Imagined Communities:
Awareness,Information Sharing,and Privacy on the Facebook Pre-proceedings version.Privacy Enhancing Technologies Workshop(PET),2006
Alessandro Acquisti1and Ralph Gross2
1H.John Heinz III School of Public Policy and Management
2Data Privacy Laboratory,School of Computer Science
watch out什么意思Carnegie Mellon University,Pittsburgh,PA15213
Abstract.Online social networks such as Friendster,MySpace,or the Facebook have experienced
exponential growth in membership in recent years.The networks offer attractive means for inter-
action and communication,but also rai privacy and curity concerns.In this study we survey a
reprentative sample of the members of the Facebook(a social network for colleges and high schools)
at a US academic institution,and compare the survey data to information retrieved from the net-
work itlf.We look for underlying demographic or behavioral differences between the communities
of the network’s members and non-members;we analyze the impact of privacy concerns on members’
behavior;we compare members’stated attitudes with actual behavior;and we document the changes
in behavior subquent to privacy-related information exposure.Wefind that an individual’s privacy
boredconcerns are only a weak predictor of his membership to the network.Also privacy concerned individ-mmdb
uals join the network and reveal great amounts of personal information.Some manage their privacy
concerns by trusting their ability to control the information they provide and the external access to it.
However,we alsofind evidence of members’misconceptions about the online community’s actual size
and composition,and about the visibility of members’profiles.
多伦多大学研究生
1Introduction
“Students living in the scholarship halls[of Kansas University]were written up in early February for pictures that indicated a party violating the scholarship halls alcohol policy”[1].“‘Stan Smith’(not his real name)is a sophomore at Norwich University.He is majoring in criminal justice even though he admits to shoplifting on his MySpace page”[2].“Corporations are investing in text-recognition software from vendors such as SAP and IBM to monitor blogs by employees and job candidates”[3].Although online social networks are offering novel opportunities for interaction among their urs,they em to attract non-urs’attention particularly becau of the privacy concerns they rai.Such concerns may be well placed; however,online social networks are no longer niche phenomena:millions of people around the world,young and old,knowingly and willingly u Friendster,,LinkedIn,and hundred other sites to communicate,find friends,dates,and jobs-and in doing so,they wittingly reveal highly personal information to friends as well as strangers.
Nobody is literally forced to join an online social network,and most networks we know about encourage, but do not force urs to reveal-for instance-their dates of birth,their cell phone numbers,or where they currently live.And yet,one cannot help but marvel at the nature,amount,and detail of the personal information some urs provide,and ponder how informed this information shar
ing is.Changing cultural trends,familiarity and confidence in digital technologies,lack of exposure or memory of egregious misus of personal data by others may all play a role in this unprecedented phenomenon of information revelation.Yet, online social networks’curity and access controls are weak by design-to leverage their value as network goods and enhance their growth by making registration,access,and sharing of information uncomplicated. At the same time,the costs of mining and storing data continue to decline.Combined,the two features imply that information provided even on ostensibly private social networks is,effectively,public data,that could exist for as long as anybody has an incentive to maintain it.Many entities-from marketers to employers to national and foreign curity agencies-may have tho incentives.
In this paper we combine survey analysis and data mining to study one such network,catered to college and high school communities:the Facebook(FB).We survey a reprentative sample of FB members at a US campus.We study their privacy concerns,their usage of FB,their attitudes towards it as well as their awareness of the nature of its community and the visibility of their own profiles.In particular,we look for underlying demographic or behavioral differences between the communities of the network’s members and
non-members;we analyze the impact of privacy concerns on members’behavior;we compare membe
rs’stated attitudes with actual behavior;and we document the change in behavior subquent information exposure: who us the Facebook?Why?Are there significant differences between urs and non-urs?Why do people reveal more or less personal information?How well do they know the workings of the network?
Our study is bad on a survey instrument,but is complemented by analysis of data mined from the network before and after the survey was administered.We show that there are significant demographic differences between FB member and non-members;that although FB members express,in general,significant concern about their privacy,they are not particularly concerned for their privacy on FB;that a minority yet significant share of the FB population at the Campus we surveyed is unaware of the actual exposure and visibility of the information they publish on FB;and we document that priming about FB’s information practices can alter some of its members’behavior.
The rest of the paper is organized as follows.In Section2we discuss the evolution of online social networks and FB in particular.In Section3we highlight the methods of our analysis.In Section4we prent our results.In Section5we compare survey results to network data.
2Online Social Networks
At the most basic level,an online social network is an Internet community where individuals interact,often through profiles that(re)prent their public persona(and their networks of connections)to others.Although the concept of computer-bad communities dates back to the early days of computer networks,only after the advent of the commercial Internet did such communities meet public success.Following experience in1997,hundreds of social networks spurred online(e[4]for an extended discussion),sometimes growing very rapidly,thereby attracting the attention of both media and academia.In particular,[5],[6], and[7]have taken ethnographic and sociological approaches to the study of online lf-reprentation;[8]have focud on the value of online social networks as recommender systems;[4]have discusd information sharing and privacy on online social networks,using FB as a ca study;[9]have demonstrated how information revealed in social networks can be exploited for“social”phishing;[10]has studied identity-sharing behavior in online social networks.
2.1The Facebook
2013考研成绩查询The Facebook is a social network catered to college and high school communities.Among online social networks,FB stands out for three reasons:its success among the college crowd;the amount and the quality of personal information urs make available on it;and the fact that,unlike other netw
orks for young urs, that information is personally identified.Accordingly,FB is of interest to rearchers in two respects:1)as a mass social phenomenon in itlf;2)as an unique window of obrvation on the privacy attitudes and the patterns of information revelation among young individuals.
FB has spread to thousands of college campus(and now also high schools)across the United States, attracting more than9million(and counting)urs.FB’s market penetration is impressive:it can draw more than80%of the undergraduate population in many colleges.The amount,quality,and value of the information provided is impressive too:not only are FB profiles most often personally and uniquely identified, but by default they show contact information(including personal address and cell phone numbers)and additional data rarely available on other networks.
FB requires a college’s email account for a participant to be admitted to the online social network of that college.As discusd in[4],this increas the expectations of validity of the personal information therein provided,as well as the perception of the online space as a clod,trusted,and trustworthy community (college-oriented social networking sites are,ostensibly,bad“on a shared real space”[11]).However,there are reasons to believe that FB networks more cloly remble imagined[12]communities(e also[4]): in most online social networks,curity,access controls,and p
rivacy are weak by design;the easier it is for people to join and tofind points of contact with other urs(by providing vast amounts of personal information,and by perusing equally vast amounts of data provided by others),the higher the utility of the network to the urs themlves,and the higher its commercial value for the network’s owners and managers.FB,unlike other online networks,offers its members very granular and powerful control on the privacy(in terms of archability and visibility)of their personal information.Yet its privacy default ttings are very permeable:at the time of writing,by default participants’profiles are archable by anybody el on the FB network,and actually readable by any member at the same college and geographical location.
2
In addition,external access to a college FB ,by non-students/faculty/staff/alumni,or by non-college-affiliated individuals)is so easy[4],that the network is effectively an open community,and its data effectively public.
高口官网3Methods
Our study aims at casting a light on the patterns and motivations of information revelation of college students on FB.It is bad on a survey instrument administered to a sample of students at a North A
merican college Institution,complemented by analysis of data mined from the FB network community of that Institution.
3.1Recruiting Methods
Participants to the survey were recruited in three ways:through a list of subjects interested in participating in experimental studies maintained at the Institution where the study took place(and containing around4,000 subscribed subjects);through an electronic billboard dedicated to experiments and studies,with an unknown (to us)number of campus community subscribers;and throughfliers posted around campus.The above two lists are populated in majority by undergraduate students.The emails and thefliers sought participants to a survey on“online networks,”and offered a compensation of$6,plus the possibility to win a$100prize in a lottery among all participants.
Around7,000profiles were mined from the FB network of the same Institution.In order to automate access to the Facebook we ud Perl scripts[13],specifically the Perl LWP library[14],which is designed for downloading and parsing HTML pages.The data was mined before and after the survey was administered.
3.2Survey Design
The survey questionnaire contained around forty questions:an initial t of screening questions;a connt ction;a t of calibration questions(to ascertain the respondents’privacy attitudes without priming them on the subject of our study:privacy questions were intersperd with questions on topics such as economic policy,the threat of terrorism,same-x marriage,and so on);and,next,FB-related questions.Specifically, we asked respondents to answer questions about their usage,their knowledge,and their attitudes towards FB.Finally,the survey contained a t of demographics questions.
Only respondents currently affiliated with the Institution were allowed to take the survey(students,staff, and faculty).Respondents received somewhat different questions depending on whether they were current FB members,previous members,or never members.The survey is available on request from the authors.
3.3Statistical Analysis
We analyzed survey results using STATA8.0on Windows and other ad hoc scripts.The study was performed on dichotomous,categorical(especially7-point Likert scales),and continuous variables.We performed a number of different tests-including Pearson product-moment correlations to study relatio
ns between con-tinuous variables,χ2and t tests to study categorical variables and means,Wilcoxon signed-rank test and Wilcoxon/Mann-Whitey test for non-normal distributions,as well as logit,probit,and linear multivariate regressions.
watch your back4Results
A total of506respondents accesd the survey.One-hundred-eleven(21.9%)were not currently affiliated with the college Institution where we conducted our study,or did not have a email address within that Institution’s domain.They were not allowed to take the rest of the survey.A parate t of32(8.2%) participants had taken part in a previous pilot survey and were also not allowed to take the survey.Of the remaining respondents,318subjects actually completed the initial calibration questions.Out of this t,278 (87.4%)had heard about FB,40had not.In this group,225(70.8%)had a profile on FB,85(26.7%)never had one,and8(2.5%)had an account but deactivated it.Within tho three groups,respectively209,81, and7participants completed the whole survey.We focus our analysis on that t-from which we further removed3obrvations from the non-members group,since we had reasons to believe that the respons had been created by the same individual.This left us with a total of294respondents.
3
4.1Participants
In absolute terms,we had exactly the same number of male participants taking the survey as female partic-ipants,147.We classified participants depending on whether they were current members of the FB campus network(we will refer to them as“members”),never members,or no longer members(we will often refer to the last two groups collectively as“non-members”).
A slight majority of F
秋装搭配技巧B members in our sample(52.63%)are male.Our sample slightly over-reprents females when compared to the target FB population,who data we mined from the network(male reprent 63.04%of the Institution’s FB network,but it is important to note that the gender distribution at the Institution is itlf similarly skewed).However,we know from the information mined from the network that 79.6%of all the Institution’s undergraduate males are on the FB(91.92%of our sample of male undergrads are FB members)and75.5%of all the Institution’s undergraduate females are on the FB(94.94%of our sample of female undergrads are FB members).In other words(and expectably),our total sample of respondents slightly over-reprents FB members.
The gender distribution of our sample is reverd among respondents who were never or are no lon
ger members of FB:56.46%are female.This gender difference between current members and current non-members is not statistically significant(Pearsonχ2(1)=2.0025,P r=0.157).However,when we test usage by contrasting actual FB urs and non-members plus members who claim to“I never login/u”their profile,the gender difference becomes more radical(54.19%of urs are male,but only40.66%of non urs are)and significant(Pearsonχ2(1)=4.5995P r=0.032).See Figure1for the gender distribution in the three FB member groups.
Fig.1.Gender distribution of the survey participants for the three FB member groups.
There is no significant difference among the distributions of undergraduate versus graduate students in our sample and in the overall FB population.
Overall,sixty-four percent of our respondents(64.29%)are undergraduate students;25.17%are graduate students;1.36%are faculty;and9.18%are staff.We did not consider alumni in our study.This distribution slightly oversamples undergraduate students when compared to the actual Institution’s population(total student population in2005/06:10,017.Undergraduate students:54.8%).This was expected,considering the available recruiting tools and the comparatively higher propensity of undergraduate students to take paid surveys and experiments.However,when checking for current FB membership in our sample,wefind that undergraduates dominate the picture(84.21%),followed by graduate students(14.35%)and staff(1.44%). The numbers are comparable to the distribution of the target population discud in[4]when correcting for alumni(91.21%were undergraduate students on the Facebook network).
Again,the distribution of non-members is reverd:graduate students dominate(51.76%),followed by staff(28.24%).The distributions of ur types(undergraduates,graduates,staff,or faculty)by FB member-ship status are significantly diver(Pearsonχ2(3)=135.3337P r=0.000).See Figure2for a breakdown of the academic status of survey participants across the three FB groups.
4
Fig.2.Distribution of survey participant status for FB members,non-members and people who never had a FB account.
Unsurprisingly,age is a strong predictor of membership(e Figure3).Non-members tend to be older (a mean of30years versus a mean of21)but their age is also more broadly distributed(sd8.840476vs. sd2.08514).The difference in the mean age by membership is strongly significant(t=-14.6175,P r<t= 0.0000).
Fig.3.Distribution of age for FB members and non-members.
4.2Privacy Attitudes
Age and student status are correlated with FB membership-but what el is?Well,of cour,having heard of the network is a precondition for membership.Thirty-four participants had never heard of the FB-nearly half of the staffthat took our survey,a little less than23%of the graduate students,and a negligible portion of the undergraduate students(1.59%).
However,together with age and student status(with the two obviously being highly correlated),an-other relevant distinction between members and non-members may ari from privacy attitudes and privacy concerns.
Before we asked questions about FB,our survey ascertained the privacy attitudes of participants with a battery of questions modelled after the Alan Westin’s studies[15],with a number of modifications.In particular,in order not to prime the subjects,questions about privacy attitudes were intersperd with questions about attitudes towards economic policy and the state of the economy,social issues such as same-x marriage,or curity questions related to the fear of terrorism.In addition,while all instruments asked the respondent to rank agreement,concern,worries,or importance on a7-point Lik
ert scale,the questions ranged from general ,“How important do you consider the following issues in the public debate?”), to more and more specifi,“How do you personally value the importance of the following issues for your own life on a day-to-day basis?”),and personal ,“Specifically,how worried would you be if”
priority是什么意思
[a certain scenario took place]).
5
Fig.4.Box-plots of age distribution for different membership status “Privacy policy”was on average co
nsidered a highly important issue in the public debate by our respon-dents(mean on the7-point Likert scale:5.411,where1is“Not important at all”and7is“very important”; sd:1.393795).In fact,it was regarded a more important issue in the public debate than the threat of ter-rorism(t=2.4534,P r>t=0.0074;the statistical significance of the perceived superiority was confirmed by a Wilcoxon signed-rank test:z=2.184P r>|z|=0.0290)and same x marriage(t=10.5089,P r>t= 0.0000;Wilcoxon signed-rank test:z=9.103P r>|z|=0.0000);but less important than education policy (mean:5.93;sd:1.16)or economic policy(mean:5.79;sd:1.21).The slightly larger mean valuation of the importance of privacy policy over environmental policy was not significant.(The results are comparable to tho found in previous studies,such as[16].)
The same ranking of values(and comparably statistically significant differences)was found when asking for“How do you personally value the importance of the following issues for your own life on a day-to-day basis?”The mean value for the importance of privacy policy was5.09.For all categories,subjects assigned slightly(but statistically significantly)more importance to the issue in the public debate than in their own life on a day-to-day basis(in the privacy policy ca,a Wilcoxon signed-rank test returns z=3.62P r>|z| =0.0003when checking the higher valuation of the issue in the public debate).
高中数学教科书
Similar results were also found when asking for the respondents’concern with a number of issues directly relevant to them:the state of the economy where they live,threats to their personal privacy,the threat of terrorism,the risks of climate change and global warming.Respondents were more concerned(with statistically significant differences)about threats to their personal privacy than about terrorism or global warming,but less concerned than about the state of the economy.
Finally,we asked how worried respondents would be if a number of specific events took place in their lives.The highest level of concern was registered for“A stranger knew where you live and the location and schedule of the class you take”(mean of5.78,with45.58%of respondents choosing the7th point in the Likert scale,“very worried,”and more than81%lecting Likert points above4).This was followed by“Five years from now,complete strangers would be able tofind out easily your xual orientation,the name of your current partner,and your current political views”(mean of5.55,with36.39%-the relative majority-choosing the7th point in the Likert scale,and more than78%with points above4),followed,in order,by the‘global warming’scenario(“The United States rejected all new initiatives to control climate change and reduce global warming”),the curity scenario(“It was very easy for foreign nationals to cross the borders undetected”),the‘contacts’scenario(“A friend of a friend that you do not even know knew your name,
your email,your home phone number,and your instant messaging nickname”),and the‘same-x’scenario (“Two people of the same x were allowed to marry in your State”).
Privacy Attitudes and Membership Status Privacy concerns are not equally distributed across FB members and non-members populations:a two-sided t test that the mean Likert value for the“importance”of privacy policy is higher for non-members(5.67in the non-members group,5.30in the members group)
6
