internet usage effect on educational attainment: evidence...

Post on 03-Oct-2020

8 Views

Category:

Documents

0 Downloads

Preview:

Click to see full reader

TRANSCRIPT

1

Internet Usage Effect on Educational Attainment: Evidence of Benefits

JoeLeeJuly2017

IntroductionSincetheinternetbecameawidelyadoptedplatformforinformationexchange,therehasbeenavarietyofresearchintotheeffecttheinternethasoneducation.SomeresearchfindsincreasedsuccessratesforminoritiesinK-12whenprovidedcomputers,whileotherresearchfocusesontheroleofmobileplatformsandsocialmedia.1,2,3Educationhasseenlargechangesduetotheinternet,includingmodificationstocurriculumanddifferencesinthewayeducationisdelivered.Thispaperprovidesevidenceofapositiverelationshipbetweenthehoursspentontheinternetperweekandeducationalattainment.Thisanalysisshowsthattherelationshipisrobusttospecificationchangesanddisplaysstabilityinbothmagnitudeanddirectionalityofthestatisticalrelationship.Controllingforoutsideinfluencesinlinewithpreviousresearch,thereisevidencethattheinternetcanbeatooltoimproveeducationaloutcomes.Expandingthescopeofeducationalattainmentresearchwillgivepolicy-makersabetterunderstandingofhowtoincreaseeconomicopportunityforall.

1Fairlie,W,Robert.“Academicachievement,technologyandrace:Experimentalevidence”EconomicsofEducationReview31,(2012):pg663-679.https://people.ucsc.edu/~rfairlie/papers/published/eer%202012%20-%20minority%20computers.pdf2Fairlie,W,Robert,London,A,Rebecca.“TheEffectsofHomeComputersonEducationalOutcomes:EvidencefromaFieldExperimentwithCommunityCollegeStudents”TheEconomicJournal122(2012),http://escholarship.org/uc/item/4v87w4693Bulman,George,Fairlie,W,Robert.“TechnologyandEducation:Computers,Software,andtheInternet”NBERWorkingPaperSeries,(2016)no22237,http://www.nber.org/papers/w22237.pdf

2

FIGURE1.HighestDegreeAttained

Source:U.S.GeneralSocialSurvey,2000-2016,highestdegreeattained,18-89 Recently,pastacademicworkhasbeenusedtoqualifyproposedpoliciesseekingtoincreaseeducationalattainmentinpursuitofexpandingeconomicprosperity.In1974,JacobMincer’sresearchproducedacorrelationthatillustratedthateducationandexperiencehaveapositiveeffectonincome.4,5Overtime,changesintheeconomycreatedaneedforamoreskilledworkforce.Thisshiftispredicatedontheexpandingroleoftechnologyandtheever-increasingneedforspecialization.Acquiringskillsthrougheducationcanhelppeopleadapttothesechanges.Figure1showsthatbetween2000and2016,27.1percentofthepopulationhadattainedabachelor’sdegreeorhigher.Theinternethasfundamentallytransformedhowpeopleexperienceculturalandeconomicchanges.Asanewmediumforindividualopinionsandinformationexchange,theinternethaschangedthewaywecommunicatebyincreasingthespeedandreachofideas.

Internetusagehasaffectedhoweconomicactivityconcentratesinclusters,andindustryinteractionswithintheseclusters.6,7Astheshiftfrommanufacturingoccupationstoknowledge-basedoccupationshaschangedtheemploymentlandscapeinmetroclusters,urbanconcentrationsofhighskilledservicejobshavegrown.Thecontinuedriseofscienceandtechnologyhasdevelopednewfieldsinhealthcare,

4Mincer,Jacob,andSolomonPolachek."FamilyInvestmentsinHumanCapital:EarningsofWomen."JournalofPoliticalEconomy82,no.2(1974):76-108.http://www.jstor.org/stable/1829993.5Mincer,Jacob.“SchoolingandEarnings”NBER(1974):http://www.nber.org/chapters/c1765.pdf6Leamer,E,Leamer,Storper,Michael.“TheEconomicGeographyoftheInternetAge”JournalofInternationalBusinessstudies32,no.4(2001):641-665,http://www.jstor.org/stable/pdf/3069470.pdf7Forman,Chris,Goldfarb,AVI,Greenstein,Shane.“GeographiclocationandthediffusionofInternettechnology”ElectronicConsumerResearchandApplications,4(2005)pg1-13,http://s3.amazonaws.com/academia.edu.documents/43895172/ecra.pdf?AWSAccessKeyId=AKIAIWOWYYGZ2Y53UL3A&Expires=1490915087&Signature=iG8J%2F2aQ0LM3yngGQRMs4oPuBFI%3D&response-content-disposition=inline%3B%20filename%3DGeographic_location_and_the_diffusion_of.pdf

0

10

20

30

40

50

60

NoDegree HighSchool Associate'sDegree Bachelor'sDegree GraduateDegree

ChartTitlePercent

3

computerscience,andengineering.ThesefieldscreateneweconomicpowercentersandredefinethecompositionofskillsneededtoincreaseeconomicopportunitythroughouttheU.S.economy.Througheducationwecanimpartskillstoanexpandingworkforce.

Asasociety,wehavecreatednewopportunitiesastechnologyadvances,andtheinternetcontinuestoredefinehowtheU.S.economyfunctions.Theriseofsocialmedia,cordcutting,andvideogamesareallobservablechanges.Thoughwehavecreatedmanyofnewwaystospendfreetime,theinternethasdonemorethanjustgiveusnewwaystoconsumecontent.8,9Theinternethasreducedthecostoftransmittinginformationbyloweringthefrictionofaccesstothatinformation.10Newcapabilities,suchasaninstitution’scapacityforpostinglecturesonlinetowebsitesofferingonlinecertificates,isevidencethataccesstoknowledgehasincreased.Beyondinstitutionsandothereducationaloutlets,peoplecanlearnmusicalinstrumentsorlearntocodethroughonlinetutorials.Theinternethasgivenaspacetothosepeoplewhowishtoshareinformationonanytopic,bothpositiveandnegative.

Thegenesisoftheinternetswiftlyandpermanentlyusheredinanewera.However,therearesectionsoftheU.S.thatstillhavelimitedinternetaccess.Previousresearchexaminingtheeffectofaccesstotheinternetinaresidenceshowsapositiveeffectonstudentschoosingtopursehighereducation.11

However,therearenotabledifferencesininternetaccessandspeedwhencomparingurbanpopulationcentersandruralareas.Figure2shows22.8percentofU.S.householdsdonothavesometypeofinternetsubscription.Thisincludesdial-up,DSL,broadband,andmobileinternetservices.Cableinternetsubscriptionsrepresent41percentofU.S.households,whilethesecondlargestsubscriptiontypeismobilebroadband(38.1percent).Inmajorurbancenters12percentofhouseholdsdon’thaveaccesstotheinternetcomparedwith20.5percentofruralhouseholds.Thedifferentialofinternetspeedandaccessbetweenurbanandruralareasreducethebeneficialimpactofinternetusageoneducationalattainmentofthepopulation.

8Junco,Reynol.“TherelationshipbetweenfrequencyofFacebookuse,participationinFacebookactivities,andstudentengagement”Computers&Education58,(2011):162-171.http://citeseerx.ist.psu.edu/viewdoc/download?doi=10.1.1.716.205&rep=rep1&type=pdf9Aguiar,etal.,“LeisureLuxuriesandthelaborsupplyofYoungMen”,NBERworkingpaperseries,no23552,(2017):http://www.nber.org/papers/w23552.pdf10Torberg,Falch,Lujala,Päivi,Strøm,Bajarne.“Geographicalconstraintsandeducationalattainment”RegionalScienceandUrbanEconomics43,no.1(2012):164-176.http://www.sv.ntnu.no/iso/torberg.falch/Articles/Geographical%20constraints.pdf11Bowden,P,Mark,Doughney,James.“Theimportanceofculturalandeconomicinfluencesbehindthedecisiontoattendhighereducation”TheJournalofSoci-Economics41,(2012):pg95-103.http://vuir.vu.edu.au/9370/1/Cultural%20and%20econ%20influences%20and%20decision%20to%20attend%20he.pdf

4

FIGURE2.Urbanvs.RuralInternetSubscriptionsofU.S.Households

Source:IPUMSAmericanCommunitySurvey2015Urbanvs.RuralDivideinAmericanInternetAccess TheU.S.hasincreasingconcentrationsofeconomicactivityinurbanareas,makinglargermetropolitanareasmajoreconomicforces.Thisconcentrationofeconomicactivitygeneratesinnovationandcandrivegrowth,whichincreasinglyshiftsresourcestowardtheseclusters.SlowinternetreducestheefficiencyofknowledgetransferthroughinternetusageandismorecommoninruralareasoftheU.S.thaninurbanareas.

Onlineeducationalresourcescanhelpreducethecostofeducationthroughminimizingboarding,travel,materials,allocationoftime,andfacilityexpendituresduetoonlineeducationalresources,howevertheeffectivenessandqualityoftheseprogramsarestillsubjecttofurtherinquiry.Somehighschool,college,andgraduateprogramsarenowconductedentirelyonline,openingaccesstoeducationinanewway.Figure3showsthatabout78percentofhouseholdsinruralAmericahaveaccesstohigh-speedinternetcapableofsupportingonlineclassrooms.Broadeningaccessalsoallowspeopletoovercomebothgeographicbarriersandtimeconstraintsthroughincreasedflexibilityofscheduling.TheabilityforsomeoneinruralAmericatoobtainadegreefromamajorinstitutionontheothersideofthenationprovidesanavenueforeconomicopportunitynotpreviouslyavailable.However,thebenefitsofthisneweducationaccessarelosttoover1in10urbanhouseholdsand1in5ruralhouseholdswithoutaninternetconnection(seeFigure3).

0

10

20

30

40

50

60

NoAccess Dial-Up Other Satellite DSL MobileBroadband

CableInternet FiberOptic

Percent

Urban Rural

Low-Speed

High-Speed

5

FIGURE3.HouseholdswithBroadbandInternetSubscriptionsbyGeographicCategory

Source:CensusAmericanCommunitySurvey2015Astudentcanquicklyaccessprogramrequirementsandcurriculuminformationontheinternet.Accesstoacollege’sadmissionrequirementsallowspotentialapplicantstoassessprerequisiteconditionstodecidewhatcoursestheyshouldcompleteorwhatSAT/ACTscorestheyneedtomaximizetheirlikelihoodofacceptance.Thisreductionoftimeandresourcesrequiredtoobtaininformationisprovidedbytheeaseofaccesstoinformation.Evenduringthecollegeexperience,quickanduser-friendlycommunicationplatformsenablepeerinteractionandteachersupport.12Throughincreasedpeerinteractionandaccesstoteachers,theinternethasincreasedthespeedofcommunicationandthespreadofknowledge.

Data and ResultsGeneralSocialSurvey ThemainresultsofthispaperusetheGSSNationalDataProgramfortheSocialSciences,whichisapooledcross-sectionoftheU.S.population.ABayesianOrderedProbitModel(adiscretechoicemodelfortheprobabilityofpredictingthechoicewheretheorderofoutcomesisimportant)isusedforthisanalysisexaminingtherelationshipbetweeneducationalattainmentandinternetusage.Thesampleperiodfrom2000to2016encapsulatestheriseoftheinternetintoitscurrentform.ItcoversmajoreventsfromthecollapseoftheDotComBubblethroughtheGreatRecessionanditssubsequentrecoveryyears.Controllingforthreemajorgroupsofvariables,themodelshowsapositiverelationshipbetweenthehoursspentontheinternetperweekandthehighestdegreeattained.Thegraduate

12Gikas,Joanne,grant,M,Michael.“Mobilecomputingdevicesinhighereducation:Studentperspectivesonlearningwithcellphones,smartphones&socialmedia”InternetandHigherEducation19,(2013):18-26http://s3.amazonaws.com/academia.edu.documents/38919269/gikas_grant_mobile_devices.pdf?AWSAccessKeyId=AKIAIWOWYYGZ2Y53UL3A&Expires=1490891993&Signature=Zze0kmHjwaeqirWVH0Rek2NrayE%3D&response-content-disposition=inline%3B%20filename%3DMobile_Computing_Devices_in_Higher_Educa.pdf

21.6%30.3% 36.3%

0102030405060708090

100

MetropolitanStatisticalArea MicropolitanStatisticalArea OutsidePopulationCenter

PopulationCentershigh- speedinternetsubscribers

WithBroadband WithoutInternetorDial-Up

Percent

6

degreecategoryincludesanyhighereducationdegreerequiringabachelor’sdegreeforprogramentry.Familyinfluences,demographics,time,andregionaleffectshavebeencontrolledforwiththeintenttoimprovethisanalysisbyaccountingforotherconfoundinginfluencesthatmightwashouttheeffectofinternetusage.ThemainfindingsofthispaperarereportedincolumnthreeofTable2. IndividualandHouseholdCharacteristics Thefirstsetofcontrolvariablesarerelatedtohouseholdcharacteristics.Thesevariablescaptureimportantexternalinfluencesonanindividual.Theanalysiscontrolsforbothparents’highestdegreeattainedtoexplainhoweducationalattainmentisvalued.13,14Bothparenteducationvariablesarehighlysignificantandshowapositiverelationshipwitheducationalattainment.Thesevariablesarehighlycorrelatedbetweengenerationsandinfluencetheexpectedoutcomeofeducationattainmentinthismodel.15

Realindividualincomealsoshowsapositiveandhighlysignificantimpactoneducationaloutcomes.16Higherpersonalincomeincreasesaccesstotechnologyaswellassomeone’sabilitytoinvestineducation.Whenlookingatyoungeragerangesitisimportanttorememberthatpeoplewillgenerallyhavelowerlevelsofincome.Thebenefitsfromprivateeducationandfinancialsupportforhighereducationareincreasinglyinfluential.Resourcesdevotedtohighereducationincreasesthelikelihoodofgraduationbyreducingfinancialconstraints.Thisresultissupportedbypreviousresearchonthepositiverelationshipbetweenparenteducationalattainment,andfinancialresourcesavailableforeducation.17

Inthisvein,increasesinhouseholdsizehaveahighlysignificantnegativeeffectoneducationaloutcomes.Asthenumberofpeopleinahouseholdincreases,fewerresourcesareavailabletoeachpersonforeducationinvestment.Theopportunitycostofanindividualisincreasedasageincreasesbecausepeoplewillhavehigherincomesbutalsotheneedtomaintainanincomeashouseholdsizesincreasefinancialconstraints. 13Haveman,Robert,Wolfe,Barbara.“TheDeterminantsOfChildern’sAttainments:AReviewofMethodsandFindings”JournalofEconomicLiterature33,no.4(1995):1829-1878https://www.researchgate.net/profile/Barbara_Wolfe/publication/44819892_The_Determinants_of_Children%27s_Attainments_A_Review_of_Methods_and_Findings/links/0c96053bc08dcac165000000/The-Determinants-of-Childrens-Attainments-A-Review-of-Methods-and-Findings.pdf14Hung,Jin.“Intergenerationaltransmissionofeducationalattainment:Theroleofhouseholdassets”EconomicsofEducationReview33,(2013):112-123.https://www.researchgate.net/profile/Jin_Huang23/publication/257106204_Intergenerational_transmission_of_educational_attainment_The_role_of_household_assets/links/550d8eb40cf2ac2905a7e724.pdf15Taubman,Paul."RoleofParentalIncomeinEducationalAttainment."TheAmericanEconomicReview79,no.2(1989):57-61.http://www.jstor.org/stable/1827730.16Bowden,P,Mark,Doughney,James.“Theimportanceofculturalandeconomicinfluencesbehindthedecisiontoattendhighereducation”TheJournalofSoci-Economics41,(2012):pg95-103.http://vuir.vu.edu.au/9370/1/Cultural%20and%20econ%20influences%20and%20decision%20to%20attend%20he.pdf17Hung,Jin.“Intergenerationaltransmissionofeducationalattainment:Theroleofhouseholdassets”EconomicsofEducationReview33,(2013):112-123.https://www.researchgate.net/profile/Jin_Huang23/publication/257106204_Intergenerational_transmission_of_educational_attainment_The_role_of_household_assets/links/550d8eb40cf2ac2905a7e724.pdf

7

DemographicCharacteristics Thesetofvariablesrepresentingdemographicsareintendedtoisolateculturalandgenerationalinfluencesthataffecteducationalattainment.Thisanalysiscontrolsforageandage-squared.Ageishighlysignificant,servingseveralpurposesinthisanalysis.Age-squaredisnotsignificantwhichisunexpected(seeappendix).Thefirstuseofageistocontrolfortheincreasedlikelihoodofhighereducationalattainmentduetoincreasesinage.Thesecondreasonistoaddressconcernsofdirectionalcausalityofhighereducationleadingtohigherinternetuse.Ageiscorrelatedtobothexperienceandseniorityofoccupationsincombinationwithpersonalincome,whichishighlycorrelatedtooccupationalchoice.Theuseofthesetwovariablesinintendedtomitigatetheimpactofmulti-directionalcausalityofinternetusageandeducationalattainment.Thethirdistoisolatetheeffectofpeopleofdifferentages’tendencytousetheinternet.FIGURE4.Agevs.HoursonInternetPerWeek

Source:GeneralSocialSurvey2000-2016 Ata99percentconfidencelevel,femaleshaveagreaterlikelihoodofobtainingahigherdegree.Thisisreflectedbythelargerfemaleenrollmentnumbersinhighereducationinstitutions.18,19Theracevariableincludesallnon-whiteethnicitiesandishighlysignificantwithapositiverelationshiptohighereducationalattainment.Whileencompassingallminoritiesisnotagreat indicatorandcouldbe reflectingthatthesampleis19percentnonwhiteandfamilyincomeispositivelyskewed.Forthisdataset,whiteindividualshaveahighermeanLn(personalincome)whichisstatisticallysignificantatthe99percentconfidencelevelseeTable3.

18NationalCenterforEducationStatistics,Table318.30(2013to2016)https://nces.ed.gov/programs/digest/d13/tables/dt13_318.30.asp19NationalCenterforEducationStatistics,Table310(2011)https://nces.ed.gov/programs/digest/d12/tables/dt12_310.asp

0

20

40

60

80

100

120

140

17 27 37 47 57 67 77 87AgeinYears

AgevsinternetusageHours PerWeek

8

Thereasonfortheaggregationisthatallotherminoritiesotherthanblacksarealreadyaggregatedinthisdataset.Inordertoaddressinfluenceofculturaldifferencesineducationalattainment,ethnicity,andreligionvariablesareincluded.

Religionhasbeenbrokenoutintobinaryvariablesthatarenotaggregatesofmultiplereligiousdoctrines.ThesevariablesarenormalizedagainstProtestantsbecausetheyarethelargestreligiousgroupinthesample.Judaism,BuddhismandHinduismareallhighlysignificantandhaveapositiverelationshipwitheducationalattainment.Thesedemographicvariablesareincludedtohelpcontrolforculturalinfluencesnotaccountedforbythenonwhitevariable.20RegionalandTimeEffects Thethirdgroupofvariablesusedascontrolsareregionandtimeeffects.Theregionsarerelatedtozipcoderegionswhicharenormalizedagainstthenortheast.Allregionaldummyvariablesarehighlysignificantandnegativelysigned,whichindicatesthenortheasthasanadvantageforincreasededucationalattainment.Thisoutcomemaybeduetoanumberoffactorssuchaspopulationdensity,concentrationofpercapitaGDP,employmentopportunitiesthatdrawinpeoplewithhigherlevelsofeducation,largenumbersofhighereducationalinstitutionsprovidinglocalhumancapitaltolabormarketsduetoin-statetuitionloweringcosts,ormoreopportunityforplacementbecauseoftheamountofschools.TheregionvariablesareincludedtocapturethevariationofeducationalopportunityacrosstheU.S.

Therearemacroeconomiceffectstobecontrolledforaswell,asthesampledatacoverstwomajorrecessions.Startingin2000,Ihaveincludedallyearsbut2008duetoalackofinternetusagedataforthatyear.Alltimevariablesaresignificantandnegativelysigned.Thetimevariablesareincludedbecauseperiodsofrecessioncauseddifferencesintheroleoftheinternetintheeconomy.Thestructureofthedataallowsfornewpeopletoenterthesamplewhichincludesyoungerpeoplewhohaveearliercontactwiththeinternet.Theadditionofyearvariablesallowsfortheseeffectstobeaccountedfor.InternetUsage Thecontrolsinthismodelarebeingusedtoisolatetheeffectthathoursperweekonlinehasoneducationalattainmentoutcomes.Thefamilyanddemographiccharacteristicsareusedtoconditionforexternalinfluenceswhiletimeandregionaleffectsareusedtoconditionformoreabstractfactors.Thisanalysisaddressesconcernsaboutpossiblereversecausalityofinternetusageandeducationalattainmentbycontrollingforageandincome.Incomeishighlycorrelatedtooccupationandhelpscontrolfordifferencesinexposuretotechnologyandtheinternet.Thisissueofdirectionalcausalityisacommonproblemwithalleconometricsbuthasbeenaddressedasmuchaspossiblewiththedataavailableforthisanalysis.Thenecessaryrobustnesschecksoftheseinfluencesconfirmthattheeffectofaverageweeklyinternetusageissignificantanddemonstratesalargepositiveeffectontheprobability

20Bowden,P,Mark,Doughney,James.“Theimportanceofculturalandeconomicinfluencesbehindthedecisiontoattendhighereducation”TheJournalofSoci-Economics41,(2012):pg95-103.http://vuir.vu.edu.au/9370/1/Cultural%20and%20econ%20influences%20and%20decision%20to%20attend%20he.pdf

9

ofincreasededucationalattainment,onaverage.Internetusageshowsaremarkablystableeffectwitheachadditionalspecificationofthefullmodel.Theincreaseofonehourweeklyinternetuseimpliesoveraonepercentincreaseintheprobabilityofattainingbachelor’sdegreesandgraduatedegrees,onaverage.Sincethemeasureofinternetusageisspreadovertheweek,anincreaseofanhouronaverageisnotahugetimeinvestment.Thisisevidencethatinternetusagehasahighlysignificantpositiveeffectoneducationalattainment. FIGURE5.EffectofanHourIncreaseofAverageWeeklyInternetUseonEducationalAttainmentOutcome

Source:MilkenInstitute

ConclusionThisanalysisshowsthatthetimespentontheinternethasapositiveeffectoneducationalattainmentconditionalonfamily,demographic,regionalandtimeeffects.Ashouseholdsincreaseinsize,theprobabilityofapersonachievingalowerlevelofeducationalsoincreases.Themodelprovidesfurtherevidencethatparentaleducationalattainmentishighlycorrelatedwithanindividual’seducationalattainmentlevel.Thisanalysisprovidessomeevidencethatmoretimespentontheinternetcanmitigatenegativeexternaleffectsforindividualsinhouseholdswithlowerparentaleducationlevels,lowerincome,orlargefamilies.Inthefuture,beingabletodifferentiatehoursspentontheinternetforentertainmentratherthanproductivitywouldenhancethisanalysis.Thereisevidencethatforpeoplebetweentheagesof18-22,internetusagedoesnotaffecteducationalattainment.However,thisagerangeismuchlesslikelytohaveanydegreebeyondhighschoolwhichtruncateseducationalattainmentoutcomeswhichprecludes

0

10

20

30

40

50

60

OnAverage OnHourIncreasefromAverage

ChartTitle

Nodegree HighSchoolOnly Associate'sDegreeOnly Bachelor'sDegreeOnly GraduateDegreeOnly

Percent

10

theusefulnessofthissubsetofages.Thepopulationshifttourbanclustersgeneratesincreasinglydisproportionaleconomicactivitywhichdrivesupskillrequirementsthroughcompetitioninthelabormarket.Thisnotonlyputsmoreemphasisonbeingabletoperformatahigherlevel,butalsorequiresindividualstosignalahigherlevelofability.Highereducationcanbeawaytodoboth.

Inthecontextofequitableaccess,spreadinghighspeedinternetaccesstobothurbanandruralresidentscanhelpincreasethechancesforimprovingeconomicopportunity.Giventheimportanceoftheinternetinmodernlife,equalaccesstoinformationandexpandingaccesswillbeevermoreimportanttoencouragehighereducation.

11

Appendix, Tables, Data, and Methodology TABLE1.DescriptiveStatisticsFullGSSdata

StandardVariables Observations Mean DeviationsAge 62,245 45.99 17.51HighestDegree 62,293 1.35 1.17Father’sdegree 46,985 0.89 1.17Mother’sDegree 54,451 0.84 0.99Householdsize 62,460 2.65 1.51Female 62,466 0.56 0.50Non-white 62,466 0.19 0.40Ln(FamilyIncome) 56,142 10.33 0.99Ln(RespondentIncome) 36,524 9.51 1.11Ln(MeanInternetHoursPerWeek)

11,119 1.52 1.19

Protestant 59,044 0.61 0.49Catholic 59,044 0.29 0.45Judaism 59,044 0.02 0.14Atheist 59,044 0.05 0.23Buddhist 59,044 0.002 0.04Hindu 59,044 0.0017 0.04Muslim 59,044 0.0024 0.05Orthodox-Christian 59,044 0.0021 0.05Christian 59,044 0.01 0.08OtherReligions 59,044 0.01 0.11NewEngland 62,466 0.05 0.21Mid-AtlanticU.S. 62,466 0.14 0.35EastNorthCentralU.S. 62,466 0.18 0.39WestNorthCentralU.S. 62,466 0.07 0.26South-AtlanticU.S. 62,466 0.19 0.39EastSouthCentralU.S. 62,466 0.07 0.25WestNorthCentralU.S 62,466 0.09 0.29Mountain 62,466 0.06 0.24Pacific 62,466 0.13 0.34 Min Max Year 1972 2016

Thisdataisthefulldatasetwithoutremovalofanyobservations

12

TABLE2.RegressionResultsofBayesianOrderedProbit21,22

VariablesBase Full Full Full

Degree Degree Degree Degree Ln(InternetUsage) 0.08*** 0.09*** 0.09*** 0.10*** (0.01) (0.01) (0.01) (0.01)Mother’sDegree 0.18*** 0.18*** 0.18*** 0.19*** (0.02) (0.02) (0.02) (0.02)Father’sDegree 0.21*** 0.22*** 0.21*** 0.23*** (0.01) (0.01) (0.01) (0.01)Ln(realpersonalincome)2000USD 0.29*** 0.28*** (0.02) (0.02) Ln(Familyincomeperearner) 0.47*** (0.02) Ln(realpersonalincome)1997USD 0.32*** (0.03)HouseholdSize -0.04*** -0.04*** -0.03*** -0.05*** (0.01) (0.01) (0.01) (0.01)Female 0.22*** 0.22 0.17 0.12*** (0.03) (0.01) (0.03) (0.03)Non-white -0.01 0.01 0.06 -0.01 (0.04) (0.04) (0.04) (0.04)Age 0.03*** 0.02*** 0.04*** 0.05*** (0.005) (0.01) (0.01) (0.01)Age2 -0.0001 -0.0001 -0.0003*** -0.0004*** (0.00001) (0.00001) (0.00009) (0.0001)Catholic 0.06* 0.05 0.05 0.07** (0.03) (0.03) (0.03) (0.03)Judaism 0.64*** 0.58*** 0.58*** 0.65*** (0.10) (0.10) (0.10) (0.10)Atheist -0.04 0.05 0.04 0.04 (0.06) (0.06) (0.06) (0.06)Buddhist 0.71*** 0.69*** 0.70*** 0.76*** (0.17) (0.18) (0.18) (0.18)Hindu 1.11*** 1.05*** 0.83** 1.07*** (0.21) (0.21) (0.21) (0.21)Muslim 0.51** 0.48** 0.51** 0.46** (0.17) (0.21) (0.227) (0.21)Otherreligion 0.04 0.03 0.09 0.06 (0.15) (0.16) (0.16) (0.16)Orthodox-Christian 0.29 0.17 0.20 0.16 (0.21) (0.22) (0.22) (0.22)

21Imai,Kosuke,GaryKing,andOliviaLau.2008.“TowardACommonFrameworkforStatisticalAnalysisandDevelopment.”JournalofComputationalandGraphicalStatistics,Vol.17,No.4(December),pp.892-913,http://j.mp/msE15c.22Choirat,Christine;ChristopherGandrud,JamesHonaker;KosukeImai;GaryKing;OliviaLau.2017.Zelig:Everyone'sStatisticalSoftware,Version5.0-15,URL:http://ZeligProject.org.

13

VariablesCont.Base Full Full Full

Degree Degree Degree Degree Christian -0.20 -0.13 -0.15 -0.11 (0.13) (0.13) (0.13) (0.13) (0.08) (0.08) (0.08)Mid-Atlantic -0.09 -0.09 -0.09 (0.08) (0.08) (0.08)EastNorthCentral -0.24*** -0.23*** -0.23*** (0.07) (0.07) (0.07)WestNorthCentral -0.25*** -0.23*** -0.27*** (0.08) (0.08) (0.09)South-Atlantic -0.19** -0.17** -0.19** (0.08) (0.08) (0.08)EastSouthCentral -0.20** -0.18* -0.22*** (0.09) (0.09) (0.09)WestSouthCentral -0.39*** -0.37*** -0.38*** (0.08) (0.08) (0.08)Mountain -0.19** -0.19** -0.22*** (0.08) (0.08) (0.08)Pacific -0.28*** -0.32*** -0.28*** (0.08) (0.08) (0.08)2002 -0.18*** -0.15*** -0.16*** (0.06) (0.06) (0.06)2004 -0.12*** -0.10 -0.11 (0.06) (0.06) (0.06)2006 -0.12*** -0.12** -0.14** (0.06) (0.06) (0.06)2010 -0.14*** -0.12 -0.19*** (0.07) (0.07) (0.07)2012 -0.26*** -0.22*** -0.30*** (0.07) (0.07) (0.07)2014 -0.31*** -0.27*** -0.37*** (0.07) (0.07) (0.06)2016 -0.28 -0.26*** -0.35*** (0.06) (0.07) (0.06)Constant -2.12*** -1.70*** -4.03*** -0.37** (0.18) (0.20) (0.26) (0.18)Gamma2 2.10*** 2.13 2.15*** 2.10*** (0.04) (0.03) (0.04) (0.04)Gamma3 2.39*** 2.42*** 2.45*** 2.39*** (0.04) (0.03) (0.04) (0.04)Gamma4 3.42*** 3.45*** 2.45*** 3.38*** (0.05) (0.03) (0.04) (0.04)Observations 5,715 5,715 7,034 5,688

Standarderrorsinparentheses***p<0.01,**p<0.05,*p<0.1

14

TABLE3.T-TestsGeneralSocialSurvey2000-2016

Two-SampleT-TestofEqualVariancebyNon-White

Variables T-score P-Value Ln(RealRespondentIncome) 17.43 0.00

DataCleaning Foranyestimatedparameterstohavevalidityforstatisticalinference,therearesomeassumptionsthatneedtobemet.Tomaintaintheseassumptionsandachievearobustanalysis,adjustmentsweremadetothedataset.TheworkingsubsetoftheGSSiscreatedbyfirstsub-settingtheyearswhereinternetusagewasprevalent,2000to2016except2008wherethereisnodata.OutlierswerethenidentifiedbyusingDffitsvaluestoconformtotheassumptionofasymptoticnormality.Thebasemodelspecificationwithonlythenaturallogstakenofinternetusageandfamilyincome,notincludingage-squared,wasusedinthisestimation.Theregionalortimebinaryvariablesinthismodelarekeptintheirrawcollapsedform.

Twoobservationswereremovedduetorealityconstraints.Internethoursperweekhastwoobservationsthatareunreasonable,giventhatthereare168hoursinaweek.Observationsrecordinganaverageof168hoursorgreaterperweekarenotbereasonableandareremoved.Binaryvariableswerecreatedforalluseableyearsandregions.Thisisdonebecausethereisnoordinalinterpretationofeitherofthesevariablesandbinaryvariablesallowtheestimatedparameterstohaveaninterpretableoutcome.Thereligionvariableswerebrokendownintobinaryvariablesforallrecordedreligionsthatarenotaggregates.Yearsarenormalizedby2000andthereligionsarenormalizedbyProtestants.Duetothelackoforderingandthesubjectivityofreligionandusefulnessofseparatingoutculturalinfluence,binaryvariablesaremoreinformative.Separatingoutyearsfollowsthesamelogicinthattheaggregationofyearsintoasingleparametermaywashoutmacroeconomiceffectsandreducethespecification’sabilitytobeinterpreted.Regionsarenormalizedagainstthenortheast.Familyincomeisinreal2000dollarsandistransformedwithanaturallogarithmforbetterfunctionality.Theaveragehoursspentperweekontheinternetishighlypositivelyskewedandiscorrectedforbyusinganaturallogarithmtransformation.Thesechangesinthemodel’sfunctionalformareintendedtoconformtotheassumptionofconditionalmeanzeroofmaximumlikelihoodestimation(MLE).Allofthechangesinthefunctionalformarealsointendedtoincreasethepoweroftheparameterestimationforinterpretation. ModelTesting AnOrderedProbitModelwaschosenbecauseofthediscretenatureofeducationaloutcomesaswellastheexistenceofaclearordinalcomponentofeducationalcredentials.ThefullmodelpassesfrequentistspecificationtestsofMLEassumptions.Themodellikelihoodratiochi-square(χ2)statisticof2153.63andspecificationlinktestwithlinearpredictedvalue(ŷ)andlinearpredictedvaluesquared(ŷ2)

displayingsignificanceandnon-significance,respectively.Nosignsofmulticollinearitywhereobserved

15

withallvarianceinflationfactor(VIF)valuesundertwowhenusingnobrokenoutbinaryvariables.Whenbreakingoutregion,year,andreligionsomevariableshaveVIFincreasesbutarealllessthanfive.Theoneexceptionisageandage-squared,whichisexpected.Withoutthesquaredtermtherearenosignsofmulticollinearity.Theprovidedsetofsurveyweightscorrectsforanysamplingissuesthatmightoccur.Therewerenomajorchangesinanyofthefrequentistestimationsusinganyoftheprovidedsampleweights.Largeoutlierswhereremoved(seedatacleaningsection).ThemodeldoesfitwithinthestandardMLEassumptions,butaBayesianOrderedProbitwaschosentoincreasetheprecisionoftheestimationhavingestablishedaccuracy.UsingGewekeandRaftery-Lewisdiagnostictestsforconvergence,Ifoundthatthevariablesarestationaryandhaveweakautocorrelation.23TheacceptancerateoftheMarkovchainMonteCarlo(MCMC)iterationsaverageis0.5percentandissupportedbytestsofMCMCconvergencebeingclosetooneexceptclassificationcutoffpoints.Anormalpriorwasusedtoreflecttheunderlyingdistributionofyearsofeducation,forwhichdegreeattainmentisanalternativefunctionalform.

Startingwiththebasemodelwithonlyfamilydynamicsandexpandingtothefullmodelshowsthenaturallogofaveragehoursperweekspentontheinternetneverlostsignificanceandmaintainsdirectionalityandmagnitude.Thisisconfirmedbyarelativelystablet-statisticwithincreasesinthecoefficient’smagnitude.Overall,thesensitivityanalysisdemonstratesthatthenaturallogarithmofaveragehoursspentontheinternetisrobustinitspositiveandlargeeffectontheprobabilityofachievinghighereducationalattainmentoutcomes.Moreimportantly,inthebasemodelwithoutyearandregionalcontrols,internetusageishighlysignificantandremainshighlysignificantwhenaddinginthosecontrols.Internetusageremainsstableinthisspecificationthroughoutallrobustnesschecks.Figure6showsthemarginaleffectsofanhourincreaseintheaveragenumberofhoursspentperweekontheinternet,allelseequal,calculatedatmeans.Aonehourincreaseintheaverageweeklyusageoftheinternetonaverageincreasestheprobabilityofobtainingabachelor’sdegreeby1.42percent.Thiseffectforagraduatedegreeisa1.39percentincreaseonaverage.

TheunexpectedresultthatagedoesnotexhibitdecreasingreturnsisanissueandthroughasetofrobustnesschecksseemstobeananomalywiththemainresultsseeTable2columns4and5.Therobustnesschecksgivemoreevidencethatoverallcontrollingforage,withtheadditionofthequadraticterm,andtheuseofindividualincomedoescontrolforendogeneityofeducationalattainmentandinternetuseasbestaspossible.Withthedifferentincomevariablesusedintherobustnesschecks,whichincreasesamplesizeandrepresentationofhigherincomes,thequadratictermissignificantatthe99percentconfidencelevel.ThetwodifferentincomevariablesareaLn(nominalrespondent’sincome)oraLn(realincomeperearner)variable.Therearereasonsthatanyoneoftheincomevariablescouldfitintothismodel,butthereportedresultsarestillthebestoptionforuseascontrolsinthisspecification.ThedifferenceinthedataoftheincomevariablescomefrommorevariationatthehigherendLn(Familyincome)orthetruncationofhigherincomeearnersandthecategoricalfictionalformofthevariable.Theresultoftherobustnessanalysisgivestheaddedpowerofadditionalobservationsalsoaddstotheeffectivenessofincomeandagecontrollingforendogeneitybias.

23ibid

16

FIGURE6.TheEffectofanHourIncreaseinInternetUsagefromSampleAverages Source:MilkenInstitute

17

AcknowledgmentsIwouldliketothankRossDeVol,MinoliRatnatunga,JessicaJackson,andKenSagynbekovfortheirhelpinthedevelopmentprocessofthispaper.

About the Author JoeLeeisaresearchanalystwiththeMilkenInstituteontheregionaleconomicsteam.Hespecializesinlaboreconomicswithafocusonhumancapitalandeconomicdevelopment.Hisrecentworkincludesregionalworkforceissuesandhumancapitalinteractionwithurbanclusters.

BeforejoiningtheInstitute,LeeworkedasalabinstructorinthedepartmentofeconomicsatCaliforniaStateUniversity,LongBeach(CSULB)andworkedforAmazoninSeattle,Washington.LeereceivedhisMAineconomicsfromCSULBandgraduatedfromTheEvergreenStateCollege.

About the Center for Jobs and Human Capital TheMilkenInstituteCenterforJobsandHumanCapitalpromotesprosperityandsustainableeconomicgrowtharoundtheworldbyincreasingtheunderstandingofthedynamicsthatdrivejobcreationandpromoteindustryexpansion. About the Milken Institute TheMilkenInstituteisanonprofit,nonpartisanthinktankdeterminedtoincreaseglobalprosperitybyadvancingcollaborativesolutionsthatwidenaccesstocapital,createjobsandimprovehealth.Wedothisthroughindependent,data-drivenresearch,action-orientedmeetingsandmeaningfulpolicyinitiatives.©2017MilkenInstituteThisworkismadeavailableunderthetermsoftheCreativeCommonsAttribution-NonCommercial-NoDerivs3.0UnportedLicense,availableatcreativecommons.org/licenses/by-nc-nd/3.0/

top related