Aren’t we never invulnerable for modern deceptive user interfaces?
Yes
No
The behavioural factors contributing to the vulnerability of web users
towards the deceptive interface technique Trick Questions.
The design of a Trick Question in the eyes of a web designer with Sublime Text: text editor for code, mark-‐up and prose.
Thomas Kocken 10541748 20-‐10-‐2014
Behavioural Economics and Game Theory
Faculteit Economie en Bedrijfskunde, Universiteit van Amsterdam. Supervisor: Joep Sonnemans
Table of contents
Page
1. Introduction 2
2. The format of the question 4
2.1 The default effect (status quo effect) 4
2.2 The framing effect 5
2.3 The confusion of linguistics 5
2.4 The trammel net 6
3. Before the question 7
3.1 The halo effect and trust 7
4. After the questions 7
4.1 Cognitive dissonance 7
4.2 Loss Aversion 8
5. The fragility of web reading 8
5.1 Satisficing 8
5.2 Scan reading 9
5.3 The vulnerability of scan reading 9
5.4 Eye gaze patterns 10
5.4.1 Absence of visual signposts 10
5.4.1.1 The F-‐pattern or golden triangle 10
5.4.2 Presence of visual signposts 11
6. Experimental design and research methods 12
6.1 General architecture 12 6.2 Question phase 13 6.3 Design experiment 14 6.3.1 Textual design 14 6.3.2 Layout webpages 14 6.3.3 Technical design 14 6.4 Hypothesis 15 7. Results 15
7.1 Product and Project choice 16
7.3 Content questions 18 7.4 Interface questions 19 8. Conclusion 21 8.1 Reflection 21 8.2 Discussion 21 8.3 Conclusion 22 9. Literature 24 Appendixes 27 Appendix 1 27 Appendix 2 27 Appendix 3 28 Appendix 4 29 Appendix 5 30 Appendix 6 31 Appendix 7 32 Appendix 8 33 Appendix 9 34 Appendix 10 36
1.
Introduction
The rise of the World Wide Web
The rise of the Internet has been changing communication and the accessibility to information significantly. Websites have become a vital part of business activities and have become
indispensable for the social life of many individuals. The World Wide Web has become the cheapest and fastest way to get a tremendous load of information 24/7. Rising Internet-‐use comes together with a growth of online social and financial activities (Chun-‐Der and Huang, 2011). The total global consumer spending rises every year and reached in 2011 total annual amount of 690 billion euros (USD 961 billion) with Europe being the largest E-‐commerce market in the world (June 2012). The E-‐commerce grows approximately 20% a year (AadWeening, 2012). Between 2005 and 2010, not only online spending rose but also the percentage of people spending relatively to the number of Internet users (Turban et al, 2010). In the second quarter of 2012 counted the Internet 2.4 billion users. This is a rise of 566.4 per cent compared to the year 2000 (Internet World Stats, 2012).
Website characteristics
The rising worldwide use of the Internet increased the importance of websites of companies. Company websites play nowadays a crucial role in consumption, information-‐gathering and
consumer services. The online consumer behaviour highly depends on the web content of a website, such as the text, pictures, layout, graphics, sound and motion. These web content characteristics have been identified as one of the most important factors contributing to repeat visit (Rosen and Purinton, 2004). Next to the overall presentation of the website (website visibility, credibility, website interface and payment security) also privacy, convenient time, educational level,
information comparison and experience with the network are important factors influencing online consumer behaviour (Wang, Liu and Chang, 2008).
Dark Patterns
Websites can be designed with almost no restrictions. This can make websites a vulnerable place for deception (Hunton, 2009). The user interface (UI) of many websites is designed in a way that the user or consumer gets an incorrect representation of the circumstances (Grazioli and Jarvenpaa, 2003). This can either be done on purpose by the web designer or be an unintentional usability mistake. Deception in the user interface is closely related to the intentions of the design pattern, which is a proven and documented reusable solution to software engineering and interaction design problems (Gamma, Helm, Johnson and Vlissides, 1994; Borchers 2001). The user interface is called a Dark Pattern if the design pattern is intentionally designed to trick web users into doing things and only in the interest of the web designer (Harry Brignull, Darkpatterns.org). Brignull states that: “A Dark Pattern is a type of user interface that appears to have been carefully crafted to trick users into doing things [where these user interfaces] are carefully crafted with a solid understanding of human psychology, and they do not have the user’s interests in mind”. Psychological tricks and exploiting of cognitive biases are ways used in web design to manipulate and trick web users. A cognitive bias is according to Haselton, Nettle and Andrews (2005) a pattern of deviation from rational behaviour and can take many different forms. Essential for a Dark Pattern is the difference in interests between the web designer and the web user. A different case is a poorly designed website or interface that unintentionally confuses, misleads or even harms users. This last example is an anti-‐ pattern. Anti-‐patterns may resemble Dark Patterns a lot but trick users, on the other hand, not deliberately (Zagal, Bjork and Lewis, 2013).
Legislation
A solution that is probably most effective protecting web users against Dark Patterns is by
legislation. On 13 June 2014 the Directive on Consumer Rights replaced two previous Directives in the European Union (Official Journal of the European Union, L series, 304, 22.11.2011. p64). These two Directives protect consumers in respect of contracts negotiated away from business premises and in respect of distance contracts (OJ L 372, 31.12.1985, p31; OJ L 144, 4.6.1997, p19). The content of these two Directives has been simplified and applicable rules have been updated in the
new Directive on Consumer Rights (OJ L 304, 22.11.2011, p64). The goal of this new EU legislation is to strengthen consumer rights online in all 27 Member States. A small proportion of the Dark
Patterns have become punishable due to this law changes. According to Brignull (2014) these illegal Dark Patterns are situations where: additional non-‐free products and services are added by
defaults, website fails to inform the costumer about any kind of costs, lying to the customer about what will happen after a particular action and, lastly, after free trail automatically start charging money without any warning.
Trick Questions
The EU law changes mentioned before only affect a small proportion of the Dark Patterns
concerning e-‐commerce and only within the 27 EU Member States. The new Directive on Consumer Rights does not affect Dark Patterns concerning privacy, information disclosure, sharing and advertising (Brignull, 2014). One of the Dark Patterns that is hardly affected by the new Directive and probably the most appearing Dark Pattern used are Trick Questions. Trick Questions,
introduced by user experience designer Harry Brignull, are questions and checkboxes intended to deceive the user. At first these questions seems to be innocent but after reading carefully, mean something entirely different. The deceptive interface technique Trick Questions are often used on the web in order to subscribe users to the newsletter, trick users into accepting software updates and installing unwanted toolbars or giving permission to send information to third parties. The last point is essential in the new multi-‐billion dollar industry of consumer data (Steel, 2013). Companies buy from each other consumer data and are able with this information to make calculations to determine how to predict and influence consumer behaviour. Some data is probably not harmful for consumers such as gender, age and location. This kind of information is traded for half a dollar per thousand consumers. On the contrary, information of web shops could be harmful for consumers and much more expensive to buy as a third party (Steel, 2013). For example, if a particular consumer buys a lot of cigarette’s or alcoholic drinks online and this information is sent to the health insurance company of this particular person, then this company wants to increase the premium of this consumer. Trick Questions can next to indirectly financial consequences also have directly financial consequences outside the EU by tricking users to pay for additional unwanted services. Differently, Trick Questions is not a Dark Pattern that omits or lies about information but uses a specific way of showing information. It makes the Trick Questions interesting because they are designed with focus on different cognitive biases and therefore hard to ban with legislation (example shown in Appendix 1).
Research question
A better understanding of Trick Questions is needed in order to protect consumers against it. Growing literature about Dark Patterns helps to increase web users’ general awareness but also recognizing and understanding modern treats on the web. It is, however, not clear why different Dark Patterns are so effective on a user level. Why do consumers specifically fall for these tricks? For behavioural economists it is interesting how this phenomena influences behaviour in order to understand what is happening on a macro level. The following research question will be answered in the thesis: What behavioural factors make a web user vulnerable to the deceptive user
interface technique Trick Questions?
The first part of this thesis will cover different behavioural factors explaining why web users are vulnerable for the deceptive user interface technique Trick Questions. These behavioural factors are divided into the following categories: the format of the question, influencing behaviour before the question, influencing behaviour after the question and web reading. The first chapter about the format of the questions explains that the question itself can be deceptive due to cognitive biases or tricks from linguistics. The second and third chapter investigates how a user can be influenced up front or afterwards increasing the effectiveness of the Trick Question. Lastly, the fourth chapter investigates web reading: a less commonly known vulnerability of web users. These categories are not by definition deceptive. Different factors from various categories are mostly combined and used in real Trick Questions and Dark Patterns existing nowadays. The factors that are covered in this thesis are not certainly the only factors. After extensive web search, investigation on the website darkpatterns.org and all twitter examples (#darkpatterns), are different factors selected as most
important influential factors and thereby researched in this thesis. An experiment is made to further investigate the last category: web reading. This experiment tries to answer the following question:
To what extent can web reading-‐patterns of web users make the deceptive interface
technique ‘Trick Questions’ a more effective technique misleading web users? The answer to
this question helps answering the research question investigating the behavioural factors making users vulnerable for Trick Questions.
2.
The format of the questions
This chapter focuses on the format of the question. How the question is written down, the present possible answers and the choice of words can be deceptive and confusing for users. The following factors will be discussed in this chapter: the default effect, the framing effect, the confusion of linguistics and the trammel net.
2.1 The default effect (status quo effect)
The format of the online questions can make a big difference in the answer of users when their values are not well articulated (Kahneman, Ritov, Jacowitz and Grant, 1993). This is because a users’ response to a Trick Question is not a pre-‐calculated preference. Instead, the response to the
question is generated on the spot (Fischhoff 1991; Payne, Bettman & Johnson, 1992; Slovic 1995).
Thereby are revealed preferences and contact with websites influenced majorly by defaults
(Johnson, Bellman and Lohse, 2002). Ticking box is a useful tool for web designers on the web when asking a question to web users. The designer can however already tick one of the preferable boxes instead of letting the user choose. In this case the user can either change his decision or stay to the defaults (Jakob Nielsen, 2005). The difference between the defaults, opting-‐in and opting-‐out can make a sizable difference in web users’ choice and is known in Behavioural Economics as the status quo effect (Samuelson and Zeckhauser, 1988). This is evident from the research of Johnson,
Hershey, Meszaros and Kunreuther (1993) who compared two car insurance options. The choice of the first user group in the experiment was either to opt-‐in for the car insurance or, in the other group, to opt-‐out in order to not get the car insurance. The result in this field experiment that involved significant amounts of money, was that the insurance participation was 20% in the opting-‐ in case compared to 75% participation in the opting-‐out case.
Some opting-‐in and opting-‐out questions are shown in table 1 and are commonly applied by web designers on numerous of different websites (Johnson, Bellman and Lohse, 2002). These questions with participation results indicate the effect of defaults, framing and a combination of the framing and the default effect. The default effect is noticeable comparing question 1 and 3, and comparing question 2 and 4. The results show that participation is higher when participation is already ticked and that, the other way around, participation is lower when ‘not participate’ is already ticked.
Johnson, Eric J., Steven Bellman, and Gerald L. Lohse. "Defaults, Framing and Privacy: Why Opting In-‐Opting Out1." Marketing Letters 13.1 (2002): 5-‐15. p7
According to Samuelson and Zeckhauser (1988) do status quo options inflate the attractiveness of that particular option, even though it was a randomly assigned option. The Status Quo effect influences choice and thereby contributes to the effectiveness of the Dark Pattern: Trick Questions (Schweitzer 1994; Schweitzer 1995). The reason of the existence of this effect can be that the decision-‐maker does nothing because of physical and cognitive laziness (Chapman and Johnson, 1999). This reason contradicts, however, with the theory that people satisfice on the web: using their time efficiently and going for the ‘good-‐enough’ outcome (Krug 2006; Whitenton 2014). Opting-‐in or opting-‐out would be a minimum amount of effort. This means that opting-‐in or opting-‐ out to answer the question the way the users want it to be answered would be time efficient and worth doing (Johnson, Bellman and Lohse, 2002). Also, table 1 shows that 48.2 percent of the participants participate even though they have to opt-‐in actively. This percentage is remarkably high which means that laziness is probably not the biggest reason. Another reason for making the defaults options a popular option is the increased choice probability because the default option is considered subject of comparison (Houston, Sherman and Baker, 1989). Lastly, web users might perceive the defaults as recommended by the questioner (Chapman and Johnson, 1999).
Figure 1
Johnson, Eric J., Steven Bellman, and Gerald L. Lohse. "Defaults, Framing and Privacy: Why Opting In-‐Opting Out1." Marketing Letters 13.1 (2002): 5-‐15. p7
2.2 The framing effect
Next to the default effect is the expression of alternatives as either negative or positive a strong influential factor. This effect is called the framing effect. In table 1 occur two different frames. The positive frame is question 1 and 3 and the negative frame question 2 and 4. The framing effect is hard to measure in table 1 because more than just framing is at issue. In contrast with table 1 does table 2 with the corresponding figure 1 show the framing and default effect separately and with multiple combinations of both effects. Figure 1 clearly shows that the participation rate is always higher with a positive frame compared to the negative frame. This finding is in line with more research done on the framing effect (Johnson, Bellman and Lohse, 2002). According to the results of the participation experiment does the framing effect also exist when it is very subtle (table 1; Johnson, Bellman and Lohse, 2002). Framing differences arise because of emphasizes on either a loss or a gain. Cost of a certain loss hurts more than the amount of pleasure gained from the alternative frame. Revealed preferences have obvious reversals caused by these differences (Johnson, Bellman and Lohse, 2002). Multiple experiments and research has shown that framing effects can be significant (Kahneman and Tversky 1984; Tversky and Kahneman 1986). Frames and defaults are commonly applied together on numerous different websites, which is further discussed in chapter 2.4 (Johnson, Bellman and Lohse, 2002).
2.3 The confusion of linguistics
The misleading part of Trick Questions can be the details of certain words and phrases. Words are left out, added or substituted with difficult or confusing words. Consequently, the question does at first seem to be normal. After reading carefully it turns out to mean something entirely different. Scan reading web users fixate only on some of the words and do not read the entire sentence or
question (Nielsen, Whitenton and Pernice, p14). These users become thereby vulnerable for deception of wording that focuses on details within the question. A relatively common example of this kind of Trick Questions is the double negative question. Table 2 contains a double negative question that got highest participation of the negative framed questions (question 1). A more sophisticated example is the ‘limit advertisement’ option in the settings of an iPhone (Appendix 2). Double negative questions are questions with two negations in one clause, which results in losing the negative aspect of the question (Righarts, 2007). Negations can be divided into three different variants (Benamara et al, 2012). First of all, the negative operator: these words are adverbs or conjunctions such as “not”, “no more” and “neither”. Secondly, the negative quantifier, which expresses both quantification and negation and can both be a noun or pronoun, such as “no” and “nobody”. Lastly, implicit negative words called lexical negations such as “absence of” and “deficiency”. The concept of sentences with multiple negations is broadly studied in Behavioural Studies, Philosophy, Law, Linguistics and Logics because of the complexity of this linguistic phenomenon (Benamara et al, 2012).
Results from Law and Human Behavioural studies (Perry et al. 1995) show that not only children but also college students (from 18 years to 22 years old) have great difficulties answering negative, double negative and difficult vocabulary questions. Furthermore, adults seem to have significant more problems with negative questions compared to positive questions (deVilliers & deVilliers, 1979; Walker, 1994). Questions with more than one negation are substantially harder to
understand than single negative questions (Charrow & Charrow, 1979; Matthews & Saywitz, 1992; Walker, 1994). Perry et al. (1995) found in their experiment that college students answered on average 1.6 double negative questions out of 5 correctly about a short movie they just watched. Interestingly, the college students answered on average 3.4 questions out of 5 correctly when the same questions were asked without the negations (Perry et al. 1995).
Perry et al. (1995) conclude that not only questions with double negations but also with difficult vocabulary and single negatives result in significant problems trying to answer the question. Single negatives and difficult vocabulary is slightly easier to answer; 2.3 and 2.2 on average correct answers out of series of 5. This finding is supported by a lot of other researchers (e.g., Cashmore & Bussey, 1990; Charrow & Charrow, 1979; Flin et al., 1989; Gaer, 1969; Melton, Limber, Jacobs, & Oberlander, 1992; Pierre-‐Puysegur, 1985; Saywitz, 1989; Saywitz, Jaenicke, Comparo, 1990; Saywitz & Snyder, 1993; Stevens & Berlinger, 1980; Walker, 1993; Warren-‐Leubecker, Tate, Hinton and Ozbek, 1989). These results confirm that inappropriate questioning could obfuscate communication (Perry et al. p625, 1995).
According to Moore (p306, 1992), people misinterpret negations because the cognitive capacities of the receiver are being overloaded and therefore greater effort is required. Also, receivers are facing an intricate problem recognizing the question as negative or as positive, mainly because double negation increases the logical difficulty. Lawyers can, because of their skills in language, manipulate thoughts and opinions of others in court (Philbrick, cited O’Barr, 1982). This manipulative language use is very similar to the questioning techniques, mostly double negative, used in web interfaces by web designers nowadays.
2.4 Trammel Net
A trammel net is a variation of the gill fishing net that consists of two or three layers of netting with an inner network of light and supple yarn. The problem with these nets is that not only the targeted fish get stuck but also almost all other sea creatures such as dolphins and whales (Reeves, Read and Notarbartolo-‐di-‐Sciara, 2001). Web designers of user interfaces use comparable techniques
(Brignull, 2013b). The trammel net is a technique used to increase the effect of Trick Questions by using different techniques together, such as: the framing effect, defaults effect and confusing wording. This implies that multiple (Trick) questions are asked after each other in order to catch the user either with the first, second, third question or somewhere in between. Appendix 1, example B, is an example of a small online trammel net with two questions directly after each other.
3. Before the question
Web users have beliefs and thoughts about a particular company before answering a Trick Question. This belief can be created on the spot because of a visit on the website. On the other hand, can it be a result of a long relationship between the web user and the company brand. Both extreme cases of web users have thoughts and beliefs that influence their behaviour when answering the Trick Question and need to be taken into account.
3.1 The halo effect and trust
Another reason for the effectiveness of defaults and the effectiveness of trick questions is the halo effect. Introduced by Edward Thorndike (1920) explaining that the overall impression of a person influences the thoughts and feelings about that person’s character. The halo effect can, next to judgments about other people, also impact our judgments about organizations, locations, products and services (Cardello and Nielsen, 2013). One favourable aspect of a website can generate good judgments from users. Conversely, a bad experience or bad aspect of a website can form the belief that the website will threat them poorly in the future as well. The impression of one particular attribute can therefore form the web user’s overall judgment (Cardello and Nielsen, 2013). The overall judgment of a website seen for the first time is according to Lindgaard en Dudek (2002) based on the visual appearance. Web users are subsequently reluctant to revise their judgments resulting in the confirmation bias, which implies that people tend to search or interpret information such that their beliefs or hypotheses are confirmed (Plous, 1993). Lindgaard and Dudek (2002) found with their experiment that the look and feel of websites has a halo effect on the total website experience, even with websites with very poor usability. In this way trust is created due to a nice homepage design. Web users believe because of the halo effect that the website will threat them good in the future. This thought makes users subsequently vulnerable for Trick Questions because users do not expect to get tricked. An example of such a good attribute in web design that could trigger the halo effect is the quality of the internal search results that tend to influence the judgment of users on the overall quality of the site, brand and its products or services (Cardello and Nielsen, 2013). According to Nodder (2013, p22-‐24), increase certification logos on websites trust even though the certification logo is not real. Users think that these images are provided by third parties and that the certification means the endorsement to the website (Nodder, 2013 p23). The marginal level of additional reassurance increases trust and is just enough to let users scan read the question and trust it. Not trusting users are more sceptical and read everything word by word without taking any risk.
The halo effect explains why big companies as Philips, Apple and Windows use Trick Questions. These companies spend a lot of money in order to increase brand awareness and brand preference which raises the overall impression to a higher level than just the trust created by the website. Trust is in this case even higher and the Trick Question therefore more effective. An example of effective abuse of created trust is found in Appendix 3.
4. After the question is answered
The answer of the web user to a Trick Question is a vital point in time for this thesis. However, the web user is also after the question is answered affected by different behavioural factors making Trick Questions more effective and thereby increasing the damage (financially or privacy) or irritation to users. Two important factors in this stage are the effects of cognitive dissonance and loss aversion.
4.1 Cognitive dissonance
An important cognitive bias to take into account is cognitive dissonance. According to Festinger (1962) is cognitive dissonance the discomfort and mental stress due to the confrontation of new information that contradicts with existing beliefs, ideas or values. Cognitive dissonance can play an important role influencing behaviour after the Trick Question is answered. This feeling can,
however, only occur if users find out that they were misled or answered differently than they wanted. First, the user answers the question. This is a contradictory answer compared to the users beliefs and ideas and, thereby, creates discomfort (Festinger, 1962). The cognitive dissonance increases the ‘damage’ and improves the effectiveness of Trick Questions. In this case users, firstly, get tricked and give an answer to the question that contradicts with their individual beliefs, ideas or values. Later on they find out that they answered the question differently than they wanted it to be answered. The mental stress and discomfort of the cognitive dissonance influences the users’ attitude towards his action. The individual is motivated to reduce the dissonance in order to achieve consonance. In addition, users ignore the problem and avoid further dissonance instead of trying to reduce the dissonance by taking action (Festinger, 1962). They convince themselves that their bought flight insurance or newsletter subscription was not a mistake but instead useful.
Subsequently, they do not complain because the result is good for the web user. People are biased to see their choices as correct. They would however not have done the same if they had noticed the trick before they answered the Trick Question (Nodder, 2013 p4-‐5).
4.2 Loss Aversion
Another cognitive bias influencing behaviour after the user is tricked is loss aversion. After
answering a Trick Question the reference point of the user changes. For example, users start with an option to either book the insurance or do not book it. When they find out that they are booking the insurance their reference point changed. Their reference point is changed to the situation where the user can either stay with the safe insured option or change to the unsafe non-‐insured option. This shift in reference point makes the decision a little different, psychologically. Web designers can, and do, abuse this switch and focus on the fear that is present with loss aversion (Nodder, 2013 p67). Loss Aversion indicates that people tend to strongly prefer avoiding losses to acquiring gains (Kahneman and Tversky, 1984). This means in the insurance case that getting insurance makes the user less unhappy than losing the insurance makes him happy. Loss aversion has overlapping elements with cognitive dissonance in the case of a Trick Question. Two strong behavioural factors that can influence the web users at the same time.
5. The fragility of web reading
Various behavioural factors discussed in previous chapters either play a role before, during or after the answering process of a Trick Question. Web reading is the final behavioural category discussed in this paper and is, on the contrary to the other categories, an overarching behavioural factor. The way people read, influences web users in every stage: before, during and after answering a Trick Question.
5.1 Satisficing
According to Krug (2006) and Whitenton (2014), web users do not investigate all information available on websites when making a decision. Web users, most of the time, do not follow the optimal decision making principle. This principle implies that people would put effort and time to find the optimal option or solution (Krug, 2006; Whitenton, 2014). Instead, they quickly search for a hint, answer or link that matches approximately the way to their solution. In the latter case, users accept the ‘good-‐enough’ answer. This strategy is called satisficing, which is a combination of the words ‘satisfy’ and ‘suffice’ introduced by the economist, psychologist and sociologist Herbert Simon (1955, 1956). Simon referred to this approach as bounded rationality (Simon, 1957). This cognitive heuristic is dependent on the individual and his personal searching threshold. More information and more options increase the probability of satisficing in a particular situation. The principle of a too great range of choices, choice overload phenomenon, makes people dissatisfied according to many laboratory results (Iyengar and Lepper, 2000). Choice overload is a common problem on websites making satisficing a frequently applied problem solving technique (Whitenton, 2014). Web users do not always go for the ‘good-‐enough’ option, but carefully weigh options from time to time. This approach is dependent on the frame of mind, the confidence users have in a particular website and the pressure of time present at a particular moment (Krug, 2006).
Web users commonly satisfice because, firstly, they do not have a lot of time; satisficing is more efficient (Klein, 1999). Secondly, carefully weighing different choices against each other does not always deliver the preferred outcome. Users know that there is a large amount of websites and information, users do, however, not know if the current webpage gives them the information that they are looking for (Nielsen, 1997). Lastly, the user does not have a big problem when making a wrong choice (Krug, 2006). This last reason does count for most cases but contradicts however with the Dark Patterns with financial consequences.
5.2 Scan reading
Reading is one of the most frequent and important activities of web users. The way users read on screens of computers, phones or tablets differs significantly from how people read on paper (Nielsen, 2010). According to Nielsen (1997) scan read 79% of the web users. This scan reading activity means that users look for words, phrases or images that catch the attention instead of reading the whole page systematically (Krug, 2006). Generally, scan reading web users only read a little more than 20% of the words on a web page (Nielsen, 2008).
Satisficing and scan reading are closely related. Satisficing is the strategy picked aware or unaware and scan reading is the manner of execution. People who satisfice tend to scan read on the web as a strategy and not as a random act. They use this strategy to quickly sample specific information, shapes, text or words (Nielsen, Whitenton and Pernice, p42). This explains why most web users leave a web page after 10-‐20 seconds (Nielsen, 2011). There are, however, more reasons why users scan read next to the reasons to satisfice. In 1997 Nielsen found that web users scan read on the web because reading from a screen is tiring. People read 25% slower on the web compared to non-‐ screen pages. In addition, the web is user-‐driven; users want to feel active on the Internet.
Steve Krug (2006) adds the explanation that users not only want fast information but are simply not interested in everything written on the webpage, they do not need all the information. Secondly, web users scan read because they are very good and successful at it. Nielsen concludes in (2010) after new results from usability research that, on the contrary, the increased efficiency of
information gathering is the one main reason for scan reading on the web. This habit is something every individual web user learned on their own by web surfing a lot and finding out better ways (Nielsen, 2010).
5.3 The vulnerability of scan reading
On the contrary to the efficiency reasons for scan reading introduced by Krug (2006) and Nielsen (2010) is scan reading most certainly not always an efficient strategy. Moreover, harm can most definitely be done to the user. Krug (2006) states that users just click buttons or think they understand but have actually no idea what they are doing. Most websites that score high on usability are designed in a way to effectively help the satisficing user to reach their goal or find where the user is looking for (Whitenton, 2014). In many cases satisficing will not harm the user. The deceptive interface Trick Questions is however able to harm or trick this user. Scan reading makes users vulnerable for two different cases of Trick Questions. Firstly, users are more vulnerable for deception within the question. Users read rarely word-‐by-‐word; a wrong presentation can thereby be created when vital words are skipped in questions with confusing linguistics (chapter 1.3) or with framing effects (chapter 1.2). Secondly, when the questions are put in big blocks of text. An example of this kind is shown in Appendix 4 and 9. This method will be discussed further in chapter 4.4 with results from multiple eyetracking studies. Background information of eyetracking research is provided in Appendix 5.
5.4 Eye gaze patterns
Nielsen, Whitenton, and Pernice did extensive research investigating how people read the web with eyetracking techniques. For this research multiple experiments were conducted using more than 300 web users (participants between the age of 18 and 64) and tested hundreds of websites. An important result from this research is that web users do not have a universal way of looking at a webpage. The gaze pattern of web users on a web page is mainly dependent on the layout, content and user motivation. These variables result in web users reading only paragraph titles, scanning paragraphs or selecting one or more paragraphs and reading that intensively word-‐by-‐word. The layout of the page can vary considerably, influencing user behaviour. This is dependent on the type of webpage and the related design patterns of these pages (Nielsen, Whitenton and Pernice, p13).
5.4.1 Absence of visual signposts
Nielsen, Whitenton and Perinice did also eyetracking research on SERP (Search Engine Result Page) and Article pages. Most of these pages have heavy textual content with the absence or low presence of visual signposts. As far as is known are Trick Questions most prevalent on webpages without visual signposts. In absence of any visual signposts are web users left with text and have to create their own shortcuts. This is done by scan reading a web page with multiple different fixations: eye gaze patterns (Nielsen, Whitenton and Perinice, p11-‐12). Regularly, a combination of different eye gaze patterns arises, because the user motivation changes during the scan reading process (Nielsen, Whitenton and Pernice, p11-‐12).
5.4.1.1 The F-‐pattern or golden triangle
The most dominant and most occurring gaze pattern is the F-‐pattern. This gaze pattern does not only occur on highly textual webpages such as ‘about us’ or ‘product information’ but also on SERP (Appendix 7, image 3; Nielsen, 2006). This gaze pattern on SERP is also known as the golden triangle introduced by Google. The golden triangle is literally a triangle in the upper left corner of SERP that involves most eye fixation activity according to eyetracking results (Granka, Feusner and Lorigo, 2008). The F-‐pattern or golden triangle implies that web users start at the upper left point of the textual content and read word-‐by-‐word horizontally to the most right part of the sentence. They continue reading underneath the point where they have initially started and generally stop reading a little bit earlier than they did in the first sentence. The third sentence is read the same way as the second but the users stop reading even earlier than they did in the second sentence. This pattern continues until users read nothing or almost nothing of the sentence (Nielsen, Whitenton and Pernice, p54). A graphical representation is found in figure 2.
Nielsen (2006) states that the F-‐pattern occurs in different forms but overall three fixed
characteristics exist: firstly on the top of the page a horizontal reading line, secondly, slightly under the first line a second shorter reading line. This pattern is finished with a vertical bar on the left that makes the shape of the letter F complete. This letter F is not fixed: it can differ in layout, content and motivation of users. The F-‐pattern can, for example, take erratic forms due to faster scan reading by web users (Nielsen, 2006). On a webpage with four of more paragraphs does 81% of the web users look at the first paragraph, 71% at the second paragraph, 63% at the third and only 32% at the last fourth paragraph. In this case looking at a paragraph can either mean: taking a quick look or reading the whole paragraph (Nielsen, Whitenton and Pernice, p55-‐56). This explanation is also graphically shown in figure 2. These results indicate that the F-‐pattern is most certainly not always an effective technique to scan a particular article. Commonly, information outside the golden triangle or F-‐ pattern is missed which can be crucial if important information is put there (Nielsen, Whitenton and Pernice, p56-‐57).
Figure 2
This figure shows a textual web page without visual signposts. The F reading pattern is applied here. The red area is read or scanned. The first paragraphs are read more than the last paragraphs.
Nielsen, J., Whitenton, K., & Perinice, K. How People Read on the Web. The Eyetracking Evidence. p55.
5.4.2 Presence of visual signposts
The location of page elements affects the viewing pattern of web users. Strong factors are images, blurbs and the headline placement. After testing and analysing 8 different homepage designs, Outing and Ruel (2004) found that web users particularly start scanning with many fixations on the logo or headlines in the upper left area of the page. After perusing this particular area, users
generally move on to the lower part of the upper right hand quadrant. This movement is followed with a movement to the lower left side. This back and forth pattern continues until users end-‐up at the bottom of the page after which they scan the right-‐hand side of the page before leaving (Outing and Ruel, 2004). This last move by web users is probably the consequence of the learning effect from the emerging practices and standards of web designers. Advertisements generally appear and are expected on that side of the web page, shown in Appendix 8 (Lynch and Horton, 2009; Bernard, 2000, 2001 and 2002; Bernard and Sheshadri, 2004). Usability trends that have emerged become clearer and start forming the basis for good web page composition (Lynch and Horton, 2009). Users prefer to avoid online advertisements. Advertisements are one of the most frustrating factors when surfing on the web (Retail Forward, 2002). The results of the scan reading behaviour of the 8 different homepages is put together in a webpage figure with priority zones shown in Appendix 6 (Outing and Ruel, 2004). These results suggest that certain information is generally read and recorded more in yellow areas and even more in red areas compared to the case where this particular information was placed in a green area. This assumption can be an opportunity for deception.
Unclear from the eyetracking experiments
Nielsen, Whitenton and Pernice did not state in their research report how many participants participated. They only mention that more than 300 individuals participated which is confusing because more than 300 could be 301 or another extreme of 600 participants (Nielsen, Whitenton and Pernice, p351).
Important to take into account is that according to Nielsen, Whitenton and Pernice (p38-‐39) the gaze pattern of non-‐experienced web users is almost the same as of experienced web users. They did however not mention the possible difference in reading behaviour between users who do and users who do not suffer from the specific reading disability dyslexia. Dyslexic people tend to read more slowly and have more problems with nonsense word reading and common spelling (Ferrer, Shaywitz BA, Holahan, Marchione and Shaywitz SE, 2010). Precise numbers are unknown but estimated is that between 5 and 10 percent of a population suffers from dyslexia (Birsh, 2005). This may have resulted in wrong conclusions or inexplicable data in the eyetracking research. It might be