Description
Systematically examines behavioural datasets through distributions, plots and summaries before modelling, to find structure, anomalies and questions worth testing.
Exploratory data analysis, formalised by John Tukey, is the disciplined practice of looking at data before deciding what it means. For behavioural UX data, analysts inspect distributions of key metrics, check missing and duplicated values, examine relationships with scatterplots and cross-tabulations, compare segments and time periods, and look for anomalies that indicate tracking errors or unexpected user strategies. EDA deliberately generates hypotheses rather than testing them, so its outputs are questions and candidate explanations. Rigour comes from following a checklist (univariate, bivariate, temporal, segment views), documenting data quality issues, using robust summaries such as medians and quantiles, and clearly separating exploratory findings from confirmatory evidence to avoid the garden of forking paths.
Attributes
- Discipline
- Statistics & Modelling
- Loss type
- Latent
- Stage
- Post-launch, Discovery
- Data
- Quantitative
- Effort
- Days
Steps
- EDA protocol for behavioral data:
- Distribution plots for all continuous variables (histogram + density): look for bimodality, extreme skew, or unexpected modes.
- (2) Pairplots for all variable pairs: look for non-linear relationships, clustered patterns.
- (3) Outlier analysis: flag any observation > 3 IQRs from Q3.
- (4) Missing data pattern: visualize missingness across variables.
- (5) Time trends: if longitudinal data, plot each variable over time before assuming stationarity.
Sources
- Advancing into Analytics — George Mount
- Practical Statistics for Data Scientists (Bruce & Bruce)
- Myatt, G., Making Sense of Data: A Practical Guide to Exploratory Data Analysis and Data Mining (2007)
Also known as
- EDA
- Exploratory analytics
Categories
- Analytics & Behaviour
- Statistics & Modelling