Writing Survey Questions | Pew Research Center

Designing the questionnaire is complicated because surveys can ask about topics in varying degrees of detail, questions can be asked in different ways, and ... Politics&PolicyInternationalAffairsImmigration&MigrationRace&EthnicityReligionGenerations&AgeGender&LGBTFamily&RelationshipsEconomy&WorkScienceInternet&TechnologyNewsHabits&MediaMethodologicalResearchFulltopiclist Asia&thePacificEurope&RussiaLatinAmericaMiddleEast&NorthAfricaNorthAmericaSub-SaharanAfricaMultipleRegions/Worldwide FeaturesFactSheetsVideosDataEssays Seeourresearchon:Russia|SupremeCourt|COVID-19 HomeOurMethodsU.S.Surveys Perhapsthemostimportantpartofthesurveyprocessisthecreationofquestionsthataccuratelymeasuretheopinions,experiencesandbehaviorsofthepublic.Accuraterandomsamplingwillbewastediftheinformationgatheredisbuiltonashakyfoundationofambiguousorbiasedquestions.Creatinggoodmeasuresinvolvesbothwritinggoodquestionsandorganizingthemtoformthequestionnaire. Questionnairedesignisamultistageprocessthatrequiresattentiontomanydetailsatonce.Designingthequestionnaireiscomplicatedbecausesurveyscanaskabouttopicsinvaryingdegreesofdetail,questionscanbeaskedindifferentways,andquestionsaskedearlierinasurveymayinfluencehowpeoplerespondtolaterquestions.Researchersarealsoofteninterestedinmeasuringchangeovertimeandthereforemustbeattentivetohowopinionsorbehaviorshavebeenmeasuredinpriorsurveys. Surveyorsmayconductpilottestsorfocusgroupsintheearlystagesofquestionnairedevelopmentinordertobetterunderstandhowpeoplethinkaboutanissueorcomprehendaquestion.Pretestingasurveyisanessentialstepinthequestionnairedesignprocesstoevaluatehowpeoplerespondtotheoverallquestionnaireandspecificquestions,especiallywhenquestionsarebeingintroducedforthefirsttime. Formanyyears,surveyorsapproachedquestionnairedesignasanart,butsubstantialresearchoverthepastfortyyearshasdemonstratedthatthereisalotofscienceinvolvedincraftingagoodsurveyquestionnaire.Here,wediscussthepitfallsandbestpracticesofdesigningquestionnaires. Questiondevelopment Thereareseveralstepsinvolvedindevelopingasurveyquestionnaire.Thefirstisidentifyingwhattopicswillbecoveredinthesurvey.ForPewResearchCentersurveys,thisinvolvesthinkingaboutwhatishappeninginournationandtheworldandwhatwillberelevanttothepublic,policymakersandthemedia.Wealsotrackopiniononavarietyofissuesovertimesoweoftenensurethatweupdatethesetrendsonaregularbasistobetterunderstandwhetherpeople’sopinionsarechanging. AtPewResearchCenter,questionnairedevelopmentisacollaborativeanditerativeprocesswherestaffmeettodiscussdraftsofthequestionnaireseveraltimesoverthecourseofitsdevelopment.Wefrequentlytestnewsurveyquestionsaheadoftimethroughqualitativeresearchmethodssuchas focusgroups,cognitiveinterviews,pretesting(oftenusingan online,opt-insample),oracombinationoftheseapproaches.Researchersuseinsightsfromthistestingtorefinequestionsbeforetheyareaskedinaproductionsurvey,suchasontheATP. Measuringchangeovertime Manysurveyorswanttotrackchangesovertimeinpeople’sattitudes,opinionsandbehaviors.Tomeasurechange,questionsareaskedattwoormorepointsintime.Across-sectionaldesignsurveysdifferentpeopleinthesamepopulationatmultiplepointsintime.Apanel,suchastheATP,surveysthesamepeopleovertime.However,itiscommonforthesetofpeopleinsurveypanelstochangeovertimeasnewpanelistsareaddedandsomepriorpanelistsdropout.ManyofthequestionsinPewResearchCentersurveyshavebeenaskedinpriorpolls.Askingthesamequestionsatdifferentpointsintimeallowsustoreportonchangesintheoverallviewsofthegeneralpublic(orasubsetofthepublic,suchasregisteredvoters,menorBlackAmericans),orwhatwecall“trendingthedata”. Whenmeasuringchangeovertime,itisimportanttousethesamequestionwordingandtobesensitivetowherethequestionisaskedinthequestionnairetomaintainasimilarcontextaswhenthequestionwasaskedpreviously(see questionwording and questionorder forfurtherinformation).Allofoursurveyreportsincludeatoplinequestionnairethatprovidestheexactquestionwordingandsequencing,alongwithresultsfromthecurrentsurveyandprevioussurveysinwhichweaskedthequestion. TheCenter’stransitionfromconductingU.S.surveysbylivetelephoneinterviewingtoanonlinepanel(around2014to2020)complicatedsomeopiniontrends,butnotothers.Opiniontrendsthataskaboutsensitivetopics(e.g.,personalfinancesorattendingreligiousservices)orthatelicitedvolunteeredanswers(e.g.,“neither”or“don’tknow”)overthephonetendedtoshowlargerdifferencesthanothertrendswhenshiftingfromphonepollstotheonlineATP.TheCenteradoptedseveralstrategiesforcopingwithchangestodatatrendsthatmayberelatedtothischangeinmethodology.Ifthereisevidencesuggestingthatachangeinatrendstemsfromswitchingfromphonetoonlinemeasurement,Centerreportsflagthatpossibilityforreaderstotrytoheadoffconfusionorerroneousconclusions. Open-andclosed-endedquestions Oneofthemostsignificantdecisionsthatcanaffecthowpeopleanswerquestionsiswhetherthequestionisposedasanopen-endedquestion,whererespondentsprovidearesponseintheirownwords,oraclosed-endedquestion,wheretheyareaskedtochoosefromalistofanswerchoices. Forexample,inapollconductedafterthe2008presidentialelection,peoplerespondedverydifferentlytotwoversionsofthequestion:“Whatoneissuematteredmosttoyouindecidinghowyouvotedforpresident?”Onewasclosed-endedandtheotheropen-ended.Intheclosed-endedversion,respondentswereprovidedfiveoptionsandcouldvolunteeranoptionnotonthelist. Whenexplicitlyofferedtheeconomyasaresponse,morethanhalfofrespondents(58%)chosethisanswer;only35%ofthosewhorespondedtotheopen-endedversionvolunteeredtheeconomy.Moreover,amongthoseaskedtheclosed-endedversion,fewerthanone-in-ten(8%)providedaresponseotherthanthefivetheywereread.Bycontrast,fully43%ofthoseaskedtheopen-endedversionprovidedaresponsenotlistedintheclosed-endedversionofthequestion.Alloftheotherissueswerechosenatleastslightlymoreoftenwhenexplicitlyofferedintheclosed-endedversionthanintheopen-endedversion.(Alsosee “HighMarksfortheCampaign,aHighBarforObama” formoreinformation.) Researcherswillsometimesconductapilotstudyusingopen-endedquestionstodiscoverwhichanswersaremostcommon.Theywillthendevelopclosed-endedquestionsbasedoffthatpilotstudythatincludethemostcommonresponsesasanswerchoices.Inthisway,thequestionsmaybetterreflectwhatthepublicisthinking,howtheyviewaparticularissue,orbringcertainissuestolightthattheresearchersmaynothavebeenawareof. Whenaskingclosed-endedquestions,thechoiceofoptionsprovided,howeachoptionisdescribed,thenumberofresponseoptionsoffered,andtheorderinwhichoptionsarereadcanallinfluencehowpeoplerespond.OneexampleoftheimpactofhowcategoriesaredefinedcanbefoundinaPewResearchCenterpollconductedinJanuary2002.Whenhalfofthesamplewasaskedwhetheritwas“moreimportantforPresidentBushtofocusondomesticpolicyorforeignpolicy,”52%chosedomesticpolicywhileonly34%saidforeignpolicy.Whenthecategory“foreignpolicy”wasnarrowedtoaspecificaspect–“thewaronterrorism”–farmorepeoplechoseit;only33%chosedomesticpolicywhile52%chosethewaronterrorism. Inmostcircumstances,thenumberofanswerchoicesshouldbekepttoarelativelysmallnumber–justfourorperhapsfiveatmost–especiallyintelephonesurveys.Psychologicalresearchindicatesthatpeoplehaveahardtimekeepingmorethanthisnumberofchoicesinmindatonetime.Whenthequestionisaskingaboutanobjectivefactand/ordemographics,suchasthereligiousaffiliationoftherespondent,morecategoriescanbeused.Infact,theyareencouragedtoensureinclusivity.Forexample,PewResearchCenter’sstandardreligionquestionsincludemorethan12differentcategories,beginningwiththemostcommonaffiliations(ProtestantandCatholic).Mostrespondentshavenotroublewiththisquestionbecausetheycanexpecttoseetheirreligiousgroupwithinthatlistinaself-administeredsurvey. Inadditiontothenumberandchoiceofresponseoptionsoffered,theorderofanswercategoriescaninfluencehowpeoplerespondtoclosed-endedquestions.Researchsuggeststhatintelephonesurveysrespondentsmorefrequentlychooseitemsheardlaterinalist(a“recencyeffect”),andinself-administeredsurveys,theytendtochooseitemsatthetopofthelist(a“primacy”effect). Becauseofconcernsabouttheeffectsofcategoryorderonresponsestoclosed-endedquestions,manysetsofresponseoptionsinPewResearchCenter’ssurveysareprogrammedtoberandomizedtoensurethattheoptionsarenotaskedinthesameorderforeachrespondent.Rotatingorrandomizingmeansthatquestionsoritemsinalistarenotaskedinthesameordertoeachrespondent.Answerstoquestionsaresometimesaffectedbyquestionsthatprecedethem.Bypresentingquestionsinadifferentordertoeachrespondent,weensurethateachquestiongetsaskedinthesamecontextaseveryotherquestionthesamenumberoftimes(e.g.,first,lastoranypositioninbetween).Thisdoesnoteliminatethepotentialimpactofpreviousquestionsonthecurrentquestion,butitdoesensurethatthisbiasisspreadrandomlyacrossallofthequestionsoritemsinthelist.Forinstance,intheexamplediscussedaboveaboutwhatissuematteredmostinpeople’svote,theorderofthefiveissuesintheclosed-endedversionofthequestionwasrandomizedsothatnooneissueappearedearlyorlateinthelistforallrespondents.Randomizationofresponseitemsdoesnoteliminateordereffects,butitdoesensurethatthistypeofbiasisspreadrandomly. Questionswithordinalresponsecategories–thosewithanunderlyingorder(e.g.,excellent,good,onlyfair,poorORveryfavorable,mostlyfavorable,mostlyunfavorable,veryunfavorable)–aregenerallynotrandomizedbecausetheorderofthecategoriesconveysimportantinformationtohelprespondentsanswerthequestion.Generally,thesetypesofscalesshouldbepresentedinordersorespondentscaneasilyplacetheirresponsesalongthecontinuum,buttheordercanbereversedforsomerespondents.Forexample,inoneofPewResearchCenter’squestionsaboutabortion,halfofthesampleisaskedwhetherabortionshouldbe“legalinallcases,legalinmostcases,illegalinmostcases,illegalinallcases,”whiletheotherhalfofthesampleisaskedthesamequestionwiththeresponsecategoriesreadinreverseorder,startingwith“illegalinallcases.”Again,reversingtheorderdoesnoteliminatetherecencyeffectbutdistributesitrandomlyacrossthepopulation. Questionwording Thechoiceofwordsandphrasesinaquestioniscriticalinexpressingthemeaningandintentofthequestiontotherespondentandensuringthatallrespondentsinterpretthequestionthesameway.Evensmallwordingdifferencescansubstantiallyaffecttheanswerspeopleprovide. [ViewmoreMethods101Videos] AnexampleofawordingdifferencethathadasignificantimpactonresponsescomesfromaJanuary2003PewResearchCentersurvey.Whenpeoplewereaskedwhethertheywould“favororopposetakingmilitaryactioninIraqtoendSaddamHussein’srule,”68%saidtheyfavoredmilitaryactionwhile25%saidtheyopposedmilitaryaction.However,whenaskedwhethertheywould“favororopposetakingmilitaryactioninIraqtoendSaddamHussein’srule evenifitmeantthatU.S.forcesmightsufferthousandsofcasualties,”responsesweredramaticallydifferent;only43%saidtheyfavoredmilitaryaction,while48%saidtheyopposedit.TheintroductionofU.S.casualtiesalteredthecontextofthequestionandinfluencedwhetherpeoplefavoredoropposedmilitaryactioninIraq. Therehasbeenasubstantialamountofresearchtogaugetheimpactofdifferentwaysofaskingquestionsandhowtominimizedifferencesinthewayrespondentsinterpretwhatisbeingasked.Theissuesrelatedtoquestionwordingaremorenumerousthancanbetreatedadequatelyinthisshortspace,butbelowareafewoftheimportantthingstoconsider: First,itisimportanttoaskquestionsthatareclearandspecificandthateachrespondentwillbeabletoanswer.Ifaquestionisopen-ended,itshouldbeevidenttorespondentsthattheycananswerintheirownwordsandwhattypeofresponsetheyshouldprovide(anissueorproblem,amonth,numberofdays,etc.).Closed-endedquestionsshouldincludeallreasonableresponses(i.e.,thelistofoptionsisexhaustive)andtheresponsecategoriesshouldnotoverlap(i.e.,responseoptionsshouldbemutuallyexclusive).Further,itisimportanttodiscernwhenitisbesttouseforced-choiceclose-endedquestions(oftendenotedwitharadiobuttoninonlinesurveys)versus“select-all-that-apply”lists(orcheck-allboxes).A2019Centerstudyfoundthatforced-choicequestionstendtoyieldmoreaccurateresponses,especiallyforsensitivequestions. Basedonthatresearch,theCentergenerallyavoidsusingselect-all-that-applyquestions. Itisalsoimportanttoaskonlyonequestionatatime.Questionsthataskrespondentstoevaluatemorethanoneconcept(knownasdouble-barreledquestions)–suchas“HowmuchconfidencedoyouhaveinPresidentObamatohandledomesticandforeignpolicy?”–aredifficultforrespondentstoanswerandoftenleadtoresponsesthataredifficulttointerpret.Inthisexample,itwouldbemoreeffectivetoasktwoseparatequestions,oneaboutdomesticpolicyandanotheraboutforeignpolicy. Ingeneral,questionsthatusesimpleandconcretelanguagearemoreeasilyunderstoodbyrespondents.Itisespeciallyimportanttoconsidertheeducationlevelofthesurveypopulationwhenthinkingabouthoweasyitwillbeforrespondentstointerpretandansweraquestion.Doublenegatives(e.g.,doyoufavororoppose not allowinggaysandlesbianstolegallymarry)orunfamiliarabbreviationsorjargon(e.g.,ANWRinsteadofArcticNationalWildlifeRefuge)canresultinrespondentconfusionandshouldbeavoided. Similarly,itisimportanttoconsiderwhethercertainwordsmaybeviewedasbiasedorpotentiallyoffensivetosomerespondents,aswellastheemotionalreactionthatsomewordsmayprovoke.Forexample,ina2005PewResearchCentersurvey,51%ofrespondentssaidtheyfavored“makingitlegalfordoctorstogiveterminallyillpatientsthemeanstoendtheirlives,”butonly44%saidtheyfavored“makingitlegalfordoctorstoassistterminallyillpatientsincommittingsuicide.”Althoughbothversionsofthequestionareaskingaboutthesamething,thereactionofrespondentswasdifferent.Inanotherexample,respondentshavereacteddifferentlytoquestionsusingtheword“welfare”asopposedtothemoregeneric“assistancetothepoor.”Severalexperimentshaveshownthatthereismuchgreaterpublicsupportforexpanding“assistancetothepoor”thanforexpanding“welfare.” Weoftenwritetwoversionsofaquestionandaskhalfofthesurveysampleoneversionofthequestionandtheotherhalfthesecondversion.Thus,wesaywehavetwo forms ofthequestionnaire.Respondentsareassignedrandomlytoreceiveeitherform,sowecanassumethatthetwogroupsofrespondentsareessentiallyidentical.Onquestionswheretwoversionsareused,significantdifferencesintheanswersbetweenthetwoformstellusthatthedifferenceisaresultofthewaywewordedthetwoversions. Oneofthemostcommonformatsusedinsurveyquestionsisthe“agree-disagree”format.Inthistypeofquestion,respondentsareaskedwhethertheyagreeordisagreewithaparticularstatement.Researchhasshownthat,comparedwiththebettereducatedandbetterinformed,lesseducatedandlessinformedrespondentshaveagreatertendencytoagreewithsuchstatements.Thisissometimescalledan“acquiescencebias”(sincesomekindsofrespondentsaremorelikelytoacquiescetotheassertionthanareothers).Thisbehaviorisevenmorepronouncedwhenthere’saninterviewerpresent,ratherthanwhenthesurveyisself-administered.Abetterpracticeistoofferrespondentsachoicebetweenalternativestatements.APewResearchCenterexperimentwithoneofitsroutinelyaskedvaluesquestionsillustratesthedifferencethatquestionformatcanmake.Notonlydoestheforcedchoiceformatyieldaverydifferentresultoverallfromtheagree-disagreeformat,butthepatternofanswersbetweenrespondentswithmoreorlessformaleducationalsotendstobeverydifferent. Oneotherchallengeindevelopingquestionnairesiswhatiscalled“socialdesirabilitybias.”Peoplehaveanaturaltendencytowanttobeacceptedandliked,andthismayleadpeopletoprovideinaccurateanswerstoquestionsthatdealwithsensitivesubjects.Researchhasshownthatrespondentsunderstatealcoholanddruguse,taxevasionandracialbias.Theyalsomayoverstatechurchattendance,charitablecontributionsandthelikelihoodthattheywillvoteinanelection.Researchersattempttoaccountforthispotentialbiasincraftingquestionsaboutthesetopics.Forinstance,whenPewResearchCentersurveysaskaboutpastvotingbehavior,itisimportanttonotethatcircumstancesmayhavepreventedtherespondentfromvoting:“Inthe2012presidentialelectionbetweenBarackObamaandMittRomney,didthingscomeupthatkeptyoufromvoting,ordidyouhappentovote?”Thechoiceofresponseoptionscanalsomakeiteasierforpeopletobehonest.Forexample,aquestionaboutchurchattendancemightincludethreeofsixresponseoptionsthatindicateinfrequentattendance.Researchhasalsoshownthatsocialdesirabilitybiascanbegreaterwhenaninterviewerispresent(e.g.,telephoneandface-to-facesurveys)thanwhenrespondentscompletethesurveythemselves(e.g.,paperandwebsurveys). Lastly,becauseslightmodificationsinquestionwordingcanaffectresponses,identicalquestionwordingshouldbeusedwhentheintentionistocompareresultstothosefromearliersurveys.Similarly,becausequestionwordingandresponsescanvarybasedonthemodeusedtosurveyrespondents,researchersshouldcarefullyevaluatethelikelyeffectsontrendmeasurementsifadifferentsurveymodewillbeusedtoassesschangeinopinionovertime. Questionorder Oncethesurveyquestionsaredeveloped,particularattentionshouldbepaidtohowtheyareorderedinthequestionnaire.Surveyorsmustbeattentivetohowquestionsearlyinaquestionnairemayhaveunintendedeffectsonhowrespondentsanswersubsequentquestions.Researchershavedemonstratedthattheorderinwhichquestionsareaskedcaninfluencehowpeoplerespond;earlierquestionscanunintentionallyprovidecontextforthequestionsthatfollow(theseeffectsarecalled“ordereffects”). Onekindofordereffectcanbeseeninresponsestoopen-endedquestions.PewResearchCentersurveysgenerallyaskopen-endedquestionsaboutnationalproblems,opinionsaboutleadersandsimilartopicsnearthebeginningofthequestionnaire.Ifclosed-endedquestionsthatrelatetothetopicareplacedbeforetheopen-endedquestion,respondentsaremuchmorelikelytomentionconceptsorconsiderationsraisedinthoseearlierquestionswhenrespondingtotheopen-endedquestion. Forclosed-endedopinionquestions,therearetwomaintypesofordereffects:contrasteffects(wheretheorderresultsingreaterdifferencesinresponses),andassimilationeffects(whereresponsesaremoresimilarasaresultoftheirorder). AnexampleofacontrasteffectcanbeseeninaPewResearchCenterpollconductedinOctober2003,adozenyearsbeforesame-sexmarriagewaslegalizedintheU.S.Thatpollfoundthatpeopleweremorelikelytofavorallowinggaysandlesbianstoenterintolegalagreementsthatgivethemthesamerightsasmarriedcoupleswhenthisquestionwasaskedafteroneaboutwhethertheyfavoredoropposedallowinggaysandlesbianstomarry(45%favoredlegalagreementswhenaskedafterthemarriagequestion,but37%favoredlegalagreementswithouttheimmediateprecedingcontextofaquestionaboutsame-sexmarriage).Responsestothequestionaboutsame-sexmarriage,meanwhile,werenotsignificantlyaffectedbyitsplacementbeforeorafterthelegalagreementsquestion. AnotherexperimentembeddedinaDecember2008PewResearchCenterpollalsoresultedinacontrasteffect.Whenpeoplewereasked“Allinall,areyousatisfiedordissatisfiedwiththewaythingsaregoinginthiscountrytoday?”immediatelyafterhavingbeenasked“DoyouapproveordisapproveofthewayGeorgeW.Bushishandlinghisjobaspresident?”;88%saidtheyweredissatisfied,comparedwithonly78%withoutthecontextofthepriorquestion. Responsestopresidentialapprovalremainedrelativelyunchangedwhethernationalsatisfactionwasaskedbeforeorafterit.AsimilarfindingoccurredinDecember2004whenbothsatisfactionandpresidentialapprovalweremuchhigher(57%weredissatisfiedwhenBushapprovalwasaskedfirstvs.51%whengeneralsatisfactionwasaskedfirst). Severalstudiesalsohaveshownthataskingamorespecificquestionbeforeamoregeneralquestion(e.g.,askingabouthappinesswithone’smarriagebeforeaskingaboutone’soverallhappiness)canresultinacontrasteffect.Althoughsomeexceptionshavebeenfound,peopletendtoavoidredundancybyexcludingthemorespecificquestionfromthegeneralrating. Assimilationeffectsoccurwhenresponsestotwoquestionsaremoreconsistentorclosertogetherbecauseoftheirplacementinthequestionnaire.WefoundanexampleofanassimilationeffectinaPewResearchCenterpollconductedinNovember2008whenweaskedwhetherRepublicanleadersshouldworkwithObamaorstanduptohimonimportantissuesandwhetherDemocraticleadersshouldworkwithRepublicanleadersorstanduptothemonimportantissues.PeopleweremorelikelytosaythatRepublicanleadersshouldworkwithObamawhenthequestionwasprecededbytheoneaskingwhatDemocraticleadersshoulddoinworkingwithRepublicanleaders(81%vs.66%).However,whenpeoplewerefirstaskedaboutRepublicanleadersworkingwithObama,fewersaidthatDemocraticleadersshouldworkwithRepublicanleaders(71%vs.82%). Theorderquestionsareaskedisofparticularimportancewhentrackingtrendsovertime.Asaresult,careshouldbetakentoensurethatthecontextissimilareachtimeaquestionisasked.Modifyingthecontextofthequestioncouldcallintoquestionanyobservedchangesovertime(see measuringchangeovertime formoreinformation). Aquestionnaire,likeaconversation,shouldbegroupedbytopicandunfoldinalogicalorder.Itisoftenhelpfultobeginthesurveywithsimplequestionsthatrespondentswillfindinterestingandengaging.Throughoutthesurvey,aneffortshouldbemadetokeepthesurveyinterestingandnotoverburdenrespondentswithseveraldifficultquestionsrightafteroneanother.Demographicquestionssuchasincome,educationorageshouldnotbeaskednearthebeginningofasurveyunlesstheyareneededtodetermineeligibilityforthesurveyorforroutingrespondentsthroughparticularsectionsofthequestionnaire.Eventhen,itisbesttoprecedesuchitemswithmoreinterestingandengagingquestions.OnevirtueofsurveypanelsliketheATPisthatdemographicquestionsusuallyonlyneedtobeaskedonceayear,notineachsurvey. U.S.Surveys OTHERRESEARCHMETHODS 1615LSt.NW,Suite800Washington,DC20036USA (+1)202-419-4300|Main (+1)202-857-8562|Fax (+1)202-419-4372|MediaInquiries Politics&PolicyInternationalAffairsImmigration&MigrationRace&EthnicityReligionGenerations&AgeGender&LGBTFamily&RelationshipsEconomy&WorkScienceInternet&TechnologyNewsHabits&MediaMethodologicalResearchFulltopiclist EmailNewsletters Facebook Twitter Tumblr YouTube RSS AboutPewResearchCenterPewResearchCenterisanonpartisanfacttankthatinformsthepublicabouttheissues,attitudesandtrendsshapingtheworld.Itconductspublicopinionpolling,demographicresearch,mediacontentanalysisandotherempiricalsocialscienceresearch.PewResearchCenterdoesnottakepolicypositions.ItisasubsidiaryofThePewCharitableTrusts. Copyright2022PewResearchCenter About Terms&Conditions PrivacyPolicy Reprints,Permissions&UsePolicy Feedback Careers