Abstract
This study investigates the effect of several types of anti ad block strategies on the behaviour of website visitors. The analysis comprises an experiment where loyal versus non-loyal visitors’ behaviour of NU.nl was measured with regard to two forms of anti ad block strategies; one where an informative banner was shown on top of the website and one where it was forced to turn off ad block in order to visit the website. These strategies were compared to a reference group. Overall visitors reacted negatively to the anti ad block strategies by rating the website lower and neither were likely to visit the website again. Their intention to actually disable ad blocking tools remained the same across strategies. Side effect was a strong increase in privacy-related irritation among visitors, so website owners should be careful implementing an anti ad block strategy.Keywords: Online advertising, online privacy, e-loyalty, online behaviour,
Management Summary
The aim of this study is to address the increasing problem content providers face searching for a profitable business model. Currently the most important source of income for content providers is the use of online advertisements. However, due to the increasing use of ad blocking tools among Internet users, the main source of income for these website owners is under pressure. As a response to this development, the first anti ad block strategies have been born. In this research the effect of two frequently used anti ad block strategies are measured. The first one is an informative banner where in a banner on top of the website the user is alerted about their ad blocker use and informed about the business model of the website. The other one is a full-page pop up where users are forced to turn off their ad blocker in order to access content or the complete website.
Acknowledgements
This thesis is the final hurdle of my long but fruitful master experience. Through many ups and downs, hours of work and litres of coffee, the moment has finally come to graduate and find a suitable job. First of all, I want to thank my supervisor Lara Lobschat for the moments of giving clarity and direction to my research and providing me with numerous deadlines to keep me on track. Subsequently I want to thank my thesis group for the moments we shared our experiences and kept each other motivated during the journey. Also a special word of thanks to my MARUG committee for coping with my reduced social time but kept supporting me with substantive comments. Lastly thanks to my parents for coming to Groningen several times and cheering me up during this journey as time to travel home was scarce.
I hope you enjoy reading my thesis and look forward to seeing you all in the future.
Sebastiaan Jansen
1.
Introduction
Online content providers such as social networks and news sites nowadays collect a large heap of data from their users. This is where the actual value of these websites depends on, their user base (Tucker 2014). Although content providers offer their services mostly for free, they commonly generate an income by showing advertisements to the visitors (Nason 2010; Punj 2015; Schumann, von Wangenheim, and Groene 2014). The collected data is used by advertisers to personalize the advertisements shown on the website. However, collecting all this data causes reactance among visitors, as they feel invaded on their privacy (Tucker 2014). Reactance is a psychological behaviour where consumers resist to something they find intriguing by behaving in the opposite way (Middleton, Buboltz, and Sopon 2015). In this case, instead of getting exposed to advertisements, visitors block the advertisements. Installing an ad blocker is such an example of ad avoidance behaviour visitors exert if exposed to too intriguing advertisements (Seyedghorban, Tahernejad, and Matanda 2016).
If asked, most consumers are against the collection of privacy sensitive online data to be used by the advertisers (Alreck and Settle 2007). This so-called misuse of personal information is the main reason for users to install ad blocking software (Adobe 2015). Content providers will lose revenue if consumers decide to block advertisements out of privacy protection, because advertisements they paid for will not be displayed on the ad block user’s browser. This is considered to be an increasing problem for content providers, as it is their main source of income (Punj 2015). The cost of ad blocking for content providers is expected to be a staggering 20,3 billion dollar in 2016. This number is considered even low if you take into account that major websites relying on advertising revenue, for example Facebook, reported already 17 billion dollars on ad revenue in 2015 (Statista 2016).
As a response to declining online advertisement revenues due to ad blocking, some content providers on their turn start to take counter measures (Cummings 2015). By detecting users on their website that use an ad blocker, these content providers target them with a different approach. By implementing an anti ad block strategy, content providers can inform users about their revenue model or even force users to turn off their ad blocker. Examples of websites that have already implemented such a strategy are Forbes and Wired. Both websites force users to whitelist the website so advertisements are shown again. Alternatively, Wired also offers an ad-free and track-free subscription to the website for 1$ per week. Unknown, however, is how users react to an anti ad block strategy. How will they perceive a website compared to websites without such a strategy and will they actually comply by turning off their ad blocker? Generally speaking the more functionality a website uses, the more users experience this as clutter (Mallapragada, Chandukala, and Qing Liu 2016). Therefore, an interesting research question is whether the user experiences the anti ad block strategy as clutter or do they understand the content providers’ choice and what factors play a role in this relationship.
Next to that, there are important research questions to be answered in this area that are still unknown; what value is it for consumers to accept the website’s advertisement strategy and disable their ad blocker? Or will they stop visiting and search for alternatives? And will the visit frequency play a role here, as in, does loyalty to a website increase the intention to turn off the ad blocker? Finally this study controls for privacy concerns, as this is one of the main drivers for users to install an ad blocker (Adobe 2015). Hence, the main research question is: “Which anti ad block strategy is the most effective for content providers to secure their advertising income pressured by ad avoidance and what is the resulting behaviour and perception of visitors?”
The rest of the paper is organized as follows. First the status quo of content providers’ revenue models and the drivers of privacy concerns are discussed. Secondly, the variety of ways for a firm to deal with ad avoidance is being summarized. Then the conceptual framework and appropriate analysis to measure behavioural effects of different anti ad block strategies is developed. Lastly this study discusses the managerial implications and limitations of this
2. Literature Background
This section provides an overview of the existing literature about the constructs involved in this study. First the status quo of content publishers’ revenue model and the privacy concerns among consumers this induces is discussed. Second, the response of firms to consumers’ privacy concern and their corresponding behaviour will be assessed. Thirdly, the privacy trade-off consumers have to make when complying with content providers’ revenue model and how loyalty plays a role in this is evaluated. Throughout this review, hypotheses will be formulated and finally a conceptual model will be presented.
2.1 The Status Quo of Online Content Publishers
Rapp et al. (2009) also argues that advertisers cleverly use the lack of interest in lengthy privacy disclosure warnings and the resulting behaviour of users to blindly agree without reading. This enables advertisers to merge data from several sources to build a complete profile of the visitor. Content providers do not offer their services for a monetary return and rely on the advertisers for their income, thus maintaining this business model.
One of the anti ad block strategies proposed in the experiment in this study is showing an informative banner to consumers where the revenue model of the content provider is explained. In this banner the need for targeted advertising is explained together with the urge not to block advertisements, as it is the primary source of income for content providers. 2.2 Online Privacy Concerns As coined in the previous chapter, targeted online advertising can evoke privacy concerns among consumers (Jianqing Chen and Stallaert 2014; Rapp et al. 2009). The feeling of loss of personal data increases a consumers’ intention to protect privacy. Privacy concerned visitors actively secure their personal information by limiting access to the data and tend to disclose less personal information (Chen and Chen 2015). Privacy concerns can be defined as consumer concerns about the use of their revealed information for marketing purposes, beyond its intended purpose (Mothersbaugh et al. 2012). Especially for content providers, visitor data is not necessarily needed to improve website functionality but is primarily used to serve the advertisers. Therefore privacy concerns could potentially be a problem content providers need to deal with; as privacy concerned users are likely to exert reactance behaviour.
In the introduction it is already addressed that reactance behaviour can cause visitors to avoid advertisements (Johnson 2013). By blocking advertisements, consumers try to resolve their privacy concerns and protect their personal information (Chatterjee 2007; Johnson 2013; Seyedghorban, Tahernejad, and Matanda 2016).
On old media such as television this was already possible by using a digital video recorder (DVR)(Anderson and Gans 2011) or by simply zapping away. Since the development of online ad blockers advertisement avoidance also became possible on the Internet. The definition of ad avoidance encompasses all actions by media users that differentially reduce their exposure to ad content (Seyedghorban, Tahernejad, and Matanda 2016). This study states that using an ad blocker as the most used tool to accomplish ad avoidance. An ad-blocking tool is generally used to block third party content on webpages for example the content (ads) advertisers provide. In appendix A the technical working of ad blockers is explained. The current revenue model of content providers of using online advertising thus suffers under the increasing use of ad blockers (Adobe 2015).
Previously already one anti ad block strategy used in this experiment was already explained, namely the informative banner. The other strategy that will be tested for its effects is the force unblock strategy, where a pop-up is shown to ad-block users forcing them to disable their ad blocking tool or whitelist the website before they are able to visit the website.
Since these strategies force visitors to let their data to be collected by turning off their ad blocker, expected is that privacy-concerned users to have a negative attitude towards these anti ad block strategies. It is also expected that this will be reflected in their perception of the website as the first hypothesis proposes:
2.3 Content Providers’ Response to Ad Avoidance Behaviour
To battle the decrease in revenue due to ad blocker users, content providers need to find new ways to monetize their content. One of these ways is changing the revenue model from targeted advertising to a subscription-based one. This change brings a great challenge, because current visitors of the site value the reference price to pay zero in the current situation. Therefore, they are not willing to pay extra to access the content based on subscription (Punj 2015). Punj (2015) also argues that any price above zero will be considered a loss compared to the gain accessing the content gives. Another study suggests that as long as advertising-based ‘free’ services exist, subscription-based services will be considered too expensive (Papies, Eggers, and Wlömert 2011). They also state that it is very hard to convince consumers of the advantages of subscription-based services when their willingness to pay is low. Reason that consumers know about ‘free’ advertising-based alternatives (Papies, Eggers, and Wlömert 2011).
While subscription-based revenue models may not be the perfect solution, content providers need to find other ways to minimize ad avoidance. Pressured by ad blocking tools visitors have installed, content providers are driven to implement an anti ad block strategy (Cummings 2015). Such a strategy comes in different degrees. Some sites such as Nu.nl, Wired or The Guardian use informative banners to politely ask their visitors to turn-off ad blocking tools or buy a subscription to support the website. The informative banner contains a short text message indicating you, the visitor, is blocking their ads. Often the banner also explains the revenue model and the need for (not blocking) advertisements on the webpage.
Other websites use a more coercive strategy by limiting access or even exclude ad block users from the site, for example see the websites of The Washington Post, RTLXL, De Telegraaf, and Forbes. They detect ad block users and show them a pop-up forcing visitors to disable their ad blocker or whitelist the website so advertisements will be shown again.
Expectation of this study is that a coercive strategy is to encounter more resistance from visitors, as forcing visitors to do something will lead to more reactance behaviour. This means that implementing an anti ad block strategy is expected to have the opposite effect. Instead of visitors disabling their ad block, their intention to comply with the strategy decreases. Therefore this study hypothesizes the effects of the force unblock strategy to be more negative than showing an informative banner as is stated in the following hypotheses:
H2: Higher level anti ad block strategies decrease the intention to disable
ad block compared to the control group
H3: Higher level anti ad block strategies decrease the rating of the
website compared to the control group
H4: Higher anti ad block strategies decrease the intention to revisit the
website compared to the control group
2.4 Online Loyalty and the Privacy Trade-Off
The aim of content providers using an anti ad block strategy is to make their visitors aware of the revenue model and ultimately make them disable their ad blocker. If users actually do so depends on whether they are willing to trade their privacy for access to the website. Visitors confronted with an anti ad block strategy need to make the decision for themselves if they want to give access to personal information in order to be able to visit the website. This internal consideration is called the privacy trade-off (Savitz and Snyder 2012).
Previous research suggested that the willingness to trade privacy sensitive information is dependent on a number of factors. First context, degree and type of benefit play an important role. Second loyalty and trust in the firm determines the willingness to trade privacy for access (Cottrill and “Vonu” Thakuriah 2015). Another important factor is quality of the content. If the content provider serves high quality content, this induces loyalty. This eventually may outweigh the privacy concerns visitors have, because then visitors are possibly more willing to trade their privacy to access the high quality content (Li and Unger 2012). Therefore loyal visitors are potentially more willing to turn off their ad blocker.
Next to loyalty resulting in desired behaviour, like turning off the ad blocking tools, attitudinal loyalty also plays an important role. Trust is an important antecedent for (attitudinal) loyalty. Higher trust results in a more positive perception of the brand or product (Chuan Huat Ong, Salniza Md. Salleh, and Yusoff 2016). Besides that loyal consumers’ behaviour may be more willing to turn off their ad blocker, their attitude towards the website may also be more positive than non-loyal visitors. Both these types of loyalty will be measured using interaction effects with the dependent variables proposed in the conceptual framework in chapter 3.
To incorporate loyalty into this model, first loyalty needs to be defined. E-loyalty or online loyalty is defined as the intention to revisit a website and/or to consider purchasing from it in the future (Arya and Srivastava 2015). So a loyal visitor has a higher visit frequency than a non-loyal visitor. In this study we manipulate the loyalty variable by assigning participants into high and low loyalty scenarios based on the visit frequency. The aim is to find out whether the high loyalty group has a different effect on the perception of the website when an anti ad block strategy is implemented compared to the low loyalty group. Hypothesizing this results in the following proposition:
• H5(a,b,c): High loyalty strengthens the relationship of anti ad block strategy and the intention to revisit the website (a), strengthens the relationship of anti ad block strategy and intention to turn off ad block (b), weakens the relationship of anti ad block strategy and rating (c) compared to low loyalty
So although high loyalty strengthens the intention to revisit and the intention to comply with the anti ad block strategy, this study expects the rating of the website to be lower. This due to the trade-off concerning privacy visitors have to make in order to have access to the website.
3. Conceptual Framework
In an attempt to answer the hypotheses this study constructs a conceptual framework covering all the to be researched relationships. Figure 1 shows this research model. The main relationship is the effect of an anti ad block strategy of a website on consumer behaviour. This strategy consists of three levels where ‘none’ is the control group. This will be covered in the methodology chapter (4) more extensively.
This relationship is moderated by using loyalty to the website as an interaction variable where participants are assigned into low and high loyalty scenarios. Furthermore several demographic control variables and a privacy concern variable is included in the model.
In table 1 is a summary of all hypotheses as reference.
Table 1. Summary of Hypotheses
Hypothesis Sign Expectation
H1(a,b,c) - Privacy concerns negatively influence intention to disable ad block (a), intention to revisit (b) and rating of the website (c)
H2 - Higher level anti ad block strategies decrease the intention to disable ad block compared to control group
H3 - Higher level anti ad block strategies decrease the rating of the website compared to control group
H4 - Higher anti ad block strategies decrease the intention to revisit the website compared to control group
H5a + High loyalty strengthens the relationship of anti ad block strategy and the intention to revisit the website compared to low loyalty
H5b + High loyalty strengthens the relationship of anti ad block strategy and the intention to disable ad block compared to low loyalty
H5c - High loyalty weakens the relationship between the anti ad block strategy and rating of the website compared to low loyalty
In the following chapter the research design and analysis methods will be
4. Methodology
In this section the analysis and design of this study is presented in order to answer the proposed hypotheses.
4.1 Research Design
The independent variable ‘anti ad block strategy’ consists of three levels. This research includes a control group that is not exposed to an anti ad block strategy, a ‘low’ level that shows an informative banner to the participant and a ‘high’ level that forces ad block users to turn off their ad blocker. Taken together with the moderator loyalty, which consists of the two levels ‘low’ and ‘high’, this results in a 3*2 full-factorial design. This means every combination of variables will be represented in the experiment. Thus, this study is made up of six different conditions participants are assigned to. This also implicates that there will be six different versions of the survey. Participants will be randomly presented with one version of the survey. Each survey version features a scenario where users will be presented with the manipulation of the independent variable anti ad block strategy and the moderator loyalty.
By including both behavioural (disabling ad block, intention to revisit) and attitudinal (rating) variables, goal is to get an as complete as possible image of the visitors’ perceptions. The rating of the website is asked before the manipulation and after, this enables us to filter out the initial attitude towards NU.nl by calculating the difference. Finally privacy concern is measured as control variable using the IUIPC scale which consists of 12-items (Malhotra, Sung S. Kim, and Agarwal 2004). Below is a simplified example of the difference between the ‘low’ and ‘high’ anti ad block strategy. In the actual survey, the layout and the brand of NU.nl are also added to the scenarios in order to have a more realistic look (da Silva e Cruz and da Ascensão Marques 2014). On the left hand, the ‘low’ manipulation is shown, an informative banner on the website explaining the revenue model and politely asking visitors to turn off their ad blocker. On the right, the ‘high’ manipulation, where access to the website is restricted until users turn of ad block.
4.2 Data Collection and Pre-Test
The survey is created in Qualtrics, a survey tool where it is possible to present the respondents with different versions of the same survey. To gather respondents, the survey is spread via email, social media and personal contacts. The respondents received a link to complete the survey online. They were automatically randomized into one of the six conditions. A condition consists of the loyalty scenario followed by the anti ad block strategy manipulation. After the experiment the control variable questions were asked and if the respondent was willing to leave their email address. This is not a mandatory field, but required if participants were willing to participate in winning the prize of a Bol.com voucher. In order to test whether the survey design is accurate, a pre-test was conducted with 25 respondents. In the pre-test, only the manipulation was presented and the three dependent variables (intention to revisit, intention to turn off ad block and rating). Then was tested whether groups differ per condition. This however, was not the case for the loyalty manipulation; therefore, it was decided to increase the contrast between the two loyalty conditions to ‘at least daily’ and ‘less than once a month’.
4.3 Scales for Analysis
4.4 Data Characteristics
The collected data sample consists of 330 respondents. However due to several reasons elaborated later, some respondents had to be removed from the research. This leaves 273 valid respondents, more than the threshold of 210 respondents enabling to draw conclusions. This threshold was calculated by having at least 35 respondents in each condition (35*6). The table below displays the general sample characteristics. For reference, the demographics of the Netherlands are also displayed (Netherlands 2016).
Table 2. Data characteristics
Gender Frequency Percentage Population
Familiar with Ad Block Frequency Percentage Yes 191 70.0% No 82 30.0% Installed Ad Block Yes 108 56.5% No 74 38.7% Do not know 9 4.7% Category counts Low loyalty 133 48.7% High loyalty 140 51.3% None group 86 31.5% Informative banner 94 34.4% Force ad unblock 93 34.1%
Two-third of the sample is familiar with an ad blocker (n=191 70.0%) but only a small sample has installed an ad blocker (n=108, 56.6%). The other participants did not have an ad blocker installed (n=74, 38.7%) or did not know.
Finally, it is tested if the participants were evenly divided between groups by Qualtrics. Where the low/high loyalty group a deviation of 50% was desirable, the actual numbers correspond well (48.7%/51.3%). Also the anti ad block strategy groups were evenly divided (31.5%/34.4%/34.1%). To check whether the participants within conditions are the same as to demographics, histograms were made. This proves that age, education and gender demographic means were the same across categories. The histograms are in appendix C. To statistically confirm these histograms, an ANOVA test is performed with the demographic variables and use the six conditions as factor. The results desirably need to be insignificant, so the null-hypothesis cannot be rejected.
Table 3. ANOVA for Conditions
Variable SS Df Mean Sq F Sig.
As table 3 shows the result for all variables is insignificant; therefore it is concluded that the demographic variables are the same across groups. The assignment to conditions works as expected.
4.5 Data Preparation
This section elaborates the steps needed to prepare the data for the actual analysis. To make the data fit the model as well as possible, any inconsistencies in the data have to be analyzed and/or deleted first.
4.5.1 Partial Responses
Because this research was conducted using a survey, it happens that there is a dropout rate of 13%. These respondents have not (entirely) filled in the survey and therefore bias the model. Normally this should not immediately causes concerns, however, some participants did not fill in the essential questions for measuring manipulations. For the model to be more accurate, those respondents will therefore be filtered out from the analysis. In total 48 of the total 330 respondents were deleted before the analysis. This leaves us now with 284 valid responses.
4.5.2 Outliers
Most likely the experiment has not worked well if it took 5937 minutes to complete a survey so this case was deleted and then the analysis is run again. Now the mean is 4,5 minutes, which is a nice average time for this survey. For the lowest values we take a cut-off point of 1 minute, which is already four times faster than the mean. All values below are deleted. This leaves us with 273 valid cases. 4.5.3 Reliability Analysis Before starting to conduct further analysis, it is tested whether the scales used were reliable. For this the Cronbach’s 𝛼 (alpha) was calculated using a reliability analysis. For the first scale analysis all items of the IUIPC-scale were included in the reliability analysis after recoding of several variables. This resulted in a Cronbach’s 𝛼 of .893, which is well above the acceptable threshold of .7 (Cho and Kim 2015). In appendix D is the output of the items used for the reliability analyses. This table also displays the Cronbach’s Alpha for each item if they were to be deleted from analysis. However, with the current score of .893, the only improvement will come from deleting item two. Even then the alpha only increases by .001 so it was decided not to delete the item.
Next the loyalty items are tested on reliability. The visit frequency, connection with NU.nl and preference for NU.nl are the included items to calculate if they measure loyalty well. The test results in a Cronbach’s 𝛼 of .766, which is also above the threshold of .7. Deleting items would not result in a better alpha. Again the results can be found in appendix D. 4.5.4 Factor Analysis Now that the IUIPC-scale is tested to be reliable it can used this in factor analysis. Factor analysis is used to extract factors from the 12-item IUIPC scale (Credé and Harms 2015). By using factor analysis, it is tried to identify underlying variables and compress the number of items. These are the so-called latent variables, the actual variable measured by the 12-item scale. Then the factor scores for each individual are computed and will be used in the further analysis.
All the variables of the IUIPC scale are inserted in the factor analysis. Also Varimax rotation is selected to obtain better interpretable scores (Field 2009: p. 642). This rotates the axes so that variables are loaded maximally to only one factor. Finally a Barlett’s test of sphericity is conducted to assess whether factor analysis is appropriate. Result is that it is (𝜒! = 1455, p < .05). According to the scree plot (Appendix E) and table of eigenvalues the point of inflexion is at two factors. A cut-off value for 0.3 is and lower is chosen for the loadings to be left out of the analysis. Values above .364 are for this sample size considered to be significant (Field 2009: p. 644). Table 4. Factor Analysis Items Loadings
I believe my privacy is violated when control is lost or decreased as consequence of use for marketing purposes
.402 .495
Companies that process personal information should reveal how information is collected and stored
.782
It is important for me that I am aware how my personal information is used
.713
It annoys me when websites ask me for my personal information
.302 .682
I find it annoying to give away my personal information to different websites
.376 .638
I worry that websites collect too much personal information about me
.748 .391
I find that other people worry too much about online privacy issues
.405 .445
From table 4 it can be seen which items load high on which factor. These are highlighted in green. Looking at the items that load high on each factor can identify the latent variables. It can be seen that next to finding a privacy concern variable, several other variables about irritation and data collection correspond with each other.
Respondents that load high on factor 1 (left column) correspond with the privacy concern items, they worry about their privacy and privacy in general is important to them. Respondents that load high on factor 2 (right column) correspond with irritation about privacy, they find it annoying that companies collect data and asking them for their personal information. So factor 1 is named ‘Privacy Concern’ and factor 2 ‘Privacy Irritation’. Then values of individual participants are assigned to these factors by multiplying their scores on the individual items (6 for factor one and 6 for factor two) times the factor loadings, summing them up and finally taking the average. These two variables are ready to be included in the regression analysis.
4.5.5 Manipulation Check
From first analyses of the sample it was seen that with regard to frequency all experimental groups were fairly even assigned. To check whether the manipulation of the experiment actually worked, manipulation checks were done during the survey and afterwards. For example in the experiment was asked to imagine a low and high loyalty scenario. Before and afterwards a question was asked how frequently respondents visit NU.nl, to compare if it corresponds with the manipulation. Asking on beforehand measures their actual relationship with NU.nl (VisitFreq). Asking afterwards measures whether the manipulation worked (VisitFreq2).
For this a T-test is used with loyalty as grouping variable to compare the means of the low and high loyalty group with regard to the manipulation check question. The results are displayed in table 5.
Table 5. T-test for Loyalty Manipulation
Group Stats Loyalty N Mean
VisitFreq2 Low 125 2,28 High 125 3,62 T-Test T Df Sig. VisitFreq2 4.708 248 < .01 From the group statistics table it is seen that the mean for the low loyalty group is 2.28 and 3.62 for the high loyalty group. These values correspond respectively with “less than once a week” and “4 – 6 times per week”. This corresponds well with the actual values where the low loyalty scenario text was “Less than once a month” and high loyalty visit frequency was “Daily”. The T-test also proves to be significant; therefore the two loyalty groups are different with regard to the visit frequency.
Now that all the data is analysed for inconsistencies and the experiment has proven to work next is the statistical method for analysis and the presentation of the actual results.
4.6 Analysis Method
4.6.1 Variables for Analysis
Before selecting a technique to be used for analysis, first is summarized which variables will be entered into the analysis and the format they have. This overview can be found in appendix H.
4.6.2 Analysis Technique
From the conceptual model it can seen that in this experimental setup there are three dependent variables, namely the rating of the website, the intention to turn off ad blocker, and the intention to revisit. Therefore simple multiple regression does not suffice. To include multiple dependent variables in one analysis, multivariate analysis is required (Field 2009: p. 585). Multivariate analysis has several advantages over conducting just three separate regression analyses. First with conducting several univariate regression analyses you do not take correlations between the dependent variables into account. Secondly this does not give you any simultaneous tests for all regressions. Fortunately multivariate regression solves these problems.
To explain how multivariate regression works, first look at simple multiple regression equations where there is a set of r predictors of X and measured Y1, Y2, … Yp. 𝑌! = 𝛽!" + 𝛽!!𝑋! + ⋯ + 𝛽!!𝑋! + 𝜀! 𝑌! = 𝛽!" + 𝛽!"𝑋! + ⋯ + 𝛽!!𝑋! + 𝜀! ⋮ ⋮ 𝑌! = 𝛽!! + 𝛽!!𝑋! + ⋯ + 𝛽!"𝑋! + 𝜀!
Where the error terms ε = ε!, ε!, ⋯ , ε! ′ have an expectation of 0. These
separate regressions can be combined into one equation as follows: 𝑌!×! = 𝑋!×(!!!)𝛽(!!!)×! 𝜀!×!
𝑌!"#$%&,!"#$%%,!"#$%&
= 𝛽!" + 𝛽!!𝐴𝑛𝑡𝑖𝐴𝑑𝑆𝑡𝑟𝑎𝑡!+ 𝛽!!𝐺𝑒𝑛𝑑𝑒𝑟! + 𝛽!!𝐴𝑔𝑒!
+ 𝛽!!𝐸𝑑𝑢𝑐𝑎𝑡𝑖𝑜𝑛! + 𝛽!!𝑃𝑟𝑖𝑣𝐼𝑟𝑟! + 𝛽!!𝑃𝑟𝑖𝑣𝑎𝑐𝑦𝐶𝑜𝑛𝑐!
+ 𝛽!!𝐿𝑜𝑦𝑎𝑙𝑡𝑦!∗ 𝐴𝑛𝑡𝑖𝐴𝑑𝑆𝑡𝑟𝑎𝑡!+ 𝛽!"!𝑉𝑖𝑠𝑖𝑡𝐹𝑟𝑒𝑞!" + 𝛽!!!𝐴𝑡𝑡𝑎𝑐ℎ𝑚𝑒𝑛𝑡!!+ 𝛽!"!𝑃𝑟𝑒𝑓𝑒𝑟𝑒𝑛𝑐𝑒!"+ 𝜀!
Where n is the sample size and p is the number of the dependent variable category. So in the current model three separate equations are combined together for multivariate regression analysis. 4.7 Assumptions Multivariate Regression Before performing the actual analyses, assumptions need to be checked whether the data is suited for this method. 4.7.1 Multivariate Normality Just like normal regression, it is assumed that the dependent variable is normally distributed. Because there are multiple dependent variables, this means that they all need to be normally distributed. Constructing histograms with normality plots show this (Appendix G). These graphs show results as expected, where all three dependent variables are normally divided. However, this method relies on visual inspection, so to be sure the dependent variables are really normally divided a Kolmogorov-Smirnov test was conducted (Field 2009: p. 147). In table 6 are the results.
Table 6. Kolmogorov-Smirnov Test
Kolmogorov-Smirnov Test
4.7.2 Box’s Test of Equality
Box’s test is used to compare the variance matrices across groups. The result of this should be non-significant if the matrices are the same (Field 2009: p. 604). This statistic is standard output when conducting the analysis and resulted in a Box M of 16.702 with a corresponding non-significant p-value of .174. Thus the matrices are the same and the assumption is met. Now the actual multivariate
5. Results
Within this chapter, the results of the multivariate regression analysis are presented. First the main effects model is presented, after which the interaction effects are included.
5.1 Multivariate Tests
When conducting multivariate analysis, four test statistics are calculated. These are Pillai-Barlett trace, Hotelling’s T2, Wilks’s Lambda and Roy’s Largest Root (Field 2009: p. 604). This study does not dive deeper into the way these statistics work, but overall they all test the power of the model. Therefore it is desirable that they are significant, as this means there are differences in the dependent variable groups with regard to the independent variables. Only outcomes of the Pillai’s Trace test are displayed in the table below, as all tests generate the same p-values.
Table 7. Multivariate Tests
Test Value F Sig.
5.2 Parameter Estimates
The parameter estimates in table 8 are different per dependent variable. Therefore this is considered a univariate test. For simultaneous model results the multivariate tests are assessed.
Table 8. Parameter Estimates
DV IV B Std.Err. T Sig.
A lot of results for each regression of the dependent variables are found in table 8. The main findings are the differences of the DVs in the different anti ad block strategy groups. For intention to revisit (IntRev) it is observed that the estimate decreases per category. Where zero is the reference category with a value of .926, for the info banner category (AntiAdStrat=1) the value drops to .619. The visit frequency also is significant with a positive estimate of .146 (p < .05). So, if the visit frequency to NU.nl is higher, the intention to revisit also increases.
The note with this is that these parameter estimates do not say a lot about the differences of the DVs between groups as told before these are univariate results. Looking at the estimated marginal means and pairwise comparisons gives a clearer image of the results between groups. By constructing a graph of the estimated marginal means, the differences between groups can be visualized.
Figure 3. Estimated Means per Dependent Variable
From the graph above it is observed that the intention to revisit the website decreases per increase of anti ad block strategy. Whereas it was 3.546 for the reference category, it is a point lower at 2.621 in the ForceBlock category. For intention to turn off ad block it is different, whereas in the reference category it is 2,71, it becomes lower for the Infobanner strategy (2,48). However, for the ForceBlock strategy it increases again to 2,718, which is higher than the reference category. So participants do comply with the anti ad block strategy if -2 -1 0 1 2 3 4
None (Reference) Infobanner ForceBlock
5.3 Interaction Effects
Next is to include the loyalty moderator and analyse how this influences the multivariate model. Including the interaction effect of AntiAdStrat times Loyalty and again estimate the means per DV does this. Inclusion of loyalty still generates a significant multivariate model (Pillai’s Trace value .283, p < .01). So the analysis is continued by looking at the estimated means. The results are given in the table below.
Table 9. Interaction Effects Loyalty * AntiAdStrat
DV AntiAdStrat Loyalty Mean Std. Error
IntRev None Low 3.709 .144
High 3.341 .150
Infobanner Low 3.188 .141
High 3.287 .134
ForceBlock Low 2.738 .136
High 2.518 .139
IntOff None Low 2.794 .172
High 2.549 .180
Infobanner Low 2.487 .169
High 2.482 .160
ForceBlock Low 2.801 .163
High 2.621 .166
Rating None Low -.200 .240
High -.273 .250 Infobanner Low -.556 .235 High -.621 .223 ForceBlock Low -1.290 .226 High -1.923 .231
However, for the Infobanner group, in the high loyalty manipulation there is a slight increase. This is the same for intention to turn off ad block, whereas there is a decrease in the reference and ForceBlock categories. In the Infobanner category the decrease is minimal. With regard to the rating of the website, a decrease in all categories is observed when you compare high loyalty to low loyalty. The decrease is the largest for the ForceBlock category. So having high loyalty has more impact on the rating than low loyalty. To visualise these results the following graph has been drawn. Figure 4. Interaction Effects Loyalty
6
Discussion
The aim of this study was to analyse the effects of different types of anti ad block strategies on the behaviour of website visitors. The key findings with regard to the drafted hypotheses from the conceptual framework are displayed in the table below. Table 10. Key Findings of the AnalysisHypothesis Conclusion Notes
H1
Privacy concerns negatively influence intention to disable ad block (a), intention to revisit (b) and rating of the website (c) Partially supported Only privacy irritation for (a) (p < .05) H2
Higher level anti ad block strategies decrease the intention to disable ad block compared to the control group Not supported H3
Higher level anti ad block strategies decrease the rating of the website compared to the control group
Supported (p < .01)
H4
Higher anti ad block strategies decrease the intention to revisit the website compared to the control group
Supported (p < .01)
H5a
High loyalty strengthens the relationship of anti ad block strategy and the intention to revisit the website
Rejected Weakens the effect
(p < .01)
H5b
High loyalty strengthens the relationship of anti ad block strategy and the intention to disable ad block Rejected Weakens the effect (p < .01) H5c
Of all of the hypotheses that have been proposed for this research, three are fully supported, one is partially supported, one is not supported and two are rejected; However, the two rejected hypotheses proved to be significant for the opposite direction of effect. 6.1 Conclusions From this research a lot of interesting conclusions can be drawn. The conclusions of this research are divided into three groups; the anti ad block strategy, loyalty and privacy concerns. 6.1.1 Anti Ad Block Strategy
The effects of several anti ad block strategies were the foremost important reason to conduct this research. Literature review shows that content providers are forced to use different methods to battle ad block usage. This research made it clear how visitors react to this by measuring the three dependent variables. First the rating decreases with implementation of the anti ad block strategy, especially in the ForceBlock category. So anti ad block strategies are not appreciated by the visitor, and websites that implement one get a lower rating compared to the reference situation. Reason for this might be that it causes website visitors to perform an extra action, to read the informative banner or to turn off their ad blocker opposed to just visit the website for the content.
On the other hand, the intention to disable ad block shows a different pattern. Analysis showed that the intention to turn off ad block decreases for the infobanner strategy, however, increases again for the forceblock strategy. So although other dependent variables show a downward trend, visitors of NU.nl are somewhat willing to comply with the anti ad block strategy and actually turn off their ad block. However, this relationship has been found non-significant, therefore it can only concluded that the intention to disable ad block does not significantly change per strategy. A legit reason is that if consumers are limited in their freedom, they will show reactance behaviour (Clee and Wicklund 1980). Next to that an interesting relationship has been found where attachment to the website also increases the intention to disable ad block. It can thus be said that if visitors are attached to the website, they are willing to comply with the anti ad block strategy in order to keep visiting the site. On the other hand, privacy irritation does have a negative influence on the intention to turn off ad block. So users that are irritated about privacy questions, probably get even more irritated by anti ad block strategies and will therefore not intend to turn off their ad block.
6.1.2 Loyalty
6.1.3 Privacy Concerns
Privacy concerns were expected to have an influence on how people rate the anti ad block strategy implementation and their intention to disable ad block. For privacy reasons visitors implemented an ad blocker to protect their personal information as literature research have shown (Kumar and Reinartz 2012). For this study privacy concern was divided into actual privacy concern and privacy irritation by using factor analysis. Privacy concerns itself have not proven to be of any significant influence on anti ad block strategy. A possible explanation is that the context sensitivity with content providers is lower than for example on a webshop where more personal information disclosure is required (Bansal, Zahedi, and Gefen 2016). Content provider websites might therefore induce less privacy concerns.
Privacy irritation was found significant. The level of irritation has a negative effect on the intention to turn off ad block. If irritation increases, the intention to turn off ad block decreases. This group of visitors is probably irritated by online advertisements overall and will not comply with anti ad block strategies. That their ad blocker does not work anymore after an anti ad block strategy of a content provider causes even more irritation, because they now have to face ads again. Ad irritation’s effect on ad avoidance has already been found to have a positive effect on ad avoidance (and thus using ad blockers) (Baek and Morimoto 2012).
6.2 Managerial Implications
This study shows that website owners and content providers need to be careful when deciding to implement an anti ad block strategy. It not only decreases the rating of the website and lowers the intention to revisit the website, it also causes irritation among visitors. Although visitors tend to comply with the website’s request to turn of ad block, websites should consider other possibilities to generate revenues. Implementing a subscription-based revenue model might be a better solution. If websites still want to implement an anti ad block strategy, they should pay great attention to how privacy sensitive data is handled to minimize irritation and keep visitors to come back. Implementing an informative banner is then the best way to go, as this strategy has the lowest negative effects (and might even cause an increase in revisits for loyal visitors).
Forcing visitors to do something they do not wish to do as it causes privacy concerns will result in reactance behaviour. This behaviour is undesirable for content providers as it has a negative effect on the perception of the website. Therefore informing visitors about the revenue model and the need for advertisements on the website might be a middle ground where both visitors and owner of the website can compromise.
Another advice for managers is to control the ads they serve to the visitors. If you choose to cancel the ad blockers’ effect by implementing an anti ad block strategy, the ads you serve must be good. Otherwise visitors do not see the benefit of visiting your website and accepting the ads over searching for alternatives. So irritating ads such as pop-ups and animations are a no-go. Moreover, the content you provide must be of high quality so that visitors’ focus is on the content, not the ads.
6.3 Limitations and Further Research
As with any scientific study, this research has some limitations that researchers should consider in future research. First of all, the used sample has some limitations concerning generalizability since the majority of the sample is students. A better-divided sample with regard to education levels might yield different results. The demographic control variables also proved not to be significant at all; perhaps including other control variables might make a better model.
References
Adobe (2015), “The 2015 Ad Blocking Report,” Inside PageFair.
Alreck, Pamela L. and Robert B. Settle (2007), “Consumer reactions to online behavioural tracking and targeting,” Journal of Database Marketing & Customer Strategy Management, 15 (1), 11–23.
Anderson, Simon P. and Joshua S. Gans (2011), “Platform Siphoning: Ad-Avoidance and Media Content,” American Economic Journal: Microeconomics, 3 (4), 1–34. Arya, Sandeep and Sandeep Srivastava (2015), “Effects of user’s primary need on relationship between e-loyalty and its antecedents,” Decision (0304-0941), 42 (4), 419–49. Baek, Tae Hyun and Mariko Morimoto (2012), “Stay Away From Me,” Journal of Advertising, 41 (1), 59–76.
Bansal, Gaurav, Fatemeh Mariam Zahedi, and David Gefen (2016), “Do context and personality matter? Trust and privacy concerns in disclosing private information online,” Information & Management, 53 (1), 1–21.
Chatterjee, Patrali (2007), “Forced vs. Voluntary Exposure Web Ads: Immediate and Long-Term Impact of Ad Avoidance on Communication Outcomes,” Advances in Consumer Research, 34, 304–304.
Chen, Hsuan-Ting and Wenghong Chen (2015), “Couldn’t or Wouldn’t? The Influence of Privacy Concerns and Self-Efficacy in Privacy Management on Privacy Protection,” CyberPsychology, Behavior & Social Networking, 18 (1), 13–19.
Cho, Eunseong and Seonghoon Kim (2015), “Cronbach’s Coefficient Alpha Well Known but Poorly Understood,” Organizational Research Methods, 18 (2), 207–30. Chuan Huat Ong, Salniza Md. Salleh, and Rushami Zien Yusoff (2016), “The Role of Emotional and Rational Trust in Explaining Attitudinal and Behavioral Loyalty: An Insight Into SME Brands,” Gadjah Mada International Journal of Business, 18 (1), 1–19.
Cottrill, Caitlin D. and Piyushimita “Vonu” Thakuriah (2015), “Location privacy preferences: A survey-based analysis of consumer awareness, trade-off and decision-making,” Transportation Research Part C: Emerging Technologies, 56, 132–48.
Credé, Marcus and Peter D. Harms (2015), “25 years of higher-order confirmatory factor analysis in the organizational sciences: A critical review and development of reporting recommendations,” Journal of Organizational Behavior, 36 (6), 845–72.
Cummings, Chris (2015), “Publishers Fight Back: How The Top 50 Websites Combat Adblock,” PubNation Blog, (accessed March 6, 2016), [available at http://blog.pubnation.com/publishers-fight-back-how-the-top-50-websites-combat-adblock/]. Field, Andy (2009), Discovering statistics using SPSS. Gauzente, Claire (2010), “The intention to click on sponsored ads—A study of the role of prior knowledge and of consumer profile,” Journal of Retailing and Consumer Services, 17 (6), 457–63. Goldstein, Daniel G., Siddharth Suri, R. Preston Mcafee, Matthew Ekstrand-Abueg, and Fernando Diaz (2014), “The Economic and Cognitive Costs of Annoying Display Advertisements,” Journal of Marketing Research (JMR), 51 (6), 742–52.
Hong, Ilyoo B. and Hwihyung Cho (2011), “The impact of consumer trust on attitudinal loyalty and purchase intentions in B2C e-marketplaces: Intermediary trust vs. seller trust,” International Journal of Information Management, 31 (5), 469–79.
joshi, Deepal and Sarla Achuthan (2016), “E-commerce Buying Behavior in India: The Role of Website Features in E-loyalty,” South Asian Journal of Management, 23 (1), 56–88.
Kumar, Abhash (2014), “The ultimate master list of revenue models used by Internet companies,” YourStory.com.
Kumar, Prof V. and Werner Reinartz (2012), “Customer Privacy Concerns and Privacy Protective Responses,” in Customer Relationship Management, Springer Texts in Business and Economics, Springer Berlin Heidelberg, 279–300.
Li, Honglei, Nnanyelugo Aham-Anyanwu, Cemal Tevrizci, and Xin Luo (2015), “The interplay between value and service quality experience: e-loyalty development process through the eTailQ scale and value perception,” Electronic Commerce Research, 15 (4), 585–615.
Li, Ting and Till Unger (2012), “Willing to pay for quality personalization? Trade-off between quality and privacy,” European Journal of Information Systems, 21 (6), 621–42.
Luzak, J. A. (2014), “Privacy Notice for Dummies? Towards European Guidelines on How to Give ‘Clear and Comprehensive Information’ on the Cookies’ Use in Order to Protect the Internet Users’ Right to Online Privacy,” Journal of Consumer Policy, 37 (4), 547–59.
Malhotra, Naresh K., Sung S. Kim, and James Agarwal (2004), “Internet Users’ Information Privacy Concerns (IUIPC): The Construct, the Scale, and a Causal Model,” Information Systems Research, 15 (4), 336–55.
Mallapragada, Girish, Sandeep R. Chandukala, and Qing Liu (2016), “Exploring the Effects of ‘What’ (Product) and ‘Where’ (Website) Characteristics on Online Shopping Behavior,” Journal of Marketing, 80 (2), 21–38.
Mansfield-Devine, Steve (2015), “When advertising turns nasty,” Network Security, 2015 (11), 5–8.
Mothersbaugh, David L., William K. Foxx, Sharon E. Beatty, and Sijun Wang (2012), “Disclosure Antecedents in an Online Service Context The Role of Sensitivity of Information,” Journal of Service Research, 15 (1), 76–98. Nason, Scott D. (2010), “Free: The future of a radical price,” Journal of Revenue &
Pricing Management, 9 (5), 479–80.
Netherlands, Statistics (2016), “CBS,” webpagina, , (accessed May 15, 2016), [available at https://www.cbs.nl/en-gb/society/education].
Papies, Dominik, Felix Eggers, and Nils Wlömert (2011), “Music for free? How free ad-funded downloads affect consumer choice,” Journal of the Academy of Marketing Science, 39 (5), 777–94.
Parasuraman, A., Valarie A. Zeithaml, and Arvind Malhotra (2005), “E-S-QUAL A Multiple-Item Scale for Assessing Electronic Service Quality,” Journal of Service Research, 7 (3), 213–33.
Pauwels, Koen and Allen Weiss (2008), “Moving from Free to Fee: How Online Firms Market to Change Their Business Model Successfully,” Journal of Marketing, 72 (3), 14–31.
Punj, Girish (2015), “The relationship between consumer characteristics and willingness to pay for general online content: Implications for content providers considering subscription-based business models,” Marketing Letters, 26 (2), 175–86.
Rapp, Justine, Ronald Paul Hill, Jeannie Gaines, and R. Mark Wilson (2009), “Advertising and Consumer Privacy,” Journal of Advertising, 38 (4), 51–61. Savitz, Eric and Scott Snyder (2012), “Mobile Devices: Facing The ‘Privacy Vs.
Benefit’ Trade Off,” Forbes.com, 14–14.
da Silva e Cruz, Raquel and Alzira Maria da Ascensão Marques (2014), “Determinants OF LOYALTY TOWARD Booking.com BRAND,” International Journal of Management Science & Technology Information, (11), 98–126.
Statista (2016), “Facebook’s advertising revenue worldwide from 2009 to 2015 (in billion U.S. dollars).”
Thorbjørnsen, Helge and Magne Supphellen (2004), “The impact of brand loyalty on website usage,” Journal of Brand Management, 11 (3), 199–208.
Tucker, Catherine E. (2014), “Social Networks, Personalized Advertising, and Privacy Controls,” Journal of Marketing Research (JMR), 51 (5), 546–62. Yoo, Chan Yun (2014), “Branding Potentials of Keyword Search Ads: The Effects
of Ad Rankings on Brand Recognition and Evaluations,” Journal of Advertising, 43 (1), 85–99.
Appendices
Appendix A: Technical Background on Ad blocking
Most ad blockers currently on the market are extensions for web browsers that act like a firewall between the web browser and all known ad servers. (Adobe 2015) Content served through ad servers are being blocked automatically leaving only the ‘pure’ website content.
Appendix B: Malvertising
Malvertising has been around a while, and although we do not often hear about the vicious attacks, sometimes they can be quite severe. Malvertising works through criminals that place ads within normal advertising networks that serve the advertisements to be placed on websites and apps. (Mansfield-Devine 2015) Because ads are customized to the individual visitor, so if two different visitors load a single page they see different ads, malicious ads are hard to find. Figure 6. Malvertising
Even trusted websites are not safe because of the chaotic nature of online advertising. There is no website that has full control over which ads that are shown on their websites because of the intermediary role of ad brokers. And ad brokers are unable to check every ad due to the immense volume of advertisements processed everyday.
Appendix C: Histograms for Condition Demographics
Below are the histograms for different condition demographics. As it can be sen, for these demographic means are the same, except for age in condition three. However, as it is only one point higher it is no reason for concern.
Figure 8. Gender per Condition
Appendix D: Reliability Analysis Table 11. Reliability Analysis Current Cronbach’s Alpha .893 Cronbach's Alpha if Item Deleted I believe my privacy is violated when control is lost or decreased as consequence of use for marketing purposes .887 Companies that process personal information should reveal how information is collected and stored .894
It is important for me that I am aware how my personal information is used
.885
It annoys me when websites ask me for my personal information .885
I find it annoying to give away my personal information to different websites
.883
I worry that websites collect too much personal information about me
.875
I think the internet causes serious privacy issues .884
Compared to others I am more sensitive concerning the way websites handle personal information .888 For me it is very important to secure my privacy with respect to websites .880 Compared to other subjects personal privacy is very important to me .881 I worry this day about privacy infringement .880
Appendix E: Factor Analysis Output
Appendix F: Outlier Analysis
Appendix G: Normality Plots
Figure 16. Normality Plot Rating
Appendix H: Variables Entered in Analysis
Table 13. Variables in Analysis
Variable Description Notation
AntiAdStrat Anti ad block strategy (0 = None / reference, 1 = infobanner, 2 = Force ad block turnoff) Categorical 0 - 2 Loyalty Loyalty manipulation 0 = low loyalty, 1 = high loyalty Binary 0 - 1 Gender Gender of participant 0 = male, 1 = female Binary 0 – 1
Education Education level of the participant according to international standard classification of education
Categorical 1 – 6
Age Age of participant divided into 5 levels Categorical 1 – 5
PrivIrr Privacy irritation variable from factor analysis Numeric
PrivConc Privacy concern variable from factor analysis Numeric
VisitFreq Visit frequency of NU.nl Likert
1 – 5
Attachment Attachment to NU.nl Likert
1 – 5
Preference Preference for NU.nl over other sites Likert
