社交媒体的福利影响（英文）-facebook-201901.pdf
with deactivation by regularly checking participants' public profile pages. We measured a suite of outcomes using text messages, surveys, emails, dircct mcasurcmcnt of activity on Faccbook and Twitter, and administrative records on voting and electoral contributions. Less than two percent of the sample failed to complete the endline survey, and the Treatment groups compliance with deactivation exceeded 90 percent Our study offers the largest-scale experimental evidence available to date on the way facebook ffects a range of individual and social welfare measures. We evaluate the extent to which time on Facebook substitutes for alternative online and offline activities, with particular attention to crowd out of news consumption and face-to-face social interactions. We study Facebook's broader political externalities via measures of news knowledge, awareness of misinformation, political engagement and political polarization. We study the iimpact on individual utility via measures of sub jective well being, captured through both surveys and text, messages. Finally, we analyze the extent, to which behavioral forces like addiction and misprediction may cause sub-optimal consumption choices, by looking at how usage and valuation of Facebook change after the experiment Our first sct of results focuscs on substitution patterns. A kcy mcchanism for cffccts on indi- vidual well-being would be if social media use crowds out face-to-face social interactions and thus deepens loneliness and depression(T wenge 2017). A key mechanism for political externalities would be if social media crowds out consumption of higher-quality news and information sources. We find evidence consistent with the first of these but not the second. Deactivating Facebook freed up 60 minutes per day for the average person in our Treatment group. The Treatment group actually spent less time on both non-Facebook social media and other online activities, while devoting more time to a range of offline activities such as watching television alone and spending time with friends and family. The Treatment group did not change its consumption of any other online or offline news sources and reported spending 15 percent less time consuming news Our second set of results focuses on political externalities, proxied by news knowledge, political engagement, and political polarization. Consistent with the reported reduction in news consump tion, we find that Facebook deactivation significantly reduced news knowledge and attention to politics. The Trcatmcnt group was less likely to say thcy follow news about politics or the pres- ident, and less able to correctly answer factual questions about recent news events. Our overall index of news knowledge fell by 0. 19 standard deviations. There is no detectable effect on political engagement, as measured by voter turnout in the midterm election and the likelihood of clicking on email links to support political causes. Deactivation significantly reduced polarization of views on policy issues and a measure of exposure to polarizing news. Deactivation did not statistically significantly reduce affective polarization(i.e. negative feelings about the other political party )or polarization in factual beliefs about current events, although the coefficient estimates also point in that direction. Our overall index of political polarization fell by 0. 16 standard deviations. As a point of comparison, prior work has found Chat a diferent index of polilical polarization rose by 0.38 standard deviations between 1996 and 2018(Boxell 2018) Our third sct of results looks at subjective wcll-bcing. Dcactivation causcd small but significant improvements in well-being, and in particular on self-reported happiness, life satisfaction, depres sion, and anxiety. Effects on subjective well-being as measured by responses to brief daily text messages are positive but not significant. Our overall index of subjective well-being improved by 0.09 standard deviations. As a point of comparison. this is about 25-40 percent of the effect of psychological interventions including self-help therapy, group training and individual therapy, as reported in a meta-analysis by Bolier et al.(2013 ). These results are consistent with prior studies suggesting that Facebook may have adverse effects on mental health. However, we also show that the magnitudes of our causal effects are far smaller than those we would have estimated using the correlational approach of luch prior literature. We find litlle evidence to support Llie hy pothesis suggested by prior work that Facebook might be more beneficial for " active users-for example users who regularly comment on pictures and posts from friends and family instead of just scrolling through their news feeds Our fourth sct of rosults considcrs whether deactivation affected pcoplc's demand for Faccbook after the study was over, as well as their opinions about Facebook's role in society. as the experi- ment ended, participants reported planning to use Facebook much less in the future. Several weeks later, the Treatment group's reported usage of the Facebook mobile app was about 12 minutes(23 percent) lower than in Control. The Treatment group was more likely to click on a post-experiment email providing information about tools to limit social media usage, and five percent of the treat- ment group still had their accounts deactivated nine weeks after the experiment ended. Our overall index of post-experiment Facebook use is 0. 61 standard deviations lower in Treatment than in Con trol. In response to open-answer questions several weeks after the experiment ended, the Treatment roup was inore likely to report that they were using Facebook less, had uninstalled the Facebook app from their phones, and were using the platform more judiciously. Reduced post-experiment use aligns with our finding that deactivation improved subjective well-being, and it is also consistent with the hypotheses that Facebook is habit forming in the sense of Becker and Murphy(1988)or that pcoplc lcarncd that thcy cnjoy lifc without Facebook morc than thcy had anticipated Deactivation caused people to appreciate Facebooks both positive and negative impacts on their lives. Consistent with our results on news knowledge, the Treatment group was more likely to agree that Facebook helps people to follow the news. The great majority of the Treatment group agreed that deactivation was good for them, but they were also more likely to think that people would miss Facebook if they used it less. In free response questions, the Treatment group wrote more text about how Facebook has both positive and negative impacts on their lives. The opposing effects on these specific metrics cancel out, so our overall index of opinions about Facebook is unaffected 3 Corrclation studics on active vs. passive Faccbook usc includc Burkc, Marlow, and Lcnto(2010), Burkc, Kraut and Marlow(2011), Burke and Kraut(2014), and Krasnova et al(2013), and randomized experiments include Deters and Mehl(2012)and Verduyn et al.(2015 Our work also speaks to an adjacent set of questions around how to measure the economic gains from frcc online scrviccs such as scarch and media. In standard modcls with consumers who correctly optimize their allocation of time and monev, researchers can approximate the consumer surplus from these services by measuring time use or monetary valuations, as in Brynjolfsson and Oh (2012), Brynjolfsson, Eggers, and Gannamaneni(2018). Corrigan et al.(2018), and others. But if users do not understand the ways in which social media could be addictive or make them unhappy. these standard approaches could overstate consumer surplus gains. Sagioglu and greitemeyer (2014) provide suggestive evidence: while their participants predicted that spending 20 minute Facebook would make them feel better, it actually caused them to feel worse To quantify the possibility that a period of deactivation might help the Treatment group to understand ways in which their use had Inade CheIn unhappy, we elicited WTA al three separate points, using incentive-compatible Becker-De groot -Marschak(1964,BDM)mechanisms. First, on October llth, we elicited willingness-to-accept to deactivate Facebook between October 12th and November &th, which we loosely call"month 1. We immediately told participants the amount that thcy had bccn offered to deactivate($102 for the Trcatment group, SO for Control), and thus whether they were expected to deactivate over that period. We then immediately elicited WTA to deactivate Facebook for the next four weeks after November Sth, which we call"month 2. When November Sth arrived, we then re-elicited WTa to deactivate in month 2. The Treatment group,s change in valuation for month 2 reflects a time effect plus the unanticipated effect of spending time off of Facebook. The Control group's parallel valuation change reflects only a time effect. Thus the difference between how Treatment vs. Control change their WTAs for deactivation in month 2 reflects projection bias, learning, and similar unanticipated experience effects, which we collectively ll“ misprediction. Afler weighting our sample to malch the average US Facebook user On observables, the median and mean willingness-to-accept to deactivate Facebook for the initial four weeks were $100 and $180 respectively. These valuations are larger than most estimates in related work by Brynjolfsson Eggers, and Gannamaneni(2018), Corrigan et al.(2018), Mosquera et al. (2018), and Sunstein (2019). Aggregated across an estimated 172 million US Faccbook uscrs, this could bc intcrprctcd to mean that Facebook generates several hundred billion dollars of consumer surplus per year in the us alone. Consistent with our other results that deactivation reduced demand for Facebook deactivation caused month 2 wta to drop by 13 percent, al though this may be an upper bound on misprediction for reasons we discuss later. While such misprediction may be substantial in absolute See, for example, Brynjolfsson and Saunders(2009), Byrne, Fernald, and Reinsdorf(2016), Nakamura, Samuels and Soloveichik(2016), Brynjolfsson, Rock, and Syverson(2018), and Syverson(2017) SOur measurement of state dependence and misprediction connects us to the literature on habit formation and projection bias, including Acland and Levy(2015), Becker and Murphy(1988), Becker, Grossman, and Murphy 1991), Busse et al.(2015), Charness and Gneezy(2009), Conlin, O'Donoghue, and Vogelsang(2007), Fujiwara, Meng, and Vogl(2016), Gruber and Koszcgi(2001), Hussam ct al.(2016), Locwcnstcin, O'Donoghuc, and Rabin (2003), and Simonsohn(2010) terms, it would not reverse the conclusion that Facebook generates enormous Hows of consumer surplus Our results should be interpreted with caution, for several reasons. First, effects could differ with the duration or scale of deactivation. A longer period without Facebook might have less impact on news knowledge as people find alternative news sources, and either more or less impact on subjective well-being. Furthermore, a larger-scale experiment in which a greater share of the population deactivated could have a different impact due to network effects and equilibrium adjustments Second, our sample is not fully representative. Our participants are relatively young, well-educated and left-leaning compared to the average Facebook user, and we included only people who reported using Facebook more than 15 minutes per day. In addition, although we went as far as possible to avoid telegraphing the experimental design and research questiOns, deacTivatiOn could have different effects on the average Facebook user than on the type of person who was willing to participate in our experiment. Third, many of our outcome variables are self-reported, adding scope for both measurement error and experimenter demand effects. This latter concern is mitigated somewhat by the fact that thc non-sclf-rcportcd outcomos we mcasurc(c g, post-cxpcrimcnt Faccbook usc) paint a similar picture to the survey responses The causal impacts of social media have been of great interest to researchers in economics psychology, and other fields. We are aware of 12 existing randomized impact evaluations of face book 6 The most closely related is the important paper by Mosquera et al. (2018), which was made public the month before ours. They also use Facebook deactivation to study news knowledge and well being, finding results broadly consistent with those reported here. Appendix Table Al details these experiments in comparison to ours. Our deactivation period is substantially longer and our sample size an order of magnitude larger than most prior work, including Mosquera et al.(2018) We neasure impacts on a relalively conmprehensive range of outcomes, and we are the only one of these randomized trials to have submitted a pre-analysis plan. Given the effect sizes and residual variance in our sample, we would have been unlikely to have sufficient power to detect any effects if limited to the sample sizes in previous experiments Scctions 2 and 3 dctail thc cxpcrimcntal design and cmpirical strategy. Scction 4 presents the impact evaluation, and Section 5 presents measurements of the consumer surplus generated b Facebook These studies sit within a broader media effects literature that uses experimental and quasi-experimental methods to quantify the effects of media technologies such as television, media providers such as Fox News, and content such as political advertising(Bartels 1993; Besley and Burgess 2001; Della vigna and Kaplan 2007: Enikolopov, Petrova and Zhuravskaya 2011; Gentzkow 2006; Gerber and Green 2000; Gerber et al. 2011; Gerber, Karlan, and Bergan 2009; Huber and Arceneaux 2007; Martin and Yurukoglu 2017; Olken 2009; and Spenkuch and Toniatti 2016): fc reviews, see Della vigla and Gentzkow(2010), Napoli(2014). StrOmberg(2015), Enikolopov and Petrova(2015),and Della vigna and La Ferrara(2015 2 Experimental Design 2.1 Experiment Overview Figure 1 summarizes our experimental design and timeline. We timed the experiment so that the main period of Facebook deactivation would end shortly after the 2018 US midterm elections, which took place on November 6th. The experiment has eight parts: recruitment, pre-screen, baseline survey, midline survey, endline survey, post-endline survey, post-endline emails, and daily text messages. Table 1 presents the sample sizes at each part of the experiment Between September 24th and October 3rd, we recruited participants using Facebook ads. Our ad said, "Participate in an online research study about internet browsing and earn an easy $30 in electronic gift cards. Appendix Figure Al presents the ad. To minimize sample selection bias, the ad did not hint at our research questions or suggest that the study was related to social media or Facebook deactivation. We targeted the ads by demographic cells in an attempt to gather an initial sample that was approximately representative of Facebook users on gender, age, college completion and polilical ideology. 1, 690,076 unique users were shown the ad, of whoM 30,064 clicked on it This 1.8 percent click-through rate is about twice the average click-through rate on facebook ads across all industries(Irvine 2018) Clicking on the ad took the participant to a brief pre-screen survey, which included several background demographic qucstions and thc consent form. The prc-scrccn, basclinc, midline, and endline surveys were hosted on the site stanforduniversity qualtrics. com. 14, 324 people passed the pre-screen, by reporting being a US resident born between the years 1900 and 2000 who uses Facebook more than 15 minutes and no more than 600 minutes per day. Of those people, 5, 974 consented to participate in the study After completing the consent form. participants began the baseline survey. The baseline recorded email addresses, additional demographics, and a range of outcome variables. We also asked for each participant's name, zip code, Twitter handle, and phone number order for us to send you text messages during the study"), as well as the URL of their Facebook profile page (which we would use"solely to observe whether your Facebook account is active") To minimize selective attrition, we asked all participants regardless of subsequent treatment status to deactivate their Facebook accounts for 24 hours following the midline and endline surveys In the baseline, we informed people that " As part of this study, we will ask you to deactivate you Facebook account twice for a period of 24 hours. You will keep your access to Facebook Messenger If you deactivate, you can choose to come back whenever you want with your content and friends network unchanged, and asked, "Are you willing to deactivate your Facebook account twice for 24 hours(once after Survey 2 and once after Survey 3 ? In all, 3, 234 people finished the baseline survey and responded that they were willing to de activate. Of those, 183 were dropped from the experimental sample because of invalid data(for example, invalid Facebook profile URLS) or low-quality baseline responses(for example, discrepan cics bctwccn avcrage daily Faccbook usage rcported in thc prc-scrccn vS. basclinc survcy, completing the survey in less than ten minutes, no text in short-answer boxes, and other patterns suggesting careless responses). The remaining 3,051 participants had valid baseline data, were included i In our stratified randomization. and were invited to take the midline survey On October 1lth, we sent email invitations to the midline survey. In that survey, we first asked people to deactivate their Facebook account for 24 hours, and guided them through the process. Second, we used a Becker-Degroot-Marschak(BDM) mechanism to elicit willingness-to accept (WTA)to stay deactivated for four weeks rather than 24 hours. We then revealed the BDM price offer. Participants whose WTa was strictly less than the price draw were informed that they should deactivate for the full four weeks aller midline. Third, we reminded people that we would again ask them to deactivate for 24 hours after the endline survey, and used a second bdm mechanism to elicit Wta to stay deactivated for the four weeks after endline instead of just 24 hours. We informed people that We will check continuously whether your account is deactivated for the cntirc 24 hours/ 4 wicks in which it is supposed to bc by pinging the URL associatcd with your profile. For expositional purposes, we will loosely refer to the four weeks after midline as month1,” and the four weeks after endline as“ month2.” On November &th, two days after the midterm election, we sent an email invitation to the endline survey. The endline survey first measured the same outcome variables as the baseline survey. All questions were identical, with the exception of cases discussed in Section 2.3 below, such as using updated news knowledge questions and rephrasing questions about the midterm election to be in next 24 hours, and again elicited WTa to stay deactivated for the next four weeks (i.e, monther o the past tense. We then asked all participants to again deactivate their Facebook accounts for the instead of the next 24 hours. Participants were told, " With a 50%o chance we will require you lo abide by the decision you made 4 weeks ago; with 50% chance we will ignore the decision you made 4 weeks ago and we will require you to abide by the decision you make today. We gathered data from two post-endline emails. On November 20th, we sent an email with links to information on ways to limit smartphone social media usc, and on Novcmbcr 25th, wc scnt an email with links to donate to, volunteer for, or sign petitions related to political causes. Clicks on these emails provide additional non- self-reported measures of interest in reducing social media use and political engagement. Appendix Figures A2 and A3 present the two emails TThe survey explained, " The computer has randomly generated an anount of money to offer you to deactivate your Facebook account for the next 4 weeks. Before we tell you what the offer is, we will ask you the smallest offer you would be willing to accept. If the offer the computer generated is above the amount you give, we will ask you deactivate for 4 nt if you do. If the offe t. we will ask you to deactivate. We then asked several comprehension questions to make sure that participants understood the mechanism. We did not tell participants the distribution or support of the offer prices, both because we did not want to artificially truncate the distribution of elicited WTA and bccause prior studics have found that providing information on the bounds of the offer price distribution can affect BDM valuations(Bohm, Linden, and Sonnegard 1997; Mazar, Koszegi, and Ariely 2014 On December 3rd, we invited participants to a short post-endline survey in which we asked how many minutes pcr day they had uscd thc Facebook app on thcir smartphones in the past seven days We asked participants with iphones to report the facebook app time reported by their phones Settings app, and we asked other participants to estimate We also asked several open-answer questions, such as"How has the way you use Facebook changed, if at all, since participating in this tudy For the approximately six weeks between baseline and endline, we sent daily text message surveys to measure several aspects of subjective well-being in real time rather than retrospectively We rotated three types of questions, measuring happiness, the primary emotion felt over the past ten minutes, and loneliness. Appendix Figure A4 presents the three questions We verified deactivation by checking each participants Facebook profile page URL regularly at random times. While a user can limit how much content other people can see in their profiles they cannot hide their public profile page, and the public profile trl returns a valid response if and only if their account is active. This is thus our measure of deactivation. For all participant we verified deactivation approximately oncc pcr day for the soven days bcforc midline and al days between endline and late January 2019. Between midline and endline, we verified deactivation approximately four times per day for the participants who had been randomly assigned to deactivate (i.e., the Treatment group) and once every four days for participants who had not been assigned to deactivate. During the post-midline and post-endline 24-hour deactivation periods. we generally verified deactivation within about six hours of when each participant completed the survey. If participants were not deactivated when they were supposed to be, our program immediately sent an automated email informing them that they should again deactivate as soon as possible along with a survey asking them to explain why they were not deactivated. We discuss the way we handle imperfect COmpliance in our eMpirical analysis in Section 3 below All participants received $5 per completed survey, paid via gift card immediately upon com pletion. All participants were told that they would receive a $15 "completion payment?" if they completed all surveys, responded to 75 percent of text messages, kept their accounts deactivated for the 24 hours aftcr midline and cndlinc, and, if the deactivation offer price was abovc thcir reported WTA, kept their accounts deactivated for the full period between midline and endline. The latter re- quirement(making the completion payment contingent on complying with the BDMs deactivation assignment)makes it a strictly dominant(instead of weakly dominant) strategy to truthfully report. By default, Facebook profile URLs end in a unique number which is the numeric ID for that person in the Facebook system. Users can update their default. URI to be something customized, and they can change their customized URL as often as they want. In the baseline survey, participants reported their profile URLs, which could have been either the default or customized version. Shortly after the baseline survey, we checked if each participant's Facebook profile URL was valid by pinging it and looking in the page source for the string containing the person's numcric ID. If the numcric ID cxisted, we knew that the URL was valid. After that point, we used participants numeric IDs to construct their default numeric URLs which allowed us to correctly measure deactivation even if they changed their customized uRL valuations in the BDM. These payments were in addition to the $102 that treatment participants rcccivcd in cxchange for deactivation 2.2 Randomization We use the bdm mechanism described above to randomly assign participants to facebook deactiva- tion. Figure 1 illustrates the randomization. Participants with valid baseline data were randomized into thrcc groups that dctcrmincd thc BDM offer price p for deactivation in month 1(ic, the wccks between midline and endline): p=$102(approximately 35 percent of the sample), p= $0(ap- proximately 65 percent), and p drawn from a uniform distribution on [ $0, $170(approximately 0.2 percent ). We balanced the p=$102 and p= so group assignments within 48 strata defined by age, average daily Facebook use, heavy vs. light news use(those who get news from Facebook fairly often or very often vs. never, hardly ever, or sometimes). active vs. passive Facebook use and Democrat, Republican or independent party affiliation The effects of Facebook deactivation in month 1 are identified in the sample of participants who were allocated to p=$102 or p-o and were willing to accept less than $102 to deactivate in Inonth 1. We call Chis the "impact evaluation sample. Within the impact evaluauion sample, we call p=$102the“ Treatment” gToUp,andp=$0the“ ontrol"group For deactivation in month 2(i. e, the four weeks after endline), 0. 2 percent of participants were randomly selected to a BDM offer price drawn randomly from pE[0, 170, while the remaining 99. 8 percent rcccivcd offer =0. We balanced this month 2 offer price p bctwccn the month 1 offer price groups, so two participants who were offered p=$102 and four participants who were offered p=S0 were assigned to positive month 2 offers P'E0, 17 This approach allows us to maintain incentive compatibility in the BDM mechanism, have balance between Treatment and Control groups, and use a simple regression to estimate treatment effects of post-midline deactivation 2.3 Outcome variables For the impact evaluation, we consider the outcome variables in the nine families described below Appendix b. 1 presents survey question text and descriptive statistics for each outcome variable and moderator, grouped by family. We also construct indices that combine the outcome variables within each family, weighting by the inverse of the covariance between variables at endline, as described in Anderson(2008 ). In constructing these indices, we orient the variables so that more As discussed above, we did not inform participants of the BdM offer price distribution. Thus, more precisely truthfully reporting valuations is a strictly dominant strategy only within the support of the offer price distribution that participants expected us to use $170 was chosen because it was the InaxinuIn that we could pay participants without requiring tax-related paperwork