1687_sri-understanding-society-july-2014

July 2014

Results from our Global Trends Survey,a 20 country study of over 16,000 online adultsWho do we trust?Trust in different institutions varies considerably – social media sites andmedia companies are only slightly more trusted than foreign governments.Public sector healthcare providersBanksPrivate sector healthcare providersYour national governmentSupermarketsCredit card companiesInsurance companiesTelecommunications companiesSocial media sitesMedia companiesForeign governmentsTo what extent, if at all, do you personally trust the following to use the information they have about you in the right way?Social media tends to be trustedmuch less in developed countriesKnowing moredoesn’tincrease trust45%45%38%33%32%31%31%25%20%19%15%India48AustraliaRussia30Poland30Argentina12 12 10 9 8 726Turkey24Brazil22GreatBritain Germany Belguim France SwedenTrust social mediawith informationknow what information isheld on them19don’t know what informationis held on them23Trust governmentwith informationknow what information isheld on them34don’t know what informationis held on them32

Scan the QR code belowto download an electronicversion of this report.Editorial:Daniel CameronBobby DuffyAalia KhanGideon SkinnerInformationwww.ipsos-mori.comaalia.khan@ipsos.comFollowwww.twitter.com/IpsosMORI01.12.22.Foreword by Bobby Duffy02.Data privacyPolling shows that peopleare either unconcerned,pragmatists or privacyfundamentalists. Wheredo we draw the line?Bobby Duffy08.No place to hideAn interview with theEditor-in-Chief of TheGuardian, Alan RusbridgerBen PagePrivacy or security:a false choice?European citizens' perceptionsof privacy, personal data,surveillance and security.Carolina Haita andDaniel Cameron17.Is people-powereddata possible?The future of citizen controlGeoff Mulgan19.The trouble with numbersAn interview with Nate SilverThe rise of theforecaster-punditBig data and data aggregationin the 2012 US presidentialelection.Clifford Young and Elisa Bernd27.The data rich societyBig data is a new way ofworking with numbers, but hasbeen challenged for being toovague a term. Can big datamake big mistakes?Henri WallardBen Page

Ipsos MORI - Understanding Society July 2014Data privacy:Where do peopledraw the line?Bobby Duffy,LondonData privacy feels like a very modernproblem. The media is full of storiesabout government surveillance and whathappens to the data we all provide toprivate companies. But of course it’s farfrom a new concern, and closely relateddiscussions on threats to data privacygo back decades, at least since moderndatabases were developed.Yet superlatives abound in currentdiscussions. Various think tanks,¹consultants and trade groups² have toldus that the latent demand for privacyhas never been greater, that privacy isthe next key consumer rights issue³,and that our personal data are morevaluable than gold or oil.Our surveys also tell us the greatestsin a company can commit in the eyesof the public is losing their personaldata, followed by selling that data,even if it’s anonymised – much worsethan exploiting foreign workers or evencharging more than competitors.There is no onepublic opinionon data privacyOur new 20 country study 4 releasedat an event with King’s College Londonthis year adds more international depthto this picture. It confirms some of whatwe knew, and in particular that there isno one public opinion on data privacy –in three separate ways.2.

Data privacy – where do people draw the line?First, there is just variety in howconcerned different people are. Indeedthere is remarkable consistency inhow populations segment across timeand in different developed nations: wetend to find around 10% are 'privacyunconcerned', 60% are pragmatists,where concern depends on thecircumstances, and 30% are 'privacyfundamentalists'. 5But, second, this gives a false senseof certainty in opinion. In practice,it’s not just the pragmatists whochange their views depending on thecircumstances – experiments show thatlarge proportions of the two groupsat either end of the spectrum can beshifted, depending on what they areoffered or how they are reassured. 6And, third, stated concernabout data privacy and how peopleactually behave are barely noddingacquaintances. We can see the massivedisconnect between what people sayand how we know they act in this newsurvey. For example, nearly half ofpeople across the 20 countries say theyare willing to pay for increased levelsof privacy for their data. But at sametime, in the same survey, only a quarterof the same people say they have takenbasic steps to increase the privacysettings on their browser. This meansthat three quarters of those who saythey would pay for additional privacyhaven’t changed a simple setting ontheir computer.Figure ONE.Q: To what extent do you agree or disagree...'I am willing to pay extra for a service or product to keep my details private'?BrazilIndiaChinaS AfricaTurkeyS KoreaPolandUSItalyCanadaTotalGermanySpainArgentinaAustraliaFranceRussiaJapanGBBelgiumSweden64%63%60%58%54%53%49%48%46%45%45%44%44%40%39%36%35%34%34%33%30%29%32%34%37%42%43%41%44%45%46%46%50%49%47%49%60%56%51%52%57%58%The latentdemand forprivacy hasnever beengreater, privacyis the nextkey consumerrights issue,and ourpersonal datamore valuablethan goldor oil.AgreeDisagree3.

Ipsos MORI - Understanding Society July 2014Do you reallyread termsand conditions?We also asked the softest questionpossible on whether people read termsand conditions or user agreementson websites - giving people the bestchance to come clean and admit theydon’t always. But still a third of peopleinsisted they do always read them. Thisranged from a quarter of (marginallymore honest) Canadians to over fourin ten in Spain, Brazil and India. Theevidence suggests a lot of people arekidding themselves or us: server-sidesurveys show that barely 1% actuallydo read them. And this is no surprise,when some service agreements are over30,000 words, longer than Hamlet. 7There are many examples of peoplebeing tested on how much attentionthey pay to what they’re signing upfor – and failing. The best of these isprobably Gamestation’s famous clausewhich they included in their termson April 1st 2010 and which gavethe computer game retailer the 'nontransferableoption to claim now andforever more your immortal soul'. 88%of people signed up.Figure TWO.Q: To what extent do you agree or disagree...'I often don’t bother fully reading terms and conditions on a website beforeaccepting them.'?100%90%80%70%60%50%40%30%20%10%0%CanadaAustraliaGreat BritainItalyChinaFranceDifferentcountries are atdifferent stageson data privacyThe international aspect to our newstudy also highlights the different stagesthe public are at around the world, andgives some clues to likely futures. Inparticular, people in more developedmarkets are more focused on privacy,at least in stated attitudes – and lesslikely to want to give it up for greaterpersonalisation. This seems to be mostlya function of exposure to personalisedservices and recommendations, whichpeople often find more irritating than3363Totalthe USGermanyRussiaSwedenAgreeSpainBrazilIndiaDisagreehelpful. People in the West are alsomore focused on anonymity: promisingthis does little to opinion in countrieslike Brazil, India and China, but it shiftsviews in the West.The survey also shows the realproblems that some industries havedeveloped in a short space of time.Social media and media companiesare only slightly more trusted withour data than foreign governments.And it’s clear from this and previouswork that revelations about the scaleof security service surveillance doesset some of the tone - but it’s not asdirect an effect on opinion of dataprivacy as we might expect from thesignificant media coverage. Moreimportant are the everyday interactionswhere people are surprised by howtheir online activities can be tracked4.

Data privacy – where do people draw the line?Figure THREE.Q: Which comes closest to your own opinion...A - I am happy sharing information about online activities so thatI get personalised services/relevant recommendations.B - I would rather keep information and online activities private evenif I do not get personalised services and relevant recommendations.100%90%80%70%60%50%40%30%20%10%0%IndiaBrazilChinaItalyRussia5226Totalso that recommendations follow themaround the web. Indeed it’s that senseof connection across spheres thatunsettles people the most – the worldand our data may be linked in waysthat we couldn’t imagine a few yearsago, but that worries many of us, aswe like to keep different aspects of ourlives separate.And this provides a hint of thetricky future ahead. In particular, moreexposure to and understanding of whatcan be done with our data doesn’t makeus any more comfortable – in fact, formany of us it increases our concern.the USGreat BritainSpainCanadaAustraliaGermanyFranceHow shouldgovernmentand businessesrespond?ASwedenBoth business and government havedecisions to make. How open shouldthey be about what happens to ourdata, and how far do they chase thebenefits of knowing more about us?It seems blindingly obvious thathonesty will be the best policy. Manystudies (including our own) haveshown transparency is key to trust,and trust is related to all sorts of othergood outcomes. But things are not sostraightforward in this case.BMore exposureto andunderstandingof what canbe done withour datadoesn’t makeus any morecomfortable– in fact, formany of us itincreases ourconcern.5.

Ipsos MORI - Understanding Society July 2014First, we shouldn’t kid ourselves thatopenness will automatically lead to trust.Technological capabilities have racedso far ahead of public knowledge thatthe implications of greater transparencyare unpredictable – many will get moreworried as they find out more.And while there is great concern,there is still little action by mostindividuals to protect their privacyor control their data. There aretherefore weak incentives for privatecompanies and governments to reallypush the issue. Being a first mover,highlighting concerns that peopledon’t really have at the front of theirmind and encouraging them to restrictaccess to their valuable data, is adubious advantage.Of course, while business andgovernments have similar incentives,they have very different interests – anda key dynamic will be the extent towhich one shifts attention to the other.We’re already seeing how legislativebodies are acting to protect citizens byenforcing people’s 'right to be forgotten'online. Internet and other companiesare at the same time championingtheir role in protecting our data fromgovernment intrusion.6.Figure FOUR.Q: To what extent, if at all, do you personally trust the following to use theinformation they have about you in the right way?Public sector healthcare providersBanksPrivate sector healthcare providersYour national governmentSupermarketsCredit card companiesInsurance companiesTelecommunications companiesSocial media sitesMedia companiesForeign governmentsBut as we have seen there is alsosuch variety in views across situations,individuals and countries that thereis no one 'right' response - it needscareful tailoring.Companies need to be clear aboutwhere on the spectrum they stand,but be flexible in how they act andWhile there is great concern,there is still little action by mostindividuals to protect their privacyor control their data.0% 10% 20% 30% 40% 50%Global Totalcommunicate – and careful not to betoo clever.For example, there is talk inthe academic and think tank worldsabout how the combination of moresophisticated analysis and insights frombehavioural economics could allow us to'personalise privacy', defaulting peopleto what we think they’d prefer. Butmany will find it particularly creepy thatdecisions on privacy (of all subjects)can be made for us by connecting ourdata and predicting what we want.There seem to be two factors thatwill unsettle this uneasy balance –events and legislation, which will beclosely linked. It is difficult to see manyevents gaining sufficient public attentionto topple governments or companies– but as we’ve already seen, they areleading to regulatory change.But getting that legislative balanceright is going to be tough.

Data privacy – where do people draw the line?In the end, as Geoff Mulgan remindsus later in this edition, what we’retrading our privacy against is notso much personalised services orrecommendations as all the wondersof an internet provided (largely) forfree: 'Surveillance is the businessmodel of the internet' as onetechnology commentator describes it.Governments and legislators need tobe mindful that where the public drawthe line will depend on what they areasked to give up.Companies need to be clearabout where on the spectrum theystand, but be flexible in how theyact and communicate – andcareful not to be too clever.7.

Ipsos MORI - Understanding Society July 2014No place to hideAn Interview with The Guardian’s Editor-in-Chief, Alan RusbridgerBen Page,LondonThe Guardian and the WashingtonPost were recently awarded thePulitzer Prize for Public Service fortheir coverage of the National SecurityAgency’s surveillance activities basedon the leaks of Edward Snowden. 8The Pulitzer committee praisedthe Guardian for its 'revelation ofwidespread secret surveillance bythe National Security Agency, helpingthrough aggressive reporting to spark adebate about the relationship betweenthe government and the public overissues of security and privacy'. 9BP: In a relatively young digital age,how do we marry the instinct to bothprotect our data and embrace theadvantages of using it? Is the concernfor privacy a growing trend or atemporary blip?Oh no, I think this is going to be oneof the big issues of the 21st century,simply because it is so self-evidentlyproblematic. That doesn’t mean that,in the end, people will not settle for thebenefits over the disadvantages andmake the compromises when necessary.But I think that the potential for thesethings to be quite toxic is always there.I suppose I think of myself on theutopian side of the argument - I’m sort oflike the Clay Shirkys of this worldinstinctively, more than the Morazovs.Clay Shirky is the guy who wrote thesunny books about the internet 10 , andMorazov wrote the gloomy books aboutthe dangers and the threats 11 . I’vealways thought of myself on the sunnierside of the argument, mostly because Ilike technology and I like the things it hasenabled us to do in the news business.Even so, I think the problems ofconnected data and the unthinking wayin which governments and commercialorganisations are stumbling into this,is potentially quite problematic.BP: So do you want to seemore regulation?Well let’s start with transparency andthink about regulation down the line. Ithink there has to be education for astart, because I think most people arequite shockingly ignorant about it all.BP: Our research shows that peoplesay they are anxious about datasharing, but they also admit mostlythat they don’t actually read any ofthe terms and conditions they tick onwebsites. There are great examples ofterms and conditions where, on page33, there is a phrase, ‘Claim $1,000 ifyou have read this far’. And I think ittakes 3,000 people to download itbefore the first person rings up toget the $1,000.That’s interesting. The producers ofthe play Privacy at the DonmarWarehouse worked out that the Appleterms and conditions were the samelength as The Tempest. The actor SimonRussell Beale read The Tempestalongside Eric Walter reading the Appleterms and conditions.I think there needs to be aneducation all round for politicians,business leaders and consumers -we’ve all got to get smarter.BP: So are we right to be anxiousabout privacy, transparency anddata sharing?I think we’re right to be anxious.The Snowden story drove home tome the potential menace of how thesetechnologies can be used and theirpower is considerable. We’re all relyingon oversight mechanisms to say -well, that’s the thing that stops thisfrom being bad, is that firmly in control?I don’t think it is firmly in control.BP: People are particularly anxiousabout specific types of data. Forexample, there is more public anxietyaround institutions like the NHS usingtheir medical data to save time andpublic money, than there is aboutcommercial businesses like Amazonknowing what they buy. Why do youthink this is?Well, it’s the same picture really.I can see the benefits of sharing thattype of data, but I think there is a lot ofreassuring yet to do. People need to bereassured that they are in control of thesecurity of these systems, and thatso-called metadata - information whichis supposedly anonymous - can’tactually be traced.When we live in a world where theNSA, the Department of Justice inAmerica and the military can’t prevent8.

No place to hidemassive leaks, what are we to makeof something like the NHS spine wheretwo million people can access our data?Until we can build safe systems, peopleare bound to feel anxious about use oftheir medical records.BP: Many people will tell you thatvirtually anything can be hacked oraccessed. Do you think it is possibleto build safe systems?Probably not. So I’m quite interestedin the people who are coming up withways in which we are much more incontrol of our personal data. I would bemore comfortable, for example, if I hadmy medical records in my own walletwhich I could personally take to thehospital or my doctor, rather than asystem which controlled it.There are people who are workingon these decentralised systems of data,where it can be centralised, but ina way that I feel in control of it. I don’tknow enough about the technology toknow where that experiment leads, butwhat’s more worrying is governments andcommercial organisations that believe theycan blunder in without seeking any form ofconsent or attempts of any kind attransparency. That is a world where youare just storing up trouble.But I think after Snowden, there willbe quite serious industries being builtaround peer to peer encryption andgauging meaningful consent. Also, thenotion of what is meant by meaningfulconsent is interesting – how much of it isconsent as opposed to how quickly can Iget this box to disappear off my screen?This is something that businesses andgovernments will have to think about.Peopleneed to bereassuredthat they arein control ofthe securityof thesesystems, andthat so-calledmetadata -informationwhich issupposedlyanonymous -can’t actuallybe traced.I think the countries which have notbeen blessed with the kind of stabilitythat we have in the UK are generallymore attuned. So when I meet mySpanish colleagues, who were not livingunder a democracy until 1975, they aremore highly attuned. America had Nixon,Hoover, McCarthy, in living memory, andhave a more rights-based society, andso they are more highly attuned. And Ithink sometimes in Britain we sort of patourselves on the back for Magna Carta,and we don’t think this really affects us.But when you sit down and explainwhat’s going on, I think people then areworried and do understand the issue.BP: Thinking a bit more aboutthe opportunities, how has newtechnology changed journalism’srelationship with its audience,and how much of a focus is thereon personalised news?We’re experimenting all the time. In1994, the Massachusetts Institute ofTechnology designed a virtual dailynewspaper, customised for anindividual’s news preferences, called theDaily Me. In fact, we tried printing off afew Daily Mes. And actually last weeksomebody came in with somethingcalled Paper Later, which is apersonalised newspaper, so it’s sort oftaken 20 years to get to what we wereimagining in the early ‘90s. We couldcompletely tailor-make anything foranybody, depending on theirpreferences. At the moment there’s alimited take up, because I think mostpeople’s idea of a newspaper is actuallysomething that challenges and broadenstheir horizons.BP: Which country has the bestapproach to privacy and security, andwhich has the worst?9.

Ipsos MORI - Understanding Society July 2014BP: Reading a paper online or on aniPad, as opposed to reading in print,means that you can miss stories,depending on your filtering system.You are also immediately drawn tosome stories through availabilityor attention bias than others.Yes, but we don’t know thatthis will always be the case. I thinkwe’re still in the very early days ofwhat location data means, and whetherpeople will want to use location servicesto limit their news or their entertainmentand sporting choices. So we willcontinue to experiment, but we don’tknow the degree to whichpersonalisation will be really important.BP: The Guardian has been atthe forefront of citizen and digitaljournalism, embracing online anddata blogs, leaving other nationalpapers to catch up. How do youtread the line between engagingyour readers as citizen journalists,whilst meeting the standards ofprofessional journalism?We call it open journalism, becauseI think citizen journalism has a rathersort of clunky feel to it. But you can sayin the world that we live in now,everybody is potentially a publisherof information.I use the analogy of MichaelBillington, our theatre critic. Being theprofessional expert, we send him intothe stalls every night. A typical night atthe theatre will have 900 people therewith him. Is it conceivable to say thatout of all 900 people, none of themhave anything interesting to say?The answer has to be no, of course.So then the question is, how do wefilter the interesting ones from theWe don’tknow thedegreeto whichpersonalisationwill be reallyimportant.uninteresting ones? Even if you onlyget 10 interesting opinions of the playout of 900, you are bound to have amore complete account. So, if it worksfor theatre, would it work for sport,science, war reporting andinvestigations?BP: And does it work?Yes, it always works. We are tryingto build a model where we acknowledgethat there are some things that we asjournalists do best, and if we are reallysmart about harnessing what otherpeople are doing, we could buildsomething that was even better. And thatis how we knit a newspaper into the ebband flow of information today, includingthe distribution and sharing of it.If you look at our science orenvironment coverage, you can seewe have a whole network of sisterbloggers, commentators and articles.What we are doing is handpicking thebest of what’s out there, some of whichis good, if not better than what we cando. When you look at Twitter, you seethat it will always be faster and morecomprehensive. But newspapers canamplify what is interesting. Suddenly,you have a different idea of what anewspaper is.BP: What’s the next big thing afterSnowden for you?The aftermath of Snowden,as I’ve said, is that we understand thatwe need some of this stuff to keep ussafe. The question is, how do we keepus safe from technologies that couldbe totalitarian? I don’t think that anyonebelieves that the system of oversightthat we currently have is up to the task,or that Parliament has ever debated it10.

No place to hideproperly. We still don’t have anythingthat you can reasonably call consentwhen it comes to sharing information.I think if you had all those conditionsin place, then yes, you could moveon from this subject and write aboutsomething else. But I think there isstill some mileage to go.BP: And of course, if you had abrilliant browser that allowed youto click here and nobody could eversee what you were looking at, youcan guarantee it would get plentyof take up.This word, metadata, is the word of2014, I think. And wheneverI hear politicians like Malcolm Rifkindor William Hague on the radio, talkingabout Snowden - politicians whose jobit is to reassure - go straight to'nobody’s reading your emails, we needa warrant to look at this, we’re notinterested', and so on. And they fail tomention the information that they don’tneed warrants for, which is metadata.There is a moment in the Privacyplay where everybody is asked to gothrough their mobile phones. The actortakes everyone who has an iPhonethrough the settings where you can seeeverywhere that your phone has trackedyou for the last three months. And youhear a gasp in the audience. I’ve done itwith my friends – you can take themthrough it and tell them ‘so you werein that house in Tottenham on that nightat 8 o’clock’. When people realise howmuch information on their lives is beingstored, they suddenly feel that it’s creepyand then the penny drops. You don’tneed a warrant to access thatinformation and nobody has any privacy.I don’t thinkthat anyonebelieves thatthe system ofoversight thatwe currentlyhave is up tothe task.And just like phone hacking trialsillustrated an explosion of bad practicesin the newspaper industries, this willhappen in other walks of life too. Orthere will be accidental leaks. Eitherway, that’s why I think it is an issue thatis not going to go away.Brilliant, thanks so much foryour time.11.

Ipsos MORI - Understanding Society July 2014Privacy or security:a false choice?European citizens' perceptions of privacy, personal data, surveillance and securityCarolina Haita,BucharestDaniel Cameron,LondonThe relationship between privacyand security is often described as atrade-off. The argument goes somethinglike this: securing against the manythreats individuals and societies facerequires people to give up at least someof their personal privacy. Many of thesecurity and surveillance technologiesused by governments and others relyon this reasoning to justify the waysthey reduce individual privacy. 12 Thereare certainly different views aboutthe extent to which privacy should betraded-off, but this basic model is rarelyquestioned in the course of public andpolitical debate.Ipsos MORI is working with aconsortium of partner organisationson an EU-funded project to investigatethis traditional privacy-security tradeoffmodel. 13 The aim is to re-examineprivacy and security in order to explorewhether there could be alternative waysto characterise the relationship. Is itpossible to enhance security without aninevitable reduction in privacy? Can newtechnologies be designed to strengthenboth privacy and security?As part of the project, Ipsos MORIconducted discussion groups in eightEuropean countries during summer2013. 14 The discussions soughtpeople’s views about specific securitytechnologies in contexts where theyare or might be used, alongside moregeneral perceptions of privacy andsecurity in today’s world. The findingswere used to help design a large-scalerepresentative quantitative surveycurrently being carried out in EUMember States.This article summarises some ofthe key findings from the qualitativeresearch. These give a snapshot ofEuropean citizens’ perceptions ofprivacy and security.The importanceand limits ofprivacyParticipants were clear that privacyis essential to them personally. Thisinstinctive sense that privacy shouldbe considered a basic human rightwas near-universal. Yet participants’understanding and expectations ofwhat privacy looks like in practicewere much more nuanced. This was12.

.Privacy or security: a false choice?often informed by what they thoughtwas realistic rather than by anyideological commitment to a specificunderstanding of privacy.For example, participants did notthink it was possible to have privacyin many circumstances where theyconsidered it desirable. They wereoften willing to accept considerableintrusions into their privacy if theythought this would strengthen local,national and international security,and if they could see limited drawbacksfor them personally.information in order to access services.They therefore saw a strong role forthe state in developing appropriateregulatory frameworks for protectingpersonal data, and for policing andenforcing agreed standards.What we do doesn’t matter…our personal data is storedin many places… when youare online all your data canbe available.Male, under 40, HungaryI think it’s scary that theyknow when and where we go,but I can understand why.Female, under 40, BelgiumProtecting personal data was theusual starting point for thinking aboutprivacy. How personal data is collected,stored and used was the most currentand contentious privacy issue.Participants discussed many differentexamples of the sorts of personal datathey knew was collected and storedabout them, from financial data tomedical records, from location data topictures of their most recent holiday onsocial media. They felt that they shouldhave the ability to control their personaldata, as it was theirs to share becauseof the benefits it brought them.I only give away the datathat is really necessary,especially online.Female, under 40, GermanyYet participants also believedthey had very limited control over thepersonal data held by organisations,including both government and privatecompanies. Often they felt they hadno choice but to provide personalGovernmentmore trustworthy,businesses morecompetent?How much participants trustedgovernment and private companiesvaried considerably. In general,there was more trust in government’sintentions but less trust in theircompetence to protect personal data.The opposite was often the case forbusinesses. There were concerns aboutthe reasons behind private companies’frequent requests for personal data.However, if participants could seeclear benefits for them, for example interms of improved service, many werecomfortable providing data if the rightreassurances were in place.13.

Ipsos MORI - Understanding Society July 2014Levels of trust in government whenit comes to privacy were differentin each country. This tended to berooted in broader attitudes to thestate: where government was viewedas largely benign there were fewerconcerns about its role in storing andusing personal data. In countries withmore recent experience of oppressiveregimes, participants were much moreguarded about trusting the state torespect privacy and use informationappropriately.Overall, some types of privacywere seen as untouchable – sensitivedata like medical records, individuals’political and religious views, theirpersonal communications, andparticularly what happens in people’shomes. Any intrusion in these areaswithout very good reasons and strictlegal controls was seen as completelyunacceptable.I don’t want to be monitored athome, they can do that all theywant in other places, but not athome.Female, under 40, DenmarkMy family and my home areprivate, and also what I tell mybest friend via mail. I want tokeep that.Female, over 40, GermanyThere was anassumptionthat privacy,at least inthe strictestsense, is veryoften difficultto maintainif you wantto be partof a modernsociety.But there was also an assumptionthat privacy, at least in the strictestsense, is very often difficult to maintainif you want to be part of a modernsociety. In particular, being able to movearound and meet with other people inpublic without being watched was notthought to be as important as some ofthe other types of privacy discussed.Many thought freedom of movementand association were now unrealisticideals and, more than that, relativelyunimportant provided appropriatesafeguards were in place. Theyassumed this sort of information wouldonly be of interest if people were doingsomething wrong.I think privacy in public spaces isnot possible anymore. There arecameras everywhere.Male, over 40, GermanyI don’t know what the problem is– if you follow the rules there’snothing to worry about. If not youdeserve to be caught.Female, under 40, UKThis reflects the clear concernsparticipants had about protectingsecurity. Threats from crime andterrorism were discussed frequently,as were the responses people thoughtwere required. The need to protectsecurity was an important priority –and for some, more important thanprotecting their privacy.I favour the use of videosurveillance in the cities becauseof safety concerns.Female, over 40, Portugal14.

Privacy or security: a false choice?We rely on government – thepolice and the army and so on– for our security, ultimately. Thereare things you can do but weneed outside help.Male, over 40, UKAttitudes to both privacy andsecurity were broadly shared acrosscountries, but there were differencesin the detail. For example, in somecountries like Hungary and the UKthere were few concerns about internetsurveillance, while in others likeGermany and Denmark participantshad considerable reservations aboutthese technologies becoming moreembedded without proper controls.Context mattersParticipants also discussed specificsecurity technologies using a numberof different scenarios. Examplesincluded airport scanners, smart metersand fingerprint recognition in schools.Discussions around these scenarioshighlighted how context-dependentattitudes to privacy and security are.One scenario that generatedgood discussions focused on internetmonitoring. The scenario described aMuslim student who researches anddiscusses terrorism on the internetas part of his studies. His parentsbecome concerned about internetmonitoring and ask him to stop hisresearch. Participants were able to seethe tensions inherent in protecting bothprivacy and security in this context.For most, monitoring this type ofbehaviour was seen as legitimate –or indeed crucially important – providedit is done by government in a controlledway. But most thought that researchingterrorism was legitimate too, evenif they thought this should be donein a transparent way that is open toquestions from government. Finally,participants also understood parentalconcern, with some saying they wouldwant to dissuade their own children insimilar circumstances.I agree with the parents…I would do the same thing.Male, over 40, PortugalSome of the scenarios were moreclear-cut. One described governmentasking citizens to record their religionon an identity card, in order to helpensure people from different religiousbackgrounds were involved in localdecision-making.Participants raised a numberof objections to this idea, pointingout that religion is a private matter.They were concerned about linkinginformation about religion to other dataon an identity card; risks around thesecurity of government databases;and possible future misuse of the databy radical organisations. More thanthat, participants could see no obviousbenefits, arguing that there werealternative ways to consult local religiousgroups. Even with reassurances, thisscenario was considered completelyunacceptable in almost all countries.The needto protectsecurity wasan importantpriority– and forsome, moreimportantthanprotectingtheir privacy.15.

Ipsos MORI - Understanding Society July 2014Participants wanted a robust,detailed case to be made fornew technologies, includingwhether they are cost effectiveand clarity of why any loss ofprivacy is necessary.Reinforcement,resignation andresistanceThere were three commonunderlying themes when it came toparticipants’ views on the relationshipbetween privacy and security, reflectingthe pragmatism evident in our GlobalTrends Survey data described earlierin this edition. While participantsresponded in different ways dependingon the specifics, most tended towardsone of these groups:• Reinforcement – many were contentwith giving up further privacy if itmade them feel more secure. Theysaid they had nothing to hide fromauthorities and would welcomefurther surveillance if this helpedprevent crime and terrorism.16.• Resignation – this group wereinstinctively uncomfortable abouthow much their privacy is currentlycompromised, and this wasexacerbated because they expectedthe situation would continuegetting worse. But they could seefew options for them to claw backprivacy, and seemed willing to livewith the situation because it madelittle difference to their daily lives,and indeed brought some benefits.• Resistance – there were someparticipants who wanted privacy tobe strengthened. This was usuallybecause they had a principledobjection to intrusion into theirprivacy. They felt things had gonetoo far in terms of surveillance andsecurity technologies, and did notaccept that these really improvedsecurity in any case.What next?Participants did not agree on thebest way forward, but there were someshared concerns. They highlightedseveral issues they thought should betaken into account when introducing newsurveillance and security technologies.They wanted better protection ofpersonal data by controlling accessto data within organisations, and bystrongly limiting data transfer to thirdparties. Many felt that individuals shouldbe treated better too, so that people arerespected, have choice and control, andare not viewed with a general suspicionas a result of new technologies. Finally,participants wanted a robust, detailedcase to be made for new technologies,including whether they are cost effectiveand clarity on why any loss of privacyis necessary. Taking these concerns onboard will be important for developingnew technologies that protect bothprivacy and security in a way that isconsistent with public priorities.

Is people-powered data possible?Is people-powereddata possible?The future of citizen controlGeoff Mulgan,Chief Executiveof NestaSo far the public seem to haveaccepted a dramatic increase inuse of their personal data because itdoesn’t impinge much on our freedom,and helps to give us a largely freeinternet. That’s because the publicgenerally take a pragmatic view ofprivacy: it’s not and never should besomething absolute. Instead we’rewilling to make trade-offs – if usinga loyalty card gives a supermarketinformation about what we eat, thatdoesn’t matter much so long as weget something in return and there’sno obvious downside. On the otherhand, we’re far more sensitive aboutdata relating to our physical health,or even more our mental health,because we can see how easily thatinformation could be used againstour interests.But the ways in which we makethese trade-offs could be changing.Edward Snowden’s NSA revelationshave fuelled a growing perception thatthe big social media firms are cavalierwith personal data (a perception nothelped by Facebook and Google’srecent moves to make tracking cookiesless visible) and the InformationCommissioner in the UK has describedthe data protection breaches of manyinternet firms, banks and othersas ‘horrifying’.Ipsos MORI’s research suggeststhat the more people are familiar withsocial media the more cautious theyare about what’s done to their data,and of course over time an ever largerproportion of the world’s population isbecoming more digitally aware.According to some this doesn’tmatter. Scott McNealy of SunMicrosystems famously dismissedthe problem: ‘you have zero privacyanyway. Get over it.’ Mark Zuckerbergclaims that young people no longerworry about making their livestransparent. We’re willing to be digitalchattels so long as it doesn’t do us anyvisible harm.That’s the picture now. But the pastisn’t always a good guide to the future.More digitally savvy young people put ahigh premium on autonomy and control,and don’t like being the dupes of bigorganisations. We increasingly live witha digital aura alongside our physicalidentity – a mix of trails, data, pictures.We will increasingly want to shape andcontrol that aura, and will pay a price ifwe don’t.That’s why the movement forcitizen control over data has gatheredmomentum. It has been 30 years sinceGermany enshrined ‘informationalself-determination’ in the constitutionand other countries are consideringsimilar rules. Brazil recently passed theworld’s first Digital Rights legislation,partly inspired by Sir Tim Berners Leewho thinks we need a new MagnaCarta fit for a digital age.Organisations like Mydex 15 and Qiy 16now give users direct control over astore of their personal data, part ofan emerging sector of Personal DataStores, Privacy Dashboards and even‘Life Management Platforms’. In theUK, the government-backed Midata 17programme is encouraging firms tomigrate data back to public control,while the US has introduced green,yellow and blue buttons to simplifythe option of taking back your data(in energy, education and theVeterans Administration respectively).Meanwhile a parallel movementencourages people to monetisetheir own data – so that, for example,Tesco or Experian would have to payfor the privilege of making money17.

Ipsos MORI - Understanding Society July 2014Privacy or security: a false choice?out of analysing your purchasesand behaviours.Few of us have a very realisticunderstanding of what actuallyhappens to our personal details.But when people are shown what reallyhappens to their data now they areshocked. That’s why we may be neara tipping point. A few more scandalscould blow away any remainingcomplacency about the near futureworld of ubiquitous facial recognitionsoftware, Google Glass and thelike, a world where more people arelikely to spy on their neighbours,lovers and colleagues.Geopolitics will also play its part.Europe is much more concerned aboutidentity and privacy than other partsof the world, and that concern hasbeen fuelled by revelations of NSA’s farreaching surveillance. Europe will setmuch more stringent rules for protectingpersonal privacy and given that the EUnow has 500m of the world’s richestcitizens it will be hard for social mediafirms to ignore these standards.Meanwhile some firms aredistancing themselves from the pack –like Microsoft, developing technologiesto help citizens control unauthorisedreuse of their data. The bigconsultancies and accounting firms seeprivacy accreditation as a huge newsource of business. And The BostonConsulting Group recently warned thattwo-thirds of the total value of greateruse of personal data, estimated toreach a potential €1trn in Europe by2020, will be lost if organisations failto establish trust.The next few years will bring afurther explosion of data, and of dataawareness in daily life. In the end wemay be able to choose more easilyjust how open our lives are to be –and some of us will be willing topay more to stay more private. It’spossible too that a new generation oftechnologies will be developed thatdesign in privacy and personal controlover data, just as a previous generationof technologies, dominated by themilitary, security services and bigbusiness, did the opposite.We’re some way off the new MagnaCarta that will at some point need toestablish the ground rules of privacy,power and identity in a digital world.But these issues are fast moving fromthe margins to the mainstream ofdaily life – and some very powerfulorganisations risk being on the wrongside of history. 18Few of ushave a veryrealisticunderstandingof whatactuallyhappens toour personaldetails. Butwhen peopleare shownwhat reallyhappens totheir datanow they areshocked.18.

The trouble with numbersThe trouble with numbersBen Page interviews Nate SilverBP: Survey research is mostly aboutasking people questions and addingthe answers up, and then trying tointerpret what they say. Increasingly,it’s moving into passive measurementas well. You have become one of theaggregators-in-chief of surveys. Whatdo you find most interesting thesedays? Are you going to try andforecast the US job market?NS: Probably not! The evidencesuggests that economic forecasts arenot very good, and have not beengetting any better.BP: The meteorologists claim to bebetter than the economists.NS: They are; the meteorologistsare kind of the heroes of my bookThe Signal and the Noise. There area few things that help: one is that wehave a good physical understandingof dynamics of the weather system,so they’re actually building a simulationof the atmosphere – it’s not reallystatistical. It’s been a success storywhere hurricanes now are forecast about250% more accurately than 25 yearsago. Hurricane Sandy was predictedalmost exactly five days in advance.The sports gamblers are also very good.Part of it is that if you’re putting moneyon the line, it has a disciplining effect.A lot of people in American politics inparticular are just pontificating – kindof like in the essay, you know ‘OnBullshit’ – where it’s not that you’re lying,so much as you’re entirely indifferenttowards whether you’re accurate or not.BP: I read you as somebody who isbig on Bayesian statistics. Tell usabout Mr Bayes.NS: So Thomas Bayes was anEnglish reverend and kind of a pioneerin the theory of statistics, and – thiswas 250 years ago – his basic insight issomething called Bayes’ Theorem, whichis simple algebraic equation. It says, firstof all, you have a probabilistic view onthe world – and it’s by the way not ametaphysical statement, it’s not sayingthe world is uncertain, it’s just sayingour knowledge of the world is uncertain.The other thing is that it assumes youstart out with a prior belief – apresumption of a bias – and you weighnew evidence against that.BP: I guess there is still an elementof subjectivity or judgement involvedin the things you’re doing?NS: I tend to see subjectivity asintrinsic to the way human beings viewthe world: we all have one narrow lensthrough which we perceive the shared‘objective’ reality and making predictionsis a way to verify whether our subjectiveview matches this reality, which is hardto do.The evidencesuggests thateconomicforecasts arenot very good,and have notbeen gettingany better.19.

Ipsos MORI - Understanding Society July 2014It’s aboutunderstandingwhen youare really ina data-richenvironment,because it’snot just thenumber ofobservationsyou have.BP: I’m interested in just how goodboth ordinary people and decisionmakersin business and ingovernment are at dealing withprobabilities.NS: Yeah, people don’t have anintuitive grasp of that at all. They thinkif you say something is 80% likely tooccur, it’s tantamount to saying it’sguaranteed to occur, when obviously it’snot. If you woke up every day and youhad an 80% probability of not beingstabbed, you wouldn’t survive very long.Those 20% outcomes occur quite a bit.If you follow sports as I do, or especiallyplaying poker, where you’re on the losingside of that 20% enough, your intuitionfor probability is better. There were acouple of years where poker was mymain source of income.BP: You’ve stopped this now? You’vefound the blog is better, the writingis better?NS: Well, the poker’s worse! In 2003you had a kind of fat accountant slobnamed Chris Moneymaker, this everymancharacter – nice guy – who won theWorld Series of Poker in Las Vegas. He’san OK player, but basically got verylucky. And ESPN would edit out all thehands that he would lose or play badly.That deceived people as to how easy itwas. So for a while, there was a lot ofdumb money in the poker economy – butnow it’s gotten very, very competitive.BP: People talk a lot about Big Data.To me, some of that is like that guyfrom Google who said, ‘Social mediais like teen sex: lots of people talkabout it, lots of people look forwardto it but when its finally achieved,most people are disappointed.’NS: If you read the Harvard BusinessReview, every other ad is about big data.They’re pitching this as the magicserum: that’s always dangerous. Theproblem with some of this data is thatit’s big, but it’s not structured. It’s aboutunderstanding when you are really ina data-rich environment, because it’s notjust the number of observations youhave. The credit rating agencies inadvance of the crunch had millions ofobservations on individual mortgages,but all from a period when housingprices were increasing. Part of the issuewith social media metrics is that you’recollecting data that’s only a couple ofyears old. But Google's product, FluTrends, searches in real time.BP: This is people typing ‘flu’ and theycan predict actual influenza cases.NS: They say they can, but theyconsiderably overestimated the flu inthe United States last year. My hunch is,you build a model that can accuratelydescribe what search terms people wereusing in, say, 2008 to 2010, but peoplechange their search habits. Now that youhave Google auto-complete, that willsuggest different queries to people thanyou might have had, and when peopleare typing on mobile phones you tendto have shorter queries, so that willchange things. Three years is a lifetimein social media.20.

The trouble with numbersBP: We know people aren’t rational:Nudge by Thaler & Sustein is verypopular here, as it is in US policycircles, but what’s your advice to theaverage human being? Most people’sfinancial planning, for example, seemsto involve them dying at about 70,when actually they should know it’svery unlikely to happen.NS: People apply a lot of attentionand a lot of bandwidth to where they’regoing to go for dinner – and I’m a foodie,so I don’t begrudge that entirely – butthose heuristics don’t work as well forthings like career or education planning,where you have to be more detached.One thing we’re seeing now is that onceyou’re unemployed for more than aboutsix months, it becomes very difficult tofind a job again. Employers begin toassume, 'A lot of other employers havepassed on this person, there must besomething wrong with them'.People thinkif you saysomething is80% likelyto occur, it’stantamountto saying it’sguaranteed tooccur, whenobviouslyit’s not.BP: So the advice is if you becomeunemployed, take any job, and don’twait for the right job? OK. Any otherexamples for people?NS: There was a column I did in theSun – slightly embarrassing! But thereis a dating site called OKCupid thatanalyses a lot of data, and a coupleof years ago, we looked at which isthe best night of the week to go out tomeet someone. They looked at whatpercentage of people had updated theirstatus to say 'I wanna get laid', basically,and the percentage peaks onWednesday night. We think the issue isthat early in the week, people go out todrink; then on Friday and Saturday,everyone goes out with their friends. It’skind of amateur hour. The headline of theSun article was ‘Maths can be fun!’This is an editedversion of anarticle originallypublished in theIpsos MORIAlmanac 2013.21.

Ipsos MORI - Understanding Society July 2014The rise of theforecaster-punditBig data, data aggregation and the 2012 US Presidential ElectionClifford Young,WashingtonElisa Bernd,WashingtonThe 2012 US presidential electionwas a watershed for a number ofreasons. Our first African-Americanpresident—Barack Obama—successfully defended his presidencyagainst Mitt Romney, the Republicanchallenger. For most, Obama’s victorywas no surprise. He did, though,outperform both the polls and theelectoral forecasts by a fair margin,beating Romney by 3.9 percentagepoints (51.1 to 47.2), when most put hisprobable victory margin at between 1and 2 points.But the 2012 electoral cycle alsomarked the rise of big data andadvanced analytics in politics. Onesuch example is the Obama campaign’suse of advanced analytics and largedatasets to push their base to the pollson election-day. Such data allowed theObama campaign, through statisticalpropensity models, to identify likelyObama voters who without a nudgewould not have shown up on electionday.Many analysts attribute Obama’sstrong electoral showing to hiscampaign’s analytics advantage overRomney and the Republicans. 19Another example of 2012’s big dataDNA is the sheer volume of polls. Theadvent of new methodologies like onlinepolling and robo-polls has significantlyreduced cost barriers when comparedwith the face-to-face and telephonepolls of previous decades.By our count, 17,058 polls wereconducted at all levels of government(local, state and national) in 2012. Thisnumber probably undercounts the totalnumber of polls by a factor of two tothree as it excludes private politicalpolls for campaigns.Front and center in this pollvolume revolution were websites, likeRealClearPolitics 20 and HuffPo’s 21Pollster, that aggregated the polls inalmost real-time using ‘web scraping’and other automated curationtechniques. These ‘aggregators’became essential in serving as avirtual meeting place both for politicaljunkies and by making poll data easilyassessable to the public.By our count,17,058political pollswereconducted atall levels ofgovernment(local, stateand national)in the USin 2012The exponential increase in thenumber of polls also made statisticalanalysis easier and more robust. Nolonger was the single poll of muchinferential interest. This innovation, inturn, led to another equally new andunique phenomenon in elections: the'forecaster-pundit'.Statistical election forecasting foundits way into the national lexicon and themedia’s election coverage in 2012. NateSliver—the best known forecaster of thebunch—together with others, like SamWang, Drew Linzer, Alan Abramowitz,and Simon Jackman, became importantstakeholders in the national narrativeand ultimately predicted quiteaccurately the final election outcome(see Figure 5). Mis-predictions bywell-established polling firms andthe Romney campaign led many totalk about the rise of the 'forecasterpundit'and the imminent demise of thetraditional political pundit and pollster. 22What did theforecasters do?In broad strokes, the electoralforecast models employed in 2012depended on data aggregation. Someaggregated individual polls—sometimeshundreds or even thousands. Othersaggregated past US presidentialelections; while still others combinedboth poll and election aggregation.However, all models used aggregateddata rather than a single poll or other‘special sauce’ heuristics.At their core, all these modelsmaximize the amount of informationemployed in order to reduce uncertainty.They also go beyond forecasting ofvote share by estimating the probability22.

The rise of the forecaster-punditof victory for a given candidate(e.g., Obama has an XX% chance ofwinning). 23 This is a radical departurefrom the ‘deterministic’ nature ofsingle poll analysis where outcomesare thought of as binary (e.g., Obamawill win or will lose). This innovation,however, only becomes possible in aworld of big data where the frequencyof an event can be more easilyascertained (e.g., how many polls showObama ahead of Romney).The first family of models—pollaggregation or model-based averagingapproaches (see Figure 6) – basicallydo two things: (1) they combine multipleFigure FIVE.Forecasters’ final predictions of the 2012 US presidential election resultFinal PredictionVote ShareProbability ofObama VictoryActual 3.9 -Linzer 3.9 99%Jackman 1.6 91%Wang 2.5 99%Silver 2.2 91%Abramowitz 3.8 80%Figure SIX.Summary of aggregated forecasting modelsPoll Aggregation Election Aggregation Poll + Election AggregationFoundational Article Jackman (2005) 24 Abramowitz (2008) 25 Linzer (2012) 26Model Definition Model-based Averaging Fundamental Model Combinatorial ModelMethodAggregating multiple polls andthen correcting for specific pollsbiasUsing past election polling andother data to predict futureelectionsAggregating multiple polls andusing extrapolation from pastelectionsDataNational and States polls100s to 1000sPast elections: 16 US electionssince 1948 (1948 is the adventof modern poll)National and States polls pluspast electionsVariablesSingle Polls and poll characteristics(polling firm, sample size,etc)Vote Share, GDP, incumbency,consumer confidence, approvalratingsSingle Polls and poll characteristics(polling firm, sample size,etc)Estimation technique Markov Chain Monte Carlo 27 Linear regression Markov Chain Monte CarloForecasters Jackman, Silver, and Wang Abramowitz LinzerStrengthsPolling information is real-timeinformation that captures thecampaign dynamicNot dependent on campaignnoise like the pollsCombines the strengths of thepoll and election aggregationapproachesWeaknessPolls often are false positives,or capture noise. If there is asystematic bias in all polls thenno way to correctVery dependent on variablesin the model. Also, assumptionthat past behavior predictsfuture behaviorMore complicated. Still isdepended on the assumption ofpoll and election aggregation—always problem of being out ofsample23.

Ipsos MORI - Understanding Society July 2014polls into an overall average and (2)they correct for biases in individual polls(known as house effects). Jackmanof HuffPo’s Pollster.com and Stanford,Wang of Princeton, and Nate Silver allemploy their own version of modelbasedaveraging.Such models are very dynamic asthey capture real-time information fromvoters via polls. However, there canbe systematic bias in polls, like undercoverageof minorities or misidentifyingwho will show on election-day.Additionally, polls can be noisy andproduce ‘false positives’ in the shortterm as they jump around due to eventssuch as party conventions or debates. 28In contrast, election aggregationapproaches (or fundamentals models)do not incorporate horse race polls,instead relying on past electoralbehavior to predict future behavior.Often such models employ bothpolitical and economic variables toestimate vote share:Vote Share = Political variables andEconomic variablesThere are many derivations of thesemodels. 29 But political variables mayinclude party-of-the-president, approvalratings, incumbency, or number ofyears in power. Economic variablesmay include unemployment, consumerconfidence, or GDP among others.Fundamentals models overcome thespecific noisiness of poll aggregationas they depend on past electionsand not current horse race polls. Thatsaid, there are three weaknesses tofundamentals models.First, they assume that past isprologue: that past behavior will predictfuture behavior, which is not always thecase. Second, fundamentals modelsdepend on the variables inputted intothem – omitting key variables is a riskwhich can be especially problematicwhen data is scarce. And third, suchmodels rely on the election as theunit of analysis. Many countries haveonly a handful of elections, leadingto challenges with small sample sizesand a heightened risk of rare events oroutlier elections.Finally, combinatorial modelsintegrate the poll and electionaggregation approaches. They includeboth the averaging of multiple pollsas well as inputs from fundamentalsmodels. Theoretically, this increasein information should improve modelaccuracy. That said, combinatorialmodels do suffer from the sameunderlying weaknesses of eachindividual model. It is always importantto remember that all models are only asgood as the data they use.Figure SEVEN.Performance of re-engineered forecasting modelsMethod Adjustment/variables DiffProb obamawin (%)Pundit forecasterModel-Based Averaging Simple average; no adjustments 0.9 80 Naive Poll Watcher PredictionModel-Based Averaging Taking out house effects 1.65 93 Jackman Model (actual diff 1.6)Model-Based AveragingFundamentals ModelCombined ModelTaking out house effectsAdjustments for N-sizeAdjustment for most trustedfirmsGDPApproval rating# of years of incumbent partyin powerCombines Model-Based Averagingwith Fundamentals Model2.1 91 Silver Model (actual diff 2.2)3.7 81* Abramowitz Model (actual diff 3.8)4.0 98 Linzer Model (actual diff 3.9)* To calculate the probability of an Obama victory for the fundamentals model (which does not count polls), we simply count the numberof elections in which the model picks the right winner for US President (or 13/16).24.

The rise of the forecaster-punditIs it possibleto replicatethese models?In Figure 7 we detail results fromre-engineered versions of the differentforecasting models employed in 2012.In order to this, we employ two differentdatasets: (1) 539 national polls fromJan 1, 2012 to November 4th and (2)16 US Presidential elections from 1948to 2008.Three benchmarks are used toassess these re-engineered models:1. The last estimate of each of the‘forecaster-pundits';2. The actual election results; and3. A naïve poll aggregation estimatewhich does not correct for housebiases.So how close do I get to theforecaster-pundits?The results in Figure 7 are basedon simplified approximations of eachforecaster’s model. Even so, we getwithin a tenth of a point of their finalestimate, showing that it is possible toreplicate these models and to validatehow well they predicted the electionoutcome.How good arethe models?All models out-perform the naïveaveraging model which puts Obamaonly at about a 1 point advantage overRomney. Remember Obama ultimatelywins by 3.9, meaning the polls as awhole were off by about 3-points!Both the Jackman and Silver modelsare improvements over the naïve model,picking the right winner and assessingan Obama victory as highly probable.But we wouldn’t call them 'highlyaccurate' being off by more than 1.5points. The Abramowitz model does doa great job of estimating the correctvote share (3.8 vs. 3.9). So why notemploy fundamentals models in placeof poll aggregation?Well, for two reasons: First, suchmodels have an error rate of around20% which is primarily a function ofthe small N-size of elections at hand(which increases the risk of an outlierelection having more influence than itshould). Second, fundamentals modelswork well until they don’t; they dependon past behavior to predict futurebehavior. For most elections, this isa reasonable assumption, but not fordiscontinuity elections.The Linzer combinatorial modelperforms the best. Such models havean advantage over their ‘fundamentals’counterparts because they combinestructural information from pastelections together with real-time polling.Again, while they don’t eliminate allproblems, they do minimize the specificflaws of each individual model.25.

Ipsos MORI - Understanding Society July 2014The power of bigdata in pollingSo what did the 2012 election teachus about big data in politics? And whatdoes it mean for polls and the pollingindustry?First and foremost, the ‘forecasterpundit’is here to stay in one form oranother. With the quantity of polling dataout there, media and political analystswill increasingly take advantage of theinherent superiority of poll aggregationover single-poll analysis. With this, anyone poll or polling firm becomes lessrelevant to the overall political narrative.The poll aggregation movement andthe 2012 election have provided manyinsights into big data as a concept.Specifically, big data does not alwayshave to be, well, BIG. The aggregateddata sets we used here are no largerthan a few gigabytes of memory – muchsmaller than those usually associatedwith big data.Cheaper fieldcosts allow formuch largersample sizesand for pollswhich canalways be inthe field.So if not size then what does2012 tell us about big data? In ourview, the primary lesson is about theaggregation of heretofore nonexistentor inaccessible information and howaggregation, in turn, can innovateour analytical reasoning. Specifically,many repeated measures of the samething increase our analytical power.It reduces the uncertainty of ourestimate and pushes us from thinkingdeterministically to probabilistically. Nolonger do we need to assess the worldin ‘binary single poll’ terms: Obamawill win yes/no. Instead, we can assigna number from 0 to 100. This gives usincredible analytical flexibility.But the 2012 Presidential electionalso shows the shortcomings of bigdata. In particular, that your modelsare only as good as the data they use.Most of the models based solely on pollaggregation underestimated the Obamavictory because the polls themselvesunderestimated Obama.So what of the traditional pollster?Will they go by the way of the typewriter?Our answer is both yes and no. Thesingle poll loses its relevance in thecontext of poll aggregation. But, at thesame time, polls still help us understandvoter motivations which often are washedout in a big data context.Cheaper field costs allow for muchlarger sample sizes and for polls whichcan always be in the field. We call thiscontinuous polling. The best exampleis the present Ipsos-Reuters onlinepoll which is in the field every-day andincludes about 11,000 interviews permonth. Such continuous polling allowsfor a much more detailed analysis ofsubgroups (e.g., blue collar democratsin Ohio) as well as the ability to capturepublic opinion on key events within thesame day.The 2012 US presidential electioncould be classified as a high informationelection: it had many polls and electionsto aggregate. Not all countries andelectoral contexts have such largevolumes of information. Would ourmodels here work in low informationenvironments? In countries with onlya few polling firms? Or with only afew elections? Our initial response tothis question is no—in low informationelections our models must be different.But, even so, aggregation in some formor another will be our solution.26.

The data rich societyThe data rich societyExploring a new way of working with numbersHenri Wallard,ParisThe basis of many publishedarticles – in science, economics,research, media and so on – includesdata to support the authors’ arguments.Yet data and statistics cause a widerange of reactions: from skepticismabout accuracy and objectivity to thecurrent hype and fascination aroundbig data. In some cases data isrevered as an idol, while in others it isdisparaged as a possible source of biasand manipulation.Is big datathe answer?In an essay entitled 'The End ofTheory' published in 2008 in Wired theauthor argues that 'with enough data,the numbers speak for themselves' ,and 'correlation supersedes causation,and science can advance even withoutcoherent models, unified theories, orreally any mechanistic explanation atall' 30 . Clearly, this is a radically differentopinion of data and statistics from therefrain popularized by Mark Twain, 'Lies,damned lies, and statistics'.Large datasets combined withalgorithms and IT architectures thatenable machine learning offer enormousopportunities, and we should definitelythink positively of this new world. It istrue that it is sometimes possible topredict well without fully understandingthe underlying mechanisms. This hasbecome our everyday experience inmany parts of our lives (e.g. spamfiltering software, and online serviceoptimization).However we should not disregardthe knowledge accumulated frompast experience with regard to thelimitations of data sources, modeldesign, and statistical expertise. Weshould grasp new thinking and welcomethe contribution of new sources andnew methods but not drop our guard.In particular, there should be healthychallenge when data are used in thefields of social and political research.It is reassuring to see that thevague term 'big data' has already beenchallenged. An article in the FinancialTimes even asked the question 'BigData: are we making a big mistake?' 31Figure EIGHTPeak ofinflatedexpectationsVisibilitySlope ofenlightenmentPlateau ofproductivityTechnologytriggerThrough ofdisillusionmentMaturity27.

Ipsos MORI - Understanding Society July 2014It is easy tomake mistakeswith data.Therefore, wemust not losea healthy doseof skepticism.The Hype Cycle is one useful way tothink about the maturity and adoption oftechnologies and applications like bigdata. 32 The cycle illustrates the usualstages new technologies go throughbefore reaching their most productive.Some of the key big data technologiesare at or beyond the 'peak of inflatedexpectations', and heading downwardto the 'trough of disillusionment'. 33It may be some time before the realbenefits of big data approaches arefelt more broadly.In any case, big 'bad' data isjust bad, as the election of FranklinRoosevelt in 1936 reminds us. Famously,a newspaper using a large but biasedsample source got its prediction wrong,while George Gallup predicted theoutcome more accurately with 800times fewer responses but on a betterselected sample. 34The importanceof knowing moreSo bigger is not always better anddefinitely data does not speak for itselfregardless of how much is available.With the proliferation of data sources,the growing use of data visualization,and the development of data journalism,we must make sure we allow citizensto be informed with the right contextand background when it comes tousing data as evidence. Data shouldnot be considered as having a life of itsown; it must be associated with otherinformation such as the source, how itwas acquired, what precision should beexpected, and any other caveats .An interesting example was thepublic debt to GDP debate. In 2010,Harvard professors Carmen Reinhartand Kenneth Rogoff published paperson public debt and growth. 35 Theysuggested that the average real GDPgrowth declines when the ratio ofpublic debt to GDP is high: from circa4% growth if the ratio is below 30%to a small or slightly negative growthwhen the ratio is above a threshold of90%. This conclusion has been widelypublicized and was used in the politicaldebate both in the US and Europe tojustify controlling public debt and topromote austerity. 36In 2013, a student from theUniversity of Massachusetts Amherstfound errors in the calculation.He discovered that countries at the28.

The data rich societybeginning of the alphabet had beenomitted because of an Excel error,noticed that some observations hadbeen rejected, and challenged thecomputation methods. 37 Together withhis professors they published their ownfindings in April 2013 rejecting the ideaof a sharper decline in growth above thethreshold of 90% for debt to GDP ratio.This has led to numerous comments inthe press and blogs.But there was a much moresubstantial issue here than justdifferences in computation. The basicproblem is that these data were toodispersed to provide a convincing linkbetween the debt to GDP ratio and theBigger is notalways betterand definitelydata do notspeak forthemselvesregardless ofhow much isavailable.rate of GDP growth. With a sufficientlylarge number of observations, you tendto find that variables are correlatedat least a little. However this doesnot provide a convincing link if thedispersion of the data is too large.Statisticians use a coefficient ofdetermination, noted R² (R squared)to measure the linear strength of arelationship between variables.This R squared value always sitsbetween 0 (no correlation) and 1(perfect correlation). In the GDP debtdataset, the R squared value was 0.04.Statisticians would say that 96% of thevariation in growth is 'unexplained' bythe debt to GDP ratio. This is far toolow in terms of explanatory power andcannot lead to convincing conclusions.As for the classic confusion betweencorrelation and causation, assumingcausality can lead even the bestintentioned towards blinkered views.For instance, people do not commitcrimes just because they are pooror belong to certain groups; otherfactors will be important in explainingcriminal behaviour. But in many articlessimple correlations are treated asa comprehensive explanation, andthese important caveats are absent.This can lead to potentially dramaticmisinterpretation. 38 So just as datadoes not speak for itself, neither doescorrelation supersede causation.Lastly, the availability of largedatasets from social networks does notmean that all data should be seen asrepresentative. For instance, there canbe multiple anonymous users or eveninvoluntary selection bias, particularlywhere only partial access to the data isavailable. As such who has access tothe data is also a source of power andinfluence over the political debate.It is easy to make mistakes withdata. Therefore, we must not lose ahealthy dose of skepticism, and alwaystake into account where data comesfrom and how they are processed.Big Data and machine learning offernew avenues of knowledge andreinforce the permanent thirst for validdata and sound interpretations. One ofour challenges will be to fight prejudice,biased, wrong or unethical use of data.As such, data scientists and statisticianswill play a key role in the political andsocial debates to come.29.

Ipsos MORI - Understanding Society July 2014REFERENCES1. http://www.demos.co.uk/files/The_Data_Dialogue.pdf2. http://www3.weforum.org/docs/WEF_IT_RethinkingPersonalData_Report_2012.pdf3. http://www.deloitte.com/assets/Dcom-UnitedKingdom/Local%20Assets/Documents/Market%20insights/Deloitte%20Analytics/uk-da-data-nation-2013.pdf4. http://www.ipsos-mori.com/newsevents/events/120/Personalisation-vs-Privacy-The-Balance-of-Public-Opinion.aspx5. See http://reports-archive.adm.cs.cmu.edu/anon/isri2005/CMU-ISRI-05-138.pdf for more details6. http://www.people-press.org/2013/07/26/governmentsurveillance-a-question-wording-experiment/7. http://conversation.which.co.uk/technology/length-ofwebsite-terms-and-conditions/8. http://www.theguardian.com/world/the-nsa-files9. http://www.pulitzer.org/citation/2014-Public-Service10. http://www.shirky.com/11. http://www.cjr.org/cover_story/evgeny_vs_the_internet.php?page=all12. http://www.pbs.org/newshour/bb/new-police-surveillancetechniques-raise-privacy-concerns/http://www.nap.edu/catalog.php?record_id=1874913. http://prismsproject.eu/14. Belgium, Denmark, Estonia, Germany, Hungary, Portugal,Romania and the UK15. Mydex: http://mydex.org/16. Qiy: https://www.qiy.nl/17. Midata: http://www.midatalab.org.uk/18. I’ve had three main involvements in this field in the past: atDemos in the 1990s we ran some big researchprogrammes on privacy, the Internet and trust which inretrospect were far ahead of their time http://www.demos.co.uk/files/thefutureofprivacyvolume2.pdf. In government Ioversaw the Cabinet Office review of privacy and datasharing in 2002 (which, ironically, can no longer beaccessed on-line). In both cases we concluded that, withthe right safeguards, the public stood to gain more fromthe free flow of personal data than they were losing. Butthe balance of the issue has changed markedly – onereason why more recently I became involved in helping thecreation of Mydex, a project to give the public much morecontrol. A good recent overview of the field which sets outcurrent European thinking is ‘Digital EnlightenmentYearbook 2013: the value of personal data’ published bythe Digital Enlightenment Forum.19. Sasha Issenberg (2013) The Victory Lab: the Science ofWinning Campaigns20. http://realclearpolitics.com/21. http://www.huffingtonpost.com/news/pollster/22. Alex Knapp (2012) http://www.forbes.com/sites/alexknapp/2012/11/07/election-forecaster-sam-wang-onthe-future-of-polling-and-punditry/23. Silver, N. (2012). The signal and the noise: Why so manypredictions fail-but some don't.24. Jackman, S. (2005). Pooling the polls over an electioncampaign. Australian Journal of Political Science, 40(4),499-51725. Abramowitz, A (2008) Forecasting the 2008 presidentialelection with the time-for-change model PS: politicalscience & politics 41: 691-5; Montgomery, J. M.,Hollenbach, F. M., & Ward, M. D. (2012). Improvingpredictions using ensemble Bayesian model averaging.Political Analysis, 20(3), 271-29126. Linzer, D. A. (2013). Dynamic Bayesian Forecasting ofPresidential Elections in the States. Journal of theAmerican Statistical Association, 108(501), 124-134.27. Jackman (2000) Estimation and Inference Are MissingData Problems: Unifying Social Science Statistics viaBayesian Simulation Political Analysis, Vol. 8, No. 4(Autumn)28. Holbrook, T. (1996). Do campaigns matter? (Vol. 1). Sage.29. Hummel, P., & Rothschild, D. (2013). Fundamental modelsfor forecasting elections.30. http://archive.wired.com/science/discoveries/magazine/16-07/pb_theory31. http://www.ft.com/cms/s/2/21a6e7d8-b479-11e3-a09a-00144feabdc0.html#axzz33fzwJ3n830.

References32. http://www.gartner.com/technology/research/methodologies/hype-cycle.jsp33. http://www.informationweek.com/big-data/softwareplatforms/big-data-meets-trough-of-disillusionmentgartner/d/d-id/89893934. http://pollingmatters.gallup.com/2010/10/75-years-agofirst-gallup-poll.html35. http://www.nber.org/papers/w1563936. http://thinkprogress.org/economy/2013/04/16/1875541/11-republicans-who-cited-a-flawed-study-to-push-for-drasticspending-cuts/37. http://www.peri.umass.edu/236/hash/31e2ff374b6377b2ddec04deaa6388b1/publication/566/38. http://www.motherjones.com/kevin-drum/2012/11/singlemothers-now-hook-70s-crime-waveFIGURE References1. Source: Ipsos Global Trends SurveyBase: 16,039 adults across 20 countries, online, 3-17September 2013, data is weighted2. Source: Ipsos Global Trends SurveyBase: 16,039 adults across 20 countries, online, 3-17September 2013, data is weighted3. Source: Ipsos Global Trends SurveyBase: 16,039 adults across 20 countries, online, 3-17September 2013, data is weighted4. Source: Ipsos Global Trends SurveyBase: 16,039 adults across 20 countries, online, 3-17September 2013, data is weighted8. http://www.gartner.com/technology/research/methodologies/hype-cycle.jsp31.

Ipsos MORI - Understanding Society July 201432.

InformationIpsos MORI79-81 Borough RoadLondon SE1 1FYt: +44 (0)20 7347 3000f: +44 (0)20 7347 3800www.ipsos-mori.comwww.twitter.com/IpsosMORIAbout Ipsos MORI’sSocial Research InstituteThe Social Research Institute works closely with national government, local public services and the not-for-profit sector.Its 200 research staff focus on public service and policy issues. Each has expertise in a particular part of the public sector,ensuring we have a detailed understanding of specific sectors and policy challenges. This, combined with our methodological andcommunications expertise, ensures that our research makes a difference for decision makers and communities.