Goldstein Microsoft Research NYC 102 Madison Ave 12th Floor New York NY 10016 dggmicrosoftcom R Preston McAfee Google Strategic Technologies 1600 Amphitheatre Parkway Mountain View CA 94043 prestonmcafeecc Siddharth Suri
Download Pdf - The PPT/PDF document "The Cost of Annoying Ads Daniel G" is the property of its rightful owner. Permission is granted to download and print the materials on this web site for personal, non-commercial use only, and to display it on your personal computer provided you do not modify the materials and that you retain all copyright notices contained in the materials. By downloading content from our website, you accept the terms of this agreement.
etarycostofannoyingads.Therstandmaincontributionofthisworkisthatwemeasurethecompensatingwagedif-ferentialofannoyingads.Thatis,wemeasurehowmuchmoreonemustpayausertodothesameamountofworkinthepresenceofannoyingadscomparedtoinnocuousadsornoads.Thecompensatingdierentialisimportanttomeasurebecauseitcapturessomeofthenegativeeectsofadvertising,whichpublishersneedtoheedasalowerboundwhensettingthepricetorunanad.Inatwo-experimentinvestigation,wecomputethecom-pensatingdierentialforannoyingads.Intherstexper-imentusersrandomlyratedeitherananimatedadoritsstaticcounterpart.Thisdesignshowsthatanimationhasanegativeimpactonuserratings.Forthoseadsthatusersrateasannoyingweaskthemtoexplaintheirthinking.Ananalysisoftheratingsandcommentsyieldsabetterofun-derstandingofwhatusersndannoyingabouttheseads.Thisanalysiswillalsoexhibithowannoyingadsnegativelyaectuserperceptionsofadvertisers.Theseanalysesareadditionalcontributionsofthiswork.Inthesecondexperiment,weusethoseadsidentiedasmoreorlessannoying,alongwiththerecentmethodolog-icalinnovationofToomimetal.[18],toestimatethepayrateincreasenecessarytogenerateanequalnumberofpageviewsinthepresenceofannoyingads,comparedtoinnocu-ousadsornoads.Thisestimateisthecostofannoyingadsinourexperiment.Wechosecategorizingemailsasthetasktoproxyforusingapublisher'ssitebecauseusersei-therimplicitlyorexplicitlyneedtocategorizetheiremailsasspamornotspaminthepresenceofadswhenusingfreeweb-basedemailservicessuchasYahoo!Mail,GMail,andMail.com.Finally,weprovideatheoreticalmodelofhowourempiricalndingscouldaectthedisplayadvertisingindustry,whichisthethirdcontributionofthiswork.2.RELATEDWORKAsmentionedinSection1,weusethemethodologicalin-novationofToomimetal.[18]forcomputingcompensatingdierentials.Toomimetal.conductedaMechanicalTurkexperimentinwhichparticipantsrandomlyexperiencedaneasy,medium,orhardversionofataskatarandomlyas-signedpayrate.Thisallowedtheauthorstocomputehowmuchmoreonewouldhavetopayaworkertodothehardtaskoverthemediumandeasytasks.Theauthorsalsoex-hibitedthistechniqueinanexperimentinwhichparticipantswererandomlyassignedtouseeitheran\ugly"ora\pretty"interfacetodoatask.Wewillusethistechniquetoisolatetheeectoftheadqualityonuserabandonment.Next,wedescribepriorexperimentalworkwhichstudiestheimpactofadqualityonbehavior.DrezeandHussherr[4]conductedanexperimentontheef-fectivenessofdisplayadvertisementsusingeye-trackingtech-nology.Theirconclusion,thatusersrarelyfocusdirectlyonbannerads,isoftenreferredtoasbannerblindness,atermcoinedbyBenway[1].Burkeetal.[2]hadparticipantsper-formvisualsearchtasksinthepresenceofnoads,astaticdisplayad,orananimateddisplayad.Theyfoundthatadsdidreducesearchtime,however,therewasnosignicantdierencebetweenanimatedandstaticads.Perhapsevenmoresurprisingly,theydidaposthoctestwhichfoundthatanimatedadswererememberedlessfrequentlythanstaticads.YooandKim[21]askedasimilarresearchquestion.Theyconductedalarger-scalelaboratoryexperimentinwhichpar-ticipantswererandomlyexposedtowebpageswithadswithnoanimation,slow-movinganimationorfast-movinganima-tion.Theyfoundthatmoreanimationdidincreaseattentiontoads.Moreover,moderateanimationincreasedadrecogni-tionratesandbrandattitudes.Highlyanimatedads,how-ever,decreasedrecognitionratesandbrandattitudes.ThisresultcomplementstheresultsofBurkeetal.[2].YooandKim[21]concludethat,\Webadvertisersshouldbeawareofthepossibilitythatexcessiveanimationcanbackreagainsttheoriginalintentionofeectivecommunication."GoldfarbandTucker[5]conductedaeldexperimentinwhichtheyfoundthatadsthatmatchedthesite'scontentoradsthatwereintrusiveincreasedparticipant'sself-reportedintenttopurchase.However,adsthatwerebothintru-siveandmatchedthewebsite'scontentreducedintenttopurchase.Adswereconsideredintrusiveif,forexample,theyproducedapopupwindow,tookoverthewholescreen,playedmusic,orobscuredthewebpagetext.Theauthorssuggestthatthereasonforthisinteractioneectisthatusersaremoresensitivetotargetedandintrusiveadswhentheproductadvertisedisprivacysensitive.Inthecontextofsponsoredsearch,Buscheretal.[3]foundthatadsthatarerelevanttothesearchtermsreceivedmorevisualatten-tionthanadsthatwerelessrelevant.ThiscomplementstheresultsofGoldfarbandTucker[5]whichwerefoundinthedomainofdisplayadvertising.Takenasawhole,thesestudiessuggesttheremaybeben-etstoasmalldegreeofanimationorintrusivenessinad-vertising,butthattoomuchanimationorintrusivenesscanhaveadetrimentalimpactontheadeectiveness.3.RATINGTHEQUALITYOFADSWenextdescribeourexperiments,bothofwhichwereconductedonAmazon'sMechanicalTurk1,anonlinelabormarket.Sinceitwasoriginallybuiltforjobsthataredif-cultforcomputersbutareeasyforhumans(e.g.,imagerecognition),jobsonMechanicalTurkarecalledHumanIn-telligenceTasksorHITs.TherearetwotypesofpeopleonMechanicalTurk:requestersandworkers.RequesterscanpostHITsandworkerscanchoosewhichHITstodoforpay.AfteraworkersubmitsaHIT,therequestercaneitheracceptorrejecttheworkbasedonitsquality.ThefractionofHITsthataworkersubmitswhichareacceptedisthatworker'sapprovalrating.Thisfunctionsasarep-utationmechanism.TheAmazonAPIgiveseachworkeraccountaunique,anonymousidentier.BytrackingtheIDsoftheworkerswhoacceptedourHITs,wecouldenforcethatparticipantswereonlyallowedtoparticipateinoneofthetwoexperiments,andtheywereonlyallowedtodothatexperimentonetime.ThereisaburgeoningliteratureonconductingbehavioralexperimentsonMechanicalTurk[12,11,16,6,7,20,13,9,17].Inthissetting,theexperimentertakesontheroleoftherequesterandtheworkersarethepaidparticipantsoftheexperiment.MasonandSuri[10]provideahow-toguideforconductingbehavioralexperimentsonMechanicalTurk.Wenowdescribethedesignandresultsofourrstexperiment,whichservedtoidentifysetsofmoreandlessannoyingads 1http://www.mturk.com Figure1:Thetoppanelranksadsbyannoyingnessandshowsthatthe21mostannoyingadswereanimatedandthe24leastannoyingadswerestatic.Thebottompanelrankspairsofadsbytheannoyingnessoftheanimatedvariant.Thestaticvariantstendtofallbelowtheiranimatedversions,suggestingthatanimationincreasesannoyingness,evenwhentheadvertiserandproductareheldconstant.Errorbarsare1standarderror. Figure4:Screenshotoftheemailcategorizationtaskshowingthebadadscondition.Atthebottomofeachemailclassicationpage,partici-pantswereshownhowmanyemailstheyhadrated,theirpayrate,andareviewoftheinstructions.Thefooterincludedtwobuttons:oneallowingthemtosubmitandrateanotheremail,andasecondallowingthemtostopcategorizingandcollecttheirpayment.Participantswereallowedtoclassifyupto1000emails.4.2ResultsLetanimpressionbeoneparticipantviewingoneemail(anditsaccompanyingads,ifany),regardlessofwhethertheparticipantclassiestheemailorquitsbeforeclassifyingit.Sinceanemailispresentedassoonastheuseracknowledgestheinstructions,eachofthe1223participantsgeneratedatleastoneimpression.Theoveralldistributionofimpressionsperpersonisskewedwithameanof61,amedianof25andrstandthirdquartilesof6and57.Beingboundedby1frombelowandeectivelyunboundedfromabove(onlytwoparticipantsreachedtheupperlimit),theseimpressionsconstitutecountdata.Inparticular,theyareoverdispersedcountdatarelativetothePoisson(observedvariance/the-oreticalPoissondatavarianceis228.7,p.0001)andthuswellsuitedtoanegativebinomialgeneralizedlinearmodel(GLM)[19].Model1inTable2providesthecoecientsofanegativebinomialGLMofimpressionsonpayrateanddummyvariablesforthepresenceof\goodads"ornoads,relativetothebaselineof\badads".Relativetoabaseline Model1Model2 (Intercept)3:433:43(0:12)(0:12)Goodads0:17(0:10)Noads0:22(0:10)Payrate26:4726:61(4:80)(4:80)Goodadsornoads0:19(0:08) AIC12158:5712156:85BIC12184:1212177:29LogLikelihood6074:296074:43Deviance1481:001481:04Numberofobservations12231223 ***p0:01,**p0:05,*p0:1Table2:NegativebinomialGLMofimpressionsonadconditionandpayrate.Badadsleadtofewerimpressionsthangoodadsornoads.Coecientsareexpressedinlogimpressions;predictedvaluesaredisplayedinFigure5.Payrateisindollarsperveimpressions(.01,.02,.03).Standarderrorsareinparentheses.of\badads",boththe\goodads"conditionandthenoadsconditionledtosubstantiallymoreimpressions(19%and25%moreimpressions,respectively).Model2isthesameasModel1butreplacesthetwoaddummieswithonenewdummyrepresentingthe\goodads"andnoadsconditionscombinedandresultsinasimilarconclusion.Astheco-ecientsinTable2areexpressedinlogterms,theeectsoftheconditionsonrawimpressionsismosteasilyseeninFigure5,whichalsomakesclearthatthedierenceinim-pressionsbetweenthe\goodads"and\noads"conditionsisnotsignicant.ThemodelexpressedinTable2andFigure5canbeusedtoestimatethecompensatingdierentialofannoyingadsinthisexperiment.Sincethecurvesareslightlynon-linear,arangeofcompensatingdierentialscouldbecalculatedacrossthepayrateandadconditions.Togetasimple,sin-gleapproximationweusethemiddle,\goodads"conditiontoestimatetheeectofpayraises.Wetaketheaverageofthe.2to.4and.4to.6centdierences,givinganestimatedincreaseof16.58impressionsresultingfroma.2centperimpressionpayraise.Whensummarizingtheeectofadquality,weusethenumberofimpressionsatthe.4centpayrate.Movingfrom\badads"tonoads,impressionsincreaseby12.68.Thepayraiserequiredtoachievea12.68impres-sionincreaseis.153centsperimpression(=:212:68=16:58)or$1.53CPM(costperthousandimpressions).Thatis,inthisexperiment,aparticipantinthe\badads"conditionwouldneedtobepaidanadditional$1.53perthousandim-pressionstogenerateasmanyimpressionsasapersonintheconditionwithoutads.Similarly,movingfromthe\badads"conditiontothe\goodads"conditionresultedinanadditional9.52impressionsperperson.Itwouldrequireapayraiseof.115centsperimpression(=:29:52=16:58)togenerate9.52additionalimpressions,meaningthatpeoplein Figure7:PhasediagramrelatingmarketsharetouserutilityasdescribedbyTheorem1Inaddition,ifxx,thenuserutilityu-383;uandisde-creasingovertime.Ifx-383;x,uuandisincreasingovertime.TheproofofthistheoremisgiventheinAppendix.ThesolutionisillustratedinthephasediagramgiveninFigure7.Theequilibriumforanystartingmarketsharexinvolvesthepathpointingtoward(x;u).Thevalueofuadjuststoputthepublisheronthispath.Startingwithalowmarketshare,thepublishersetsahighuserutilitywhichisacombinationoflowadvertisingandhighcontentquality,andthengradu-allydegradesuserutilityandincreasesadvertisingintensity.Incontrast,apublisherwhostartswithahighmarketsharewillsetaverylowcontentqualityandhighadvertisingin-tensity,andthengraduallyimprovetheuserexperience.Anincreaseintheinterestratedecreasesx,theasymptoticmarketshare.Anincreaseinthecompetitiveleveluin-creasesxwhenislog-convexandvice-versa.Thereareseveralconclusionsonecandrawfromthismodel.First,sincetheterminalmarketsharepredictedinTheo-rem1dependson,whichdependsonAandu,themodeljustiestheratiooftherevenuetousercostasthekeymetricforadvertisingselection.Second,inacompetitiveadvertis-ingmarket,alladswillsellforaconstanttimestheusercost.Annoyingadswillrunonlywhentheirrevenueisveryhighorthepublisherisextremelywillingtosacriceuserexpe-rienceforrevenue.Third,alegacypublisher,whosemarketshareislargebecausetheyinitiallyfacedlittlecompetition,willstartwithaloweruserexperienceinvolvingbothmoreadsandworsecontentthananentrant.Thiswillresultinthelegacypublisherseeingafastdeclineinuserbase.Thelegacypublisherscontentwillgraduallyimproveuntilasta-blepointisreached.Finally,ifconsumersreactsucientlyslowlytochangesincontent(thatis,issmall),alegacypublisherwillgraduallygoextinctratherthanoerabetteruserexperience.6.CONCLUSIONTherststudyreportedhereshowedthatpeoplendani-matedadvertisementsmoreannoyingthanstaticones,hold-ingallelseconstant.Thisstudyalsoidentiedvecategoriesofcomplaintsaboutannoyingadsprovidingarstpassatidentifyingundesirablefeatures.Weusedthegoodandbadadsfromthisstudytomeasurethecompensatingwagedif-ferentialinthesecondstudy.Themainresultofthispa-peristhatannoyingadsleadtositeabandonmentandthusfewerimpressionsthangoodadsornoads.Inwhatmightbeseenasgoodnewsforpublishers,goodadsandnoadsledtoroughlyequalnumbersofimpressions.Annoyingadsimpairedpeople'sabilitytocarryoutanemailclassicationtask,suggestingthatannoyingadshavearealcosttousersbeyondmereannoyance.Finally,weprovidedatheoreticalmodelthatcomputesadynamicequilibrium,whichpermitsstudyingnotonlypropertiesofthesteadystate,butthead-justmenttothatstateaswell.Thismodelcanbeusedtounderstandthebehavioroflegacypublishers,whoinheritedalargemarketshare,inthefaceofcompetitionfromnewentrants.Wecalculatedthecompensatingwagedierentialinourexperimentofbadadstonoadstobe$1.53CPM,badadstogoodadstobe$1.15,andgoodadstonoadstobe$.38CPM.Somecaremustbetakenininterpretingthesenumbers.Whilewepickedatask|classifyingemails|thatshouldbefamiliarandcommonformostinternetusers,thistaskmaynotberepresentativeofotherinternettaskslikereadingnewsstoriesorsearchingforproductstopurchase.Abandonmentratesmaydierwithdierenttasksandtheeectsofadvertisingmayvaryaswell.Whilevirtuallyev-erywebservicefeaturescompetition,theswitchingcostsvaryfromverylowinconsumingnewstorelativelyhighinchangingemailservices.BecauseourusersonMechanicalTurkhaveanoutsideoptionofworkingonanalternativeHIT,weexpectourresultstobemostapplicabletosit-uationsinvolvinglowerswitchingcosts.Nevertheless,weexpectthatourndingthatannoyingadscosttheuseratleast$1CPMovermorepleasantadswillbeobtainedinsomeotherenvironments.Forthesereasons,wesuggestfurtherstudiesbedoneonMechanicalTurk,aseldexperiments,andinlaboratoriestomeasurethisdierentialonsimilaranddierenttasks.Ifstudiesacrossvariousdomainswithavarietyoftasksandoutsideoptionsarriveatsimilardierentials,morecredencecanbeplacedonthesenumbers.Weviewthisworkasarststepinthisdirection.Iffutureworkarrivesatsimilarestimatesacrossavarietyofpublishers,suchestimatescouldserveasausefullowerboundforwhatapublishershouldchargetoruntheseads.Moreover,itwillbevaluabletousethecompensatingdierentialsapproachtopricethevariousbadaspectsofads,includinganimationandpooraesthetics.Thisworkalsosuggestsavarietyofpolicyrecommenda-tions.Mostdirectly,the$1CPMusercostofbadadshaspracticalconsequencesforpublishers,especiallyasbadadsoftencommandlowerCPMs.Itisareasonthatpublishersshouldinsistonasubstantialpremiumforannoyingadver-tisements.Moreover,apublishercouldrandomizewhichusersseewhichadsandtrackbothtimespentonthepageandthefrequencywithwhichusersreturntothesite.Thistypeofexperimentationwouldcapturelongertermeectsofannoyingadsthanthosestudiedhere.Also,publisherscouldgiveusersanoptiontocloseorreplaceanad.Areplace-menteventwouldallowthepublishertoinferthatauserwouldpreferarandomadovertheadcurrentlyshown.Ad-vertiserswithahighclosurerateshouldbechargedmore.Furthermore,itwouldbereasonabletoassumethatmore annoyingadswouldbeclosedorreplacedfasterthanlessannoyingads.Adreplacementwouldhelptheuserbyre-movingtheannoyingadandthepublisherbymakingitpos-sibletochargefortwoimpressions.7.ACKNOWLEDGMENTSWethankRandallA.Lewis,JustinM.Rao,andDavidH.Reileyforhelpfulconversations.APPENDIXInthissectionwegivetheproofofTheorem1.Proof.Deney=logx 1x.Note,y0=x0 x+x0 1x=x0 x(1x),andx=ey 1+ey.Furthermore,1+ey=1+x 1x=1 1x.Thuswecanreformulatethepublisher'soptimizationproblemasthatofmaximizingR10ertey 1+eyu+1 y0dt.LetF(y;y0;t)=ertey 1+eyu+1 y0.TheEulerequationforthisproblemis0=@F @yd dt@F @y0=ertey (1+ey)2u+1 y01 d dtertey 1+ey0u+1 y0=x ert[(1x)(u)+r0(u)0(u)(uu)(1x)00(u)u0]Thus,00(u)u0=(1x)(u)+r0(u)(uu)(1x)0(u):Asteadystateofthesystemholdswhenx0=u0=0,oru=uand0=(1x)(u)+r0(u).Thisisequivalentto1x=r 0(u) (u):Ifr 0(u) (u)1,alloptimalpathsinvolvex!0asthereisnointernalsteadystate.Whenr 0(u) (u)1,thereisaninteriorsteadystate.Theu0=0curveoccurswhen0=(1x)(u)+r0(u)(uu)(1x)0(u):Thus,near(x;u),du dxu0=0=(u)(uu)0(u) (1x)0(u)+r00(u)(1x)0(u)(uu)(1x)000(u)=(u)(uu)0(u) +r00(u)(uu)(1x)000(u)(u) r00(u)0:WecanobtaininsightaboutthepathsnearthissolutionbyarstorderTaylorapproximation.Thestrategylookslikethis.Writex0u0= (uu)x(1x)(1x)(u) 00(u)+r0(u) 00(u)(uu)(1x)0(u) 00(u)!=g(x;u)h(x;u):x0u0 @g/@x@g/@u@h/@x@h/@u!(x;u)=(x;u)xxuu.Locallythebehaviorofthegeneralsystemisapproximatedbythebehaviorofthelinearsystem.Theonlychallengingterminthematrixis@h @uu=u;x=x=@ @u(1x)(u) 00(u)+r0(u) 00(u)(uu)(1x)0(u) 00(u)u=u;x=x=(1x)(u)000(u) 00(u)2+r10(u)000(u) 00(u)2=rThus,x0u00x(1x)(u) 00(u)rxxuuTheeigenvaluesofthelinearsystemaredeterminedbyso-lutionsto0=detx(1x)(u) 00(u)r0=2r+2x(1x)(u) 00(u)solvingforgives,=1 2 rs r242x(1x)(u) 00(u)!Because00(u)0,thereisonepositiveandonenegativeeigenvalueandbotharereal.Thus,thebehaviorofthesystemisasaddle,asillustratedinFigure7.Thereareinnitelymanypathsconsistentwithequilibriumgivenbythedierentialequations.Whichoneistherightone?Inthecasewhenr 0(u) (u)1,allpathsthatdon'tviolatetransversalityleadtox=0.Supposexisacandidatelimit.Considersettingu=u+fortunitsoftime.Thermearns Zt0ersdsx(u+)+Zt0ersds(x+x(1x)t)(u)y=1 r1ertx(u+)+1 rert(x+x(1x)t)(u)1 t@ @=0=1 t1 r1ertx0(u)+ rertx(1x)t(u)= rx(u)1ert t0(u) (u)+ert(1x)= rx(u) r0(u) (u)+(1x)Thus,itpaystoincreaseaconvergentxifandonlyifx1r 0(u) (u),implyingthatthisisonlycandidateforconvergentpathswhenr 0(u) (u)1.
Goldstein Microsoft Research NYC 102 Madison Ave 12th Floor New York NY 10016 dggmicrosoftcom R Preston McAfee Google Strategic Technologies 1600 Amphitheatre Parkway Mountain View CA 94043 prestonmcafeecc Siddharth Suri ID: 2197 Download Pdf
Goldstein Microsoft Research NYC 102 Madison Ave 12th Floor New York NY 10016 dggmicrosoftcom R Preston McAfee Google Strategic Technologies 1600 Amphitheatre Parkway Mountain View CA 94043 prestonmcafeecc Siddharth Suri Microsoft Researc
GOLDSTEIN Microsoft Research and R PRESTON MCAFEE Microsoft Corporation and SIDDHARTH SURI Microsoft Research Display advertisements vary in the extent to which they annoy users While publishers know the payment they receive to run annoying ads
GOLDSTEIN Microsoft Research and R PRESTON MCAFEE Microsoft Corporation and SIDDHARTH SURI Microsoft Research Display advertisements vary in the extent to which they annoy users While publishers know the payment they receive to run annoying ads
(applause) Now it is also a moment for me to thank my staff and my support staff and all the extraordinary army of interns I have, and first and foremost I would like to thank these
http://www.popguard.org/ BROWSE FASTER AND SAFER WITH POPGUARD By blocking all ads, the Pop Guard protects your data from online trackers and loads webpages 60% faster. That's more time for you to browse online. What is PopGuard? Blocks unwanted popups and popunders on sites you visit. You will see a notification when any popup is blocked. You can add sites to whitelist to ignore this. Filter lists enabled by default include: An ad-blocking list selected based on your language (Easy List) Filter lists are enabled to get you started. You can remove them or add others - it's up to you.
Ivanecká. Simona. Fill. . the. . gaps. . with. . the. . adjectives. in . brackets. ... 1. . . He's. . such. a . monotonous. speaker. I . was. ________________ . stiff. . (. bored. / . boring.
eunuchs…. Daniel 1.14-16. … So . he listened to them in this matter, and tested them for ten . days. At . the end of ten days it was seen that they were better in appearance and fatter in flesh than all the youths who ate the king's food. So the steward took away their food and the wine they were to drink, and gave them vegetables..
Regulating Co-owner Behavior. “Immoral” or “Improper”. Immoral - violating moral principles; not conforming to the patterns of conduct usually accepted or established as consistent with principles of personal and social ethics. (.
Regulating Co-owner Behavior. “Immoral” or “Improper”. Immoral - violating moral principles; not conforming to the patterns of conduct usually accepted or established as consistent with principles of personal and social ethics. (.
Message. Times of the Gentiles are revealed prophetically (2, 7, 8-12) and ethically (1, 3-6). CHAPTER 2. CHAPTER 3. CHAPTER 4. CHAPTER 5. CHAPTER 6. CHAPTER REVIEW. Synthetic Outline. Historical (1-7):.
© 2021 docslides.com Inc.
All rights reserved.