an electoral system in crisis - wordpress.com › 2016 › 07 › ... · 7/18/2016 · an...
TRANSCRIPT
AnElectoralSysteminCrisis
luluFries’dat&AnselmoSampietro
incollaborationwithFritzScheuren
July18,2016
Havingconfidenceinourelectionsiscentraltoourfaithinourgovernmentandallofthedecisionsthatwemakecollectivelyasanation.Butarethecandidateswhowintheonesweactuallyvotefor? AlargeandgrowingbodyofresearchprovidesconvincingevidencethatU.S.electronicvotingequipmentinmanyareasthroughoutthecountryisnotcountingthevotesaccurately.Thiscouldbeduetomalfunctionsincomputerequipmentthatin43statesisoveradecadeold,andlongpastitsnatural
life.However,inmanycases,theevidencestronglysuggeststhatfraudisthelikelyexplanation.Theseproblemshavebeenoccurringsinceatleast2004,andarecertainlypresentinthecurrent2016presidentialprimaries. Thedocumentationconsistsofstatisticalgraphsanalyzingdatafromfivepresidentialcycles,aswellasoff-yearracesfromacrossthecountry.Thedataillustratesthatthereareunusuallylargediscrepanciesbetweensmallprecinctandlargeprecinctelectionreturns,andnoticeabledifferencesbetweenhand-
countedandmachine-countedprecinctresults.Eveninisolation,thedatagivescauseforconcern.Thestatisticalevidenceisreinforcedbyphysicalevidenceandcongressionalhearings:manualrecountsthatdonotmatchthetotalsofthemachinesbeingaudited;andtestimonyunderoathaboutdirect
knowledgeoftamperingwithelectronicvotingequipment.
Weexaminedtheelectionresultsofthe2016presidentialprimaries,andfoundirregularitiesintheoverwhelmingmajorityofthetwenty-onestatesthatweanalyzed.Thedataindicates,inparticular,thatthetotalsreportedintheDemocraticracebetweenHillaryClintonandBernieSandersmaynotbe
correct.Instateafterstate,independentexaminationbytwoseparateanalystsfoundsuspectstatisticalpatternsgivingClintoninflatedpercentages,thatinalllikelihoodarenotfullybasedonactualvotes;andleavingSanderswithwhatappeartobeartificiallydepressedtotals.
Thedifferencebetweenthereportedtotals,andourbestestimateoftheactualvotetotals,varies
considerablyfromstatetostate.However,thesedifferencesaresignificant—sometimesmorethan10%—andcouldchangetheoutcomeofthe2016Democraticpresidentialprimary.Wefoundirregularitiesinthe2016Republicanpresidentialprimaryaswell,andwhileconcerning,wedonot
believetheyarelargeenoughtochangetheoutcomeofthatrace.
FritzScheuren,amemberofthestatisticsfacultyatGeorgeWashingtonUniversity,andaformerpresidentoftheAmericanStatisticalAssociation,hasbeenacollaboratorinthisresearch.Examiningthedatafromthestudy,Scheurensaid,“Asastatistician,Ifindtheresultsofthe2016primaryvoting
AnElectoralSysteminCrisis
unusual.Infact,Ifoundthepatternsunexpected[andpossiblyeven]suspicious.Thereisagreaterdegreeofsmoothnessintheoutcomesthantheroughnessthatistypicalinraw/realdata.”
Itisimportanttonotethatthefactthatacandidatebenefitsfromirregularitiesdoesnotimplythata
candidateisresponsibleforthem.
InJanuary2014,ThePresidentialCommissiononElectionAdministrationpublishedareportstating,“PerhapsthemostdirewarningtheCommissionheardinitsinvestigation…concernedtheimpendingcrisisinvotingtechnology.Well-knowntoelectionadministrators,ifnotthepublicatlarge,this
impendingcrisisarisesfromthewidespreadwearingoutofvotingmachinespurchasedadecadeago(p.62.)”Thisreportwasissuedovertwoyearsago,butunfortunatelyverylittlehasbeendonesincethentorectifytheproblem.Sotheissueswearereportinghere,ofsecurityproblemsonoldandfailing
machines,arenotsurprising.Howeverwedidfindsecurityissueswithevennewerelectronicvotingequipment,suchasthemachinesinNewYorkState.
Atacongressionalbriefingonvotersuppression,heldonApril21,2016,Rep.HankJohnson(D-Georgia)expressedgraveconcernaboutthesecurityofthevotingequipment:“Thereisaveryinsidious,
treacherousanddeceitfulmethodofvotersuppression,andithastodowiththeintegrityofthevotingprocessitself…onepossibility,andIthinkit'saverygoodone,isthatsomeone'smanipulatingthecountingofthevotes.Someoneishackingintothesecomputersthattabulatethevotes."
AnEnvironmentofCorruption
Theportraitofanelectoralsystemincrisisisfurthersupportedbyreportsfromelectionintegrity
organizations,mediaoutlets,andindividualsonsocialmediathatvotingisincreasinglytakingplaceinacorruptenvironment.Thiscontextualevidenceofvoterspurgedfromtherolls,registrationslostinthe
mail,partyregistrationsbeingchangedwithoutavoters’knowledgeorintent,votersbeingsentincorrectballots,ashortageofballots,pollingplacesbeingclosed,discouraginglylonglinesintargetedprecinctsandstates,anddisturbinglylargedisparitiesbetweeninitialexitpollsandofficialresults,lends
credencetotheargumentthatifoneformoffraudisalreadyinplay,anotherformoffraudismoreplausible.ThisinformationisbeingaggregatedbyelectionintegritygroupssuchasElectionJusticeUSA,throughvotertestimonialsandlawsuitsthatareinprogressaroundthecountry.
Figures1and1Aareexamplesofdisenfranchisedvotersfromthe2016presidentialprimaries.Stories
likethesehavebeenubiquitousinmanystates,includingArizona,NewYorkandCalifornia.MoreoftheseinstancesaredocumentedinthisarticleonHeavy.com.
AnElectoralSysteminCrisis
Fig.1—Facebookpost:BeckyDillon,aCaliforniavoterforcedtovoteviaprovisionalballot,June7,2016
Fig.1A—CourtorderrequestofChloePecorino,afirst-timeNewYorkvoterwhoseregistrationwaslostinthemail.ShewasunabletocastaregularballotintheDemocraticpresidentialprimarydespite
requestingacourtorderonApril19,2016
DocumentscourtesyofElectionJusticeUSA
AnElectoralSysteminCrisis
ThisispartofthetextoftheaffidavitthatMs.PecorinofiledwithElectionJusticeUSA:
“I,ChloePecorino,remainunregisteredinthestateofNewYorkasaDemocrat.IhavemademultipleattemptstoconfirmmyvoterregistrationwiththeDMVandtheBrooklynBoardof
Elections…IregisteredthroughachangeofaddressformwiththeDMVinearlyMarch.IregisteredasaDemocrat.IknowtheDMVreceivedtheformbecausemynewaddressisonfileasofMarch18th,2016.IneverreceivedconfirmationaboutmyregistrationwhichiswhyI've
beencallingtheBrooklynBoardofElectionsforthepastthreeweeks.”
SomeofthesetamperingissuesmayberelatedtoaDecember2015massivedataleak,reportedbyresearcherChrisVickery,thatincluded“personal,public,andsomenon-publicinformationon191millionregisteredvoters.”AccordingtoaForbesarticlebyThomasFox-Brewster,Vickeryfound“300GB
ofvoterdata,whichincludesnames,homeaddresses,phonenumbers,datesofbirth,partyaffiliations,andlogsofwhetherornottheyhadvotedinprimaryorgeneralelections.Thedataappearstodatebackto2000.”Forbesstatedthattheinformationwasopenlyavailableonline,andthat“Itwouldappear
everyregisteredUSvoterisincludedintheleak.”
HowWouldWeKnowiftheVotingMachinesWereNotCountingtheVotesCorrectly?
Thebestwaytocheckwouldbetocounttheballotsbyhand,orexamineanypaperorelectronictrailavailableinathoroughandpublicaudit.Thisisnothappening.AccordingtoadatabasecompiledbyCitizensforElectionIntegrity,only12statesrequireapost-electionauditof“everycontestandballot
issuevotedontheballot.”Eventhatestimateisgenerous.Forexample,NewYorkislistedasoneofthose12states,butin2015,itspost-electionauditlawwaschangedfromrequiringathree-percenthandcountaudittosimplyrunningthoseballotsthroughthemachineagain.AlanGoldston,aNewYork
electionlawconsultant,said“thisisnotarecountatall.”
Sotheshortanswertothisquestionis:Wewouldn’tknowifthetotalswerewrong.Orwouldwe?Wouldtherebeotherindicationsthatthemachine-countisnotaccurate?
Evidence
Ifvoting-machineresultswereinaccurateonaregularbasis,therewouldbesomeevidenceofit.One
indicatorwouldbethatvotescountedbymachineswouldgivedifferentresultsthanvotescountedbyhand.Infact,thisisnowbeingseeninelectionsalloverthecountry.
Inthe2016DemocraticprimaryinKingsCounty,NewYork(Brooklyn,)agroupofaffidavitballotswerehand-countedbyagroupofvolunteers.Comparingthehand-countswiththemachine-counts,thereisa
noticeabledifference(Figure2).Ineverysingleassemblydistrictweexamined,exceptone,HillaryClintonperformedbetterwhenthevoteswerecountedbymachine.Thisisasmallsampleoftheoverallballotscast,buttheconsistencyoftheresultsmakesaconvincingcasethatsomethingisamiss.
AnElectoralSysteminCrisis
Fig.2—Hand-countedballotsshow aconsistentlyhigherreturnforSandersinthe2016NewYorkpresidentialprimary
GraphbyAnselmoSampietro
Comparisonshavebeenmadepreviouslybetweenvotingresultsinhand-countedprecinctsandmachine-countedprecincts.Whentherehavebeendiscrepancies,theyhavebeenpassedoffasthe
resultofdemographics.Thisisareasonableconcern,sinceitispossiblethatvotersofaparticularpoliticalperspectivecouldtendtochooseaparticulartypeofvotingequipment.However,inthisinstance,becausethetwosamplesets(hand-countedandmachine-counted)arefromidentical
precincts,withvotersparticipatinginthesameelectiononthesameday—thereisnodemographicvariabletotakeintoaccount.
Theseaffidavitballotsarefromvoterswhowerenotabletovotebyregularballot.IntheNewYork2016primary,over120,000voterswerepurgedfromtherollsinBrooklynalone,andalargenumberofvoters
alsohadtheirvoterregistrationchangedwithouttheirknowledgeorintent.
Sandersvoterstendtobeyoungerandmoreindependent,soonemightthinkthattheywouldbelesslikelytoregisteraheadoftime,andmorelikelytoshowupintheaffidavitsample.Howeveroftheover120,000affidavitballotscast,onlyabout30,000wereactuallycertifiedandcounted.Itisthatfinal
“approved”subsetbeingcountedinourstudy.ThosevoteswouldhaveonlyincludedofficiallyregisteredDemocrats,notindependentsorlateregistrants.Thoseofficiallyapprovedaffidavitvotes,whencountedbyhand,areshowingaconsistentlyhigherpercentageforSandersthanwhenthevotes
arecountedbymachine.Therearetwopossibleexplanationsforthis.Oneisthatthemachinesarecountingthevotesdifferently.Theotheristhatthevoterswhowereforcedtouseaffidavitballotswere
targetedSandersvoters.Possibly,bothofthesefactorsareatwork.Eitherway,thedataindicatesthefootprintofmanipulationintheelection,andcallsintoquestionthevalidityofthereportedresults.
AnElectoralSysteminCrisis
Figure2AshowstheresultsofarecountinHillsboroughCountyinthe2008NewHampshireDemocratic
primary.Thereweredifferencesinalmosteveryprecinctbetweentheoriginalmachinecountandthemanualrecount.
Fig.2A—2008NHDem.presidentialprimary Themanualre-countshowslargediscrepancieswiththeoriginaltotals
Source:TheBradblog
Inthe2016WisconsinandMassachusettspresidentialprimaries,therehavealsobeenstarkdifferences
betweenthecandidates’percentagesinhandcountandmachinecountprecincts.
AnElectoralSysteminCrisis
Hacking
Theexamplesprovidedabove,showingdifferencesbetweenhandcountsandmachinecounts,cannotbeexplainedbydemographics.Ifthediscrepanciesarenotduetodemographics,thereiseithersome
issuewiththevotingmachinesorthehandcounts.Informationsurroundingthosetwoprotocolssuggeststhattheproblemwouldbewiththemachinesandnotthehandcounts.
Inhiswell-researchedpostontheoddresultsofthe2016MassachusettsDemocraticprimary,TheodoredeMacedoSoarespointsoutthatCanada,Australia,Denmark,France,Ireland,Italy,Sweden,andSpain
areamongthe59countriesthatrelyonhand-countedpaperballotstodeterminetheirresults.Incontrast,concernsaboutsecurityandaccuracyhaveplaguedelectronicvotingmachineswherevertheyhavebeenimplemented.
In2009,Germany’shighestcourtbannedtheuseofcomputersinthevotingprocessamidstconcerns
thattheprocesswasnottransparent.JonathanSimon,aHarvard-educatedattorneywhoistheco-directoroftheElectionDefenseAlliance,saysonhiswebsite,“There’svirtualunanimityamongtheexpertswhohavestudiedelectronicvotingmachinesthatinsidersorhackerscanchangetheresultsof
electionswithoutleavingatrace.”HecitesstudiesfromJohnsHopkins,Princeton,UniversityofMichigan,TheBrennanCenterForSocialJusticeatNYU,thestatesofCaliforniaandOhio,andeventheU.S.GovernmentAccountabilityOfficetobackuphisclaim.
J.AlexHaldermanteachescomputerandnetworksecurityattheUniversityofMichiganandhas
successfullycompromisednumerousvotingsystems.Hepaintsavividandunnervingdescriptionofonehack,“Within36hoursofthesystemgoinglive,ourteamhad…almosttotalcontroloftheserver
software,includingtheabilitytochangevotesandrevealvoters’secretballots.”Haldermanpointsoutthatthethreattoourelectionscouldbecomingfrompoliticalplayersinsideourcountry—orevenfromabroad.TestifyingbeforetheD.C.BoardofEthicsandElectionsaboutoneofhismanyvoting-machine
hacks,Haldermanmakesitclearthattherisksarenottheoretical.“WhilewewereincontrolofthesesystemsweobservedotherattackattemptsoriginatingfromcomputersinIranandChina.Theseattackerswereattemptingtoguessthesamemasterpasswordthatwedid.Andsinceitwasonlyfour
letterslong,theywouldlikelyhavesoonsucceeded.”
YoucanviewDr.Haldermanperformingasuccessfulhackonavotingmachineinthisclipfromthedocumentary“HollerBack—[not]VotinginanAmericanTown.”
WhoisResponsible?
Atthispoint,weareunabletosaywhomightberesponsibleforanydatabreachestothevotingequipment.Therecouldbeanynumberofindependentplayerswhowouldbenefitfromthevictoryofa
AnElectoralSysteminCrisis
particularcandidateandwouldbewillingtotakeactiontoinfluencetheresults.Ourresearchalsoindicatesthatinsomeelectionsthefootprintofmorethanoneunofficialplayerisevident.
WhatisthatPattern?
Wearenowgoingtoutilizeadifferenttechniqueinoursearchforevidenceofelection-result
irregularities.Wewillfocusprimarilyonstatisticalirregularities,andbythatwemeanresultsthatdefystatisticallaws.Thetechniqueweareusingiscalledthe“CumulativePrecinctVoteTallyChart,”alsoknownasaCVT(cumulativevotetally)graph,orCVS(cumulativevotestudy).TheCVTgraphhasa
numberofadvantagesinexaminingelectionresults.Exitpollsandthediscrepanciesbetweenthemandtheofficialresultshavereceivedalotofattentioninthe2016presidentialcycle.TheCVTgraphusesactualvotes,andnotpost-votesurveys,sotheresultsaremoreconclusivethanexitpollcomparisons.
Secondly,althoughitisbasedonsolidstatisticalprotocols,itdoesnotrequirestatisticaltrainingtounderstand,andisthereforesuitedtohelpingboththestatisticalandthenon-statisticalcommunityfullygraspthelargedistancebetweenthevotetotalscurrentlybeingreportedandthestatisticalnorm.
Third,statisticianBethClarksonexplainsthatelectiondatatendstohavealotof“noise.”ShelikestheCVTanalysisbecause,“Itallowsyoutoseeatrendthatisdifficulttospotinanoisydataset.“
ThetechniqueisbasedontheLawofLargeNumbers(Figure3.)Investopediaprovidesastraight-forwardexplanation,“Aprincipleofprobabilityandstatisticswhichstatesthatasasamplesizegrows,
itsmeanwillgetcloserandclosertotheaverageofthewholepopulation.”
Fig.3—IllustrationoftheLawofLargeNumbersusingrollsofasingledie WikigraphbyNYKevin
AnElectoralSysteminCrisis
Interpretingthislawforelections,thesamplesizeisthenumberofvotes,andthemeanisthecandidate’spercentage.Inpracticewhathappensisthatthelargerasampleofvotesthatyoucollect,
thecloseryoushouldgettothecandidate’saveragepercentageofsupportinthatlocale.Thisiseasytoseeinaction.Ifyouandyourfriendssupportacandidate,itdoesnotmeanthecandidatehasthatlevelofsupportoverall.Butabroadersampleofvotersinyourcommunitywillgenerateamoreaccurate
pictureofthecandidate’sactuallevelofsupport.Thisisthebasicconceptbehindallpolling;andthisistheprinciplethatisthefoundationfortheCVTgraphs.
UsingCVTgraphstodemonstrateirregularitiesinelectionresultshasbeencontroversial.Asaresult,we’regoingtorelaythemethodologyandbackstoryofthetechnique,confirmthatitaccurately
demonstratesastatisticalpatternthatexists;investigatewhetherthereisademographicexplanationforthatpatternandexplorewhat,ifanything,thepatternsignifies.
TheCVTgraphshowstheprecinctsaddedtogethercumulativelyfromthesmallesttothelargestalongtheX-axis.OntheY-axisitshowsthetwocandidates’percentages(Figures4and4A).Intheseinstances
from2000and2004,theCVTgraphresemblesthegraphillustratingtheLawofLargeNumbers.Becausetheprecinctsareaddedtogethercumulativelyasyoumovefurtherrightonthegraph,itbecomesharderandharderforanyindividualprecincttoovercometheaveragepercentageofallthevotesthat
havebeenaddedupsofar,andthedatatendstochartasaflatline,atleastitdiduntil2004.Sometimearound2004,orpossiblyalittleearlier,otherpatternsemergethatwewilldiscussshortly.
Fig.4—2000AlachuaFloridaDemocraticpresidentialprimaryGraphbyAnselmoSampietro
AnElectoralSysteminCrisis
Fig.4A—2004AlachuaFloridaDemocraticpresidentialprimary(Thesecondgraphisazoomed-inviewofthebottomhalfofthefirstgraph.)
fromRonPaulForumby“Liberty1789”
Youmaybesurprisedtoseesomeoftheabovegraphscreditedto“Liberty1789.”Oneofthereasons
forthecontroversysurroundingtheCVTgraphisthatitwasdevelopedontheInternetbynon-professionalsoutsideofacademicstatisticalcirclesbyforumuserspostingunderpseudonyms.Youcouldn’treallyaskforaworsestartforastatisticalmethodtobetakenseriously.
Thegraphwasfirstusedin2012byagroupofRonPaulsupporterswhohadstronganalyticaland
engineeringskills.ThefirstformalpresentationofthetechniquewasmadebytwoofthoseRonPaulsupporters,ChoquetteandJohnson,intwoonlinepapers.ButaccordingtoChoquette,theideaofchartingtheprecinctsfromthesmallesttothelargestwasconceivedbyanengineernamedPhilEvans,
whousedtheonlinehandle“TheMan.”
Evansremembersthenighthefirststartedtonoticeanunusualpatternintheelectionreturns.“In2012IwaswatchingCNNreportontheGOPprimaryresultsinNewHampshireandwhatstruckmewasthat[Ron]Paulreceiveddoublethepercentinsmallprecinctsasinlarge.Iwonderedwhatthatcouldbe.”
Evansdesignsandbuildsindustrialmachinery,andhisworkinvolvescomplexdataanalysis.Hebecamefascinatedwiththequestion:Whywouldonecandidategetsuchalargerpercentageofthevotesinthelargeprecincts?
Afterstudyingthedataintensivelyforsixweeks,Evanscametoaconclusionthatstunnedhim—but
alsomadesense.Hebecameconvincedthatinthelargeprecincts,someofthecandidates’voteswerebeingshiftedtoanothercandidate.Whyonlyinthelargeprecincts?Itwouldbeeasiertodisguisethe
differences,hethought.Inthesmallprecinctswithonlyafewvoters,theshiftwouldbemuchmorenoticeable.Therewereatleasttwowaysitcouldbedone—throughsoftwareinthemachines;orthroughthesoftwareusedwhenthetotalswerecentrallytabulated.Hewantedtoillustratethevote-
switchinghebelievedwasoccurring.Hesays,“SixweekslaterIhadfiguredoutamethodforexpressingthisusingExcelandreleasedapaperthatisstillonlinetoday.”
AnElectoralSysteminCrisis
Evanssayshisinitialgraphsfromthatpaperweremodifiedbyanotherforumuser,“Liberty1789,”intothe“CumulativePrecinctVoteTallyChart.”EvansandhisfellowRonPaulsupportersbeganusingitto
graphmanyoftheelectionresultsofthe2012Republicanprimary.Hereisthepatterntheysawinstateafterstate:acandidatereceivesahigherpercentageofvotesinlargeprecinctsthanhe/shereceivesinsmallprecincts.Thisincreaseoccursinamathematicallyproportionatepattern,inotherwords,asthe
precinctsgetlargerthecandidate’ssupportgetslargerattheexpenseofothercandidates.Oftenthisincreaseisenoughtochangetheoutcomeorthedynamicoftheelection.Whichcandidatereceivesincreasedsupportinthelargerprecinctsdependsontheparticularrace.In2012,thecandidatethat
benefitedfromthepatterninalmosteveryracewasMittRomney(seeFigure5).
Fig.5—2012IowaRepublicanpresidentialcaucus source:“RepublicanPrimaryElection2012Results:AmazingStatisticalAnomalies”
graphbyChoquetteandJohnson
AfterRonPaullosttheelection,EvanssuspectedthatPaulhadbeencheated.Evanssays,“Itwasfrustratingbecausehewasgivingspeechesinlargevenueswiththousandslinedupoutsidebeyond
capacity,whiletheothercandidate[s]weresomewhatlonely.”Itdidn’tmakesensetoEvans,buthehadnowaytoprovethatPaul’svoteshadbeenstolen.
Inthefallof2012,ChoquetteandJohnsonwroteupthefindingsoftheforumandcirculatedtheir
paperswidelyontheInternetandviaemail.Inthosepaperstheyoccasionallyusetheword“alleged;”butforthemostparttheymadeboldclaimslike,“WhencandidateMittRomneyisontheballothealwaysgainsvotesthroughVoteFlipping”(v1.6p.4)and,“Thisdocumentexposeswhatmayverywell
bethegreatestcaseofelectionfraudevertooccurinUShistory(v1.6p.20.)”Theysparkedconsiderablediscussionwithintheelectionreformcommunity.However,theirstudywasreceivedinthestatisticalcommunitywith(understandable)skepticism.
AnElectoralSysteminCrisis
ThemostobviousflawinChoquetteandJohnson’spaperistheirclaimthatDemocraticpartyelections“don’tshowthisproblem.”ItturnsouttherearemanyDemocraticpartyelectionsthatexhibitthis
patterntoo.However,despitethisweakness,theirstatisticalgraphshavebeenconfirmedtobeaccurateinthreeseparatestudies(Clarkson,PlattsmouthNebraskaHighSchoolScientificLogicClass,andLindeman).Eachofthesefoundmoreelectionswherethepatternappears.
WeaskedKellieOttoboni,agraduatestudentatUCBerkeley,toconfirmtheaccuracyofthegraphsinall
threeofthesestudiesandshereplicatedandconfirmedtheaccuracyofonegrapheachofClarkson;theNebraskaHighSchoolScientificLogicClass;andtwoofMarkLindeman’sgraphs.
BethClarkson,whoconductedoneofthestudies,isaqualitycontrolengineerwithadoctorateinstatistics.ShereadChoquetteandJohnson’spaperandtestedtheirtechniqueherselfonanumberof
elections.IntheelectionsClarksonexaminedinKansas,Ohio,andWisconsin,shefoundthesameunusualincreaseforonecandidateinthelargeprecincts.Asastatisticianshefoundtheresults“terriblysurprising.”
ClarksonpublishedanarticleinSignificance,affirmingboththestudies’analysis,andconclusions,
saying,“ThedataI’veanalyzedsupportstheirhypothesisthatwehaveaserious,pervasive,andsystematicproblemwithelectronicvotingmachines.”SheiscurrentlysuingKansaselectionofficialsforpermissiontoauditthepapertrailofoneoftheelectionssheanalyzedinordertocomparethe
machine’spaperrecordswiththerecordedresults.Sofarheraudithasnotbeenpermitted,andsheremainsconcerned:“Iffraudwereoccurring,thesearethekindofpatternswewouldexpecttosee.”
Clarksondiscoveredthatstatisticalpatternsandacandidate’spercentageofthevoteshare,varybetweendifferentmodelsofelectronicvotingequipment(Figure6).Shealsofoundthattherewere
statisticalirregularitiesfavoringmorethanonecandidate,leadinghertosurmise,“themanipulationisnotlimitedtoasinglepowerfuloperator.Myassessmentisthatthedatarevealsmultiple(asleasttwo)
agentsworkingindependentlytosuccessfullyaltervotingresults.”
AnElectoralSysteminCrisis
Fig.6—2012Ohiopresidentialgeneralelection(MittRomneyvs.BarackObama)Statisticalpatternsandcandidates’percentagesvarywithdifferentmodelsofvotingequipment
GraphbyBethClarkson
InFigure6,theES&SDS200vote-scannershowsanirregularstatisticalpatternthatfavorsRomney(reddots.)ButtheES&SDS100andtheHartEscanshowanirregularpatternthatfavorsObama(bluelines.)
ClarksonisgraphingthepercentagesastheyimpacttheRepublicanvote,sowhenalineisgoingup,itisfavoringtheRepublican(Romney)andwhenalineiscomingdown,itisfavoringtheDemocrat(Obama).KellieOttoboniofUCBerkeleyreplicatedClarkson’sresearchandconfirmeditsaccuracy(Figure6A).
Fig.6A–2012Ohiopresidentialgeneralelection(MittRomneyvs.BarackObama)
Clarkson’sresearchwasreplicatedandconfirmedaccuratebyKellieOttoboniofUCBerkeley GraphbyKellieOttoboni
AnElectoralSysteminCrisis
TherearesomeDREsinthiselectiondemonstratinganormal(relativelyflat)statisticalpattern.Butitwouldbeill-advisedtoconcludethatthosemachinesaresecure.Basedonconversationswithsecurity
expertslikeMattBishopatUCDavisandHaldermanattheUniversityofMichigan,itisalmostalwayspossibletobreachthesecurityofthesemachines.AfterhackingtheWashington,D.C.Internetvotingpilotprogram,Haldermansaid,“Ifthisparticularproblemhadnotexisted,I’mconfidentthatwewould
havefoundanotherwaytoattackthesystem.”ColumbiaUniversitypoliticalscientistMarkLindemananddatascientistLeviBowleshavebothpublishedworkconfirmingtheexistenceofthepattern,butarguingthatitisnotindicativeoffraud.We
foundtheirresearchflawedandtheirlogicunconvincing,andhaveprovidedadetailedbreakdownoftheseissueslaterinthispaper.
OnefactthatisclearfromallthesestudiesisthatinmanyU.S.elections,certaincandidatesarereceivinganincreasedshareofthevoteastheprecinctsgetlarger.Thecrucialquestionis,Why?Isthere
aninnocuousdemographicexplanationfortheincrease?Orisitsomethingthatisindicativeoferrororfraud?
WhenDidthePatternBegin?Weareunabletopindownexactlywhenthepatternoriginated.Inafascinatingtripdownelectionfraudmemorylane,writerVictoriaCollierdescribesnumeroustroubledU.S.elections.Itwouldbeinstructive
todoastatisticalanalysisononeoftheracesthatshecitesasan“up-set”likeChuckHagel’s1996NebraskaSenatevictory.”Threedaysbeforetheelection…apollconductedbytheOmahaWorld-Heraldshowedadeadheat[but]HageltrouncedNelsonbyfifteenpoints,”Colliersays.“Thisdivergence
frompre-electionpollingwasenoughtoraiseeyebrowsacrossthenation.”
Fornow,wecanstatethatracesthatweexaminedfrom2004andearlierdidnotshowthepatternofincreasedcandidates’percentagesinlargeprecincts.LookingagainatFigure4,weseethatinthe2000and2004racesinAlachua,Florida,eachcandidate’sshareofthevotesisroughlythesameinsmalland
largeprecincts.However,by2008,thisisnotthecaseinmanyracesaroundthecountry(seeFigure7).
AnElectoralSysteminCrisis
Fig.7—2008NewHampshireandMinnesotaDemocraticpresidentialprimaries
Inracesaroundthecountrycandidates’beginreceivinglargerpercentagesofthevoteinlargeprecincts GraphsbyMarkLindeman
IntheNewHampshireandMinnesotaDemocraticprimaries,aswellasinotherracesin2008,
candidatesreceivealargerpercentageofthevotesastheprecinctsgetlarger.InNewHampshire,thepatternbenefitsClinton.InMinnesota,Obamaistheonewhogainsvotesharesinthelargeprecincts.The2008NewHampshireDemocraticpresidentialprimarywasalsotheracewelookedatinitiallywhere
themanualrecountdidnotmatchtheoriginalmachinetotals(Figure2A).
Figure8showstheWisconsinRepublicanprimaryintwodifferentelectioncycles.In2000,nocandidatehasmuchofanincreaseinthelargeprecincts.Butin2016,TedCruz’spercentagenoticeablyincreasesinthelargeprecincts,whileDonaldTrump’spercentageofthevotegoesdown.
ThesetwocomparisonsalsodemonstratethatthepatternishappeninginbothDemocraticand
Republicanraces.
AnElectoralSysteminCrisis
Fig.8–2000and2016WisconsinRepublicanpresidentialprimariesshowdifferentstatisticalpatterns GraphsbyAnselmoSampietro
Doingfurtherresearchonhistoricalraceswillhelpidentifypossibleearlyappearancesofthepattern.Colliersays,“Throughoutthe1980sand1990s,theuseofopticalscannerstoprocesspaperballotsbecamewidespread.”Butprobablythemostseminalyearforelectronicvotingequipmentwas2002,
whenstatesacrossthecountryexperiencedalargeinfluxofcomputer-basedvotingsystems,withthepassageofthe(perhapsironicallynamed)HelpAmericaVoteAct.
Smooth,Unidirectional,andMathematicallyPredictable.
Figure9showsagraphofthe2016LouisianaDemocraticPrimary.TheanalysisisbyBethClarksonandAnselmoSampietroconfirmeditsaccuracy.
AnElectoralSysteminCrisis
Fig.9—2016LouisianaDemocraticpresidentialprimary
Candidates’votesharesvarybyasmuchas37%betweensmallandlargeprecincts GraphbyBethClarkson
ThisgraphisincompleteviolationoftheLawofLargeNumbers.Foracandidatetoreceivethislevelof
increasedsupportinthelargeprecincts,eachnewprecinctmustbesoheavilyweightedthatitdefiestheaverageofalltheotherprecinctsthathavealreadybeenaddedtogether.Thisisamajorstatisticalirregularity.
Inthesmallprecincts,thedifferencebetweenClintonandSandersisapproximately10%(Clinton48%–
Sanders38%).However,inthelargestprecinctsthedifferencebetweenthecandidatesis47%(Clinton70%–Sanders23%.)Thatisadifferenceof37%supportbetweenthesmallestprecinctsandthelargestprecincts.
Toseehowheavilyweightedthelargeprecinctsare,wegraphedthemseparately,countybycounty,
dividingthelargest25%fromtheremaining75%(Figure10).Withinalmosteverycounty,Clinton
receivesahigherpercentageofthevoteinlargeprecinctsbyunusuallyhighmargins,sometimesby
closeto40%.InWashingtonCountyyoucanliterallyseethemomentthatthedatastartstochange
around600votes.
AnElectoralSysteminCrisis
Fig.10–2016LouisianaDem.presidentialprimary,countyleveldata-thedifferenceinthecandidates’percentagesbetweensmallandlargeprecinctsisunusuallyhigh-insomecasescloseto40%
GraphsbyAnselmoSampietro
Therearethreeothercharacteristicsofthisdatathataresuspect:
1) Thedataissmooth.Thelinesintheoverallstatechartgostraightupandstraightdown;andlinesofdatainthelargeprecinctsarealsoquitestraight.ThisiswhatDr.Scheurenisreferringtointheopeningofthepaperwhenhesays,“Thereisagreaterdegreeofsmoothnessinthe
outcomesthantheroughnessthatistypicalinraw/realdata.” 2) Thedataisunidirectional.Inthestatewideresults,thedataonlymovesinonedirection:
Clintongoesup;Sandersgoesdown.Thepercentagesneverdemonstratethekindofupsand
downscausedbyorganicvotingbehavior.3) Thedatafollowsamathematicallypredictablepattern.Clinton’ssupportisincreasingina
mathematicallypredictableway.Ineachprogressivelylargerprecinctshegetsaslightlylarger
levelofsupport.Thisisapossibleindicationthatamathematicalalgorithmhasbeenappliedtotheresults.
DemographicFactors
Thedatawe’veexaminedsofarshowsthatthepatternemergedinvariouslocationssometimeafter2000.Thisarguesagainstademographicexplanation,sincethedemographicsareconsistent,anditis
thetimeperiodthatischanging.
AnElectoralSysteminCrisis
DiggingdeeperintotheLouisianadatafurtherunderminesthetheorythatdemographicfactorsareresponsiblefortheincreasedvotesharesinlargeprecincts.Here,thesheersizeofthedifference
betweenthesmallandlargeprecinctshasnoplausibledemographicexplanation.
NateCohnattheNewYorkTimeshasinsistedthatthecandidates’percentagesbeingcorrelatedwithprecinctsizeintheSouthisduetoalargeconcentrationofblackvotersintheprecinctsthatsupportClinton.
Fig.11–NateCohn’stweetmakesaweak,sarcasticargument,implyingthatallofthestatisticalirregularitiesintheSouthcanbeexplainedbyClinton’sleadamongblackvoters
InatweetpostedonJune27,2016(Figure11),Cohntakesanunfortunatelyflipattitudetowardtheevidence—whichallowshimtopassonanalyzingthedata,butdoesnothelpanyoneelsesortout
whetherthisisapatternthatweneedtobeconcernedabout.TheunderlyingquestionofCohn’stweetisthis:Isthereademographicexplanationthatcanaccountfora37%shiftinsupportforClintonfrom
thesmallestprecinctstothelargestprecincts?Forinstance,isthiscorrelationhappeningbecauseClinton’ssupportissostrongamongblacksandtherearesomanyadditionalblacksinthelargeprecinctsthatitcanaccountfora37%increase?
JustabrieflookatthedemographicsinLouisiana,contradictsthistheory.The2010censussummaryfile
(searchWashingtonParish,Louisiana–andthenclickon“RaceandHispanicorLatinoorigin,forexample)indicatesthatbothWashingtonandWinnparisheshaveablackpopulationof31%(Figure11A).Soevenif100%ofblackvotersweresupportingClinton,(whichtheyarenot)andeveniftheyall
livedinthelargeprecincts,itcouldnotexplaina37%increaseinhersupportinthelargeprecincts.
AnElectoralSysteminCrisis
Fig.11A–CensusdatashowsthereisnotalargeenoughblackpopulationinWinnorWashingtonParish
toexplainthepercentageofsupportClintonisreceivinginthelargeprecincts
EvenifyoucouldexplaintheincreaseinClinton’ssupportinLouisianabysayingthattherearemoreblacksinthoselargeprecincts,itdoesnotexplainCruz’sincreasedsupportinlargeprecinctsinWisconsin,orRomney’sincreasedsupportinlargeprecinctsinalmosteverystatein2012.Whileone
mightbeinclinedtolooktodemographicstoexplaindiscrepanciesinonestateoranother,thefactthatthesamepatternispresentacrossmultiplestateswithverydifferentdemographicsarguesstronglyagainstitbeingcausedbydemographicfactorsalone.
Despitethelackofharddata,theideathatthesecorrelationscanbeexplaineddemographicallyremains
popular.EitanHersh,amemberofthepoliticalsciencefacultyatYale,toldusthathewouldbetmoneyonademographicexplanation.Weappreciatehisgamblingspirit,let’sseeifhisconfidenceiswell-placed.
First,let’stakeacloselookattheprevailingnarrativethatthemostpredictivedemographicfactorof
whetheravoterwillsupportClintonorSandersisrace.Thisisactuallynotthecase.ArticlesinTheAtlantic,theLosAngelesTimes,andVoxhaveallreportedthatthesinglemostsalientfactorindeterminingwhetheravoterwillsupportClintonorSandersinthe2016primaryisage.Voxquoted
politicalscientistAlanAbramowitzassaying,"Itwasage,andbeyondthatnothingmattered,”afterhe“ranamultivariateanalysistohelpfigureoutthisquestion.”
InthesameVoxarticle,JeffSteinrefinesthenarrativethatClintonisbeatingSandersamongblackvotersbypointingoutthat“severalpollshaveputSandersaheadofClintonamongyoungAfrican-
Americans;intheReuterspollingdata,forinstance,SandersbeatsClintonby25pointsamongblackvotersaged18to29.”AccordingtoLeeMiringoff,directoroftheMaristCollegeInstituteforPublicOpinion,intheLosAngelesTimespiece,"Theagefactorseemstotrumpeverything."
HowmuchadvantagedoesClintonhaveamongoldervoters?Wetooktheaverageofthreepollsto
determineClinton’sapproximatestatisticalleadwitholdervoters(which,basedonthesepolls,we’re
definingas45andolder):
• NBCNews/WallStreetJournalpollfromApril18,2016(quotedintheLosAngelesTimes)givesClintona27%leadwithvotersover50
• IowaentrancepollgivesClintona23%leadwithvoters45–64(TheAtlantic)• IowaentrancepollgivesClintona43%leadwithvotersover65(TheAtlantic)
AnElectoralSysteminCrisis
AveragingthesethreepollsshowClintonwitha31%leadamongoldervoters.SoevenifthelargestprecinctsinLouisianaconsistedofonlyvotersovertheageof45(whichclearlytheydonot),youstill
couldnotexplainashiftof37%,usingthemostimportantdemographicfactorinthecontest.
Isitpossiblethatacombinationoffactorssuchasraceandagecouldexplainthedifferencesinthesepercentages?Intheory,yes.Buttheprecinctsthatwouldhavetoexisttojustifythesepercentageshavenobearingonreality.Theywouldneedtobeunusualneighborhoodswhereolderblackandwhite
voterslivetogetherinmixed-race,largeprecincts,withveryfewyoungpeople.Inactuality,blackneighborhoodstendtoconsistoffamiliesandpeopleofallages;andmixed-raceneighborhoodsarenottypicallymadeupprimarilyofpeopleoverforty-five–notinLouisiana–andnotinotherpartsofthe
countrywhereweseethispatternoccurringoverandoveragain.
InLouisiana,thedemographicargumentseemstoevaporatewhenthedataiscarefullyinvestigated.Whatexplanationissupportedbythedata?Let’slookatthe2016Wisconsinprimarytoanswerthat.
PhilEvans,theengineerwhooriginatedthe“CumulativePrecinctVoteTallyChart,”didananalysisofthe2016WisconsinRepublicanPrimary(Figure12).Thegraphcompareshand-countedcountiestothe
machine-countedcountiesintherestofthestate.Thepatternoflargervotesharesinlargerprecinctsisnotpresentinthehand-countedcounties.Allcandidatesreceiveapproximatelythesamepercentageofthevoteinbothsmallandlargeprecincts.However,inthemachine-countedcounties,Cruzdoesbetter
astheprecinctsgetlarger;Trumpfaresworsethanhedidinthehand-countedcountiesand17%worseinthelargestprecincts.Sampietroconfirmedtheseresultsandalsochartedthemasscatterplots(Figure12A).
Fig.12–2016WisconsinRepublicanpresidentialprimary Hand-countedcountiesandmachine-countedcountiesshowdifferentresults
graphsbyPhilEvans
AnElectoralSysteminCrisis
Fig.12A–2016WisconsinRepublicanpresidentialprimary Hand-countedcountiesvs.machine-countedcounties(scatterplots)
graphsbyAnselmoSampietro
Wehavecreatedscatterplotsofourresearch,butingeneralaredemonstratingthelinechart,becausethepatternismorevisiblethere.Eachdotonthescatterplotrepresentsthepercentageofvotesacandidatereceivesinasingleprecinct:thehigherthedot,thelargerthepercentageofvotesforthe
candidateinthatprecinct.Thefurthertotherightadotis,thelargertheprecinct.Inthehand-countedWisconsinresults,thedotsarelevel:precinctsizedoesn’tinfluencethepercentageofvotesforeachcandidate,andthepercentagesarerelativelyconsistent.Inthemachine-countedresults,Cruz’s
percentageofvotessystematicallyincreasesastheprecinctsgetlarger.ThedescendingshareofvotesforTrumpisalsovisible.
Onceagain,onecandidatedoesinexplicablybetterinthelargeprecincts,butthistimethereisalsoanassociationwiththevotingequipment—becauseunlikeLouisiana,Wisconsincountssomevotesby
hand—allowingforanaddedlayerofcomparison.Wehaveanassociation,butisitacauseandeffectassociation?Arethemachinescausingthisincreaseinacandidate’stotals,oristheresomeotherexplanation?
Thechoiceofvotingequipmentisnotrandom.Forexample,itcouldbethattheprecinctsthatchooseto
counttheirvotesbyhandarealsoprecinctsthatincludemore“independentthinkers”andarethereforemoreattractedtothemaverickcampaignofTrump.
Ifthatisthecase,thenwewouldexpectthesameindependentthinkersinthesameprecinctsontheDemocraticsidetosupportthemaverickcampaignofSanders.Isthatwhatwefind?DotheWisconsin
hand-countedprecinctsgiveSandersagreaterpercentageofsupportthantheWisconsinmachine-
AnElectoralSysteminCrisis
countedprecincts?No,theydonot(Figure13).Sandersreceivesapproximately57%ofthevoteinbothhand-countedandmachine-countedprecinctsinthatrace.
Fig.13–2016WisconsinDemocraticpresidentialprimary-unliketheRepublicanpresidentialprimary
thisraceshowsverylittlecorrelationbetweenprecinctsizeandanycandidates’percentageofthevote graphsbyAnselmoSampietro
Thisseemsstrange.Ifit’sademographicpattern,whyisn’titevidentonbothsides?Themediahas
commentedonthestrikingsimilaritiesbetweentheTrumpandSanderscampaigns.Bothcandidatesarepoliticaloutsiders,deliveringscathingcriticismsofthestatusquoandopposingtradedeals.Bothcampaignshavedonewellwithwhitemalevoters.Itwouldseemlikelythatademographicpatternlarge
enoughtoaffecttheRepublicanraceby17%wouldalsohavesomeimpactonvotersintheDemocraticrace.
KeyArgumentsPart1
Wehavenowestablishedsixkeyargumentsthatthepatternwearewitnessingisirregular;isnotdueto
demographicfactors;andisinsomewayconnectedwiththeuseofelectronicvotingequipment:
1) Datashowsthatelectionsinthesamelocationdevelopedthepatternafter2000,contradictingademographicexplanation.*
2) ThepatternviolatestheLawofLargeNumbers.
3) Thepatternissmooth,unidirectionalandmathematicallypredictable.4) Thelargediscrepanciesinthepercentagesbetweensmallandlargeprecinctsdefyaplausible
demographicexplanation.
AnElectoralSysteminCrisis
5) Thepatternisevidentinawidevarietyofgeographies,withverydifferentdemographics,furthercontradictingademographicexplanation.WehavealreadyshownthepatterninIowa,
Ohio,Wisconsin,Florida,andLouisiana.Thepatternhasbeenconfirmedbytwoanalystsinthe2016presidentialprimaryinConnecticut,Delaware,Florida,Illinois,Kentucky,Louisiana,Massachusetts,NewYork,NorthCarolina,Ohio,SouthCarolina,Tennessee,WestVirginia,and
Wisconsin.Ithasadditionallybeenconfirmedbyatleastoneanalystinthe2016presidentialprimaryinAlabama,California,Georgia,Oklahoma,andTexas.
6) Thepatternappearsinmachine-countedraces,butnothand-countedraces.Publichandcounts
withasecurechainofcustodyareconsideredinternationallythemostsecureformofcountingvotes,sothispointstothevotingmachinesasalikelysourceoftheproblem.
*PhilEvanstoldusthattheonlyracestheyfoundthepatterninpriorto2000,wereinelections
whereanindividualorofficialshadbeenconvictedoffraud.Wehavenotyethadachancetoverifythatinformation.
NewYorkState—MultipleIssuesRaiseSeriousConcerns
Inthe2016NewYorkDemocraticpresidentialprimary,theCVTgraphsofcertaincountiesareirregular
(Figure15).BothRichmondCounty(StatenIsland)andKingsCounty(Brooklyn)revealastrongcorrelationbetweenprecinctsizeandcandidates’percentages.
Fig.15—2016NewYorkDemocraticpresidentialprimary RichmondandKingsCounty,bothinNewYorkCity,showirregularCVTgraphs
graphsbyPhilEvans
Earlierinthepaperwedemonstratedthathand-countedandmachine-countedballotsinKingsCountygivedifferentpercentagesforthecandidates.ThesuspectCVTgraphinKingsCountyreinforcesthe
AnElectoralSysteminCrisis
findingsofirregularelectionresultspresentedearlier,andpaintsanincreasinglytroubledportraitofpotentiallycompromisedvotetotals.Butthereisanotheroddfactorinthisstate’sresults.
PrecisePercentages
WeareshowingtheNewYorkCitygraphs(Figure15)toillustrateaveryspecificpoint.Thedata
supportstheideathattheoverallstatevotetotalsarebeingmassagedtoachieveapredeterminedpercentage.
ThefinalreportedtotalsinNewYorkstatewerealmostexactly58%Clintonto42%Sanders.Itisnecessarytogotothethirddecimaldigittoseeadifference:57.995versus42.005.
DougJohnsonHatlemreportedonthisissue,pointingoutthat,“TheoverallresultsinNewYork,asannouncedonelectionnight,deviatedfromaperfect58–42splitby0.005345.That’s97votesoutofover1.8million.”
InKingsCounty,thereportedtotalswerealmostexactly60%Clinton,to40%Sanders:(59.72%Clinton,40.27%Sanders.)The.27differenceiscausedbyabout800votesoutof300,000.IntheBronx,the
percentageswerealmostprecisely70%to30%(69.59%Clintonto30.41%Sanders),thedifferencebeingjust616outof151,908totalvotes.TheBronxCountyCVTgraphisalsoirregular(Figure16).
Fig.16—2016NewYorkDemocraticpresidentialprimary-theBronxCountyCVTgraphshowsastrong
correlationbetweenprecinctsizeandcandidatepercentage graphbyPhilEvans
AnElectoralSysteminCrisis
Inaninterview,J.AlexHaldermanwasconfidentthatitwouldbepossibletoassignapercentagetoa
particularcandidate,“Ifyou’remanipulatingthecentralcountingsystems,thenyoucanmakeanoveralladjustment.”Askedwhetherit’spossibletogetaccesstothecentralcountingsoftware,hereplied,“Probably.It’sbeenourexperienceinthelastdecadewiththeDieboldsystems,forinstance,thatitwas
certainlypossibleinthemachinesandcountingsystemsweexamined.”
NewYorkCityalreadyhadtwounusualpiecesofdatagivingcauseforconcern:
• Thedifferencebetweenhand-countandmachine-countresults• IrregularCVTgraphsinmultiplecounties
Addtothat:
• Precisepercentagesinthetotalelectionresultscouldbeanindicationthattheresultshavebeenmanipulatedtoachieveaspecificpercentageforthecandidates.
Thesearethreeseparatebutreinforcingfacts,illustratingwhythetotalsforthisstatearesuspect.
AnElectoralSysteminCrisis
ANormalGraph
ColumbiaCounty,NewYork—ahand-countedcounty—offersagoodpointofcomparisontotheabovegraphs(Figure17).
Fig.17—2016NewYorkDemocraticpresidentialprimary,ColumbiaCounty Thishand-countedcountyhasanormalCVTgraph
graphbyAnselmoSampietro
Inthisgraph,thereisalargedegreeoffluctuationbothupanddownontheleftsideofthegraph.Then,byabout1,300votes,thegraphsettlesintoafairlyevenstraightline,withsmall,randomfluctuations.It
maintainsabasicallyflatlinethroughallofthelargestprecincts.Thisisaverynormal-lookingCVTgraph.
ColumbiaCountyDemocraticElectionCommissionerVirginiaMartinsaystheylookateverysingleballotincompetitiveraces.InaninterviewwiththeRegister-Star,shedescribestheirprocess:“Whenavoterscansaballot,itdropsdownintoaballotbagatthebottomoftheopticalscanningmachine.Attheend
oftheday,twoinspectors,oneRepublicanandoneDemocrat,openthemachine,pulloutthebagandzipitshut.”Thentheyproceedtocount100%oftheballotsbyhand.Inaninterviewonpublicradio,shesaidthatColumbiaCounty’svote-countingprocessisfocusedonsecurity,accuracy,andtransparency:
“Wearevery,verycarefulaboutthechainofcustody…Everythinggetshandcounted,andtheresults
AnElectoralSysteminCrisis
thatwecertifyarebasedonthathandcount.”Sheconcludedbypointingoutthatthehandcountisopentothepublic.
Opportunity,Incentive,andRationalization
Fig.18–TheFraudTriangle:opportunity,incentive,andrationalization
Fraudinvestigationsinotherindustrieslookforthreefactors:opportunity,incentive,andrationalization.TheeasewithwhichDr.Haldermanandothersecurityexpertshavebeenableto
penetrateandcontrolthesoftwareofthevotingequipment,makesitapparentthattheopportunityexists.
Theotherfactors,incentiveandrationalization,arealsopresent.“There’salotatstake,”saysJonathanSimonoftheElectionDefenseAlliance.“ControloftheAmericangovernment,controloftheeconomy,
controlofthemilitary.Whenthere’salotatstakethere’sanincentivetofraud.”
Rationalization?It’sunfortunate,butthereareindividualsandgroupsactiveinthepoliticallandscapewhofeeljustifiedtakingactionsthatcrossethicalboundaries.OntheDemocraticside,leakeddocumentsrevealedthattheDemocraticNationalConventionsupportedClintonastheeventual
nominee,althoughtheirpublicpositionwasoneofneutrality.TheGucciferdocument,“2016GOPpresidentialcandidates,”says,“Ourgoalsinthecomingmonthswillbetoframe…theeventualnomineeearlyandtoprovideacontrastbetweentheGOPfieldandHRC[HillaryRodhamClinton.]”
OntheRepublicanside,formerFloridaRepublicanPartychairmanJimGreertestifiedin2012that
Republicanshadmeetingsabout“keepingblacksfromvoting.”
AnElectoralSysteminCrisis
Thesearejusttwoexamples.Therealityisthatpoliticalcampaignshaveoperativeswhoworktodiscreditandtargettheotherside,andthatthesecampaignssometimescrossethicalboundaries.
RisingThreatofDataAttacks
Theadvancedageofmanyofthevotingmachinesbeingusedintheareaswestudiedmakesitquite
possiblethatatleastsomeofthestrangepatternswearewitnessingareduetohardwareorsoftwarefailures.Buttheunidirectionalresultsofthestatisticalpatternsmakeitunlikelythaterroristhecentralsourceoftheseissues.
Inansweringthequestionofwhetherornotthedatainourvotingmachinesisbeingbreached,itis
importanttobeawareofhowmanymajorgovernment,financialinstitutions,andhigh-profileindustriesinthecountryhavebeensuccessfullyattacked.InMay2016,NewYorkAttorneyGeneralEricSchneidermanreleasedastatementsaying,“hisofficehasreceivedanover40%increaseindatabreach
notificationsinvolvingNewYorkerssofarthisyear.”IRSCommissionerJohnKoskinentoldFortunemagazine,“Wearebasicallyattackedoratleastprobedoveramilliontimesaday.”OneofthoseattacksontheIRSresultedinaspectacularlysuccessfulMay2015hackinwhich,“hackershadusedGet
Transcript[anIRStool]tostealthepersonalinformationof724,000people.”TheU.S.ArmyandNavy,thePentagon,andNASAwereallsuccessfullypenetratedbyahackernamedGaryMcKinnon.TheGuardianquotedMcKinnonassayinghecouldscan“65,000machinesinlessthannineminutes.”This
wasbetween1999and2002;hackershavedevelopedconsiderablymoresophisticatedprotocolssincethen.
Banksandcorporationsarealsodefendingagainstconstantattacksontheirdataandfinancialresources.InFebruary2016,CNNreportedthathackersstolecashfrom100banksandriggedATMsto
spewcashin“oneofthelargestbankheistsever,”totalingapproximately$1billioninstolenfunds.OtherfinancialinstitutionsthathavebeensuccessfullyattackedincludeJPMorganChase,Citigroup,the
FederalReserveBankofNewYork,andsecurityanalystsworkingforBankofAmerica.ThehackercollectiveAnonymoushas,bythemselves,hackedtheChurchofScientology,HiddenWiki,SanFranciscoBART,theDepartmentofJustice,andtheWorldTradeOrganization.Majorcorporateattacksinclude
theTargetdatabreachthatexposedthefinancialinformationof40millioncustomers,theSonyemailscandal,andAdrianLamo’sattackontheNewYorkTimes,whichwasapparentlysoeasythataccordingtogeek.com,Lamo“createdanentryinthe[NewYorkTimes]Op-Eddatabaseforhimself,complete
withcellphonenumber,realname,ande-mailaddress.Inthedescriptionfieldofthedatabase,hesimplyentered’Computerhacking,nationalsecurity,communicationsintelligence.’”
EvaluatingtheOpposition Tworeportshavebeenfiledthat,whileconfirmingthepatternofincreasedcandidatepercentagesinlargeprecincts,arguethatitisnotanindicationoferrororfraud.
AnElectoralSysteminCrisis
InNovember2015,MarkLindeman,apoliticalscientistatColumbiaUniversity,confirmedtheexistenceofthepattern.Buthetookissuewiththeconclusionsofpreviousstudies,dismissingtheiranalysisas
“unsupported”andhaving“nofoundation.”Lindeman’sanalysisdoesnotholduptoclosescrutiny.HeattacksChoquetteandJohnsonwithoutpresentinganyrealdatatobackhisassertions.Heoftenmisinterpretsevidenceandselectivelyignoresfactsthatdon'tmatchwithhistheory.Hisaddendum
critiquingClarksonisequallyweak.Clarksonagreesthathisanalysisisnotstatisticallysupported,writingtous,“MyownworkincludingshareofregisteredRepublicansshowsthatevenwhenthatdataisincluded,thenumberofvotescastremainsasignificantfactor,whichcontradictshisanalysis.”
Aseconddataanalyst,LeviBowles,coveredClarkson’sworkinaseriesoffiveblogpoststitled,Kansas
ElectionFraud.Bowlesisconcerned,asarewe,thatthepatternswearewitnessingcouldbeduetodemographicissues,andhemakeshispointemphaticallyinhiscomments,“…thereisn'tgoodstatisticalevidencethatthemachinesareworkingincorrectly.Theevidence,isthatthereisanunderlying
correlation[ofincreasedpercentageswithlargerprecincts],thoughafterwerecognizetheworldiscomplexandcreation/existenceofprecinctsisnotarandom,stochasticprocess,weseethatcorrelationcompletelydisappear.”
Bowles’critiquedoesnotprovideanexplanationfortheappearanceofthepatternsincetheyear2000.
Precinctshaveneverbeenrandomlycreateddistricts.Sowhywasn’tthispatternpresentinearlierelections?
Furthermore,thecreationofprecinctsisimpactedbyavarietyoffactors,butthosefactorsvaryfromstatetostate.Yetwearewitnessingthepatternacrossabroadselectionofstates,withstrikingly
differentdemographicssuchasWisconsinandLouisiana.Eachofthosestateshasitsownprocessforcreatingprecincts,sotheexistenceofthepatternacrosssuchabroadarrayofstatesarguesagainstit
beingduetofactorsthataffectprecinctformation.
Tosummarizeourcritiqueofthiseffort:Bowlesgoesbeyondrhetoricalargumentsandutilizesastatisticalanalysistocomparethecorrelationofdemographicfactorswiththecorrelationofprecinctsize;buthefailstoapplytherightstatisticalmodel,deliversweakornon-relevantevidence,and
eventuallydoesvalidateClarkson'swork,againsthisownintent.
LargeRepublicanPrecinctTheoryNotValid
Bowlesdoescomeupwithoneintriguingtheorywefoundworthpursuing.HespeculatesthatthereasonforRepublicancandidatesgaininglargerpercentagesinthelargerprecinctsisthatthelargestprecinctsareactuallyconservative-leaningsuburbanprecincts.Hecreatesamapthatshowsthatat
leastsomelargeprecinctsareconservativeintheirpoliticaloutlook.WeaskedClarksontoseparateoutthesuburban“conservative”precinctsfromthe“democratic-leaning”urbanprecinctsandgraphthemseparately.IfBowles’theoryiscorrect,thatthecorrelationofhigherRepublicanpercentagesisdueto
largeconservative-leaningsuburbs,thenthecorrelationwillshowupevenmorestronglyinthoseconservativesuburbs.Isthatwhathappens?
AnElectoralSysteminCrisis
No.IntheRepublican-leaningsuburbs,thereisnoincreaseintheRepublicancandidate’spercentageinthelargeprecincts(Figure19).TherearefewerRepublicansinsidethecityofWichita(37%vs.48%),but
theincreaseintheshareoftheRepublicanvoteinthelargerprecinctsisplainlyevidentthere.Bowlestheory,thatthisstatisticalpatternisduetotheexistenceoflargeconservativeprecinctsinthesuburbs,isnotcorrect—atleastinthisinstance.Whateveriscausingthepattern,hereitisvisiblyimpactingonly
theinner-cityDemocratic-leaningprecincts.Thatraisesthequestionofwhethertheseinner-cityprecinctshavebeentargetedinsomeway.
Fig.19—2014KansasSenaterace-theincreaseofacandidate’spercentageinthelargeprecinctsisonlyseenintheinnercityprecincts,notthesuburbanprecincts
GraphbyBethClarkson
KeyArgumentsPart2
Continuingwithourlistofthekeypointsthatwehaveestablished:
7) Electionresultswithprecisepercentagepoints,inracesthatarealreadyshowingstatisticalirregularities,raisefurtherconcernthattheresultsmaynotbebasedonactualvotes.
8) Theconsistencywithwhichonecandidatebenefitsfromthepatternarguesagainstitbeinggeneratedbyarandomcomputererror.
9) IntheKansas2014Senateracetheirregularstatisticalpatternisevidentinsidethecitylimitsof
Democratic-leaningWichita,butnotinthemoreRepublicansuburbanareas.ThisarguesagainstthetheorythatthepatternisduetolargesuburbanprecinctsbeingmadeupofagreaterpercentageofRepublicanvoters.
AnElectoralSysteminCrisis
LeftSide/RightSideFluctuation
TheretendstobealotoffluctuationontheleftsideoftheCVTgraph.Thisistobeexpected,becauseinsmallprecincts,beforealotofvotesareaccumulated,thegraphwillswingfromtheinfluenceofjusta
fewvotes.Occasionallywehaveseengraphswithsuchsteepunidirectionalmovementontheleftsideofthegraphthatitisconcerning.The2016DelawareDemocraticpresidentialprimaryisanexampleofthis(Figure20).
Fig.20—2016DelawareDemocraticpresidentialprimary Theslopeontheleftisunusuallysteepandunidirectional
graphbyAnselmoSampietro
WefoundscreenshotsonlineconfirmingthattheactualvotecountforSandershaddecreasedsharplyafterearlyreporting(Figure21).ThesetwofactorstogetherleadustosaythattheresultsinDelaware
bearfurtherinvestigationandmaynotbeaccurate.
AnElectoralSysteminCrisis
Fig.21—2016DelawareDemocraticpresidentialprimary,SussexCounty
Sandersvotesgodownfrom6,247to5,630 screenshotcourtesyofRenoBerkeley
ReversedEffect
Inthe2016primaryelection,ontwooccasionswesawirregularchartsthattiltedinSanders’sfavor
(Figure22).Butthiswasrare.Ourtheoryisthatvotetotalsarebeingmanipulatedtoachieveapre-determinedpercentage,andthatinthisprocessvotesmightbeshiftedatvarioustimesbetweenmultiplecandidates.
AnElectoralSysteminCrisis
Fig.22—2016NewYorkDemocraticpresidentialprimary,BroomeCounty TheCVTgraphshowsSanderswithanunusualgaininvotesinthelargestcounties
Ascreenshotshows112.8%reporting,followedbybothcandidatesexperiencingalossofvotes graphbyAnselmoSampietro&screenshotcourtesyofAimeeRoxColeman
VoterFraudvs.ElectionFraud
Thereisoneothercritiquethatisworthtakingthetimetoexamine.ProfessorJ.CelesteLay’scommentinaninterviewwithDougJohnsonHatlemisagoodexampleofthis:
“MostofthisdiscussionisdrivenbySanderssupporterswhoaredisappointedheisnotwinningand
wanttoclaimhehasmoresupportintheDemocraticPartythanheactuallydoes…Untilprovenotherwise,I’llgowiththenumerousstudiesdemonstratingtheinfinitesimalamountofvoterfraudinU.S.elections.”
Layisconflatingvoterfraudwithelectionfraud.Theyareoppositebehaviors.Voterfraudiswhena
singlevoteristryingtoscamthesystembyvotingtwice,orvotingwhenheorsheisnotauthorizedtodoso.IthasbeendemonstratedtoberareintheU.S.Electionfraudreferstotheillegalinterferencewiththeprocessofanelection.Inelectionfraud,thevotersarethevictimsofanorganizedefforttorob
themoftheirrightfulinfluence.Howcommonisit?
AccordingtoWalterR.Mebane,Jr.andAllenHicken,bothonthepoliticalsciencefacultyattheUniversityofMichigan,electionfraudoccursinamajorityofdemocracies.Intheir2015GuidetoElectionForensics,theywrite,“Duringthefirsthalfof2015alone,forexample,allegationsofelection
fraudoccurredinBangladesh,India,Israel,Macedonia,Nigeria,Pakistan,Russia,Togo,theUnitedKingdom,theUnitedStates,andZambia…datacompiledbyKelleyandKolev(2010)onnational
AnElectoralSysteminCrisis
electionsconductedinmorethan170countriesfrom1978–2004indicate…61%ofcountriesexperiencedsomedegreeof(known)cheating.”
Incorrespondence,Mebanewasfirmthat,“theproblemofdetectingelectionfraudreallyisextremely
difficult.”Thisraisesthelargerissue:Isitincumbentonthevotingpublictopoliceelectionsandprovethatfraudistakingplace?Orisitmoreappropriatelyincumbentonelectionofficialstoprovideconvincingevidencethattheelectionresultsareaccurateandsecure?
TestimonyUnderOath
Itwouldbehelpfulifsomeonewouldcomeforwardandtestifyunderoaththattheyhadtamperedwith
orbeenaskedtotamperwiththeelectionresultsofvotingmachines.InhisarticleonMedium,SpencerGundertremindedusthatClintCurtisdidexactlythat.
In2004,computerprogrammerClintCurtistestifiedunderoathbeforetheU.S.HouseJudiciaryCommitteethathe“wroteaprototypefor[atthattime]presentCongressmanTomFeeney,”ofa
programthat“wouldflipthevote51–49.Whoeveryouwantedittogotoandwhicheverraceyouwantedtowin.”Curtistestifiedthathewastoldbyhiscompanythat,“weneededtocontrolthevoteinSouthFlorida(5:23.)”
Inthisclipfromthedocumentary,Uncounted,Curtisdemonstratesthe“flip.”At4:17hesays,“Twenty-
fourlinesofcode.Youneverseeit.”
FutureElections
Evenasdoubtsriseabouttheaccuracyandsecurityofelectronicvotingmachines,countiescontinuetoinvestmillionsofdollarspurchasingthem.InMarchMarionCountyIndianaspent1.4milliondollars
purchasingelectronicvotingequipmentthatincludedtheES&SDS200in-precincttabulators;thesamemachinesthatproducedirregularelectionresultsinthe2016NewYorkprimary.InSedgwickCountyKansas,thecountycommissionersapprovedameasurethisweektospend7.8millionovertenyearson
newES&Smachinesaswell.BethClarksonwasatthemeeting,andexpressedherconcerns,butthevotetopurchasethemachinesstillpassed.Thatvotewascountedbyhand.
The2016Primary
Themajorityofthedataweexaminedsuggeststhatthetwocandidatescurrentlyslatedtoaccepttheirparty’snominationinthe2016presidentialprimaryraces,receivedadifferentnumberofvotesthan
whathasbeenofficiallyreported.
AnElectoralSysteminCrisis
OntheRepublicanside,statisticalanalysisindicatesthatDonaldTrumpprobablyreceivedmorevotesthanwhathasbeenreportedandcertified.Becausehewasabletoovercomehisopposition,evenwith
theirregularities,hisselectionasthepresumptiveRepublicannomineeissupportedbythedata.
Aswestatedintheopening,thisisnotthecaseontheDemocraticside.Theoverwhelmingmajorityofthealmosttwodozenstatesthatweanalyzed,demonstratedirregularities.InalmosteveryinstancethediscrepanciesfavoredHillaryClinton.Inalllikelihoodthecurrentresultshaveassignedheragreater
percentageofthevotethanshemayhaveactuallyreceived,whilesimultaneouslyunder-reportingBernieSanders’legitimatevoteshare.
Thedifferencebetweenthereportedtotals,andourbestestimateoftheactualvote,variesconsiderablyfromstatetostate.Howeverthesedifferencesaresignificant–sometimesmorethan10%
–andcouldchangetheoutcomeoftheelection.Weintendtoreportonthepercentagethateachstatemaybeoff,basedonastatisticalanalysisofasmanystatesaspossible.
Itishardtoconceiveofalegitimatetransferofpowerfollowinganelectionthathasbeenthisflawed.Werecommendthatmanyoftheseelectionsbeexamined,andiffoundtobeinaccurate,decertified.
Wherepaperballotsareavailable,itwouldbeinformativetocountthembyhand.Wherepaperballotsarenotavailable,itwouldbemoreequitabletoholdasecondvote,onethatdoesnotutilizeanyelectronicvotingequipment.Itwouldbebestifthecountsarevideotaped,andforobserverstobeable
towatchincloseenoughproximitytoverifytheaccuratecountofeveryballot.Possiblytheballotsthemselvesneedtobeindividuallyphotographedanditemized.Handcounts,historically,havealsobeensubjecttoelectionfraud,sotheprotocolsmustfocusontransparency,chainofcustody,and
secureandverifiableresults.
Weunderstandthatthisisunprecedented.WhiletheU.S.doeshavealonghistoryofelectionfraud,wedonotbelieveithaseverbeenthiswell-documentedpriortotheendoftheelectioncycle.
PhilipB.Stark,astatisticsprofessoratUCBerkeley,whohasbeeninstrumentalindesigningnew
auditingtechniquesforelectionsagreesthat,“Closerscrutinyofelectionsisnecessaryandwelcome…thatrequiresapapertrail,convincingevidencethatthepapertrailiscompleteandaccurate,andarisk-limitingauditofthepapertrailorafullmanualtallytoprovideconvincingevidencethatthepapertrail
matchestheannouncedresult.”
BethClarksonadds,“It’spossiblethattodoagoodauditwouldbemoreexpensiveandlesstransparentthanahandcountbecausetodoagoodauditrequiresexpertise.”
SecuringOurDemocracy
Thelackofaccuracyinourelectionsistrulyabetrayalofourancestorswhofoughtanddiedforthedemocraticprocess.ItisabetrayalofthesoldierswholosttheirfeettoamputationintheValleyForge
winter.Itisabetrayalofthewomenwhowenttoprisonandstarvedthemselvestojointhefranchise.Itisabetrayalofthecivilrightsworkerswhodiedfortherighttoregistertovote.Therecanbenodebate
AnElectoralSysteminCrisis
aboutwhetherornotthevoteisaccurate.WemustknowthatitisaccuratethewayweknowthattheEarthrevolvesaroundthesunandnotviceversa.
Thepeoplewhowintheseelectionswilldecidewhetherornotwegotowar,howmanypeopleget
whatjobs,whereourchildrenattendwhatqualityschools,thequalityoftheairwebreatheandthewaterwedrink,andsomuchmore.Theentireidentityofournationrestsonourself-imageasaself-correctingdemocracywhoseleadersareaccountabletothevotersthatelectthem.Tonotknowwith
100%confidencethatthoseleadersaretheleadersthatweactuallyvotedforistheveryessenceofanexistentialcrisis.Thisisastateofemergency.Wemustmoverapidlytosecuretheintegrityofthevote.
Weneedtoimmediatelyimplementrobustauditprocedures.Then,asrapidlyaspossible,wemustjointheotherlegitimatedemocraciesoftheworldandimplementasystemofpaperballots,hand-counted
inasecureprocessthatisopentothepublic,invitesmediascrutiny,andhasstrongchainofcustodyprotocols.Inthisway,wecanachieveaccurate,verifiableresults.EachcitizenoftheUnitedStates,andindeedtheworld,deservesthisfromus.
AbouttheTeam
luluFries’datisanEdwardR.Murrowaward-winningjournalistanddocumentaryfilmmaker.She
receivedaBestDocumentaryawardforherfirstfeature-lengthdocumentary(producer/director)HollerBack—[not]VotinginanAmericanTown;afilmthatexploressystemicissuesinourelectionsthatdiscouragevoterparticipation.Clipsareavailableforviewinghere.Hernetworknewsexperience
includesassignmentsasaneditorforCBSEveningNews,Nightline,SundayMorning,TheTodayShow,andGoodMorningAmerica;SheproducedandeditedprofilesofDemocraticcandidatesforMSNBCand
hasdonelong-formatdocumentaryworkwithNBCNewsandCNBC.ShewasontheeditingteamofGideon’sArmy,anEmmy-nominateddocumentarythatfollowsthepersonalstoriesofpublicdefendersintheDeepSouth.Forachangeofpace,shecreatedanddirectedvoterregistrationPSAsforVH1that
featuredTriumphtheInsultComicDog.ShehasproducedanddirectedfrequentlywithherindependentproductioncompanyShugahWorks,includingthreeshortfilmsforNewYorkCityschools.Herfullbioisavailablehere.Followherontwitter@shugahworks.
AnselmoSampietroholdsaMasterofStatisticsdegreefromtheUniversityofBologna,Italyandhasalso
studiedattheUniversityofTechnologySydney,inAustralia,andtheUniversityofWarsaw,inPoland.HecurrentlyleadsateamofdataanalystsforacompanybasedinLondon.Hecollaborated(throughInnoCentive)withGeneralFusion,aCanadianstart-upthatisdevelopingcleanandreliablenuclear
fusionreactors,tobuildastatisticalmodelpredictingplasmaperformance.HehasspecializedintheuseofstatisticalanalysissoftwareR,whichistheanalytictoolusedthroughoutthisarticle.
FritzScheurenisaSeniorFellowandVicePresidentatNORCintheCenterforExcellenceinSurveyResearch.Scheurenhasanunparalleledrecordofworkoncomplexsubstantivetasksrelatedto
AnElectoralSysteminCrisis
samplingandtotheanalysisofdatafromgovernmentagenciesandprivatesectorinstitutions.NotableworkatNORCincludesalandmarkprojectresolvingissueswithIndianTrustFundaccountsattheU.S.
DepartmentoftheInterioraswellasworkwiththeFederalReserve,particularlytheSurveyofConsumerFinancesandattheCensusBureau.ScheureniscurrentlyleadingimpactevaluationsofMillenniumChallengeCorporation(MCC)programsinGeorgiaandLesotho,andformerlyadvisedthe
MCConimpactevaluationsinArmeniaandVanuatu.Hebringsaprovenabilitytosolvehard,non-standardproblemsandthetheoreticalbackgroundtobackthesesolutionsupwithsoundstatisticalarguments.
ScheurenalsoservesontheStatisticsFacultyatTheGeorgeWashingtonUniversity,wherehehadbeen
aVisitingProfessorofStatistics.AtGWU,hecreatedasuccessfulsurveysamplingcertificateprogramwhichhestillteaches.Mostofhisover450appliedandtheoreticalpapers,presentations,monographs,andbooksareonsamplingaspectsofdatacollectionprimarilyinasurveycontext,withanemphasison
sampledesignandestimation,processquality,andthehandlingofmissingdata.Hehaswrittenoverhalfadozenbooks.
Scheurenservedatthe100thPresidentoftheAmericanStatisticalAssociationandchairedtheASASectionsonSurveyResearchMethodsand,later,theSocialStatisticsSection.Hehasbeentheir
ScientificSecretaryattheInternationalAssociationofSurveyStatisticians.HeisaFellowoftheAmericanStatisticalAssociationandtheAmericaSocietyfortheAdvancementofScience.
ElectionJusticeUSA–thepaperwasassistedinitsresearchanddevelopmentbyElectionJusticeUSA,anationalnon-partisancoalitionofseasonedelectionintegrityexperts,statisticians,attorneys,journalists
andactivists,whosemissionistomakesureeachAmerican'srighttovotecanbeexercisedwithoutissueinaccurateandhonestelections.
AnElectoralSysteminCrisis
LinkstoElectionResults
2000WIRepublicanpresidentialprimary
2000FLAlachuaCountyDemocraticpresidentialprimary
2004FLAlachuaCountyDemocraticpresidentialprimary
2016DEDemocraticpresidentialprimary
2016LADemocraticpresidentialprimary
2016NewYorkDemocraticpresidentialprimary:
http://vote.nyc.ny.us/downloads/csv/election_results/2016/20160419Presidential%20Primary/01000200000Citywide%20Democratic%20President%20Citywide%20EDLevel.csv
2016NYBroomeCountyDemocraticpresidentialprimary
2016NYColumbiaCountyDemocraticpresidentialprimary
2016WIRepublicanandDemocraticpresidentialprimary
LinkstoChoquetteandJohnson’sElectionData
Version1.5
Version2.1
LinkstoClarkson’sElectionData
LinkstoLindeman’sElectionData
VotingEquipmentbyType
Wisconsinvotingequipment