Effect of hypermedia structure on acquired knowledge organization.
Numerous research works have explored the effects of hypermedia structure on information comprehension and use. In a review paper dedicated to this topic, Tricot (1997) summarized existing findings in the following way: "The principal limit of these studies was put forward by Dillon (1991), who showed the importance of the rhetoric structure of the document no matter what its formal structure is" (Tricot, personal translation). Despite Tricot's conclusion, this issue is still worth exploring for two reasons.
First, even if textual structure appears to be the main factor that influences the construction of a coherent representation of the system, the aspects of the system that are specifically hypertextual seem to have a facilitating or perturbing effect whenever they respectively corroborate or contradict what is presented by the textual materials. This issue will be addressed later in this article.
Second, the "rhetorical structures" studied by Dillon are not exclusively textual: they are related to types of discourse or genres, and include textual characteristics as well a content-related aspects, media properties, and so forth. The user's understanding and use of the system relies on all of these features. The type of structure that has been studied the most in the works referred to by Tricot (1997) is that of scientific papers: a structure that is tightly coupled with the paper medium. It is probably still too early to identify such types of structure in the realm of hypermedia (although Dillon & Gushrowski (2000) have described the homepage as the first digital genre).
The role played by hypermedia structure in the process of comprehension and mental organization of the knowledge acquired through navigation is thus still largely to be specified. In this article, results from an experiment are presented, in which 40% subjects browsed of an educational hypermedia system, to evaluate the influence of the system's structure on the organization of the knowledge they acquired.
In the view advocated in this article, navigation is the activity through which the user discovers the system in all its dimensions (informational content, formal organization, pragmatic enunciation position, etc.). Hence, it is an inherently interpretative activity, through which the user attempts to built coherent mental representations of the system he is using (bearing on all the dimensions just mentioned without distinction), based on his successive navigation choices, in a constructive and incremental fashion.
The next section of this article presents the theoretical rationale for a potential influence of hypermedia structure on the organization of knowledge acquired through browsing. The second section introduces the experimental protocol as well as the educational hypermedia system used for the experiment. Finally, the last section details and discusses the results of the experiment.
THEORETICAL FRAMEWORK AND HYPOTHESES
Structure and Hypermedia Systems
What do we mean when we talk about hypermedia "structure"? Where is this structure located? Hypermedia structure will be defined here not as an entity that exists objectively in the system, but rather as a reconstruction operated by the user, based on different types of cues that will be called factors. Factors correspond to all the components of the system image that are likely to influence the user in building mental representations of the system.
The designer and the user both hold a model of the system, of its functioning and of the appropriate way to use it. Ideally, these two models match perfectly. Although, the designer can only communicate his/her model to the user through the system itself, or--more precisely--through the system image, which corresponds to "its appearance, its operation, the way it responds, and the manuals and instructions that accompany it" (Figure 1; [Norman, 1988, p. 190]).
[FIGURE 1 OMITTED]
In that regard, the system image is a semiotic intermediary whose function is to communicate the design model to the user, who relies on it to build a host of mental representations of the system, its functioning, its formal organization, its informational contents, and so forth. The hypermedia system's structure is part of these representations. It does not exist outside the user's interpretative activity (or the designer's design activity, but this article will focus on the structure as elaborated by the user based on the system image).
System Image and Factor Classes
Within the system image, let us distinguish between several classes of factors. A first class--the rhetorical factors--groups the factors related to the internal organization of the different semiotic materials involved in the system: text, images, sounds, animations, and so forth. In this category, the textual factors (i.e., linguistic and typographic cues that structure any given text, often referred to as "rhetorical structure") have received the most attention. Most research works focused on these rhetorical factors in hypermedia use (Rouet, 1995, 1997, 1999; Foltz, 1996) are based on discourse comprehension theories in which building a mental representation of the situation described in a text relies on the construction of a representation of the text itself, that is, of the propositional structure that corresponds to it (van Dijk & Kinstch, 1983).
The two next classes--formal factors and interfacial factors--are specifically hypertextual. Formal factors correspond to the hypermedia's content segmentation into nodes, and to the implementation of links connecting them. Interfacial factors include the way information is laid out and presented on screen, as well as the different navigation tools, that give different statuses to the links and nodes.
The segmentation defined by formal factors delimits (supposedly) coherent content units, that correspond either to a screen or to a "page" (when the node's content exceeds what can be displayed on the surface of the screen, requiring--for example--the user to scroll down). Within each node, the relative positioning of information units (titles, texts illustrations, menus, to name only a few), the use of color, shape, contrast, etc. allow the user to assign each node a status, a position in the overall structure he is reconstructing ("Is it a section head?, an example for a notion described in another node?, a glossary entry?," and so forth.). The same is true for the navigation tools assigning distinct statuses to the links that are available from one node ("this menu allows me to go between section heads, that one lists the nodes that are subordinated to the active node," etc.). All this information supporting the user in categorizing nodes and links corresponds to interfacial factors.
Of course, the notion of interface cannot be reduced to what is defined here as interfacial factors. Rather, interfacial factors are components of the interface that make the structural characteristics of the hypermedia system visible. An interface has other functions (that will not be discussed here) as well.
Finally, let us consider semantic factors, which are related to the informational content presented by the system. Semantic factors are the way notions, concepts, information chunks, and so forth, are articulated together and organized within the knowledge domain the system presents. For example, a temporal process (a sequence of actions, for instance) entails an order that characterizes the chain of steps that compose it, however it is presented in a document.
Factor Congruence and Representation Construction
In terms of what was previously described, the research question this article seeks to answer is the following: What influence do formal factors (node segmentation and link implementation) and interfacial factors (aspects of the interface making the status of the relationships between nodes visible) have on the construction by the hypermedia user of structured representations of the system, and more specifically of the informational contents it presents?
One could, from a naive point of view, strictly separate informational and navigational tasks (on this distinction, see Thuring, Hannemann, & Haake, 1995): rhetorical and semantic factors would allow the user to understand and integrate the informational contents of the system, while formal and interfacial factors would allow him/her to understand its formal organization and to navigate it efficiently.
However, this view is too simple. If some classes of factors can be seen as a privileged material for understanding such and such dimension of the system (its contents, its formal organization, the pragmatic relationship it induces towards its user, etc.), the user is likely to found his/her understanding of these dimensions in all of these factor classes. As a matter of fact, the congruence of the different classes of factors appears to have a positive effect on different types of tasks. Three examples from the literature will illustrate this point. (1)
Edwards and Hardman (1989) compared their subjects' ability to represent the structure of the hypertext they had them browse, for three hypertexts sharing the same rhetorical, semantic and interfacial factors, but with different formal factors: the links that could be activated in each version were different, although all of them were presented as links (i.e., underlined colored text) in all three conditions. The three systems presented their contents (a directory of leisure facilities in Edinburgh) in a hierarchical fashion. The subjects who had access to the version of the system that had hierarchical links (vs. links in an index or both) performed better. No matter what condition they were in, subjects all represented the system's structure as being hierarchical. Hence, it appears that subjects are more able to represent the hypertext's formal structure when formal factors are congruent with other classes of factors.
Dee-Lucas (1996) presented the results of several experiments testing the added value of graphical interactive overviews (as compared to text-only hyperdocuments) in the construction of coherent representations of the contents presented by the system. According to her results, using overviews (which are part of the interfacial factors) coupled with the information contained in the textual materials presents a double advantage: one the one hand, it facilitates the selection of relevant information, and on the other hand, it helps elaborating appropriate study strategies.
Giroux, Bergeron, and Lamarche (as cited in Rouet, & Tricot, 1998) showed how congruent interfacial and semantic factors can support the selection of information: confronted with menu pages including eight or nine options each, their subjects selected their target information faster when links were grouped according to semantic criteria.
Andrew Dillon has lead a series of research projects that corroborate the view developed here, and question the traditional distinction between informational and navigational tasks by hypothesizing that "comprehension is not something 'other than' navigation, some form of task that is independent of the process of moving through the information space. Rather it is an intrinsic component of information use" (Dillon & Vaughan, 1997). Dillon introduces the concept of "shape" in replacement of that of navigation. To perceive a shape in the informational space being manipulated, the user relies both on the interface and on content-related cues. "The concept of shape assumes that an information space of any size has both spatial and semantic characteristics. That is, as well as identifying placement and layout, users directly recognize and respond to content and meaning" (Dillon, 2000, p. 523).
Dillon's spatial cues are related to screen layout and interface. His semantic cues are related to informational contents. This distinction overlaps nicely with the one between formal and interfacial factors on the one hand, and semantic and rhetorical factors on the other. Indeed, the use of semantic cues he describes essentially correspond to the identification of structural schemas that are specific to one peculiar type of discourse (scientific papers), based on the way the presented text is formulated.
Whereas Dillon demonstrated the influence of semantic factors on navigation, the work presented in this article goes the other way around and asks the question of the influence of formal and interfacial factors on informational tasks. The question is "do cues related to the hypermedia system's formal structure and interface leave a mark on the way users organize the knowledge they acquire through browsing?"
In the previous section, the idea was introduced that understanding the formal organization of a hypermedia system and understanding its informational contents are two tightly articulated activities from the user's standpoint. Although, all is not in all: hypermedia users do not build one and only one big unified and coherent mental representation of the system, including both its formal and semantic aspects. The notion of schema (Rumelhart & Norman, 1988) will help clarify this issue.</p> <pre> Schemata are data structures for representing the generic concepts stored in memory. (...) Schemata represent encyclopedic knowledge rather than dictionary-like definitions; [they] are active recognition devices whose processing is aimed at evaluating how well they fit to the data being processed. (p. 537) </pre> <p>Schemas correspond to information packets varying in complexity and in abstraction level, organized and articulated together according to abstraction--instantiation relationships on the one hand, and inclusion (or metonymical) relationships on the other hand. In this regard, the notion of schema describes an organization principle for knowledge in memory rather than knowledge itself. With schema theory as a model of the way our knowledge is stored in memory, the choice between strict separation and total equivalence between representation of form and representation of contents loses its meaning. Schemas are embedded and hierarchically structured in multiple fashions: in a schematical network, one concept can be part of several super-ordinate categories: the schema for "cat" is subsumed by the schema for "feline" (which also includes "lion," "tiger," etc.), as well as by the schema for "pet" (with "dog," "parrot," etc.).
In our case, hypermedia users hold a number of representations in memory, composed of "information packets" (about the hypermedia system they are browsing, its contents, formal organization, etc.) and organized together in a schematic way, so that the representation of the whole system includes formal and semantic aspects, even though some specific subschemas only represent one or another of these aspects.
The central hypothesis tested in the experiment presented in this article can be summarized as follows: modifying the formal and interfacial factors of a hypermedia system (without any content modification) entails a modification in the organization of the user's mental representations of the system (and of its informational contents in particular).
The hypothesized modifications are modifications in the relationships within the schematic network of the user's representations of the system. By fragmenting its contents in nodes that are limited in size, by grouping them in sections, subsections, and so forth, also by arranging them according to some organizing principle(s), the system makes some semantic relationships more salient than others. A fair share of the relationships that connect bits of information is supposed to be reflected by hypertext links available in the system, each link representing an important association between two chunks of information. This modulation of semantic relationships must--and this is the hypothesis--have an influence on the schematic organization of the user's representations.
The two versions (HyperDoc 1 and HyperDoc 2) share the same contents (text, images, animations, and headings are identical), which are fragmented into nodes in identical ways, as well as a number of screen layout principles. Nodes are grouped into sections and subsections, each section corresponding to a color. Figure 2 shows a screenshot from HyperDoc 1. This node is located at the third depth level: it is part of the subsection called "the machines that make the Internet," which itself is part of the section called "Computers and the Internet."
[FIGURE 2 OMITTED]
The way nodes are grouped into sections and subsections differs between the two versions of HyperDoc. In HyperDoc 1, nodes are grouped in two series of sections. The first series congregates the issues that are related to computer hardware, network protocols, Internet (Net) applications and services, and social stakes associated with the use of the Internet. The second series gathers more transversal issues: the way information travels on the net, the history of the Internet, and Internet's authorities (Figure 3).
HyperDoc 2 uses the roundtrip journey of data between a client and a server on the Net as a structuring principle. Nodes are grouped and located according to where on the network the issues they are related to are raised. For example, the node dedicated to routers is accessible from every 'crossroads' on the map-like network presented on the homepage (Figure 4), which is where routers are expected to be found. Three extra sections complement this organization: a series of generalities about computers, the way information travels on the Net, and the history of the Internet.
[FIGURE 3 OMITTED]
[FIGURE 4 OMITTED]
Organizational links (Carey, Hunt, & Lopez-Suarez, 1990) differ from one version to another, unlike associative links (within texts and schematic maps), which are identical. The main difference between HyperDoc 1 and HyperDoc 2 resides in their interfacial factors (section and subsection groupings), formal factors being a slighter subsequent difference (section groupings entail changes in the organizational links).
Forty students from the Catholic University of Louvain, Belgium, were used as subjects for the experiment. They all had previous experience with computers and common services offered by the Internet (Web, e-mail, etc.). Although, none of them was an expert user with a deep former knowledge of the hardware and software functioning of computers and the Internet. Subjects received a 12.5 EUR gift certificate for their participation.
Subjects were split into two conditions, each condition having access to one and only one version of HyperDoc. Two variables were controlled to compose the conditions: gender and former education in computer science (students who had taken a computer science class as part of their program were distinguished from those who had not).
Each subject was asked to browse one version of HyperDoc (without being informed that the other version existed) on an iMac with an 800 by 600 pixel monitor. This task included two steps. Subjects were first invited to browse the system freely for no longer than 20 minutes. They were then given four questions that they had to find answers to by looking for relevant information in the system. No time limit was set for that second step. Subjects were instructed that once they thought they could answer all four questions, they were expected to call the experimenter.
The four questions were:
1. What is DNS? How does it work?
2. What does one mean when they say Internet is an intelligent network, designed to withstand partial damage? What transmission process and what machines in particular allow for this?
3. What are the functions of a firewall? In what context is it used?
4. What is a browser? How can one define it? How does it work?
Subjects were informed from the start that the task would be followed by a structured interview during which they would have to answer the four questions (part A) and explain how they had proceeded to find answers (part B). As they received the questions, subjects were also informed that none of the answers could be found in one single node, and that they would be invited to widen their answers as much as possible during the interview. A notepad and a pen was at their disposal during the task. The order in which subjects underwent the two parts of the interview that followed the task was controlled within the two conditions, so as to have an identical number of subjects undergoing the interview in the two possible orders.
Collected Data and Analysis
The remainder of this article will discuss the results from part A of the interview, which focused on the answers to the questions of the task. A partial analysis of the log files from the subjects' sessions (captured from the HTTP server on which HyperDoc was running) will complement this discussion.
The complete transcripts of part A of the interviews were analyzed using a content analysis method developed for this project: Relational Content Analysis (RCA [Fastrez, 2002]). RCA is inspired by Propositional Discourse Analysis (APD [Ghiglione & Blanchet, 1991; Ghiglione, Kekenbosch, & Landre, 1995]), although it does not share most of its epistemological assumptions.
In both methods, the analyzed discourse is split into clauses (that are called context units), and the analysis relies on the identification of key-referents in the clauses and on the inventory of the relationships that connect these referents together (most often corresponding to verbs). The referents are items that belong to the referential universe communicated by the analyzed discourse; they typically correspond to substantives or pronouns. In our case, referents are key notions that are part of the answers subjects formulate.
Unlike APD, the basic context unit of RCA is not the proposition, but the relational predication, as defined by Ron Langacker (1987) in his cognitive grammar. Consequently, the analyzed discourse is split into clauses based on semantic criteria (e.g., a clause needs to contain two referents connected by some relationship), not on grammatical criteria (e.g., a clause needs to contain a subject, an object and a verb). Predications are encoded into the RCA matrix. Each encoded predication lists the referents it includes and the role assigned to each of them in the relationship (Table 1). A predication includes two (or exceptionally three) referents.
Each encoded predication also includes a unique identifier (a primary key, in the database terminology). Hence, some predications can have the identifier of another predication as a referent. Through computer-based crisscrossing, complete embedded predications (with multiple levels of embeddedness) can be reconstructed. Table 1 shows the following excerpt from one of the subject interviews encoded in the RCA matrix. (The codes that appear in parentheses are the relationship abbreviations used for coding--see Table 2). The subject answers question #3 ("What are the functions of a firewall?"):</p> <pre> In fact a firewall is a ... is a ... yes, it's a computer that is part of a network ... in fact ... and ... eventually it controls whatever gets into this ... this network of ... into this local network of computers. (Subject #31, transcript excerpt translated from French) </pre> <p>In this example, the embedded predication corresponding to "a firewall is a (...) computer that is part of a network" would be noted [ID_Firewall * ID_[E-_Computer * E+_Network]]. For the experiment presented in this article, the subjects' interview transcripts were coded into 3970 predications in the RCA matrix: 2017 predications for condition 1, and 1953 for condition 2.
Table 2 shows a summary of the relationship typology used to characterize how predication referents are connected (2).
Analyses based on the RCA matrix, focuses either on the referents (average frequency and order of use within one answer), or on the types of relationships used in predications (average frequency and order of use within one answer), or on both, yielding answers to the following questions: in the subjects' answers to a given question, what referents are the most commonly used? In what order? What are the most common relationships present in the answers, predications? In what order? What roles are assigned to each referent? What referents are the most frequently associated? What types of relationships characterize these associations? Are there recurring expressions (i.e., embedded predications) in the answers for each question?
RCA coding adopts the speaker's point of view. The goal is to be able to describe the organization among the components of the answers. This organization is taken as an index of the way subjects organize and understand their own knowledge in that domain. In the next section, differences in the organization of answers from subjects of the two conditions (what referents they use to answer, how they associate these referents, etc.) will be correlated with differences in terms of formal and interfacial factors between the two versions of HyperDoc, in order to put the working hypothesis defined earlier to the test.
As a complement of the RCA undertaken on the answers to the four questions of the task, a more global and less formalized analysis of those same answers was carried out. The goal of this analysis is to condense discourse fragments (i.e., answers to the questions) into outlines that include keywords corresponding to the key steps of the answer as well as transitions between these steps. These "condensations" can account for the global organization of the subject's discourse, whereas RCA can only represent discursive configurations that are limited in scale. For the answers that allow it, the condensation itself is kept separate from the definitions that are assigned by the subject to the question's keyword (i.e., DNS, firewall, and browser).
RESULTS AND DISCUSSION
Measures Based on the RCA Matrix and Results Presentation
One of the disadvantages of RCA is that it produces abundant results. Consequently, only a part of these results can be presented in this article. Table 3 shows some overall features of the subjects' answers to the task questions as encoded in the RCA matrix.
The subjects used a large number of referents (ranging from 43 to 87 depending on the question--see Table 3) to answer the task questions. Two measures were used to assess these referents' frequencies of use in the answers to each question of the task in each condition. The first measure is the average (for all subjects of a given condition n) of the proportion between the number of occurrences of a given referent i and the number of referents in a given answer 1 by a given subject m:
[F.sub.an[l]co[n].sup.ref[i]] = [F.sub.answer[l]condition[n].sup.referent[i]] = [1/[N.sub.subjectcondition[n]]][summation over (m)][[N.sub.rejerent[i].answer[l].subject[m]]/[N.sub.referents.answer[l].subject[m]]]
The second measure is the number of different subjects within condition n who used referent i in their answer to question 1.
[N.sub.an[l].co[n].sup.ref[i]] = [N.sub.subjectsanswer[l]condition[n].sup.referent[i]]
The same two measures were computed to assess the relationships' frequencies of use. Applying these two measures to all referents and relationships used in the answers to all four questions gave way to 1348 figures to be analyzed (i.e., (62 + 87 + 43 + 73) referents + (18 + 18 + 19 + 17) relationships * 2 conditions * 2 measures).
In addition, the average order of use of all referents and relationships in the subject answers was evaluated. The measure used to estimate the referents and relationships' order of appearance in the subject's answers was the Order Score (OS):
O[S.sub.an[l].co[n].sup.ref[i]] = O[S.sub.answer[l]condition[n].sup.referent[i]] = [1/[N.sub.subjectscondition[n]]][summation over (m)][[[summation over (i)][R.sub.referent[i].answer[l].subject[m]]/[[N.sub.referent[i]answer[l].subject[m]] x [N.sub.predicationanswer[l].subject[m]]]
To obtain the Order Score for referent i, one needs to compute the average rank of referent i (i.e., the sum of the ranks for the different occurrences of i in subject m's answer to question 1 divided by the number of occurrences of i in this answer) and divide it by the number of predications in subject m's answer to question l. This measure is then averaged between subjects of condition n. (The same OS can be computed for any relationship i.) Applied to the RCA matrix, this generated 674 figures for all referents and relationships.
The referents used by at least 10 subjects in one of the two conditions were selected for further analysis (see Table 3). These referents were analyzed in terms of the roles subjects assigned to them in their predications, as well as in terms of the other referents that co-occurred with them in the subjects' predications. The frequency of use of each role-referent association was computed using the following formula:
[F.sub.an[l].co[n].sup.ro[k].ref[i]] = [F.sub.answer[l]condition[n].sup.role[k]referent[i]] = [1/[N.sub.subjects.condition[n]]][summation over (i)] [[N.sub.role[k].referent[i].answer[l].subject[m]]/[N.sub.referent[i]answer[i].subject[m]]]
The number of occurrences of referent i playing role k in subject m's answer to question 1 is divided by the number of occurrences of i in this answer. The average of this proportion is then computed for all subjects of condition n. The co-occurrence of referents' frequency was computed as follows: the number of predications where referent i and referent j co-occur in subject m's answer to question 1 is divided by the number of occurrences of i in this answer. The average of this proportion is then computed for all subjects of condition n.
[F.sub.an[l].co[n].sup.ref[i]xref[j]] = [F.sub.answer[i].condition[n].sup.referent[i]xreferent[j]] = [1/[N.sub.subjects.condition[n]]][summation over (i)] [[N.sub.referent[i]xreferent[j].answer[l].subject[m]]/[N.sub.referent[i].answer[l].subject[m]]]
To combine both referent co-occurrence and role-referent association, the frequencies of the different relationships under which the referents co-occurred were computed, using a formula that combines the two previous ones:
[F.sub.an[i].co[n].sup.rel[k].ref[i]xref[j]] = [F.sub.answer[l]condition[n].sup.relationshp[k]referent[i]xreferent[j]] = [1/[N.sub.subjects.condition[n]]][summation over (i)][[N.sub.relationshp[k].referent[i]xreferent[j]answer[l].subject[m]]/[N.sub.referent[i]xreferent[j]answer[l].subject[m]]]
Finally, the most frequent embedded predications in the subject answers were listed for each of the task questions. The predications appearing in at least four subject answers in one of the conditions (see Table 3) were selected for comparison between conditions.
As stated earlier, the collection of figures generated by the RCA matrix analysis is too large to be presented in its entirety. Hence, only the most relevant differences that came out of the analysis are presented in the next paragraphs. For each result, the specific measures used (among those which were introduced in this paragraph) will be indicated in each case. All the differences mentioned in the next section (based on mean comparisons or proportion comparisons) are statistically significant (p < 0,05 or even lower thresholds in cases where Bonferonni adjustments are necessary).
Similarities Between Conditions
From a global standpoint, the subjects' browsing sessions were comparable between the two conditions (Table 4). The average total time spent browsing HyperDoc was similar between conditions, and so was the average number of nodes browsed by subjects. The average time spent on each node was also similar, and so was the average number of mouse clicks per session.
Also, the subjects' answers shared a number of common features across the two conditions (Table 5). Frequencies of use of the different referents encoded in the RCA matrix were highly correlated between conditions, across all questions: among all referents cited in the answers for one question, the most frequent referents were the same in the two conditions, and so were the least frequent referents. The same was true for the number of subjects using each of these referents, and for the average order of use of these referents. The average number of referents used in the answers to each question was also similar between conditions, except for question #4 (for which subjects from condition 2 used a greater number of referents on average--see Table 3 for averages). There was also a strong correlation between frequencies of use of relationships between conditions.
Despite these similarities on a global level, answers to the task questions differed between conditions on a number of specific elements.
Differences Between Conditions
Two types of differences appeared between the answers of subjects from the two conditions, that can be correlated with differences in terms of formal and interfacial factors between the two versions of HyperDoc.
Node hierarchy. The most obvious kind of modification to formal and interfacial factors that had a fairly specific effect was the difference in position in the hierarchy of nodes. Positioning a node or a whole section one level higher in HyperDoc's hierarchy, as well as making one or several nodes an autonomous section, seemed to cause subjects to pay more attention to these nodes or sections. Several instances were present in the analyses results (Figure 5).
[FIGURE 5 OMITTED]
When answering question #2 ("Internet as an intelligent network"), subjects from condition 2 used information related to routers and information packets more frequently:
[F.sub.an.co.sup.ref[inf.pack.]] = 0.0285 < [F.sub.an.co.sup.ref[inf.pack.]] = 0.0913 and [N.sub.an.co.sup.ref[inf.pack.]] = 7 < [N.sub.an.co.sup.ref[inf.pack.]] = 14
[F.sub.an.co.sup.ref[router]] = 0.0151 < [F.sub.an.co.sup.ref[router]] = 0.0579 and [N.sub.an.co.sup.ref[router]] = 6 < [N.sub.an.co.sup.ref[router]] = 10
In HyperDoc 2 (see Figure 5) information packets and routers each have their own independent section, which are directly accessible from the homepage. In HyperDoc 1, these same nodes correspond respectively to a subsection in the "Exchanges on the Net" section and to one of the nodes in the "Machines" subsection of the "Computers and the Net" section.
The answers to question #2 actually contained a number of referents that were used equally frequently across conditions (i.e., "pathways," "information," "Internet," "damage," and "network" were the most frequent ones). Those were general referents that were related to the most frequent answer step in the condensations: the one about the principle of alternative paths (in case of partial damage, the information can always travel on an alternative line).
Although, some referents were used with different frequencies in the two conditions. For example, predications stating the existence of routers and of information packets were among the most recurrent predications of condition 2. The same was true for the predication presenting information and information packets as being acted upon together.
[N.sub.an.co.sup.pred["[ID_router]"]] = 2 < [N.sub.an.co.sup.pred["[ID_router]"]] = 7 and [N.sub.an.co.sup.pred["[ID_inf.pack]"]] = 0 < [N.sub.an.co.sup.pred["[ID_inf.pack]"]] = 4
[N.sub.an.co.sup.pred["[A-3_inf.*A-3_inf.pack.]"]] = 2 < [N.sub.an.co.sup.pred["[A-3_inf.*A-3_inf.pack.]"]] = 6
Also, in the condensations, routers ([N.sub.co] = 3 < [N.sub.co] = 9), information packets ([N.sub.co] = 6 < [N.sub.co] = 11), and the TCP protocol ([N.sub.co] = 3 < [N.sub.co] = 7) were more frequently used as major steps of the answer in condition 2.
The differences described so far were corroborated by the comparison between average times spent on HyperDoc's nodes depending on the condition. Subjects from condition 2 spent significantly more time on nodes about "information packets" and "packet sending management," on animations about information packets and on the two nodes dedicated to routers (Table 6).
The differences in terms of hierarchical level of one node in the two versions of HyperDoc were not only reflected in the importance allotted to some pieces of information in the answers to the task questions. In the case of question #1 ("What is DNS?"), the "name server" node is the head of the DNS section in HyperDoc 2, and belongs to another subsection in HyperDoc 1 (Figure 6). As a result, subjects from the two conditions interpreted the question itself differently.
[FIGURE 6 OMITTED]
Globally, subjects from condition 1 usually defined DNS as a conversion system, whose acronym they detailed in most cases ([N.sub.co] = 14 > [N.sub.co] = 7); whereas subjects from condition 2 tended to identify it as the name server, the key agent in the system ([N.sub.co] = 3 < [N.sub.co] = 13). This difference showed up in the analysis of the answer condensations for this question. Also, the [ID_DNS * ID_Domain Name System] equivalence was among the most recurrent predications in condition 1, as was the [ID_DNS * ID_name server] equivalence in condition 2.
[N.sub.an.co.sup.pred["[ID_DNS*ID_DomainNameSystem]"]] = 2 < [N.sub.an.co.sup.pred["[ID_DNS*ID_DomainNameSystem]"]] = 9
[N.sub.an.co.sup.pred["[ID_DNS*ID_nameserver]"]] = 8 < [N.sub.an.co.sup.pred["ID_DNS*ID_nameserver]"]] = 3
In terms of relationships between referents, the RCA matrix showed another difference in the association between "DNS" and "domain name:" in condition 1, "DNS" equaled "domain name" more frequently in condition 1 than in condition 2, and it acted on "domain name" in condition 2 more often than in condition 1.
[F.sub.an.co.sup.rel[ID].ref[DNS]xref[DomainName]] = 0.357 < [F.sub.an.co.sup.rel[ID].ref[DNS]xref[DomainName]] = 0.591
[F.sub.an.co.sup.rel[A3].ref[DNS]xref[DomainName]] = 0.643 > [F.sub.an.co.sup.rel[A3].ref[DNS]xref[DomainName]] = 0.364
Furthermore, in terms of roles assigned to referents, the "name server" appeared later in the answers of subjects from condition 1, where it was more often presented as an agent, in comparison with the answers of subjects from condition 2, where it was more often presented as an equivalent to another referent.
O[S.sub.an.co.sup.ref[Name Server]] = 0.578 > O[S.sub.an.co.sup.ref[Name Server]] = 0.357
[F.sub.an.co.sup.ro[A+3].ref[NameServer]] = 0.373 > [F.sub.an.co.sup.ro[A+3].ref[NameServer]] = 0.252
[F.sub.an.co.sup.ro[ID].ref[NameServer]] = 0.196 < [F.sub.an.co.sup.ro[ID].ref[NameServer]] = 0.501
The reason for this is the following: in condition 1, the name server was not part of the definition of DNS that was given early in the answer, even though it remained an important agent in the DNS system explained later in the same answers. In condition 2, this role of agent was assigned either to the DNS itself (since it is the name server), or to a series of less determined referents, such as "one" or "the user."
Moreover, it seemed that the extra attention subjects from condition 2 paid to the name server entailed a decreased interest in the conversion process between domain names and IP addresses. In the condensations, the conversion process was present in a greater number of answers of subjects form condition 1([N.sub.co] = 15 > [N.sub.co] = 11), especially in its detailed version (which went through all the steps a request from a client to a server goes through--[N.sub.co] = 11 > [N.sub.co] = 5).
In terms of relationships between referents, the consecution relationship (used to describe the aforementioned steps from client to server) was more frequently used by subjects from condition 1:
[F.sub.an.co.sup.rel[T]] = 0.081 > [F.sub.an.co.sup.rel[T]] = 0.032.
This difference of definition reflected a difference in the way HyperDoc's nodes were browsed (Table 7): subjects from condition 1 spent more time on the "IP-DNS conversion" node, whereas subjects from condition 2 spent more time on the "name server" node, which is the head of the DNS section in HyperDoc 2 (and does not belong to this section in HyperDoc 1).
Information contextualization. Another type of formal and interfacial factors modification implying different answers between conditions was the integration of a node within a section that contextualizes its contents. This is the case of the "Firewall" and "Browser" nodes.
Let's start out with the firewall. In HyperDoc 1, the "Firewall" node is located in the "Machines" subsection, part of the "Computers and the Net" section, along with three other nodes on local area networks (LAN), intranets and extranets. In HyperDoc 2, the same "Firewall" node is part of the "Local Area Network" section, along with the "Intranet" and "Extranet" nodes (Figure 7).
Henceforth, subjects from condition 1 answered to question #3 ("what are the functions of a firewall?") with more precise information on the firewall itself. The condensation analysis revealed that subjects from condition 1 offered a definition of the term "firewall" more often than subjects from condition 2 ([N.sub.co] = 12 > [N.sub.co] = 9). As a matter of fact, it was defined as a server (or a computer) more often in condition 1 than in condition 2 ([N.sub.co] = 10 > [N.sub.co] = 6). The [ID_Firewall * ID_[E-_Server * E+_LAN]] equivalence was also relatively frequent in the RCA matrix, especially in condition 1 ([N.sub.co] = 9 > [N.sub.co] = 5).
[FIGURE 7 OMITTED]
On the opposite, subjects from condition 2 detailed what the two types of LAN are (i.e., intranets and extranets) more frequently, and were more specific in locating the firewall within a LAN (which corresponded to the second part of question #3: "in what context is it used?"). Based on the condensations, it appeared that a majority of subjects from condition 2 located the firewall at the exit of the LAN (or as "belonging to the LAN but placed at its exit"--[N.sub.co] = 9 < [N.sub.co] = 14). The opposite was true in condition 1: the majority of subjects located the firewall inside the LAN, or as "belonging to the LAN but placed at its exit" ([N.sub.co] = 13 > [N.sub.co] = 11).
These observations were corroborated by the comparison between average times spent on HyperDoc's nodes in each condition (Table 8): subjects from condition 1 spent more time on the "Firewall" node, whereas subjects from condition 2 spent more time on the "Intranet" and "Extranet" nodes.
The "Browser" node is in the "World Wide Web" subsection of the "Services of the Net" section in HyperDoc 1. In HyperDoc 2, it is part of the "Server and Client: Hardware and Software" subsection, inside the "Terminals" section, where it is coupled with the "HTTP Client and Server" node (Figure 8).
[FIGURE 8 OMITTED]
Subsequently, the condensation analysis revealed that there were more subjects from condition 2 than from condition 1 raising the issue of exchanges between server and client to contextualize their answer to question #4 ("what is a browser?"), as a complement to other parts of their answer ([N.sub.co] = 3 < [N.sub.co] = 7). In condition 1, when this explanation showed up, it did so on its own, after a definition and an example of what a browser is.
Apart from the one just detailed, the differences between conditions that can be attributed to variations in the formal and interfacial factors are weak for answers to question #4. This question was the one for which the least differences appeared in the collected data. Correlatively, the average times spent on nodes related to browsers or the Web were not significantly different between conditions.
Other differences between conditions (which will not be developed here), are harder to interpret in terms of specific changes in the formal or interfacial factors in HyperDoc. This is the case for the different ways in which subjects ranked the firewall's functions (all listed and explained in the same node) depending on the conditions; or for the historical perspective (presented in an autonomous section available from the homepage in both HyperDoc versions) that was more frequently included to the answers of subjects from condition 2 (for answers to question #2 and #4). At most, these variations can be attributed to the way these factors oriented the subjects' reading paths, although the collected data does not allow to say any more.
Several significant differences between conditions were presented in the previous section. First, the differences in the way HyperDoc 1 and 2 are structured (in terms of formal and interfacial factors) appear to have entailed different navigational behaviors. Although they spent a similar amount of time (in total and on average by node) browsing HyperDoc, subjects from each condition focused on different specific nodes, that seemed directly related to the things they stressed in their answers to the task questions.
On the other hand, depending on which condition they were in, subjects used the available navigation tools differently (Table 9). Subjects from condition 1 made more frequent use of the links in the text body, the menus and the pretitles than subjects from condition 2, who used the "back" button, the homepage (which does include more links to specific nodes than the one in HyperDoc 1), the schematic maps (in the heading part of the nodes) and the animations more frequently. The seven tested proportions are significant with p < 0.0071 (i.e., 0.05 / 7, following Bonferonni's adjustments for seven simultaneous comparisons).
The differences between conditions appeared to point to a distinction between a type of use that is essentially focused on the textual navigation tools (text links, menus, pretitles) in condition 1, and another that takes more readily advantage of graphic or analog tools (homepage, maps, and animations) in condition 2. The presence of a presentation map as a homepage in HyperDoc 2 can probably account (at least in part) for such a difference.
Moreover, these different navigational behaviors implied different understandings of the acquired information. All the differences that were observed in terms of browsing time of specific HyperDoc nodes corresponded to differences in the answers produced by the subjects: conversion system versus name server for question #1, information packets and routers in condition 2 for question #2, definition of firewall versus intranet for question #3. By contrast, question #4, which showed the least noticeable differences in the way subjects organized their answers depending on their condition, is also the question for which average browsing times did not show any significant difference between conditions.
Not only did the difference in terms of formal and interfacial factors between the two versions of HyperDoc have an effect on the navigational behavior of the subjects, it also affected the way HyperDoc's users understood and organized the knowledge they acquired by browsing its nodes; the differences between the answers of subjects from the two conditions identified earlier provide evidence for this.
This effect--which mainly depends on the way information is hierarchized and contextualized in HyperDoc's structure--occurred through the navigation process, which was described earlier as the inherently interpretative activity through which the user discovers the system in all its dimensions.
As previously noted, answers of subjects from both conditions to the four task questions shared a number of features: similar frequencies of use and ranks for referents, similar frequencies of use for relationships, analogous number of referents per answer, and so forth. The analyses on the collected data showed differences that were sometimes faint, and results that were subtle.
But these subtleties probably are an interesting conclusion themselves: if the observed differences in terms of understanding and organization of acquired knowledge were indeed subtle, it is probably because the hypertext medium allows for subtleties. Earlier in this article, I posed as a hypothesis the ability of hypermedia systems to modulate semantic relationships within a body of knowledge (stressing a part of them and dimming the others). In all cases, all these semantic relationships stay present and integrated in the system, which only puts emphasis on some of them. In the case presented here, the two versions of HyperDoc assign different statuses to their nodes (and henceforth to content chunks) and to the links that connect them, thereby modulating the semantic relationships between chunks of presented information. Still, the entirety of the relationships reified by links in one of the two versions stays available in the other.
In this perspective, hypermedia systems are best viewed as a medium that allows the learner to manage complexity (a view advocated by the supporters of Cognitive Flexibility Theory--Jacobson & Spiro, 1995; Spiro, Feltovich, Jacobson, & Coulson, 1991), to grasp a complex web of semantic relationships, and to consider some of them as more important than others, without forgetting about those. Hence, the results presented in this article are in multiple shades of grey, rather than in black and white.
This article started out with a claim: information comprehension and use in hypermedia systems does not depend solely on rhetorical factors. As the results presented earlier show, formal and interfacial factors do have an influence on acquired knowledge organization. This brings us back to the notion of factor congruence: all classes of factors have a part to play in the user's interpretive activity, and factor congruence has a positive effect on understanding. This yields an important conclusion both for designers and educators. As far as designers are concerned, it means content editing cannot be separated from system and interface design. Designers need to take a wholesome approach towards educational hypermedia development, where the pedagogical project acts as a rationale both for the way contents are selected, presented and organized, and for the shape the system's structure, interface, and mode of interaction take (Tricot & Rufino, 1999). As for educators, it reminds them that technology is not a neutral element in the learning process. The techno-semiotic properties of hypermedia need to be used as yet another means of supporting the learner in constructing coherent representations of the knowledge domain presented in the system. Following the results presented, this can be achieved by placing the most important pieces of information higher in the hypermedia's hierarchical structure, or by surrounding them with relevant contextual information.
Carey, T.T., Hunt, W.T., & Lopez-Suarez, A. (1990). Roles for tables of contents as hypertext overviews. In D. Diaper, G. Cockton, D. Gilmore, & B. Shackel (Eds.), Human-Computer Interaction--INTERACT '90. Proceedings of the IFIP TC 13 Third International Conference (pp. 581-586). Amsterdam & New York: North Holland.
Dee-Lucas, D. (1996). Effects of overview structure on study strategies and text representations for instructional hypertext. In J. F. Rouet, J. J. Levonen, A. Dillon, & R. J. Spiro (Eds.), Hypertext and cognition (pp. 73-108). Mahwah, NJ: Lawrence Erlbaum.
Dillon, A. (1991). Reader's models of text structures: The cases of academic articles. International Journal of Man-Machine Studies, 35, 913-925.
Dillon, A. (2000). Spatial-semantics. How users derive shape from information space. Journal of the American Society for Information Science, 51(6), 521-528.
Dillon, A., & Gushrowski, B. A. (2000). Genres and the Web: Is the personal home page the first uniquely digital genre? Journal of the American Society for Information Science, 51(2), 202-205.
Dillon, A., & Vaughan, M. (1997). "It's the journey and the destination": Shape and the emergent property of genre in evaluating digital documents. New Review of Multimedia and Hypermedia, 3, 91-106.
Edwards, D. M., & Hardman, L. (1989). Chapter 7: "Lost in Hyperspace." Cognitive mapping and navigation in a hypertext environment. In R. McAleese (Ed.), Hypertext: Theory into practice (pp. 105-125). Oxford, UK: Blackwell Scientific.
Fastrez, P. (2002). Navigation hypertextuelle et acquisition de connaissances. Approche semio-cognitive [Hypertextual navigation and knowledge acquisition. A semio-cognitive approach]. Unpublished doctoral dissertation, Catholic University of Louvain, Belgium.
Foltz, P. W. (1996). Comprehension, coherence and strategies in hypertext and linear text. In J. F. Rouet, J. J. Levonen, A. Dillon, & R. J. Spiro (Eds.), Hypertext and cognition (pp. 109-136). Mahwah, NJ: Lawrence Erlbaum.
Ghiglione, R., Kekenbosch, C., & Landre, A. (1995). L'analyse cognitivo-discursive [The cognitive discursive analysis]. Grenoble, France: Presses Universitaires de Grenoble.
Ghiglione, R., & Blanchet, A. (1991). Analyse de contenu et contenus d'analyse [Content analysis and contents of analysis]. Paris, France: Dunod.
Jacobson, M. J., & Spiro, R. J. (1995). Hypertext learning environments, cognitive flexibility, and the transfer of complex knowledge. An empirical investigation. Journal of Educational Computing Research, 12(4), 301-333.
Langacker, R. (1987). Foundations of Cognitive Grammar. Volume I: Theoretical prerequisites. Stanford, CA: Stanford University Press.
Norman, D. A. (1986). Cognitive engineering. In D. A. Norman & S. W. Draper (Eds.), User centered system design (pp. 31-61). Hillsdale, NJ: Laurence Erlbaum.
Norman, D. A. (1988). The design of everyday things. New York: Basic Books.
Rouet, J. F. (1995). Navigation et orientation dans les hypertextes. Quelques aspects du fonctionnement cognitif de l'utilisateur [Navigation and orientation in hypertext. Some aspects of the user's cognitive functioning]. In E. Bruillard, G. L. Baron & B. de la Passadiere (Eds.), Hypermedias, education et formation, Proceedings from the seminar (pp. 11-25).
Rouet, J. F. (1997). Le lecteur face a l'hypertexte [The reader facing hypertext]. In J. Crinon & C. Gautelier (Eds.), Apprendre avec le multimedia, ou en eston? (pp. 165-180). Paris: Retz.
Rouet, J. F. (1999). Interactivite et compatibilite cognitive dans les systemes hypermedias [Interactivity and cognitive compatibility in hypermedia systems]. Revue des Sciences de l'Education, 25(1), 87-104.
Rouet, J. F., & Tricot, A. (1998). Chercher de l'information dans un hypertexte: vers un modele des processus cognitifs [Searching for information in a hypertext: towards a model of the cognitive processes]. In A. Tricot & J. F. Rouet (Eds.), Les hypermedias. Approches cognitives et ergonomiques (pp. 57-74). Paris: Hermes.
Rumelhart, D., & Norman, D. A. (1988). Chapter 8: Representation in memory. In R. C. Atkinson, R. J. Herrnstein, G. Lindzey, & R. D. Luce (Eds.), Stevens' handbook of experimental psychology: Vol.2. Learning and cognition (2nd ed., pp. 511-587). New York: Wiley.
Spiro, R. J., Feltovich, P. J., Jacobson, M. L., & Coulson, R. L. (1991). Cognitive flexibility, constructivism and hypertext. Random access instruction for advanced knowledge acquisition in ill-structured domains. Educational Technology, 31, 24-33.
Thuring, M., Hannemann, J., & Haake, J. M. (1995). Hypermedia and cognition, designing for comprehension. Communications of the ACM, 38(8), 57-66.
Tricot, A. (1997). Que savons-nous sur l'activite mentale de l'utilisateur d'un hypermedia? [What do we know about the hypermedia user's mental activity?]. In Hypermedias, Education et Formation, Proceedings from the Seminar (pp. 1-10).
Tricot, A., & Rufino, A. (1999). Modalites et scenarios d'interaction dans des hypermedias d'apprentissage. [Interaction modalities and scenarii in hypermedia for learning.]. In Revue des Sciences de l'Education, 25(1), 105-129.
van Dijk, T. A., & Kintsch, W. (1983). Strategies of discourse comprehension. New York: Academic Press.
The work presented in this article was funded by a PhD Research Fellowship from the Belgian National Funds for Scientific Research.
(1) The references cited here do not necessarily stress what is highlighted here from their analyses.
(2) In order not to exceed the length limit assigned to this article, this typology (Fastrez 2002) will not be presented in further detail.
(3) The proportions were computed for each subject, and then average proportions were computed for each condition, to give an equal weight to each subject in the final measure. Eventually, it appeared the differences between proportions were statistically significant even when the proportions were computed for the two conditions without balancing the subjects.
University of Louvain
Table 1 An Example of Predications Encoded as RCA Matrix Lines Predication Relationship Ref. #1 Role Referent #1 p1 Identity (ID) Identical entity (ID) firewall p2 Containment (E) Contents (E-) computer p3 Entering Into (M4) Entering entity (M+4) 'something' p4 Action (A3) Agent (A+3) Firewall Predication Ref. #2 Role Referent #2 p1 Identical entity (ID) p2 p2 Container (E+) network p3 Entered entity (M-4) network p4 Patient (A-3) p3 Table 2 Relationship Typology Used with RCA Atemporal Relationships Temporal Relationships Different-level relationships: [T] Consecution [S] Schematicity [E] Embeddedness / Relative motion: Containment [M1] Coming From [M2] Going To Same-level relationships: [M3] Going Through [ID] Identity [M4] Entering Into [IN] Difference [M5] Exiting [C] Collocation [L] Logical Entailment Agency: [P] Prototype-to- [A1] Motivational Causality (Will) Peripheral-Instance [A2] Psychological Causality Relationship (Knowledge, Ability) [A3] Physical Causality (Action) [A4] Conditional Causality (Permission) [A5] Compulsion [A6] Blockage Table 3 Global Characteristics of the Subject Answers Encoded in the RCA Matrix Question Number #1 #2 #3 #4 Number of different referents 62 87 43 73 used by the subjects Average number of Condition 1 15.35 11 10.9 8.421 referents used by the subjects Condition 2 14 12.85 10.474 12.7 Number of different relationships 18 18 19 17 used by the subjects Number of referents used by at 11 9 11 10 least ten subjects in one condition Number of predications used by at 17 11 15 7 least four subjects in one condition Table 4 Global Subject Sessions Characteristics Condition Condition T statistic p 1 (N=20) 2 (N=20) Average total time 5195.9 4885.8 1.713 < 0.05 (sec.) Average number of 223.55 194 1.365 < 0.05 clicks Average number of 77.3 71.25 1.086 < 0.05 nodes browsed Average time per 73.3618 70.3701 0.4028 < 0.05 page (sec.) Table 5 Similarities Between Conditions Question Number Measure #1 [r.sub.F] Correlation between 0.966 [F.sub.an.co.sup.ref[i]] and (p>0.05) [F.sub.an.co.sup.ref[i]] [r.sub.N] Correlation between 0.923 [N.sub.an.co.sup.ref[i]] and (p>0.05) [N.sub.an.co.sup.ref[i]] [r.sub.OS] Correlation between 0.475 O[S.sub.an.co.sup.ref[i]] and (p>0.05) O[S.sub.an.co.sup.ref[i]] t Mean Comparison between average number 0.664 of referents used by subjects of each (p>0.05) condition (see Error! Reference source not found.) [r.sub.F] Correlation between 0.974 [F.sub.an.co.sup.ref[i]] and [F.sub.an.co.sup.ref[i]] Question Number Measure #2 #3 #4 [r.sub.F] 0.849 0.992 0.948 (p>0.05) (p>0.05) (p>0.05) [r.sub.N] 0.923 0.965 0.887 (p>0.05) (p>0.05) (p>0.05) [r.sub.OS] 0.388 0.497 0.426 (p>0.05) (p>0.05) (p>0.05) t -0.982 0.467 -3.185 (p>0.05) (p>0.05) (p>0.05) [r.sub.F] 0.918 0.988 0.981 Table 6 Average Time Spent on Nodes Related to Question #2 Average time (sec.) spent browsing nodes on ... Condition 1 Condition 2 t p "Information packets" 60.75 99.3 -2.284 0.014 "Packet sending management" 25.95 36.65 -2.578 0.008 "Information packets" 79.65 146.85 -1.909 0.032 (animations) "Router" (part 1) 82.7 126 -2 0.026 "Router" (part 2) 29.2 93.9 -3.478 0.0007 Table 7 Average Time Spent on Nodes Related to Question #1 Average time (sec.) spent browsing nodes on ... Condition 1 Condition 2 t p "Name Server" 69.65 178.5 -4.103 <0.001 "IP-DNS 195.6 115.45 2.19 0.017 Conversion" Table 8 Average Time Spent on Nodes Related to Question #3 Average time (sec.) spent browsing nodes on ... Condition 1 Condition 2 t p "Firewall" 288.55 162.25 3.412 0.001 "Intranet" 8.05 27.2 -2.244 0.017 "Extranet" 1.9 9.3 -2.099 0.023 Table 9 Average Proportions of Link Type Usage per Condition (3) Condition 1 Condition 2 [delta] ([F.sub.1]) ([F.sub.2]) ([F.sub.1]- Link Type (N=2839) (N=2902) [F.sub.2]) p "Home" Button, 0.2764 0.2328 0.0436 < 0.0071 Pull-down Menus and Tabs Links Opening a 0.0146 0.031 -0.0164 < 0.0071 Pop-Up Animation Links from the 0.0408 0.1167 -0.0759 < 0.0071 Homepage Schematic Maps 0.0458 0.0752 -0.0295 < 0.0071 Pretitles 0.0428 0.026 0.01686 < 0.0071 Back Button 0.106 0.1352 -0.0292 < 0.0071 Links in Text Body 0.4736 0.383 0.09054 < 0.0071