behavioural levers to enhance data visualisation
TRANSCRIPT
Behaviouralleverstoenhancedatavisualisation.JoseVila1andJoseL.Cervera-Ferri2
1.DevStatchairatUniversityofValencia.UniversityofValencia,ERI-CESandIDAL.2.DevStat
Abstract
Thispaperpresentsdifferentexperimentsimplementedbytheauthors,which hels understand critical behavioural insights conditioning theeffectivity of visualisation methods. Specifically, the paper surveysrecentresearchonthebehaviouralpatternsintheprocessingofsimplestatistical data in different visual formats (Arribas, Comeig, UrbanoandVila2014;Gomez,Martinez-MolesandVila2016;VilaandGomez2016)and the impactofgamification (Attanasi, Cervera,Hernandezand Vila 2014) in the understanding and utilisation of quantitativeinformation. The surveyed papers apply different experimentalmethods (economic experiments, eye tracking techniques,gamification,etc.).
1. IntroductionOne of the key implementation areas of the modernisation of official statistics, asrecalled in the European Statistical System Vision 2020, is the improvement ofdissemination and communication of official statistics, including data visualisation(Cervera-Ferrietal,2016).Datavisualisationaimstoaidexpertandnon-expertusersinexploring,understanding,andanalysingdatathroughiterativevisualexploration.Withthedevelopmentofuser-friendlyandpowerful ITtoolsfordatavisualisationandtheboominbigdataanalytics,datavisualisationisspreadinginavarietyofapplications,including official statistics.However, besides the technical aspects of preparing datavisualisations,itisimportanttorecallthatitisnotfreeofhumancognitivebiasesanddecision-heuristics.Thespecificwayinwhichstatistical informationispresentedmayhavearelevantimpactinhowthehumanbrainprocessestheinformation.Visualisationconditions both the perception process and the decision-making of the users of theinformation.Inthiscontext,visualisationmethodscanbeunderstoodasimplicitnudges,inthesenseofThalerandSunstein(2008).Section2surveysfourbehavioural-economicexperimental experiments implemented by the authors, which helps to understandcriticalbehaviouralinsightsconditioningtheeffectivityofvisualisationmethods.Section3presentsabriefdiscussionandimplicationsofthesestudiestoenhancevisualisationfromabehaviouralviewpoint,especiallyforofficialstatisticsproducers.
2. Reviewpapers2.1. Statistical formats to optimize evidence-based decision making: A behavioralapproach
Arribas,Comeig,UrbanoandVila(2014)assesstheimpactofalternativedisseminationformats in the presentation of quantitative data on both the optimality of decision-makingandthetimerequiredtoperformthedecision-makingprocess.Aneconomic
experiment provides the data for this study. The experiment presents statisticalinformationinsimplefrequenciesandrelativefrequenciesusingnumericalandpictorialrepresentations in the context of different informational environments, as shown inFigure1.
Figure 1. Example of the statistical formats
Thekeyfindingsarethatstatisticalinformationpresentedintermsofrelativefrequencyformats(boththenumericalandpictorialrelativeformats)giverisetomoreaccuratedecision-makingthandatapresentedintermsofsimplenumericalfrequencies.Whentime is the relevant variable, numerical formats lead to a faster interpretation thanpictorialones.2.2. Spanish regulation for labelling of financial products: a behavioral-experimentalanalysis.
Gomez,Martinez-MolesandVila (2016)assess the impactof theSpanishMinistryofEconomy and Competitiveness’ (Board of Executives (BOE) Order ECC/2316/2015.Economy and Competitiveness Ministry, Spain, 2015) new regulation for financialproduct labelling. Theydesign and conduct an economic experimentwhere subjectsmakeriskyinvestmentdecisionsunderthreedifferenttreatments(Figure2):acontrolgroupwhere subjectshaveonlyobjective informationabout thekey featuresof theproducts they must select and two treatment groups introducing visual labelsresemblingthelabelsrequiredunderthenewSpanishregulation.
Theresultsoftheexperiment,analysedwithintheframeworkofrank-dependentutilitytheory(Wakker,2010),shownthatvisuallabelsdonotchangetheutilityfunctionofthesubjects, but they do significantly affect the subjects’ weighting functions. Theintroductionofnumericalandcolor-codedlabelssignificantlyincreasestheconcavityoftheweightingfunctionsandincreasespessimismandrisk-aversionincaseswheretheprobabilityofobtainingthebestoutcomeishigh.Labelswidenthedifferencebetweenrealsubjects’behaviourandthatoftheperfectlyrationalagentsdescribedbyexpectedutilitytheory.Consequently,theseempiricalfindingsraisedoubtsastowhetherthenewregulation actually achieves its objectives. The regulation seeks to empower retailinvestorsbyenhancingtheirunderstandingoffinancialproducts.Introducingthevisuallabels,however,seeminglyincreasesthedifferencesbetweenactualrisklevelsandthedecisionweightsappliedbysubjectswhenmakingdecisions.Moreover,labelsincrease
investors’pessimismandrisk-aversionwhenthebestoutcomeislikelyandfailtoalterinvestors’risk-aversionwhentheworstoutcomeislikely.
Figure 2. Framings applied to present the information of a risky event
2.3.Extractingbusinessinformationfromgraphs:Aneyetrackingexperiment.
Although data visualization trough basic statistical graphs is a common practice tosupportevidence-drivenbusinessdecisionmaking,VilaandGomez(2016)showthattheextractionofrelevantinformationfromsuchgraphsmaybecomeadifficulttaskeveninverysimplesituations.Thispaperappliesthemethodologyofexperimentaleconomicsto the analysis of graph reading and processing to extract underlying information.Specifically, they design and implement a behavioural-economic experiment whosebaseline treatment includes graphical and numerical information. The experimentapplieseye-trackingtechnologytouncoversubtlecognitiveprocessingstagesthatareotherwisedifficulttoobserveinvisualizationevaluationstudies.
The experiment presented information in a bar graph and asked the participant toanswer a question related to the information in the graph. Therewas an economicincentive for right answer. Themain result of the paper is the identification of twodifferentvisualisationpatterns,linkedtothelevelofunderstandingoftheinformationandtheeffectivity indecision-making:visualizationpatternsofthe43.3%ofsubjectswhoanswer thequestionproperly (for short, effective visualizationpatterns) andof56.7%whoarenotabletoprovidetherightanswer(forshort,ineffectivevisualizationpatterns).Thekeyfeaturedistinguishingbothpatternsisthatsubjectsrespondingthequestionproperlyfocusedtheirattentioninasmallsubsetofrelevantandinformativepartsofthegraph,whereastheotherssparetheirattentioninawidersetofareas,someofthemcompletelyirrelevanttoanswertheproposedquestion,asshownintheheatmapsofFigure3.
Figure 3. Heat maps showing the effective and ineffective visualisation patterns. Colour represent the eyes
fixation time at each point of the graph, from light green (shortest fixation time) to intense red (longest fixation time).
a) Effectivevisualisation b) Ineffectivevisualisation
Theresultsoftheexperimentprovideanempiricalfoundationfortwokeyguidelinestoimprovechartpresentation.Firstly,theexperimentshowsthatthosesubjectswhoarenot able to answer the question properly seem to have difficulties to discriminaterelevantfromirrelevantelements.Theapplicationofgraphicalframingsthat(1)stressthekeypiecesof informationof thechart (for instance, thetopofeachbar)and(2)avoidhighlightingotherelementsinthegraphthatprovidenoinformationandcoulddistract the subjects during the scanning process could be useful to increase theeffectivenessofdatavisualization.
2.4.Canexpertiseclosetheexperience-descriptiongap?
InaframeworkofProspectTheory(KahnemanandTversky,1979),Attanasi,Cervera,Hernandez and Vila (2014) designed and implemented a behavioural-economicexperiment to estimate theweighting function. Following Abdellaoui, L'Haridon andParaschiv(2011),subjectdisclosetheirwillingnesstopayforaprospect(lotteries)undertwo alternative treatments. In the first one, named as “description treatment”, theinformationofthelotteryispresentedinanexplicitandobjectiveway,providingthenumeric value of the possible outcome and their actual probabilities. In secondtreatment, named as “gamified treatment”, subjects are not provided with explicitinformationonoutcomesandprobabilitiesandareinvitedtoplayagamewheretheycanobserveasampleofoutcomesofrealisationsofthelotteries.Theexperimentwasreplicatedtwice:oncewithanexpertgroupsincludingparticipantswithahighstatisticalbackground(staffofEurostat,NationalStatisticalOffices,etc.)andasecondtimewitha non-expert groups including participants with no quantitative background. Thisexperimental design provided with four alternative estimations of the weightingfunction,asshowninFigure4.
Figure 4. Estimation of the weighting function for the non-expert and expert groups. a)Non-expertsubjects
b)Expertsubjects
Figure4showsthat,fornon-expertusers,gamificationenhancestheunderstandingoftheinformationandimprovesitsusefordecision-making.However,forexpertusers,gamification does not enhance understanding and can be evenmisleading. In otherwords, the experiment shows that, gamification can reduce cognitive biases in theprocessofstatisticalinformationinspecificsegmentsofusers.
3. ConclusionsThefourbehavioural-economicexperimentsreviewedinthispapershowthattheroleofvisualisationgoes farbeyond than justmakeclearer thequantitative information.Visualisationisalsoabletoactivatecognitiveleversandconditiondecisionheuristics.Since visualisation impacts the understanding of information and is able to nudgedecision-making in an automatic and unconscious way (i. e. through System 1 ofreasoning,asdefinedinKahneman,2011),behaviouraleconomicexperimentsprovidewithaneffectivetooltoanalysetheimpactofalternativevisualisationformatsandtodefinetheoptimalvisualisationtechniqueineachdecisioncontextandforeachuserpersona.Untilnow,theapplicationofthisbehavioural-economicexperimentalmethodsto improve data dissemination, and specifically data visualisation for statisticaldissemination,isscarce.Inparticular,thefundamentalroleofofficialstatisticsintheprovisionofevidenceforpolicydesignandevaluationimpliesthatbetterunderstandingisneeded,throughtheresearchoncognitiveaspects,oftheimpactofdatavisualisationon(1)theawarenessabout existing statistical information (2) the understanding of complex issues (data
0,00#
0,25#
0,50#
0,75#
1,00#
0,00# 0,25# 0,50# 0,75# 1,00#0,00#
0,25#
0,50#
0,75#
1,00#
0,00# 0,25# 0,50# 0,75# 1,00#
Probability Probability
Dec
isio
n w
eigh
t
Dec
isio
n w
eigh
t
Description Gamification
0,00#
0,25#
0,50#
0,75#
1,00#
0,00# 0,25# 0,50# 0,75# 1,00#
Description
0,00#
0,25#
0,50#
0,75#
1,00#
0,00# 0,25# 0,50# 0,75# 1,00#
Gamification
Probability Probability
Dec
isio
n w
eigh
t
Dec
isio
n w
eigh
t
literacy)bypolicy-makers(3)theeffectiveuseofstatisticsinthedesignandevaluationofpolicyinstruments.Inthecaseofbusinessregistersandbusinessstatistics,thefollowingresearchavenuescanbesuggested:
- Impactofvisualisationontheidentificationbyanalystsoftrendsinthebusinesssector (e.g. emergence of economic activities, value-chain linkages betweensectors);
- Impactofgeographicvisualisationofthebusinessactivityforthedesignoflocaldevelopment policies (e.g. business clusters, Smart Specialisation strategies,localemploymentpolicies);
- Impactofvisualisationofbusinessstatisticsandregistersonbusinessmanagersto identify opportunities (e.g. location of establishments, design of supplyroutes);
- Impact of visualisation by job-searchers (students, unemployed, prospectiveentrepreneurs)ofbusinessstatisticsandregistersforthepersonalstrategyforjobsearchandskillacquisition.
Insummary,thereisacriticalresearchgaptobefilledanditsinnovationpotentialtobelevered for dissemination and official statistics and visualisation of information toimprovedecision-makinginbothpolicy-makingandbusinessdevelopment.References:
Abdellaoui,M.,L'Haridon,O.,&Paraschiv,C.(2011).Experiencedvs.describeduncertainty:Doweneedtwoprospecttheoryspecifications?.ManagementScience,57(10),1879-1895.
Arribas, I., Comeig, I., Urbano, A., and Vila, J. (2014). Statistical formats to optimize evidence-baseddecisionmaking:Abehavioralapproach.JournalofBusinessResearch,67(5),790-794.
Attanasi,G.,Cervera,J.,Hernandez,P.,andVila,J.(2014).Canexpertiseclosetheexperience-descriptiongap?.ERICESworkingpaper.
Cervera,-Ferri,J.L.,DeJonge,E.,V.Kasperiuniene,V.Dinculescu,P.Votta(2016).ESSVisualisationWorkshop–Valencia2016.Summaryandconclusions.http://ec.europa.eu/eurostat/cros/system/files/technical_workshop_report_valencia2016_final3006.pdf_en
Gómez,Y.,Martínez-Molés,V.,andVila,J.(2016).Spanishregulationforlabellingoffinancialproducts:abehavioral-experimentalanalysis.EconomiaPolitica,33(3),355-378.
Kahneman,D.,&Tversky,A.(1979).Prospecttheory:Ananalysisofdecisionunderrisk.Econometrica:Journaloftheeconometricsociety,263-291.
Kahneman,D.(2011).Thinking,fastandslow.Macmillan.
Thaler, Richard H.; Sunstein, Cass R. (2008). Nudge: Improving Decisions about Health, Wealth, andHappiness.YaleUniversityPress.
Vila, J.,&Gomez,Y. (2016).Extractingbusiness informationfromgraphs:Aneyetrackingexperiment.JournalofBusinessResearch,69(5),1741-1746.
Wakker, P. P. (2010). Prospect theory: For risk and ambiguity. Cambridge university press.