The Capacity Limited, Cognitive Constructionist (CLCC) model of presence is proposed as an information processing model of presence, which is demonstrated to have more theoretical power than extant models. The CLCC model defines information processing paths between attention, working memory, declarative memory and procedural memory, which operate to create and maintain a semantic context or bias. Bottom-up information entering the sensory cortices is filtered by attention into working memory where it forms temporary structures encoding the subject’s experience of the VE. These structures also receive top-down information, which arises in declarative memory. This interaction of top-down and bottom-up data gives the entire model a semantic bias which attempts to keep the subject’s construction of the environment semantically coherent. This allows for inferences and decision making, which translates into high presence. A semantically incoherent construction, or one which does not have enough working memory capacity allocated to it will result in poorer inferences about the environment, and reduced presence. If, as the CLCC model contends, presence involves information processing rather than simple perception, then one would expect to see working memory interference effects and semantic content effects in presence. Six studies were conducted to test these conjectures and validate the CLCC model. Studies 1 – 3 examined the role of working memory and attention on presence (the bottom half of the model), while Studies 4 – 6 examined semantic content and processing effects on presence (the top half of the model). Study 1 manipulated working memory (WM) load during VE exploration. The CLCC prediction was that WM load would interfere with presence. Data from 177 subjects showed smaller effects than predicted: No WM effects on spatial presence, lower naturalness under spatial WM load, and lower engagement under verbal WM load. This suggests that spatial presence makes no use of WM, and that engagement and naturalness make limited use of it. While engagement seems to make use of semantic processing as predicted, naturalness seems to use spatial processing. Study 2 examined WM use by media decoders by repeating Study 1 with a text-based VE. Data from 114 subjects shows no WM effects exist on any of the four ITC-SOPI factors. This supports Study 1’s finding that spatial presence does not use WM, but 3 contradicts results engagement and naturalness. Study 3 examined the relative contribution of attention and WM. 46 subjects viewed VE walkthroughs in three conditions: Viewing one walkthrough only (baseline), viewing two walkthroughs simultaneously (WM load condition), or viewing one walkthrough and a jumbled video simultaneously (attention load condition). The CLCC model predicted the WM load condition would interfere with presence the most, followed by the attention load condition, followed by the baseline. No difference was found across conditions, although naturalness and engagement predicted task performance, indicating that spatial presence is distinct from these factors, in agreement with the findings of Study 1 and 2. Study 4 was a survey of semantic and processing effects on presence. Data from 101 computer gamers indicate that it is how often gamers play presence games (and not how many years they have been playing) that predicts how important they consider presence to their gaming experience. This suggests a moderate term activation effect rather than a long term learning effect. Furthermore, gamers with a high thematic inertia rate presence as important to gaming, indicating a processing effect. Finally, gamers who are capable of integrating non-diegetic music into their experiences rate presence as more important, which supports the CLCC notion that information processing of both semantic and perceptual information is important to presence. Study 5 followed up Study 4 by focusing on one specific content area. 461 flight simulation gamers completed the survey. Findings largely agree with those of Study 4, and strongly support the CLCC model prediction that highly specific expectations of content will reduce presence, while generalized expectations will increase it. Thematic inertia and priming were are also positively associated with presence, as predicted by the CLCC model. Study 6 manipulated non-diegetic information (background music) and semantic priming to test semantic processing in presence. The CLCC model predicted that all VE related information (semantic or perceptual) contributes to presence, particularly engagement and naturalness. 181 subjects were primed with materials semantically relevant or irrelevant to VE content, and then experienced the VE with no background music (baseline), music which semantically fit the VE, or VE music which was not a semantic fit. Priming did not influence presence as predicted, but non-diegetic music which fit the VE increased naturalness as predicted. The no-fit music produced the same presence scores as the baseline 4 condition, indicating that it was filtered out by attention, as predicted by the CLCC model. Overall, the CLCC model and data show that content effects occur in presence, and how these are mediated by declarative memory. It also shows that presence is a complex multi-level processing phenomenon. Spatial presence is at a cognitively low level, relying on perceptual (bottom-up) information, while engagement and naturalness are heavily dependent on conceptual (top-down) information, operating as a set of expectation-content comparisons which, when met by the content, lead to enhanced presence. These high and low cognitive forms of presence are largely independent, but do share some semantic effects, likely due to a reliance on common underlying cognitive processes such as priming and thematic inertia. The top half of the CLCC model (which encodes semantic meaning and explains content effects) is better supported that the bottom half (which predicted interference and attention effects). This finding is highly unexpected, as the literature on almost all extant models predicts an important role for attention in presence. From the data however, one must conclude that spatial presence makes no use of working memory, while cognitive higher forms of presence make use of limited amounts of working memory.

South Africa has one of the highest HIV+ prevalence rates in the world [1]. Furthermore, social support is beneficial to HIV+ people. Informational support is a type of social support which is used to increase one’s knowledge base [2]. Hayes et al. state that informational support is especially beneficial for those in the early stages of HIV infection [3]. Computer technologies have been used successfully in providing informational support to their users. However, virtual reality (VR) is a relatively unexplored technology in South Africa, and we feel it is a highly appropriate medium for a context where users have little or no prior computing experience. Traditionally, computer interfaces require users to have a learned skillset, but a VR interface does not necessarily require this as it maps more directly to users’ natural interaction techniques with the real world. A key benefit of a virtual environment (VE) is the interactivity and user involvement that it offers through a high degree of navigation and interaction with objects [4]. VR may, initially, seem to be an expensive technology to use in a developing country but it is possible to make use of desktop VR on a consumer-grade PC relatively affordably. This dissertation presents a comparison of the effects of two media, VR and paper (i.e. pamphlets) in communicating supportive information to an HIV+ sample group. We created a VE to provide social and informational support for HIV+ people in the South African context. The design of the VE placed emphasis on creating a typically South African space which users could recognize and find familiar. Our research focused on two rooms containing virtual agents and points of possible interaction: the lounge and the kitchen. In the lounge, a HIV/Aids support group was simulated while the kitchen contained two areas which presented nutritional informational support: Diet and Cleanliness & Hygiene. We conducted a pre- post-test study with 22 HIV+ women at two clinics in Cape Town. Participants were randomly assigned into one of three groups. One group experienced the informational VE (VE), one group received information pamphlets (Text), the control group who received no information until the end of the study (Ctrl). Participants attended three experiment meetings over a five week period. Participants completed two 3-day food diaries and completed questionnaires that provided measurement for two sets of variables: Food Safety Behaviours (a measure of knowledge of correct food and water safety practices to prevent food-borne illnesses) and Dietary Quality (measure of the diet quality – in terms of quantity, variety, water intake and vitamin supplements, as well as specific food items for the prevention of stomach ailments, a common complaint of HIV infection). While we found no differences between the Text and Ctrl groups, the VE group showed a significant improvement in consuming two (of three) specific food items recommended for the prevention of stomach complaints. This is a particularly striking result given that more than half the participants stated that they routinely did not have enough money to buy food let alone specific healthy foods. The area that contained the information related to stomach complaints was the last imagery experienced by all VE participants. That it was the only area that showed improvement highlights how careful VE authors should be in choosing the actual content for the environment, as well as how that content is delivered. Despite very minimal computing experience and only short training sessions, all participants mentioned that they found the VE easy to use and enjoyed their experience of it. Our results show that VR can indeed be used to deliver informational content to HIV+ women in South Africa.

Digital video games are an immensely popular form of entertainment. The meaningful positive experiences that games facilitate are fundamental to the activity; players are known to invest a lot of time playing games in search of those experiences. Digital games research is polarized. Some studies find games to be a healthy hobby with positive effects; games promote well-being through regular experience of positive psychological experiences such as flow and positive emotions. Others have identified rare problematic use in those players who devote excessive amounts of time to gaming, associating them with social dysfunction, addiction, and maladaptive aggression. While it remains unclear if games cause these effects, or merely coincide with play, the negative effects historically receive more attention in both popular media and academia. Some authorities attempt to reduce the harms associated with games to such an extent that their methods have become national policy affecting all players including those who exhibit no negative outcomes. In South Korea and Taiwan, policing authorities employ a behaviour policy that sets strict daily limits on session length, thereby controlling the amount of time people spend playing games each day. In China, the General Administration of Press and Publication employ a design policy requiring that games service-providers fatigue their games’ mechanics after a period to coax them to take a break sooner than they ordinarily would. Both policy types alter player interaction with games in any given session and it is unclear how these policies affect players in general. This research aims to compare sessions affected by the behaviour policy, design policy, and policy-free sessions in terms of session length, measurable subjective user-experience, the player's intention to return to the game, and their reasons for choosing to stop playing in a particular session. For use in a repeated-measures experiment, we modified the action RPG Torchlight II to simulate both policies. Participants had one session at the same time each week for three consecutive weeks. In varied sequences, participants played a control session unaffected by policy, a one-hour shutdown session representing behaviour policy, and a fatigue session representing design policy. After each session, we recorded their session's length, their user-experience in terms of flow and affect, their intention to return to the game, and their reason for ending the current session. We found that our shutdown condition successfully decreases session length, when compared to the other conditions. The condition facilitates strong flow, moderate positive-affect, and weak negative-affect. The shutdown event does not appear to degrade positive experiences and makes participants slightly more upset (statistically significant) than they would be after choosing to stop playing. This is because players do not get to make that decision, and because players are unable to complete the goals they have set for themselves. Most players intended to play the game again immediately or sometime later in that same day, much sooner (statistically significant) than they would after choosing to stop. This also may be due to satisfaction associated with choosing to stop, or being unable to complete their self-set goals. We found that our fatigue condition increases session length when compared to the other conditions. This result contradicts the intentions associated with design policies: shorter sessions. The fatigue mechanics make the game more difficult, which increases the time required for players to complete the goals they have set for themselves, whether it is to complete a level, quest, or narrative sequence. The condition facilitates high levels of flow, moderate positive-affect and low negative-affect; the condition does not appear to degrade these positive experiences, nor increase negative experience. Most players intended to take the longest breaks between sessions of at least one day, and although we observed that these were longer than the control condition, the differences is not statistically significant. We found that most participants chose to stop playing when the game stopped providing them with positive experiences, or begins to generate discomfort. A large group of participants chose to stop because another activity took priority. Few participants chose to stop because they were satisfied with their session. Less than one third of players explicitly referenced the fatigue mechanics in their decision to stop. Neither policy is holistically better than the other. Both provide strong positive experiences, and have different effects on session length. Whereas it appears that the fatigue condition fails to reduce session length, it also appears that players intend taking longer breaks between sessions, which may reduce total play-time across all sessions. Similarly, the shutdown condition may increase total play time, or at least bring it closer to normal amounts of play-time while also making players more upset. Our operational definition of user-experience is bi-dimensional, and does not include many experiential constructs commonly associated with digital games. During this research, several reliable and valid, and more representative experience measures became available. Any future work on this topic should make use of one of these. Our experiment tested the effects of player experience associated with a single game, genre, and context. Future research should reduce the variation of player factors by focusing on single personalities, typologies, or risk-factors rather than generalizing to all players. We tested out participants only as they played in the early stages of Torchlight II. It is possible that the game's narrative elements, rather than the gameplay mechanics fatigued by the design policy, motivated continued play. We suggest a longitudinal study of the individual policies to explore their effects over many sessions.

This dissertation reports on the results on an exploratory investigation into the potential efficacy of VR as both a support mechanism to people living with HIV/AIDS, as well its capabilities as an emotive medium. Two hypothesis were presented viz. (1) VR will be a form of social support and (2) VR will have an emotional impact on participants. The research builds up on findings which demonstrate the therapeutic effectiveness of telling personal and collective narratives in an HIV/AIDS support group. This fact, together with the tested ability of VR as a therapeutic medium, let to the development of a virtual support group with an aim to test its therapeutic efficacy. A low cost, deployable desktop PC based system using custom software was developed. The system implemented a VR walkthrough experience of a tranquil campfire in a forest. The scene contained four interactive avatars who related narratives compiled from HIV/AIDS patients. These narratives covered the aspects of receiving an HIV+ diagnosis, intervention, and coping with living with HIV+ status. To evaluate the system, seven computer semi-literate HIV+ volunteers from townships around Cape Town used the system under the supervision of a clinical psychologist. The participants were interviewed about their experiences with their system, and the data was analyzed qualitatively using grounded theory. The group experiment showed extensive qualitative support for the potential efficacy of the VR system as both a support mechanism and an emotive medium. The comments received by the participants suggested that the VR medium would be effective as a source of social support, and could augment real counselling sessions, rather than replace them. The categories which emerged from the analysis of the interview data were emotional impact, emotional support, informational support, technology considerations, comparison with other forms of support, timing considerations and emotional presence. The categories can be grouped according to the research questions viz. + The efficacy of VR as an emotive medium (Presence, Emotional Impact, Computer Considerations) + The efficacy of the VR simulation as a source of social support (Emotional and Informational Support) Other themes not anticipated by the data included the following: Timing considerations and Comparison with other forms of counselling. The interviews suggested that both hypothesis 1 and 2 are correct viz. that the VR system provided a source of social support, and has an emotional impact on the participants.

Personal informatics are important tools in health self-management as they support individuals to quantify and self-reflect on their lifestyle. Human-computer interaction researchers have devoted resources on studying how to design such tools. Various motivational strategies have been explored for their capabilities in improving user engagement. However, such strategies are developed with an assumption that the targeted consumer of information is the one directly manipulating user interfaces of the system that has information. This may not always be the case for users in developing regions. As a result, such systems may not scale well in contexts where a targeted consumer (beneficiary) may use technology through the facilitation of another person (intermediary) whom is responsible for manipulating user interfaces, because such facilitators are not recognized as part of the system, hence motivational strategies don't cater for them. In order to uncover design implications for intermediated technology use in the context of personal health informatics (PHI), the researcher started with the theoretical framing of the work followed by a contextual enquiry which led to development of mobile applications' prototypes for tracking nutrition and physical activity. Evaluation of the prototypes revealed that a familial relationship is a prerequisite for such an intervention. The most promising combination involves family members, possibly a child and a parent working together. The study used self-determination theory to understand how a collaborative gamified system can increase engagement. The result revealed that gamification as the source of a significant increase in perceived competence in intermediary users whom also tended to consider themselves as co-owners of the interaction experience. Therefore, gamification was found to be a catalyst for increasing collaboration between an intermediary and beneficiary user of technology, provided that the two users that formed a pair had a prior social relationship. In the absence of gamification, intermediary users tended to be less engaged in the intervention. The study highlights both the positive and negative aspects of gamification in promoting collaboration in intermediated use and its general implications in health settings. Design considerations required in order to improve the overall user experience of both users involved are proposed. In general, this work contributes to both theory and empirical validation of factors for, supporting proximate-enabled intermediated use of personal health informatics.

This dissertation explores virtual storytelling for conveying cultural stories effectively. We set out to investigate: (1) the strengths and/or weaknesses of VR as a storytelling medium; (2) the use of a culturally familiar introductory VE to preface a VE presenting traditional storytelling; (3) the relationship between presence and story experience. We conducted two studies to pursue these aims. Our aims were stated in terms of effective story experience, in the realm of cultural heritage. This was conceptualised as a story experience where story comprehension, interest in the story’s cultural context and story enjoyment were achieved, and where boredom and confusion in the story were low. This conceptualisation was empirically validated by our studies. Three storytelling scenarios were created to tell a traditional San story: text (T); a storytelling VE with no introductory VE (VR+NI); a storytelling VE with a hip-hop themed introductory VE (VR+I). These scenarios comprised our experimental conditions. Questionnaires, measuring interest in hip-hop and the story experience aspects identified above, were developed and psychometrically validated. Study 1 was conducted with a sample of 44 high-school learners and Study 2 with 98 university students. Both studies used a between-subjects design. Study 2 was a refined version of Study 1, improving Study 1’s questionnaires for use in Study 2 and considering two additional variables: attention to the story and perceived strangeness of the story. For our first aim, story experience in the text and VR storytelling scenarios were compared. In Study 1 and 2, comprehension was significantly higher in the T condition than in the two VR conditions combined and attention was higher in Study 2’s T condition. Therefore, we conclude that text is better for achieving story comprehension. In Study 1, interest and enjoyment were significantly higher in the VR condition, while boredom was higher in the T condition. But, no significant differences between text and VR were noted for these variables in Study 2. Comparisons of the T and VR conditions across Study 1 and 2 showed a particularly poor story experience in Study 1’s T group; we speculate that this was due to differences in Study 1 and 2’s samples and procedures. Barring this, there were no interest, enjoyment or boredom differences between T and VR across Study 1 and 2. Thus, we conclude, conservatively, that text and VR are equally good in terms of interest enjoyment and boredom. Confusion was higher in Study 1’s T condition, but this result was counter-intuitive since this condition had also shown higher comprehension. In contrast, Study 2’s VR condition showed significantly higher confusion and lower strangeness. We conclude that Study 1’s participants had reported strangeness rather than confusion and, while virtual storytelling resulted in more confusion, it also resulted in less perceived strangeness of the story. Presence and story experience in the VR+NI and VR+I storytelling scenarios were compared for our second aim. The introductory VE only had an effect for participants who showed a pre-existing interest in hip-hop. In Study 1’s VR+I condition, hip-hop interest was a significant predictor of enjoyment. In Study 2’s VR+I condition, those who identified hip-hop as a favourite music genre showed significantly higher presence than those who identified other genres as a favourite. This suggests that strongly themed introductory VE’s do not benefit virtual storytelling, and that content familiarity and preference interact with VE content to influence virtual experiences. Regarding our third aim; we did not find strong evidence of a relationship between presence and story experience since presence only correlated significantly with interest in Study 1.

Intermediaries are necessary to overcome challenges of Internet use for many users in the developing world. However the need for co-presence with intermediaries can be inconvenient for beneficiaries, and the process is time consuming for intermediaries. We work with an NGO programme called Link which wanted to expose high school students from low income urban communities of Cape Town to Internet career guidance content, but did not have the staff power for regular in-person meetings with all of the students they wanted to reach. We present Internet-supported intermediation, in which intermediaries create a source of content that is tailored to beneficiaries and is accessible using the most appropriate Internet technologies for the context. We discuss the use of two technologies, the web as accessed by conventional computers (preferred by the NGO), and the mobile Internet accessed through low end feature phones. In the target demographic the mobile Internet is very popular for entertainment, especially because of low cost for communication via instant messaging, but the web is more frequently used for tasks outside of entertainment. Using an Action Research approach we implement two Internet systems to support Link intermediation, one a conventional website and the other a text interface suitable for access through mobile instant messaging. We evaluate the systems to determine whether they increase the impact of Link intermediation, and compare the usage of each to determine relative adoption of the technologies for a task outside of entertainment. Students demonstrated capable but slow website use in controlled evaluations, but almost no use occurred outside of our presence. Most students were experienced mobile Internet users, and some began unsolicited use of the mobile system and demonstrated it to their peers. In eight months of simultaneous deployment website users demonstrated minimal engagement, while mobile system users made repeated visits at all hours of the day from varied locations such as homes spread across the city. Students’ most frequent use of computers took place at venues where many users competed, and they prioritised other activities over the website during that access. Mobile use could take place when these restrictions did not apply. The mobile system demonstrated the benefit of Internet support for intermediation: the number of students who viewed career guidance content through it no longer affected the Link team’s effort, while students no longer had to travel to a single meeting place to access content. The consistent higher use of the mobile system than the website shows that the mobile Internet is suitable for non-entertainment use cases by low income urban youth.

Recent studies have shown that the social context in which people play digital multiplayer games has an effect on their experience. Whether co-players are in the same location (―co-located‖) or in different locations (―mediated‖) changes how they interact with the game and with one another. We set out to explore how these complex psychological dynamics played out in a collaborative multiplayer game, since most of the research to date has been focused on competitive gameplay scenarios. To this end, we designed a two-player puzzle-based gaming apparatus called COLAB, implementing specific features that have been proven to foster collaboration and preclude competition between players. The independent variable was player location; the dependent variable was game experience, as measured by the Social Presence in Gaming Questionnaire and the Game Experience Questionnaire, two comprehensive self-report instruments. We found a significant difference in the game experiences of players collaborating in the same location versus players collaborating in different locations. Specifically, co-located players of the collaborative game experienced significantly higher scores for negative experience than mediated players did, while mediated players experienced significantly higher levels of three key game-experience measures: positive affect, immersion, and flow.

This dissertation builds upon the ideas introduced by Liben-Nowell and Kleinberg in The Link Prediction Problem for Social Networks [42]. Link prediction is the problem of predicting between which unconnected nodes in a graph a link will form next, based on the current structure of the graph. The following research contributions are made: • Highlighting the difference between the link prediction and link detection problems, which have been implicitly regarded as identical in current research. Despite hidden links and forming links having very highly significant differing metric values, they could not be distinguished from each other by a machine learning system using traditional metrics in an initial experiment. However, they could be distinguished from each other in a "simple" network (one where traditional metrics can be used for prediction successfully) using a combination of new graph analysis approaches. • Defining temporal metric statistics by combining traditional statistical measures with measures commonly employed in financial analysis and traditional social network analysis. These metrics are calculated over time for a sequence of sociograms. It is shown that some of the temporal extensions of traditional metrics increase the accuracy of link prediction. • Defining traditional metrics using different radii to those at which they are normally calculated. It is shown that this approach can increase the individual prediction accuracy of certain metrics, marginally increase the accuracy of a group of metrics, and greatly increase metric computation speed without sacrificing information content by computing metrics using smaller radii. It also solves the “distance-three task” (that common neighbour metrics cannot predict links between nodes at a distance greater than three). • Showing that the combination of local and temporal approaches to link prediction can lead to very high prediction accuracies. Furthermore in “complex” networks (ones where traditional metrics cannot be used for prediction successfully) local and temporal metrics become even more useful.

Achieving sustainability requires major changes in several areas in which society makes use of technology to meet human needs and while doing so influences the environment, such as agriculture, mobility, power production and water management. The awareness of a need for radical changes is accompanied by an increasing recognition of the interconnectedness of technological, socio-cultural and environmental elements and processes. This has led to an increasing amount of research on system innovations. System innovations refer to changes to a "structurally different" system involving radical changes in the technological and socio-cultural domains and are often contrasted to incremental (technological) change. System innovations involve many actors and many factors, and developments at multiple levels interact. Control over such processes is distributed, they are laden with uncertainty and they exhibit sometimes surprising and unexpected behaviour due to non-linear dynamics and emergent properties involved. Our current understanding of system innovations is limited and the need for an enhanced understanding has clearly been recognized. Computer simulation models seem a promising tool to that end as they already proved to be useful to enhance the understanding of complex systems in many fields like complex chemistry, ecosystems and physics. However, system innovations are mostly processes in social systems. In the social sciences, the application of formal simulation models has a far shorter history and the availability of formalized (and widely accepted) theories and generalizations is low compared to the natural sciences. It is thus not clear-cut which role computer simulation models can play with respect to system innovations. This thesis fathoms the potential of computer simulation models for enhancing our understanding of system innovations and takes some first steps towards fruitful application of models. A theoretical and methodological discussion outlines how models can in principle contribute to an understanding of social macro-processes through facilitating a causal reconstruction of processes that account for the respective observed phenomenon. The view adopted regarding the representation of the social world thereby is that of reciprocity of agency and structure. Compared to the sociological literature the perspective is extended beyond comprising actors and institutions but encompasses also other entities, especially technological artefacts. The thesis then relates the current state of empirical and conceptual work in the field of transition research (the terms "transition" and "system innovation" are used interchangeably) to insights from modelling of complex systems. The intrinsic characteristics of system innovations and the knowledge base available to study them are explicated and implicated challenges and opportunities for model application are discussed. This is complemented by a review of the few existing models of system innovations. The thesis further develops a specification of the regime concept. A regime refers to a dominant structure which originates incremental change but resists system innovations. The concept of "regime" is at the heart of the multi-level perspective, the most widely used framework of transition research, but it is yet only loosely defined. The absence of shared definitions, concept specifications and operationalizations of key concepts of transition research is a major obstacle for defining (and especially for comparing) models. In this thesis, five defining characteristics of regimes are developed and a method to structure and graphically represent knowledge about a regime is introduced. Furthermore, theoretical and conceptual work has been complemented by hands-on experience to make methodological and theoretical deliberations tangible. An agent-based model has been developed which addresses the transition from rainfed to irrigated agriculture in the Upper Guadiana, Spain. The purpose of the model is to bridge a gap in the explanation for the observed process. Case specific literature provides information on driving forces (technological development, changes in regulations) and consequences (amount of irrigation). The model focuses on the farmers which "translate" driving forces into practices of irrigation and water use. It studies the effect of weights farmers attach to a list of priorities. The main findings are that interactions of factors have to be considered and that it is important to acknowledge heterogeneity of farm types to understand empirically observed land-use changes. Based on the outlined work, different possibilities to model system innovations have been abstracted and discussed with respect to their advantages and limitations: a) functional subsystems, b) interacting structures (niches, regimes and landscape) as suggested by the multi-level perspective and c) micro-level entities (actors, technological artefacts, institutions, etc.). None of these representations is superior to the other ones per se but each features certain advantages and drawbacks. The model purpose is a necessary guideline to choose an appropriate representation and to distinguish those parts and aspects of a system which need to be captured from negligible ones. The main findings of this thesis can be summarized as follows: System innovations feature several characteristics which put model-based approaches to this topic on the most challenging edge of the broader endeavour of understanding and modelling social systems. Those are the significance of emergent decay and re-creation of structure during system innovations; the vast scope of system innovations involving several types of subsystems (consumption, production, governance, and nature); the intertwinement of system innovations with their governance – a field which is hardly accessible to modelling; the complexity of the topic; and the unpredictability of innovations. Still, it is concluded that models can be useful as thinking tools. In any case, given the complexity of the topic and the underdeveloped knowledge base, adhering to transparency is essential. In a field as vast and complex as system innovations this requires either very strong simplifications or restricting a model's scope to some parts or aspects of an overall process. This thesis proposes to make use of existing building blocks of understanding of an intermediate level of complexity – e.g. timing and kind of multi-level interactions - to define abstractions and model scope. The challenge to identify, specify, understand and relate conceptual building blocks, to identify the contexts and situations in which each of them becomes relevant and to explicate their role in the overall system innovation could be an agenda for transition modelling for the coming years. Modelling system innovations will remain a huge challenge in the near future. However, this thesis fathoms that models can be valuable tools contributing to the enhancement of the knowledge base of the field; little by little adding to answers of the "big questions". The specific role(s) models of system innovations can play in this endeavour needs to be further explored and discussed.