Conference PaperPDF Available

‘Let us work Together’ – Insights from an Experiment with Conversational Agents on the Relation of Anthropomorphic Design, Dialog Support, and Performance


Abstract and Figures

In the human interaction with CAs, research has shown that elements of persuasive system design, such as praise, are perceived differently when compared to traditional graphical interfaces. In this experimental study, we will extend our knowledge regarding the relation of persuasiveness (namely dialog support), anthropomorphically designed CAs, and task performance. Within a three-conditions-between-subjects design, two instances of the CA are applied within an online experiment with 120 participants. Our results show that anthropomorphically designed CAs increase perceived dialog support and performance but adding persuasive design elements can be counterproductive. Thus, the results are embedded in the discourse of CA design for task support.
Content may be subject to copyright.
Let us work Together’ – Insights from an Experiment with Conversational Agents on the Relation of
Anthropomorphic Design, Dialog Support, and Performance
This is the author’s version of a work that was published in the following source:
Chair of Business Informatics, esp.
Intelligent Systems and Services
Prof. Dr. Alfred Benedikt Brendel
Helmholtzstraße 10
01069 Dresden
Digital Work Research Group
Sascha Lichtenberg, M. Sc.
Helmholtzstraße 10
01069 Dresden
Please note: The copyright is owned by the author and / or the
publisher. Commercial use is not allowed.
This work is licensed under a Creative Commons Attribution-NonCommercial-
NoDerivatives 4.0 International License.
Lichtenberg, S.; Bührke, J.; Brendel, A.B.; Trang, S.; Diederich, S.; Morana, S (2021): Let
us work Together’ – Insights from an Experiment with Conversational Agents on the
Relation of Anthropomorphic Design, Dialog Support, and Performance, Proceedings
of the 16th International Conference on Wirtschaftsinformatik.
16th International Conference on Wirtschaftsinformatik,
March 2021, Essen, Germany
‘Let us work Together’ – Insights from an Experiment
with Conversational Agents on the Relation of
Anthropomorphic Design, Dialog Support, and
Sascha Lichtenberg2, Johannes Bührke1, Alfred Benedikt Brendel2, Simon Trang1,
Stephan Diederich1, Stefan Morana3
1 Georg-August-Universität Göttingen, Chair of Information Management, Goettingen,
{Strang} {johannes.buehrke,
2 Technische Universität Dresden, Chair of Business Informatics, esp. Intelligent Systems and
Services, Dresden, Germany
{ Sascha.Lichtenberg, Alfred_Benedikt.Brendel}
3 Universität des Saarlandes, Chair of Digital Transformation and Information System,
Saarbrücken, Germany
Abstract. In the human interaction with CAs, research has shown that elements
of persuasive system design, such as praise, are perceived differently when
compared to traditional graphical interfaces.
In this experimental study, we will extend our knowledge regarding the relation
of persuasiveness (namely dialog support), anthropomorphically designed CAs,
and task performance. Within a three-conditions-between-subjects design, two
instances of the CA are applied within an online experiment with 120
participants. Our results show that anthropomorphically designed CAs increase
perceived dialog support and performance but adding persuasive design elements
can be counterproductive. Thus, the results are embedded in the discourse of CA
design for task support.
Keywords: Conversational Agents, Persuasive System Design, Task
Performance, Dialog Support, Chatbot, human computer interaction
1 Introduction
Information Systems (IS) can be designed to attain various goals. Following Benbasat
[1], one of the goals is to increase the effectiveness and efficiency of users in the
completion of a task, such as finding and purchasing a product online. However, IS also
exhibits substantial potential to influence individual beliefs and behavior [2], for
instance, regarding environmental sustainability [3] or health [4]. Studies in the context
of persuasive systems and their design have received increasing attention recently,
which is reflected in calls for more research [5].
16th International Conference on Wirtschaftsinformatik,
March 2021, Essen, Germany
While the vast majority of studies in the area of persuasive system design focuses on
software with graphical user interfaces [6], we follow the notion that conversational
agents (CAs) offer the opportunity to design even more persuasive IS. CAs, defined as
software with which users interact through natural language (i.e. written or spoken
word) [7], have been shown to trigger mindless social responses (i.e. users treat
computers like it is a human being [8]) as formulated in the paradigm of computers-
are-social-actors (CASA) [8], [9]. Due to the social nature of human interaction with
CAs, we argue that elements of persuasive system design, such as praise or social roles
[10], can be leveraged to influence individual behavior.
Initial work in the area of persuasive and anthropomorphic CAs underlines this
potential. For example, Diederich, Lichtenberg, et al. [11] investigated how persuasive
messages of a CA can influence an individual’s environmental sustainability beliefs,
finding that a anthropomorphic design of a CA increase the perceived persuasiveness.
Similarly, Gnewuch et al. [12] argue that CAs can be a useful means to enable more
sustainable energy consumption behavior of consumers, due to their feedback provided
for the user. However, we still lack an understanding of whether persuasive CAs can
extend beyond the scope of emotion and cognition, influencing actual user behavior
(e.g., task performance).
In this experimental study, we address this research gap regarding the relation of
persuasive, anthropomorphic CAs, and actual behavior in the form of performance. The
performance of an individual can be measured by the number of completed tasks (e.g.,
in the context of gamification, by completed rounds [13], or the number of steps per
day [14]). We conducted an experiment with three different treatment groups (no CA,
anthropomorphic CA and anthropomorphic CA extended with persuasive features) in a
task completion setting. Specifically, participants had to complete a certain number of
tasks, with the option to voluntarily complete more of them. Against this background,
this study aims to answer the following research question:
RQ: How can persuasive and anthropomorphic design of conversational agents
positively influence performance?
2 Research Background
The following section contains the relevant background information for understanding
this work: (1) persuasive system design and performance and (2) anthropomorphic
conversational agents and social response theory.
2.1 Persuasive System Design and Performance
The observation that technology can influence human cognition, emotion, and behavior
has been made around two decades ago. On this basis, the paradigm of CASA [9], [11],
[16] has been formulated. The paradigm of CASA posits that individuals mindlessly
apply social rules and expectations to computers once they receive cues associated with
human traits or behavior [17]. Against this background, research in the domain of
persuasive design investigates the social responses people show to computers [9], [10].
16th International Conference on Wirtschaftsinformatik,
March 2021, Essen, Germany
Research in this context entails the development and application of design elements
intended to shape user perception and promote desired behavior. An example of this is
the display of anthropomorphic communication features, such as humor, empathy, and
praise, to trigger social dynamics, such as competition or cooperation [10].
These persuasive design elements can be distinguished into five types of social cues
Fogg [2]: physical (e.g., touch, facial expressions, movement), psychological (e.g.,
empathy, humor), language (e.g., written or spoken language, turn-taking in a
conversation), social dynamics (e.g., praise, judgment), and social roles (e.g., guide,
partner). In sum, designers are provided with a wide selection of design elements and
cues that can be used to persuade individuals in a variety of application domains, such
as environmental sustainability, work, or education [18]. Regarding the effects of these
social cues, four different categories can be distinguished [10]: (1) primary task support
(e.g., individual tailoring of information), (2) dialog support (e.g., providing praise), (3)
credibility support (e.g., displaying trustworthiness), and (4) social support (e.g.,
referring to social facilitation).
In the domain of work and performance, persuasive design offers the opportunity to
incline individuals to perform their primary task [10]. In the context of performance,
for instance, this can mean enabling an individual to measure their primary task
progress via self-monitoring [6] (e.g., displaying heart rate while exercising to ensure
progress and commitment [19]). Similar examples can be found in the context of the
academic performance of students [20], promoting physical activity at the workplace
[4] and provoke “work-like” performance in experimental contexts [21], [22]. Dialog
support has shown that users are encouraged to use the enhanced IS and consecutively
motivated to perform their primary task [23]. One example is praise in the form of
images, symbols, or words [6] to support a person in achieving his or her goals (e.g.,
increase the number of steps per day [14]).
2.2 Anthropomorphic Conversational Agents and Social Response Theory
Through technological progress regarding natural language processing and machine
learning, CA-related technology has become widely available [24]. Consequently, CAs
are currently attracting strong interest from research and practice [7], [24], [25] . Users
can interact with CAs using written (e.g., chatbots) or spoken language (e.g., personal
assistants like Siri or Microsoft Cortana). Furthermore, CAs can be disembodied, have
a virtual embodiment [26], or a physical embodiment, e.g. service robots [27]. Through
various means, CAs can display human characteristics, such as having a human name
and participating in a dialogue with turn-taking [28]. These anthropomorphic
characteristics trigger mindless social responses by users [28], [29], as postulated in the
social response theory [17], [30].
The intensity of these social responses varies according to the degree of perceived
anthropomorphism (i.e., human-likeness) of a CA [31]. Current studies on CA design
found that a higher degree of anthropomorphism can lead to various positive effects,
such as an increase in service satisfaction [32], trustworthiness [33], and persuasiveness
[11]. In order to better understand the relation of anthropomorphic CA design,
perceived anthropomorphism, and related benefits, CAs are studied in various
16th International Conference on Wirtschaftsinformatik,
March 2021, Essen, Germany
application areas, such as customer service (e.g., marketing and sales [34]), and
healthcare [35]). Synthesizing current research on anthropomorphic CA design, Seeger
et al. [15] developed a conceptual framework that comprises three dimensions: (1)
human identity, (2) verbal cues, and (3) non-verbal cues. The dimension of human
identity includes cues regarding the representation of the agent, for example, having an
avatar [31]. The second dimension of verbal cues comprises the language used by a
CA, for instance, using self-references (“I think that…” [36]), expressing artificial
thoughts and emotions (“In my experience…” [37]), or variability in syntax and word
choice [15]. The third dimension of non-verbal cues includes conveying information
on attitudes or emotional state [38], such as indicating thinking through dynamic
response times depending on message length and complexity [32] or using emoticons
to express emotions [39].
3 Research Model and Hypotheses
Our research will contribute to a better understanding of the relation between CA
design, its perception, and user performance. Our research model is depicted in Figure
1. Specifically, we hypothesize that CAs equipped with social cues as part of an
anthropomorphic design [15] persuade users to complete a higher number of tasks when
combined with persuasive design elements, such as dialog support [40].
Figure 1. Research Model
Based on the paradigm of CASA [17], [30], technology influences individual beliefs
and behavior [2]. CAs equipped with anthropomorphic characteristics, such as a human
name and participating in a dialogue with turn-taking [28], trigger social responses by
users [28], [29]. The human appearance leads individuals to perceive the CA as more
persuasive, giving it the potential to influence the beliefs and behavior of individuals.
Specifically, CAs provide users with the option to interact with the system via written
dialog, providing dialog support [23]. Thus, we formulate the following hypothesis:
H1a: An anthropomorphically designed chatbot yields a higher level of perceived
dialog support than no chatbot.
In the context of this study, we focus on CAs that are praising the user for their
performance and award points for certain achievements, thereby providing dialog
support [23]. Kamali et al. [41] were able to show that praise was expected (i.e., for
specific behavior) when elderly people interact with a CA. Similarly, receiving points
16th International Conference on Wirtschaftsinformatik,
March 2021, Essen, Germany
for certain behavior increases participation [42]. Therefore, we formulate our next
hypothesis as follows:
H1b: A persuasively and anthropomorphically designed chatbot yields a higher level
of perceived dialog support than an anthropomorphically designed chatbot.
Furthermore, CAs offer various possibilities for anthropomorphic design. An agent
equipped with a name, gender, and avatar [31], displaying emotions through verbal cues
[8], and applying nonverbal cues, such as dynamic response delays to indicate thinking
or typing [32], can contribute to the perception of the agent as more anthropomorphic,
even when users are aware of the artificial nature of it. Thus, we propose the following
H2a: An anthropomorphically designed chatbot yields a higher level of perceived
anthropomorphism than no chatbot.
Furthermore, CAs additionally displaying persuasive cues, such as praising their
user, add further to the anthropomorphic perception [10]. For instance, the study of Xu
and Lombard [43] have shown that even a small cue (e.g., the name of the CA) can
change the perception of the CA. Therefore, we hypothesize that such cues contribute
to users anthropomorphizing the agent:
H2b: A persuasively and anthropomorphically designed chatbot yields a higher level
of perceived anthropomorphism than an anthropomorphically designed chatbot.
Recent studies, which explore the interaction of anthropomorphic design of CAs and
their persuasiveness, suggest that perceived anthropomorphism can increase the
persuasiveness of the agent. For instance, Harjunen et al. [44] found that virtual offers
are more likely to be accepted when the agent shows typical human behavior, such as
smiling or touching (with a haptic glove). Similarly, Adler et al. [45] showed that a CA
displaying positive emotions leads to a higher degree of perceived persuasiveness
compared to a CA without emotionally loaded language. Against this background, we
H3: Perceived anthropomorphism positively impacts perceived dialog support.
Following Lehto et al. [23], persuasive design elements have the potential to
reinforce, change, or shape the behavior of individuals by increasing the overall
persuasiveness of information systems. Superficially, dialog support has shown to
encourage users to perform their primary task, such as increasing the amount of
physical exercise [14]. Thus, we propose the following hypothesis:
H4: Perceived dialog support positively impacts performance.
4 Research Design
To test our hypotheses, we conducted an online experiment with three-conditions (no
design, anthropomorphic design, and persuasive design) in a between-subjects design,
avoiding carryover effects [46]. We conducted an a priori power analysis using GPower
[47] to estimate the required sample size. We assume a large effect and estimated a
minimum amount of 102 participants, given an effect size f = 0.4, alpha = .05 and power
(beta) = 0.95). We collected data from the 2nd to the 15th of October 2019 until we had
at least forty observations per treatment, resulting in a total of 120 participants. Overall,
16th International Conference on Wirtschaftsinformatik,
March 2021, Essen, Germany
the sample consisted of 37% of females (5% of the participants preferred not to specify
their gender). The age of the participants ranges from 18 to 83 (mean 33), and all
participants are currently residing in Germany.
4.1 Data Collection Procedure and Sample
The experiment consisted of four steps: (1) Explanation of the experiment, (2) chat with
the chatbot, (3) perform the task, and (4) fill out the questionnaire. In the first step, the
participants received a briefing screen, which explained the context [48] and the
structure of the experiment (completing five of 15 slider tasks with a subsequent
survey) and described the tasks. Every participant received the same explanations to
make sure that all participants have the same information [49]. Following the
instructions, participants got two attempts to answer three comprehension questions.
Those who failed both attempts were excluded from the experiment. This procedure
ensures that no participant completed more tasks because the rules related to the number
of completed tasks were not understood properly. After this step, all participants were
randomly assigned to one of the three treatments and proceeded to step 2. The second
step is divided into two sub-steps: (2a) chat with chatbot and (2b) perform the task. In
step 2a, the participants had to chat with a chatbot. Via the chatbot, participants were
able to start a task and end the experiment (see Control and Treatment Configuration
section for details). If the participant was not in a chatbot treatment, the start of a task,
and the end of the experiment could be triggered by a button. In step 2b, users had to
perform slider tasks [48]. For the slider task, the participants had to set five sliders from
0 to 50 by using the mouse pointer. After completing each task, the participants returned
to step 2a. When five tasks were completed, participants had the option to proceed to
the questionnaire or complete up to ten more tasks. In step (3), participants had to fill
out a questionnaire (see Measures section for details).
Figure 2. Procedure of the Experiment
4.2 Control and Treatment Configurations
Our experiment had three conditions: (1) no chatbot (control treatment), (2)
anthropomorphic chatbot, and (3) persuasive chatbot. Every participant was randomly
16th International Conference on Wirtschaftsinformatik,
March 2021, Essen, Germany
assigned to one experimental condition (between-subjects design). For condition (1),
users did not have the option to communicate with a chatbot. For conditions (2) and
(3), two chatbots were developed via the natural language processing platform
Dialogflow by Google. Both chatbots received the same training phrases (i.e.,
exemplary statements that users might make during the interaction) to train them to
understand a user’s intent and provide the correct reply. The chatbots were able to
process different variations of sentences with the same meaning and could extract
parameters, such as the intention to proceed to the next task or to exit the experiment
and react appropriately. We further implemented a custom-built web interface to
provide convenient access to the chatbots, ensure device independence, and minimize
Figure 3. Slider Task
Figure 4. Persuasive Chatbot
16th International Conference on Wirtschaftsinformatik,
March 2021, Essen, Germany
Both chatbots were equipped with various cues for anthropomorphic CA design
according to the three dimensions (human identity, verbal, non-verbal) as suggested by
Seeger et al. [19] to establish a baseline for perceived anthropomorphism. Regarding
the human identity, we equipped the chatbot with the name “Laura,” a female gender,
and a human pictogram representing a female individual. Concerning verbal
communication, the CA was designed to use self-references, turn-taking, and a personal
introduction (“Hi! I am Laura and I will…”), including a greeting in the form of a
welcome message. Regarding the non-verbal anthropomorphic CA design dimension,
we implemented blinking dots in combination with dynamic response delays depending
on the length of the previous message to simulate thinking and typing of replies by the
CAs [32].
Overall, both chatbot instances were identical except for the addition of persuasive
messages for condition (3). The chatbot provides dialog support by using praise,
suggestions, and rewards [10]. The persuasive chatbot praises users after every task
completed (“Wow! You finished your task very quickly.”), whereas the
anthropomorphic chatbot renounces praise. Furthermore, in case users want to end the
experiment and proceed to the questionnaire, the chatbots suggests continuing and
completing more tasks (“Maybe you can hold on a little longer? Would you like to
continue?”). Lastly, the chatbot introduces a point system, rewarding the user with one
point for every completed task (“You now have a total of X points”).
4.3 Measures and Descriptive Statistics
Our research variables included experimentally manipulated variables, questionnaire-
based variables (i.e., dialogue support and control variables), and the task outcome
Table 1. Questionnaire Items (Note that the items are translated from German to English.)
Constructs and Items
Perceived Dialogue Support ( = .911)
I believe that the tool has supported me with appropriate feedback.
I believe that the tool has encouraged me to continue working on the task.
I believe that the tool motivated me to complete the task by praise.
Perceived Anthropomorphism( = .934)
I believe that the tool has a mind.
I believe that the tool has a purpose.
I believe that the tool has free will.
I believe that the tool has a consciousness.
I believe that the tool desires something.
I believe that the tool has beliefs.
I believe that the tool has the ability to experience emotions.
Perceived Persuasiveness (Single Scale)
I believe that the tool convinced me to perform the task.
FL = factor loadings, REF = reference,
= Cronbach’s alpha;
First, the effect of the experimentally manipulated variables for the different types
of chatbots. As the three treatments build on one another, we detangled the different
16th International Conference on Wirtschaftsinformatik,
March 2021, Essen, Germany
effects and coded variables that capture commonalities and differences between the
treatments. Second, dialog support, anthropomorphism, and control variables in terms
of age, gender, education, and experience with chatbots were captured using a
questionnaire. All items were measured on a scale from 1 (strongly disagree) to 7
(strongly agree). For the design of the survey, only established constructs from previous
studies were considered. Additionally, we included attention checks by asking two
questions that prompt the participant to select a specific number on a scale. If the
participant failed to answer the questions correctly, the data was not considered for the
analysis. Perceived dialog support was measured using a 7-Point Likert scale adapted
from [23]. Perceived anthropomorphism is based on a 7-Point Likert scale adapted from
[15]. Additionally, we measured perceived persuasiveness [23] as a single-scale item
to conduct a manipulation check. The items are displayed in Table 1. Third, the outcome
variable of the task was measured in terms of the number of completed tasks, where the
number of completed tasks equals the times a participant positioned all sliders correctly.
5 Results
In the following two sub-sections, we will present our results regarding the descriptive
statistics and structural model.
5.1 Descriptive Statistics
The group averages of the performance show that the anthropomorphic chatbot
(M=7.375, SD=5.309) and anthropomorphic chatbot with persuasive elements (M=4.3,
SD=2.893) differ from the control group, which yields a lower number of tasks
performed (M=3.150, SD=3.519). Similarly, we observed that the perceived dialog
support is lower for the control group (M=2.45, SD=1.693) when compared to the
anthropomorphic chatbot (M=5.15, SD=1.743) and anthropomorphic chatbot with
persuasive elements (M=2.858, SD=1.571). As for anthropomorphism, the system is
perceived lower in the control group (M=2.107, SD=1.318) when compared to the
treatments anthropomorphic chatbot (M=3.279, SD=1.734) and anthropomorphic
chatbot with persuasive elements (M=2.504, SD=1.045) (see Table 2).
To test whether our manipulation of the interface designs for the three different
treatments was successful, we assessed users’ perceived social persuasiveness. A test
for variances homogeneity was not significant (F(2, 117) = 13.467; p = .597). Based on
this result, we conducted a one-way ANOVA. The ANOVA was significant with
F(2,117) 13.467; p < .001. The result of a Tuskey HSD post hoc comparison revealed
following significant differences between for control (M=2.7; SD=1.951)
anthropomorphic chatbot (M=4.88; SD=1.977) (p < .001), and anthropomorphic
chatbot - anthropomorphic chatbot with persuasive elements (M=3.08; SD=1.789) (p <
.001). We applied PLS (partial least squares) to evaluate the measurement model and
estimate the structural model. As our analysis includes dialog support as a latent
variable, we applied a structural equation approach. We used partial least squares (PLS)
path modeling and employed SmartPLS 3.2.9. In the following paragraph, we first
16th International Conference on Wirtschaftsinformatik,
March 2021, Essen, Germany
inspect the measurement models and will then estimate and interpret the structural
Table 2. Descriptive Statistics
Dependent Variables
(N = 40 for all treatments)
Perceived Dialogue Support
Perceived Anthropomorphism
Perceived Persuasiveness
(Manipulation Check)
SD = Standard deviation, AC = Anthropomorphic, ACwPE = AC with Persuasive Elements
5.2 Measurement Model and Structural Model
The measurement model includes manifest variables in terms of the experimentally
manipulated variables, the number of completed tasks, and reflective constructs. From
the experimental treatments, we derived four variables (see Table 3). The no chatbot
variable (control treatment) was not included (reference group).
Table 3. Inter-Construct Correlations, CR, and AVE
(Latent) Variable
1. Number of Completed Tasks
2. Dialogue Support
3. Anthropomorphism
4. Anthropomorphic Chatbot Design
5. Persuasive Chatbot Design
CR = composite reliability, AVE = average variance extracted
Figure 5. PLS Structural Model ***p ≤ .001, **p ≤ .01, *p ≤ .05
We then assessed the reflective measurement model of anthropomorphism and
dialogue support for individual item reliability, convergent validity, and discriminant
16th International Conference on Wirtschaftsinformatik,
March 2021, Essen, Germany
validity. The model displays good measurement properties: all factor loadings are
meaningful and significant, the composite reliability is above .7, the average variance
extracted is above .5, and the FornellLarker criterion is satisfied. We then applied a
bootstrap resampling procedure (with 4999 samples) to test the relationships. We favor
the SEM for our research design with latent variables because it takes into account
measurement errors or multidimensional structures of theoretical constructs [50] . The
PLS estimator has advantages with respect to restrictive assumptions and is therefore
widely used in experimental research [51], [52]. The different experimental conditions
(no chatbot, anthropomorphically designed chatbot, persuasively and
anthropomorphically designed chatbot) were dummy coded for our structural model, to
compare the manipulations with a baseline condition (no chatbot). The structural model
explains variances in Anthropomorphism (R² = .213, f² = .156), Dialog Support (R² =
.503, f² = .312) and Performance (measured as number of completed tasks) (R² = .291).
The results of the PLS estimation are illustrated in Figure 5.
Table 4. Results for Hypotheses
a) An anthropomorphically designed chatbot yields a higher
level of perceived dialog support than no chatbot.
b) A persuasively and anthropomorphically designed chatbot
yields a higher level of perceived dialog support than an
anthropomorphically designed chatbot.
a) An anthropomorphically designed chatbot yields a higher
level of perceived anthropomorphism than no chatbot.
b) A persuasively and anthropomorphically designed chatbot
yields a higher level of perceived anthropomorphism than an
anthropomorphically designed chatbot.
Perceived anthropomorphism positively impacts perceived
dialog support.
Perceived dialog support positively impacts the number of
completed tasks.
s = supported, c = contradicted, ns = non-supported, B = path coefficient
In summary, we find support for hypotheses H1a, H2a, H3, and H4. We find
contradicting results for H1b and H2b, namely the role of the persuasive design (see
Table 4). Concerning, our control variables, we find significant effects for prior
experience with chatbots on Anthropomorphism (β = -.239, p < .05). Moreover, we find
a significant effect on Gender on Number of Completed Tasks (β = -.181, p < .05), with
male participants completing fewer tasks.
6 Discussion
Our experiment aimed to explore the relationship between the persuasive and
anthropomorphic design of conversational agents and performance. The results have
implications for CA and persuasive system design. In this regard, we provide empirical
16th International Conference on Wirtschaftsinformatik,
March 2021, Essen, Germany
evidence for the relation of anthropomorphism and persuasive design in CAs. We found
contradicting evidence for our hypotheses that persuasive cues (explicitly praise,
suggestion, and rewards) lead to higher perceived anthropomorphism and dialogue
support. These results can be explained from different perspectives.
6.1 Implications for Research
First, when looking at CA literature, Seeger et al. [15] state that simply applying
more social cues and anthropomorphic design elements will not automatically lead to a
higher level of perceived anthropomorphism. Selecting and combining them should be
done with caution. In this context, Clark et al. [53] see the expectations of a user as
decisive. Users are experienced with the interaction with humans and know the
mistakes they make in an interaction. However, computers make errors that can rarely
be found in humans. Therefore, these errors are unexpected. Regarding our CA design,
the anthropomorphic chatbot was well perceived, leading to higher perceived
anthropomorphism and dialog support. However, by adding the intended-to-be-
persuasive elements to the design, the perception of the chatbot is vastly different from
the other one. This observation indicates that users did not expect the added social cues.
Second, it could also be hypothesized that the persuasive chatbot appears to be
disingenuous. A slider task does not require specific skills, qualifications, or knowledge
[48]. Furthermore, unlike tasks in crowdsourcing, such as labeling pictures, performing
a slider task has no trigger for enjoyment (task enjoyment leading to increased
performance [54]), has no deeper meaning (perceived meaning is linked with
satisfaction and performance [55]), and does not enable a user to contribute to a greater
good (like voluntary work where the reward is intrinsic to the act of volunteering [56]).
Hence, we would suggest that individuals perceive the high level of praise, combined
with suggestions to keep going and receiving arbitrary point rewards, as disingenuous
and not fitting the task.
Lastly, the negative perception of the persuasive chatbot might be explained by the
cognitive fit theory [57]. The theory proposes that the fit between task and presentation
of supporting information shapes the task performance. Our results indicate that an
anthropomorphic CA provides a better information presentation in terms of dialog
support, fitting the task at hand. This fit leads to higher performance. Thus, through the
lens of the cognitive fit theory, the addition of persuasive elements appears to reduce
the fit between task and task support.
In summary, our results can be embedded in the current discourse of CA design for
task support. However, the significant negative change in the CA’s perception by
adding persuasive elements was unexpected. Thus, our results highlight a research
opportunity to investigate the design of CAs for task support. Specifically, the framing
and nature of a task appear to interact with the perception of a CA. CAs should meet
expectations, appear genuine, and be adapted to the nature of the task. However,
understanding how to design such a CA has yet to be addressed.
16th International Conference on Wirtschaftsinformatik,
March 2021, Essen, Germany
6.2 Implications for Practice
For practice, our result indicates that using a CA to frame and support tasks can be
beneficial. To be specific, we would relate our results to the context of crowdworking.
In crowdworking, crowd workers perform multiple tasks [58], which fits the
experimental setup of this study. Our participants were inclined to complete more tasks
than necessary. This indicates that adding the option to perform more tasks,
accompanied by an anthropomorphic CA, can lead crowd workers to do more tasks.
Furthermore, our study provides a blueprint regarding the design of such an
anthropomorphic CA. Specifically, we would advise against adding persuasive
messages or other design elements to an anthropomorphic CA that is intended to
provide dialog support. Therefore, our results can be used to better design chatbots in
the context of (crowdworking) tasks.
6.3 Limitations and Future Research
Our study is not free of limitations and offers different opportunities for future research.
We conducted the online experiment in a rather controlled setting, with a set of specific
tasks that every participant was asked to complete, and a single interaction with the
conversational agent. Moreover, we did not compare the provided CA’s with a CA
without any social cues. Thus, we benefitted from control yet lacked realism in our
research design [49]. Similarly, our results are limited by the selection and
reimbursement of participants. In a real-world work environment, individuals are under
the constant influence of expecting and receiving payment for work. For instance,
crowd workers primarily perform tasks to be paid [58]. In our setting, participants did
not receive a comparable form of pay. They were allowed to participate in a raffle for
10€ online shopping vouchers. Thus, it is safe to assume that participants were
motivated by other factors, such as curiosity or escaping boredom.
7 Conclusion
In this study, we set out to explore the relation of persuasive and anthropomorphic CA
design and performance (measured as the number of completed tasks). By means of a
three-condition online experiment with two chatbots and 120 participants, we find
empirical evidence for the positive influence an anthropomorphic CA has on an
individual’s perceived dialog support, mediated by the perceived anthropomorphism.
However, a CA that displays the same anthropomorphic features and additionally
provides persuasive messages, intended to provide further dialog support, is negatively
perceived. This observation supports the proposition of Seeger et al. [15] that merely
adding social cues and anthropomorphic characteristics to a CA is not always
beneficial. In this context, our results indicate that a chatbot that provides dialog support
(in our case praise, suggestions, and rewards) for simple tasks appears to be
disingenuous. Therefore, our results indicate a potential for future research regarding
the interaction of task and persuasive CA design. Our study makes three main
contributions: First, we empirically demonstrate how the application of
16th International Conference on Wirtschaftsinformatik,
March 2021, Essen, Germany
anthropomorphic characteristics and persuasive messages can influence performance.
Thereby, we add to the body of knowledge regarding the perception and influence
anthropomorphic IS has on users. Second, we present CAs as a new type of persuasive
IS that triggers social responses by users and offers new opportunities for interface and
task design. Third, we bridge the gap between knowledge on persuasions and
anthropomorphism of IS and the design of CA for dialog support.
8 Acknowledgements
We would like to thank Jonas Gehrke and Jessica Lühnen for their support during this
research project.
9 References
[1] I. Benbasat, “HCI Research: Future Challenges and Directions,” AIS Trans.
Human-Computer Interact., vol. 2, no. 2, pp. 1621, 2010.
[2] B. J. Fogg, “Computers as persuasive social actors,” in Persuasive Technology,
San Francisco, USA: Morgan Kaufmann Publishers, 2003, pp. 89120.
[3] C.-M. Loock, T. Staake, and F. Thiesse, “Motivating Energy-Efficient
Behavior with Green IS: An Investigation of Goal Setting and the Role of
Defaults,” Manag. Inf. Syst. Q., vol. 37, no. 4, pp. 13131332, 2013.
[4] M. S. Haque, M. Isomursu, M. Kangas, and T. Jämsä, “Measuring the influence
of a persuasive application to promote physical activity,” CEUR Workshop
Proc., vol. 2089, pp. 4357, 2018.
[5] P. Slattery, R. Vidgen, and P. Finnegan, “Persuasion: An analysis and common
frame of reference for is research,” Commun. Assoc. Inf. Syst., vol. 46, pp. 30
69, 2020.
[6] H. Oinas-Kukkonen and M. Harjumaa, “A Systematic Framework for
Designing and Evaluating Persuasive Systems,” LNCS, vol. 5033, pp. 164176,
[7] M. McTear, Z. Callejas, and D. Griol, The Conversational Interface: Talking
to Smart Devices. Basel, Switzerland: Springer Publishing Company, 2016.
[8] N. Wang, W. L. Johnson, R. E. Mayer, P. Rizzo, E. Shaw, and H. Collins, “The
politeness effect: Pedagogical agents and learning outcomes,” Int. J. Hum.
Comput. Stud., vol. 66, no. 2, pp. 98112, 2008.
[9] B. Reeves and C. Nass, The Media Equation: How People Treat Computers,
Television and New Media Like Real People and Places. The Center for the
Study of Language and Information Publications, 1996.
[10] H. Oinas-Kukkonen and M. Harjumaa, “Persuasive Systems Design: Key
Issues, Process Model, and System Features,” Commun. Assoc. Inf. Syst., vol.
24, no. 1, p. 96, 2009.
[11] S. Diederich, S. Lichtenberg, A. B. Brendel, and S. Trang, “Promoting
sustainable mobility beliefs with persuasive and anthropomorphic design:
Insights from an experiment with a conversational agent,” Nov. 2019.
16th International Conference on Wirtschaftsinformatik,
March 2021, Essen, Germany
[12] U. Gnewuch, S. Morana, C. Heckmann, and A. Maedche, “Designing
Conversational Agents for Energy Feedback,” in Lecture Notes in Computer
Science, vol. 10844 LNCS, Springer Verlag, 2018, pp. 1833.
[13] M. J. Koeder, E. Tanaka, and H. Mitomo, “‘Lootboxes’ in digital games - A
gamble with consumers in need of regulation? An evaluation based on learnings
from Japan,” 22nd Bienn. Conf. Int. Telecommun. Soc. "Beyond boundaries
Challenges business, policy Soc., 2018.
[14] T. Toscos, A. Faber, S. An, and M. P. Gandhi, “Chick Clique: Persuasive
technology to motivate teenage girls to exercise,” Conf. Hum. Factors Comput.
Syst. - Proc., pp. 18731878, 2006.
[15] A. M. Seeger, J. Pfeiffer, and A. Heinzl, “Designing anthropomorphic
conversational agents: Development and empirical evaluation of a design
framework,” in ICIS, 2018, pp. 117.
[16] C. Nass, J. Steuer, and E. R. Tauber, “Computers are social actors,” in ACM
CHI, 1994, p. 204.
[17] C. Nass and Y. Moon, “Machines and mindlessness: Social responses to
computers,” J. Soc. Issues, vol. 56, no. 1, pp. 81103, 2000.
[18] S. Langrial, T. Lehto, H. Oinas-Kukkonen, M. Harjumaa, and P. Karppinen,
“Native mobile applications for personal wellbeing: A persuasive systems
design evaluation,” in PACIS, 2012, pp. 116.
[19] S. Consolvo, K. Everitt, I. Smith, and J. A. Landay, “Design requirements for
technologies that encourage physical activity,” in Conference on Human
Factors in Computing Systems - Proceedings, 2006, vol. 1, pp. 457466.
[20] J. Filippou, C. Cheong, and F. Cheong, “Modelling the impact of study
behaviours on academic performance to inform the design of a persuasive
system,” Inf. Manag., vol. 53, no. 7, pp. 892903, 2016.
[21] S. Lichtenberg and A. B. Brendel, “Arrr you a Pirate ? Towards the
Gamification Element ‘ Lootbox ,’” AMCIS (Forthcoming), 2020.
[22] S. Lichtenberg, T. Lembcke, M. Brenig, A. B. Brendel, and S. Trang, “Can
Gamification lead to Increase Paid Crowdworkers Output ?,” in 15.
Internationale Tagung Wirtschaftsinformatik, 2020, no. December 2019.
[23] T. Lehto, H. Oinas-Kukkonen, and F. Drozd, “Factors affecting perceived
persuasiveness of a behavior change support system,” in ICIS, 2012, vol. 3, pp.
[24] S. Diederich, A. B. Brendel, and L. M. Kolbe, “On Conversational Agents in
Information Systems Research: Analyzing the Past to Guide Future Work,”
Proc. Int. Conf. Wirtschaftsinformatik, pp. 15501564, 2019.
[25] Oracle, “Can Virtual Experiences Replace Reality? The future role for humans
in delivering customer experience,” p. 19, 2016.
[26] N. V. Wünderlich and S. Paluch, “A Nice and Friendly Chat With a Bot: User
Perceptions of AI-based Service Agents,” in ICIS, 2017, no. 1, pp. 111.
[27] R. M. Stock and M. Merkle, “Can Humanoid Service Robots Perform Better
Than Service Employees? A Comparison of Innovative Behavior Cues,” 2018.
[28] J. Feine, U. Gnewuch, S. Morana, and A. Maedche, “A Taxonomy of Social
Cues for Conversational Agents,” Int. J. Hum. Comput. Stud., vol. 132, pp.
16th International Conference on Wirtschaftsinformatik,
March 2021, Essen, Germany
138161, Dec. 2019.
[29] T. Verhagen, J. van Nes, F. Feldberg, and W. van Dolen, “Virtual customer
service agents: Using social presence and personalization to shape online
service encounters,” J. Comput. Commun., vol. 19, no. 3, pp. 529545, 2014.
[30] B. J. Fogg and C. Nass, “How users reciprocate to computers,” in ACM CHI,
1997, no. March, p. 331.
[31] L. Gong, “How social is social responses to computers? The function of the
degree of anthropomorphism in computer representations,” Comput. Human
Behav., vol. 24, no. 4, pp. 14941509, 2008.
[32] U. Gnewuch, S. Morana, M. T. P. Adam, and A. Maedche, “Faster Is Not
Always Better: Understanding the Effect of Dynamic Response Delays in
Human-Chatbot Interaction,” in ECIS, 2018, pp. 117.
[33] T. Araujo, “Living up to the chatbot hype: The influence of anthropomorphic
design cues and communicative agency framing on conversational agent and
company perceptions,” Comput. Human Behav., vol. 85, pp. 183189, 2018.
[34] M. D. Hanus and J. Fox, “Persuasive avatars: The effects of customizing a
virtual salespersons appearance on brand liking and purchase intentions,” Int.
J. Hum. Comput. Stud., vol. 84, pp. 3340, 2015.
[35] J. Sebastian and D. Richards, “Changing stigmatizing attitudes to mental health
via education and contact with embodied conversational agents,” Comput.
Human Behav., vol. 73, pp. 479488, 2017.
[36] Y. J. Sah and W. Peng, “Effects of visual and linguistic anthropomorphic cues
on social perception, self-awareness, and information disclosure in a health
website,” Comput. Human Behav., vol. 45, pp. 392401, 2015.
[37] R. M. Schuetzler, J. S. Giboney, G. M. Grimes, and J. F. Nunamaker, “The
Influence of Conversational Agents on Socially Desirable Responding,” in
HICSS, 2018, vol. 9, pp. 283292.
[38] P. Ekman and W. V. Friesen, “The Repertoire of Nonverbal Behavior:
Categories, Origins, Usage, and Coding,Semiotica, vol. 1, no. 1, pp. 4998,
[39] R. E. Mayer, W. L. Johnson, E. Shaw, and S. Sandhu, “Constructing computer-
based tutors that are socially sensitive: Politeness in educational software,” Int.
J. Hum. Comput. Stud., vol. 64, no. 1, pp. 3642, 2006.
[40] N. Shevchuk and H. Oinas-Kukkonen, “Exploring Green Information Systems
and Technologies as Persuasive Systems: A Systematic Review of Applications
in Published Research,” in ICIS, 2016, pp. 111.
[41] M. El Kamali, L. Angelini, M. Caon, G. Andreoni, O. A. Khaled, and E.
Mugellini, “Towards the Nestore e-Coach: A tangible and embodied
conversational agent for older adults,” in UbiComp/ISWC 2018, Oct. 2018, pp.
[42] J. Hamari, “Transforming homo economicus into homo ludens: A field
experiment on gamification in a utilitarian peer-to-peer trading service,”
Electron. Commer. Res. Appl., vol. 12, no. 4, pp. 236245, 2013.
[43] K. Xu and M. Lombard, “Persuasive computing: Feeling peer pressure from
multiple computer agents,” Comput. Human Behav., vol. 74, pp. 152162, Sep.
16th International Conference on Wirtschaftsinformatik,
March 2021, Essen, Germany
[44] V. J. Harjunen, M. Spapé, I. Ahmed, G. Jacucci, and N. Ravaja, “Persuaded by
the machine: The effect of virtual nonverbal cues and individual differences on
compliance in economic bargaining,” Comput. Human Behav., vol. 87, pp.
384394, Oct. 2018.
[45] R. F. Adler, F. Iacobelli, and Y. Gutstein, “Are you convinced? A Wizard of
Oz study to test emotional vs. Rational persuasion strategies in dialogues,”
Comput. Human Behav., vol. 57, pp. 7581, Apr. 2016.
[46] M. C. Boudreau, D. Gefen, and D. W. Straub, “Validation in information
systems research: A state-of-the-art assessment,” MIS Q. Manag. Inf. Syst., vol.
25, no. 1, pp. 116, 2001.
[47] E. Erdfelder, F. FAul, A. Buchner, and A. G. Lang, “Statistical power analyses
using G*Power 3.1: Tests for correlation and regression analyses,” Behav. Res.
Methods, vol. 41, no. 4, pp. 11491160, 2009.
[48] E. Lezzi, P. Fleming, and D. J. Zizzo, “Does it Matter Which Effort Task You
Use? A Comparison of Four Effort Tasks When Agents Compete for a Prize,”
SSRN Electron. J., 2015.
[49] A. R. Dennis and J. S. Valacich, “Conducting Experimental Research in
Information Systems,” Commun. Assoc. Inf. Syst., vol. 7, no. 5, pp. 141, 2001.
[50] R. P. Bagozzi and Y. Yi, “On the Use of Structural Equation Models in
Experimental Designs,” J. Mark. Res., vol. 26, no. 3, p. 271, Aug. 1989.
[51] P. W. Fombelle, S. A. Bone, and K. N. Lemon, “Responding to the 98%: face-
enhancing strategies for dealing with rejected customer ideas,” J. Acad. Mark.
Sci., vol. 44, no. 6, pp. 685706, Nov. 2016.
[52] M. Trenz, D. Veit, and C.-W. Tan, “Disentangling the Impact of Omnichannel
Integration on Consumer Behavior in Integrated Sales Channels,” Manag. Inf.
Syst. Q., vol. 44, no. 3, Sep. 2020.
[53] L. Clark et al., “What makes a good conversation? Challenges in designing
truly conversational agents,” in Conference on Human Factors in Computing
Systems - Proceedings, 2019, vol. 12.
[54] R. M. Puca and H. D. Schmalt, “Task enjoyment: A mediator between
achievement motives and performance,” Motiv. Emot., vol. 23, no. 1, pp. 15
29, 1999.
[55] A. Wrzesniewski, J. E. Dutton, and G. Debebe, “Interpersonal
Sensemaking.pdf,” Research in Organizational Behavior, vol. 25. pp. 93135,
[56] H. Bussell and D. Forbes, “Understanding the volunteer market: the what,
where, who and why of volunteering,” Int. J. Nonprofit Volunt. Sect. Mark.,
vol. 7, no. 3, pp. 244257, 2002.
[57] R. Agarwal, A. P. Sinha, and M. Tanniru, “Cognitive Fit in Requirements
Modeling: A Study of Object and Process Methodologies,” J. Manag. Inf. Syst.,
vol. 13, no. 2, pp. 137162, 1996.
[58] D. Durward, I. Blohm, and J. M. Leimeister, “The Nature of Crowd Work and
its Effects on Individuals’ Work Perception,” J. Manag. Inf. Syst., vol. 37, no.
1, pp. 6695, 2020.
ResearchGate has not been able to resolve any citations for this publication.
Conference Paper
Full-text available
Gamification, using game-like elements in a non-game context, stands a means to trigger an individual's innate disposition to game and play, leading to enhanced engagement, enjoyment, and motivation. Gamification elements can take many forms. Amongst the most prominent and often applied gamification elements are progress bars, badges, and leaderboards. Each of these elements can be found in games, for instance, computer games. One recent development in online computer games is the so-called "lootbox." Lootboxes can be described as a mechanism that rewards gamers with a random object when a specific objective is met. Lootboxes have yet to be adapted as a gamification element. It remains unclear how users react to lootboxes, both in terms of psychological and behavioral outcomes. Against this background, this study investigates the effect of lootboxes on extrinsic and intrinsic motivation, as well as the performance (quantity of completed tasks) via an online experiment with 203 participants. The results of our research indicate that lootboxes can be an effective way of gamifying non-gaming contexts, increasing extrinsic motivation and performance, while preserving intrinsic motivation.
Full-text available
Crowd work reflects a new form of gainful employment on the Internet. We study how the nature of the tasks being performed and financial compensation jointly shape work perceptions of crowd workers in order to better understand the changing modes and patterns of digital work. Surveying individuals on 23 German crowd working platforms, this work is the first to add a multi-platform perspective on perceived working conditions in crowd work. We show that crowd workers need rather high levels of financial compensation before task characteristics become relevant for shaping favorable perceptions of working conditions. We explain these results by considering financial compensation as an informational cue indicating the appreciation of working effort which is internalized by well-paid crowd workers. Resulting boundary conditions for task design are discussed. These results help us understand when and under what conditions crowd work can be regarded as a fulfilling type of employment in highly developed countries.
Conference Paper
Full-text available
Gamifying serious work environments, such as paid crowdsourcing platforms, potentially increases crowdworkers’ task motivation, engagement and enjoyment. This, in turn, can lead to a higher willingness to contribute, higher quality of work and long-term engagement. However, it remains unclear how crowdworkers behave, when gamification is applied to motivate them to do more tasks than being paid for. In this study, we conducted an experiment on Amazon Mechanical Turk to investigate this context in a controlled setting, enabling the isolation of gamification effects. With 320 crowdworkers, we study the effect of different gamification affordances (progressbars, badges and leaderboards) on autonomous motivation and tasks performed. We find that some gamification affordances (namely badges and leaderboard) can lead crowdworkers to do more work than they are paid for. However, this is not necessarily linked to autonomous motivation because we did not consistently observe an increase in autonomous motivate together with more performed tasks.
Conference Paper
Full-text available
Sustainable mobility behavior is increasingly relevant due to the vast environmental impact of current transportation systems. With the growing variety of transportation modes, individual decisions for or against specific mobility options become more and more important and salient beliefs regarding the environmental impact of different modes influence this decision process. While information systems have been recognized for their potential to shape individual beliefs and behavior, design-oriented studies that explore their impact, in particular on environmental beliefs, remain scarce. In this study, we contribute to closing this research gap by designing and evaluating a new type of artifact, a persuasive and human-like conversational agent, in a 2x2 experiment with 225 participants. Drawing on the Theory of Planned Behavior and Social Response Theory, we find empirical support for the influence of persuasive design elements on individual environmental beliefs and discover that anthropomorphic design can contribute to increasing the persuasiveness of artifacts.
Full-text available
Conversational agents (CAs) are software-based systems designed to interact with humans using natural language and have attracted considerable research interest in recent years. Following the Computers Are Social Actors paradigm, many studies have shown that humans react socially to CAs when they display social cues such as small talk, gender, age, gestures, or facial expressions. However, research on social cues for CAs is scattered across different fields, often using their specific terminology, which makes it challenging to identify, classify, and accumulate existing knowledge. To address this problem, we conducted a systematic literature review to identify an initial set of social cues of CAs from existing research. Building on classifications from interpersonal communication theory, we developed a taxonomy that classifies the identified social cues into four major categories (i.e., verbal, visual, auditory, invisible) and ten subcategories. Subsequently, we evaluated the mapping between the identified social cues and the categories using a card sorting approach in order to verify that the taxonomy is natural, simple, and parsimonious. Finally, we demonstrate the usefulness of the taxonomy by classifying a broader and more generic set of social cues of CAs from existing research and practice. Our main contribution is a comprehensive taxonomy of social cues for CAs. For researchers, the taxonomy helps to systematically classify research about social cues into one of the taxonomy's categories and corresponding subcategories. Therefore, it builds a bridge between different research fields and provides a starting point for interdisciplinary research and knowledge accumulation. For practitioners, the taxonomy provides a systematic overview of relevant categories of social cues in order to identify, implement, and test their effects in the design of a CA.
Conference Paper
Full-text available
Conversational agents (CA), i.e. software that interacts with its users through natural language, are becoming increasingly prevalent in everyday life as technological advances continue to significantly drive their capabilities. CA exhibit the potential to support and collaborate with humans in a multitude of tasks and can be used for innovation and automation across a variety of business functions, such as customer service or marketing and sales. Parallel to the increasing popularity in practice, IS researchers have engaged in studying a variety of aspects related to CA in the last few years, applying different research methods and producing different types of theories. In this paper, we review 36 studies to assess the status quo of CA research in IS, identify gaps regarding both the studied aspects as well as applied methods and theoretical approaches, and propose directions for future work in this research area.
Conference Paper
Full-text available
The ability to engage the user in a conversation and the credibility of the system are two fundamental characteristics of virtual coaches. In this paper, we present the architecture of a conversational e-coach for promoting healthy lifestyles in older age, developed in the context of the NESTORE H2020 EU project. The proposed system allows multiple access points to a conversational agent via different interaction modalities: a tangible companion that embodies the virtual coach will leverage voice and other non-verbal cues in the domestic environment, while a mobile app will integrate a text-based chat for a ubiquitous intervention. In both cases, the conversational agent will deliver personalized interventions based on behavior change models and will promote trust by means of emotionally rich conversations.
‘Brick-and-mortar’ retailers, when expanding their businesses to online channels, can either add a separate online channel or integrate channels to enhance service offerings. Although past studies on channel choice have yielded insights into factors affecting consumers’ channel preference, there is a dearth of research that sheds light on when and why massive investments into channel integration would be preferred over online optimizations. To this end, we construct and validate a theoretical model that posits omnichannel integration services for acquisition and recovery as predictors of consumers’ online channel preference through influencing their perceptions of convenience and risk. Our experimental study reveals how distinct configurations of cross-channel service offerings affect consumers’ channel evaluations and decisions, as well as how complementarities from channel integration across transaction and post-transaction phases can prevail over pure online substitutes. Consequently, this study bridges diagnostic and prescriptive research streams on multichannel and omnichannel retail by attesting to channel integration as a viable channel differentiator. From a practical standpoint, we compare twelve distinct channel configurations with regards to consumers’ core evaluative criteria and highlight the value of omnichannel integration since efficiency improvements to the online channel can only serve as a partial substitute to channel integration.
New procedures are developed and illustrated for the analysis of experimental data with particular emphasis on MANOVA and MANCOVA designs. The authors begin with one-way designs, including overall tests of significance, step-down analyses, and the use of latent variables. Next they describe a general test of homogeneity and consider a procedure that is applicable even under conditions of heterogeneity. Two-way designs then are derived as special cases of the more general n-way case. Finally, advantages and disadvantages of the new methods are considered.