Upload
others
View
3
Download
0
Embed Size (px)
Citation preview
TSLCScholars:Amixedmethodsexamina6onofacomprehensivecollegetransi6onandsuccess
programforlow-incomestudents
November 12, 2016
AdriannaKezar,DarnellCole,Ma1Soldner,Ta4anaMelguizo,RobertReason,RosiePerez,AraceliEspinoza-Wade
PulliasCenterattheUniversityofSouthernCalifornia(USC-Pullias)and
AmericanIns4tutesforResearch(AIR)
Mixed-MethodsDesign
• Presenta6on1:Thecollegetransi4onprogramandoverviewofthemixed-methodsresearchdesign
• Presenta6on2:Anoverviewofsurveydesignand
forma4veanalysis• Presenta6on3:Summa4veresearchdesign• Presenta6on4:Qualita4veresearchdesign
TheProgramandOverviewofMixedMethodsResearch
Design
AdriannaKezar
TSLCProgram•
•Sincetheearly1960s,theFounda4onprovidedscholarshipsforlow-incomegraduatesfromNebraskahighschools
• •2007:TheFounda4onincreasedscholarshipsandcreatedlearningcommuni4esprogram(TSLC)at3UniversityofNebraskacampuses
• •TSLCprogramincludesfiveyearsoffunding,smalllearningcommunitycourses,mentoring,residen4alcomponent,firstyearexperiencecourse,andprogrammingfromstaff(2yearprogram)
• •2014:Decisiontoembarkonevalua4on
ProjectGoal
TounderstandhowthekeyelementsoftheThompsonScholarsLearningCommunity
(TSLC)facilitateengagement,thedevelopmentofacademicself-efficacy,
ma1ering/senseofbelonging,persistenceandotheroutcomes.
Framing•PriorresearchonLCsmostlydescrip4ve•Fewoutcomestudies-narrowstudiesofGPAmostly;Learning communi4esandrecentMDRCstudyoflimitedimpact
•One-4melookatoutcomes
•Focusinliteratureissenseofbelonging,self-efficacy,butnotstudied–weendedupfocusingonpsycho-socialoutcomes
•Programismuchmorethanalearningcommunity;amul4-prongedprogram–Howdoweseparateoutimpactofdifferentcomponentparts?
–Moststudiesshownoimpact
•Engagementvs.climatevs.contexts
ModelofEnvironmentalVariablesrelatedtoTSLC
StudentBackgroundCharacteris6cs
RaceGender
FamilyIncomeCollegeReadiness
ExpectedComple4onCampus
FinancialStressTimeUseMajor
Pre-test(outcome)
TSLCEngagement
PeerInterac4onFaculty-StudentAc4vi4es/
EventsFYS/SAC
MentoringGradeCheckStaff-StudentResiden4alOrienta4on
Psycho-SocialOutcomes
ResiliencySenseofBelonging
Ma1eringCollegeKnowledge
Career/MajorSelf-EfficacyAcademicSelf-EfficacySocialSelf-Efficacy
ModeladaptedfromTinto(1997)&Hurtadoetal.(2007)
AcademicSSHealthOfficesCampusSS
CampusServiceUse
ResearchQues4ons1.Howdothekeyprogramcomponentssharedacrossthethreecampuses
facilitateengagement,thedevelopmentofacademicself-efficacy,ma1ering/senseofbelonging,andtheotheroutcomes?
2.Howdodifferencesintheprogramsacrossthethreecampusesfacilitateengagement,shapethedevelopmentofacademicself-efficacy,ma1ering/senseofbelonging,andtheotheroutcomes?
3.Howdodifferencesbyloca4on–asrural,urban,commuterandmoretradi4onalsuburbancampus–shapestudents’programexperienceandoutcomes?
4.HowdoestheTLSCexperienceaffectpar4cularsubgroupsincludingminoritystudentsandstudentswithdifferentlevelsofacademicprepara4on?
5.HowdoTSLCscholarsexperiencethetransi4onoutoftheprogramandhowmightthisexperienceaffecttheircon4nuededuca4onalsuccess(reten4onandgradua4on),academicselfefficacy,andma1ering/senseofbelonging?
ResearchQues4ons6.Howdofacultyandclassroomexperiencesshapestudents’program
engagementandoutcomes?7.HowdodifferentstudentsengagewiththeTSLCprogramandhowdoesthis
engagementappeartoshapetheirexperienceandoutcomes?8.Aretheredifferencesinthepsycho-socialandtradi/onalacademicoutcomes
betweenthestudentswhoexperiencedtheTSLCcomparedtothosewhodidnot(i.e.,CollegeOpportunityScholarship(COS)andnon-recipients/control)?
9.Aretheredifferencesinthepsycho-socialandtradi/onalacademicoutcomesbetweenthestudentswhoexperiencedtheTSLCcomparedtothosewhodidnot(i.e.,CollegeOpportunityScholarship(COS)andnon-recipients/control)bytypeofTSLC(LC-Kearney,LC-Omaha,andLC-Lincoln)?
10.Aretheredifferencesinthepsycho-socialandtradi/onalacademicoutcomesbetweenstudentswhoreceivedfinancialsupport(i.e.,BuffetScholarship(BS)orCOS)andthosewhodidnot?
ResearchQues4ons11.a)Aretheredifferencesinthepsycho-sociallevels(i.e.,measuresofacademicself-efficacyormeasuresofsenseofbelonging)betweenstudentswhopar4cipatedintheprogram(i.e.,TSLC)andthosewhodidnot(i.e.,CollegeOpportunityScholarship(COS)andcontrol)?11.b)Isthereanassocia4onbetweenengagementlevels(mediator)betweenstudentswhopar4cipatedintheprogram(i.e.,TSLC)andthosewhodidnot(i.e.,CollegeOpportunityScholarship(COS)andcontrol)?11.c)Aretheredifferencesinthepsycho-socialvariablesbetweenstudentswhopar4cipatedintheprogram(i.e.,TSLC)andthosewhodidnot(i.e.,CollegeOpportunityScholarship(COS)andcontrol)anercontrollingforengagementlevels?12.Aretheredifferencesinthepsycho-socialandtradi4onalacademicoutcomesbetweenthestudentswhoexperiencedtheTSLCcomparedtothosewhodidnot(i.e.,CollegeOpportunityScholarship(COS)andnon-recipients/control)byselectedstudentcharacteris4cs(i.e.,minoritystudents)?
ParallelMixedMethodswithEmbeddedSequen4alExploratoryDesignCohort1:2015-2020andCohort2:2016-2020
Casestudy Digitaldiariesandinterviews Socialmedia Focusgroups
2015Focusgroups
• Cohort1Baseline(Fall2015)
• Cohort1Follow-up
(Spring2016)
2016Focusgoups
*Cohort2Baseline(Fall2016)
• Cohort1&2Follow-up
(Spring2017)
2017Focusgroups
• Cohort1&2Follow-up
(Spring2018)
2018Focusgroups
• Cohort1&2Follow-up
(Spring2019)
(Sequen6alExploratoryDesign)
PsychometricPilot
(Summer)2015
Cogni4veInterviews(Spring)2016
Opera4onalPilot
(Spring)2016
*Cohort1&2Follow-up
(Spring2020)
MixedMethodsImplementa4on• Retreatsthroughouttheyeartodiscuss
quan4ta4veandqualita4veconcepts,designanddatacollec4on
• Bi-monthlyteammee4ngs–mixedmethodsdiscussed
• Casestudydatausedtodesignbaselineandfollowupsurveys
–Areasofprogramtoinclude,differencesacrosscampusesandwordingofprogramelementsbyvariouscampuses
*
DesigninginRealTime•Mixedmethodsdesignwherequalita4veinformssurvey,butsurveyneedstostartASAPtoleveragecontrolgroup•TSLCprogramchangeseachyear•Ourunderstandingofprogramchangesover4me•Changeinprogramunderstandingrelatestonewconstructs,theoriesandframing
2015 2016 2017 2018 2019 2020 First 6-8 months
descriptive case study
Ethnographic case study
begins
Case study continues
Spring: Case study ends
Social media analysis
Social media analysis
Social media analysis
Social media analysis
Digital diaries/interviews
begin
Digital diaries/interviews continue
Digital diaries/interviews continue
Digital diaries/interviews continue
Digital diaries/interviews continue
STS survey STS survey STS survey STS survey STS survey STS survey Fall focus
groups Spring/Fall
focus groups Spring/Fall
focus groups Spring/Fall
focus groups Spring/Fall
focus groups
MixedMethodsImplementa4on•Focusgroupsaboutsummerexperienceinformedfollowupsurvey•Casestudyiden4fieduniqueprogramma4celementsbycampuswhichwasusedinsurveydesign•Surveyfindingsusedtodesigndigitaldiaryques4ons•Constructsrevisitedandreconsideredbasedondatainterac4on,e.g.,resiliency
TeamBuilding&Communica4on•On-boardingasteamassembled•Bi-monthlyteammee4ngs•Sub-teams–weeklymee4ngs•Basecamp•Seminars/retreats•Languagedifferences
–methods,theories,paradigms,researchexperience
TeamTraining•Mixedmethods•Constructsandtheories•Differentmembers,needsandengagement:
–Graduatestudents–Postdocs
–Na4onallyrecognizednon-profitwithkeysurveyexper4se
Forma4veResearch
DarnellCole(USC)MarkMasterton&Ma1Soldner(AIR)
DevelopingtheDataset
ScholarsSurvey
Key Design Features ● Self-administered via the web ● Optimized for mobile devices ● Incentives-based ● Reminders via email and text message
Key Methodological Features ● Dynamic, allowing targeting of student subgroups based on
pre-loaded administrative data ● Adaptive non-response follow-up ● Imputation used to address item-level missingness ● Weighted to adjust for non-response bias
ScholarsSurvey
Year1SurveyOverviewComponent Purpose Timing Sample and Respondents
Psychometric Pilot
Validate key constructs Evaluate operational protocols
July, 2015
972 prior scholars 350 respondents, RR = 36%
Baseline Survey
Capture baseline data about background, high school experiences and college expectations
August, 2015
1335 scholars in 2015-16 cohort 1074 respondents, RR = 80%
2nd Psychometric Pilot
Validate student engagement measures
January, 2016
1652 prior scholars 332 respondents, RR = 21%
Cognitive Interviews
Refine student engagement measures
March, 2016
11 prior scholars
Follow-up Survey
Capture outcome and engagement data at end of scholars’ first year
May, 2016
1297 scholars in 2015-16 cohort 951 respondents, RR ≈ 73%
ExemplarConstructsBaseline Follow-up
§ High school sense of belonging
§ High school interactions with peers
§ High school interactions with faculty
§ High school time use
§ Expectations about social self-efficacy
§ Expectations about academic self-efficacy
§ Expectations about mattering
§ Expectations about belonging
§ Expectations about likelihood of graduation
§ “Actual” academic self-efficacy
§ “Actual” social self-efficacy
§ Experience of mattering
§ Experience of belonging
§ Engagement with peers and faculty
§ Engagement with program features
§ Financial stress
§ Revised graduation expectations
Takeaway#1:Don’tSkipPilotTes4ng
• Psychometricproper4esofadaptedmeasuresareonenintheball-parkofpublishedresults,butusingConfirmatoryFactorAnalysistoverifycanaddconfidenceinyourchoices
• WeusedgradedresponseIRTmodelsasanaddi4onalcheckfollowingCFA,par4cularlywhenwewereconsideringscalereduc4on
24
Takeaway#2:BeWaryofTiming
• Ourbaselineadministra4onuncoveredatleastonephenomenonthatwasnotdetectedinthepilot,givenitspopula4onandthe4mingofadministra4on:higherthanan4cipatedbaselinescoresonself-efficacymeasures
• Whatisthetemporalnatureofyourphenomena?Iftheyare4me-variant,howbesttomeasure?
25
Takeaway#3:Don’tSkipCogni4veInterviewing
• Althoughwethoughtwehadasenseofhowtodesignengagementitemsthatwouldbe“respondentfriendly,”wehadnoideahowhardtheywouldbetoanswerinanyformatun4lwehadconductedcogni4veinterviews.
• Interviewsbringevidencetodebatesbetweenthosewhoare
arguing“weneeddetailedinforma4on”and“respondentswon’tbeabletonavigatethecomplexity,”sotherightbalancecanbestruck.
26
27
Takeaway#4:AnalyzeDataPromptlytoEvaluateConstructs
• Givenhighbaselinemeasures,wewereworriedaboutwhetherwecoulddetectasimpletreatmenteffectatfollow-up,controllingforbaselinelevel.Wecould,butdevelopednewitemsforthenextcohortthatwehopewillincreaseourabilitytodiscernbetweenmorenuancedlevelsoftheunderlyingconstruct.
28
Takeaway#5a:Don’tBeAfraidtoExperimentwithMethods
• Inalongitudinalsurvey,one(ofmany)tricksisensuringhighresponseratesateachstage—withoutbreakingthebank.Experimentscanhelpyouiden4fy“whatworks”withyourpopula4onstokeepresponserateshighandminimizedifferen4alnon-responsebygroups.
• Inthefollow-upsurvey,AIRconductedanRCTtodeterminewhetherwecouldsafelytailorourapproachtofollow-up,basedonstudents’es4matedpropensitytorespond.
29
Takeaway#5b:Don’tBeAfraidtoExperimentwithMethods
• Inthefollow-upsurvey,AIRconductedanRCTtodeterminewhetherwecouldtailorfollow-upefforts,basedonstudents’es4matedpropensitytorespond.
• Ourgoalwastoshineffortfrommostlikelyrespondentstotheleastlikely,makingitpossibletodophonereminders—orphoneinterviews—forthelowestpropensitygroup.
• Earlyevidence:a5to8percentagepointresponseratebumpinourlowestpropensitygroups.
30
Non-responseFollow-upExperiment
Contact Date Control
Condition
Experimental Condition High
Propensity Medium
Propensity Low
Propensity Very Low
Propensity Invitation e-mail 8/29 X X X X X Reminder e-mail 1 9/1 X X X X X Reminder text 1 9/6 X X X Reminder e-mail 2 9/8 X X X X X Phone reminder (R) / interview (I)
9/12-14 R R R I
Reminder e-mail 3 9/15 X X X X Reminder text 2 9/19 X X Reminder e-mail 4 9/22 X X X
31
ResponseRatebyResponsePropensityGroupandExperimentalCondi4on
Response Propensity Group
Response Rate
Control Experimental Difference Very Low 65.7% 70.3% 4.6% Low 71.9% 80.2% 8.3% Medium 87.6% 82.7% -4.9% High 96.1% 84.3% -11.8%
Summa4veResearch
Ta4anaMelguizoFrancisco(Paco)Martorell
W.EdwardChiElizabethParkAdriannaKezar
OverallResearchGoalSumma4ve
RigorouslyexplorewhethertheTSLCgenerateddifferencesinthesocialandpsychologicaloutcomesrela4vetothe
CollegeOpportunityScholarship(COS)andnon-recipients/control.
USC/AIR’SThompsonScholarLearningCommuni4es(TSLC)SUB-SAMPLE
MustFund
Control
CollegeOpportunityScholarship
(COS)
Buffe1Scholar(BS)
Applicants
EnrolledNUCampus
(3)
EnrolledNECommunityCollege(7)
EnrolledOut-of-State
EnrolledNEStateCollegeCampus(3)
DonotEnroll
Eachgroupcouldfollowoneofthese6paths
USCcanONLY“follow”1ofthese6groupsandcollectsurveydata
USC/AIRSTS
LongitudinalSurvey
EnrolledPrivate
CollegeNE
MethodologicalChallenges• Twopoten4alsourcesofbias:
– Studentself-selec4on– Changesinoriginalgroupingsresul4ngfromtherandomiza4on
– ShinintoTSLC:studentswhotargetedastateorcommunitycollegeandreceivedtheBuffe1grantcouldenrollatanNUandbecomepartoftheTSLC
– ShinoutofControl:studentswhoini4allytargetanNUanddidnotreceivetheBuffe1grantoragoodfinancialpackagemightenrollatadifferentcollegeornotenrollincollegeatall
• Thisisproblema4casthesestudentsmightdifferintermsofobservableandun-observablefactorsfromthestudentsoriginallyrandomized– Needtoassesstheproblembylookingatdifferencesinbackgroundand
demographiccharacteris4csinresul4nglongitudinalsample
MethodologicalChallenges• Surveydifferen4alResponseBias:
• Changesinoriginalgroupingsresul4ngfromsurveydifferen4alresponserate
– Cohort2015:» Baselinesurveyresponserateof80%,theresponseratevariedbetweentreatmentsandcontrol(e.g.,TSLC=87%,COS=85%,andControl=66%)
» First-followupresponserateof73.3%(e.g.,BUFFETT=82.6%,COS=82.39%,andControl=60.87%)
• Needtoiden4fythemagnitudeoftheproblembyes4ma4ngthepropor4onofstudentswhoremainedinthelongitudinalsamplefromtheoriginalassignment
– Cohort2015:TSLC=66%,COS=58%,andControl=33%
Cohort2015DatasetsandSample
• Datasets– 2015STBFScholarshipApplica4onData– SurveyofThompsonScholars(STS)
• Cohort2015Baseline• FirstFollow-upSurveyData
• Sample– Studentsfromthe2015cohortwhotargetedoneNUcollegeintheirapplica4onandwerepartofthestra4fiedrandomassignmentthatwasusedtoawardscholarships
AnalysestoAssesstheSelec4onProblems
• Twosidedt-testsofmeandifferences• Mul4plemeancomparisontests,BonferroniAdjusted
• Chi-squaretests• Regressionanalyses• Latentmeancomparison*
ProposedCorrec4onsforSelec4onProblem
ControlforObservableDifferencesAcrossGroups(RegressionandPropensityScoreWeigh4ng:HeckmanandNavarro,2004;Hirano,Imbens,andRidder,2003)
• Richinforma4ononapplicantsishelpful• But,s4llrequiresstrong‘noselec4ononobservables’
assump4onTreatmentEffectBounds
• Lee“trimming”procedureforcon4nuousoutcomes• Challengeconver4ngsocialandpsychologicaloutcomes
Likert-scaleintocon4nuousvariables
Takeaway #1 The power and challenges of using simultaneously
qualitative and psychometric methods to measure social and psychological constructs
• Thoroughreviewoftheoriesandtheliterature• Leveragecasestudytogainadeepcontextualunderstanding
oftheprogram• Relateprogramma4celementswithsocialandpsychological
outcomesofinterest• Partnerandcollaboratewithsurveyexperts• CommunicateregularlywithProgramDirectorsand
Founda4onOfficials
Takeaway #2 Partner with survey firm to maximize survey response
rates and minimize differential response rates • SurveyResponserates
- Needtogetappropriate(80%)andbalanced(similarresponseratesforTsandCgroups)surveyresponserates
• Psychometricreports
- Needtocarefullyandsystema4callytestforvalidityandreliabilityofconstructsnotonlyinthepilotbutinthebaselineandfollow-ups
Takeaway #3 Leverage disciplinary methodological expertise
• Economics- Studentself-selec4on
• PsychometricsandEduca4onalPsychology- Itemresponsetheory(IRT)- LatentVariableAnalyses- Structuralequa4onmodels(SEM)
Take Away #4 Summative Research deeply benefited from Case
Study • Longitudinalcasestudyprovidesdeepcontextualunderstanding
oftheprogramasitwasconceivedandimplementedinthedifferentsites
• Casestudyinformedlongitudinalsurvey• Focusgroupsandcogni4veinterviewsarekeyto:
–Iden4fykeyprogramma4celementsasexperiencedbythestudents–Testassump4onsofmeaningofsurveyscalesandindexesforstudents
Take Away #5 Case Study and Formative Benefitted from
Summative Research • Cohortcategoriza6on
–Earlyanalysesofsurveydatatopresentdescrip4ve portraitsforstudents
• Impactresults
–Preliminaryimpactfindingscanbeexploredfurthertogetamorenuancedunderstandingofthemechanismsthroughwhichtheprogramisworking(e.g.,promptsfordigital diarypar4cipantsrelatedtosocialself-efficacy)
Qualita4veResearch
RobertReason,RonaldHalle1,RosemaryPerez,AraceliEspinoza-Wade
OverviewofDataCollec4onAc4vi4es
• Observa4onsandInterviews• DigitalDiaries• SocialMedia• DocumentAnalysis
PurposeswithinMixedMethods
• Exploratory
• Explanatory
LessonsLearnedabouttheResearchDesign
• DigitalDiaries– Breadthv.Depth– Language– Management
• SocialMedia– Analyzingsocialmediacontent– Integra4onintoMixedMethods
LessonsLearnedabouttheResearchDesign
• LongitudinalQualita4veResearch– Engagement– Reten4on– ResearcherTransi4ons
• ResearchasInterven4on
Ques4onsLenUnanswered
• Howdowebalanceethnographicandcasestudymethods?
• Howdowedevelopacasestudyofprogramsthatcon4nuetoevolve?
• Howdowemanageandintegratesuchvariedformsofrichdatainqualita4veandmixedmethodanalysis?
51
Q&A
Wewelcomeyourfeedbackandideas!
PulliasCenterforHigherEduca4onRossierSchoolofEduca4on
UniversityofSouthernCalifornia
3470TrousdaleParkwayWaitePhillipsHall701LosAngeles,CA90089
Website:pullias.usc.edu