Clustered multiple variables graphs

Command:

Graphs Clustered multiple variables graphs

Description

Creates graphs that allow to visually compare subgroups across different variables.

The graph can be composed from different elements: Bars, Horizontal lines, Markers and or Connecting lines for mean or median, with choice of different error bars for mean (95% CI, 1 SEM, 1 SD, 2 SD, 3 SD, range) or median (95% CI, 25-75 percentiles, 10-90 percentiles, 5-95 percentiles, 2.5-97.5 percentiles, 1-99 percentiles, range), Box-and-whisker plot or Notched box-and-whisker plot, Violin plot, Dot plot (display all data), or Dot and line diagram (Ladder plot).

Required input

In this dialog box, you enter:

• Variables: the different variables of interest.
• Define clusters by: a categorical variable containing codes to break-up the data into subgroups.
• Select: a filter to include only a selected subgroup of cases in the graph.
• Graphs:
  • Bars, Horizontal lines, Markers and/or Connecting lines for means or medians.
  • Error bars: the following error bars are available if at least one of the graph types Bars, Horizontal lines, Markers and/or Connecting lines is selected:
    If mean is selected: (none), or 95% CI for the mean, 1 SD, 2 SD, 3 SD, 1 SEM, and range.
    • Note that 2 SEM is not in this list: when the number of cases is large, mean ± 2 SEM corresponds to the 95% confidence interval (CI) for the mean. When the number of cases is small, then the 95% CI interval is calculated as mean ± t * SEM, where t is taken from a t-table with DF=n−1 and area A=95%) (see also SEM).
    • Although 1 SEM gives the more narower error bar, this option is not recommended since the resulting error bar may be highly misleading, especially when the number of cases in the groups is different. Preferably the 95% CI for the mean is used for providing a valid graphical comparison of means (Pocock, 1984), or use 2 SD as an indication for the variability of the data.
    If median is selected: (none), or 95% CI for the median, 25-75 percentiles, 10-90 percentiles, 5-95 percentiles, 2.5-97.5 percentiles, 1-99 percentiles, and range
    • When the number of cases is small, it is possible that the 95% CI for the median is not defined and that it will not be displayed in the graph.
    • When you use percentile ranges, take into account the number of observations: you need at least 100 observations for 1-99 percentiles, at least 20 for 5-95 percentiles, at least 10 for 10-90 percentile and at least 4 for 25-75^th percentiles.
  • Box-and-Whisker plot (Tukey, 1977) or Notched box-and-whisker plot (McGill et al., 1978). A Notched box-and-whisker plot is a variation of the box-and-whisker plot in which confidence intervals for the medians are shown by means of notches surrounding the medians. If the notches about two medians do not overlap, the medians are significantly different at a ± 95% confidence level. For a detailed description of a Box-and-Whisker plot and Notched Box-and-Whisker plot, see Construction of a Box-and-Whisker plot.
  • Violin plot. The violin plot (Hintze & Nelson, 1998) shows the density trace of the data. It is recommended to combine the violon plot with a box-and-whisker plot (select both options).
  • When you select Dots (plot all data, all observations will be displayed in the graph.
  • In the Dot and line diagram (or Ladder plot), all observations are plotted as individual dots, and observations from the different cases (rows in the spreadsheet) are connected by a line, see the example below.

• Options:
  • If the data require a logarithmic transformation, then select the Logarithmic transformation option.
  • Complete cases only: Option to include only complete cases in the graph. If selected, only cases with valid numerical data for all variables selected in the dialog box will be included in the graph.
  • Clustered by variables: Let clusters be defined by the Variables (X-axis will display groups defined by the Categorical variable, list of variables will appear in the Legend frame).

Examples

This is the same graph, but with the option "Clustered by variables" selected:

This is an example of a graph with option "Connecting lines" selected:

Literature

• Altman DG (1991) Practical statistics for medical research. London: Chapman and Hall.
• Hintze JL, Nelson RD (1998) Violin Plots: A Box Plot-Density Trace Synergism. The American Statistician 52:181-184.
• McGill R, Tukey JW, Larsen WA (1978) Variations of box plots. The American Statistician, 32, 12-16.
• Tukey JW (1977) Exploratory data analysis. Reading, Mass: Addison-Wesley Publishing Company.

See also

• How to define and use labels for the different categories of a variable
• Multiple comparison graphs
• Clustered multiple comparison graphs
• Multiple variables graphs
• Multiple line graph
• Format graph
• Graph legend
• Add graphical objects
• Reference lines

Recommended book

Exploratory Data Analysis
John W. Tukey

Buy from Amazon

The approach in this introductory book is that of informal study of the data. Methods range from plotting picture-drawing techniques to rather elaborate numerical summaries. Several of the methods are the original creations of the author, and all can be carried out either with pencil or aided by hand-held calculator.