190 likes | 399 Views
M oderators of panel conditioning in sensitive questions. A meta-analysis. Tanja Burgard, ZPID Trier Nadine Kasten, University of Trier Michael Bosnjak, ZPID Trier. Talk at the Conference of the European Survey Research Association (ESRA) 18.07.2019, Zagreb ( Croatia ).
E N D
Moderators of panel conditioning in sensitive questions. A meta-analysis. Tanja Burgard, ZPID TrierNadine Kasten, University of TrierMichael Bosnjak, ZPID Trier Talk at the Conference ofthe European Survey Research Association (ESRA) 18.07.2019, Zagreb (Croatia)
Definition andrelevanceofpanelconditioning An example: Veroff, Hatchett and Douvan (1992) randomly assigned newlywed couples to one of two groups: one that participated in frequent and intensive interviews (the study group) about marital satisfaction [..] and another that participated in minimal and infrequent interviews […]. The authors concluded that ‘‘[b]y the fourth year . . . The marriages of the study group couples appeared to be better adjusted on several dimensions of marital quality’’ (p. 315). • Panels arenecessarytoanswer longitudinal researchquestions, but dangerstovalidity: panelmortalityandpanelconditioning • Panel Conditioning = Learning effect in panelstudies • Problem: Due totheconditioningofrespondents in formersurveywaves, theyarenolongerrepresentativefor non-respondents in laterwaves. • Examples: Avoidanceof follow-upquestions, cristallizingofattitudes, increasedattentionforsurveytopics, knowledgechanges Warren, Halpern-Manners (2012): Panel Conditioning in Longitudinal Social Science Surveys. SociologicalMethodsand Research 41(4): 491-534.
sensitivityanditsrelevanceforpanelconditioning • Threecharacteristicsof sensitive questions (Tourangeau & al., 2000): • Answercallsforsociallyunacceptedanswer • Intrusive and private, taboo in everydayconversation (religion, income, sexuality) • Concernfordataprivacy, datasecurityanduseofdata Twopossibleeffectsofpanelconditioning in caseof sensitive questions: • Desirable: More trust in surveysituation→ More honest answers, lesssocialdesirabilityeffectsforattitudequestions • Not desirable: Consistencyeffect on behaviororself-reporting→ Adjustmentofactualorreportedbehaviortogreaterconformity in caseof deviant behavior
Measurement ofpanelconditioning Time 1 Time 2 Time 3 Group 1 T T • Between-subjects design: Comparisonofparametersofexperiencedandfreshparticipants C Group 2 C Group 3
Hypotheses • H1: The answersofexperiencedrespondentsdifferfromtheanswersoffreshrespondents in caseof sensitive questions. • H2: Socialdesirabilityeffects in sensitive questionsdifferdepending on the type ofquestion. • H2.1: Experiencedrespondentsanswerlesssociallydesirable in caseofattitudequestions. • H2.2: Experiencedrespondentsanswermoresociallydesirable in caseofbehaviorquestions. • Dosageeffects • H3: The moreoftenthetreatmentgroup was interviewed, thegreaterthedifferencebetweengroups. • H4: The greaterthe time-lag betweenwaves, thesmallerthedifferencebetweengroups. • H5: The kindofsensitivityinteractswiththe type ofquestion • H5.1: Questionsthatcallforsocialdesirableanswersincrease PC effects in caseofbehaviorquestions. • H5.2: Intrusive and private questionsdecrease PC effects in caseofattitudequestions.
Information searchandselection • First literaturesearch: • CLICsearch(broadsearchinterfacecontainingforexamplePsycInfo, PsycArticles, PubMed, Sociological Abstracts) • Search terms: „panelconditioning“, „surveyconditioning“, „time in sample“, „rotationgroupbias“ and 10 morerelatedsynonyms • Eligibilitycriteria: • (Quasi-) experiments • Population: Experiencedandfreshpanelrespondents (treatmentandcontrolgroup) • Treatment: Fomerconditioningby sensitive surveyquestion • Time ofcomparison: Bothgroupsareaskedforthe same, sensitive question • Outcome: Reportedbehaviororattitudesofbothgroups SMD • Forward andbackwardsearchwithrecordsfromthefirstsearchmeetingeligibilitycriteria
Data collectionandoutcome • Coding • Information on thereport: Author, Year ofpublication, funding • Intervention: Dataset, surveymode, yearofcomparison, countryofconduction, incentives, Type ofquestion, topic, conditioningfrequency, intervalbetweenwaves • Results: Outcomes ofbothgroups, Odds Ratios, teststatistics • Effectsizemeasure: Standardizedmeandifferences • SMD > 0: Experiencedpanelistsrespondlesssociallydesirable • SMD < 0: Experiencedpanelistsrespondmoresociallydesirable
Analysis method • Toaccountforthehierarchicaldatastructure, a three-level meta-analysis isused • Source: Harrer, M. & Ebert, D. D. (2018). Doing Meta-Analysis in R: A practical Guide. PROTECT Lab Friedrich-Alexander University Erlangen-Nuremberg. https://bookdown.org/MathiasHarrer/Doing_Meta_Analysis_in_R/ • R packageused: metafor 2.0-0
Overall effect (H1) andvariancedistribution • k = 154 effectsizes, x = 85 samples, n = 19 reports • Meaneffectofpanelconditioning: -0,028*** [-0,042; -0,013] • Overall, experiencedpanelistsrespondmoresociallydesirablethanfreshpanelists H1 • Distribution ofheterogeneity: • Sampling variance: 5,26 % • Withinstudies: 80,40 % • Betweenstudies: 14,33 % True heterogeneity, thatmaybeexplainedwithmoderators
Moderating effectof Type ofoutcome (H2) • H2.1: Experiencedrespondentsanswerlesssociallydesirable in caseofattitudequestions. • H2.2: Experiencedrespondentsanswermoresociallydesirable in caseofbehaviorquestions. • 13,6 % ofthevarianceofSMD‘swithinstudiesand 37,5 % ofthevariancebetweenstudiesisexplainedbythe type ofoutcome
Dosageeffects (H3 / H4) • Gesamt
kindofsensitivityand type ofquestion (H5) • H5.1: In caseofquestionsthatcallforsocialdesirableanswers, PC effectsforbehaviorquestionsarestronger. • H5.2: In caseof intrusive and private questions, PC effectsforattitudequestionsaresmaller.
Conclusions • As expected, PC effectsdiffersignificantlybetweenattitudeandbehaviorquestions. • Noevidencefor PC effects in attitudequestions • Experiencedrespondentsreportmoresociallydesirablebehavior • Moderators ofbehaviorquestions • Frequencyslightlyincreasesthe PC effect • Contradictorytoexpectations, time lageincreases PC effect plausible, asbehaviorislearnedover time? • Questionscallingforsocialdesirableanswersincrease PC effects in behaviorquestions • What‘snext? • PC effectsforotherkindofoutcomes (demographics, wellbeing, knowledge) • Targeted experimental studiesevaluatingtheeffectsoffrequencyandtiming
Thankyouforyourattention! Questions?!?
Manifestationsofpanelconditioning Panel conditioning in thecontextoftheansweringprocess in surveys(Tourangeauet al. 2000): • Stage 1: Comprehensionofthequestion [-] Change in attitudesorbehavior due toreflection / increasedattention(Sturgis et al 2009: Cognitivestimulusmodel) [+] Less„don‘tknow“-answers • Stage 2: Information retrieval [+] More reliableanswers due tobetteraccessibityof relevant information(Bergmann, Bath 2017) • Stage 3: Assessment ofavailableinformation [-] Freezingofattitudestoappearconsistently(Waterton, Lievesley 1989) • Stage 4: Reporting / Selectionofadequateanswer[+] Reductionofsocialdesirabilitybias more honest answering(Waterton, Lievesley 1989)[-] Reductionofthecognitiveburdenofthesurveybystrategicanswering/ satisficing(Krosnick 1991): • Negative answeringoffilterquestionstoavoid follow-upquestions • Selectionofacceptableanswerswithoutprocessingthecontent Dynamics and moderators of panel conditioning. A meta-analysis
Relevanceandproblemsof Panels • Demand ofthesciencecouncil: Infrastructure in psychologyfor longitudinal datacollection Bruder et. al (2014): Nationallyfounded online lab • Existing panel-infrastructures: NEPS; GESIS Panel, LISS Panel (Blom et al 2016) ZPID: Online-Lab • Relevanceofpanelinfrastructures • Improvementofresearchpossibilities (availabledatabasefor longitudinal researchquestions) • Provision byinfrastructuremoreefficient: • Recruitmentand care forpoolofparticipants • Methodologicalandtechnicalexpertise • Dangersofpanels: panelmortalityandpanelconditioning
Rationale fortheexpected time effectofpanelconditioningeffects • Pluralism / lessbindingnessofsocialnorms • Socialdesirabilitylessimportantfornewrespondents, too • Information overloadandscarcityofattention • Cognitivestimulus due tosurveyparticipationlesspronounced • Information ofprevioussurveysarelessaccessible due toamountofinformation • Increase in surveysandscientificstudies • More familiaritywiththerulesofsurveys • Satisficingandstrategicansweringismore probable withnewrespondents, too • General tendency: Decreaseofpanelconditioning, becauserespondentsarelessaffectedbythesurveyparticipationandthus, differencesbetweennewandexperiencedrespondentstendtolevel out Dynamics and moderators of panel conditioning. A meta-analysis