Page 1 of 27

Exploring the Negotiation Behaviors of Owners and Bystanders

over Data Practices of Smart Home Devices

Ahmed Alshehri

alshehri@mines.edu

Colorado School of Mines

Golden, Colorado, USA

Eugin Pahk

epahk@mines.edu

Colorado School of Mines

Golden, Colorado, USA

Joseph Spielman

jspielman@mines.edu

Colorado School of Mines

Golden, Colorado, USA

Jacob Parker

jacobparker@mines.edu

Colorado School of Mines

Golden, Colorado, USA

Benjamin Gilbert

bgilbert@mines.edu

Colorado School of Mines

Golden, Colorado, USA

Chuan Yue

chuanyue@mines.edu

Colorado School of Mines

Golden, Colorado, USA

ABSTRACT

Bystanders (i.e., visiting friends, visiting family members, or do- mestic workers) are often not aware of the data practices in other

people’s (i.e., owners’) smart homes, exposing them to privacy risks.

One solution to avoid violating bystanders’ privacy is to increase

the data practice transparency and facilitate negotiation. In this

paper, we designed a negotiation interaction study to explore the

behaviors of owners (n1=238 participants assigned with the owner

role) and bystanders (n2=222 participants assigned with the by- stander role) when negotiating about smart home data practices

with the corresponding bystander and owner digital agents. We also

asked questions to explore factors that may potentially correlate

with or affect the observed negotiation behaviors and outcomes.

We found that owner and bystander participants differ in behaviors

regarding numbers of rounds of negotiation, final reached prefer- ences, and total number of agreements. We analyzed the correlating

factors and predictability of reaching agreements.

CCS CONCEPTS

• Security and privacy → Social aspects of security and pri- vacy; Privacy protections.

KEYWORDS

Smart home devices, data practices, privacy, negotiation behaviors.

ACM Reference Format:

Ahmed Alshehri, Eugin Pahk, Joseph Spielman, Jacob Parker, Benjamin

Gilbert, and Chuan Yue. 2023. Exploring the Negotiation Behaviors of Own- ers and Bystanders over Data Practices of Smart Home Devices. In Proceed- ings of the 2023 CHI Conference on Human Factors in Computing Systems

(CHI ’23), April 23–28, 2023, Hamburg, Germany. ACM, New York, NY, USA,

27 pages. https://doi.org/10.1145/3544548.3581360

This work is licensed under a Creative Commons Attribution International

4.0 License.

CHI ’23, April 23–28, 2023, Hamburg, Germany

© 2023 Copyright held by the owner/author(s).

ACM ISBN 978-1-4503-9421-5/23/04.

https://doi.org/10.1145/3544548.3581360

1 INTRODUCTION

Smart homes are increasing in popularity and it is estimated that

there will be more than 77 million smart homes in the United States

by 2025 [56]. Most of the previous studies such as [36, 66] focused

on the privacy concerns of smart home owners who buy and use

the devices. Recent studies have highlighted the importance of

investigating the privacy risks to bystanders, i.e., people who are

temporarily in the smart homes but do not control the devices [2,

8, 65]. For example, Mare et al. found that Airbnb hosts and guests

had different perspectives towards smart home devices, and their

privacy preferences might conflict [40].

Furthermore, researchers have recommended providing better

transparency and control for bystanders to mitigate their privacy

concerns [2, 65]. Enhancing transparency is by itself an active

area for research, and some methods or tools have already been

proposed to enhance transparency for bystanders to protect their

privacy [2, 35, 65]. However, little has been done to help bystanders

better control their data and protect their privacy. It is worth noting

that granting bystanders with more control to their data might

compromise the utility of the smart home devices. Bystander studies

that do not recognize this loss of utility of the devices might be

biased. Another group of studies focused on allowing bystanders to

control the data collection in public areas (e.g., in campuses [19, 35]).

However, expectations of privacy differ between public and private

(e.g., homes) places [20, 21].

Facilitating the negotiation of data practices between owners

and bystanders is a promising approach to protecting bystanders’

privacy. In this paper, we explore this approach with two major

research questions:

(1) RQ1: What are the negotiation behaviors of owners and

bystanders over data practices of smart home devices, and

what are the correlated factors?

(2) RQ2: Can we predict the negotiation outcomes using sit- uational factors (e.g., negotiator types, device types, and

relationship types, etc.)?

Negotiation is normally the first method to settle disagreement

between opposing parties [22]. It is more desirable than other al- ternatives such as arbitration or litigation because it allows both

parties to influence the process and the outcome. Negotiation be- haviors incorporate how owners and bystanders may perceive each

other’s benefits and losses, and how they may make their positions

known to each other in order to achieve an agreement about the

Page 2 of 27

CHI ’23, April 23–28, 2023, Hamburg, Germany Alshehri, et al.

data practices of smart home devices. For example, previous studies

show that if the first offer is perceived as fair, less rounds of negotia- tion would be needed [54]. Investigating the negotiation behaviors

of owners and bystanders can help vendors develop appropriate

negotiation features (e.g., user interfaces or functions that disclose

data practices to bystanders, collect bystanders’ preferences, and

facilitate negotiation interactions, etc.) to empower the negotiation

process for bystanders to preserve their privacy and owners to

preserve the utility of their smart home devices. Predicting agree- ments can be helpful to reduce the negotiation burden to owners

and bystanders (e.g., a small number of rounds would not take too

much time), and to explore if negotiating over smart home data

practices is worthwhile.

To answer RQ1 and RQ2, we designed a negotiation interac- tion study in which each participant acts as either an owner or

a bystander to correspondingly interact with a bystander digital

agent or an owner digital agent. We designed our study with three

between-subjects factors: the (digital) agent types (owner vs. by- stander agents), the relationship types (friends vs. strangers), and

the negotiator types (strict vs collaborative negotiators). Our within- subjects factors are the smart home device types (a smart camera

and a smart speaker) and the data practice types (data collection,

storage, and sharing practices). We also asked three sets of questions

before, during, and after the interactions to explore the factors that

may potentially correlate with or affect the observed negotiation

behaviors and outcomes.

We recruited 460 Amazon Mechanical Turk (AMT) workers as

owner (n1=238) and bystander (n2=222) participants, respectively.

Based on their behaviors and responses, we have the following

findings: (1) Owner and bystander participants took three and 1.7

rounds on average, respectively, to complete a negotiation interac- tion; owner participants took more rounds negotiating with strict

friends than strict strangers. (2) Owner and bystander participants

were more likely to reach an agreement when negotiating data

sharing practices than data collection and storage practices. (3) In

general, owner participants negotiated more aggressively than by- stander participants; owner participants took more rounds and

reached fewer agreements than bystander participants. (4) Factors

such as whether participants had negotiated over data practices,

had been to other smart homes as bystanders, and had preferred to

negotiate with other owners of smart homes over their data prac- tices were correlated with the negotiation behaviors of participants.

(5) Reaching an agreement is predictable and we explored that with

an average of 70% accuracy.

The key contributions of our work include:

(1) We explored the negotiation behaviors of owners and by- standers regarding the data collection, storage, and sharing

practices in smart homes.

(2) We designed and implemented owner and bystander dig- ital agents that can flexibly emulate humans’ negotiation

behaviors to support negotiation interaction studies; using

programmed digital agents to emulate different types of

counter-parties is a unique design in our study to the best

of our knowledge.

(3) We designed and conducted a study with 238 owners and

222 bystanders; we quantitatively and qualitatively analyzed

their negotiation behaviors and the related factors to derive

important findings.

(4) We provided recommendations for device vendors to help

owners and bystanders negotiate with each other for pre- serving both the privacy of bystanders and the utility of

owners’ smart home devices.

2 RELATED WORK

We review four main groups of related work: owners’ privacy in

smart homes, secondary users’ privacy in smart homes, bystanders’

privacy in smart homes, and negotiation over interdependent pri- vacy.

Owners’ Privacy in Smart Homes: Past research on smart

home devices has mainly targeted owners and the security and

privacy issues they tend to face [21, 34, 36, 45, 66, 68]. Surveys

and interviews have been used in these studies to analyze owners’

perceptions on security and privacy issues they may face in their

smart homes. Many owners do not realize that there are privacy

risks associated with the use of smart home devices [21, 36]. Many

owners are willing to use smart home devices even when they are

somewhat aware that their privacy is being invaded [60]. In these

studies, it was often assumed that owners are in full control of the

devices in their smart homes, ignoring the presence or potential

presence of bystanders.

Secondary Users’ Privacy in Smart Homes: Secondary users

include those who live in the same smart home (e.g., spouses, chil- dren, and roommates), and they differ from bystanders in this pa- per. Secondary users often do not have the complete control over

smart home devices, but they may still use the devices and be- come the targets of the data collection. Researchers have studied

the privacy concerns of secondary users in a number of recent

studies [28, 32, 43, 67]. Geeng et al. examined the types of privacy

tensions expressed by users in multi-user smart homes [28]. They

found that in existing smart home systems, users are often un- able to adequately mitigate differences in their expectations and

associated tensions. Koshy et al. indicated that privacy percep- tions and needs of main users (i.e., owners) differed from that of

secondary users [32]. They investigated privacy control options

in smart homes as a means to further empower secondary users.

Zeng and Roesner explored privacy tensions between people in a

multi-user home by designing a prototype smart home app with

location-based and supervisory access control mechanisms [67].

They found that the established trust between different smart home

users is one main reason for them to not employ access controls.

Meng et al. conducted semi-structured interviews to investigate

how owners and visitors perceive intelligent personal assistants

(IPA) such as smart speakers [43]. They found that secondary users

including resident owners and visitors consider themselves to be

owners, and have similar awareness and worries around IPA usage.

There are two major differences between these studies and ours.

First, secondary users might participate in the decision of buying

smart home devices and be considered co-owners who could find

smart home devices beneficial [68]. Second, we explicitly explore

the negotiation behaviors between bystanders and owners.

Bystanders’ Privacy in Smart Homes: In recent years, some

researchers started to study bystanders’ privacy in smart homes [2,

Page 3 of 27

Negotiation Behaviors of Owners and Bystanders over Data Practices of Smart Home Devices CHI ’23, April 23–28, 2023, Hamburg, Germany

8, 65]. Yao et al. deduced that social relationships, the duration

of a visit, perceived trust, and device utility establish the perspec- tives of bystanders [65]. Ahmad et al. found that some bystanders

are unsure about the potential data collection from smart home

devices, specifically if the devices actually collected data about

bystanders [2]. This unawareness could prevent bystanders from

making informed privacy related decisions. Bernd et al. carried out

a case study analyzing the presence of nannies in smart homes [8].

They delved into the socio-economic power dynamics between

nannies and smart home owners. Albayaydh and Flechais con- ducted a qualitative study with domestic workers and household

employers [3]. They focused on Jordanians and investigated how

the Islamic belief and customs influenced privacy issues regarding

smart home devices. Privacy was hypothesized to be highly re- spected due to the Islamic beliefs that forbid breaching the privacy

rights of others. However, many employers did not disclose smart

devices, either purposely or because they assumed that employees

already knew about the devices. In these two studies, a smart home

is a workspace more than a private space.

Some researchers have considered both smart home owners and

bystanders in their studies. Cobb et al. defined specific scenarios

where users are incidental [15]. The idea of incidental is similar to

how we define bystanders in this paper. They investigated which

scenarios bystanders found the most problematic and the concerns

that arose for each scenario. Marky et al. explored how owners

and bystanders are concerned about their own privacy from each

other through the qualitative analysis of 42 young adults in an

interview study [42]. Tan et al. explored owners’ perspectives on

bystanders’ privacy including secondary users such as spouses

and roommates [61], but no negotiation was discussed in their

study. Some researchers started to advocate for an open dialogue or

negotiation between owners and bystanders to settle disagreements

about data practices [4, 8, 15, 65]. However, little is known about

how negotiations between owners and bystanders in smart homes

could go as well as what are the factors related to the negotiation

behaviors and outcomes.

Negotiation over Interdependent Privacy: Interdependent pri- vacy means that the privacy of individuals depends on the actions

of others. Biczok et al. explored the existence of interdependent pri- vacy and found that third-party applications on Facebook violated

the privacy of app users’ family members and friends [9]. Symeoni- dis et al. studied the collateral damage of third-party applications

on Facebook and found similar privacy violation problems [59].

Recent studies on privacy management have focused on applying

agreement techniques to solve privacy problems before they take

place. For example, researchers have used multi-agent negotiation

techniques to resolve privacy conflicts among users [44, 58]. The

general idea is to enable users (or their software agents) to negotiate

the ways for sharing the content before it is shared. These solutions

can be helpful in smart homes as well. However, the virtual setting

of Online Social Networks (OSNs) is different from the physical and

intimate setting of smart homes. Personal data in smart homes can

be collected passively without users’ actions. Also, the utility of

using smart home devices is perceived differently from the utility

of using OSNs. For example, posting information in OSNs might

be voluntary while collecting data in smart homes might be essen- tial for safety. Last, negotiation between data subjects and sharers

in OSNs is mainly on data sharing practices, while negotiation in

smart homes can be about all data practices related to owners and

bystanders.

3 DESIGN OF THE STUDY

We answer our two research questions by designing and conduct- ing an interaction study in which participants negotiate over data

practices with digital agents, playing the role of either an owner

or a bystander in smart homes. Our digital agents play the op- posite role of the participants. An interaction study can create a

more dynamic and realistic environment than a pure survey study

can do, thus better eliciting the potential negotiation behaviors

of participants in the real world. Using digital agents that can be

flexibly programmed to emulate different types of counter-parties,

instead of pairing human participants, allows us to more easily

and comprehensively explore different experimental conditions.

It is very difficult to recruit owner and bystander participants to

directly interact in owners’ physical smart homes especially due

to COVID-19; meanwhile, it is unnatural or even awkward to ask

recruited participants to play different roles and directly interact in

hypothetical negotiation scenarios. Note that almost all negotiation

experiments (largely in economics) have humans negotiating with

each other on concrete things (Section 3.2); using programmed dig- ital agents to emulate different types of counter-parties is a unique

design in our study to the best of our knowledge. We also design pre- negotiation, during-negotiation, and post-negotiation questions to

explore factors that may potentially correlate with or affect the

observed negotiation behaviors and outcomes, as is common in

economics and other social science literature (e.g., in [5, 31, 38, 53]).

In the following subsections, we present the design of our study

from four aspects: (1) negotiation scenario generation and assign- ment; (2) owner and bystander digital agents; (3) pre-, during-, and

post-negotiation questions; (4) participant recruitment and ethical

considerations.

3.1 Negotiation Scenario Generation and

Assignment

We design our study with three between-subjects factors: the (digi- tal) agent types (owner vs. bystander agents), the relationship types

(friends vs. strangers), and the negotiator types (strict vs. collab- orative negotiators). Each participant is randomly assigned to a

single scenario (i.e., one treatment condition) with one combination

of these three between-subjects factors. Our within-subjects fac- tors are the smart home device types (a smart camera and a smart

speaker) and the data practice types (data collection, storage, and

sharing practices). Each participant will go through one scenario

with all six combinations of these within-subjects factors to have

six interactions with a digital agent.

Figure 1 illustrates an overview of the interaction scenario gener- ation and assignment in the entire study. Participants are randomly

selected to play either the bystander or the owner role. Each by- stander participant interacts with a digital owner agent of one

randomly chosen negotiator type and relationship type combina- tion (e.g., a strict negotiator who is a friend). Similarly, each owner

participant interacts with a digital bystander agent of the randomly

chosen negotiator and relationship types. The relationship type of

Page 4 of 27

CHI ’23, April 23–28, 2023, Hamburg, Germany Alshehri, et al.

Scenarios

Bystander is a

strict negotiator

and a friend

Bystander is a

collaborative

negotiator and a

friend

Bystander is a

strict negotiator

and a stranger

Bystander is a

collaborative

negotiator and a

stranger

Owner

participant

Negotiate

data

collection

Negotiate

data

storage

Negotiate

data

sharing

Negotiate

data

collection

Negotiate

data

storage

Negotiate

data

sharing

Smart speaker Smart camera

Negotiation interactions

Owner is a strict

negotiator and a

friend

Owner is a

collaborative

negotiator and a

friend

Owner is a strict

negotiator and a

stranger

Owner is a

collaborative

negotiator and a

stranger

Bystander

participant

Scenarios

Each participant will go through six negotiation interactions. The order of devices (smart speaker and smart camera) and

the order of data practices (collection, storage, and sharing)

are independently randomized.

Figure 1: An overview of the interaction scenario generation and assignment in the entire study. For each participant, only

one scenario is randomly chosen. The relationship type of the digital agent is explicitly informed to the participant, but the

negotiator type is not.

the agent is explicitly communicated to each participant but the

negotiator type is not; this design choice reflects real world smart

home environments. The order of devices (smart speaker and smart

camera) and the order of data practices (collection, storage, and

sharing) are independently randomized for each participant. The

assignment of each participant to the bystander or owner role, to

the negotiator type of an agent, and to the relationship type with

an agent was random, so potential external confounders are not a

major concern.

Theme Selection and Social Trust: We choose the same theme

(i.e., estimating the cost of installing a new dishwasher) in all scenar- ios to minimize the amount of assumptions that participants might

have in the negotiations. For example, a bystander participant in

a friend’s home is presented with the scenario that “Assume that

you are a friend of a smart home owner and you are estimating

the cost of a home project (e.g., installing a new dishwasher) at

your friend’s smart home.”. An owner participant in the “stranger”

treatment is presented with the scenario that “Assume that a com- pany worker comes to estimate the cost of a home project (e.g.,

installing a new dishwasher) at your smart home.” Holding the

theme selection constant allows us to focus on measuring the effect

of social trust between owners and bystanders due to their social

relationships. Social trust is found in previous work to be influen- tial to bystanders’ privacy perspectives and owners’ willingness

to address bystanders’ concerns [4, 15, 41]. Thus, we consider that

it may influence how participants will negotiate with our digital

agents. Note that the selected theme is simply on estimating (which

is typically free of charge) the cost of installing a new dishwasher

instead of doing the actual installation, thus we avoided introducing

the payments to bystanders as an extra factor in the scenarios.

Negotiator Types: Our consideration of negotiator types is in- formed by the literature on bargaining [48, 54, 63]. Researchers in

the Harvard Negotiation Project summarized three main types of

negotiators (i.e., soft, principled, and hard negotiators) [48]. Soft

(a.k.a. lenient) negotiators are those who are very accommodating

and tend to agree on any offer by the other party. Principled (a.k.a

collaborative) negotiators are those who would collaborate and

compromise to reach an agreement with others. Hard (a.k.a strict)

negotiators are those who are assertive and insist on their positions

without any compromise. We choose to consider only the strict

and collaborative types of negotiators in the digital agent design

because a lenient digital agent will agree on whatever a participant

would propose and will not be able to stimulate meaningful negotia- tion interactions. Meanwhile, having a strict digital agent will allow

us to explore how far human participants are willing to give up

their personal preferences in order to accommodate the preferences

of the counter-parties in the context of smart home devices; recall

that the negotiator type is unknown to participants, so they do not

know in advance that the (strict) agent will not budge.

Devices, Data Practices, and Privacy Preferences: Previous

studies have shown that different types of smart home devices and

different manufacturers can raise different privacy concerns [15,

41, 68]. As a result, we consider two popular types of smart home

devices, i.e., smart camera and smart speaker. Both types of devices

may collect sensitive data in audio and/or video recordings [4, 15,

41]. In addition, we hypothesized there might be some differences

in negotiating the data practices of these devices.

We consider three typical types of data practices: collection,

storage, and sharing. Note that we provided the background infor- mation regarding these practices and other relevant terms (such

as owner, bystander, and negotiation) to participants before they

started the negotiation interactions as shown in Figure 6 in Appen- dix F. This could help create a common ground of understanding

and avoid some influences from other factors. For example, in terms

of sharing, we introduced that “Owners can share their collected

data with other friends or third-party services for reasons such as

for fun, to help with community safety features, or as a part of a deal

with the smart home devices’ vendors.”. We also provided a scenario

page that introduces participants with some details of the scenario

and the negotiation steps as shown in Figure 7 in Appendix F.

For each type of device and each of its data practices, we identify

four privacy preferences to be negotiated between a participant and

a digital agent. Table 1 lists these detailed privacy preferences. The

lower the preference index is, the more utility of the device and the

Page 6 of 27

CHI ’23, April 23–28, 2023, Hamburg, Germany Alshehri, et al.

described in Section 3.1. Our web application records all detailed

interactions between a participant and a digital agent (choices, deci- sions, participants’ responses to open-ended questions, negotiation

outcomes, etc.) besides the number of rounds. We also followed

the modular design principle so that our web application (to be

shared with other researchers) can be easily extended in the future

to support more negotiator types, device types, data practices, and

privacy preferences.

Formal negotiation research has been translated to informal

settings in the household, and there is an enormous literature on

household economics and household bargaining. Specifically with

respect to Rubinstein Bargaining, Rubinstein and Wolinsky explic- itly framed the model in the context of marriage [52]. Binmore [11]

and Binmore et al. [10] extended the model to relationships in

which one party can terminate bargaining by ending the relation- ship. Lundberg and Pollack further extended this to familial rela- tionships [37]. These frameworks have been used more recently to

study intra-household responses to tax reform (Bargain et al. [6]),

microfinance and gender empowerment (Ngo and Wahhaj [46]),

employer-employee relationships (Collard-Wexler et al. [16]), the

division of inheritance (Gafaro and Mantilla [27]), and shares of

household income spent on children (Ringdal and Sjursen [50]).

Also note that in these and many other studies, Rubinstein Bar- gaining is frequently used to model situations where there is not

perfect initial agreement among all parties on what the outcome

should be. Whether initial preferences for outcomes are exactly

opposite or only differ slightly along a subset of relevant dimen- sions, Rubinstein Bargaining is appropriate as long as there is some

distance that must be covered between the initial preferences and

the final agreement. The experimental structure and design deci- sions in our Rubinstein Bargaining inspired study are fairly stan- dard, but the innovative part mainly lies in the topic of negotiation

- privacy vs. utility in the smart home based social interactions.

Meanwhile, using programmed digital agents to emulate different

types of counter-parties is a unique design in our study to the best

of our knowledge.

3.3 Pre-, During-, and Post-Negotiation

Questions

3.3.1 Pre-Negotiation. We ask participants a set of questions be- fore they start interacting with our digital agents. The objectives

are to examine whether these factors correlate to participants’ sub- sequent negotiation behaviors. First, we ask participants to share

their experience about smart homes and if they have negotiated

any data practices in smart home environments prior to our study.

If participants own smart home devices, we ask them about their

devices, ownership history, and their experience on interacting

with bystanders. If participants did not own smart home devices,

we ask them about what prevented them from acquiring devices,

their experience, and their preferences in regard to negotiating

data practices with smart home owners. Appendix A contains these

questions.

3.3.2 During-Negotiation. After participants are presented with

the scenarios and assigned their role, we ask the during-negotiation

questions (Appendix B) to mainly derive participants’ reactions to

digital agents during the negotiation rounds. At the end of each

negotiation interaction, we ask participants whether additional

privacy options are needed and what are their general comments

regarding the negotiation interaction. We also ask participants at

the end of each negotiation round about the clarity of the questions,

satisfaction with the interactions, and fairness of the digital agent.

3.3.3 Post-Negotiation. Following the negotiation rounds, we ask

participants about their feedback on the emulated interactions. We

also ask what participants think of the negotiation process, general

comments, improvement ideas for the negotiation, and whether

they changed their preferences or not in regard to negotiating data

practices. Lastly, we ask some demographic questions such as age,

gender, and educational background. Appendix C has the complete

list of the post-negotiation questions.

3.4 Participant Recruitment and Ethical

Considerations

3.4.1 Pilot study. We ran an initial pilot study to examine the

clarity of the questions and the negotiation process. We recruited 30

participants from AMT for the pilot. We noticed many low-quality

responses to the open-ended questions. Only 10% of responses to

our open-ended questions were high quality. We then decided to

incentivize participants with an additional $3 USD to provide high

quality responses when we ran the full study.

3.4.2 Participant recruitment. We initially recruited 683 partici- pants from AMT but we excluded 223 submissions because their

responses to all of the open-ended questions were irrelevant. Our

final sample consists of 460 AMT workers. We only recruited adults

and residents of the United States. We did not further screen partic- ipants based on demographics. We compensated participants who

completed the study with $6 USD. We also provided an additional

bonus of $3 USD to 125 participants who gave detailed responses

to our open-ended questions. The average duration for completing

our study was 32 minutes, and the national minimum wage in the

U.S. at the time of the study was $7.25 per hour. The demographic

questions are in Appendix D along with Table 5 that breaks down

some characteristics about the participants in our study.

3.4.3 Ethical considerations. Our study received the Institutional

Review Board (IRB) approval before any research activity began. We

obtained the informed consent from all participants to take part in

the study. We collected AMT IDs of participants for compensation

purposes only. We deleted all the AMT IDs after we compensated

participants. All the collected data is encrypted and stored on our

school’s machines.

4 RESULTS

In this section, we first present participants’ negotiation behaviors

to answer RQ1. We present the quantitative analysis, followed by

the qualitative analysis to provide more insights. We also illustrate

how other factors from participants’ responses correlate with some

negotiation behaviors. Second, we answer RQ2 by exploring the

predictability of negotiation outcomes. Table 2 lists the four treat- ment subgroups for each type of participants in our study with

their numbers of participants.

Page 7 of 27

Negotiation Behaviors of Owners and Bystanders over Data Practices of Smart Home Devices CHI ’23, April 23–28, 2023, Hamburg, Germany

Table 2: Owner participants (n1=238) and bystander partici- pants (n2=222) in their corresponding treatment subgroups.

Types of

partici- pants

Types of digital agents # of partici- pants

Owner

Bystander agent that is a strict friend 63

Bystander agent that is a collaborative friend 61

Bystander agent that is a strict stranger 57

Bystander agent that is a collaborative stranger 57

Bystander

Owner agent that is a strict friend 56

Owner agent that is a collaborative friend 54

Owner agent that is a strict stranger 54

Owner agent that is a collaborative stranger 58

4.1 Result Analysis Methodology

For close-ended questions, we report quantitative answer distribu- tions and between-subjects comparisons supported by statistical

testing. We consider a significance threshold of 0.05 for statistical

testing. We also perform logistic regression to predict if an agree- ment is likely to be reached. For open-ended questions, we perform

thematic analysis [12]. When we report direct quotes from partici- pants’ responses, we use OP to refer to Owner Participants, and

BP to refer to Bystander Participants.

4.2 Negotiation Behaviors

We analyze the negotiation behaviors of participants from three as- pects. First, we analyze the number of rounds that participants took

before reaching the end of the negotiation interaction, regardless of

reaching an agreement or not. Second, we analyze participants’ final

preferences and the distances between their initial and final prefer- ences. Third, we report whether participants reached an agreement

or not with our digital agents. We also compare the differences

between bystander participants and owner participants from all

these three aspects.

4.2.1 Number of rounds. We analyze the number of rounds for

the two smart home devices and the three types of data practices.

Owner participants. Owner participants took 3.3, 3.3, 3.2, and

3.5 rounds on average negotiating with a strict friend, collaborative

friend, strict stranger, and collaborative stranger, respectively, as

shown in Figure 2a. The boxes, the whiskers, and the median val- ues are similar for owner participants when they negotiated with

the agents of the same negotiator types. The median number of

rounds for owner participants when negotiating with collaborative

bystander agents is four, while that with strict bystander agents is

three.

We then explored if statistical differences existed among the four

treatment groups. A one-way Analysis of Variance (ANOVA) was

used to compare the effect of negotiator types and relationship

types on the number of rounds. We found statistically significant

differences in the mean number of rounds between the four treat- ment groups (p-value was < 0.05). These results demonstrate that

owner participants did not have the same negotiation behavior with

different bystander agents. Linear regression tests further showed

that owner participants took more rounds with strict friends than

strict strangers. Future privacy solutions that empower negotiation

features should consider the relationships between owners and

bystanders. Owner participants took on average three rounds to

conclude each negotiation interaction. This is a positive insight for

the research community and smart home vendors, indicating that

providing negotiation features in smart home devices will likely not

be burdensome to owners and bystanders because a small number

of rounds would not take too much time.

Bystander participants. Bystander participants took 1.8, 1.7,

2.0, and 1.8 rounds on average negotiating with a strict friend,

collaborative friend, strict stranger, and collaborative stranger, re- spectively, as shown in Figure 2b. The boxes had similar dimensions

for bystander participants when they negotiated with a strict friend,

strict stranger, and collaborative stranger. Bystander participants

concluded their negotiation interactions slightly faster when nego- tiating with a collaborative friend. The median number of rounds

for bystander participants when they negotiated with strict friends,

collaborative friends, and collaborative strangers is one, while that

with strict strangers is two.

ANOVA was performed to compare the effect of digital agents’

negotiator and relationship types for bystander participants. Sta- tistically significant differences in the mean number of rounds be- tween the four treatment groups were found (p-value was < 0.05).

Linear regression tests further showed that bystander participants

took more rounds with strict strangers than strict friends and more

rounds with collaborative strangers than collaborative friends. It

is worth noting that if bystander participants would only choose

the preference with the best privacy protection (i.e., Preference 4

in Table 1) and would not change their preference, it would take

four rounds at least to reach an agreement with a collaborative

owner agent. This tells us that bystander participants either did not

stick with their preferences or did not choose the most privacy pro- tecting preferences. In both cases, bystander participants showed

willingness to negotiate with owner agents.

Comparison between owner participants and bystander par- ticipants. We found that the differences between owner partici- pants and bystander participants were statistically significant across

all negotiation interactions. Taking each treatment condition sepa- rately, simple t-tests confirmed statistically significant differences

in the mean number of rounds between owner and bystander par- ticipants (p-value < 0.01 for each test). It is noticeable that owner

participants took more rounds to negotiate than bystander par- ticipants. One interpretation of this result could be that owner

participants really cared about the utility of their smart home de- vices and would not easily budge (e.g., OP11: “By turning off the

video and/or audio recordings, I would be defeating the purpose of

having the device in the first place”). In addition, the vast majority

of owner and bystander participants did not exhaust the maximum

number of rounds (i.e., seven). Only five owner participants out of

238 took seven rounds to negotiate.

It is worth noting that in Rubinstein bargaining experiments over

pots of money, often the Nash equilibrium offer is rejected in the

first round if it is perceived to be unfair, and only unfair offers go

on to further rounds of negotiation – often with counteroffers also

being unfair in the other direction [47]. So the number of rounds

can be thought of as an indicator of the perceived fairness of the

first or early offers or counteroffers – more perceived fairness early

on reduces the number of rounds [13, 49]. This is consistent with

Page 8 of 27

CHI ’23, April 23–28, 2023, Hamburg, Germany Alshehri, et al.

(a) Number of rounds for owner participants. (b) Number of rounds for bystander participants.

Figure 2: Distribution of the number of rounds that participants took when negotiating with the digital agents. The x marks

represent the mean values. The four bars in each subgraph correspond to the four treatment groups.

what we found as bystander participants used fewer rounds on

average despite first offers being more favorable to the utility of

owner agents. This could mean that bystander participants thought

of the first offers of owner agents as fair because owner agents are

allowed to exercise their rights in their smart homes. This reflects

the well-known finding that, beyond their own private benefits,

people have preferences over the perceived fairness of outcomes

as well as preferences for whether outcomes conform to social

norms [18, 23–25]; we have more discussion about this at the end

of Section 4.2.2.

4.2.2 Final preferences. We report the final preferences for

owner participants and bystander participants regardless of reach- ing an agreement or not. This can tell us what privacy preferences

owner and bystander participants are willing to accept when nego- tiating over data practices of smart home devices.

Owner participants. Figure 3a shows the final preferences

reached by owner participants when they negotiated with different

bystander agents. With strict bystander agents, owner participants

would have to choose Preference 4 – the highest privacy protection

for bystanders – in order to reach an agreement. With collaborative

bystander agents, an agreement could be reached at any preference

so less pressure would be applied to owner participants to compro- mise the utility of their smart home devices. The average distances

between initial preferences and final preferences for owner par- ticipants were 1.1, 1.6, 1.6, and 1.3 when they negotiated with a

strict friend, collaborative friend, strict stranger, and collaborative

stranger, respectively. It is worth noting that 65% of owner partici- pants did not choose Preference 1 (i.e., the best utility preference

for owner participants) as their final preference. Preference 4 was

the most common final preference for owner participants when

dealing with a strict friend, collaborative friend, and strict stranger,

while Preference 1 was the most common final preference when

dealing with a collaborative stranger.

From Figure 3a, we can see the proportion of owners with fi- nal Preferences 1 or 4 is largest in the strict bystander treatments.

It seems that most owners either stood their ground with their

initial Preference 1, or caved to the demands of strict bystanders

and changed to Preference 4. By contrast, the proportion of final

Preferences 2 or 3 is larger in the treatments with collaborative

bystander agents than with strict bystander agents. This indicates

that owner participants were willing to compromise some of their

smart home utility when less pressure was applied by the bystander

agent. This kind of compromise however could be seen as reason- able by owner participants as it was not a complete compromise

(e.g., OP19:“I think that storing the data for such a short period of

time is reasonable and I would still have a record of our interaction

in case I needed it”). This owner participant was willing to shorten

the period of storage as a partial compromise.

Figure 10 in Appendix G details the final preferences for owner

participants across all interactions. We also explored (by using the

Chi-Square Test for Association) the relationships between the

final preferences of owner participants and the four treatments.

We found the differences in final preferences to be statistically

significant across treatments (the null hypothesis of no treatment

effect was rejected with p-value < 0.05).

Bystander participants. Figure 3b illustrates the final pref- erences of bystander participants across treatments. With strict

owner agents, bystander participants would have to choose Pref- erence 1 – the highest utility for owners – to reach an agreement.

With collaborative owner agents, an agreement could be reached

at any preference so that less pressure was applied to bystander

participants to compromise their privacy in other people’s smart

homes. It is clear that Preference 1 was dominant across the four

treatments. One explanation might be that bystander participants

thought negotiating with owners was not appropriate (e.g., BP4: “It

is their house and I believe that it would be rude of me to ask them to

go out of their way so that I could come over”). Another possibility is

that bystanders might trust owners to do what was best for them

(e.g., BP41: “I believe my friend should be trusted”).

The average distances between initial preferences and final pref- erences for bystander participants were 0.43, 0.34, 0.32, and 0.4

when they negotiated with a strict friend, collaborative friend, strict

stranger, and collaborative stranger, respectively. A possible expla- nation for such short distances might be that bystander participants’

initial preference selections were not too far from the owner agents’

preferences. Figure 11 in Appendix G details all the final preferences

for bystander participants across all the interactions. For example,

bystander participants demonstrated a relatively more aggressive

approach when negotiating over data sharing practices compared

to other data practices – only two of them reached Preference 1 as

the final preference for data sharing practices (i.e., agreeing with

the owner agent on sharing without any restrictions).

Page 11 of 27

Negotiation Behaviors of Owners and Bystanders over Data Practices of Smart Home Devices CHI ’23, April 23–28, 2023, Hamburg, Germany Percentage of participants

0%

25%

50%

75%

100%

Owner participant

interacting with a strict

friend as a bystander

Owner participant

interacting with a strict

stranger as a bystander

Bystander participant

interacting with a strict

friend as an owner

Bystander participant

interacting with a strict

stranger as an owner

No agreement Partial agreement Full agreement

Figure 5: Distribution of agreement types for participants. Full agreement means that a participant reached six agreements

with the digital agent. Partial agreement means that a participant reached one to five agreement(s).

factor, preference to negotiate, we found that participants who

preferred to negotiate had a greater number of rounds, but fewer

total agreements and shorter distance between initial and final

preferences (p-value < 0.05). For the third factor, exposure to smart

homes, we found that participants who have been to a smart home

as a bystander reached fewer agreements (p-value < 0.05). It is

worth noting that we did not find significant differences between

age groups (i.e., participants who are 30 years old or older vs. the

rest) and between two major gender groups (i.e., male vs. female);

we used t-tests to evaluate differences in an outcome between binary

values of each factor, with a null hypothesis of no difference in each

comparison.

4.2.5 Qualitative analysis of participants’ negotiation be- haviors. We divided our qualitative results into two parts: one for

owner participants and the other for bystander participants. We

derive insights from the responses to our open-ended questions at

each round of the negotiation. Recall that we always ask partici- pants the during-negotiation questions for them to explain their

negotiation behaviors. These questions can be found in Appendix B.

Specifically, we performed thematic analysis [12] on participants’

responses to Question B.3.2 (“Do you have any comments about this

negotiation?”) because we are interested in participants’ comments

immediately at the end of each of the six negotiation interactions.

To analyze the responses, we further separately divided both

owner and bystander participants into four exclusive subgroups

based on the following sequence (i.e., only those who are not put

into any previous subgroup can be considered for the next sub- group). First, those who budged and reached an agreement in a

negotiation interaction (i.e., in at least one of the six interactions)

will be put into the first subgroup; these participants’ responses

corresponding to the budged and agreement-reached interactions

are analyzed for this subgroup. Note that the participants in this

first subgroup have interacted with either a strict or a collabora- tive agent (Section 3.2). The second subgroup consists of those

who did not budge but reached an agreement in a negotiation

interaction; these participants’ responses corresponding to the non- budged but agreement-reached interactions are analyzed for this

subgroup. Note that the participants in this second subgroup must

have interacted with a collaborative agent (Section 3.2). The third

subgroup consists of those who budged but did not reach an agree- ment in a negotiation interaction; these participants’ responses

corresponding to the budged but agreement-unreached interac- tions are analyzed for this subgroup. Note that the participants

in this third subgroup must have interacted with a strict agent

(Section 3.2). The fourth subgroup consists of those who did not

budge and did not reach an agreement in a negotiation interaction;

these participants’ responses corresponding to the non-budged and

agreement-unreached interactions are analyzed for this subgroup.

Note that the participants in this fourth subgroup also must have

interacted with a strict agent (Section 3.2). Thus, we have eight

exclusive subgroups of owner and bystander participants, and we

separately analyzed the corresponding responses in each subgroup.

In our thematic analysis, three coders started by reading the

responses from 20 randomly chosen owner participants (in four

subgroups) and 20 randomly chosen bystander participants (in four

subgroups), and then created an initial codebook which consists of

eight sub-codebooks for the eight subgroups of participants. One

coder analyzed bystander participants’ responses and another coder

analyzed the owner participants’ responses. The third coder ana- lyzed both types of responses, so that all responses were analyzed

by at least two coders. All coders then met to update and generate

the final version of the codebook as shown in Appendix E. After

all responses were coded, we calculated the inter-coder agreement

using Cohen’s Kappa [26] and obtained 83% as the result. Note

that any value greater than 81% is typically considered as excellent.

Codes were applied exclusively, i.e., each response is only assigned

one code. The three coders then further resolved any disagreement

that could be caused by misunderstanding or mislabeling. Thus, we

have the codes for all responses (including disagreement-resolved

responses) and we report their statistics below.

Page 12 of 27

CHI ’23, April 23–28, 2023, Hamburg, Germany Alshehri, et al.

Owner participants. There were 98 owner participants who

budged and reached an agreement in a negotiation interaction.

About 54% of them described respecting bystanders’ privacy as

the main reason (e.g., OP42:“I like to respect others’ privacy if they

request it, therefore I chose to give him what he wanted and have

my device not share the data”). Additionally, about 22% of owner

participants thought a win-win situation could be reached (e.g.,

OP49: “Storing the data for two weeks is a good compromise as these

recordings are for my protection and the bystander”). Lastly, 12% of

owner participants described that their decisions were based on

how they would like to be treated when they are bystanders in

other people’s smart homes (e.g., OP9: “I would agree to my friend’s

request for no data storage since I would like the same respect”).

We found that 33 owner participants did not budge but reached

an agreement in a negotiation interaction. About 62% of them

valued utility over the privacy of bystanders (e.g., OP13: “We pay

for monitoring service. We cannot change the vendor configuration”).

It is common for smart home owners to pay for such safety features,

so they may not want to sacrifice the utility. This owner participant

also described how it might be difficult to change such settings.

This means owners might have less control on how their smart

home services work, and these services may even have full control

over the owners’ smart home devices. Future work can specifically

look into these popular services as they might raise other privacy

concerns. Another 19% of owner participants stated “my house, my

rule”. The agreements that were reached with these participants

were only possible with collaborative bystander agents as strict

bystander agents would not budge at all (e.g., OP57: “I do not see a

reason to negotiate with others if they are in my home, then they are

subject to my rules”).

We found that 80 owner participants budged but did not an

agreement in a negotiation interaction. About 47% of them thought

there was a possibility for a win-win situation, but the bystander

agent was not collaborative (e.g., OP42: “I also disagree with some- one being unwilling to compromise when I am willing to go from

permanent storage to temporary storage of 2 weeks”). About 33% of

owner participants reached their limit in terms of compromising

the utility of their smart home devices (e.g., OP60:“I will not accept

the conditions imposed by the person. I feel that negotiating for two

week storage was enough of a sacrifice in my own home. I do not

accept their resuqest”).

We found that 27 owner participants did not budge and did not

reach an agreement in a negotiation interaction. About 55% of them

thought the bystander agents were not reasonable and did not have

the right to ask for such requests (e.g., OP27: “Such a demand by

the bystander is simply ridiculous. The purpose of me having a smart

camera in and outside my home is to collect video and audio data”).

This response was towards a strict bystander agent that kept asking

owner participants to compromise the utility of their smart home

devices. Another 25% of owner participants directly shared that “my

house, my rules”. Interestingly, about 6% of owner participants re- sponded to some of the negotiation interactions with "never thought

of this before so I am not sure". Smart home technologies are still

relatively new and the privacy of bystanders in smart homes is an

active research topic. The awareness of such potential violations of

bystanders’ privacy needs more attention from researchers.

Bystander participants. We found that 124 bystander partici- pants budged and reached an agreement in a negotiation interaction.

About 48% of them thought that since it was the owners’ smart

home, it was their right to operate the smart home as the owners

wished (e.g., BP4: “It is their house and I believe that it would be

rude of me to ask them to go out of their way so that I could come

over”). In addition to respecting the owner’s sovereignty of their

smart home, this participant showed that it could be rude to even

ask about the data practices. About 25% of bystander participants

were satisfied with the fact that the owner agent informed them

about the data practices (e.g., BP43: “Although I did not like what

they wanted I am happy they at least told me about the extent of their

data collection practices within their home”). Generally speaking,

privacy protection is composed of two major parts: transparency

and control; the best practice is to offer both. BP43 seemed satisfied

with the transparency but was concerned about the data collection

practices. About 20% of bystander participants budged and reached

an agreement but were still uncomfortable. They tried to reach

a win-win situation where the owner agent could preserve some

utility and the bystander would compromise a little (e.g., BP16:“If

storing collected data for 2 weeks is not acceptable, nor for 6 weeks,

I have to accept permanent storage because this is the wish of the

owner, although I do not see the utility”).

We found that 22 bystander participants did not budge but

reached an agreement in a negotiation interaction. About 30% of

them believed their privacy and security were very important (e.g.,

BP3: “I do not prefer sharing my data due to growing attacks, more- over smart home automation can be hacked”). This response shows

another layer of concerns that bystander participants had. In addi- tion to wanting no sharing, this bystander participant thought of

the possibilities for the smart home devices to be hacked. About

24% of bystander participants would be willing to change their

preferences if owner agents justified the purpose of the collection

(e.g., BP55:“I still insist that you do not record until there is a pur- pose”). This participant would not change the preference due to

the lack of justifications by the owner agent. In a more realistic

dialogue, owners and bystanders might ask for further justifications

while negotiating. This aligns with findings from previous work

where justifications for data collection preferences were needed for

owners to address bystanders’ privacy concerns [15]. About 22% of

bystander participants thought their friends should respect their

privacy and honor their preferences (e.g., BP52:“If we are friends,

I do not know why they are sticking to their guns so firmly. I would

suggest we go to a bar or restaurant. I would feel even more uncom- fortable that they are refusing my request or discussing it”). This

response shows that friends may expect such gestures from their

owner friends. Investigating how such disagreements might harm

the social relationships could be beneficial for future research.

We found that 60 bystander participants budged but did not

reach an agreement in a negotiation interaction. About 42% of them

shared that only some options were fair (e.g., BP50: “I think it is fair

to collect data when interacting with the device”). This participant

did not budge further towards the owner agent’s preference. About

29% of bystander participants only compromised to options that

would be necessary for a device’s functionality (e.g., BP14: “I am

only comfortable sharing with device vendor”). Sharing with device

Page 13 of 27

Negotiation Behaviors of Owners and Bystanders over Data Practices of Smart Home Devices CHI ’23, April 23–28, 2023, Hamburg, Germany

vendors can be necessary in many smart home devices because of

cloud services that require continuous sharing.

We found that 16 bystander participants did not budge and did

not reach an agreement in a negotiation interaction. About 52% of

them wanted to protect their privacy and would not compromise

at all (e.g., BP23: “I would prefer they are more lax on what they are

monitoring so i can feel at ease about my privacy”). Among these

bystanders, many used a more aggressive tone (e.g., BP39: “You

cannot invade my privacy without my permission and share the data

that include me”). About 22% of bystander participants shared that

owner agents should respect them and change the data practices

because they are friends (e.g., BP47:“I wish my friends would give in a

little rather than stand completely firm. There is almost no negotiating

period”). Some of these bystander participants would reconsider

their friendship if they were in such a situation (e.g., BP15: “It really

concerns me that a so-called friend wants a permanent record of all

our conversations. I am just not comfortable with him keeping all

data forever. I am rethinking our friendship”). The rise of smart

homes will bring more complexity to social contexts. Although it

might be seen as a rude act for bystanders to ask owners to change

their data practices, it could be also seen as a sign of disrespect in

some scenarios for bystanders when owners do not allow for some

flexibility with their data practices.

4.3 Prediction of Negotiation Outcomes

Recent reports indicate that each smart home will have up to 50

devices on average by 2022 [57]. This includes smartphones, com- puters, IoT devices, and smart cars. However, the overwhelming

majority will be IoT devices. Negotiating data practices for each

device may be burdensome and people might avoid the hassle. If we

can predict whether agreements can be reached, a lot of time can be

saved. In addition, previous research has shown that discussing data

practices with owners of smart homes can be awkward sometimes

and people might feel discouraged to initiate such a dialogue [4].

Thus, it can be helpful to explore whether or not negotiating over

smart home data practices is worthwhile. This would allow people

to anticipate the results of initiating such a dialogue. Simply, people

would answer a few questions and then a prediction model might

anticipate how the negotiation could go. From the responses of

participants, we analyzed if certain patterns could be recognized

to predict some outcomes. For example, we used the responses

to the pre-negotiation questions (Appendix A) to predict whether

agreements could be reached or not. We considered agreements for

each device and its three corresponding data practice types. This

resulted in six different dependent variables to be predicted. The

ground truth in prediction was whether a participant reached an

agreement or not with a digital agent.

The independent variables were history of negotiation, prefer- ence to negotiate, exposure (i.e., whether or not the participant has

been a bystander in a smart home), agent types, and relationship

types. With these independent variables, we predicted whether an

agreement would be reached for each of the six negotiation inter- actions. We trained six logistic regression models on 80% of the

responses and tested on the remaining 20% of the responses. Table 6

in Appendix G shows the entire regression results of predicting an

agreement for each interaction. By looking at the coefficient and

Z-values, we found that agent type was very prominent among all

the models for predicting agreements. The corresponding p-value

was significant across four out of the six models. This aligns with

our findings which show that owner participants and bystander

participants had significant differences in their negotiation behav- iors. The average accuracy for predicting whether the participant

would reach an agreement was 70%. The highest accuracy was

78.2% for the data collection practices of the smart camera. The

lowest accuracy was 54.3% for the data sharing practices of the

smart camera. These preliminary results are encouraging, and more

advanced models could be explored in the future to improve the

prediction accuracy.

4.4 Results of Post-Negotiation

We asked participants a set of questions after their interactions with

our digital agents to share their feedback about the negotiations.

4.4.1 Satisfaction and fairness. We asked participants about what

they thought about the interactions in Questions B.3.4 and B.3.5 in

Appendix B.3. About 87% and 84% of participants were very satisfied

or satisfied with the interactions regarding the smart speaker and

the smart camera, respectively. In regard to fairness, about 79% and

77% of participants thought the negotiation interactions were fair

when they negotiated data practices of the smart speaker and the

smart camera, respectively. Satisfaction and fairness ratings from

the participants provide information regarding how they feel about

the interface and the process of the negotiation interactions.

4.4.2 Most important data practice to negotiate. We asked partici- pants what would be the most important data practice to negotiate

if there was only one choice (Question C.9 in Appendix C). From

the responses, owner participants were equally split between data

collection and storage practices, with 38% of responses to each. In

the open-ended responses, owner participants shared a few reasons

explaining their choices. About 65% of owner participants thought

controlling data collection practices precedes the rest of the prac- tices (e.g., OP14: “Collection precedes the rest is the most relevant”). It

is true that stopping data collection will inherently impact data stor- age and sharing practices as there may be nothing more to negotiate

if data collection is stopped. Second, 28% of owner participants re- garded negotiating data collection as easy and straightforward (e.g.,

OP37: “It is very clear and cut. I either collect or do not collect”). This

response explains how some owner participants also considered

the applicability of the negotiable preferences. When it comes to

collection, it is clear and easy as owners of smart homes can easily

turn off their devices. For storage, some owner participants chose

to only consider negotiating over data storage practices as it had

the best chances in reaching agreements with bystanders without

entirely compromising the utility of their devices (e.g., OP57: “I

think storage would be the most negotiable practice and perhaps the

practice easiest to find a mutual agreement”).

On the other hand, bystander participants had different prefer- ences. About 40% of bystander participants would choose to nego- tiate data sharing practices. About 37% of bystander participants

understood the importance of devices’ utility but wanted to control

how their data would be shared (e.g., BP15: “The device is used to col- lect data, I would respect that, as long as the data will not be misused

Page 14 of 27

CHI ’23, April 23–28, 2023, Hamburg, Germany Alshehri, et al.

and shared without consent”). Other bystander participants thought

the privacy violation happened in data sharing practices (e.g., BP32:

“Because sharing data with anyone is a breach of my privacy”).

It is worth noting that these responses differ from the negotiation

behaviors of both types of participants. Owner participants reached

more agreements with bystander agents on data sharing practices,

yet they preferred to negotiate data collection or storage practices

when there was only one choice. Bystander participants reached

more agreements with owner agents on data collection and storage

practices, yet they preferred to negotiate data sharing practices

when there was only one choice. We only observed these differences

in our participants’ responses, and we did not further investigate

the reasons for such differences.

4.4.3 Feedback for improving the negotiations between owners and

bystanders.

Improvements for each type of data practice. We asked partic- ipants what other factors they would like to see regarding negotiat- ing data collection, storage, and sharing practices (multiple-answer

Questions C.3, C.4, and C.5, respectively in Appendix C). In regard

to the data collection practices, 70% of owner participants and 88%

of bystander participants responded that negotiating where the

data collection happens can be helpful. It is intuitive that collecting

data in a bathroom or bedroom could be more invasive than collect- ing data in the living room [40]. This also aligns with the findings

in another recent study that the location of the devices might im- pact users’ privacy expectations [33]. In regard to the data storage

practices, 55% of owner participants and 78% of bystander partic- ipants mentioned negotiating whether data would be encrypted

or not. Fortunately, encrypting the IoT traffic and the stored data

is common nowadays [55]. This issue may be relevant for own- ers who store their data locally. Regarding data sharing practices,

90% of owner participants and 85% of bystander participants chose

“sharing with my explicit approval” as their answer.

Improvement for the overall negotiation process. We asked

participants how the negotiation process could be improved as a

whole (multiple-answer Question C.6 and open-ended Question C.7

in Appendix C). Prior to entering our study, we informed all par- ticipants that the potential justifications behind the data practices

could be for safety, entertainment, automation, or others (Figure 6 in

Appendix F). About 43% of owner participants and 37% of bystander

participants mentioned that better or specific justifications would

be helpful. For example, owners could tell bystanders that the data

practices for their smart cameras are for safety reasons; bystanders

could give owners reasons on why some data practices could harm

them, which aligns with previous research where justifications

from bystanders were needed in order for owners to consider ad- justing their smart home data practices [15]. Some participants

also indicated the need to know the justifications behind individ- ual decisions when they negotiate with each other. About 33% of

owner participants and 51% of bystander participants mentioned

that having more privacy options to negotiate about (as described

above for each type of data practice) would be an improvement.

5 DISCUSSION

First, we discuss the implications of our findings. Second, we pro- vide some tentative recommendations for device vendors to improve

the privacy of bystanders while maintaining the utility of smart

home devices. Third, we discuss some limitations and potential

future work.

5.1 Implications of Our Findings

5.1.1 The Need for Control and Negotiation. Our study was moti- vated by recent suggestions regarding the need for open dialogues

between owners and bystanders when it comes to agreeing about

data practices in smart homes [4, 7, 15, 62]. Our findings show pos- itive evidence to the research community and smart home vendors

that facilitating the negotiation of data practices between owners

and bystanders is a promising approach. They indicate that pro- viding negotiation features in smart home devices might not be

burdensome to both owners and bystanders since they would only

need a relatively small number of rounds to negotiate as we re- ported in Section 4.2.1. Although owner and bystander participants

acted differently, most of them showed their willingness to negoti- ate as discussed in Section 4.2.2. Owner participants leaned more

towards preserving the utility of their smart home devices, while

bystander participants showed more willingness to negotiate by

accepting the less privacy-preserving preferences; this result could

be related to the power imbalance between owners and bystanders

as discussed in Section 4.2.2.

5.1.2 Trust between Owners and Bystanders. Social trust between

data subjects and data owners (i.e., data recipients) was found in

recent studies to be a major factor that influenced how data subjects

perceive data collection about themselves – if data subjects trusted

data owners, data collection became more acceptable [1, 29, 65].

In our study, we found similar influence when either owner or

bystander participants negotiated with the digital agents. Owner

participants reached more agreements with bystander digital agents

when the scenarios were about friends instead of strangers as re- ported in Section 4.2.3. Moreover, Preference 4 (for the strongest

privacy protection) was the most common final preference for

owner participants when dealing with either a strict or a collabora- tive friend (Section 4.2.2). Similarly, bystander participants reached

more agreements with owner digital agents when the scenarios

were about friends instead of strangers (Section 4.2.3). Future re- searchers and smart home vendors should consider social trust in

their design of privacy protection mechanisms for bystanders.

5.1.3 Owners’ Privacy Risks due to Bystanders. We acknowledge

that owners could face greater privacy risks than bystanders in

many situations. For example, a smart speaker could loudly read

an email notification or a reminder that might have sensitive infor- mation about the owner. Such privacy risks of owners have been

explored in some recent studies [15, 42]. However, owners often

have some control and knowledge of their smart home devices. To

illustrate, owners have the opportunity to consent to the privacy

policies of the devices before using them as that is a required process

in many devices. Owners can also turn the devices off without the

need to consult with bystanders or without facing social awkward- ness (which bystanders would face when asking owners to turn

Page 15 of 27

Negotiation Behaviors of Owners and Bystanders over Data Practices of Smart Home Devices CHI ’23, April 23–28, 2023, Hamburg, Germany

the devices off [4]). In contrast, bystanders are normally unaware

of the existence of smart home devices and are not provided with

the opportunity to consent to the data collection [2]. Therefore, we

focused on the privacy risks faced by bystanders in this study.

5.1.4 Participants’ Preferences, Opinions, and Social Norms. The

codes derived in our qualitative analysis (Section 4.2.5) are related

to the preferences, opinions, and social norms of participants as

well as the rationales behind their decision making. For participants

in the role of owners, we noticed three main norms and opinions

regardless of their budging behaviors: (1) My house my rules; this

means that some owner participants would more likely stick to the

preferences with the best utility (i.e., Preference 1 in Table 1). (2) A

win-win situation is possible; this means that some owner partici- pants would more likely compromise a little to reach an agreement

by choosing the preferences that preserve some utility and offer

some privacy protection to bystanders (i.e., Preferences 2 and 3 in

Table 1). (3) Immense respect to bystanders’ privacy; this means that

some owner participants would more likely compromise the utility

of the smart home devices to satisfy bystanders’ privacy desires. For

participants in the role of bystanders, we noticed three main norms

and opinions regardless of their budging behaviors: (1) Immense

respect to owners for the utility of their devices, i,e., some bystander

participants would more likely budge and accept the owner agent’s

setting without negotiating. (2) A win-win situation is possible, i.e.,

some bystander participants would more likely compromise a little

to reach an agreement by choosing Preferences 2 and 3. (3) Privacy

is too crucial to be sacrificed, i.e., bystanders would not compromise

at all because their privacy is too important.

5.2 Recommendations for Smart Home Device

Vendors

We now provide some tentative recommendations to device ven- dors, with the assumption and expectation that in the future they

would be interested in developing solutions to facilitate data prac- tice negotiations in smart homes. We definitely recommend device

vendors to leverage what we analyzed in Section 4.4.3 regarding

the feedback for improving the negotiations between owners and

bystanders. Besides, we have the following recommendations.

Temporary modes of operations. It might be a burden for own- ers to switch their settings back and forth when addressing by- standers’ concerns. Some responses (Question C.7 in Appendix C)

from our study indicate this issue (e.g., OP51: “I wish there was a

privacy option to stop recording the data only when someone enters

the room but to resume after they leave”). We recommend smart

home vendors to provide some modes of operations from the pri- vacy perspective. One mode can be the normal operation for the

owner’s preferences, and another mode can be something like the

visitor mode. This would help reduce their effort if owners decide

to change certain data practices. Previous research also indicated

that some owners desire for different modes of operations for their

own privacy settings [64]. Note that some smart home devices have

certain modes of operations (e.g., the Guest Mode on Google speak- ers and smart displays that can help stop saving Google Assistant

interactions to a user’s Google account or personalizing a user’s

experience). However, those modes are mainly for usability rather

than privacy purposes. Modes should be extended to account for

the privacy of bystanders as well as owners.

Automating the entire negotiation process. For the negoti- ation features to be effective and sustainable, they should be au- tomated with the minimum burden being put on owners and by- standers. Our digital agents can potentially be integrated into a

stand-alone app or into smartphone assistants. This means that

owners and bystanders can preset their preferences in advance.

Digital agents can then represent users by learning from situational

variables and the history of negotiations. For example, the negotia- tion features could allow users to answer some simple questions

and then click a button or use a voice command to start the nego- tiation. The digital agents can further negotiate with each other

on the users’ behalves. Automation for reducing users’ effort in

smart home privacy and security protection was also mentioned in

studies such as [64, 67]. It is unknown to us whether automating

the entire negotiation process is practicable. Research needs to be

done in the future to explore its practicability.

5.3 Limitations and Future Work

AMT workers (used in this study) might not be the best representa- tion of the U.S. population as they are generally more tech-savvy

and more privacy-conscious than the average Americans [30]. An- other limitation in our study is that the overwhelming majority

(96%) of the participants owned smart home devices. This could be

related to the tech-savviness of AMT workers and a higher likeli- hood for them to be early adopters of new technologies. As a result,

our study may reflect different preferences for privacy versus device

utility than those of the general population.

In the study design, we accounted for the differences in social

trust, but not in technical capabilities. In other words, we clearly

mentioned if the owner agent was a friend or a stranger, but we

did not mention whether the agent was technically savvy enough

to change their privacy settings. Previous research found that by- standers might trust their family and friends to have good inten- tions, but might not trust their technical capabilities on data protec- tion [4, 40]. Future work may further account for technical trust in

negotiation. In addition, we designed negotiations to have a similar

style. These interactions might seem repetitive to some participants

and might have prevented them from negotiating effectively in the

six interactions. Despite our clear message stating that the six nego- tiation interactions were to be treated as independent of each other,

some participants treated them as related (e.g., BP13: “Sharing was

not too important because I negotiated to never be recorded. So, it does

not matter to me”). It is worth noting that we minimized the order

effect by randomizing the order of the interactions.

Additionally, we only considered one owner and one bystander

in our design. However, it is likely that a smart home device might

be owned and controlled by more than one person. Likewise, there

might be more than one bystander at a time in the smart home.

Investigating interactions between multiple bystanders and/or own- ers is a topic for future work. Previous studies have shown that

different types of smart home devices and different manufacturers

can raise different privacy concerns [15, 41, 68]; therefore, future

work can explore the negotiation behaviors over more types of

devices. Lastly, it is difficult to recruit participants to observe their

Page 18 of 27

CHI ’23, April 23–28, 2023, Hamburg, Germany Alshehri, et al.

A.2 Questions for participants who do not own

smart home devices

(1) What keeps you from owning smart home devices? (Choose

the most important reason) [#Cost, #Privacy Concerns,

#Unnecessary, #Other (please specify: )]

(2) When you think of smart home devices what do you think

of?

(3) What would make you buy a smart home device?

A.3 Questions for both

(1) Have you negotiated the data practices of a smart home

device with an owner? [#Yes, #No, #I am unsure]

(2) Would you prefer to be given the option to negotiate the data

practices of smart home devices with their owner? [#Yes,

#No, #I am unsure]

(3) Have you been to other people’s smart homes? [#Yes, #No,

#I am unsure]

(4) Were you explicitly told that there were smart home devices?

[#Yes, #No, #I am unsure]

(5) If Yes, how were you told?

B DURING-NEGOTIATION QUESTIONS

During-negotiation questions are asked somewhere in the progress

of a negotiation interaction. Participants also need to explain their

answers to all close-ended and open-ended questions.

B.1 Owner Participants

These questions are only asked to participants who are assigned

the owner role. By default, we set the owner participants’ first pref- erence to the most privacy violating option. These are listed below:

Collection (Smart Camera): Collecting both audio recordings

and video footage. Collection (Smart Speaker): Collecting all

audio recordings. Storage: Permanent storage. Sharing: Sharing

without restrictions.

(1) What option are you willing to change to? (The options are

identical to those in Table 1.)

Collection (Smart Camera): [#Collecting both audio

recordings and video footage (keep it the same), #Collecting

audio recordings and periodically collecting video footage,

#Only collecting audio recordings, #No collection]

Collection (Smart Speaker): [#Collecting all audio record- ings (keep it the same), #Collecting audio recordings when

interacting with the device and periodically when not inter- acting with the device, #Only collecting audio recordings

when interacting with the device, #No collection]

Storage: [#Permanent storage (keep it the same), #Storing

collected data for 6 months, #Storing collected data for 2

weeks, #No storage]

Sharing: [#Sharing without restrictions (keep it the same),

#Sharing with the device vendor and a given list of third

parties, #Sharing with the device vendor, #No sharing]

(2) You and the bystander could not come to an agreement. Do

you think they will stay or leave? [#Stay, #Leave]

B.2 Bystander Participants

These questions are only asked to participants who are assigned

the bystander role.

(1) Are you comfortable with how my [smart camera / smart

speaker] is [collecting / storing / sharing] data? [#Yes, #No]

(2) How would you prefer my [smart camera / smart speaker]

[collects / stores / shares] data?

(The options are identical to those in B.1.1.)

(3) Will you stay or leave? [#Stay, #Leave]

B.3 Individual Interaction and Device Feedback

Questions (1) and (2) are asked at the end of each negotiation

interaction. Questions (3), (4), and (5) are asked after negotiating

each smart home device.

(1) Do you wish there were other privacy options in addition to

the four options shown in the figure above?

(2) Do you have any comments about this negotiation?

(3) How clear were the [smart speaker / smart camera] data

collection, storage, and sharing negotiation interactions to

you? [#Very Unclear, #Unclear, #Not Sure, #Clear, #Very

clear]

(4) How satisfied are you with the outcomes of the negotia- tion interactions over data collection, storage, and sharing

practices for the [smart speaker / smart camera]? [#Very

Unsatisfied, #Unsatisfied, #Not Sure, #Satisfied, #Very Sat- isfied]

(5) How fair do you think the emulated agent behaved while

negotiating the data collection, storage, and sharing practices

of the [smart speaker / smart camera] with you? [#Very

Unfair, #Unfair, #Not Sure, #Fair, #Very fair]

C POST-NEGOTIATION QUESTIONS

These questions are asked after participants go through all six

negotiation interactions regarding their overall experience. Both

types of participants had the exactly same questions, except for

Question (8) where only either the word “Owner” or “Bystander”

was presented based on the negotiator type of the digital agent.

Participants also need to explain all their answers.

(1) Do you have any suggestions regarding the overall nego- tiation interactions about the data collection, storage, and

sharing practices for the smart camera and smart speaker?

[#Yes, #No, #I am unsure]

(2) What other types of devices would you like to negotiate?

(3) In the negotiation regarding data collection practices for the

smart camera and the smart speaker, you mainly negotiated

about the types of data to be collected. What other factors

would you like to negotiate? [2Location of the data collec- tion (e.g. in different rooms), 2Whether noise is added to

obfuscate your data, 2Other, 2None]

(4) In the negotiation regarding data storage practices, you

mainly negotiated about retention policies of data collected

(e.g., how long data about you is stored). What other factors

would you like to negotiate? [2Location of data storage (e.g.

in the cloud or local), 2Protection of the collected data (e.g.

is data encrypted or not), 2Other, 2None]

Page 19 of 27

Negotiation Behaviors of Owners and Bystanders over Data Practices of Smart Home Devices CHI ’23, April 23–28, 2023, Hamburg, Germany

Table 3: Descriptive analysis of the pre-negotiation questions in A.1 and A.2.

Questions A.1.1 A.1.2 A.1.4 A.1.5 A.1.6 A.2.1 (only 17 par- ticipants did not

own smart home

devices)

Answers One device

(10%), 2-5 de- vices (76%), 6-10

devices (12%),

More than 10

devices (2%)

Smart TV (82%),

Smart speaker

(75%), smart

camera (63%),

smart appliance

(39%)

Less than a

month (0.01%),

1-6 months

(21%), 6 months

- one year (30%),

1-3 years (33%),

More than 3

years (16%)

Convenience

(80%), Safety

(70%), Enter- tainment (55%),

Automation

(48%)

Yes (51%), No

(45%), I am

unsure (4%)

Cost (30%), Pri- vacy concerns

(29%), Unneces- sary (28%), Other

(13%).

Table 4: Descriptive analysis of the pre-negotiation questions

in A.3.

Questions A.3.1 A.3.2 A.3.3 A3.4 (316 par- ticipants have

visited a smart

home as a by- stander)

Answers Yes (58%),

No (37%),

I am un- sure (5%)

Yes (64%),

No (28%),

I am un- sure (8%)

Yes (69%),

No (27%),

I am un- sure (4%)

Yes (71%), No

(28%), I am un- sure (1%)

(5) In the negotiation regarding data sharing practices, you

mainly negotiated about with whom the data is shared (e.g.,

who can access your data). What other factors would you like

to negotiate? [2Sharing with my explicit approval, 2Sharing

obfuscated data only (e.g. blurred faces in video recordings,

added noise in audio recordings), 2Other, 2None]

(6) How can the negotiation interactions improve? [2More

rounds of negotiation, 2More privacy options for each prac- tice to negotiate about, 2Owners should provide reasons

why they are not accepting bystanders’ privacy preferences,

2Bystanders should justify why they choose their privacy

options, 2Nothing to suggest to improve]

(7) Please explain your previous answer or add more ways to

improve the process of negotiation:

(8) What do you think of the [Owner / Bystander] agent’s behav- iors of negotiation? [#I feel the [Owner / Bystander] was

strict, #I feel the [Owner / Bystander] was collaborative,

#Other]

(9) If you had only one choice for negotiating data practices,

what would you negotiate about? [#Collection, #Storage,

#Sharing, #Other]

(10) What types of risks could result from the collection, stor- age, and sharing of data about you when you are in other

people’s smart homes? [2Identity Theft, 2Cyberbullying,

2Cyberstalking, 2Targeted Advertisements, 2Other]

(11) Before you started the six negotiation interactions, you an- swered (Yes/No) to this question: “Would you prefer to be

given the option to negotiate the data practices of smart

home devices with their owner?” Have you changed your

mind on this question? [#Yes, #No, #I am unsure]

D DEMOGRAPHIC QUESTIONS

(1) What is your highest education level? [#Some High School,

#High School Degree, #College Degree, #Professional De- gree, #Associate’s Degree, #Medical Degree, #Graduate

Degree, #Other (please specify: ), #Prefer Not To Answer]

(2) What is your occupation?

(3) What is your age? [#18-29, #30-49, #50-65, #65+, #Prefer

Not To Answer]

(4) What is your gender? [#Female, #Male, #Non-Binary,

#Prefer to self describe: , #Prefer Not To Answer]

Table 5: Demographic breakdown of our 460 participants. The

percentage numbers not in parentheses are from our study.

We also report the U.S. average in parentheses according to

the estimates from the census data of 2020 [14].

Gender Age Education

Male 51% (48.5%) 18-29 28% (21%) High school degree 6% (47.2%)

Female 48% (51.5%) 30-49 60% (33.4%) College degree 55% (20%)

50-65 10% (25.2%)

Over 65 2% (20.4%)

Graduate degree 37% (25%)

E CODEBOOK

We present the codebook (which consists of eight sub-codebooks

for the eight subgroups of participants) from our qualitative anal- ysis (Section 4.2.5) in this subsection. Note that “other responses”

are separate codes but their percentage numbers were too low to

include.

E.1 Owner participants who budged and

reached an agreement

• 54%: bystanders’ privacy is more important than the smart

home device’s utility.

• 22%: a win-win agreement is reachable and fair.

• 12%: would like to be treated the same if they were by- standers.

• other responses.

E.2 Owner participants who did not budge but

reached an agreement

• 62%: utility is more important than bystanders’ privacy.

• 19%: my house, my rules.

Page 20 of 27

CHI ’23, April 23–28, 2023, Hamburg, Germany Alshehri, et al.

• other responses.

E.3 Owner participants who budged but did not

reach an agreement

• 47%: a win-win agreement is reachable and fair but the by- stander agent did not collaborate.

• 33%: of owner participants compromised but reached their

limits.

• other responses.

E.4 Owner participants who did not budge and

did not reach an agreement

• 55%: bystanders did not have the rights to ask for changing

devices’ settings.

• 25%: my house, my rules.

• other responses.

E.5 Bystander participants who budged and

reached an agreement

• 48%: the owners’ smart homes, their rules.

• 25%: transparency was good enough.

• 20%: still uncomfortable even with an agreement.

• other responses.

E.6 Bystander participants who did not budge

but reached an agreement

• 30%: privacy is too important to compromise.

• 24%: justifications for data collection is important.

• 22%: of bystander participants thought their friends should

honor their request.

• other responses.

E.7 Bystander participants who budged but did

not reach an agreement

• 42%: of bystander participants thought that only some op- tions were fair and reasonable.

• 29%: of bystander participants would budge further to only

necessary options.

• 8%: their privacy was more important than the smart home

devices utility.

• other responses.

E.8 Bystander participants who did not budge

and did not reach an agreement

• 52%: their privacy was more important than the smart home

devices utility.

• 22%: of bystander participants thought their friends should

honor their request.

• 10%: of bystander participants would budge to only necessary

options.

• other responses.

F SCREENSHOTS OF OUR WEB APPLICATION

FOR DIGITAL AGENTS

Figures 6 and 7 show the screenshots of the background information

page and the scenario presentation page, respectively.

Figures 8 and 9 show the screenshots of two negotiation interac- tions.

G ADDITIONAL FIGURES AND TABLES

Figures 10 and 11 show the detailed results of participants’ final

preferences in addition to what we presented in Section 4.2.2. Ta- ble 6 shows more logistic regression details in addition to what we

presented in Section 4.3.

Page 21 of 27

Negotiation Behaviors of Owners and Bystanders over Data Practices of Smart Home Devices CHI ’23, April 23–28, 2023, Hamburg, Germany

(a) The first half of the screenshot for the background information

page.

(b) The second half of the screenshot of the background informa- tion page.

Figure 6: Screenshots of the background information page that introduces participants with the terms we used in our negotiation

interaction study.

Page 24 of 27

CHI ’23, April 23–28, 2023, Hamburg, Germany Alshehri, et al.

(a) The first round of the negotiation interaction. (b) The second round of the negotiation interaction.

(c) The third round of the negotiation interaction. (d) The fourth round of the negotiation interaction.

Figure 9: Screenshots of one completed negotiation interaction between a bystander participant and an owner agent that is a

collaborative client. The data collection practice of a smart camera is being negotiated. An agreement is reached at the end of

the interaction.

Page 25 of 27

Negotiation Behaviors of Owners and Bystanders over Data Practices of Smart Home Devices CHI ’23, April 23–28, 2023, Hamburg, Germany Percentage of participants

0%

25%

50%

75%

100%

Camera collection strict friend

Camera collection collaborative friend

Camera collection strict stanger

Camera collection collaborative stanger

Camera storage strict friend

Camera storage collaborative friend

Camera storage strict stanger

Camera storage collaborative stanger

Camera sharing strict friend

Camera sharing collaborative friend

Camera sharing strict stanger

Camera sharing collaborative stanger

Speaker collection strict friend

Speaker collection collaborative friend

Speaker collection strict stanger

Speaker collection collaborative stranger

Speaker storage strict friend

Speaker storage collaborative friend

Speaker storage strict stanger

Speaker storage collaborative stanger

Speaker sharing strict friend

Speaker sharing collaborative friend

Speaker sharing strict stanger

Speaker sharing collaborative stanger

Preference 4 Preference 3 Preference 2 Preference 1

Figure 10: The detailed final preferences selected by owner participants when interacted with different bystander agents.

Page 26 of 27

CHI ’23, April 23–28, 2023, Hamburg, Germany Alshehri, et al. Percentage of participants

0%

25%

50%

75%

100%

Camera collection strict friend

Camera collection collaborative friend

Camera collection strict stanger

Camera collection collaborative stanger

Camera storage strict friend

Camera storage collaborative friend

Camera storage strict stanger

Camera storage collaborative stanger

Camera sharing strict friend

Camera sharing collaborative friend

Camera sharing strict stanger

Camera sharing collaborative stanger

Speaker collection strict friend

Speaker collection collaborative friend

Speaker collection strict stanger

Speaker collection collaborative stanger

Speaker storage strict friend

Speaker storage collaborative friend

Speaker storage strict stanger

Speaker storage collaborative stanger

Speaker sharing strict friend

Speaker sharing collaborative friend

Speaker sharing strict stanger

Speaker sharing collaborative stanger

Preference 4 Preference 3 Preference 2 Preference 1

Figure 11: The detailed final preferences selected by bystander participants when interacted with different owner agents.

Page 27 of 27

Negotiation Behaviors of Owners and Bystanders over Data Practices of Smart Home Devices CHI ’23, April 23–28, 2023, Hamburg, Germany

Table 6: The regression results for predicting agreements. Note that * indicates statistically significant results with (a < 0.05)

p-value

Factor Coefficient Std Err Z-value p-value Accuracy

(Intercept) -1.64605 1.47453 -1.116 0.2643

History of negotiation in smart homes −0.05318 0.11941 -0.445 0.6561

Preference to negotiate in smart homes 0.64 0.47 1.2 0.064

Exposure to smart homes as a bystander −0.55634 0.32427 -1.716 0.0862

Agent type 1.74537 0.34542 5.053 0.00043*

Relationship type 0.84760 0.34301 2.471 0.0135*

Accuracy for predicting an agreement on data collection practices of a smart camera 78.2%

(Intercept) -1.26794 1.37043 -0.925 0.354852

History of negotiation in smart homes −0.05318 0.11941 -0.445 0.6561

Preference to negotiate in smart homes −0.20961 0.12233 −1.713 0.086633

Exposure to smart homes as a bystander −0.31287 0.29654 −1.055 0.291386

Agent type 1.20906 0.32099 3.767 0.00016*

Relationship type 0.58852 0.31946 1.842 0.065445

Accuracy for predicting an agreement on data storage practices of a smart camera 73.91%

(Intercept) 3.20529 1.51695 2.113 0.0346*

History of negotiation in smart homes −0.04439 0.11294 −0.393 0.6943

Preference to negotiate in smart homes −0.37958 0.28517 −1.331 0.1832

Exposure to smart homes as a bystander −0.64435 0.30641 −2.103 0.0355*

Agent type 0.46541 0.31461 1.479 0.1391

Relationship type 0.31735 0.31512 1.007 0.3139

Accuracy for predicting an agreement on data sharing practices of a smart camera 54.3%

(Intercept) 0.34056 1.45560 0.234 0.8150

History of negotiation in smart homes −0.10410 0.11623 −0.896 0.3705

Preference to negotiate in smart homes −0.17739 0.28933 −0.613 0.5398

Exposure to smart homes as a bystander −0.48691 0.31206 −1.560 0.1187

Agent type 1.31676 0.32921 4.000 0.00063*

Relationship type 0.48069 0.32683 1.471 0.1413

Accuracy for predicting an agreement on data collection practices of a smart speaker 74%

(Intercept) 1.57691 1.58340 0.996 0.3193

History of negotiation in smart homes −0.31723 0.30496 −1.040 0.2982

Preference to negotiate in smart homes −0.10531 0.11904 −0.885 0.3763

Exposure to smart homes as a bystander −0.77655 0.33367 −2.327 0.0199 *

Agent type 1.60199 0.34573 4.634 0.00035*

Relationship type 0.59218 0.33933 1.745 0.0810

Accuracy for predicting an agreement on data storage practices of a smart speaker 70%

(Intercept) 4.69780 1.62596 2.889 0.00386*

History of negotiation in smart homes −0.80295 0.30584 −2.625 0.0086*

Preference to negotiate in smart homes −0.17823 0.12130 −1.469 0.14176

Exposure to smart homes as a bystander −0.73607 0.32064 −2.296 0.02170*

Agent type 0.47268 0.31792 1.487 0.13708

Relationship type −0.02350 0.31910 −0.074 0.94129

Accuracy for predicting an agreement on data sharing practices of a smart speaker 67.3%