Chapter 12. Descriptive Statistics and Visualisation
12.1 Begin with the data, not the hypothesis test
Descriptive analysis checks the sample and reveals data problems. Report sample size, relevant participant characteristics, missingness, distributions and central tendency before presenting inferential models. Use statistics appropriate to the measurement level and distribution.
| Variable type | Useful summaries | Possible displays |
|---|---|---|
| Nominal categorical | Count and percentage | Bar chart |
| Ordinal categorical | Count, percentage, median where meaningful | Ordered bar chart |
| Approximately continuous | Mean, SD, range; median and IQR if skewed | Histogram, boxplot, density plot |
| Paired/repeated measurements | Within-person change and its distribution | Paired dot plot, line plot |
| Time series | Trend, seasonal pattern, gaps | Line chart with clear time axis |
12.2 Mean, median and dispersion
The mean is the arithmetic average and can be affected by extreme values. The median is the middle ordered value and is robust to many extreme observations. The standard deviation (SD) describes dispersion around the mean; the interquartile range (IQR) describes the middle 50% of observations. The standard error describes the estimated sampling variability of a statistic and should not be used as if it were the spread of individual observations.
12.3 Confidence intervals and effect sizes
A confidence interval communicates the precision of an estimate under the model and sampling assumptions. A 95% confidence procedure has a long-run coverage interpretation; it does not mean that there is a 95% probability that a fixed frequentist parameter lies in this particular computed interval. Effect sizes describe magnitude in interpretable units or standardised form. Report estimates and intervals, not p-values alone.
12.4 Graphical integrity
Label axes, units, sample sizes and uncertainty where relevant. Avoid truncated axes that exaggerate differences, unnecessary 3D effects, unreadable colours and decorative chartjunk. Use a chart that matches the data structure. A figure should make the result easier to evaluate, not merely make the report look polished.