The referendum 28 months on.

Dr. Chris Hanretty, Royal Holloway, University of London

Introduction InJuneof2016,theUKvotedtoleavetheEuropeanUnion.

28monthshavepassedsincethen,andpublicopinionhaschanged.

Thereferendumwascounted,notinWestminsterconstituencies,butinlocalauthorityareas.

JustlookingatcountingareasinEngland,ScotlandandWales,aroundsevenintenlocalauthorityareasvotedtoleave,or263intotal.

Howwouldthatlookifwere-ranthereferendumnow?

Wecanuseatoolcalledmultilevelregressionandpost-stratification(MRP)toanswerthatquestion.Thistoolmodelstheopinionsofrespondentstoalargenationalsurvey–inthiscase,asurveyofaround20,000people.

Thoseopinionsarelinkedtopeople’sdemographiccharacteristics–thingsliketheirage,theirhighesteducationalqualifications.Usingthecensus,it’spossibletolookuphowmanypeopleofeachtypeliveineacharea,andmakepredictionsabouthoweachtypethinks.Byaddingupthepredictionsforeachtype,wegetanestimateofhoweachareamightvote.It’snotinfallible,butit’sourbestguidetolocalopinion.

Usingthattool,howwouldtheresultsofthereferendumlooknow?Well,accordingtothisanalysis,theaveragesupportforLeaveacrossGreatBritain,lookingjustatthosewhowouldbelikelytoturnouttovote,wouldbe45.6%–somesevenpointsdownonthefigurefrom2016.

Thatmeansthatfarfewerlocalauthorityareaswouldbelikelytovoteleave:around158,orclosertofourintenareas.

Thatseemslikearadicalshift–butalotofthesechangesinopinionaregradual.TheplacesthatmostsupportedLeavein2016arestilltheplacesthatmostsupportleavenow.YoucanseethatinFigure1,whichshowsthesupportforLeavein2016alongthehorizontalaxis,againstsupportforLeaveinahypotheticalreferendumheldtoday.

SupportforLeaving,thenandnow.EachdotrepresentsalocalauthorityareainGreatBritain.Thebluelineindicatesthelineofbestfit.Thesolidlineshowshowthelineswouldfallifnoonechangedtheirmind.

Thebluelineinthefigureisatrendline.Itrunsprettyclosetothesolidline(whichshowswhatwouldhappenifnothingchanged),butitisslightlyflatter.ThatmeansthatRemainvotingareasstayasRemainastheyeverwere,whilstLeave-votingareasareactuallyswitchingathigherrates.

ThiscanbeseeninFigure2,whichratherthanplotsupportforLeavingtoday,plotstheestimatedchangeintheLeavevotesharebetween2016and2018.Thelargestpanel,onthelefthandside,plotstherelationshipbetweentheLeavevotein2016andthechange.Therelationshipisnegative:thebiggertheLeavevotein2016,themoretheareahasswungroundtoRemain,evenifitstartedfromaLeavierstartingpoint.

Threepatternsofchange.SwitchesawayfromLeavearegreaterthestrongertheinitialsupportforLeave(left-handpanel).SwitchesawayfromLeaveareweakerthestrongerinareasthatvotedConservativeinthe2014EPelections(top-rightpanel).There’snosimplerelationshipwiththeUKIPvote(bottomrightpanel).

ThetoprighthandpanelofFigure2showsthepatternsofchangeaccordingtoConservativesupportinthe2014EuropeanParliamentelections.We’reusingthe2014EuropeanParliamentelectionsbecausethesetoowerecountedatthelocalauthoritylevel.Astheplotshows,theswitchawayfromLeaveissmallerthebettertheConservativesdid.Conservative-votingareasaremorelikelytostickwithLeave–or,conversely,Labourareasaremorelikelytoswitch.

ThebottomrighthandpanelshowsthesamefigurebutforUKIPsupportin2014.Here,therelationshipisessentiallyflat:there’snosimplerelationship.However,becauseUKIPsupportin2014wassuchastrongpredictoroftheLeavevote,thismeansthatifwecontrolfortheLeavevote,wefindthatareaswhichwerestrongerbedsofsupportforUKIPare,justlikeConservativevotingareas,morelikelytostickwithLeave.

ChangeinsupportforLeavingbyregion.Thereddotgivestheaveragechangebyregion.

Youmightwonderwhethertheseestimatesarerobust.Dotheychangemuchifwechangesomeoftheassumptions?

Ifweignorequestionsofturnout,andassumethateveryonevotesinafreshreferendum,thenthefiguresdon’tchangemuch.Wegofrom45.6percentto45.9percent-atinydifference.

Whataboutourconfidenceintheestimatesforeachplace–canwesummarizetheseestimatesinasinglefigure,ordoweneedarangeoffigures?

Ifwe’dconductedastandardpollineacharea,thenthemarginoferrorforeachlocalauthoritywouldbeplusorminusthreepercent.Butbecausewehaven’tconducted380thousandinterviews,wehavetosettleforsomethingless–plusorminusfivepercentineachcase.

Thatmeansthatalthoughweknowalotofareashaveswitched,wecanonlybereallyveryconfidentaboutamuchsmallergroup.Herearetheareasmostlikelytohaveswitched:

• Nottingham(40.1%Leavenowbutwas50.8%Leavethen).• Luton(43.8%Leavenowbutwas56.5%Leavethen).• Slough(41.6%Leavenowbutwas54.3%Leavethen).• Southampton(41.8%Leavenowbutwas53.8%Leavethen).• HighPeak(44.3%Leavenowbutwas50.5%Leavethen).

• Watford(43.6%Leavenowbutwas50.3%Leavethen).• Canterbury(44.6%Leavenowbutwas51%Leavethen).• Cherwell(44.5%Leavenowbutwas50.3%Leavethen).• ReigateandBanstead(44.9%Leavenowbutwas50.5%Leavethen).• Knowsley(39.7%Leavenowbutwas51.6%Leavethen).• NorthTyneside(45.5%Leavenowbutwas53.4%Leavethen).• Birmingham(41.8%Leavenowbutwas50.4%Leavethen).• Sutton(44.9%Leavenowbutwas53.7%Leavethen).• IsleofAnglesey(44.2%Leavenowbutwas50.9%Leavethen).• Swansea(43.2%Leavenowbutwas51.5%Leavethen).• RhonddaCynonTaf(43.5%Leavenowbutwas53.7%Leavethen).

Aswellasaskinghowpeoplewouldvoteinafreshreferendum,wealsoaskedhowpeoplewouldviewparticularproposals,andhowtheywouldvoteonaneventualdeal.

Hereweretheparticularproposalsweasked,togetherwiththeoverallsupport(notincludingthosethatsaiddon’tknow)

• newchecksbeingintroducedongoodscrossingtheIrishseabetweenNorthernIrelandandtherestoftheUK(41%)

• UKandEUcitizenswhowishedtodosobeingabletoliveandworkineachother’scountries

(76%) • LimitationsontheUK’sabilitytomaketradedealswithcountriesoutsidetheEU(25%) • FollowingEUregulationsonmanufacturedgoodssuchasfridges,vacuumcleanersandlightbulbs

(62%) • Supportforthedealasitstands(43%)

Theseproposalshaveverydifferentpatternsofsupport.Thisisshowninthenextfigure,whichshowshoweachareavotedin2016,againstitscurrentsupportforeachproposal.

Theassociationbetweenhoweachareavotedin2016,andhowmuchtheywouldsupportparticularproposalsifthesewerethepriceofadealbetweentheEUandtheUK

Asyoucansee,thehigherthe2016voteforLeave,thelowerthesupportforcontinuingfreedomofmovement,andthelowerthesupportforcontinuingtoacceptEUregulations–eveniftheoveralllevelofsupportfortheseproposalsisquitehigh.

ThestoryisquitedifferentforgoodschecksintheIrishsea,andforlimitationsontradedeals.Thesehavemuchlowersupport,butthey’realsomuchlessstronglyassociatedwiththe2016Leavevote.

Thefinalplotshowstherelationshipbetweenhoweachareavotedin2016,andhowtheywouldvoteonthepresentdeal,howevertheyunderstandthat.

Supportforthecurrentdeal.

Surprisingly,it’sLeave-votingareaswhicharemorelikelytosupportthedeal,eventhoughsupportisnothigh.Remain-votingareasmaybeholdingoutforsomethingbetter.

Full data by local authority can be found here:

https://www.survation.com/wp-content/uploads/2018/11/LA-predictions-from-MRP-1.xlsx