Effective Use of Graphs Annie Herbert Medical Statistician Research & Development Support Unit Salford Royal (Hope) Hospitals NHS Foundation Trust ( ) 2227
Timetable TimeTask 60 minsPresentation 20 minsCoffee Break 90 mins Practical Tasks in IT Room
Outline Graphs for categorical data Graphs for numerical data Comparing groups Additional graphs (covered in other courses) Final tips & Computer packages
Categorical Data (1) Examples: Sex – Male/Female Blood Group – A/B/AB/O Employment Status – Unemployed/Part-time/Full-time
Categorical Data (2) Record: Frequency (discrete number) per category Summary: Frequency OR percentage/fraction/proportion Visually: - Bar Chart - Pie Chart
Example – Discharge Destination (1) Where Patient Livesn = 731 Alone339 (46.3%) Family210 (28.7%) Home180 (24.6%) Other2 (0.3%)
Example – Discharge Destination (2)
Example – Psychiatric Illness/ Discharge Destination (1) Psychiatric Illness? Where Patient Lives No n=208 Yes n=523 Alone117 (56%)222 (42%) Family81 (39%)129 (25%) Home9 (4%)171 (33%) Other1 (0%)
Example – Psychiatric Illness/ Discharge Destination (2) Where Patient Lives Psychiatric Illness? Alone (n=339)Family (n=210)Home (n=180)Other (n=2) No117 (35%)81 (39%)9 (5%)1 (50%) Yes222 (65%)129 (61%)171 (95%)1 (50%)
Example – Psychiatric Illness/ Discharge Destination Bar Chart
Stacked Bar Chart
Re-ordering categories can emphasize a certain effect:
The axis should always start from 0:
Bar Charts – Adv & Disadv Advantages: - Visually strong. - Easy to compare between more than one dataset. Disadvantages: -Categories can be ‘re-ordered’ to emphasize certain effects. -Misleading if not used for counts. -Misleading if y-axis not from 0.
Bar Charts – Things to consider: What group differences are you interested in? Frequencies or percentages? If percentage, it’s down to you to specify the totals. Is ‘Other’ a large frequency/percentage? Consider the categories as un-ordered when using a stacked bar chart.
Pie Charts
Pie Charts – Advantages: Easy to compare categories, are equidistant from each other. Ordering of categories does not emphasize certain effects as badly as stacked bar charts do.
Pie Charts – Disadvantages: No choice between frequencies and percentages (down to you to specify totals). Cannot put more than one data set into a pie chart. Lose individual values of the data. Limited space: if using more than 5 or 6 categories, chart can look complicated.
Numerical Data (1) Examples: Weight Blood Pressure Cholesterol Levels
Numerical Data (2) Record: Number/Value (discrete or continuous) Summary: - Mean (SD) - Median (IQR) Visually: - Histogram- Box plot- Spread plot
Data – Ages of Patients in Selenium Study Age
Histogram – Ages of Patients in Selenium Study
Histograms for the same data can vary:
Compromise:
Beware! Histogram is not Bar Chart
Histograms – Advantages: Visual display of interval frequencies, easy to compare intervals. Can give an idea of the distribution of the data, e.g. shape, typical value, spread.
Histograms – Disadvantages: Choice of interval width can alter appearance. Individual values lost. One data set per histogram, difficult to compare data sets.
Box Plot Upper Quartile Lower Quartile Median Extreme Outlier Outlier
Box Plots – Advantages: Defines many summary statistics in one plot. Defines ‘outliers’ explicitly. Can have more than one data set in a plot, so easy to compare data sets:
Box Plots – Disadvantages: More complicated visually than some other types of data plots. Individual values lost.
Spread Plots (1)
Spread Plots (2) Advantages: -Can give an idea of the distribution of the data, e.g. shape, typical value, spread. -Shows individual values of the data. -Can show more than one dataset in a plot. Disadvantages: -Not very widely used in journal publications. -Doesn’t explicitly summarise statistics or outliers as box plot does.
Relationships in Numerical Data
Serial Measurements
What information does this give? Mean ± SE, n ≈ 30 per group
Better to look at individual data…
…or give a sensible summary.
Kaplan-Meier Curve (step graph) Time-to-Event data.
Bland-Altman Plots (scatter plots) How well do two methods of measurement agree?
Forest Plots (Hi-Lo-Close charts) Meta-Analysis.
Final Pointers: Before plotting think about the type of data and what you would like to compare. Show all data rather than summaries where possible. Label axes clearly. Graph should ‘stand alone’. Make sure when comparing groups that outcome on the same scale. Make sure any colours used are sufficiently different from each other, and not red/green.
Using a Computer Package: PackageAdvantagesDisadvantages SPSSProduces journal quality graphs Difficult to start with Expensive StatsDirectWhen copied and pasted, these graphs may be edited in Word Difficult to draw bar/pie charts ExcelEasy to use for bar/pie charts Not a statistics package