Can online political targeting be rendered transparent? Prospects for campaign oversight using the Facebook Ad Library

Meh a, Somya; E ickson, K is o e A icle Can online poli ical a ge ing be ende ed anspa en ? P ospec s o campaign o e sigh using he Facebook Ad Lib a y In e ne Policy Re iew P o ided in Coope a ion wi h: Alexande on Humbold Ins i u e o In e ne and Socie y (HIIG), Be lin Sugges ed Ci a ion: Meh a, Somya; E ickson, K is o e (2022) : Can online poli ical a ge ing be ende ed anspa en ? P ospec s o campaign o e sigh using he Facebook Ad Lib a y, In e ne Policy Re iew, ISSN 2197-6775, Alexande on Humbold Ins i u e o In e ne and Socie y, Be lin, Vol. 11, Iss. 1, pp. 1-31, h ps://doi.o g/10.14763/2022.1.1648 This Ve sion is a ailable a : h ps://hdl.handle.ne /10419/254286 S anda d-Nu zungsbedingungen: Die Dokumen e au EconS o dü en zu eigenen wissenscha lichen Zwecken und zum P i a geb auch gespeiche und kopie we den. Sie dü en die Dokumen e nich ü ö en liche ode komme zielle Zwecke e iel äl igen, ö en lich auss ellen, ö en lich zugänglich machen, e eiben ode ande wei ig nu zen. So e n die Ve asse die Dokumen e un e Open-Con en -Lizenzen (insbesonde e CC-Lizenzen) zu Ve ügung ges ell haben soll en, gel en abweichend on diesen Nu zungsbedingungen die in de do genann en Lizenz gewäh en Nu zungs ech e. Te ms o use: Documen s in EconS o may be sa ed and copied o you pe sonal and schola ly pu poses. You a e no o copy documen s o public o comme cial pu poses, o exhibi he documen s publicly, o make hem publicly a ailable on he in e ne , o o dis ibu e o o he wise use he documen s in public. I he documen s ha e been made a ailable unde an Open Con en Licence (especially C ea i e Commons Licences), you may exe cise u he usage igh s as speci ied in he indica ed licence. h ps://c ea i ecommons.o g/licenses/by/3.0/de/legalcode Volume 11 | Can online poli ical a ge ing be ende ed anspa en ? P ospec s o campaign o e sigh using he Facebook Ad Lib a y Somya Meh a Uni e si y o Leeds K is o e E ickson Uni e si y o Leeds K.E [email protected] DOI: h ps://doi.o g/10.14763/2022.1.1648 Published: 31 Ma ch 2022 Recei ed: 5 Feb ua y 2021 Accep ed: 24 June 2021 Compe ing In e es s: The au ho has decla ed ha no compe ing in e es s exis ha ha e in luenced he ex . Licence: This is an open-access a icle dis ibu ed unde he e ms o he C ea i e Commons A ibu ion 3.0 License (Ge many) which pe mi s un es ic ed use, dis ibu ion, and ep oduc ion in any medium, p o ided he o iginal wo k is p ope ly ci ed. h ps://c ea i ecommons.o g/licenses/by/3.0/de/deed.en Copy igh emains wi h he au ho (s). Ci a ion: Meh a, S. & E ickson, K. (2022). Can online poli ical a ge ing be ende ed anspa en ? P ospec s o campaign o e sigh using he Facebook Ad Lib a y. In e ne Policy Re iew, 11(1). h ps://doi.o g/10.14763/2022.1.1648 Keywo ds: Campaigns, In e ne o e sigh , T anspa ency, Ad e ising, Poli ical echnology, Poli ical mic o- a ge ing, Online pla o ms Abs ac : A ecen policy de elopmen has been olun a y sel - egula ion o in e ne pla o ms h ough he es ablishmen o online ad a chi es. Since 2018, conce n o e he po en ial misuse o pla o ms has led o calls o e o m o online campaign moni o ing. In esponse, Google, Facebook and Twi e made a ailable eposi o ies o poli ical ad e isemen s appea ing on hei espec i e pla o ms. The in ended p omise was o make he p ocess o poli ical ad e ising less o a ‘black box’ and ende o e a ge ing mo e anspa en o public e iew. In his pape , we conside whe he he Facebook Ad Lib a y ac ually imp o es he capabili y o egula o s and he public o o e see online campaigns. Speci ically, we analyse a co pus o ads ocusing on B exi in he lead-up o he Eu opean Pa liamen a y Elec ions in 2019, o de e mine whe he hese da a a e meaning ul compa ed o epo ing o o line campaign ac i i y al eady equi ed unde UK Elec o al Commission ules. We examine some 234 indi idual ad campaigns un du ing a 14-day pe iod leading up o he elec ion using da a collec ion ools a ailable ia he Ad Lib a y in e ace. A con en analysis o indi idual ads combined wi h da a ob ained om he a chi e abou he ad sponso and demog aphic each sugges s a a coa se le el o de ail ha mic o- a ge ing has aken place. Howe e , limi a ions o he Ad Lib a y p e en i s e ec i eness as an o e sigh mechanism, as epo ing obscu es de ails abou o e all spend, each and a ge ing beha iou , key issues o online poli ical ad e ising. Based on hese indings and he me hodological challenge o in e oga ing he Facebook Ad Lib a y, we e lec on he policy e ec i eness o supposedly anspa en ad a chi es as a policy ool. Issue 1 1. In oduc ion: online poli ical ad e ising A co e issue in deba es abou he “pla o misa ion” o he web, is he ex en o which use s and egula o s can adequa ely in e oga e and obse e how pla o ms shape he low and con en o in o ma ion hey ca y (Gillespie, 2010; Helmond, 2015; Pasquale, 2015; Schwa z, 2017; Riede & Ho mann, 2020). Regula o s and he public may wish o obse e he wo kings o pla o ms o a a ie y o easons. Fo example he e a e conce ns ha beha iou al a ge ing could be used o popu- la e use s’ eed wi h con en ein o cing ce ain poli ical opinions, o ha use p o- iles could be sold o unsc upulous hi d pa ies. The opaqueness o pla o m ope - a ion p esen s challenges o esea che s wishing o be e unde s and he pa e ns o in o ma ion consump ion online. Pla o ms a e opaque o esea che s in a num- be o impo an ways: hey o en lack access o in o ma ion abou how il e ing al- go i hms ope a e, as his can be a closely-held ade sec e . Resea che s ha e also been limi ed in he ex en o which hey can e e se-enginee he wo kings o al- go i hms wi hou eliable da a abou he ou comes o algo i hmic il e ing on a la ge scale, a p oblem ha has led o de elopmen o bo om-up so wa e ools and da abases (Edelson e al., 2019; Sil a e al., 2020). This pauci y o in o ma ion has led o accusa ions ha in e ne pla o ms lack accoun abili y because hey de- ny ci izens and use s he oppo uni y o e iew he ules ha go e n hei ope a- ion (McIn y e & Sco , 2008; Pe el & Elkin-Ko en, 2017; Be aldo e al., 2021). These p oblems a e pa icula ly acu e in he case o poli ical ad e ising, whe e schola s ha e poin ed ou he inc easingly da a-d i en na u e o poli ical cam- paigns (Howa d e al., 2005; Howa d, 2006; He sh & Scha ne , 2013; Nicke son & Roge s, 2014). By all accoun s, he use o da a in poli ical ad e ising has only in- ensi ied: i exempla “hype media” campaigns (Howa d, 2006) o he 1990s em- ployed digi al echnology o iden i y and map s a egically impo an g oups o swing o e s, ad ancemen s in p edic i e a ge ing o messages o e social media ha e heigh ened he s akes o da a-d i en poli ical campaigning in ecen yea s. As Nicke son and Roge s (2014, p. 53) desc ibe, “con empo a y poli ical campaigns amass eno mous da abases on indi idual ci izens and hi e da a analys s o c ea e models p edic ing ci izens’ beha iou s, disposi ions, and esponses o campaign con ac .” Wi h he mig a ion o poli ical ad e ising on o social ne wo ks, he abili- y o asce ain o e disposi ions has u he been enhanced. The 2018 Facebook/ Camb idge Analy ica scandal d ew public a en ion o he possibili y ha highly- de e mined g oups could wield disp opo iona e in luence by analysing and le e - aging da a se s ob ained om social media apps. The scandal came o p ominence in Ma ch 2018, when a Channel 4 (2018) s ing ope a ion and whis le-blowe 2 In e ne Policy Re iew 11(1) | 2022 Ch is ophe Wylie exposed he B i ish poli ical consul ing i m o using in o ma- ion om 50 million Facebook p o iles in a majo b each o da a, o shape he ou - come o Donald T ump’s ic o y in he 2016 US P esiden ial Elec ions (Cadwallad e al., 2018). The use p o iles accessed by Camb idge Analy ica con ained in o - ma ion such as he age, gende , geog aphic loca ion, likes, dislikes and social con- nec ions o indi iduals, making i possible o e ine s a is ical models o beha iou u he han could be accomplished wi h coa se da a, he eby making poli ical mi- c o- a ge ing mo e conce ning. One esul o he Camb idge Analy ica saga was ha Facebook es ic ed access o APIs ha had been used by legi ima e e- sea che s o s udy he pla o m, u he cen alising he pla o m’s con ol o e ac- cess o da a (B uns, 2018; Be aldo e al., 2021). In esponse o public inqui ies in o hei poli ical ad e ising businesses, Google, Facebook and Twi e each es ablished poli ical ad a chi es in 2018. These a chi es we e b oadly simila in design; hey each o e ed publicly-accessible eco ds o ad e isemen s o a poli ical na u e ha an on hei pla o ms, includ- ing in o ma ion abou he sponso o he ad. Facebook de ines poli ical ads as hose ha “ e e ence poli ical igu es, poli ical pa ies, elec ions, legisla ion be o e Pa liamen and pas e e enda ha a e he subjec o na ional deba e” (Allan & Lea he n, 2018, n.p.). Facebook’s Ad Lib a y was la e expanded o include moni- o ing o addi ional speci ic issues including housing, employmen and c edi op- po uni ies. The Ad Lib a y is p oposed by Facebook as a esponse o public con- ce ns abou he po en ial o manipula ion o o e s using a ge ed ad e ising ia he social ne wo k, bu does i achie e his aim? The e a e inc easing calls o e o m campaign o e sigh ules o imp o e he ca- paci y o au ho i ies and he public o obse e and egula e online poli ical ad e - ising (Domme & Powe , 2019; Ma ge s & Domme , 2020; Neude , 2020). Cam- paign moni o ing ules, such as hose se ou by he Elec o al Commission in he UK, a e in o med by adi ional conce ns ela ed o o line poli ical communica ion, and lack he equi ed ocus and nuance o con on digi al poli ical ad e ising. Fo example, epo ing equi emen s ha e no accu a ely cap u ed spending on digi al campaigns because he e is no equi emen o di e en ia e be ween digi al and non-digi al spending (Domme & Powe , 2019, p. 259). T adi ional epo ing also does no adequa ely cap u e beha iou al a ge ing o messages. Fu he mo e, we know om esea ch in he USA con ex ha spending on indi idual Facebook cam- paigns is small, wi h much ac i i y below he h eshold o epo ing se by he Elec o al Commission (Edelson e al., 2019). In Eu ope hese issues we e aised as pa o he Eu opean Democ acy Ac ion Plan (‘EDAP’) p esen ed by he Commission 3 Meh a, E ickson in Decembe 2020, in pa icula e e ing o a “need o mo e anspa ency in poli - ical ad e ising” as well as mo e s ingen ules on he use o pe sonal da a when a ge ing poli ical messages (Eu opean Commission, 2021, p. 3). T anspa ency is a key componen o p oposed e o ms: speci ically, he e a e calls o mo e anspa - en iden i ica ion o he sou ce o poli ical messages, mo e g anula de ail abou spending on online ac i i ies, and inc eased isibili y on a ge ing p ac ices, pa - icula ly when combined wi h da a p o iling capabili ies o online pla o ms. Two key ques ions abou he ole o ad a chi es p esen hemsel es: does he Facebook Ad Lib a y ac ually succeed a making anspa en he beha iou o po- li ical ad e ise s? And secondly, i he e a e sho comings in he Ad Lib a y e- po ing, how can hese be imp o ed o enable meaning ul egula o y o e sigh o poli ical ad e ise s as laid ou in he p oposed e o ms? This a icle add esses hose wo ques ions by unde aking and epo ing on an empi ical in es iga ion o poli ical ad campaigns un in he Uni ed Kingdom in 2019. The pu pose o his in- es iga ion is o asce ain he ex en o which ex e nal publics (like us) can e i y he sou ce o messages, unde s and he spend and each o campaigns, and iden i- y he p esence o a ge ing beha iou by s udying he Ad Lib a y. Since a majo conce n o obse e s is he possibili y ha ad e ise s can a ge speci ic g oups o indi iduals, we in es iga e whe he we can in e speci ic a ge ing beha iou om he da a p esen ed by he Ad Lib a y which include o al each, demog aphics o he audience and geog aphic loca ion o use s. The a icle is s uc u ed as ollows: we b ie ly e iew he concep o anspa ency as a mechanism o in e ne egula ion o si ua e he poli ical ad a chi e ini ia i es wi hin b oade ajec o ies o policy-making. We discuss c i iques o he ans- pa ency app oach o egula ion, ocusing on wo main lines o inqui y: i s , he abili y o anspa ency ini ia i es o ende in o ma ion meaning ully isible o publics and second, he socially embedded and complex na u e o in o ma ion sha ing wi hin o ganisa ions. We hen p oceed o desc ibe he me hods o ou em- pi ical in es iga ion o he Facebook Ad Lib a y, be o e epo ing on he esul s o ha s udy. We ind ha he Facebook Ad Lib a y a o ds e y limi ed capabili y o obse e ad a ge ing beha iou , due o design ea u es which limi he use ulness o da a ha can be ob ained by bo h he web and API e sions o he ool. We con- clude by discussing he implica ions o hese indings o he e ec i e egula ion o online poli ical ad e ising. 4 In e ne Policy Re iew 11(1) | 2022 2. Li e a u e e iew: anspa ency, isibili y and poli ical ac ion The poli ical ad a chi es de eloped by Facebook, Twi e and Google a e con inu- ous wi h a model o anspa en sel -go e nance ha has a long his o y in policy app oaches o he in e ne (F eedman, 2016; Lee ssen e al. 2019). The e ec i e- ness o his egula o y model es s on assump ions abou he ex en o which anspa ency can, on one hand, e ec i ely ende p ac ices isible and on he o he hand, be use ully employed by a ious ac o s o hold pla o ms and ad e ise s o accoun . A key deba e among schola s o in e ne policy has been he ex en o which sel - egula ion o online communica ion se ices is app op ia e o bes se e he public in e es , and wha al e na i e o ms egula ion migh ake (Lessig, 2009). As de- sc ibed by Cu an (2016), go e nance o he in e ne has p oceeded h ough se e - al phases in which di e ing o ms o egula ion ha e been dominan . Du ing he ea ly de elopmen o he in e ne , scien i ic no ms o openness and in o ma ion sha ing a ou ed anspa ency and collabo a ion be ween pa icipan s. Wi h la e comme cialisa ion and ma ke enclosu e, limi ed access o p op ie a y code be- came a mo e dominan ea u e o web se ices (Lessig, 2009). While code may ha e been ende ed mo e opaque compa ed o open sou ce and collabo a i ely buil al e na i es, he egula o y app oach s ill a ou ed by comme cial ac o s on he in e ne has has ended owa ds sel -go e nance and a de e al o echnologi- cal openness, a leas wi h ega ds o enginee ing s anda ds (Cu an, 2016). These app oaches a e consis en wi h a p e e ence by comme cial pla o ms o laissez- ai e ma ke -based egula ion as hey in i e minimal di ec in e e ence om he s a e (F eedman, 2016). One example o quasi- anspa en sel - egula ion is he Wo ld Wide Web Conso ium (W3C), he main s anda ds o ganisa ion esponsible o de elopmen o echnical p o ocols o he web. W3C consis s o issue- ocused wo king g oups whose membe ship is open o academic, public sec o and com- me cial ep esen a i es, and whose mul i-s akeholde decision-making p ocess is o en published anspa en ly o public e iew (Do y & Mulligan, 2013). A p ede- cesso o he poli ical ad a chi e a e anspa ency epo s published by Google and o he companies, which summa ise ac ions aken by he companies in esponse o eques s o emo al o in o ma ion, such as copy igh akedown no ices he Digi- al Millennium Copy igh Ac (DMCA) o go e nmen eques s made o na ional secu i y pu poses. Publishing his in o ma ion may help o in o m he public o he comme cial ope a o s’ compliance wi h legal obliga ions bu may also se e he unc ion o publicising po en ial su eillance o e each by go e nmen au ho i ies 5 Meh a, E ickson (Losey, 2015). T anspa ency has also been a co e o ganising p inciple o hacke subcul u es (‘In o ma ion wan s o be ee’) as well as o collabo a i e p ojec s like Wikipedia and open sou ce so wa e de elopmen eams (Anannay & C aw- o d, 2018, p. 977). T anspa ency is egula ly applied in esponse o eme gence o new online social ha ms. Fo example, he UK Ad e ising S anda ds Au ho i y (ASA) has p oduced guidelines encou aging in luence s o be mo e anspa en abou paid sponso ship in pos s and o he online ad e ising (ASA, 2018). Non- b oadcas p in poli ical ad e ising in he UK is egula ed by he Elec o al Com- mission, which equi es ha p in ads anspa en ly bea he iden i y o he o igi- na ing pa y. Gi en he exis ing egula o y ocus on anspa ency, i is expec ed ha online poli ical ad e ising (and o he ha ms ela ed o opaci y o digi al pla - o ms) would be app oached simila ly. Online poli ical ad a chi es a e he e o e a logical ex ension o a egula o y model ha has been a a ious imes he domi- nan app oach o in e ne go e nance, al hough i emains o be seen whe he na- ional policymake s and publics will iew hese anspa ency measu es as su i- cien o add ess public conce ns. Wi h ecen e en s highligh ing he po en ial o manipula ion o poli ical opinion online, c i ics ha e a gued ha sel - egula ion ails o p o ide adequa e checks and balances o sa egua d he public in e es in a democ a ic online public sphe e (Zalnie iu e & Milan, 2019; Lee ssen e al., 2019). Some legal schola s sugges ha e o s o egula e pla o ms such as Ge many’s Ne zDG law and simila ap- p oaches may ha e unin ended chilling e ec s on eedom o exp ession and could lead o unwan ed p i a ised en o cemen by en i ies like Facebook, due o o e -en o cemen (Coche, 2018; Two ek & Lee ssen, 2019). In he UK, he DCMS (2019) epo on misin o ma ion u ges g ea e anspa ency as a key aim o egula- ion in he ealm o poli ical disin o ma ion, s a ing, “[w]ha does need o change is he en o cemen o g ea e anspa ency in he digi al sphe e, o ensu e ha we know he sou ce o wha we a e eading, who has paid o i and why he in o ma- ion has been sen o us.” (2019, p. 5). Ano he key ecommenda ion o he epo is ha any independen egula o should be g an ed he powe o obse e he algo- i hmic ope a ion o social media companies o gain in o ma ion om hem abou da a usage p ac ices, as well as he a ge ing o poli ical messages (2019, p. 96). The e ec i eness o me e isibili y as a sel - egula o y measu e may be o e s a ed (Riede & Ho mann, 2020). T anspa ency p oposals es on se e al assump ions abou in o ma ion, o ganisa ions and publics which a e no ully empi ically e i- ied. D awing on his li e a u e, we add ess se e al o he key assump ions abou anspa ency below. These a e: he assump ion ha ini ia i es ende hei objec s ully isible, and he assump ion ha publics can and will be able o moni o and 6 In e ne Policy Re iew 11(1) | 2022 ac on he in o ma ion p esen ed. 2.1 T anspa ency and isibili y When applied o he managemen o ins i u ions, anspa ency has o en se ed as a “ empla e o sc ip ” o o ganisa ional esponse o public calls o accoun abili y (Edwa ds, 2020). In he con ex o poli ical e o m, anspa ency has been deployed as a eac ion o p oblems such as bu eauc a ic opaci y, cap icious and unaccoun - able public au ho i y, co up ion and nepo ism (Lessig, 2009). T anspa ency has been associa ed wi h eelings o ai ness, p oo ha ins i u ions a e wo king in he public’s bes in e es , and educ ion o hos ili y and suspicion owa ds o ganisa- ions (Fai banks e al., 2007, p. 29). A co e assump ion is ha hese ou comes a e a di ec esul o inc eased isibili y. As Ananny and C aw o d a gue, “[ ]he implici assump ion behind calls o anspa ency is ha seeing a phenomenon c ea es op- po uni ies and obliga ions o make i accoun able and hus o change i . (2018, p. 974). Howe e , he e a e po en ial limi s o he e ec i eness o anspa ency in p omo ing he public in e es , ela ed on one hand o he abili y o he public o ac e ec i ely on he basis o in o ma ion, and on he o he hand o he in luence o a ious ac o s – including o ganisa ions hemsel es – on he in o ma ion ha is made isible. 2.2 C i iques: anspa ency as a complex and nego ia ed social p ocess C i iques o anspa ency ini ia i es ha e d awn a en ion o he way ha such e - o s a e embedded in complex social ela ions. These app oaches, “emphasize he complexi y o communica ion and in e p e a ion p ocesses and ocus on he com- plica ions and pa adoxes gene a ed by anspa ency p ojec s” (Albu & Fly e bom, 2019:281). Fo example, anspa ency applied o algo i hmic o AI decision-making p ocesses may be incomple e i only one aspec o he p ocess is made isible, since “ he challenge is one o ela ions be ween da a and algo i hms, eme gen p ope ies o he machine lea ning p ocess, e y likely o be uniden i iable om a e iew o he code” (La sson & Hein z, 2020). Fens e (2015) a gues o a mo e nu- anced eading o he complex decision-making p ocesses wi hin o ganisa ions, and he impac o his complexi y on he abili y o o ganisa ions o main ain sec ecy o anspa ency. O ganisa ions a e “unwieldy and incohe en , sec e i e and ye leaky, se led and e e -changing” (2015, p. 158). So, while “clea and hones ” (Fai banks e al, 2007) communica ion o objec i e ac s may be a poli ical ideal and guiding aim, ac ual mechanisms o anspa ency may ail o be clea o hones in ac uali y. These goals may be impeded by ins i u ional ailings, opposing incen i es o pow- 7 Meh a, E ickson e ela ionships be ween a ious ac o s. The socially embedded app oach is a en i e o he possibili y ha pla o ms like Facebook may be esis an o anspa ency, may beha e in ways ha p e en mechanisms om ope a ing ideally, o may be incapable o implemen ing ans- pa ency measu es. T anspa ency can in en ionally occlude, such as when impo - an in o ma ion is hidden in he ‘de i us’ o a high olume o eleased in o ma- ion (Ananny & C aw o d, 2018, p. 979). E en esea ch conduc ed wi h go e nmen employees and ci il se an s has e ealed media ing e ec s ha can dis o ans- pa ency (An hopoulos e al., 2016). In o ma ion is o en media ed h ough p o es- sional communica o s, wi h bo h pe sonal and p o essional o ien a ions ha can shape anspa ency beha iou (Fai banks e al., 2007, p. 29). O ganisa ional leade - ship migh wi hhold sensi i e in o ma ion om o he s a , including communica- o s o epu a ional o s a egic easons (Fens e , 2015). Indeed, anspa ency i - sel may be pa o a b oade hyb id publici y- anspa ency s a egy designed o inc ease ma ke sha e o imp o e he s a us o he o ganisa ion (Edwa ds, 2020). In hei in e iews wi h go e nmen wo ke s, Fai banks e al. ound ha espon- den s ea ed anspa ency could be misused o dis o ed by bad ac o s o show he o ganisa ion in a poo ligh , causing hem o be cau ious abou eleasing in o ma- ion. Recogni ion o he socially-embedded na u e o anspa ency ini ia i es ale s us ha “o ganiza ional decisions, poli ical in e es s, and con lic ing iewpoin s may unde mine ideals o in o ma ional quali y and quan i y” (Albu & Fly e bom, 2019, p. 286). Addi ionally, he con es ed poli ical s a us o whis leblowe s means ha hose who sha e p o ec ed in o ma ion, e en when ac ing in he public in e - es , may be h ea ened when doing so, despi e no ma i e appeals o anspa ency in o he domains (Fens e , 2015). E en i in o ma ion is anspa en ly p esen ed o he public, his may no esul in e ec i e poli ical ac ion. O en ope a ing wi hin a “ma ke place o ideas” pa adigm (Coe, 2015), anspa ency ini ia i es assume a public ha is “ eady, willing, and able o ac in p edic able, in o med ways in esponse o he disclosu e o […] in o - ma ion” (Fens e , 2015, p. 152). A su plus o da a emo ed om con ex , wha Lessig (2009b) calls “naked” anspa ency, can lead o a si ua ion in which a mul i- ude o co ela ions could be d awn om he same in o ma ion dump. An o e sup- ply o anspa ency can esul in poli ical pa alysis: “We need o see wha compa - isons he da a will enable, and whe he hose compa isons e eal some hing eal.” (Lessig, 2009b). When in o ma ion is anspa en ly eleased o he public domain, he e is no gua an ee ha ecipien s will beha e in a ionally-assumed ways. Ma - ke -based heo ies o exp ession assume ha “in o ma ion is easily disce nible and 8 In e ne Policy Re iew 11(1) | 2022 ADVERTISER ESTIMATED AD SPEND (£) 38 Deg ees 5950 Conse a i es 900 Lo ing Eu ope 2 750 I Lo e EU 450 Sco ish En i onmen 450 Bes o Doncas e 300 Al hough he me ic o ad imp essions does no p o ide he numbe o unique iews pe ad, making i a a he a skewed measu emen o he ac ual po en ial each o a pa icula ad, i is he mos widely used indica o o cos -e ec i eness o an online ad e ising campaign. Facebook, in pa icula , uses a CPM (cos - pe -1,000 imp essions) me ic (Ads Help Cen e, n.d.). Ou da a show ha “Lea e” campaign ads incu ed he highes CPM o £13.89, while he lowes CPM (£4.95) was obse ed o neu al ads wi h no disce nible poli ical s ance. These da a sug- ges , as pe obse a ions by Edelson e al. (2019) ha i may be mo e cos ly o each speci ically a ge ed audiences wi h poli ical messages on Facebook, o ha he a ge ing me hods used a y ac oss poli ical ac o s and yield di e en esul s in e ms o imp essions and engagemen . While limi ed by he epo ing o agg ega e da a in anges o o al spend and im- p essions, he Facebook Ad Lib a y da a p o ides some insigh in o he online campaign ac i i y ha is no cu en ly a ailable om o line epo ing. The UK Elec o al Commission o e sees poli ical campaigns including hose o he 2019 EU Pa liamen a y elec ions and se s spending limi s o campaigns (Elec o al Com- mission, n.d. a). These limi s apply du ing he “ egula o y pe iod” o 4 mon hs in ad ance o he elec ion. The limi s a y be ween Sco land, Wales and No he n I e- land, bu o England hey depend on he numbe o MEPs e u ned o a gi en e- gion. Pa ies and non-pa y campaigne s mus epo expenses ela ed o a ange o ac i i ies, which include i ems like unsolici ed ma e ial sen o o e s, ad e is- ing using lis s (online o o line), ma ke esea ch and allies o e en s (Elec o al Commission, n.d. a, p. 12). The headline ac i i ies, concei ed in a p e-digi al wo ld, do no p o ide meaning ul in o ma ion abou he in ensi y o esou ces de o ed o online ac i i ies in any o he ca ego ies which may include a ange o media ypes. Fo example, since Facebook does no p o ide in o ma ion abou how ads we e a ge ed, i is impossible o know whe he a gi en online ad was aimed a a p ede e mined lis o o e s o was unsolici ed. Fu he mo e, ad e ising and e- po ing limi s es ablished in he p in wo ld may no adequa ely co espond o he 15 Meh a, E ickson in luence possible wi h a ge ed online ad e ising a a signi ican ly lowe cos . The Elec o al Commission h eshold o epo ing in England is £20,000. In he Facebook con ex i is possible ha smalle ad e ise s could each a signi ican digi al audience wi h poli ical messages wi hou ha ing o epo ha ac i i y. We ound a p eponde ance o e y small campaigns (<£100) in ou sample, indica ing he possible p esence o poli ical ac o s no picked up by adi ional moni o ing ules. 4.1 Who is paying o online poli ical ad e isemen s ela ed o ‘B exi ’ on Facebook? Ou analysis o Ad sponso s e eals ha a wide ange o di e en poli ical ac o s made use o online ad e ising on Facebook. The main ac o s ha made use o po- li ical ad e ising ela ed o “B exi ” on Facebook’s pla o m, o he wo-week ime pe iod, we e poli ical pa ies, poli ical issue o ganisa ions, non-go e nmen o gan- isa ions, ci izen communi y g oups and poli ical candida es. The ca ego ies used o compa e ad sponso s we e de i ed om assigned ca ego ies p o ided on he ad- e ise s’ Facebook pages as well as by u he checking sponso websi es o con- i m he ca ego ies. These ca ego ies, while coa se, align wi h expec ed poli ical ac o s ha unde ake o line me hods o poli ical ad e ising, as indica ed by p e- ious esea ch (ICO, 2018a), bu wi h some excep ions, discussed below. This e- sea ch encoun e ed some o -p o i businesses and o he non-candida e indi idu- als ha did no ha e a di ec ela ionship o he Pa liamen a y elec ions, and we e g ouped as “o he s”. Edelson e al. (2019) also obse ed in hei s udy o online po- li ical ad e ising in he USA ha he Facebook Ad A chi e con ained “a wide b ead h o poli ical ad ypes and sponso s,” han ypically moni o ed, including comme cial pa icipan s ha c ea ed poli ically-cha ged ads as well as o ganisa- ions wi h no clea in en . In he 2016 P esiden ial elec ion in he USA, hese in- cluded a Russian company named he In e ne Resea ch Agency (IRA) (Ribei o e al. 2019; Sil a e . al., 2020). 16 In e ne Policy Re iew 11(1) | 2022 FIGURE 2: Ad imp essions by sponso ca ego ies FIGURE 3: Ad spend by sponso ca ego ies Figu e 2 shows he dominance o poli ical pa ies accoun ing o he la ges amoun o Facebook ad imp essions (60.2%) and he highes ad spend (72.9%) as compa ed o he es o he poli ical ad e ise s. In he case o poli ical o ganisa- ions and non-go e nmen o ganisa ions, one can also see ha hei imp essions a e highe in he p opo ion o hei ad spends (Figu es 2 & 3), and in he case o poli ical pa ies, i is he opposi e. As discussed abo e, he CPM a es o poli ical- ly- a ge ed messages we e highe han o poli ically neu al messages, as hey a e o poli ical pa ies compa ed o NGOs and o he o ganisa ions. The highe CPM 17 Meh a, E ickson a es sugges he p esence o a ge ing, as i may cos mo e o a ge ha d- o- each demog aphics and audiences in mo e con es ed ma ke s (Domme & Powe , 2019, p. 263). A lack o isibili y in o he way ha Facebook and o he online pla - o ms p ice hei ads means he e is po en ial o algo i hmic bias, which may un- in en ionally occu i ad e ising p ac ices o a ge audiences a e di e en be- ween poli ical ac o s, esul ing in highe p icing. O e all, he highe CPM a es o poli ical pa y messages sugges s ha his ype o ad e ising is comme cially a - ac i e o online pla o ms. FIGURE 4: Ad imp essions by sponso ca ego ies o “Lea e” ads We u he examined whe he he ype o poli ical sponso s di e ed by B exi s ance. Acco ding o he esul s shown in Figu e 4, 93.1% o he o al imp essions c ea ed by “Lea e” Ads we e om Poli ical Pa ies, as opposed o 53.5% in he case o “Remain” ads. In he case o “Remain” ads, he imp essions we e mo e e enly dis ibu ed amongs poli ical pa ies and poli ical o ganisa ions, indica ing he p e alence o bo h pa y and non-pa y o ganisa ions campaigning o “Remain”, as opposed o he pa y-based dominance isible wi hin he “Lea e” ads (Figu e 5). Howe e , an o e all analysis clea ly indica es ha poli ical pa ies a e he mos p e alen ca ego y o poli ical sponso s o online Facebook ads ela ed o B exi , o e he cou se o wo weeks ha a e assessed in his esea ch. 18 In e ne Policy Re iew 11(1) | 2022 FIGURE 5: Ad imp essions by sponso ca ego ies o “Remain” ads The B exi Pa y, a new p o-Lea e pa y led by Nigel Fa age, was he op ad e ise by imp essions and ad spend, ollowed closely by Libe al Democ a s, a p o-Remain pa y (See Table 2). Lowe spend on he B exi issue by Labou and he Conse a- i es is no able. Online ad pla o ms like Facebook may p o ide smalle poli ical pa ies wi h an oppo uni y o conduc a ge ed communica ions on speci ic issues a lowe cos , making online poli ical ad e ising a mo e signi ican o m o cam- paigning o hese o ganisa ions. 4.2 Con en o online poli ical ad e ising ela ed o B exi : wha is being shown? Some 80% o he ads con ained in ou sample con ained simple images wi h ex elemen s. A u he 10% con ained ideo con en while 10% we e “ca ousel” s yle ads. The ads con ained in ou sample displayed a ange o di e en image y and used a ange o he o ical echniques ( o one example see Figu e 7). They also d i ed away om B exi in o o he ela ed poli ical opics as well as o he poli ical con es s happening a he na ional and Eu opean scale (see Figu e 6 below). 19 Meh a, E ickson FIGURE 6: Addi ional opics co e ed in B exi ela ed ads on Facebook Ad Lib a y Compa ed o adi ional poli ical ad e ising in p in o ele ision, some o he con en obse ed in he sample con ained ex eme o e nacula language ha made i ha de o iden i y he con en as a poli ical ad e isemen . Fo example, one “Lea e” ad used a e nacula po man eau o pain he suppo e s o emain- ing in he EU as “moane s”: “Remoan a e an i-democ a ic, like he e il empi e hey se e,”. Sugges ing he in e na ional, c oss-bo de na u e o online poli ical mes- sages, ano he ad implo ed o e s o “Make B i ain G ea Again”. Some ads consis - ed o images o p o-Lea e poli ical candida es pain ed as clowns. Ads om “Peo- ple’s Vo e UK”, a p o- emain poli ical o ganisa ion, an mul iple ad campaigns wi h a ack image y. These po ayed Bo is Johnson, using a “Saa chi and Saa chi” aes- he ic s yle o poli ical ad e ising, eminiscen o he “New Labou , New Dange ” campaign in he 1997 Gene al Elec ion (Scammell and Lange , 2006). One issue aised by he UK Elec o al Commission and also appa en in ou s udy is ha ads may no ha e a clea a ge in e ms o a speci ic elec o al ace o e e endum in he UK, and a e no labelled as such by Facebook. Consequen ly, campaign spend- ing moni o ing may be di icul , as ads can se e mul iple pu poses and each di - e en elec o a es simul aneously. The cu en sys em o ad spend moni o ing e- qui es ha campaigns epo hei in oices agains a speci ically iden i iable poli i- cal con es wi hin a p ede e mined egula o y ime pe iod o he moni o ing body (Elec o al Commission, 2021), bu his may inadequa ely e lec he con inuous and dynamic online poli ical ad e ising en i onmen ha is eme ging.. 20 In e ne Policy Re iew 11(1) | 2022 FIGURE 7: Example o a p o-“Remain” ad and me ics p o ided by he Facebook Ad Lib a y The majo i y o he ads analysed (70%) included a call o ac ion. The mos com- mon call o ac ions we e “Sign Up” and “Sign Pe i ion”, bo h o which consis ed o links o he ad e ise ’s websi e, asking he use o p o ide hei pe sonal in o ma- ion. While he use may no wish o sign up, o he calls o ac ion also included “liking” he ad e ise 's page, o o op -in o ecei e u u e communica ions om he ad e ise . Ano he call o ac ion included “Take he quiz”, which was used only by one ad e ise , 38 Deg ees, o pe suade use s o ake he pla o m’s quiz on B exi , which appea ed designed o ob ain p ecise B exi - ela ed use in o ma ion and pe sonal da a o u u e use. 4.3 Who is being shown online poli ical ad e isemen s ela ed o B exi on Facebook? We nex examined he co pus o ads o de e mine whe he i was possible o de- ec e idence o a ge ing om he da a made a ailable abou campaigns in he Facebook Ad Lib a y. Demog aphic p o iling has become a common p ac ice o online ad e ise s, o e ed h ough dynamic ad e ising ea u es such as “Looka- like” and Cus om Audiences on Facebook which ma ch ads o indi iduals based on ce ain selec ed c i e ia such as age, gende , in e es s o beha iou (Faizullabhoy & Ko olo a, 2018). Facebook p o ides e y limi ed in o ma ion abou a ge ing in i s Ad Lib a y. I does no e eal whe he ad e ise s ha e used any selec i e c i e- ia o a ge an Ad, only gene al in o ma ion abou he gende , age and geog aphic 21 Meh a, E ickson loca ion o people who we e shown he ad. The Ad Lib a y does no e en e eal, a a me a le el, whe he any a ge ing has aken place, o example whe he he ad- e ise used an uploaded lis o a Lookalike audience. In e ms o poli ical mes- saging, schola ship in poli ical communica ion has iden i ied ha a he mos coa se le el, age and gende a e co ela ed wi h di e en o ing p e e ences in p e ious poli ical con es s (Bu le and S okes, 1974; K osnick & Alwin, 1989; No - is, 1996). I may hen be expec ed o see poli ical campaigne s on Facebook using a leas hese cha ac e is ics o age and gende o a ge ecipien s based on hei an icipa ed p e e ences. FIGURE 8: Es ima ed ad imp essions by gende and age ange based on B exi s ance Figu e 8 shows he demog aphic (gende and age) b eakdown in numbe o im- p essions. A pa e n o di e ence acco ding o bo h gende and age, in ela ion o poli ical messages om bo h “Lea e” and “Remain”, is appa en . “Remain” ads we e shown o younge , emale Facebook use s. Con as ingly, “Lea e” ads skewed o- wa ds olde demog aphics (45+ yea s), wi h an inclina ion owa ds male audiences. B exi polling showed ha younge o e s and emales we e mo e likely o sup- po he “Remain” posi ion (Ipsos 2016). Ou indings sugges ha i Lea e/Remain ads we e a ge ed, he s a egy may e lec a goal o appealing o exis ing poli ical belie s a he han a emp ing o change exis ing poli ical p e e ences. Howe e , since he Ad Lib a y does no e eal wha a ge ing c i e ia ad pu chase s selec ed when building hei campaigns, we canno be ce ain whe he he obse ed pa - e n e lec s a ge ing in en ion on he pa o ad e ise s o pe haps Facebook’s own in e nal il e ing o accoun s mo e likely o espond posi i ely o he ads be- ing shown. Ano he possibili y is ha some o he , unobse ed a iable chosen by 22 In e ne Policy Re iew 11(1) | 2022 ad pu chase s when designing hei campaigns is also co ela ed wi h age and gende cha ac e is ics o a ge ed use s, con ounding he esul s. Finally, we e alua ed whe he i was possible o in e geog aphic a ge ing om he obse ed pa e n o imp essions. While geog aphic-le el da a a e coa se, he Ad Lib a y shows na ional-le el popula ions being a ge ed wi hin he coun ies o England, Sco land, Wales and No he n I eland. Figu e 9 shows ha 87.9% o he o al imp essions we e loca ed wi hin England, indica ing an England-cen ic geo- g aphical sp ead o B exi ads shown on Facebook. In he 2011 census, England made up 84% o he o al popula ion o he UK. Simila ly o age and gende demo- g aphics, he geog aphic loca ion is epo ed in he Ad Lib a y as a pe cen age, a he han absolu e numbe so he a e age pe cen ages o each o he 4 coun ies we e used o p oduce he cha in Figu e 9. While analysing ads based on geo- g aphical sp ead, we obse ed ha he Ad Lib a y da a o some o he o he coun- ies, like he US and India, con ains mo e de ailed insigh s abou geoloca ion han he UK. In bo h o hose coun ies, he imp ession pe cen ages a e b oken down a he s a e le el, allowing sligh ly highe esolu ion geog aphic analysis. Howe e , as wi h he abo e discussion, We don’ know om he da a made a ailable by Facebook whe he geog aphic loca ion was a ac o in a ge ing speci ic ads abou B exi in he 2019 EU Pa liamen a y elec ions. FIGURE 9: Geog aphic dis ibu ion o ads based on a e age numbe o ad imp essions pe campaign 4.4 Limi a ions o he Facebook Ad Lib a y F om he pe spec i e o poli ical communica ion esea ch, he Ad Lib a y does p o- ide some use ul da a o esea che s and public obse e s abou he ads displayed on he pla o m, and help ully collec s hem in one place. The Lib a y p o ides an 23 Meh a, E ickson indica ion o he iden i y o he o ganisa ion ha bough a pa icula ad (al hough no a ull add ess o iden i ica ion numbe ), he ime pe iod he ad was shown and a ough es ima e o he o al ad spend and each. These da a do no ex end be- yond wha is cu en ly collec ed by he UK Elec o al Commission, whe e he ocus has been on policing he o al amoun spen on ad e ising by poli ical candida es and o ganisa ions in an elec ion. The Facebook Ad Lib a y does cap u e smalle poli ical ad e ise s who migh no mee he h eshold o epo ing o he Elec- o al Commission. The use ulness o he Lib a y is signi ican ly limi ed by epo - ing only anges a he han absolu e numbe s o money spen and numbe o peo- ple eached. I is likely ha hese igu es a e obscu ed because o conce n by Face- book ha hey would e eal in o ma ion abou how he company p ices hei ads. Exac CPM a es could be wo ked ou and compa ed ac oss di e en ad buye s i exac numbe s we e epo ed. As p e iously discussed, he agg ega ion o da a e- po ed by he Ad Lib a y in unequal in e als poses challenges o s a is ical analysis. Pe haps mo e impo an ly, he Ad Lib a y o e s no addi ional insigh abou he po- en ial o a ge ing, beyond b oad epo ing on imp essions by age, gende and wide geog aphic loca ion. This is disappoin ing conside ing he le el o de ail a ailable o Facebook bo h om i s Ad Builde ool, used by ad buye s o a ge speci ic demog aphics beyond basic ca ego ies like age and gende , and om Face- book’s knowledge abou he beha iou o use s ha leads o hem being shown pa icula ads. A pla o m like Facebook ha has in es ed hea ily in use expe i- ence on i s social pla o m could be expec ed o be able o p o ide he same le el o sophis ica ion in i s ad a chi e, and doing so would conside ably imp o e he p ospec s o moni o ing online campaigns. Ano he key limi a ion o he Facebook Ad Lib a y is he lack o con ex su ound- ing online poli ical ad e ising. Use s do no iew hese ads in isola ion, de oid o any con ex . The digi al ads o m a key pa o an indi idual's pe sonalised News Feed on Facebook, which Adshead e al. (2019, p. 25) desc ibe as being “in eg a ed in o he su ounding con en in a non-in e up i e way, ollowing he o m and unc ion o he use expe ience in which i is placed”. “In- eed” ad e ising hen becomes a way o con ex ualise p omo ional con en amids he ne wo k’s cul u al a e ac s, ca e ully combining “ alues o a en ion, popula i y, and connec i i y” ( an Dijck, 2013, p. 62). In o ma ion ha would be o impo ance o esea che s o poli ical mic o- a ge ing include: he ads ha a use was shown p e iously and subsequen ly, he le el o engagemen and a en ion gi en by a use o a speci ic ad, and any o he p io beha iou s o ele ance, such as ollowing a ela ed poli i- 24 In e ne Policy Re iew 11(1) | 2022 Two ek, H., & Lee ssen, P. (2019). An Analysis o Ge many’s Ne zDG Law (T ansa lan ic High Le el Wo king G oup on Con en Mode a ion Online and F eedom o Exp ession) [Wo king Pape ]. h p s://www.i i .nl/publica ies/download/Ne zDG_Two ek_Lee ssen_Ap il_2019.pd Van Ba el, J. J., Ra hje, S., Ha is, E., Robe son, C., & S e nisko, A. (2021). How social media shapes pola iza ion. T ends in Cogni i e Sciences, 25(11), 913–916. h ps://doi.o g/10.1016/j. ics.2021.07.01 3 an Dijck, J. (2013). The Cul u e o Connec i i y: A C i ical His o y o Social Media. Ox o d Uni e si y P ess. h ps://doi.o g/10.1093/acp o :oso/9780199970773.001.0001 Wang, Z., Hale, S. A., Adelani, D., G abowicz, P. A., Ha mann, T., Flöck, F., & Ju gens, D. (2019). Demog aphic in e ence and ep esen a i e popula ion es ima es om mul ilingual social media da a. h ps://doi.o g/10.48550/ARXIV.1905.05961 Zalnie iu e, M., & Milan, S. (2019). In e ne a chi ec u e and human igh s: Beyond he human igh s gap. Policy & In e ne , 11(1), 6–15. h ps://doi.o g/10.1002/poi3.200 in coope a ion wi hPublished by 31 Meh a, E ickson