The Role of Psychological Reactance in Human–Computer Interaction

Classical usability and user-experience testing involve a thorough and systematic evaluation of the system that is under evaluation. This may be done with user-centric, or expert-centric methods [10, p. 24]. The goal of such an evaluation is usually to test if a system acts as expected and meets the requirements that its users have [8, p. 319].

The state of the art chapter gives an overview on psychological reactance. Psychological reactance can be split up into two concepts. The first concept is state reactance. State reactance is a motivational state that people can enter after experiencing a threat to their freedom of choice or them being in control. The other concept is trait reactance. Trait reactance is a personality trait and can be regarded as a person’s proneness of entering a reactant state. State reactance is the main focus of this book. At first, possible consequences of state reactance are introduced. These are typical behaviors that people engage in when they enter a reactant state. Afterward, the construct of state reactance is introduced in detail. It is argued, that state reactance can be regarded as an intermixture of negative cognitions and anger. After the concept of state reactance has been introduced, possible moderator variables are described which might influence the effect of state reactance on the acceptance of technical devices or services. Following the moderators, different techniques of assessing state reactance are introduced, which may be used to measure state reactance in the course of this work. The next section introduces the concept of trait reactance. Trait reactance is a personality trait that is relatively stable over time and does not depend on the current situation. Afterward, different questionnaires that can be used to measure a person’s level of trait reactance are introduced. One of these questionnaires has been undergoing an extensive amount of tests and revisions since its introduction. In order to illustrate its problematic factor structure, the history of this test is presented in detail. The identified assessment tools for state and trait reactance are discussed in terms of their usefulness for reactance research in the context of human–computer interaction at the end of Part I. It is concluded that two of the trait reactance measures are adequate to be used as unidimensional questionnaires in the further course of this work. However, no measure for state reactance is regarded as adequate, therefore a new questionnaire has to be constructed.

Part II deals with the question whether psychological reactance, a construct that stems from social psychology, is of relevance for human–computer interaction. In order to answer the research question, a literature review, a qualitative study and a laboratory experiment had to be conducted.

The first step in the literature search was finding relevant papers that dealt with reactance in the context of human–computer interaction. In order to find such papers, a literature search was performed in the ACM digital library (https://​scholar.​google.​com/​) publication database, Google Scholar (https://​dl.​acm.​org/​), as well as in ScienceDirect (https://​www.​sciencedirect.​com/​) and also in the references of identified papers. The results of the research that is described in Chaps. 2 and  3 have been published in  [7]. The texts in this chapter roughly follow the published paper but data has been reanalyzed. That publication also includes the publications [5, 6] which were published by the author of this thesis. These publications cover experiments that are presented in Chaps. 4 and  8 and are therefore not included in the current literature analysis.

A qualitative survey among 21 usability experts was conducted to integrate the situations from the scientific literature into observations from practice. The expert survey resulted in five clusters of situations that showed a different distribution of situations. While five of the twelve scientific papers from the literature search investigated state reactance in connection with persuasive attempts, none of the experts from the expert survey reported such a situation. Instead, multiple experts mentioned reduced freedom of choice, high effort or costs or scarceness as situations that triggered reactance, all of which were only covered by one scientific paper. Also, five experts reported to have experienced state reactance as a consequence of system errors, a state reactance trigger that has not been investigated in scientific literature, at all. The following research question asks for the consequences for the interaction if technical systems triggered state reactance in their users. The largest amount of information regarding this question was obtained from the scientific literature. Numerous consequences, including boomerang effects and avoidance behavior have been reported in the reviewed papers. In contrast to this, the reported consequences for interaction with technical systems from the survey were scarce but drastic. Two experts reported that they completely refrained from using the systems that caused state reactance, anymore.

Analyzing the identified literature came up with little evidence about whether trait reactance influences human–computer interaction. A laboratory experiment was conducted to further clarify this question. Results showed that highly reactant participants perceived a higher threat to freedom when interacting with a system that adapted itself, compared to when they interacted with a system that they could control. However, the experiment did not produce any evidence that this freedom threat results in any negative consequence for the rating of the system itself. Part II closes with a discussion on the results of the literature review, the qualitative survey, and the laboratory experiment. Also, methodological shortcomings of the describedresearch methods are discussed.

This chapter discusses the findings and shortcomings of the research done to answer Research Question 1. The results from the literature search and the qualitative survey showed some discrepancies between the situations that are investigated in the available literature on reactance and the situations that were mentioned in the qualitative survey. Methodological considerations are discussed, as well.

In order to investigate state reactance and its role in human–computer interaction, it is important to be able to measure it easily and reliably. Also, it would be of advantage to have a measurement tool, that can be easily included into the standard test battery of usability and user experience tests. All existing tools were deemed as unsuitable for easy and reliable assessment of state reactance during the evaluation of technical systems, so a new measurement tool had to be constructed in the form of a questionnaire. A series of thoroughly planned and executed studies was conducted to create a questionnaire following the intertwined model by [10]. The series of studies started with a review of existing questionnaires and the collection of fitting phrases from anonymous user comments on news websites. Afterward, a brainstorming session with experts in human–computer interaction was conducted to provide a pool of phrases for the items. These phrases were then used to design a first set of items that were used in an online study to collect a large number of user responses. The resulting dataset was split in two, the first half was used to conduct a factor analysis based on maximum likelihood factor analysis. It resulted in a questionnaire consisting of three dimensions that aimed at assessing levels of Anger, negative cognitions and the strength of the perceived freedom threat. The second part was used for a confirmatory factor analysis, based on structural equation modeling. The resulting questionnaire satisfied all criteria for reliability and model-fit that were set in the course of construction.

After development of the Reactance Scale for Human–Computer Interaction (RSHCI) was finished, its validity in terms of whether it really measures state reactance and perceived freedom threat was investigated. To this end, a number of criteria to assess criterion validity and inner criterion validity was formulated. Validation was performed on the basis of data collected in the online study that was also used to select the items which were suitable for the questionnaire. Also, a small laboratory experiment was conducted, in which participants interacted with intelligent personal assistants. The gathered data was used to correlate the state reactance dimensions of the RSHCI with another, established state reactance metric and a trait reactance metric. The results show that the state reactance part of the RSHCI correlates highly with the state reactance measurement of the established method. Therefore, inner criterion validity can be assumed. Also, the results regarding freedom threat indicate that the freedom threat dimension of the RSHCI really measures the level of perceived freedom threat.

Part IV of this work investigates factors which influence state reactance and the effect that state reactance has on interaction. The first study investigated state reactance as a consequence of system errors. System errors were the only item on the list of identified situations from the expert survey, described in Chap. 5, that has not been investigated in the context of human–computer interaction in the literature before. The study measured state reactance and acceptability of a smart television set (TV) that could be controlled via voice commands. The smart TV was able to display several steps of the processing chain of a command via different feedback types. Conditions were manipulated in a way, that artificial errors were introduced. In one error condition, the error could be identified as an interpretation error. In another error condition, the interpretation was correct and participants could not explain why the error had occurred. Results show that participants experience significantly higher levels of state reactance when the smart TV produced errors. However, when the participants could understand why the error had occurred, their state reactance levels were lower, compared to when they could not. This trend is also visible in the ratings of acceptability, which highly correlated with the state reactance ratings.

The persuasive assistant study presented 72 participants with a persuasive to-do list application that was controlled via a spoken dialogue assistant. The investigated moderator variables were trait reactance, social agency, and involvement. Results show that trait reactance only moderated the effects of negative cognitions, while social agency interacts with both components of state reactance. No moderating effect of involvement could be shown. Furthermore, the study showed that state reactance can improve the prediction accuracy of the subjects’ global opinion about the technical device.

This chapter provides a discussion on the results and weaknesses of the Smart TV Study and the Persuasive Assistant Study. It is discussed if the postulated Hypotheses are supported by the observed data or not.

Multiple Research Questions have been formulated to address different aspects of the role of psychological reactance for Human–Computer interaction. This Chapter discusses the Research Questions, the findings of the current book, as well as methodological considerations and limitations of the research that was done.

This chapter concludes the current book. It can be concluded, that psychological reactance has a two-fold role in human–computer interaction. As a motivational state, reactance can have negative implications for the user’s judgment of a technical system. As a user trait, reactance could not be shown to have a direct effect on the user’s judgment. However, high trait reactance might result is a higher proneness of perceiving a threat to autonomy when a proactive or adaptive technical system. The chapter closes with an outlook on potential future research on psychological reactance.