Transcription
Practical Statisticsfor UX & Customer ResearchJeff Sauro, PhD & Jim Lewis, PhD
InstructorsJeff Sauro, PhDJim Lewis, PhD, CHFPFounding PrincipalDistinguished User Experience ResearcherJeff Sauro PhD, is the founding principal of MeasuringU. For overtwenty years he’s been conducting UX research, includingbenchmarking studies for clients.Jim is a Certified Human Factors Professional with a Ph.D. inExperimental Psychology (and M.A. in Engineering Psychology, minorin applied statistics).Jeff has published over twenty-five peer-reviewed research articlesand five other books, including Benchmarking the User Experience,Customer Analytics for Dummies and Quantifying the UserExperience.Before joining MeasuringU Jim worked at IBM for nearly 40 years. Heis an IBM Master Inventor ( 90 US patents) and has published over100 articles and papers. 2
Cluster Analysis
Topics Covered History of Cluster Analysis Different Types of Cluster Analysis Preview of the Steps for Conducting a Cluster Analysis 4
Overview of advanced UX analytical methodsAnalysis ofDifferencesAnalysis ofStructure Basic ANOVA Advanced ANOVA Basic Linear Regression Advanced Regression Top Task Analysis Discriminant Analysis Kano Model Logistic Regression Conjoint/MaxDiff Correlation Cluster Analysis Factor Analysis Latent Class Analysis t-Test A/B Two-Proportion TestLower ComplexityMore info: 5 Advanced Stats Techniques & When to Use ThemHigher Complexity 5
Applications of advanced analytical methodsQuestionMethodsSample ApplicationAre there significant differences?Two Proportion Testt-TestANOVAA/B testingTest two designsTest multiple designs and interactionsAre there significant similarities?CorrelationAssess relationships (e.g., CSAT and age)Are there significant predictors?Linear RegressionKey driver analysisIs there latent (hidden) structure?Cluster AnalysisFactor AnalysisLatent Class AnalysisPersona/segmentation and card sortingDevelop standardized UX questionnaireAdvanced persona developmentCan we determine membership in classes?Discriminant AnalysisLogistic RegressionCustomer segment classification toolStatistical basis for feature prioritizationWhat are the most important features/tasks?Conjoint AnalysisMaxDiff AnalysisKano ModelTop Task AnalysisExhaustive feature prioritizationStreamlined feature prioritizationAlternative feature prioritization methodIdentify most important tasksMore info: 5 Advanced Stats Techniques & When to Use Them 6
Earliest Use Of Cluster AnalysisAlfred Kroeber & Unknown PeruvianDriver and Kroeber 1932“We inquire whether the cultures carried or possessed by such ethnic groups are more or less similar to one another, on the basis oftheir containing or not containing traits such as matrilineate, avoidance, self-torture vows, the fire drill, sinew-backed bow, twinedweaving, ridged houses, etc.”More info: Quantitative Expression of Cultural Relationships [PDF] 7
Cultural groups were clustered by their shared attributes 8
Different algorithms used for groupingOver 100 Different Algorithms and the “best” one depends on the context and resultsCommon “Hierarchical” Clustering AlgorithmsWard’s Method is a popular approach used in SPSS based on the Squared Euclidian distanceMore info: 5 Advanced Stats Techniques & When to Use Them 9
Cluster Analysis Application 1: Card SortingSimilarity MatrixDendrogram 10
Cluster Analysis Application 2: SegmentationDendrogram from a similarity matrix on howsmartphone users rated features 11
Steps to Running a Cluster Analysis1.Examine Correlation/Similarity Matrix2.Select Cluster Algorithm in Software3.Inspect Dendrogram for ClustersCovered inExtended Examples4. Determine Clusters Based on Visual Inspection 12
SummaryThere are many approaches to cluster analysis Cluster analysis includes a variety of methods for identifying groups of items Two-step, k-means, hierarchical Hierarchical cluster analysis using Ward’s method works well for UX researchCluster analysis useful in UX research to infer groups from data Grouping cards after card-sorting study Grouping people in persona research Some subjectivity involved in determining the number of clustersInterpret cluster analysis with visual inspection Graphs useful in interpretation include dendogram and profile of means 13
Moderated UX Studies(in our Denver labs or remotely)Unmoderated Studies(using our MUIQ platform)Participant Recruiting(US & International)Eye Tracking &Facial Expression AnalysisNavigation Testing(Card-Sorting/Tree-Testing)Survey Design & Analysis(including MaxDiff & Kano)Statistical Analysis &Measurement AdvisingTraining &WorkshopsMeasuringU is a research firm based in Denver, Coloradofocusing on quantifying the user experience.
There are many approaches to cluster analysis Cluster analysis includes a variety of methods for identifying groups of items Two-step, k-means, hierarchical Hierarchical cluster analysis using Ward's method works well for UX research Cluster analysis useful in UX research to infer groups from data Grouping cards after card-sorting .
