people and in the number of faces leads to higher levels of attention. This resulted in the following hypotheses:
H3a: The presence of multiple people in an Instagram post increases the gaze duration on the total image compared to posts with a single face
H3b: Images with multiple people with faces increases the gaze duration on the total image compared to multiple people without faces
In relation to the effect of the gaze direction, it was predicted that the effect of the gaze direction will work similarly as only one person is presented in the image. Therefore the hypotheses concerning the gaze direction and multiple people were stated as followed:
H4a: Multiple persons with mutual-gazed faces increase the gaze duration on the face itself compared to averted-gazed faces
H4b: Multiple persons with averted-gazed faces increase the gaze duration on the brand-related AOI compared to mutual-gazed faces
Figure 2. The Conceptual Framework of this study
3. Research Methods
In the following section, the analytical strategy of this study is explained. This includes the participants, design, equipment, stimuli development, and procedure.
3. 1 Participants
This study involved 171 participants aged between 19 and 61 years. As some participants were attained organically (N=71), other respondents participated through the test panel of the RealEye platform (N=100). These testers received a financial reward of $1 for their particiaption. RealEye assured that the testers are human and ethically attained (see Appendix C).
3.2 Design
The experiment involved a within 2 X 3 X 2 mixed design. This design had three independent variables, of which two included wo levels (i.e., face and no-face conditions and the one-person or multiple-persons conditions) and one with three levels (i.e., no face condition, mutual-gaze condition and averted-gaze condition). The mixed design refers to the comparison of the
Firm-generated Instagram Post
Presence of a single Face
• No Face • Face H1 Dependent variable Independent variables H2a Gaze Direction • Mutual Gaze • Averted Gaze H3a Number of People • One person • Multiple persons Gaze Direction • Mutual Gaze • Averted Gaze H4a Attention to Brand-related Elements • Logo • Description • Product
Attention to Total Image
Attention to Face
H2b H3b
manipulations on the attention of the same tester and the comparison between different testers. The dependent variable in this experiment is the attention of the tester, measured in total gaze duration (measured in seconds). The gaze duration was measured for the total post and for four detailed Areas Of Interest (AOI): the face, the product, the logo and the description (Figure 2). These AOI included the total gaze duration within that specific area in the image.
Figure 3. Example of the detailed Areas Of Interest (AOI) that were used in this study
3.3 Equipment
The online eye-tracking platform RealEye was used to collect eye-tracking data. This platform provides eye-tracking software that uses the webcams of the respondents’ computers. By sharing a link, respondents are directed to the platform and can partake in an experiment by enabling their webcam. The process of eye-tracking always starts with calibration, where the tester follows different crosses on the screen for one minute. In this way, the platform learns how the eyes of the tester look like and can start to measure the eye position while displaying various stimuli (RealEye, 2018). The precision of their measure is proven to be around 64px
into the precision of the size of a single button and therefore allows for accurate measuring of the respondents’ attention. For the presentation of stimuli, RealEye creates a screenshot of a page. This meant that the randomizer of the Qualtrics survey was done once and then presented to all testers.
3.4 Stimuli development
The content on Instagram is seen as lifestyle-oriented, and therefore involves themes such as food, fitness and fashion. (Hu, Manikonda, & Kambhampati, 2014; Manikonda, Meduri, & Kambhampati, 2016). Thus, these three themes were selected for the posts included in this study. Previous studies have used fictitious brands and posts in their eye-tracking studies, resulting in limited external validity. According to Sajjacholapunt and Ball (2014), research is needed that uses professional advertisements so that the external validity can be increased. This study answers this need by using previously posted images by real firms on Instagram.
For the selection of these firm-generated posts, the most followed brands on Instagram were selected. Previous studies that included memory tests or variables such as brand recall often chose to exclude familiar brands from their research or used fictitious brands, as people might recall a familiar brand better than an unfamiliar brand. However, as this research will only include objective data of the viewer’s attention (i.e., gaze duration), this will not limit the purpose of this study. Moreover, by only incorporating very familiar brands (i.e., the most followed brands on Instagram), this study aimed to equalize this familiarity effect for all stimuli. According to Statista (2017), the three most followed brands include National Geographic (72 million followers), Nike (70 million followers) and Victoria’s Secret (52 million followers). After inspection of these three accounts, National Geographic was excluded from the research as the majority of posts of the brand account involves scenery and animal
most food and beverage pictured brand on Instagram and has a total of 16,3 million followers. To match the food, fitness and fashion themes, this brand was additionally selected. Therefore, this study included Instagram posts of the Instagram accounts of Starbucks, Nike and Victoria’s Secret.
For each brand, 3 posts were selected from their Instagram feed that included at least one face. After this selection, the images were manipulated three times so that each image includes a set of different gaze directions: one image with a mutual gaze, one image with an averted gaze and one image without a face (i.e., the no-face condition) (see Figures 4, 5 & 6). This manipulation was done by the researcher in Photoshop and consisted of a manipulation of the eyes. The no-face condition of each image was created by cutting the face of the picture. This approach was chosen so that the features of the pictures (i.e., the colors of the images) remain relatively the same between the manipulations. In this way, this study aims to minimize the effect of feature complexity. Furthermore, by including a large number of images (36 pictures in total), this study aimed to reduce the effect of recognition on the level of attention.
Figure 4. Example of image with mutual gaze condition of Starbucks
Figure 6. Example of image with averted gaze condition of Starbucks
Figure 5. Example of image with no-face condition of Starbucks
all images were edited so that they would contain a similar number of likes (184.225 likes) and comments. All hashtags and emoticons that titles were included in the post description were removed, as well as the comments below the pictures. All titles were edited so that each post of a particular brand received a similar title of an average length. This was done so that the titles, comments or emoticons would not trigger the attention of the viewer. In this way, this study controlled for other variables that might influence the attention level of the viewer.
For the representation of these images, a survey in Qualtrics was created. All images were uploaded and randomized. To create a more realistic interface that is similar to the Instagram interface, a header of the Instagram Timeline was added. Two additional images were added in the beginning and at the end of the survey so that the eye-tracking data would not be affected by the transition to the page or the scrolling pace of the tester. These images are excluded from the analysis.
3.5 Procedure
Every participant received a link to the RealEye platform. Before the experiment started, the respondents were shown instructions on what the study entailed and how the eye-tracking works. The respondents were briefed to scroll down the page. Also, they were asked not to scroll back up so that they would only see the pictures once. After giving their consent, the calibration started so that the platform could position the eyes of the tester. The calibration entailed that the respondent had to fixate on five small white crosses on the computer screen without moving their head or body. When the calibration was completed, every tester was shown two timelines of 20 photos for 60 seconds. This set-up was chosen as the RealEye platform allows for a maximum eye-tracking of 60 seconds for each recording after which the tester is automatically directed to the next timeline.
A total of 40 pictures were shown. These pictures included three brands and similar posts with different manipulations. After the two timelines, the respondent was asked to state their age and gender and could immediately see their own results in the form of a heat map.
4. Results
In this chapter, this study will present the results of the statistical analysis. Concerning the analytical strategy, the sample characteristics, missing values, normality and outliers and recoding of the variables will be discussed. This section will be followed by various analysis that will answer the earlier stated hypotheses. This includes tests for Age, Gender, Brands, Face Directions, detailed AOI and the Number of People.
4.1 Sample Characteristics
The experiment concluded with a total of 171 respondents. An example of a heat map with eye-tracking data can be found in Appendices A and B. When the data-collection was completed, the Areas of Interest (AOI) were created for the total image (i.e. total view of the image) and for detailed areas of each post such as faces, logos, descriptions, and products. The results of these AOI were scraped from the platform and structured in Excel according to their TesterID. Due to the differences in speed of scrolling among testers, all results were examined for their completion and coded according to their progress. Images that respondents were unable to see due to too little time were treated as missing values and excluded from the analysis. Furthermore, cases that did not include their age or gender (N=17) or the tester’s eye-tracking results showed invalid results due to calibration errors (N=20) were also excluded from this study. This resulted in a total of 134 respondents. This included 74 male and 61 female
After summarizing the data, it was prepared for the statistical analysis. In this experiment gaze duration (measured in seconds) is the dependent variable and the AOI are the independent variables, with gender and age as control variables. The measurement unit is a tester per AOI, which resulted in a dataset of 1 tester x 176 Areas of Interest (N=20653). The mean and standard deviations were .469 seconds and .0546 seconds.
4.2 Missing values
The dataset was checked for missing values by running a frequency test. A total of 90.4% (N=20778) of the testers completed the entire experiment during the maximum time of 60 seconds. The number of missing values (< 10%) were excluded from the analysis.
4.3 Normality and Outliers
Violation of normality was confirmed through the inspection of Q-Q Plots, Histograms and the use of a Kolmogorov-Smirnov test (p<.00). The mean gaze duration data was examined for skew and deviations from the normal distribution. Similar to other time-based experiments, an extreme positive skew was found (Sajjacholapunt & Ball, 2014). This extreme positive skew was also seen in a skewness of 3.527 (SE=.017) and a Kurtosis of 19.769 (SE=.034). These violations were dealt with by examining for outliers. Accordingly, Z-scores of Gaze duration were, and all outliers (i.e. scores smaller than -3 or greater than 3) were excluded. This resulted in an acceptable Skewness of 2.228 (SE= .017) and a Kurtosis of 4.798 (SE=.034) with N = 20653.
4.4 Recoding of the variables
Through automatic recoding, all measuring units received a value, which resulted in 176 conditions (see Appendix D). These conditions (N=176) were recoded into different variables (i.e., FaceDirection, Brands, AOI_detailed and NumberPeople). For the conditions FaceDirection, Brands and NumberPeople, two variables for each condition were recoded. One
variable included data from the total image (FaceDirection_TOTImage) and the other variables included data from only the detailed AOI (FaceDirection_AOIdetailed). This was done so that this study can test the effect of conditions on the total image and on more detailed AOI (e.g. Logo). Due to this approach, the detailed AOI dataset included 16847 measuring units and the AOI of total images included 3806 measuring units.
4.5 Mean Gaze duration of Age and Gender
Firstly, it was examined if the demographics of the respondents had an effect on Gaze duration. An independent T-test was conducted to compare gaze duration and the gender of the respondents (Table 1). There was a significant differences in the scores for males (M = .361, SD = .535) and females (M = .379, SD = .557) conditions; t(20651) = -2.418, p = .001. From this analysis, we can conclude that female respondents have significant longer gaze duration than male respondents. To account for this effect, Gender was added as a control variable in all further analyses.
A frequency check for Age showed that the majority of the respondents are younger than 30 years (i.e., 59 %). Therefore, Age was recoded, and an independent t-test was conducted to compare respondents younger than 30 years and respondents who are 30 years or older (table 1). There was no significant difference in GazeDuration between respondents who are younger than 30 years (M = .372, SD = .547) and respondents who are 30 years or older (M = .776, SD = .541) conditions; t(20651) = 1.406, p = .160). From these results, we can conclude that the age of the respondents did not have a significant influence on the analyses.
Table 1. T-tests with Mean (M), Standard Deviation (SD), T-score and p-value on total image
Variables Group N M SD T score p-value
Gender Male 11294 .361 .535 -2.418 .001 Female 9359 .379 .557 Age < 30 years 12942 .372 .547 1.406 .160 30 years or older 7711 .776 .541
Note. N = 20778. Gender was coded as 0 = Male and 1 = female. Age was coded as 1 = < 30 years and 2 = 30 years and older.
4.6 Mean Gaze duration of Brands
After the checking for significant effects of the control variables, it was examined if a difference in the brand (e.g., Nike, Starbucks, Victoria’s secret) influences the duration of the gaze on the total images. A factorial ANOVA was conducted to compare the effect of Brands on the GazeDuration on total images while controlling for Gender and Age (Tables 2 & 3). The analysis showed that there is no significant main effect of Brands on the GazeDuration on total images. This means that respondents did not look significantly longer at a total image of a particular brand.
Table 2. Factorial ANOVA with number (n), Mean (M), Standard Deviation (SD) by Brand on total image
Brands n Mean Std Dev
Nike 1.258 1.204 .664
Starbucks 1.301 1.207 .693
Victoria's Secret 1.247 1.251 .705
Total 3.806 1.220 .688
Note. n = 3806. Brands was coded as 1 = Nike, 2 = Starbucks, 3 = Victoria’s Secret. Gender and Age were included as control variables.
Table 3. Factorial ANOVA with Sum of Squares (SS), degrees of freedom (df), mean squares (MS), F-scores, p-values by Brand on AOI total image
Source SS df MS η F p Age 17.722 1 17.772 .010 38.068 .000 Gender 1.569 1 1.569 .001 3.360 .067 Brands 1.777 2 .888 .001 1.903 .149 Error 1.774.487 3801 .052 Total 7.469.605 3806
Note. n = 3806. Brands was coded as 1 = Nike, 2 = Starbucks, 3 = Victoria’s Secret. Gender and Age were included as control variables.
For the more detailed AOI, another factorial ANOVA was conducted to assess any difference between group means of brands and more specific image aspects, while controlling for Gender and Age (Tables 4 & 5). Contrary to the results of the effect of Brands on total images, the results showed a significant main effect of Brands_detailed AOI (F(2,16487) = 55.686, p = .000) on GazeDuration. Also, the control variables Gender (p = .002) and Age (p = .000) showed significant effects on GazeDuration in this analysis.
To follow up on this significant main effect, a series of post hoc Bonferroni test were pursued for Brands_detailedAOI. These results revealed that the gaze duration for all the brands on detailed AOI significantly differ from each other (p = .000). Images of the brand Victoria’s Secret (M = .201, SD = .260) had a significant higher mean gaze duration than images of Nike (M = .154, SD = .218) and Starbucks (M = .173, SD = .235). Therefore, Brands_AOIdetailed was included in the future analysis for the detailed AOI to account for its significant main effect on the gaze duration.
Table 4. Factorial ANOVA with number (N), mean (M) and standard deviation (Std Dev) of Gaze duration by Brands on AOI detailed
Brands n Mean Std Dev
Nike 5.212 .154 .218
Starbucks 5.762 .173 .236
Victoria's Secret 5.873 .201 .260
Table 5. Factorial ANOVA with Sum of Squares (SS), degrees of freedom (df), Mean Squares (MS), Partial Eta Squared (η), F – value and p–value of Gaze duration by Brands on AOI detailed
Source SS df MS η F p Age 5.772 1 5.772 .006 101.402 .000 Gender .531 1 .531 .001 9.328 .002 Brands 6.340 2 3.170 .007 55.686 .000 Error 958.718 16842 .052 Total 1.497.021 16847
Note. n = 16847. Brands on AOI detailed was coded as 1 = Nike, 2 = Starbucks, 3 = Victoria’s Secret. Gender and Age were included as control variables.
4.7 Mean Gaze Duration of Face Direction on total image
In line with the first hypothesis, it was examined if images containing a face increases the level of attention (i.e., gaze duration) on the total image. To test these predictions, the variable FaceDirection_TOTimage was included as the independent variable, so that the dataset would only include the AOI of total images and exclude the AOI with more detailed information. This resulted in a dataset with N =3806.
A factorial ANOVA was conducted to compare the effect of FaceDirection_TOTimage on GazeDuration of the total image while controlling for Age, Gender and Brands (Figure 7, Tables 6 & 7).
Figure 7. Comparison of group means of Face Direction on Total Image
There was a statistically significant main effect of FaceDirection_TOTimage on GazeDuration, F (2,3803) = 9.185, p = .000. Follow-up Bonferroni tests revealed that the gaze
duration was significantly higher for the mutual condition (p = .036) and averted condition (p = .000) compared to the no-face condition. The mutual and averted conditions did not significantly differ (p = .138). These findings are in line with H1 and confirm the predictions that a face, both in mutual or averted face condition, increases the gaze duration on the total image.
Table 6 Factorial ANOVA Factorial ANOVA with number (n), Mean (M) and Standard Deviation (Std Dev) of Gaze duration of FaceDirection on GazeDuration on Total Image
Face direction on total image n Mean Std Dev
Mutual gaze 1281 1.226 .689
Averted gaze 1255 1.278 .696
No-face 1270 1.158 .674
Total 3806 1.220 .688
Note. n = 3806. FaceDirection_TOTImage was coded as 1 = Mutual gaze, 2 = Averted gaze and 3 =No-face. Gender and Age were included as control variables.
Table 7. Factorial ANOVA with Sum of Squares (SS), degrees of freedom (df), Mean Squares (MS), Partial Eta Squared (η), F – value and p–value of FaceDirection on total image on GazeDuration
Source SS df MS η F p
Gender 1.619 1 1.619 .001 3.483 .062
Age 17.709 1 17.709 .01 38.093 .000
Face direction on Total image 9.185 2 4.592 .005 9.878 .000
Error 1.767.079 3801 .465
Total 7.469.605 3806
4.8 Mean Gaze Duration of the number of people on the total image
In relation to predictions of the number of people, a factorial ANOVA was conducted to compare gaze durations on total images that included one person (N=2841) and gaze durations on total images that included multiple people (N=965). This analysis also included the no-face condition as still two bodies are displayed in an image with the no-face condition due to the cutting of the face from the image in the no-face condition. As this analysis tested for gaze duration on the total images, 3806 measuring units were included (Figure 8, Tables 8
Figure 8. Comparison of group means of Number of People in Total Image
A significant main effect on GazeDuration on total images was found for NumberPeople on the total image; F(1,3801) = 11.603, p = .001). To follow-up on this main effect, post hoc Bonferroni tests were pursued and revealed that images with multiple persons received significant (p= .001) longer gaze durations on the total image (M= 1.285, SD = .699) than images with only one person (M = 1.199, SD = .683). This confirms the prediction that multiple people increase the attention to an image compared to an image with a single person and thus, H3a is confirmed.
Table 8. Factorial ANOVA with number (n), Mean (M) and Standard Deviation (Std Dev) of Gaze duration of Number of People on GazeDuration of Total Image
Number of People on Total Image n Mean Std Dev
One person 2841 1.198 .683
Multiple persons 965 1.285 .699
Total 3806 1.220 .688
Note. n = 3806. NumberPeople total image was coded as 1 = one person and 2 = multiple persons. Gender and Age were included as control variables.
Table 9. Factorial ANOVA with Sum of Squares (SS), degrees of freedom (df), Mean Squares (MS), Partial Eta Squared (η), F – value and p–value of FaceDirection on Total Image on GazeDuration
Source SS df MS η F p
Gender 1.617 1 1.617 .001 3.472 .063
Age 17.652 1 17.652 .01 37.899 .000
Number of People on Total Image 5.404 1 5.404 .003 11.603 .001
Error 1770.859 3802 .466
Total 7469.605 3806
Note. n = 3806. NumberPeople total image was coded as 1 = one person and 2 = multiple persons. Gender and Age were included as control variables.
4.9 Mean Gaze Duration on the number of faces and Face direction on total image
From the results of the previous analysis, it was observed that increasing number of people increases attention on the total image. Accordingly, a more detailed look was taken into possible interactions between the number of people and their face direction. A factorial ANOVA was conducted that included GazeDirection as the dependent variable, NumberPeople and FaceDirection on the total image as dependent variables and Age and Gender as control variables (Figure 9, Tables 10 & 11).
The analysis showed a significant main effect of NumberPeople F (1, 3806) = 11.601, p =.003, η2 = .003) and FaceDirection (F(1,3806) = 3.555 , p =.002, η2 = .002) on GazeDuration. Also the control variables Age (F(1, 3806) = 38.149, p = .001, η2 = .010), Gender (F(1,3806) = 3.399 p =.023, η2 = .001) showed a significant main effect on GazeDuration.
Moreover, there was a significant interaction effect between NumberPeople and FaceDirection_TOTimage (p = .002). In the one face condition, a mutual gaze (M = 1.226, SD = .689) and an averted gaze (M = 1.267, SD = .692) showed higher gaze durations than the no-face condition (M = 1.099, SD = .657). Yet, in the multiple no-faces condition, the no-no-face
condition (M = 1.306, SD = .708) and the mutual condition (M = 1.225, SD = .690). This means that a single person with an averted gaze yields the highest gaze duration, but images that include multiple persons received the most attention in the no-face condition. However, the Partial Eta Squared of these effects shows that all of these effects are very low-sized (η <.06). From these results, we can only partially confirm the prediction that images with multiple people with faces receive greater gaze durations than images with multiple people without faces. This is because this prediction is only true for the averted gaze condition and not for the mutual gaze condition. Thus, H3b is partially accepted.
Table 10. Factorial ANOVA with Mean (M), Standard Deviation (SD) and number (N)
Number of People TOT Image Face Direction TOT image Mean SD N
One person Mutual gaze 1.226 .689 966
Averted gaze 1.267 .692 940
No-face 1.099 .657 935
Total 1.210 .683 2841
Multiple person Mutual gaze 1.225 .690 315
Averted gaze 1.306 .708 315
No-face 1.323 .697 335
Total 1.285 .699 965
Total Mutual gaze 1.226 .689 1281
Averted gaze 1.278 .696 1255
No-face 1.158 .675 1270
Total 1.220 .688 3806
Note. N = 3806 NumberPeople total image was coded as 1 = one person and 2 = multiple persons. Face Direction was coded as 1= Mutual gaze, 2= Averted gaze and 3= No-face. Gender and Age were included as control variables.
Table 11. Factorial ANOVA with Sum of Squares (SS), degrees of freedom (df), mean squares (MS), F-scores, p-values of the Number of People and the Face Direction on the total image
Source SS df MS η F p
Age 17.623 1 17.623 .010 38.149 .000
Gender 1.570 1 1.570 .001 3.399 .065
Number of People in total image 5.359 1 5.359 .003 11.601 .001 Face direction 3.285 2 1.643 .002 3.555 .029 Number of People*Face direction 6.942 2 3.471 .004 7.514 .001 Error 1754.548 3798 .462 Total 7469.605 3806
Note. N = 3806 NumberPeople total image was coded as 1 = one person and 2 = multiple persons. Face Direction was coded as 1= Mutual gaze, 2= Averted gaze and 3= No-face. Gender and Age were included as control variables.
Figure 9. The interaction effect NumberFaces and FaceDirection on total image
4.10 Mean Gaze Duration of Face Direction and Number of People on detailed AOI.
As the previous results showed a significant effect of FaceDirection and NumberPeople, these effects were also assessed for the detailed AOI. To pursue this approach, the variables FaceDirection, NumberPeople and Brands were recoded into FaceDirection_AOIdetailed, NumberPeople_AOIDetailed, and Brands_AOIdetailed, so that this analysis only included the data from more detailed AOI, excluding the gaze duration on the total image (N=16847). Here, Brands was included as the earlier analyses confirmed its significant effect on the detailed AOI.
A factorial ANOVA was conducted that included AOI_detailed, Facedirection_AOIdetailed, Brands_AOIdetailed and NumberPeople_AOIdetailed as independent variables, GazeDuration as dependent variable and Age, Gender as control variables (Table 12).
Similar as to findings from the AOI that included total images, this analysis revealed that there exist significant main effects on gaze duration of Age (F(1,16780) = 107.487, p =000,
256.544, p = .000,
η
2p = .044), Face Direction_AOIdetailed (F (2,16780) = 30.663, p =.000,
η
2p= ,004) and Brands_AOIdetailed (F (2,16780) = 30.843, p <=.000,
η
= .004). All of the effects were again low sized (η
< . 06), with AOI_detailed yielding the highest Partial Eta-Squared. These results indicate that the gaze duration on the detailed AOI differs itself between the different gaze conditions, between different brands and among the detailed AOI themselves.However, different from the analysis on the gaze durations of total images, there was no significant main effect of NumberPeople_AOIdetailed on GazeDuration (F (1,16780) = .085, p =.770,
η
= .000). This means that a difference in the number of people in the image did not result in significantly different gaze durations for more detailed AOI. Therefore, H4a and H4b are rejected.To follow-up the main effect of AOI_detailed, a series of Bonferroni tests revealed that the AOI logos captured the least amount of attention (M = .087) followed by descriptions (M = .189), faces (M = .201) and products (M = .227). Except for the mean difference between descriptions and faces (p = .163), all of these effects were significant at p-value =.000.
Note. N=16847
Concerning the prediction of the effect of a gaze on the attention to detailed AOI, this analysis showed that there are significant interactions effects between AOIdetailed and FaceDirection (p =.000), AOIdetailed and Brands (p = .000) and between FaceDirection and Brands (p = .000). In relation to AOI detailed x FaceDirection, post hoc Bonferroni tests revealed that the mean gaze duration of Logo (M= .103, SD = .116), Faces (M= .229, SD = .176), and Description (M= .216, SD = .176) were highest in the averted gaze condition. For the AOI Product, the highest mean gaze duration was in the no-face condition (M=.293, SD =
Source SS df MS η F p Age 5.549 1 5.549 .006 107.487 .000 Gender .536 1 .535 .001 10.385 .001 Face Direction 3.166 2 1.583 .004 30.663 .000 Brands 3.184 2 1.592 .004 30.843 .000 AOI Detailed 39.729 3 13.243 .044 256.544 .000 Number of People .004 1 .004 .000 .085 .770 Face Direction*Brands 8.775 4 2.194 .010 42.498 .000
Face Direction*AOI Detailed 5.068 5 1.014 .006 19.498 .000 Face Direction*Number of People 1.204 2 .602 .001 19.634 .000 Brands*AOI Detailed 1.573 6 .262 .002 11.662 .000 Brands*Number of People .835 2 .417 .001 5.077 .000 AOI Detailed*Number of People .967 3 .322 .001 8.087 .000 Face Direction*Brands*AOI Detailed 2.125 10 .213 .002 6.244 .098 Face Direction*Brands*Number of People .405 4 .101 .000 4.117 .000
Face Direction*AOI Detailed* Number of People
1.770 5 .354 .002 1.960 .000
Brands*AOI Detailed* Number of People
4.298 6 .716 .005 6.858 .000
Face Direction*Brands*AOI Detailed* Number of People
1.530 9 .170 .002 13.878 .001
Error 866.196 16780 .052 3.292
Total 1.497.021 16847
the averted condition results in higher gaze durations for logos, descriptions, and faces. Moreover, images that included a mutual gaze showed significant lower gaze durations on the AOI Face than imaged with averted gaze. These results reject the prediction that a face with mutual gaze increases the gaze duration on the face itself (H2a), and only partially confirm the prediction that an averted gaze increases attention to brand-related aspects (H2b).
Figure 10. The interaction effect between detailed AOI and Facedirection
As previous results showed that Brands had a significant effect on the detailed AOI, the interaction between these two variables was also further investigated. A significant low-sized interaction effect was found between Brands_AOIdetailed and AOI_detailed (F (6,16780) = 5.077, p = .000,
η
2p = .002). Images of the Victoria’s Secret brand received longest gaze durations
on Description (M= .199, SD= .214), Product (M=.262, SD=.300) and Face (M=.237, SD= .304). The gaze durations of Logo were the highest for the Starbucks brand (M=.095, SD= .268). Overall, the images of Victoria’s Secret received most attention on their detailed AOI (M= .200, SD=.26) compared to the Nike brand (M= .154, SD= .218) and the Starbucks brand (M= .174, SD=.235).
Figure 11. The interaction effect between detailed AOI and Brands
4.11 Mean Gaze Duration for detailed AOI: Products
As the previous analyses showed that products of the brands Nike and Victoria’s Secret received the most attention, follow-up analyses were conducted in order to investigate if the nature of this product itself could be the cause of this. As images of Nike and Victoria’s Secret included products of lingerie wear, it was assessed if these products significantly differed in their attention levels from other products (e.g., a coffee cup from Starbucks). In this respect, a new variable was computed called ProductNature where products including lingerie were coded as 1 and all other products as 2.
(N=4375) (Figure 12, Tables 13 & 14). Results showed a significant difference in gaze duration means between Lingerie Products (M= .255, SD=.290) and Other Products (M= .208, SD= .276) conditions; F (1,4371) = 30.845, p-value = .000). Therefore, this analysis shows that the respondents looked significantly longer at lingerie products than to all other products.
Table 13. Factorial ANOVA with number (n), Mean (M), Standard Deviation (Std Dev) of Nature of Product on GazeDuration
Nature of Products n Mean Std Dev
Lingerie Products 2125 .255 .290
Other Products 2250 .208 .276
Total 4375 .230 .284
Note. N = 4375. Products on AOI detailed was coded as 1 = Lingerie products, 2 = other products
Table 14. Factorial ANOVA with Sum of Squares (SS), degrees of freedom (df), mean squares (MS), F-scores, p-values for detailed AOI Product on GazeDuration
Source SS df MS η F p Gender .557 1 .557 .002 6.986 .008 Age 1.875 1 1.875 .005 23.521 .000 Products 2.460 1 2.460 .007 30.845 .000 Error 384.511 4371 .080 Total 352.957 4375
Note. N = 4375. Products on AOI detailed was coded as 1 = Lingerie products, 2 = other products
In sum, from all of these results, we can conclude the following regarding the hypotheses:
H1. An Instagram post with a face receives a higher level of attention than an Instagram post without a face (accepted)
H2a: A face with mutual gaze increases the gaze duration on the face itself compared to a face with averted gaze (rejected)
H2b: A face with averted gaze increases the gaze duration on the brand-related aspects ((Logo, description, Products) compared to a face with mutual gaze face (partially accepted)
H3a: The presence of multiple people in an Instagram post increases the gaze duration on the total image compared to posts with a single face (accepted)
H3b: Images with multiple people with faces increases the gaze duration on the total image compared to multiple people without faces (partially accepted)
H4a: Multiple persons with mutual-gazed faces increase the gaze duration on the face itself compared to averted-gazed faces (rejected)
H4b: Multiple persons with averted-gazed faces increases the gaze duration on the brand-related AOI compared to mutual-gazed faces (rejected)
5. Discussion
This experiment used eye-tracking data to examine if attention (measured by gaze duration in seconds) is affected by (1) a face compared to images without faces (i.e., mutual face versus averted face versus no face), (2) the gaze of the person in the image (i.e. mutual gaze versus averted gaze) and (3) the number of people (i.e., images with one person versus images with multiple persons). The primary measure that was used was the mean gaze duration on a Areas of Interest (AOI). These different AOIs included the total image and more detailed aspects such as logos, descriptions, products, and faces.
In relation to the use of faces in Instagram content, previous eye-tracking research for other advertising media (e.g., print and web pages) showed that advertisements with faces result in higher gaze durations (Adil, Lacoste-badie, et al., 2018; Pieters & Wedel, 2004; Sajjacholapunt & Ball, 2014). No such evidence was provided for Instagram yet. However, a content analysis of Instagram images by Bakhshi et al. (2014) showed an increased level of engagement in terms of likes and comments. Thus, it was predicted that images including a face
In relation to the gaze of a face, evidence from previous studies showed that the direction of a gaze can have a cueing effect, meaning that it helps the viewers to orientate towards other areas in the image (Frischen et al., 2007; Hutton & Nolte, 2011; Sajjacholapunt & Ball, 2014). This helped people in locating search tasks and other image elements such as products or text elements. On Instagram, images often include faces and products and are surrounded by many other elements such as logos and descriptions. By including images with either a single face or multiple faces with different gaze directions (i.e., mutual gaze or averted gaze), this study aimed to assess this finding and predicted that averted gazes in facial images lead to higher gaze durations on brand-related aspects (i.e., Logo, Product, Description). Furthermore, based on evidence from Sajjacholapunt & Ball (2014), it was predicted that people tend to look longer at mutual-gazed faces compared to averted-gazed faces.
In relation to the number of faces, Bakhshi et al. (2014) indicated that multiple faces do not lead to changed levels of engagement. Yet, these results were based on engagement in the forms of likes and comments, rather than the level of attention. Therefore, this study hypothesized according to evidence from research on visual complexity that states that increasing the number of objects (i.e., increased design complexity) leads to higher levels of attention (Pieters et al., 2010). This experiment aimed to investigate this by using eye-tracking data for the prediction that images with multiple faces increase the gaze duration on total images compared to images with a single face.
5.1 The effect of a Face on Instagram
The critical finding of this experiment is the significant difference in the average gaze duration between an image with a face and an image without a face. Images with faces, both with mutual or averted gaze, have a higher gaze duration and therefore capture more attention than images without faces. This is in line with findings of previous studies that have similarly
