Click as Fast as You Can:
A Duration Model on the Timing of Online Video Ad Skipping
January 11, 2021
University of Groningen Faculty of Economics and Business
Master’s Thesis
MSc Marketing Intelligence & Marketing Management
Evi Gorter Paterswoldseweg 81a 9727 BB Groningen S2928175 e.s.gorter@student.rug.nl
2
Abstract
YouTube is one of the largest online video platforms with more than 2 billion visitors each month. The most used form of advertising on this platform are pre-roll advertisements. Pre-roll advertisements are gaining popularity among marketers and consumers because of the possibility to skip the ad after 5 seconds. Giving the consumer the power to determine the duration of the ad reduces the negative attitude against advertisements. However, very little is known about the factors influencing consumer’s skipping behavior. Furthermore, much uncertainty still exists about the timing of skipping. This study investigates the effect of advertisement attitude, product involvement, and type of device on consumer’s skipping behavior. The results show that advertisement attitude negatively affects skipping. Moreover, the probability to skip the ad increases when the visitor is watching the video on a smartphone or tablet, compared to a desktop. Besides, significant effects of age and gender are found. Younger visitors and males are more likely to skip the online video advertisement. Interestingly, Germans and Dutch react differently to the investigated factors. Germans respond more modest compared to Dutch visitors. No differences are found between Dutch and British visitors. This study helps marketers to better understand the analysis of their online video campaigns. The importance of a good advertisement is emphasized, advertisement attitude is the most important factor affecting skipping behavior.
3
Table of Content
1. Introduction ... 5
1.1 Traditional to Online Media ... 5
1.2 Online Video Advertising ... 6
1.3 Research Question and Relevance ... 7
1.4 Structure of the Paper ... 8
2. Theoretical Framework ... 9
2.1 Changing Role of Visitors ... 9
4
3.4.7 Nationality ... 21
3.5 Data Cleaning and Outliers ... 21
3.6 Model Specification ... 21
3.7 Descriptive Analysis ... 22
3.8 Multicollinearity ... 24
4. Results ... 25
4.1 Analysis of Skipping Behavior ... 25
4.2 Kaplan-Meier Method ... 25
4.3 Cox Model ... 27
4.3.1 Model Fit ... 27
4.3.2 Estimations ... 27
4.4 Interaction Effects ... 29
4.4.1 Interaction Effect with Type of Device... 29
4.4.2 Interaction Effect with Nationality ... 30
4.5 Proportional Hazards Assumption ... 31
5. Discussion ... 35
5.1 Conclusions ... 35
5.2 Managerial Implications ... 37
5.3 Limitations and Suggestions for Further Research ... 38
References ... 39
Appendices ... 45
Appendix 1: Ad Attitude ... 45
Appendix 2: Product Involvement ... 47
Appendix 3: Boxplots ... 50
Appendix 4: Survival Curves ... 51
5
1. Introduction 1.1 Traditional to Online Media
Every day, consumers see thousands of advertisements (Elsen, Pieters, & Wedel, 2016). A lot has changed since the beginning of advertising. For a long time, television was the most important medium for brands to show their advertisements to consumers (Li & Lo, 2015). While traditional television advertising is still growing (Deng & Mela, 2018), advertisers encounter the trend of digital advertising. In 2017, $209 billion was spent on digital advertising compared to $178 billion on television advertising (Kafka & Molla, 2017). Since the first online ad appeared in 1994, traditional media are making way for online advertising (Liu-Thompkins, 2019). Moreover, the traditional print media (i.e., newspapers and magazines) are developing online items to cope with this trend (Goodrich, Schiller, & Galletta, 2015).
The development can also be observed in the behavior of consumers. The trend of spending less time on traditional television is occurring across all generations (Nielsen, 2016). Nowadays, online platforms (i.e., Twitter, Facebook, YouTube) attract hundreds of millions of visitors (Grewal, Bart, Spann, & Pal, 2016). The amount of time consumers spend on the internet is increasing every year (Nielsen, 2016). In 2012, 60 hours of video were uploaded on YouTube every minute (Pashkevich, Dorai-Raj, Kellar, & Zigmond, 2012). This number has grown to more than 500 hours of video every minute in 2020 (YouTube, 2020). According to Zenith (2018), internet consumption will overtake television consumption in 2020. Therefore, brands should adjust their advertising strategy to remain relevant in the world of millions of advertisements.
Given that social media has become widely used by consumers, brands are gradually responding to this development (Dehghani, Niaki, Ramezani, & Sali, 2016). Over the years, more formats within online advertising became available, such as banner ads, mobile advertising, and retargeted advertising (Liu-Thompkins, 2019). As a result, brands are shifting their advertising investments from traditional media to online media. Chakraborty, Basu, and Ray (2020) predict that more than 50% of the global total advertising expenditures are made in the online advertising category by 2020.
6
video advertisements that users watch is observed (Li & Lo, 2015). Consequently, the most increasing revenue is found in the online video advertising category compared to the other categories. In the US, the revenue from online video advertisements has grown by 37.2% in 2018 (Chakraborty et al., 2020). These developments show that brands are investing more in online video advertisements.
1.2 Online Video Advertising
Numerous online platforms are compatible with video advertisements, for instance, YouTube, Facebook, Vimeo, and news websites. According to Dehghani et al. (2016), the biggest global brands acknowledge YouTube as a crucial aspect of their marketing strategy. Founded in 2005, the number of YouTube visitors is increasing every year. In 2012, more than 12 million visitors searched for videos on the platform (Pashkevich et al., 2012). Nowadays, YouTube attracts more than unique 2 billion visitors every month. Each day, more than a billion hours of videos are watched, and billions of views are generated (YouTube, 2020). Thus, YouTube offers brands interesting advertising opportunities.
7
to watch the video. In contrast to other forms of online advertising, pre-roll ads will cause a bigger feeling of intrusiveness because the desired content of the visitor is delayed and interrupted (Campbell et al., 2017; Li & Lo, 2015). Therefore, the consumer can decide to immediately skip the video ad when possible to diminish this negative feeling.
Advertisers only pay for TrueView when the visitor clicks on the ad or when the visitor has watched at least 30 seconds of the ad. In the case of a video shorter than 30 seconds, the advertiser only pays when the ad is watched completely (Pashkevich et al., 2012). This prerequisite makes TrueView attractive for brands and has led to an increase in skippable ads on YouTube (Pashkevich et al., 2012). Moreover, brands have more control over online advertisements than over traditional advertisements (Li & Lo, 2015). The vast majority of the advertisements are skippable: approximately 85% of the video ads on YouTube are skippable ads (Belanche, Flavián, & Pérez-Rueda, 2017b). Thus, it is important to understand how skippable pre-roll ads work and how consumers respond to these ads.
1.3 Research Question and Relevance
The shift from traditional advertising towards online video advertising requires an understanding of the consumer online watching behavior. However, a lot is still unknown about this emerging medium and therefore the research question that this study aims to answer is: When do consumers skip online video advertisements?
This question will be answered by investigating different factors. First, Puccinelli, Wilcox and Grewal (2015) state that the attitude of consumers towards the ad influence their watching time: What is the effect of ad attitude on the skipping behavior of consumers?
Second, as Bellman, Beal, Wooley, and Varan (2020) stated in their article, it is expected that the type of device has an effect on the viewing time. This leads to the following sub question: What is the effect of the type of device on the skipping behavior of consumers?
Third, the advertisements that will be shown in this study are from different degree of product involvement types (Jeon et al., 2019) and different ad lengths (Belanche, Flavián, & Pérez-Rueda, 2017a):
8
What is the effect of the ad duration on the skipping behavior of consumers?
Fourth, it is possible that the skipping behavior can be explained by demographics of visitors such as countries (Li & Lo, 2015; Seyedghorban et al., 2016). Next to this, the respondents of this study will vary in age and gender:
What is the effect of country of origin on the skipping behavior of consumers? What is the effect of age and gender on the skipping behavior of consumers?
The relevance of this study will be the contribution to the current literature on the skipping behavior of online video advertisements (Belanche et al., 2017a; Campbell et al., 2017). Although the studies focus on the same dependent variable, this study will focus on different factors influencing the behavior of consumers such as ad attitude, product involvement, and type of device. Specifically, the effect of these factors on the timing of skipping the online video ad will be investigated. Little is currently known about skipping pre-roll advertisements while it is an emerging form of online advertising.
1.4 Structure of the Paper
9
2. Theoretical Framework 2.1 Changing Role of Visitors
Traditional television advertisements and modern online video advertisements are both concentrated on activating the same sensory systems. However, the ads contrast in the way they are received; they are experienced in different media environments (Li & Lo, 2015). While traditional media is focused on a “captive audience”, online media is centred towards an “active audience” (Belanche et al., 2017a). When watching television, consumers are in a passive state of mind, considering that they cannot influence the duration and the variety of advertisements (Liu & Shrum, 2002). When browsing on the internet, consumers are actively searching for information or pleasure. They are fully focused on the online content they will consume (Campbell et al., 2017). Pre-roll advertisements are a form of interactive advertising since the user is in control over the course of the ad (Gao, Rau, & Salvendy, 2010). Visitors are given the power to skip the ad after a few seconds or to watch the ad for some more time, or to watch the ad completely. While traditionally consumers are zapping to the next TV channel during a disliked or annoying advertisement, consumers are now skipping online video ads when it is perceived as a nuisance (Tuchman, Nair, & Gardete, 2018).
Online video advertisements lead to opportunities for interactivity between consumers and the brand. Interactivity can be conceptualized as the ability of the consumer to control and modify the content (Liu & Shrum, 2002). Hoeck and Spann (2020) define online advertising interactivity as the extent to which consumers experience a two-way communication with the brand. Advertisers give the consumer control over the advertisement by, for instance, allowing the consumer to click on it. In the case of online video advertisements, visitors are not authorized to fast-forward the advertisement, nonetheless, they have the possibility to skip the ad after a few seconds (Li & Lo, 2015).
10
on the goal of the internet activity. Liu and Shrum (2002) argue when the visitor is searching for information, a high amount of control is preferred. However, when the customer is browsing for pleasure, a lower amount of control satisfies. The level of control is low in traditional advertising and high in online advertising. Hence, the majority of visitors prefer interactive online video advertising over traditional advertising.
For brands, the major benefit of skippable online video advertising is the reduced negative attitude towards the advertisement by enabling interactivity (Belanche et al., 2017b). Furthermore, the perception of the website, in this case YouTube, advances. This could lead to a higher customer return rate, which gives brands more advertising opportunities.
2.2 Skipping Behavior
The online advertising trend allows brands to gain rich data about online advertisements. It is possible to see who watched the ad, for how long, and whether the customer clicked on the ad. Therefore, brands can analyse the performance of their online campaigns (Johnson, Lewis, & Nubbemeyer, 2017). The advertisement itself and the goals of the campaign determine the effectiveness of the ad (Belanche et al., 2017a). Effective advertising affects brand awareness, associations, and attitudes (Teixeira, Wedel, & Pieters, 2010). Moreover, a lower skipping rate leads to higher brand recall (Pashkevich et al., 2012). Hence, brands are trying to balance achieving their marketing goals and avoiding user irritation (Goodrich et al., 2015).
11
Puccinelli et al. (2015) and Teixeira et al. (2010) state that ad watching is an important indicator of ad acceptance and thus the skipping behavior of consumers. Ad acceptance can be defined as the willingness of the visitor to obtain the information of the advertisement and to engage in activities (i.e., time or effort) for a relationship with the brand (Sultan et al., 2009). The ad acceptance of a video advertisement is high when a visitor wants to see the video and watches it for some time. Thus, when the user does not skip the video advertisement immediately. Brands are not always aware of the potential negative occurrence consumers go through when being interrupted by a video ad (Goodrich et al., 2015). The user’s response to this interruption is determined by several characteristics of interruption. For example, frequency, duration, timing, context, and complexity influence the reaction of the visitor to the in-stream ad. The interruption may lead to information overload and cognitive fatigue (Kirmeyer, 1988). Consequently, visitors might leave the site early or they might not return due to this negative experience (Pashkevich et al., 2012). Moreover, Kirmeyer (1988) argues that people try to reduce the duration of the interruption. This would mean that users who perceive the pre-roll ad as intrusive, would skip the ad as soon as possible. This intrusiveness is diminished by giving the visitors control over the ad, by giving the ability to skip the advertisement after 5 seconds (Pashkevich et al., 2012). Moreover, this indicates that brands should have knowledge about the factors that influence the skipping behavior and especially which elements decrease ad skipping (Campbell et al., 2017).
2.3 Conceptual Model
12 Figure 1: Conceptual Model
2.4 Advertising Attitude
MacKenzie and Lutz (1989) define ad attitude as “a predisposition to respond in a favourable or unfavourable manner to a particular advertising stimulus during a particular exposure occasion”. While this definition is based on traditional advertising, it is still relevant for online advertising. Factors that influence the advertising attitude are fun, entertainment, humour, relevance, and meaningfulness. Annoyance and intrusiveness negatively affect advertising attitude (Goodrich et al., 2015). Thus, an interruptive ad will not result in a high ad attitude. To obtain positive results of the advertisement, viewers should like the ad (Goodrich et al., 2015). Moreover, Puccinelli et al. (2015) found that the emotions of the visitor influence the watching time of the advertisement. Therefore, brands should try to make the advertisement as relevant and meaningful as possible. Then, visitors will keep watching the ad. However, this is a challenge to do as brands only have 5 seconds to convince the consumer.
13
H1: Advertisement attitude negatively affects the customer’s probability to skip pre-roll
advertisements
2.5 Product Involvement
Product involvement can be defined as “the perceived value of a goal object that manifests as interest in that object” (Stewart, Kammer-Kerwich, Auchter, Dunn, & Cunningham, 2019). The product category determines the product involvement; a high-involvement product category leads to a higher product involvement (Bellman et al., 2020). More risks are associated with buying a involvement product compared to a low-involvement product. Therefore, a high-involvement product requires an effortful, information-driven decision-making process while a low-involvement product does not require this as much (Andrews, Durvasula, & Akhter, 1991). High-involvement product categories have a high level of personal importance (Stewart et al., 2019). A low-involvement situation does not require much effort in the decision-making process, such as the daily visit to the supermarket (Hoyer, 1984) or buying fast food (Bellman et al., 2020). Thus, the degree of product involvement differs per product category (Brown & Stayman, 1992). Examples of high-involvement product categories are banking, cars, computers, and wireless services. Examples of low-involvement product categories are hair-care products, pet food, and public service announcement (Bellman et al., 2020; Trivedi, Teichert, & Hardeck, 2019). After all, it takes more cognitive effort to buy a computer than food and drinks.
A high involvement product has a positive effect on attention (Petty, Richman, Schumann, & Strathman, 1993) and therefore on the watching time (Bellman et al., 2020). YouTube visitors are in a high state of attention because they want to see a particular video. Therefore, the attentive visitors will be more interested in the high-involvement product category advertisement and will watch this ad longer and skip less frequently compared to ads of low-involvement product categories.
14
consumer responses compared to high-involvement products. This would mean that the watching time is higher for low-involvement products and fewer consumers skip the ad of a low-involvement product. On the other hand, Belanche et al. (2017a) and Campbell et al. (2017) contradict this finding of Stewart et al. (2019). They argue that traditional television advertising and online video advertising differs because consumers are more actively searching for content in the online world than in the offline world. These contradicting findings support the need for more research on the effect of product involvement on ad effectiveness. Based on the above information, the following hypothesis is formed:
H2: Product involvement negatively affects the customer’s probability to skip pre-roll
advertisements
2.6 Type of Device
Consumers watch YouTube videos on computers (desktops), tablets, or mobile phones. The preferred device has shifted over the years. Consumers used to access the internet on their desktop computers. Currently, laptops, tablets, and mobile phones are widely used for online browsing (Brasel & Gips, 2014; Pescher, Reichhart, & Spann, 2014). This development is also noticeable in the behavior of YouTube’s visitors. More than 70% of YouTube watch time is spent on mobile devices (YouTube, 2020). Moreover, the revenues from video advertisements grow differently for various devices. In 2018, the revenue of online video ads watched on desktops has grown by only 6.6% while the revenue of online video ads watched on mobile has grown by 65% (Chakraborty et al., 2020). Thus, it may be beneficial for brands to investigate which device will guarantee the highest ad watching.
Brasel and Gips (2014) argue that the device used to reach online content is as important as the content itself in online consumer behavior. The most obvious difference between desktops and mobile phones is the screen size. The screen of a desktop is much larger compared to the screen of a mobile phone. This difference impacts the way video advertisements are perceived (Stewart et al., 2019). Next to this, consumers use their laptop or computer to search for information and their mobile devices for relaxation (Stewart, et al., 2019). Therefore, consumers will more actively look at the ad on a desktop compared to a mobile device.
15
2016; Stewart et al., 2019). Mobile devices or tablets can be defined as touch-based devices. When touching something, the perceived ownership of the object increases (Peck & Shu, 2009). As a result, touchscreens lead to a higher feeling of psychological ownership and endowment in comparison to touchpads and non-touch interfaces such as a computer desktop (Brasel & Gips, 2014). While not much research has been done on the effect of type of device on skipping behavior, it can be argued that seeing an ad while wanting to watch a video may be perceived as more intrusive on a mobile phone or a tablet because consumers are touching the object. Therefore, the expectation is that a touchscreen gives the visitor more reason to skip the pre-roll advertisement. This results in the following hypothesis:
H3: Watching an online video ad on a touch-based device (vs. non-touch-based devices)
positively affects the customer’s probability to skip pre-roll advertisements
2.7 Advertising Duration
16
H4: The advertising duration positively affects the customer’s probability to skip
pre-roll advertisements
2.8 Gender
When watching commercials on TV, men are more likely to skip the ad than women (Teixeira et al., 2010). The same effect is expected for online advertisements. Belanche et al. (2017b) found in their study that women watch the ad for a longer period than men. They argue that men are more eager to reduce the ad watching time because they are more focused on task performance. Since the ad is not what they desire to see, it is expected that men watch the ad shorter. Hence, the following hypothesis is made:
H5: Men (vs. women) have a higher probability to skip pre-roll advertisements
2.9 Age
Social media is a new phenomenon that developed in the past years. Because the usage of online media varies between generations, it is expected that they handle it differently. No less than 90% of the younger people in Europe makes use of the internet daily (Fritz, Sohn, & Seegebarth, 2017). Moreover, millennials spend more time on online media compared to older generations (Nielsen, 2016). Since younger people make more use of online media (Dehghani et al., 2016), they are likely more skilled in this area (Gao et al., 2010). They see for example immediately the skip button and will, when possible, click on the skip button. The older generation, on the other hand, is generally not that skilled in online media. This may result in a slower reaction towards the skip button or no action at all.
According to Belanche et al. (2017b), age has a positive effect on watching time of the advertisement. The study found that young people watch an ad shorter compared to older people. Moreover, Heeter and Greenberg (1985) observed that young adults are zapping more often during the commercial on TV than older adults. Based on the literature, the following hypothesis is formed:
17
3. Research Design 3.1 Data Description
To answer the research question, three datasets from the marketing and research agency DVJ Insights will be used. The datasets contain information about a questionnaire of skippable online videos. These three datasets include the same variables but represent three different countries: The United Kingdom, Germany, and The Netherlands. DVJ Insights recruits consumers who answer questions against payment. Respondents can fill in these questionnaires from home on any device. The datasets with data from the three different countries are merged into one large dataset (n = 9,691).
The questionnaire aims to look like a normal browsing session. Customers can browse several websites and are exposed to different advertisements in random order. They can browse and switch between websites as the respondent normally does. Two of the six websites are video content platforms such as YouTube. Here, an online video advertisement is shown to the respondents before the content video. The respondent can skip the ad after 5 seconds by clicking on the “Skip advert” button. The respondent is asked different questions about the ad they watched. After these questions, the respondent sees the first video advertisement again without the possibility to skip the ad after 5 seconds. Now, the respondent evaluates the digital advert. 16 statements are shown, and the respondent answers these questions on a 5-point Likert scale (1: totally disagree – 5: totally agree). Examples are “this advert is likeable” and “this advert is irritating”. This process is repeated for the second video advertisement.
3.2 Duration Models
18
Thus, it is unknown what the respondent would do if the duration of the ad would be longer, after the observation period. For instance, how would the customer react to an advertisement of 1 hour? Therefore, the data is right-censored (Leeflang et al., 2015). Since only a certain time period is observed for this study, the event of skipping may not have taken place during time T. To account for right-censoring, a hazard model is appropriate (Leeflang et al., 2015). In this study, the probability that the customer has not skipped during the time interval (the duration of the ad), given that it has not occurred before t, is specified as:
𝑃(𝑡 ≤ 𝑇 ≤ 𝑡 + ∆𝑡|𝑇 ≥ 𝑡) (1) Where,
𝑇 = random interpurchase time variable
𝑡 = time
Since the duration variable is right-censored, the interpurchase time is not fully observed (Leeflang et al., 2015). The survival function shows the probability that skipping has not yet occurred at time t, which is formulated as:
𝑆(𝑡) = 𝑃(𝑇 > 𝑡) (2)
3.3 Dependent Variable
The dependent variable is the time of skipping, which is defined as the number of seconds the consumer has watched the online video advertisement before abandoning the content. It is observed whether a consumer skipped the ad, and the moment of skipping. Two new variables are made for the dependent variable. First, the variable time is generated. To start, the variable watching time shows the time in seconds that a respondent watches the ad. It is calculated by multiplying the ad duration (in seconds) with the percentage of the ad watched (engagement). However, watching time includes the first 5 seconds of the ad. This cannot be skipped by the visitor and therefore 5 seconds are subtracted. This results in the variable time.
19
entirely, the engagement score is 1 and hence the dummy skip has the value 0. If the engagement score is lower than 1, the consumer has skipped the ad and consequently, the dummy skip will be a 1.
3.4 Independent Variables 3.4.1. Ad Attitude
Some statements used in the questionnaire for investigating the ad attitude of respondents correspond to the literature (Goodrich et al., 2015; Lee & Lim, 2010; Mackenzie & Lutz, 1989; Ng & Houston, 2009). Other statements are expected to influence the ad statement because they are comparable to the items found in the literature. However, it is necessary to check if this holds also in this study. All the statements are measured on a 5-point Likert scale. First, question 2 (“This advert is irritating”) is reversed. After that, various tests are conducted to measure the validity of the variables. The KMO is 0.97, which is largely above the required 0.5 (Malhotra, 2009) and hence the sampling is adequate. The Bartlett’s test is significant (p-value < 0.001), which indicates that all the variables are correlated. However, two questions are further investigated because of the low correlations with the other questions. The questions deviate too much from the rest of the questions and are therefore deleted from the survey.
This leads to the principal component analysis (PCA). The results show that the first principal component explains a large part of the data (61.57%). After the first component, the proportion of the variance explained is very small (< 5%). The eigenvalues are also very low (< 1). Furthermore, the scree plots and the parallel analysis show that 1 factor is adequate (see Appendix 1). Hence, 1 factor is appropriate. Next, the PCA is again performed but now with the oblique rotation. This method is chosen because it gives a more realistic view of the data. The proportion of the variance explains by this factor is 61.6%, which is sufficient (Malhotra, 2009). Next, the Cronbach’s alpha is calculated to measure the internal consistency of this new variable. The Cronbach’s alpha has a value of 0.95. No improvement will be reached when one of the items will be removed. Based on these tests, the 14 items are merged into the variable ad attitude, which is the unweighted average. An overview of the statements and the corresponding loadings can be found in Table 1.
Statement Loading
20
This advert is funny 0.660
This advert is relevant to me 0.809
This advert is credible 0.720
This advert is distinctive 0.720
This advert is one I would like to share with others 0.806
This advert gives me a more positive impression of [BRAND] 0.843
This advert fits [BRAND] 0.640
This advert gives me energy 0.800
This advert gives me a positive feeling 0.856
This advert makes me feel close to the brand 0.847
This advert has made me more interested in [BRAND] 0.862
This advert makes me plan to buy (at) [BRAND] 0.826
This advert tells me something new 0.712
Table 1: Statements Ad Attitude
3.4.2 Product Involvement
To determine the product involvement, the 94 advertisements are analysed and placed into 29 different product categories. These product categories are based on the research of Bellman et al. (2020), Campbell et al. (2017), Hoyer (1984), and Trivedi et al. (2019). The most common product categories are food and drinks (10), supermarkets (8), insurance (7), cars (6), and telecommunications (6). An overview of all the product categories can be found in Appendix 2. Based on the product category, the product involvement is determined. A dummy high involvement is made which indicates high involvement (= 1) or low involvement (= 0).
3.4.3 Type of Device
Respondents can fill in the questionnaire on the type of device they prefer, smartphone, tablet, or desktop. A dummy variable device is created to indicate whether the pre-roll advertisement is watched on a touch-based device (smartphone or tablet = 1) or on a non-touch-based device (desktop = 0).
3.4.4 Duration of the Ad
The 94 advertisements have different lengths, from 10 to 20.9 seconds.
3.4.5 Gender
The questionnaire contains a variable which indicates whether it was filled in by a male or female. This variable is changed into a dummy variable: male (= 0) and female (= 1).
3.4.6 Age
21
3.4.7 Nationality
Before the three datasets were merged, a new variable nationality is made to indicate the nationality of the respondent (The United Kingdom = 1, Germany = 2, and The Netherlands = 3). Next, two dummies are created: United Kingdom (= UK) and Germany with reference level The Netherlands.
3.5 Data Cleaning and Outliers
Since the dataset is already prepared by DVJ Insights, not many striking values are found. For instance, DVJ Insights already deleted two video advertisements where respondents were not able to skip the ad. However, when investigating the dataset, 36 NA’s are detected for the variable engagement. According to DVJ Insights, these missing values can be explained by the fact that some respondents use an outdated browser. The mechanism used to count the number of seconds that the respondent watches the video is not working smoothly in old browsers. Therefore, it is impossible to measure the watching time of these respondents accurately. These respondents are deleted from the dataset. Moreover, the variable “watching time” contains some odd values. The watching time should be at least 5 seconds since visitors only are allowed to skip the ad after 5 seconds. However, 16 values are below 5 seconds. The reason behind this is the same as for the NA’s, people with an old browser can modify the timer of the ad watched. Hence, these observations are not reliable and are deleted from the dataset. After cleaning the dataset, 9,639 observations are left. Furthermore, the data is checked for consistency. This is done by making boxplots (see Appendix 3 for some examples) and checking for extremely high or low values. No outliers were found in the dataset.
3.6 Model Specification
A Cox proportional hazard model will be applied as it allows including effects of explanatory variables (Cox, 1972). The outcome of the model is the hazard rate, which is the probability that the event happens at or before time t. Based on the abovementioned variables, the model of this study is specified as:
22 Where,
ℎ𝑖 = probability that respondent i skips the advertisement at time t,
𝑡 = time in seconds,
𝑖 = respondent,
𝑉𝑇𝑖 = viewing time of the advertisement by respondent i in seconds, 𝑆𝑖 = event of skipping of respondent i,
ℎ0(𝑡) = baseline hazard at time t,
𝐴𝐴𝑖 = ad attitude of the advertisement by respondent i,
𝑃𝐼𝑖 = product involvement of the advertisement watched by respondent i, 𝐷𝑢𝑟𝑖 = duration of the advertisement in seconds watched by respondent i,
𝐷𝑒𝑣𝑖 = type of device on which respondent i watched the advertisement (non-touch-
= based device = 0, touch-based device = 1), 𝐺𝑒𝑛𝑑𝑖 = gender of respondent i (male = 0, female = 1), 𝐴𝑔𝑒𝑖 = age of respondent i,
𝑈𝐾𝑖 = dummy indicating whether respondent i is from the UK (= 1) or not (= 0), and 𝐺𝐸𝑖 = dummy indicating whether respondent i is from Germany (= 1) or not (= 0).
3.7 Descriptive Analysis
The dataset includes 30 to 32 video advertisements of different brands per country. Each respondent answers questions about two randomly assigned video advertisements. This results in approximately 100 observations per video advertisement.
As many men (49.2%) as women (50.8%) took part in the questionnaire. The youngest respondent is 18 years old and the oldest
respondent is 90 years old. As can be seen in the boxplot, there are few elderly respondents. Most of the respondents are between 18 and 74
years old. Figure 2: Boxplot Age
23
First, differences are noticed between the three countries. As can be seen in Table 2, the majority of the respondents from the United Kingdom and Germany used a smartphone to fill in the survey. On the contrary, in the Netherlands, the majority of the respondents fill in the survey on a desktop. In all three countries, a tablet is the least used for the survey.
Smartphone Tablet Desktop
United Kingdom 56.54% 6.16% 37.30%
Germany 61.45% 3.39% 35.16%
The Netherlands 28.01% 3.52% 68.46%
Table 2: Type of Device and Nationality
Second, Table 3 below shows that there is a small difference in the usage of devices between men and women. On average, women fill in the survey on a smartphone and men on a desktop.
Smartphone Tablet Desktop
Male 43.73% 3.55% 52.72%
Female 53.89% 5.12% 40.99%
Table 3: Type of Device and Gender
Third, the younger the respondent, the higher the chance that he or she uses a smartphone to fill in the survey (see Table 4). For example, in the category of 18-24 years old, 76% of the respondents used a smartphone. The older the respondent, the higher the chance that he or she uses a desktop for the survey. For example, 86% of the respondents between 85 and 90 years old used a desktop. A small percentage of the respondents used a tablet, this percentage grows when respondents are older compared to younger respondents.
Smartphone Tablet Desktop
18-24 75.88% 0.79% 23.33% 25-34 68.24% 1.89% 29.87% 35-44 64.66% 3.29% 32.05% 45-54 45.00% 4.36% 50.64% 55-64 27.79% 6.18% 66.02% 65-74 13.42% 8.80% 77.78% 75-84 8.37% 11.41% 80.23% 85-90 0.00% 14.29% 85.71%
Table 4: Type of Device and Age
Table 5 shows all the descriptive statistics for all the variables of the study.
Variable Descriptives Explanation
Time Mean: 6.318
Sd: 5.325
24
Skip 0: 4052
1: 5587
0 = no 1 = yes
Advertisement Attitude Mean: 2.644
Sd: 0.782
5-point Likert scale: 1 = totally disagree, 5 = totally agree
Product Involvement 0: 4589 1: 5050 0 = low involvement 1 = high involvement Type of Device 1: 4713 3: 419 5: 4507 1 = mobile 3 = tablet 5 = desktop
Advertisement Duration Mean: 16.56
Sd: 3.473
In seconds, including the first 5 seconds Gender 0: 4738 1: 4901 0 = male 1 = female Age Mean: 45.13 Sd: 16.950 In years Nationality 1: 3182 2: 3305 3: 3152 1 = United Kingdom 2 = Germany 3 = The Netherlands Table 5: Summary of the Variables
3.8 Multicollinearity
It is important to check for correlation between the explanatory variables since the presence of multicollinearity leads to unreliable parameter estimates (Leeflang et al., 2015). To check for multicollinearity issues, the variance inflation factor (VIF) scores are measured. According to O’Brien (2007), the right threshold value is somewhat arbitrary but most common values between 4 and 10. In Table 6, the VIF scores of the variables are shown. Since all variables have a VIF score around 1, no multicollinearity issues are assumed.
Ad Attitude Product Involvement Type of Device Ad Duration
Gender Age UK Germany
1.028 1.007 1.384 1.092 1.016 1.317 1.494 1.642
25
4. Results 4.1 Analysis of Skipping Behavior
Respondents watched on average 6,318 seconds after the compulsory first 5 seconds. 58% of the respondents (5,587) skipped the video advertisement, whereas 42% of the respondents (4,052) watched the full length of the video ad. Figure 3 shows when respondents click on the skip button. The dotted lines indicate the first, second, and third quartile. 75% of the skipped advertisements are skipped within 4 seconds after the skip button became available. Thus, if respondents want to skip the ad, they only wait a few seconds to do this.
Figure 3: Cumulative Number of Ads Skipped
4.2 Kaplan-Meier Method
26 Figure 4: Survival Curve
Furthermore, the Kaplan-Meier survival function compares the observed and expected number of events of each group over time. The null hypothesis states that there is no difference between the groups. This hypothesis cannot be rejected for gender (p-value = 0.099) and product involvement (p-value = 0.12). Thus, the survival probability is the same over time for men and women, and for products with a high and a low involvement (see Figure 5). However, the survival curve is significantly different for type of device value < 0.0001) and nationality (p-value < 0.0001). Respondents watching the ad on a touch-based device have a higher probability to skip the advertisement and thus have a lower survival probability compared to respondents who watch the ad on a desktop. Moreover, respondents from Germany (nationality = 2) are more likely to skip the ad (see Figure 6). Customers from the United Kingdom and The Netherlands have the same survival probability at the beginning of t. However, their survival probability also changes over time. To test whether these similarities or differences are significant, more tests will be performed. The survival curves of all the variables can be found in Appendix 4.
27
4.3 Cox Model 4.3.1 Model Fit
A model is created to calculate the consumer’s probability to skip: Cox model. To determine the model fit, the pseudo R2s, the likelihood ratio, and the log-likelihood and are estimated (Gupta, 1991; Leeflang et al., 2015). Moreover, the values of the estimated model and the null model are compared. The null model has the same dependent variables but does not have any independent variables. The likelihood ratio shows that the Cox model is significantly better than the null model (p-value < 0.001, see Table 7). Moreover, the AIC and the BIC both indicate that the Cox model has a better model fit than the null model. The Pseudo R2s are low, suggesting that there are other variables outside the model influencing the skipping behavior. The explanatory power of the variables of the Cox model is therefore not very high, but the explanatory power is higher than the null model. The concordance is a measure of goodness-of-fit in survival models. A pair of observations is defined as concordant if the prediction behaves similarly to the data (Therneau & Atkinson, 2020). A random guess would result in a concordance of 0.50. According to Therneau and Atkinson (2020), a value below 0.55 is not very good. The model shows a concordance of 0.657, a sufficient result for survival data. The model has a probability of 65.7% to predict the data correctly.
Pseudo R2
McFadden
Pseudo R2
Cox & Snell
Pseudo R2
Nagelkerke
AIC BIC LL Likelihood
ratio test
Cox model 0.013 0.120 0.120 96,556.74 96,609.77 -48,270.37 1,237***
Null model - - - 97,778.21 97,778.21 -48,889.11 -
*** p-value < 0.001
Table 7: Model Comparison
4.3.2 Estimations
The estimations of the Cox Model are shown in Table 8. The product involvement, the ad duration, and the dummy UK do not have a significant effect (p-value > 0.05) on the probability to skip. The other variables show a highly significant effect (p-value < 0.01). The dummy Germany has a weakly significant effect (p-value < 0.1).
touch-28
based device (smartphone or tablet) vs a desktop. The probability to skip the ad for someone using a smartphone or tablet increases by 19.6% compared to someone using a desktop. Hypothesis 3 is accepted. Third, gender has a significant negative effect (beta = -0.086; exp(beta) = 0.917; p-value = 0.001). The probability to skip the ad is 8.3% lower for women vs men. Thus, men skip online video advertisements more often than women do. Hypothesis 5 is accepted. Fourth, age has a significant negative effect (beta = -0.025; exp(beta) = 0.975; p-value < 0.001). The probability to skip the ad decreases when age increases. The probability to skip an ad decreases by 2.5% for each year a customer is older. Thus, the younger the visitors, the higher the chance that he or she will skip the online video advertisement. Hypothesis 6 is accepted.
Interestingly, the dummy Germany is weakly significant (beta = -0.064; exp(beta) = 0.938; p-value = 0.078). This means that there is a small difference between visitors from Germany and the reference level of the Netherlands. When the visitor is German, the probability is 6.2% lower that he or she will skip compared to a visitor from the Netherlands. No difference is found between customers from the United Kingdom and the Netherlands. Their skipping behavior is thus comparable. The product involvement does not influence the customer’s probability to skip. No significant effect (p-value = 0.596) is found of products with high or low involvement. Hypothesis 2 is not supported. Moreover, no significant effect (p-value = 0.780) is found of the advertising duration on the probability to skip the ad. Hypothesis 4 is not supported.
Variable Coefficient Exp(Coef) St. Error Z-value P-value
Ad Attitude -0.264 0.768 0.015 -17.529 0.000 Product Involvement 0.014 1.014 0.027 0.531 0.596 Type of Device 0.179 1.196 0.031 5.763 0.000 Ad Duration 0.001 1.001 0.004 0.279 0.780 Gender -0.086 0.917 0.027 -3.199 0.001 Age -0.025 0.975 0.001 -26.779 0.000 UK 0.045 1.046 0.035 1.296 0.195 Germany -0.064 0.938 0.036 -1.768 0.078 Log-likelihood -48,270.37
Likelihood ratio test 1,237***
AIC statistic 96,556.74
BIC statistic 96,609,77
Concordance 0.657
*** p-value < 0.001; ** p-value < 0.05; * p-value < 0.01
29
4.4 Interaction Effects
4.4.1 Interaction Effect with Type of Device
Next to the analysis of the main effects discussed in the previous chapter, an analysis of possible interaction effects will be performed (see Table 9). The interaction effects improve the model fit significantly (𝑋2 = 24.64, p-value < 0.001); the AIC of the new model is 96,539.85 (previously 96,556.74). The BIC has not improved. However, the values are very close to each other. The concordance of the model is 0.658, which is comparable to the previous model and a common result.
A significant interaction effect (beta = 0.117; exp(beta) = 1.124; p-value < 0.001) is found for ad attitude and type of device. The negative effect of ad attitude becomes less strong when the respondent watches the ad on a mobile phone or tablet vs desktop. In other words, the probability to skip the ad decreases with 28.2% for each unit increase in ad attitude when watching it on a desktop while it decreases with 19.3% when watching it on a smartphone or tablet. Moreover, the interaction effect between product involvement and type of device is investigated, which is not significant (p-value = 0.186). Thus, the type of device does not change the effect of product involvement on skipping behavior. Furthermore, the interaction effect between age and type of type is significantly positive (beta = 0.004; exp(beta) = 1.004; p-value = 0.025). This indicates that the negative effect of age becomes weaker when the respondent uses a touch-based device versus a non-touch-based device. The probability to skip decreases by 2.8% for each age year when using a desktop, this is 2.2% when using a smartphone or tablet. Both reductions of the negative effect can be explained by the finding that using a touch-based device increases the skipping probability and thus lowers the negative effect of variables on the skipping probability.
Variable Coefficient Exp(Coef) St. Error Z-value P-value
Ad Attitude -0.331 0.718 0.023 -14.676 0.000 Product Involvement -0.030 0.971 0.042 -0.713 0.476 Type of Device -0.464 0.629 0.0133 -3.493 0.000 Ad Duration 0.001 1.001 0.004 0.304 0.761 Gender -0.084 0.919 0.03 -3.130 0.018 Age -0.028 0.972 0.001 -21.660 0.000 UK 0.048 1.049 0.035 1.369 0.171 GE -0.045 0.956 0.036 -1.236 0.216
Type of Device x Ad Attitude 0.117 1.124 0.030 3.852 0.000
Type of Device x Product Involvement
30
Type of Device x Age 0.006 1.006 0.002 3.373 0.001
Log-likelihood -48,258.05
Likelihood ratio test 1,262***
AIC statistic 96,538.10
BIC statistic 96,611.01
Concordance 0.658
*** p-value < 0.001; ** p-value < 0.05; * p-value < 0.01
Table 9: Estimation Results with Interactions Type of Device
4.4.2 Interaction Effect with Nationality
As discussed in Chapter 3.7, some differences have been found between the three countries. To investigate the influence of these differences, interactions with the variables are added to the model. The results can be found in Table 10. The likelihood ratio test shows that the model is significantly better than the null model (p-value < 0.001). Adding the interaction effects of nationality improves the model fit significantly (𝑋2 = 39.457, p-value < 0.001) compared to the model without interactions. Moreover, it decreases the AIC from 96,556.74 to 96,529.29. Furthermore, the model with nationality interaction has a significant (𝑋2 = 14.817, p-value = 0.002) better model fit and a lower AIC than the previous model with the interaction of type of device. The BIC did not improve. The concordance of the model is the same as the previous interaction model: 0.658. Thus, the model predicts the probability sufficient.
31
The main effect of product involvement is not significant (p = 0.116). However, the interaction with Germany is significant (beta = 0.155; exp(beta) = 1.168; p = 0.020). When the respondent is from Germany, the effect of product involvement becomes stronger. The probability to skip increases by 8.4% when the respondent is from Germany and the ad promotes a high involvement product, compared to a low involvement product. There is no significant effect found of product involvement when the customer is Dutch. Moreover, no significant effect is found between Dutch and British respondents.
Variables Coefficient Exp(Coef) St. Error Z-value P-value
Ad Attitude -0.305 0.737 0.026 -11.613 0.000 Product Involvement -0.074 0.928 0.047 -1.572 0.116 Type of Device 0.308 1.360 0.050 6.210 0.000 Ad Duration 0.002 1.002 0.004 0.437 0.662 Gender -0.083 0.920 0.03 -3.086 0.020 Age -0.026 0.975 0.001 -27.263 0.000 UK 0.039 1.039 0.119 0.324 0.746 Germany -0.321 0.725 0.124 -2.584 0.010 UK x Ad Attitude -0.006 0.994 0.036 -0.155 0.876 Germany x Ad Attitude 0.118 1.126 0.037 3.176 0.001 UK x Type of Device -0.094 0.910 0.070 -1.338 0.181
Germany x Type of Device -0.304 0.738 0.069 -4.389 0.000
UK x Product Involvement 0.110 1.116 0.067 1.648 0.099
Germany x Product Involvement
0.155 1.168 0.067 2.333 0.020
Log-likelihood -48,250.64
Likelihood ratio test 1,277***
AIC statistic 96,529.29
BIC statistic 96,622.08
Concordance 0.658
*** p-value < 0.001; ** p-value < 0.05; * p-value < 0.01
Table 10: Estimation Results with Interactions Nationality
4.5 Proportional Hazards Assumption
32
increase over time. Thus, the assumption might be violated. In that case, the effect of the covariates on survival is not constant over time. To test whether the model satisfies the proportionality assumption, the Schoenfeld residuals are estimated, plots are created, and the interaction effect is investigated. The Schoenfeld residuals are based on the observed and the expected values of the covariates at each time point in period t (Schoenfeld, 1982).
First, the Schoenfeld residuals indicate that the proportional hazards assumption is violated by the variables age, duration of the ad, type of device, ad attitude, and the nationality dummy Germany (p-values < 0.05, see Appendix 5). This means that for example, the hazard ratio of touch-based devices vs. non-touch-based devices is not a fixed number. The hazard ratio should constant over time and not dependent on period t. These variables cause the whole model to violate the proportionality assumption.
Second, the plots show an estimate of the time-dependent coefficient beta(t). A horizontal line indicates that the proportionality assumption is satisfied (Therneau, Crowson, & Atkinson, 2020). However, as can be seen in Figure 7, the beta for age is not constant over time. The effect goes up over time, which means that the beta is overestimated. The other plots can be found in Appendix 5. For some variables, it is debatable whether the beta is indeed not constant as the Schoenfeld residuals suggest. For example, the beta of ad attitude is relatively constant over time. According to Therneau et al. (2020), the proportional hazards assumption is never precisely true. Therefore, the interaction effects of these variables with time will be studied.
33
Third, the interaction effects are investigated. The correct method to investigate survival data when non-proportional hazards are detected is dependent on the research question. This study predicts the survival time and therefore a time-dependent effect should be modelled (Dunkler et al., 2018). This is done by making the variables time-varying which are significantly violating the proportional hazards assumption according to the Schoenfeld residuals. An interaction effect is added with log(t+1) (Therneau, Crowson, & Atkinson, 2020). The only variable with a significant main and interaction effect is age. Significant interaction terms indicate that the respective coefficient violates the proportional hazard assumption (Dunkler et al., 2018). No significant interaction effect is found for ad attitude, type of device, ad duration, and the nationality dummy Germany. Thus, the insignificant interactions are excluded as they do not violate the proportional hazards assumption. This leads to the model with an interaction effect for age with time included.
The results of the new model are displayed in Table 11. The new model has a better model fit (decrease in AIC and BIC) compared to the previous three models. The significant likelihood ratio test indicates that the model is significantly better than the null model (p-value < 0.001). Moreover, the pseudo R2s have improved (McFadden pseudo R2 was 1.30%, now 1.60%; Cox & Snell and Nagelkerke pseudo R2 was 12%, now 15%). The explanatory power of the model has increased. Next to this, the concordance shows that the model predicts sufficient (concordance = 0.66). To conclude, the model with the time-varying interaction has a better model fit compared to the model without the time-varying interaction.
The effect of the variable age has changed a bit with the varying function. The time-dependent coefficient for age is estimated to be β(t) = -0.049 + 0.022 * log(t + 1) (Therneau et al., 2020). Thus, the beta of age at t = 0 is -0.049. The effect of age increases over time, with 0.022 for each unit of ln(time).
Variable Coefficient Exp(Coef) St. Error Z-value P-value
34
Log-likelihood -48,107.32
Likelihood ratio test 1,564***
AIC statistic 96,232.63
BIC statistic 96,292.29
Concordance 0.660
*** p-value < 0.001; ** p-value < 0.05; * p-value < 0.01
Table 11: Estimation Results with Time-Varying Interaction
Since the variable age depends on t, the covariate should have an index t. Therefore, the adjusted Cox proportional hazard model is as follows:
35
5. Discussion 5.1 Conclusions
In this study, the effects of several characteristics of the consumer and the advertisement on the timing to skip an online video advertisement are investigated. The results show that 58% of YouTube visitors skips the pre-roll advertisement. Moreover, when they skip the advertisement, the majority clicks on the button within 4 seconds. An overview of the confirmed and rejected hypotheses can be found in Table 12.
Hypothesis Supported?
H1 Advertisement attitude negatively affects the customer’s probability to skip pre-roll advertisements
Yes H2 Product involvement negatively affects the customer’s probability to
skip pre-roll advertisements
No H3 Watching an online video ad on a touch-based device (vs.
non-touch-based devices) positively affects the customer’s probability to skip pre-roll advertisements
Yes
H4 The advertising duration positively affects the customer’s probability to skip pre-roll advertisements
No H5 Men (vs. women) have a higher probability to skip pre-roll
advertisements
Yes H6 Age negatively affects the customer’s probability to skip pre-roll
advertisements
Yes Table 12: Summary of the Hypothesis-Testing Results
36
Holding a smartphone or tablet increases the perceived ownership of it (Peck & Shu, 2009). Therefore, seeing an online video ad on a smartphone or tablet is perceived as more intrusive compared to a desktop (Brasel & Gips, 2014). Consumers will try to reduce this feeling by skipping the ad. Fourth, hypothesis 4 is not confirmed indicating that the ad duration does not influence the skipping behavior of consumers. A possible explanation for this finding might be the lack of variety in the duration of the ads of this research. Only advertisements with a maximum duration between 10 and 20.9 seconds are investigated. However, customers perceive ads from 30 or 60 seconds as too long (Jeon et al., 2019). Hence, the ads used in this study might not give a realistic view of skipping behavior due to the lack of variety in ad duration. Fifth, hypothesis 5 is confirmed. This means that men skip online video ads more frequently than women. This effect has been found for traditional television (Teixeira et al., 2010), but not yet for pre-roll advertisements. It can be explained by the fact that the task orientation of men leads to a higher skipping rate than women. Men do not want to postpone the desired content and therefore they have a higher probability to skip the ad. Hypothesis 6 is confirmed, concluding that age affects the skipping behavior of the consumer. The younger the consumer, the more likely that he or she will skip the ad. Younger people are more skilled in the online world (Gao et al., 2010) and will therefore remark the skip button more often or faster compared to older consumers. Age has a time-varying effect on skipping behavior. The negative effect of age weakens over time, meaning that the probability to skip is stronger influenced by age at the beginning of the ad compared to the end of the ad.
37
higher involvement, they will skip less fast. On the other hand, Germans control the gratification of their desires (Hofstede, 2020). Therefore, this will cause a less strong reaction of German on a pre-roll advertisement.
5.2 Managerial Implications
During the past years, new advertising opportunities have gained popularity among marketers. A strong growth is found for online video advertisements and in particular pre-roll advertisements (Chakraborty et al., 2020). One problem encountered by marketers is the unknown motivations of watchers to skip the advertisement. Understanding why, and especially the timing of different target groups to skip the ad could improve the marketing results (Pashkevich et al., 2012). This research can help marketers by analysing and understanding their advertising results. The most important insights of this research are:
• Advertisement attitude has the highest impact on skipping behavior,
38
5.3 Limitations and Suggestions for Further Research
39
References
Andrews, J. C., Durvasula, S., & Akhter, S. H. (1991). A Framework for Conceptualizing and Measuring the Involvement Construct in Advertising Research. Journal of
Advertising, 19(4), 27-40.
Belanche, D., Flavián, C., & Pérez-Rueda, A. (2017a). Understanding Interactive Online Advertising: Congruence and Product Involvement in Higly and Lowly Arousing, Skippable Video Ads. Journal of Interactive Marketing, 37, 75-88.
Belanche, D., Flavián, C., & Pérez-Rueda, A. (2017b). User Adaptation to Interactive Advertising Formats: The Effect of Previous Exposure, Habit and Time Urgency on Ad Skipping. Telematics and Informatics, 34, 961-972.
Bellera, C. A., MacGrogan, G., Debled, M., de Lara, C. T., Brouste, V., & Mathoulin-Pélissier, S. (2010). Variables with Time-varying Effects and the Cox Model: Some Statistical Concepts Illustrated with a Prognostic Factor Study in Breast Cancer. BMC Medical Research Methodology, 10(20), 1-12.
Bellman, S., Beal, V., Wooley, B., & Varan, D. (2020). Viewing Time as a Cross-Media Metric: Comparing Viewing Time for Video Advertising on Television and Online. Journal of Business Reseach, 120, 103-113.
Brasel, S. A., & Gips, J. (2014). Tablets, Touchscreens, and Touchpads: How Varying Touch Interfaces Trigger Psychological Ownership and Endowment. Journal of Consumer Psychology, 24(2), 226-233.
Brown, S. P., & Stayman, D. M. (1992). Antecedents and Consequents of Attitude Toward the Ad: A Meta-Analysis. Journal of Consumer Research, 19, 34-51.
Campbell, C., Thompson, F., Grimm, P. E., & Robson, K. (2017). Understanding Why Consumers Don't Skip Pre-Roll Video Ads. Journal of Advertising, 46(3), 411-423. Chakraborty, S., Basu, S., & Ray, S. (2020). Advertisement Revenue Management:
Determining the Optimal Mix of Skippable and Non-Skippable Ads for Online Video Sharing Platforms. European Journal of Operational Research.
40
Cox, D. R. (1972). Regression Models and Life-Tables. Journal of the Royal Statistical Society: Series B, 34(2), 187-202.
Dehghani, M., Niaki, M. K., Ramezani, I., & Sali, R. (2016). Evaluating the Influence of YouTube Advertising for Attraction of Young Customers. Computers in Human Behavior, 59, 165-172.
Deng, Y., & Mela, C. F. (2018). TV Viewing and Advertising Targeting. Journal of Marketing Research, 55(1), 99-118.
Dunkler, D., Ploner, M., Schemper, M., & Heinze, G. (2018). Weighted Cox Regression Using the R Package coxphw. Journal of Statistical Software, 84(2), 1-26. Elsen, E., Pieters, R., & Wedel, M. (2016). Thin Slice Impressions: How Advertising
Evaluation Depends on Exposure Duration. Journal of Marketing Research, 53(4), 563-579.
Fritz, W., Sohn, S., & Seegebarth, B. (2017). Broadening the Perspective on Mobile Marketing: An Introduction. Psychology and Marketing, 34(2), 113-118.
Gao, Q., Rau, P. L., & Salvendy, G. (2009). Perception of Interactivity: Affects of Four Key Variables in Mobile Advertising. International Journal of Human-Computer
Interaction, 25(6), 479-505.
Gao, Q., Rau, P. L., & Salvendy, G. (2010). Measuring Perceived Interactivity of Mobile. Behaviour and Information Technology, 29(1), 35-44.
Goodrich, K., Schiller, S. Z., & Galletta, D. (2015). Consumer Reactions to Intrusiveness of Online-Video Advertisements: Do Length, Informativeness and Humor Help (or Hinder) Marketing Outcomes? Journal of Advertising Research, 55(1), 37-50. Grewal, D., Bart, Y., Spann, M., & Pal, P. P. (2016). Mobile Advertising: A Framework and
Research Agenda. Journal of Interactive Marketing, 34, 3-14.
Gupta, S. (1991). Stochastic Models of Interpurchase Time with Time Dependent Covariates. Journal of Marketing Research, 28, 1-15.
41
Heeter, C., & Greenberg, B. S. (1985). Profilling the Zappers. Journal of Advertising Research, 2, 15-19.
Hoeck, L., & Spann, M. (2020). An Experimental Analysis of the Effectiveness of Multi-Screen Advertising. Journal of Interactive Marketing, 50, 81-99.
Hofstede, G. (2020). Hofstede Insights. Retrieved 12 30, 2020, from Hofstede Insights: https://www.hofstede-insights.com/country-comparison/germany,the-netherlands/ Hoyer, W. D. (1984). An Examination of Consumer Decision Making for a Common Repeat
Purchase Product. Journal of Consumer Research, 11, 822-829.
Jeon, Y. A., Son, H., Chung, A. D., & Drumwright, M. E. (2019). Temporal Certainty and Skippable In-Stream Commercials: Effects of Ad Length, Timer and Skip-ad Button on Irritation and Skipping Behavior. Journal of Interactive Marketing, 47, 144-158. Johnson, G. A., Lewis, R. A., & Nubbemeyer, E. I. (2017). Ghost Ads: Improving the
Economics of Measuring Online Ad Effectiveness. Journal of Marketing Research, 54(6), 867-884.
Kafka, P., & Molla, R. (2017). 2017 Was the Year Digital Ad Spending Finally Beat TV. Retrieved November 19, 2020, from Recode:
https://www.vox.com/2017/12/4/16733460/2017-digital-ad-spend-advertising-beat-tv Kaplan, E. L., & Meier, P. (1958). Nonparametric Estimation from Incomplete Observations.
Journal of the American Statistical Association, 53(282), 457-481.
Kirmeyer, S. L. (1988). Coping with Competing Demands: Interruption and the Type A Pattern. Journal of Applied Psychology, 73(4), 621-629.
Lee, Y. H., & Lim, E. A. (2010). When Good Cheer Goes Unrequited: How Emotional
Receptivity Affects Evaluation of Expressed Emotion. Journal of Marketing Research, 47(6), 1151-1161.
Leeflang, P. S., Wieringa, J. E., Bijmolt, T. H., & Pauwels, K. H. (2015). Modeling Markets: Analyzing Marketing Phenomena and Improving Marketing Decision Making. New York: Springer.
42
Li, H., & Lo, H. Y. (2015). Do You Recognize Its Brand? The Effectiveness of Online In-Stream Video Advertisements. Journal of Advertising, 44(3), 208-218.
Liu, Y., & Shrum, L. J. (2002). What is Interactivity and is it Always Such a Good Thing? Implications of Definition, Person, and Situation for the Influence of Interactivity on Advertising Effectiveness. Journal of Advertising, 31(4), 53-64.
Liu-Thompkins, Y. (2019). A Decade of Online Advertising Reseach: What We Learned and What We Need to Know. Journal of Advertising, 48(1), 1-13.
MacKenzie, S. B., & Lutz, R. J. (1989). An Empirical Examination of the Structural
Antecedents of Attitude toward the Ad in an Advertising Pretesting Context. Journal of Marketing, 53(2), 48-65.
Malhotra, N. (2009). Marketing Research, an Applied Orientation. Person Education.
Ng, S., & Houston, M. J. (2009). Field Dependency and Brand Cognitive Structures. Journal of Marketing Research, 46(2), 279-292.
Nielsen. (2016). Their Generation: From Location to Listening Habits a Media Divide Exists According to Age. Retrieved October 29, 2020, from Nielsen:
https://www.nielsen.com/us/en/insights/article/2016/their-generation-from-location-to-listening-habits-a-media-divide-exists/
O'Brien, R. M. (2007). A Caution Regadering Rules of Thumb of Variance Inflation Factors. Quality and Quantity, 41(5), 673-690.
Pashkevich, M., Dorai-Raj, S., Kellar, M., & Zigmond, D. (2012). Empowering Online Advertisements by Empowering Viewers with the Right to Choose: The Relative Effectiveness of Skippable Video Advertisements on YouTube. Journal of Advertising Research, 52(4), 451-457.
Peck, J., & Shu, S. B. (2009). The Effect of Mere Touch on Perceived Ownership. Journal of Consumer Research, 36, 434-447.
43
Petty, R. E., Richman, S. A., Schumann, D. W., & Strathman, A. J. (1993). Positive Mood and Persuasion: Different Roles for Affect Under High- and Low-Elaboration Conditions. Journal of Personality & Social Psychology, 64(1), 5-20.
Prins, R., & Verhoef, P. C. (2007). Marketing Communication Drivers of Adaption Timing of a New E-service among Existing Customers. Journal of Marketing, 71(2), 169-183. Puccinelli, N. M., Wilcox, K., & Grewal, D. (2015). Consumers’ Response to Commercials:
When the Energy Level in the Commerical Conflicts with the Media Context. Journal of Marketing, 79(2), 1-18.
Robinson, H. R., & Kalafatis, S. P. (2020). Why Do People Choose to Multitask with Media? The Dimensions of Polychronicity as Driver of Multiple Media Use - A User
Typology. Journal of Advertising Research, 60(3), 251-270.
Rojas-Méndez, J. I., Davies, G., & Madran, C. (2009). Universal Differences in Advertising Avoidance Behavior: A Cross-Cultural Study. Journal of Business Research, 62(10), 947-954.
Schoenfeld, D. (1982). Partial Residuals for the Proportional Hazards Regression Model. Biometrika, 69(1), 239-241.
Segijn, C. M., Voorveld, H. A., Vanderberg, L., Pennekamp, S. F., & Smit, E. (2017). Insight into Everyday Media Use with Multiple Screens. International Journal of Advertising, 36(5), 779-797.
Seyedghorban, S., Tahernejad, H., & Matanda, M. J. (2016). Reinquiry into Advertising Avoidance on the Internet: A Conceptual Replication and Extension. Journal of Advertising, 45(1), 120-129.
Stewart, K., Kammer-Kerwich, M., Auchter, A., Koh, H. E., Dunn, M. E., & Cunningham, I. (2019). Examining Digital Video Advertising (DVA) Effectiveness: The Role of Product Category, Product Involvement and Device. European Journal of Marketing, 53(11), 2451-2479.
