To date, Janet H. Murray has given the most widely accepted definition of agency as “the satisfying power to take meaningful action and see the results of our decisions and choices” [1, p.126]. A further distinction between theoretical and perceived agency was made by Thue et al. . Theoretical agency can be viewed as the user’s objective ability to act and change the outcome of events within the story, while perceived agency depicts the user’s subjective perception of their ability to do so. In this paper we will discuss mostly perceived agency in the context of the interactive storytelling system Regicide [3, 4].
This work is organized as follows. We start by describing how the system behind Regicide works. This is followed by an analysis of test results and a discussion on their general implications for interactive storytelling systems.
Regicide is a text-based, event-based role-playing game where the user takes the role of Duke Blackwood, the young heir of an ancient noble family, as he seeks revenge for the murder of his father . The goal of the system behind Regicide is to combine separate story elements (called events) into a loosely related and coherent plot, engaging the user with decisions that feel meaningful. This creates a story that can present itself in hundreds of thousands of different ways, followed by an endgame which can also have thousands of different manifestations.
The core gameplay is based on three different stages: the event text, choice, and result stages. In the event text stage, the user is presented with a new part of the story (an event) which requires them to make a choice. These events are chosen by a drama manager trying to uphold a dramatic arc and create tensions between the user and the characters in the story. Each event has a set of 3-5 pre-determined choices which are presented to the user. The choices themselves have a varying amount of possible outcomes, and the most fitting one is always chosen by the system based on the user’s resources and other variables. The event result stage follows the user’s choice, and will display a text pertaining to the chosen result. After this, the user is taken to a new event text stage or the endgame stage, depending on how many story events have passed. Figure 1 shows the progress of a game of Regicide:
The endgame will always occur once 13 events have taken place. Here, the user makes the final choice on how to ascend to the throne, and the result is separated into three different stages. The final result text is the combination of the result texts from each stage together as one long ending for the game. The user needs to succeed in each stage to successfully “win” the game. Stage one consists of the choice the user makes, and if they are successful in that, stage two result is calculated, and if this is successful, eventually stage three result is calculated. The user does not have control over stages two and three at this point, but the results depend entirely on their resources, so the user’s agency is well represented.
One of the key purposes of the research conducted with Regicide was to test the differences in user experiences between three different game modes. The user was unaware of which game mode they were playing and of their differences.
The first game mode acts as a basis of reference; a normal interactive experience the other two game modes can be compared to, while the other two implement different ways of influencing perceived agency while attempting to guide the story towards a certain direction.
The data used in the research was gathered from game sessions and an online survey which the user was prompted to take after playing the game. In total, the survey was answered 108 times out of a total of 768 unique downloads. The values on the x-axis of the following graphs represent the five mandatory scaled (1-5) questions in the online survey, in the following order:
In the graph in Figure 2 the average results for these five questions are organized by the game mode used.
Interestingly, we can notice fairly small differences between modes 1 and 3 in general. Mode 2 garnered more positive results across the board compared to mode 1. The larger differences might be explained by the fact that users who played mode 2 were on average significantly more experienced with role-playing games in general than the users who played mode 1, and as we determined with a control question in the survey, previous experience with role-playing games had an expectedly positive correlation towards the overall user experience.
Interestingly it seems users who played mode 2 were slightly more satisfied with the ending of the game, even though some of these results were forced failures. We can also make the observation that for question #2 the averages were markedly higher for game mode 3. Perhaps the handpicked events created a more cohesive story overall, as was intended; forcing a continuity of events instead of relying on the drama manager to handle it ensures that a certain connection forms between the events of the story.
Based on the small differences between the game modes, the illusion of agency seems to have held up well and the endgame system has managed to create perceived agency in all situations, including situations where the user’s actions always had a predetermined result. An interesting excerpt to the contrary can however be taken from a specific user who answered the survey three times. The first two times the user played game mode 2 and the third time they played game mode 1. What is noteworthy is the feedback the user left after their third game: “I already started to suspect that the game can't be beaten but then I collected a huge military strength and managed to take over the throne.” This is the only recorded instance of a user suspecting that the game cannot be beaten on mode 2. Interestingly however, neither of this user’s game mode 2 losses were actually failures that were forced by the system, so this user’s suspicions are coincidental, but certainly not meaningless; if the user truly believes an illusion of agency exists, their experience will be diluted regardless of the precision of their suspicions. Therefore, it is important to understand the reasons behind such suspicions even when they are non-founded. In this case, we believe the relative difficulty of the game overall may have played a large part in their suspicions.
It is worth noting that replayability and illusion of agency might not be the best combination. Certainly the possibility of noticing the illusion grows after each replay. Regicide’s endgame system was designed in a way that attempted to mask the absence of theoretical agency as best it could according to each game session uniquely with thousands of different combinations of results available.
In the graph in Figure 3 the survey results for the five major questions have been organized by which endgame stage the user reached. This includes all of the survey results, even the ones with a forced ending, all of which happened in endgame stage 2.
We can see endgame stage 2 as generally being the most pleasing for the users. The largest difference can be noted in the question #3. Stage 3 seems to have been a lot less dramatically pleasing for the users than the other two stages, while stage 2 seems to have been in turn much more satisfying than stage 1. The reasons for this might lie in the writing of the endgame results, and the fact that whatever was the last thing to happen doubtless had the most effect on the answer to this question; how dramatically satisfying a story is depends heavily on the ending of the story. The purpose of the endgame of Regicide was the conclusion of all relevant storylines woven into one epic, satisfying ending, and a great many users singled out the ending as a particularly dramatic moment in the story.
While stage 1 and 2 results often portrayed the user’s character dying on the field of battle or being forced into exile after a spectacular failure, stage 3 results in comparison may have felt more subdued and neutral; in many stage 3 results the ultimate result is a peace-treaty between the warring factions. Perhaps if we had pushed for more dramatic conclusions to stage 3 such as a crushing victory or a dramatic defeat at the hands of the user’s enemies instead of more neutral, middle-of-the-road results the survey might reflect more favorably on stage 3. The reason that a stage 1 ending is significantly less dramatic to the users than a stage 2 ending might be because the ending can feel a bit abrupt if the user fails in endgame stage 1. In this regard endgame stage 2 seems to have worked as intended: to create a strong sense of global agency to the end of the game, giving the user’s previous actions meaning and consequence. Users who got past endgame stage 1 felt that the ending was a significantly more fitting conclusion to the story than the ones who had their game end in stage 1. Essentially the further you got in the endgame stages, the more your actions got recognition.
Interestingly, as we can see from the graph in Figure 4, succeeding in the game did not result in enjoying the game more on average. In fact the contrary seems to be true.
Fascinatingly, a failure seems to have delivered a more cohesive and dramatic story. This might be due to the same reasons we mentioned about endgame stage 3; every success result always ended in a stage 3 result which may not have been as dramatically pleasing as the failure results in endgame stage 2 specifically. Several users also specifically said that they enjoyed it when things did not go according to their wishes. As a certain user said: “Pretty much any failed decision in this game was dramatic—to me at least.”
The graph in Figure 5 takes into account only those survey results that correspond to endgame stage 2 failures. These results have further been divided into two categories: normal endgame 2 failures and forced failures experienced during game mode 2 sessions. Endgame stage 2 is the only place where the forced failures could occur, so all of them are represented here. There is a relatively small amount of the forced failure results (7) due to most playthroughs ending in a failure without outside assistance. By comparison, there were 13 results which ended in a normal endgame stage 2 failure.
The differences between the forced and normal endgame stage 2 failures do not seem very conclusive apart from the significant drop in the averages for question #2. The forced failures seemed to suffer from a loss of cohesion compared to the normal endgame stage 2 failures. The most likely reason for this is a slight loss of story cohesion when executing the illusion of agency at the end of the game. It would certainly be within the realm of possibility that some grating sequences of events could happen when trying to create the illusion. The biggest challenge for the developer is to eliminate or dilute such situations to such that they do not stand out for the user.
Influencing perceived agency is not only about creating an illusion of agency; more importantly, it should be about ensuring that the user is aware of all the theoretical agency they possess and can fully appreciate the relevance and power of their own actions and choices. This, in essence, boils down to the communication cycles between the system and the user; proper methods of communication are bound to increase the user’s perception of agency.
In the case of Regicide, this did not always actualize in an ideal way. Some users were confused about the amount of results for each given choice, and whether the resources played any part in the resulting outcomes of choices. It is up to the system to communicate these types of things clearly to the user; any and all theoretical agency not accompanied by perceived agency is utterly useless in an interactive storytelling system created for entertainment purposes. Perceived agency is the most important element when it comes to feelings of relevance in the user, which in turn leads to a more positive user experience.
Because of this it is also important that the user knows how to properly operate the interactive storytelling system and knows what to expect when communicating with it. Therefore, it may be necessary to train the users to use the interactive storytelling system. It is worth noting that Regicide had a very minimal, entirely skippable tutorial that probably was not satisfactory in letting them know exactly how much agency they have in the game.
Additionally, instead of unrestricted freedom, the user should be provided with contextual and important decisions, and the choices should be carefully crafted to serve the context. In the case of Regicide, the context could have been more defined and focused. The broadness of some events caused sub-optimal situations where the user felt they could not act the way they wanted to. It is worth noting that focusing the context of agency is entirely dependent on the interactive storytelling system in use, and the story being told; there is no one-size-fits-all solution.
Overall, the illusion of agency in Regicide was successful and gave sufficient feedback to the user to keep them feeling relevant even in situations where their actions ultimately had only one possible end result. The illusion of agency held up better mid-game (in game mode 3) than in the endgame (in game mode 2), but even the mode 2 endings in which the interactive storytelling system forced the user to fail were generally enjoyable for the users. In the end, I believe that this research speaks volumes for influencing perceived agency through communication and illusion of agency; give the user’s actions relevance and consequence even when they do not affect the ultimate outcome.
 Janet H. Murray, Hamlet on the Holodeck: The Future of Narrative in Cyberspace, 1997
 David Thue, Vadim Bulitko, Marcia Spetch, Trevon Romanuik, Player Agency and the Relevance of Decisions (2010) in Ruth Aylett, Mei Yii Lim, Sandy Louchart, Paolo Petta, Mark Riedl (eds.), Interactive Storytelling, Third Joint Conference on Interactive Digital Storytelling, ICIDS 2010, Edinburgh, UK, November 1-3, 2010 Proceedings, pp. 210–215, 2010
 Eero Itkonen, Influencing Perceived Agency: A Study into User Experiences in Digital Interactive Storytelling, Master’s Thesis, University of Turku, 2015
 Juhani Kyrki, Metrics for Predicting User Behavior and Experience in an Interactive Storytelling System, Master’s Thesis, University of Turku, 2015