A Data Protection Framework for Learning Analytics · A Data Protection Framework for Learning Analytics Andrew Cormack, Jisc, UK [email protected] ABSTRACT: Most studies

(2016).Adataprotectionframeworkforlearninganalytics.JournalofLearningAnalytics,3(1),91–106.http://dx.doi.org/10.18608/jla.2016.31.6

ISSN1929-7750(online).TheJournalofLearningAnalyticsworksunderaCreativeCommonsLicense,Attribution-NonCommercial-NoDerivs3.0Unported(CCBY-NC-ND3.0) 91

A Data Protection Framework for Learning Analytics

AndrewCormack,Jisc,[email protected]

ABSTRACT:Moststudiesontheuseofdigitalstudentdataadoptanethicalframeworkderivedfrom human-subject research, based on the informed consent of the experimental subject.However,consentgivesuniversitieslittleguidanceonusinglearninganalyticsasaroutinepartofeducationalprovision:whichpurposesarelegitimateandwhichanalysesinvolveanunacceptableriskofharm.Obtainingconsentwhenstudentsjoinacoursewillnotgivethemmeaningfulcontrolovertheirpersonaldatathreeormoreyearslater.Relyingonconsentmayexcludethosemostlikely to benefit from early intervention. This paper proposes a new framework based on theapproachusedindataprotectionlaw.Separatingtheprocessesofanalysis(pattern-finding)andintervention(pattern-matching)givesstudentsandstaffcontinuingprotectionfrominadvertentharm during data analysis. Students have a fully informed choice whether or not to acceptindividual interventions. Organizations obtain clear guidance: how to conduct analysis, whichanalysesshouldnotproceed,andwhenandhowinterventionsshouldbeoffered.Theframeworkprovidesformalsupportforpracticesalreadybeingadoptedandhelpswithseveralopenquestionsinlearninganalytics,includingitsapplicationtosmallgroupsandalumni,automatedprocessing,andprivacy-sensitivedata.Keywords:Learninganalytics,privacy,dataprotection,consent,legitimateinterests

1 INTRODUCTION

Learninganalyticshasbeendefinedas“themeasurement,collection,analysisandreportingofdataaboutlearnersandtheircontexts,forpurposesofunderstandingandoptimizinglearningandtheenvironmentsinwhich it occurs” (Long& Siemens, 2011, p. 33). Analyzing learner data could inform a university’smanagementoffinances,resources,andenrollment;enhanceitscoursepresentationandmaterials;orenablepersonalizedguidanceandinterventionforindividualstudentsandstaff(Leece,2013).Analyticscouldimproveboththeprovisionoflearningtofuturestudentsandthesupportandguidanceofferedtocurrentteachersandstudents,bothatthecohortandattheindividuallevel.ExamplessuchasPurdueUniversity’sCourseSignalsshowhowdata frompaststudentscanhelpnewundergraduatesmakethetransitionfromschooltouniversitylearning—“likesittingnexttosomebodywhohasbeenthroughtheclassbefore”(Mathewson,2015).Tickle(2015)reportsthatanalysisofcurrentinteractionscantriggertimelysupportbeforeastudent’sproblembecomesinsurmountable.TheNationalUnionofStudents(2015)identifythe“massivepowerandpotential”totacklechallengesfacingUKhighereducation,butconcernsarealsoraisedabout“studentsundersurveillance”(Warrell,2015).Analyticscan,indeed,useofaverywiderangeofbothhistoricandcurrentpersonaldata“fromformaltransactions(such as assignment submissions) to involuntarydata exhaust (such as building access, system logins,keystrokesandclickstreams)”(Kay,Korn,&Oppenheim,2012,p.9).AsinotherapplicationsofBigData,



thetechnologyitselfmaybeethicallyneutralbuttheuseofitisnot(Wen,2012).UniversitiesandtheirstudentsneedananswertoTickle’squestion:“Howcomprehensiveandintrusiveshoulddatacollectionbe?”Todate,learninganalyticshaslargelybeenconductedwithinanethicalframeworkoriginallydevelopedfor medical research. Individuals’ informed consent provides the foundation for all analysis andintervention.However,aslearninganalyticsbecomesanormalpartofeducationalprovision,thispapersuggests that a different ethical basis is required. Treating large-scale learning analytics as a form ofhuman-subjectresearchmaynolongerprovideappropriatesafeguards.InsteadthepapersuggeststhatthebroaderapproachtakenbyEuropeanlawtoprotectpersonaldataoffersclearerguidance,helpingbothorganizationsandstudentsassesswhenandhowanalyticsshould,andshouldnot,beused.Differentmeasuresmay be needed at different stages of an activity. In particular separating the processes ofanalysisandinterventionprovidesclearerguidanceandstrongersafeguardsforboth.Examplesaregivenof how this new ethical framework can inform questions commonly raised in learning analytics. Thisapproach should help learning analytics processes to be trusted by both students and organizations,deliveringthegreatestbenefitforcurrentandfuturestudents,teachers,educationalorganizations,andsociety.2 THE NEED FOR A LEARNING ANALYTICS FRAMEWORK

Learninganalytics sharesmanycharacteristicswith thewell-establishedprocessesofeducationaldatamining(Prinsloo&Slade,2013)andwebsitepersonalization(Pardo&Siemens,2014).Likethem,itseeksto“understandandoptimizelearningandtheenvironmentinwhichitoccurs”(Pardo&Siemens,2014,p.443).However,boththenatureandscaleofdataavailableandtheexpectationsofstudentsandsocietyarechangingtocreatenewopportunitiesandnewrisks.Thismayrequireanewapproachtomaintaintrustbetweeneducationalorganizations,staff,andstudents.Prinsloo and Slade see “the increasingdigitisationof education, technological advances, the changingnatureandavailabilityofdata”(2013,p.244)ascreatingimportantopportunities.Analyzingthesedatacouldhelpresearchersunderstandthefactorsthatcontributeto“theeffectivenessoflearning,studentsuccessandretention”(ibid.)andleadtoimprovementsinthefutureprovisionofeducation.Thedatatrailscreatedbystudentsinteractingwithdigitalsystemsalsoraisethepossibilityofanalysisinnearreal-time.Detailedinformationabout individualstudents’ learningandengagementmightbeusedtooffer“personalized and customized curricula, assessment and support to improve learning and retention”(2013,p.240),thusbenefittingcurrentstudentsaswell.Suchusesofpersonaldatamaybeincreasinglyaccepted,evenexpected,bothbyindividualstudentsandby society.According toKayetal., “[u]sers,especiallyborndigitalgenerations,appear increasingly toexpectpersonalized services thatare responsive toprofile,needand interestandare thereforemorelikelytobecontentfortheirdatatobeusedtothoseends”(2012,p.4),whilePardoandSiemensfind



“societyseemstobeevolvingtowardasituationinwhichtheexchangeofpersonaldataisnormal”(2014,p.440).Learninganalytics,however,byconsumingincreasingquantitiesofvariedandreal-timedata,alsosharescharacteristicswithBigData.Itcreatesthepossibility,identifiedbyMayer-SchönbergerandCukier,“toextractnewinsights”(2013,p.6)thatchangeorganizationsandtheirrelationshipwithindividuals.Inthepast,datahavebeenanalyzedtotestpriorhypothesesthathavebeenassessedagainstethicalguidelines.Beforeanalysis (orevendata collection)begins, all possibleoutcomeshavealreadybeenchecked forethical acceptability. But “big data thrives on surprising correlations and produces inferences andpredictionsthatdefyhumanunderstanding”(Ohm,2014,p.100).AccordingtotheArticle29WorkingParty of European data protection authorities (2014b) the possibility of analysis revealing entirelyunexpectedinformationaboutpeople“raisesimportantsocial,legalandethicalquestions,amongwhich[are]concernswithregardto…privacyanddataprotectionrights.”Forexample,combiningdatasetsfromdifferentsourcescanresult inaccidentalre-identificationof individuals(Ohm,2010)oranunexpectedcorrelationbetween two factorsmight suggestacausal link thatdoesnot in factexist.Thismaybeaparticular challenge ineducation.ContrastingwithBigData collected in the retail sector, forexamplethroughloyaltycards,PardoandSiemensconsiderthat

Educational institutionsposeanewscenariowithspecific requirements.Students interactveryintensively with the university (or its computational platforms) during a concrete time frame,carryingoutveryspecifictasks,andproducehighlysensitivedataduringtheprocess.Thesespecialconditionsprompttheneedforarevisionoftheprivacyprincipleswithrespecttoanalyticsandtheirapplicationineducationalsettings.(2014,p.448)

Buttherelationshipbetweentheindividualandtheorganizationisalsoverydifferent.Universitiesandtheirstudentshavealong-termrelationshipandastrongmutualinterestinimprovinglearningprocesses.Givenasuitableethicalframework,itshouldbepossibleforuniversitiestouselearnerdatasafely,inwaysthatbenefitbothstudentandorganizationfarmorethanasimpleeconomicbargainexchangingshoppinghabitsfordiscountvouchers.Itisclearthat,asPardoandSiemenssuggest,inlearninganalytics“adelicatebalancebetweencontrolandlimitsneedstobeachieved”(2014,p.440).However,PrinslooandSladeworrythat“currentpolicyframeworksdonotfacilitatetheprovisionofanenablingenvironmentforlearninganalyticstofulfilitspromise” (2013, p. 240). Those frameworks have largely been based on practice in human-subjectresearch.ThusKayetal.findtheNurembergCode’sprinciplethat“researchparticipantsmustvoluntarilyconsenttoresearchparticipation”is“highlyrelevant”toanalytics.Theyregardconsentas“fundamental”(2012,p.20).AccordingtoSclater,“informedconsentisrecognizedaskeytotheanalysisoflearnerdatabymanycommentatorsand isabasicprincipleofscientificresearchonhumans”(2014,p.16).AttheUniversityofSouthAfrica“thePolicyonResearchEthicsmakesinformedconsentandanonymitynon-negotiable”(Prinsloo&Slade,2013,p.242).



HoweverKayetal.notethatinotherfieldsdifferentethicalapproachesareconsidered“self-evidentgoodpractice”(2012,p.26).Forexample,

The practice adopted by leading business to consumer players provides a clear and legallygroundedapproachthatislikelytobereadilyunderstoodbythepublicinmuchoftheworld.Inparticular, the development of a sense of mutual gain, recognized and shared by a serviceorganizationanditscustomers,issomethingtobelearnedfromsuch[organizations]asAmazonandNectar.(p.24)

Theyconclude,

Thechallenge iswhether theeducationcommunity,not least in theemerging fieldof learninganalytics,shouldreviseitsethicalpositiononaccountofthewidespreadchangesofattitudeinthedigitalrealmfromwhichlearnersandresearchersareincreasinglydrawn.(p.26)

Thefollowingsectionsofthepaperwillconsiderwhatproblemsarelikelytoarisefromcontinuingtorelyon“informedconsent”as learninganalytics techniques transfer fromuniversity research touniversityoperationsandwhetheramorefruitfulethicalframeworkcanbederivedfromEuropeandataprotectionlaw.3 LIMITATIONS OF “INFORMED CONSENT” Todatelearninganalyticshaslargelybeenasubjectforeducationalresearch.However,thetechniquesareincreasinglybeingadoptedaspartoftheroutineoperationofuniversitiesandcolleges(forexample,OpenUniversity[n.d.]).Suchprocessesmayaffectallcurrentandfuturestudentsandstaff—notjustthosewhoparticipateinresearchstudies—throughchangestohoweducationisprovidedingeneralandthrough specific individual interventions.With this significantly increased impact, “informed consent”maynolongerprovideadequateprotectionandguidanceeitherforindividualsorfororganizations.Both law and ethics require that for consent to be valid itmust be both informed and freely given.1However,inBigDataapproaches“[t]hepotentialvalueofthegathereddatabecomesclearonlyaftertheyaresubjectedtoanalysisbycomputeralgorithms,notbeforehand”(VanderSloot,2015,p.46).Whenusinganinductive,data-drivenapproachratherthanadeductive,hypothesis-drivenone,itishardfortheorganizationtopredictevenwhatcorrelationswillemergefromthedata,letalonewhattheirimpactontheindividualwillbe.RichardsandKingseeasignificantchange:

Before big data, individuals could roughly gauge the expected uses of their personal data andweighthebenefitsandthecostsat thetimetheyprovidedtheirconsent.Even if theyguessedwrong, they would have some comfort that the receiving party would not be able to makeadditionaluseof theirpersonaldata.Thegrowingadoptionofbigdataand itsability tomakeextensive,oftenunexpected,secondaryusesofpersonaldatachangesthiscalculus.(2014,p.414)

1Directive95/46/ECoftheEuropeanParliamentandoftheCouncilontheprotectionofindividualswithregardtotheprocessingofpersonaldataandonthefreemovementofsuchdata(DataProtectionDirective)[1995]OJL281/281,Article2(h).



An educational organization may, and probably should, tell students and staff when it collects theirinformationthatsuchsecondaryuseswillbelimitedtoimprovingeducationalprovision.Howeverthelawrequiresthatforconsenttobevalidindividualsshouldhave“anappreciationandunderstandingofthefactsandimplicationsofanaction…includ[ing]alsoanawarenessoftheconsequencesofnotconsentingtotheprocessinginquestion”(Article29WorkingParty,2011,p.19).Asimplestatementofpurposeisunlikelytoprovidethis.Butattemptingtomakeconsentvalidbybeingmorespecificbeforecollectionoranalysishastakenplaceinvolvessecond-guessingtheresults.Thiscouldstopboththeorganizationanditsstudentsbenefitingfromunexpectedfindings.Consentobtainedatthetimeofjoiningtheorganizationwillonlycoverusesandprocessingthatwere“reasonableexpectations”atthattime(Article29WorkingParty,2011,p.17).Sinceexpectationsofhowdatawillbeusedarechangingveryrapidly,organizationsmight evenhave tousedifferentprocesses for individualswhogave consent at significantly differenttimes.Furthermore,thelawincreasinglypresumesthatconsentisnotfreelygiveninsituationswherethepartyrequestingconsenthassignificantpowerovertheindividualgrantingit.TheEuropeanCommission’sdraftGeneralDataProtectionRegulationstatesthat“[c]onsentshouldnotprovideavalidlegalgroundfortheprocessingofpersonaldata,wherethereisaclearimbalancebetweenthedatasubjectandthecontroller”(2012, Recital 39). Although the Nuremberg Code requires that research subjects “be allowed todiscontinuetheirparticipationatanytime,”Kayetal.observethatineducationitmaybehardtooptout;once analytics has become part of university operations the only way to prevent your data beingprocessedmay,infact,betoleavetheuniversity(2012,pp.20,27).ItishardtoreconcilethiswiththeArticle29WorkingParty’stestthat“[c]onsentcanonlybevalidifthedatasubjectisabletoexercisearealchoice,andthereisnoriskofdeception,intimidation,coercionorsignificantnegativeconsequencesifhe/shedoesnotconsent”(2011,p.12).Atapracticallevel,theuseofconsentmaywellbiastheresultsoflearninganalytics,potentiallyexcludingthosewhohavemosttogainfromtheprocess.Kayetal.notethatwhetherthechoiceispresentedasopt-inoropt-outwillaffect theoutcomebecause“theuser’s inertia”(2012,p.18)willmeanthat themajoritysimplyacceptthedefault.Tothisisaddedtheriskofself-selection:thatdifferentgroupswilloptinoroutindifferentproportions.Thiscouldskewtheresultseitherinfavouroforagainstthosegroups(Clarke&Cooke,1983).Forexample,analyticsoftenaimtohelpthosedisengagedfromtheeducationalprocessbutwillgetlittleinformationaboutthisgroupfromdatacollectedonanopt-inbasis.Consentrequiresindividualstudentsto“takeresponsibilityfortechnologiesandbusinesspracticesthattheydonotthemselvescreatebutfindthemselvesincreasinglydependentupon”(Richards&King,2014,p. 430). Theonly responsibility the lawassigns toorganizations is toensure that individualshave theinformation required tomake their consent valid. From an ethical perspective, this is unsatisfactory:organizations shouldalsobehave responsibly in their adoptionanduseof thenewpractices theyaskstudents to agree to. However using consent as a basis provides little guidance onwhat responsible



behaviourmightbe.Thelawdoesrequirethatprocessingofpersonaldatabe“fair”2butthatis,again,largelyamatteroftheinformationandcontrolsprovidedtoindividuals(InformationCommissioner,n.d.).Aconsentapproachrunstheriskofdiscussionsfocusingonwhethernewanalyticsprocessesandresultsarewithin“thescopeofthedatasubject’sconsent”(Article29WorkingParty,2014a,p.13).Anethicalframeworkshouldinsteadbeconsideringwhetherinferencesmayperhapsbe“inaccurate,discriminatoryorotherwiseillegitimate”(Article29WorkingParty,2013,p.45)andshouldnotbeusedatall.These doubts about the use of consent are not limited to the education context. In discussing thedevelopmentofBigDataoverthenextfiveyears,theEuropeanDataProtectionSupervisoralsoseesthefocusmovingtothebehaviouroforganizations:

Big data that deals with large volumes of personal information implies greater accountabilitytowards the individuals whose data are being processed. People want to understand howalgorithmscancreatecorrelationsandassumptionsaboutthem,andhowtheircombinedpersonalinformationcanturnintointrusivepredictionsabouttheirbehaviour.(2015,p.10)

Thisisaparticularconcerninrelationshipswherevoluntarilyprovidedinformationmaybesupplementedbythat“observedandinferredwithouttheindividual’sknowledge”(ibid.).Herethereisatendencyfor“opaqueprivacypolicies,whichencouragepeopletotickaboxandsignawaytheirrights”(2015,p.11).Ratherthanrelyingonpriorconsent,organizationsshouldbeprovidingclearinformationonhowandwhyinformationcanbeusedandprovidingindividualswithcontinuingopportunitiestodetectandchallenge“mistakesintheassumptionsandbiases[dataanalytics]canmakeaboutindividuals”(ibid.).Suchanapproachseemsmoresuitableforeducationwhere,accordingtoPrinslooandSlade,

Itisacceptedthattherearecertaintypesofinformationandanalyses(e.g.,cohortanalyses)thatfallwithinthelegitimatescopeofbusinessofhighereducation.Thereisthoughanurgentneedtoapproachpersonaldatadifferentlywhen it is used to categorise learners as at-risk, inneedofspecialsupportorondifferentlearningtrajectories.(2013,p.244)

Kayetal.callforaframeworkthatrecognizesthedifferencesbetweenthesetwousesofpersonaldataand provides appropriate protections for both. This would “enable Further and Higher Educationinstitutionstoprogresstheiruseofanalyticswhilstmanagingrisk,”providing“benefittotheindividual,theinstitutionandthewidermissionofeducation”(2012,p.8).Thenextsectionsproposeanewsourceforsuchaframeworkandidentifysomeofthequestionsitanswersandposesabouttheoperationaluseoflearninganalytics.

2DataProtectionDirective,Art.6(1)(a).



4 A DATA PROTECTION FRAMEWORK

Ifhuman-subjectresearchnolongerprovidesasuitableethicalmodelfortheoperationaluseoflearninganalytics, thenanewreferencemodel isneeded.Thispaperproposes that learninganalyticspracticecouldinsteadbeguidedbyEuropeanlaw’sapproachtoprotectingpersonaldata.Whileresearchgivesconsentauniquepositioninauthorizingtheuseofdataabouthumansubjects,underdataprotectionlaw,consentisoneofsixjustificationsforprocessingpersonaldata.ThesearesetoutinArticle7oftheDataProtectionDirectiveandSchedule2oftheUKDataProtectionAct1998and,atleastunderUKandEUlaw(Article29WorkingParty,2011),haveequalstatus.Eachjustificationhasitsownparticular mechanisms to protect the interests of data subjects. The UK Information CommissionersuggeststhatforBigDataanalyticsthemostrelevantjustificationsare“consent,whetherprocessingisnecessaryfortheperformanceofacontract,andthelegitimateinterestsofthedatacontrollerorotherparties”(2014,para.55).3Furthermorethe lawrecognizes(Article29WorkingParty,2011)thatasingletransactionmay involveprocessingunderseveraldifferentjustifications.ThismatchesKayetal.’sdivisionoflearninganalyticsintotwocategories:“[d]atausedforinstitutionalpurposes”and“[d]atausedforpersonalpurposes”(2012,p.10).TheOpenUniversity (n.d.)giveexamples:redesigningmodules“totakeaccountoftopicsseentocauseparticularissuesofunderstanding”versus“identify[ing]pointsonthestudypathwhereindividualsorgroupsmayneedadditionalsupport.”IndiscussingBigData,theArticle29WorkingPartymakesthesame distinction between processing to “detect trends and correlations in the information” versussituations where “an organisation specifically wants to analyse or predict the personal preferences,behaviourandattitudesofindividualcustomers,whichwillsubsequentlyinform‘measuresordecisions’thataretakenwithregardtothosecustomers”(2013,p.46).Differentsafeguardsareseenasappropriate:for trend analysis, “functional separation is likely to play a key role” in protecting individuals (ibid.)whereasforindividualinterventionsopt-inconsentwillnormallyberequired.Data protection therefore suggests that an ethical framework should treat learning analytics as twoseparatestages,usingdifferentjustificationsandtheirassociatedwaysofprotectingindividuals:

• the discovery of significant patterns (“analysis”) treated as a legitimate interest of theorganization,whichmustincludesafeguardsforindividuals’interestsandrights;and

• the application of those patterns tomeet the needs of particular individuals (“intervention”),whichrequirestheirinformedconsentor,perhapsinfuture,acontractualagreement.

The following sections examine the implicationsof this separation, andhowexisting guidanceon thelegitimateinterestsandconsentjustificationscanfurtherinformtheethicalframeworkandpracticeoflearninganalytics.

3Respectively,Articles7(a),7(b),and7(f)oftheDirectiveandclauses1,2,and6oftheAct’sSchedule2.



4.1 Analysis Like most other processing of personal data the “legitimate interests” justification requires that theindividualbeinformedwhatdatawillbeprocessed,bywhomandforwhatpurpose(s)(DataProtectionDirective,Art.10;DataProtectionAct1998,Sch.1PartII2(3)).Thepurposeoftheprocessingmustbealegitimateinterestoftheorganizationandtheprocessingmustbenecessarytoachievethatinterest:inother words no “less invasive means are available to serve the same end”; unlike the “consent”justification,however,theindividual’sagreementdoesnotneedtobeobtained(Article29WorkingParty,2014a).Instead,individualsareprotectedbytherequirementthattheorganization’sinterestmustnotbeoverriddenbytheindividual’sinterestsorfundamentalrights(DataProtectionDirective,Art.7(f)).4Anindividual with “compelling legitimate grounds relating to his particular situation” may object toprocessingifadifferentbalanceofinterestsappliesinhiscase(DataProtectionDirective,Art.14(a)).TheArticle 29Working Party describe the legitimate interests justification as involving “an additionalbalancing test, which requires the legitimate interests of the controller … to beweighed against theinterestsorfundamentalrightsofthedatasubjects”(2014a,p.48).TheWorkingPartyrecognizesthata“broadrange”ofinterestsmaybelegitimate,including“thebenefitthatthecontrollerderives—orthatsocietymightderive—fromtheprocessing,”solongastheclaimedinterest“correspondswithcurrentactivitiesorbenefitsthatareexpectedintheverynearfuture”(2014a,p.24).However,astheInformationCommissionerconfirms,organizations“needtobeabletoarticulateattheoutsetwhytheyneedtocollectandprocessparticulardatasets.Theyneedtobeclearaboutwhattheyexpecttolearnorbeabletodoby processing that data” (2014, para 73). Since theWorking Party quotes as legitimate a business’s“interestingettingtoknowtheircustomers’preferencessoastoenablethemto…offerproductsandservicesthatbettermeettheneedsanddesiresofthecustomers”(2014a,p.26),thereseemslittledoubtthat the interests of a university or college in improving its educational provision would qualify aslegitimate.Thattheseinterestsaresharedbysocietyandcurrentandfuturestudents isrecognizedas“addingweight”totheinterest(2014a,p.35).Forexample,auniversitymightwishtomakeonemoduleapre-requisiteforanotherifanalyticsindicatedthatthiscombinationproducedbetterlearningoutcomes.Howeverlegitimateinterestsdonotgivecarteblancheforanykindofanalysis:theWorkingPartywarnsagainst “creat[ing] … complex profiles of the customers’ personalities and preferences without theirknowledge”(2014a,p.26)sincethisinterferencewithindividualrightscouldnotbejustified.Havingidentifiedandarticulatedalegitimateinterestoftheorganization,thebalancingtestrequiresthatthisbeweighedagainstpossibleharmto“allrelevantinterests”ofindividuals(Article29WorkingParty,2014a,p.29).Relevantinterestsinclude,butarenotlimitedto,theirfundamentalrights.TheWorkingParty’sguidanceonreducingtheriskofharmisparticularlyhelpfulforhowtheanalysisstageoflearninganalyticsshouldbeconducted.

4TheUKDataProtectionActtransposesthisas“prejudicetotherightsandfreedomsorlegitimateinterestsofthedatasubject”[Sch.26(1)].



First, theremust be a clear “functional separation” between the analysis and intervention processes:“data used for statistical purposes or other research purposes should not be available to ‘supportmeasures or decisions’ that are taken with regard to the individual data subjects concerned (unlessspecificallyauthorizedbytheindividualsconcerned)”(Article29WorkingParty,2014a,p.30).Thisislikelytoinvolveorganizationalmeasures—suchasconfidentialityagreementsandguidancetostaffwithaccesstorawdata—aswellastechnicalonestoprotectusers’privacyduringanalysis.Thequantityofdatacollected,andaccesstothatdata,mustbelimitedtowhatisrequiredtoachievethestatedpurpose.TheWorkingPartynotethatthis“isparticularlyrelevant…toensurethatprocessingofdatabasedon legitimate interestswill not lead to anundulybroad interpretationof thenecessity toprocessdata”(2014a,p.29).Technicalmeasuresshouldbeputinplacetoreducetherisktoindividualusers.Anonymizedorstatisticaldatashouldnormallybeusedforanalysis.Wherefullanonymizationisnotpossible,directlyidentifyinginformationsuchasnamesorstudentnumbersshouldbereplacedwithkey-codeswith the index linkingkeys to individuals stored separatelyand,preferably,under separatecontrol(Article29WorkingParty,2013).TheWorkingPartynotethat

Theuseoflessriskyformsofpersonaldataprocessing(e.g.,personaldatathatisencryptedwhileinstorageor transit,orpersonaldata thatare lessdirectlyand less readily identifiable)shouldgenerallymeanthatthelikelihoodofdatasubjects’interestsorfundamentalrightsandfreedomsbeinginterferedwithisreduced.(2014a,p.42)

Safeguardsthat“unquestionablyandsignificantlyreducetheimpactsondatasubjects”may“chang[e]thebalanceofrightsandintereststotheextentthatthedatacontroller’slegitimateinterestswillnotbeoverridden”(Article29WorkingParty,2014a,p.31),thusallowingprocessingtoproceed.Forexample,successfulcombinationsofmodulescannotbe identifiedusingfullyanonymizeddata,asthisanalysisrequireseachstudent’sperformancetobelinkedacrossmodules.However,thesequencesofmodulesandresultscanbeconstructedandanalyzedusingkey-codeddata,orevenone-wayhashedidentifiers, so that there is no index thatwould allowan identifier to bedirectly linked to a student.Combinedwithorganizationalcontrolstopreventre-identification,thisinvestigationshouldeasilysatisfythe balancing test. If the organizationwishes to intervenewith those individualswho have chosen acombinationlikelytobeunsuccessful,thisshouldbeaseparateprocessfromtheanalysis,conductedafterobtainingconsentasdescribedinthenextsection.Anotherfactorthatmaytipthebalanceiswhetherprocessingmatchesusers’reasonableexpectationsofhowtheirdatawillbeused:“themoreclearlyacknowledgedandexpecteditisinthecommunityandbydatasubjectsthatthecontrollercantakeactionandprocessdatainpursuitoftheseinterests,themoreheavily this legitimate interestweighs in thebalance” (Article29WorkingParty,2014a,p.35).Thatauniversityorcollegewoulduseanalyticsto improvetheeducationalservices itprovides—benefitting



boththestudentswhosedataareanalysedandtheirsuccessors—shouldnotbeunexpected.Indeed,asnotedbyKayetal.(2012),currentstudentsmaypositivelyexpectthattheservicestheyreceivewilladapttotheindividual’sactivity,solongastheirinterestsandrightsareprotected.InaJiscworkshoptoidentifyappropriateusesofanalytics,studentsproposedprovidingpersonalizedsuggestionstothosewhowerehavingdifficultyengagingwithcoursematerial(Sclater,2015).An ethical framework may take account of both positive and negative impacts on individuals whenassessingthebalanceofrightsandinterests.However,dataprotectionlawidentifiesuniversities’closerelationshipwiththeirstudentsandemployeesascreatingparticularrisksofharmfrominappropriateprocessing(Article29WorkingParty,2014a).Ratherthanruntheserisks,ifanactionmaydirectlyaffectindividuals—eveninwaysthatareexpectedtobebeneficial—itisusuallybettertoseektheconsentofthoseindividuals.4.2 Intervention Theanalysisstagecanprovideagreatdealofinformationtohelpuniversitiesenhancetheireducationalprovision. Improvements to university facilities and processes, recruitment, courses, and teachingmaterialsaremostlikelytobebasedonaggregatedoranonymizeddata.Thediscussionabovesuggeststhat an ethical framework concentrating on continuing safeguards and the balance of interests willprovidebothbetterprotectionforindividualsandmorecompletedatatoinformsuchchanges.However, learning analytics can also be used to support and guide individual students and staff. Forexample,studentsmightbeofferedpersonalizedreadinglistsbasedonhowtheirprogresscompareswithothers’(Sclater,2015).Heretheuniversity’saimistohaveapositiveeffectontheindividualsoanethicalframework based onminimizing individual impact is no longer appropriate. Instead, interventionwillnormallybebasedontheindividual’sconsent.Thepossibilityofinterventionbeingarequirementofacontractoralegaldutyisdiscussedfurtherbelow.Again,discussionofconsentinEuropeanlawprovideshelpfulethicalguidanceforwhenandhowthisshouldbedone. Inparticular,consentmustbe“freelygiven,specificandinformed”(Article29WorkingParty,2011,p.5).Consentwill only be freely given, according to the Article 29Working Party, if there is “no risk of…coercionorsignificantnegativeconsequencesif[theindividual]doesnotconsent”(2011,p.12).Sinceauniversityorcollegehastheultimatepowertograntorwithholditsstudents’qualifications,thereisariskthatstudentswillfeelcompelledtoaccept.Interventionsshouldthereforebeofferedasachoicebetweenreceivingpersonalizedsupportorstandardeducationalprovision.Avirtuallearningenvironmentmightinvite users to enable “people like you…” suggestions; studentsmight be offered classes or seminarsappropriatetotheir learningstyle.Suchchoicesshouldavoid“significantnegativeconsequences”andtherebycontributetovalidconsent.TheWorkingPartyexplainsthatforconsenttobeinformed“allthenecessaryinformationmustbegiven



at themoment the consent is requested, and that this shouldaddress the substantiveaspectsof theprocessingthattheconsentisintendedtolegitimise”(2011,p.9).Tobespecific,consent“shouldreferclearlyandpreciselytothescopeandtheconsequencesofthedataprocessing”(Article29WorkingParty,2011,p.17).Theserequirementsaremucheasiertomeetafteranalysishasidentifiedapatternthatmaysuggestaparticularbeneficialintervention.Studentscannowbegivendetailedandpreciseinformationonwhattheinterventionisintendedtoachieve,andontheimplicationsofeithergrantingorwithholdingconsentforit.TheWorkingPartyrecognizethatconsentrequested“‘downstream’,whenthepurposeofthe processing changes” allows information to be provided that “focus[es] onwhat is needed in thespecific context” (2011,p.19). Thus, rather thanobtaininggeneral consent for “appropriate support”whenacoursebegins,waitingtillanalysisofastudent’sperformancehasidentifiedtheirmosteffectivelearningstyleletsthemgivefullyinformedconsenttoaspecifickindofhelp.Anindividual’sconsentmustbesignifiedbysome“indication”(Article29WorkingParty,2011,p.5).SincetheWorkingPartyconsidersthat“[t]henotionof ‘indication’ iswide,but itseemsto implyaneedforaction” (2011, p. 12), universities shouldobtain consent through somepositive actionby the studentratherthanpresumingitfromthestudent’ssilenceorinaction.Studentsshouldnormallybeinvitedtoopt-in to an intervention. If circumstances require an opt-out approach then itmight be argued thatconsentwasobtainedearlier—forexamplewhenthestudentsigneduptothecourseormodule—andthatanopportunity towithdrawthatconsentwasbeingofferedat the timeof the intervention.This,however,wouldrequirethecoursedescriptiontostateclearlythatpersonalizedinterventionswouldbemade. Interventions on this basis could only use personal data that were either stated or obviouslyrelevantat the timeof signingup: the informationprovided to the studentmustbe such that joining“lead[s]toanunmistakableconclusionthatconsentisgiven”(Article29WorkingParty,2011,p.23).Acoursemightstate,forexample,thatskillswouldfirstbeassessedandappropriatetopicsthenchosentoaddressthoseareasneedingimprovement.TheWorkingPartyplacesnospecificlimitonhowlongconsentmaybepresumedtolast.Asingleconsentcould therefore cover a series of interventions (2011, p. 17) so long as the student knows this, theinformationtheyweregivenaboutthescopeandconsequencesofprocessingremainsaccurate,andtheydonotindicatethattheyhavechangedtheirmind.TheWorkingPartydoessuggestthatorganizationsshouldperiodicallyremindeachindividual“oftheircurrentchoiceandoffer[…]thepossibilitytoeitherconfirm or withdraw”; how often such reminders are provided will depend on the “context andcircumstances”(2011,p.20)buttherapidpaceofdevelopmentsinlearninganalyticssuggeststhattheyshouldberelativelyfrequent.Remindersareparticularlyimportantwhereconsenthaseffectoveralongperiod;forexample,whereitwasobtainedatthestartofacourse.Whereastudenthasactivelyoptedout(asopposedtonotoptingin)regularremindersmaybethebestwaytoinformthemoftheoptionsavailable.Ifanorganizationplanstouselearninganalyticstointervenewithindividualmembersofstaff,ratherthanstudents,thenthelawwarnsthatconsentmaynotbeanappropriatebasis.Ifanemployee“mightfear



thathecouldbetreateddifferentlyifhedoesnotconsenttothedataprocessing”(Article29WorkingParty,2011,p.13),thenconsentwillnotbefreelygivensoisnotvalid.Interventionswithstaffshouldthereforebe limitedtothosenecessaryforthepurposeofthe(employment)contractorthosewherethere are sufficient safeguards that, despite the strong presumption to the contrary, genuinely freeconsentcaninfactbeobtainedfromtheemployee(Article29WorkingParty,2011).Finally,shouldalegaldutyrequireauniversitytointervenewithanindividual,bothethicsandlawindicatethatanyuseofpersonaldatamustbelimitedtothatstrictlynecessarytofulfilthatduty.5 APPLYING THE FRAMEWORK

The ethical framework proposed here — particularly the balancing test required by the legitimateinterests justification—appearstomatchthefeelingsofparticipants in learninganalyticsstudies.Forexample,PardoandSiemensfoundthat“theconcernsofusersaboutprivacyvarysignificantlydependingon what is being observed, the context and the perceived value when granting access to personalinformation” (2014, p. 440). Furthermore the balancing test protects all the rights and interests ofindividuals,notjustprivacy:theArticle29WorkingPartynote,forexample,that“[t]hechillingeffectonprotected behaviour, such as freedom of research or free speech, that may result from continuousmonitoring/tracking,mustalsobegivendueconsideration”(2014a,p.37).Thissectionconsidershowtheframeworkmightinformsomespecificquestionsraisedbylearninganalyticspractitioners:historicdata,smallgroups,fullyautomatedprocessing,andsensitivedata.PardoandSiemensidentifyissueswithhistoricdata:“Howlongwillthedatabekept?Willthedatabeusedafterstudentsgraduate?”(2014,p.446).Theyconsiderlong-terminformation“willbehelpfulfortheuniversitytorefineitsanalyticsmodels,trackthedevelopmentofstudentperformanceovermultipleyearsandcohortsorsimplyforinternalorexternalqualityassuranceprocesses”butretainingtoomuch,orfortoolong,maywelldamagetrust.Applyingthebalancingtestoftheframework,thelikelihoodofdirectpersonalbenefittoastudentdecreasesoncetheyhavecompletedtheirmoduleorcourse,long-termretentionincreasestheriskofinformationbecomingout-of-dateorsufferingasecuritybreach.Theuniversity must demonstrate that continued processing generates sufficient benefit to balance thedecreasedbenefitandincreasedrisktotheindividual’srightsandinterests.Inparticular,accordingtotheInformationCommissioner,“[i]forganisationswishtoretaindataforlongperiodsforreasonsofbigdataanalyticstheyshouldbeabletoarticulateandforeseethepotentialusesandbenefitstosomeextent,evenifthespecificsareunclear”(2014,para.75).Theymayalsoneedtoimplementadditionalsafeguards;for example, anonymization rather than pseudonymization (Information Commissioner, 2012). Thus,processingofhistoricdatashouldstillbepossiblegivenasufficientlyclearandstrongjustificationbut,because the direct benefit to the individual is less, the range of acceptable purposes is likely to benarrower than for current students.Where universitieswish to continue to collect data from formerstudents—forexamplefortheHigherEducationStatisticsAgency’s(2015)DestinationofLeaversfromHigherEducationsurvey—thisreliesinanycaseonvoluntaryparticipation,sofree,informedconsenttocontinuedprocessingofrelevanthistoricdatacouldbeobtainedatthesametime.



Kayetal.(2012)notethatpatternsderivedfromsmallnumbersofindividualsrepresentanincreasedriskofaccidentalordeliberatere-identification.Theysuggestaminimumgroupsizeoftwenty,thoughthelaw,insomecircumstances,hasconsideredthataveragesacrossgroupsassmallasfiveindividualscanbedisclosed safely (Information Commissioner, 2012). The legitimate interests balancing test can takeaccountofthevariousfactorsaffectingtheriskofharmfulre-identificationandensurethatallgroupsretainalevelofprotectionappropriatetotheriskandbenefitoftheprocessing.Inmostcases,therewillbeathresholdgroupsizebelowwhichtheriskoutweighsthebenefits.Thisshouldindicatethatsuchfine-grainedprocessingshouldnotcontinue.Oneparadoxhighlightedbythebalancingtestisthat,wherethereisaclearbenefittotheindividualandanegligible risk,automatedprocessingmaybe lessprivacy-invasive than revealingpersonaldata toahumanmediator. Re-ordering the choice of topics or modules based on a student’s performance inpreviouscoursesmightbeanexampleofthiskindofintervention.Underdataprotectionlaw,individualsareentitledtoknowofanyfullyautomatedprocessingthatmayaffectthemandthe logicused(DataProtectionDirective,Art.12(a));theyalsohavetherighttoobjecttosuchprocessing(DataProtectionDirective, Art. 15(1)). The law only requires these “profiling” safeguards where there is no humanmediation(EuropeanCommission,2012,Art.20).However,byrequiringpriornotificationandseekingconsent at the time of any intervention, the framework extends them to cover both automated andmediatedprocesses.PardoandSiemensconsidertherisksofusingprivacy-sensitivedatainlearninganalytics:

Forexample,inahypotheticalscenario,istheimprovementoftheoveralllearningenvironmentavalidreasontorecordtheexactlocationofstudentswithintheinstitutionandshareitwithpeerstofacilitatecollaborativelearning?(2014,p.439)

The lawrecognizesvarious typesofdata that representan increased risk to the individual.Anethicalframework should reflect these and apply appropriate safeguards. For example, location data can beprocessedbasedonlegitimateinterests,butthee-PrivacyDirectivewarnsofsignificantprivacyrisks,soclearbenefitsandstrongsafeguardsmustsatisfytheframework’sbalancingtest.5Anonymizedlocationdatamightbeused,forexample,toimprovethedesignoflearningspacesormakemoreefficientuseofrooms,providedtherewasstrongprotectionagainstre-identificationofindividuals.However,PardoandSiemens’(2014)collaborativelearningexampledirectlyaffectsindividualssotheframework,likeArticle9(1)ofthee-PrivacyDirective,wouldrequirethefree, informedconsentofthe individuals involved. IfinformationiscategorizedbytheDataProtectionDirective(Art.8(1))asSensitivePersonalData—forexample,raceorreligion(butnotlocation)—thenlegitimateinterestscannotbeusedevenifthereisnoeffect on the individual (Data Protection Act 1998, Sch. 3(4)). A university that wished to use thesecategoriesinanalysisshouldobtaintheinformedconsentofeachindividualbeforedatawerecollected.Such information could then only be processed for purposes and in ways envisaged at that time. In

5Directive2002/58/ECoftheEuropeanParliamentandoftheCouncilconcerningtheprocessingofpersonaldataandtheprotectionofprivacyintheelectroniccommunicationssector[2002]OJL201/37Article9.



addition,consent forprocessingSensitivePersonalDatamustbeexplicitandcannotbe inferred fromsomeotheraction(DataProtectionAct,Sch.3(1)).6 CONCLUSION Aslearninganalyticsmovesfromaresearchcontexttobecomeakeytoolinuniversityoperations,theuseofpriorconsentasthesoleethicalbasisappearsinappropriate,asitoffersneithertheinformationnortheguidancethatstudentsandthoseconductinganalysesneed.TheArticle29WorkingPartywarnsthat

Theuseofconsent“intherightcontext”iscrucial.Ifitisusedincircumstanceswhereitisnotappropriate,because theelements that constitutevalid consentareunlikely tobepresent, thiswould lead togreatvulnerabilityand…wouldweakenthepositionofdatasubjectsinpractice.(2011,p.10)

Instead,thepaperproposesanewtwo-stageethicalframework,basedontheapproachtakenbydataprotectionlaw.Analysisof learnerdataisconsidereda legitimateinterestofauniversitythatmustbeconductedunderappropriatesafeguards.Theuniversity’sinterestsmustbecontinuallytestedagainsttheinterestsandrightsofindividuals;interferencewiththoseinterestsandrightsmustbeminimized;analysismustceaseiftheycannotbeadequatelyprotected.Ifanalysissuggestsaninterventionthatmayaffectindividual studentsorstaff, theconsentof those individualsshouldbesought.Since theycannowbeprovidedwith full informationabout thenatureandconsequencesof the intervention, their choice ismuchmorelikelytobeethicallyandlegallysound.The approach should help both universities and students to benefit from developments in learninganalytics.Itavoidsartificiallylimitingtheprovisionanddevelopmentofeducationtowhatwasknownorforeseen at the time when consent was originally obtained. This is particularly important in a fast-developingfield.Itreducestheriskofself-selectionbiasaffectingthedatausedforplanning.Italsooffersstrongprotectionfor individualstudentsandstaffbyprovidingbothclearguidanceontheconductofcurrentandnewanalysesanddetailed,relevantinformationwhenindividualinterventionsareoffered.Allprocessingwillincludesafeguardsappropriatetothelevelofriskstoprivacyandotherinterestsandrights.Studentsandstaffwillbeofferedspecific,meaningful,informedchoicesatthetimewhenthesemayaffectthem.Aboveall,aslearninganalyticsmovesfromtheresearchlaboratoryintothedailybusinessofeducation,theframeworkdrawsattentiontoethicallyandpracticallyimportantquestions:Whattopicsandmethodsareappropriateusesoflearnerdata?Howcanthesecanbeprotected?Howcaninterventionsbestbedesignedtobenefitstudents?Learninganalyticspracticesthataddresstheseissuesaremuchmorelikelytobuildandmaintainthetrustofthosewhorelyonthem.



REFERENCES Article29WorkingParty (2011).Opinion15/2011on thedefinitionof consent (01197/11/ENWP187).

Retrieved from http://ec.europa.eu/justice/data-protection/article-29/documentation/opinion-recommendation/files/2011/wp187_en.pdf

Article29WorkingParty.(2013).Opinion03/2013onpurposelimitation(00569/13/ENWP203).Retrievedfrom http://ec.europa.eu/justice/data-protection/article-29/documentation/opinion-recommendation/files/2013/wp203_en.pdf

Article 29Working Party. (2014a).Opinion 06/2014 on the notion of legitimate interests of the datacontroller under Article 7 of Directive 95/46/EC (844/14/EN WP 217). Retrieved fromhttp://ec.europa.eu/justice/data-protection/article-29/documentation/opinion-recommendation/files/2014/wp217_en.pdf

Article29WorkingParty.(2014b).StatementoftheWP29ontheimpactofthedevelopmentofbigdataon theprotectionof individualswith regard to theprocessingof theirpersonaldata in theEU(14/EN WP221). Retrieved from http://ec.europa.eu/justice/data-protection/article-29/documentation/opinion-recommendation/files/2014/wp221_en.pdf

Clarke,G.,&Cooke,D.(1983).Abasiccourseonstatistics.London:EdwardArnold.EuropeanCommission(2012).ProposalforaRegulationoftheEuropeanParliamentandoftheCouncilon

the protection of individuals with regard to the processing of personal data and on the freemovementofsuchdata(GeneralDataProtectionRegulation)COM(2012)11Final.

European Data Protection Supervisor (2015). Leading by example: The EDPS strategy 2015–2019.Retrievedfromhttps://secure.edps.europa.eu/EDPSWEB/edps/site/mySite/Strategy2015

HigherEducationStatisticsAgency (2015).Destinationsof leavers fromhighereducation in theUnitedKingdomfortheacademicyear2013/2014.Retrievedfromhttps://www.hesa.ac.uk/sfr217

InformationCommissioner(n.d.).Processingpersonaldatafairlyandlawfully(Principle1).Retrievedfromhttps://ico.org.uk/for-organisations/guide-to-data-protection/principle-1-fair-and-lawful/ - fair-processing

Information Commissioner (2012). Anonymisation: Managing data protection risk code of practice.Retrieved from https://ico.org.uk/media/for-organisations/documents/1061/anonymisation-code.pdfInformation Commissioner (2014). Big data and data protection. Retrieved fromhttps://ico.org.uk/media/for-organisations/documents/1541/big-data-and-data-protection.pdfKay,D.,Korn,N.,&Oppenheim,C.(2012,November).Legal,riskandethicalaspectsof analytics in higher education, JISC CETIS Analytics Series, 1(6). Retrieved fromhttp://publications.cetis.ac.uk/2012/500

Leece,R.(2013,July).Analytics:Anexplorationofthenomenclatureinthestudentexperience.Proceedingsofthe16thInternationalFirstYearinHigherEducationConference(FYHE13),7–10July2013,TePapa Tongarewa, Wellington, New Zealand. Retrieved fromhttp://fyhe.com.au/past_papers/papers13/14E.pdf

Long,P.,&Siemens,G.(2011).Penetratingthefog:Analyticsinlearningandeducation.EducauseReview,46(5),31–40.

Mathewson,T.G.(2015,August21).Analyticsprogramsshow“remarkable”results—andit’sonlythebeginning. Education Dive. Retrieved from http://www.educationdive.com/news/analytics-



programs-show-remarkable-results-and-its-only-the-beginning/404266/Mayer-Schönberger,V.,&Cukier,K.(2013).Bigdata:Arevolutionthatwilltransformhowwelive,work

andthink.London:JohnMurray.National Union of Students (2015). Learning analytics: A guide for students’ unions. Retrieved from

http://www.nusconnect.org.uk/resources/learning-analytics-a-guide-for-students-unionsOhm,P.(2010).Brokenpromisesofprivacy:Respondingtothesurprisingfailureofanonymization.UCLA

LawReview,57,1701–1777.Ohm,P.(2014).Changingtherules:Generalprinciplesfordatauseandanalysis.InJ.Lane,V.Stodden,S.

Bender,&H.Nissenbaum(Eds.),Privacy,bigdataandthepublicgood(pp.96–111).NewYork:CambridgeUniversityPress.

Open University (n.d.). Using information to support student learning. Retrieved 9 July 2015 fromhttp://www.open.ac.uk/students/charter/sites/www.open.ac.uk.students.charter/files/files/ecms/web-content/using-information-to-support-student-learning.pdf

Pardo,A.,&Siemens,G. (2014).Ethical andprivacyprinciples for learninganalytics.British JournalofEducationalTechnology,45,438–450.http://dx.doi.org/10.1111/bjet.12152

Prinsloo,P.,&Slade,S.(2013).Anevaluationofpolicyframeworksforaddressingethicalconsiderationsinlearninganalytics.Proceedingsofthe3rdInternationalConferenceonLearningAnalyticsandKnowledge,240–244.http://dx.doi.org/10.1145/2460296.2460344

Richards,N.,&King,J.(2014).Bigdataethics.WakeForestLawReview,49,393–432.Sclater,N.(2014).Codeofpracticeforlearninganalytics:Aliteraturereviewoftheethicalandlegalissues.

Jisc. Retrieved from http://repository.jisc.ac.uk/5661/1/Learning_Analytics_A-_Literature_Review.pdf

Sclater, N. (2015).What do students want from a learning analytics app? [Web log post, April 29].Retrieved from http://analytics.jiscinvolve.org/wp/2015/04/29/what-do-students-want-from-a-learning-analytics-app/

Tickle,L.(2015,June30).Howuniversitiesareusingdatatostopstudentsdroppingout.TheGuardian.Retrieved from http://www.theguardian.com/guardian-professional/2015/jun/30/how-universities-are-using-data-to-stop-students-dropping-out

VanderSloot,B.(2015).Privacyaspersonalityright:WhytheECtHR’sfocusonulteriorinterestsmightproveindispensableintheageof“bigdata.”UtrechtJournalofInternationalandEuropeanLaw,31(80),25–50.http://dx.doi.org/10.5334/ujiel.cp

Warrell, H. (2015, July 24). Students under surveillance. Financial Times. Retrieved fromhttp://www.ft.com/cms/s/2/634624c6-312b-11e5-91ac-a5e17d9b4cff.html

Wen,H.(2012).Bigethicsforbigdata.Radar:InsightAnalysis,andResearchaboutEmergingTechnologies[Web log post, June 11]. Retrieved from http://radar.oreilly.com/2012/06/ethics-big-data-business-decisions.html

Documents

A Data Protection Framework for Learning Analytics · A Data Protection Framework for Learning Analytics Andrew Cormack, Jisc, UK [email protected] ABSTRACT: Most studies