XLSTAT: Notched box plots and box plots with two categorical variables in Excel

 

Dataset for Example

The data used in this tutorial are based on a study from the National Health and Nutrition Examination Survey (NHANES), which conducted a representative survey of the adult US population. We therefore have, for 150 individuals, their total blood cholesterol levels (in mg/dL) and two categorical variables: smoking status and gender.

 

Setting Up notched box plots

• Open XLSTAT.

• Click on the Visualizing data menu and select the Univariate plots feature.

• The dialog box appears.

In the General tab:

• Select the Cholesterol column as Quantitative Data.

• Select the Smoking column as Subsamples.

In the Options tab:

• Enable the option to Group plots and select Categories as the grouping criterion.

• Enable the Notched option.

• Click on the OK button to launch the computations.

• The results are displayed on a new Excel sheet.

 

Interpreting the results from notched box plots

On a notched box plot, the red crosses represent the means, whilst the central horizontal bars represent the medians. The notched section represents a confidence interval around the medians. Finally, the (vertical) length of the boxes corresponds to the interquartile range and illustrates the spread of 50% of the data points around the median.

The key observation to be made from the previous chart is that the notched sections for non-smokers and current smokers do not overlap, which therefore indicates a significant difference in the medians of these two groups. In contrast, for ex-smokers, there is a slight overlap of this notched section with both the non-smokers’ and the current smokers’ sections, which appears to illustrate a gradual effect of quitting smoking on total cholesterol. One could also consider conducting a Mood’s test to test the hypothesis of equality of medians, in order to validate these findings.

Now, to take the analysis further, we wish to study the impact of a second categorical variable on total cholesterol: the gender variable.

 

Setting Up an analysis with two categorical variables

In the General tab:

• Select the Cholesterol column as Quantitative Data.

• Select the Smoking and Gender columns as Subsamples.

In the Options tab:

• Enable the option to Group plots and select Categories as the grouping criterion.

 

Interpreting the results from box plots with two categorical variables

When a second categorical variable is present, the different categories of the second categorical variable that have been specified are distinguished on the box plot by colour-coding.

After analysing the various statistical indicators shown in this chart (median, mean, quartiles), three main points can be identified. Firstly, it is noted that the finding made in the previous section – namely that tobacco use appears to be associated with an increase in total cholesterol – is confirmed for both men and women. The second observation is that, across the three different smoking statuses, women have values that are quite significantly higher than those of men. Finally, it is noted that, for both sexes, there is indeed an effect of giving up smoking on the reduction of total cholesterol in the blood.

In conclusion, the use of a notched box plot combined with the analysis of a box plot with two categorical variables enabled us to demonstrate that smoking is associated with a significant increase in total cholesterol in both men and women. We also found that the beneficial effect of smoking cessation on total cholesterol levels was evident as soon as the individual became an ex-smoker, although a complete return to non-smoker levels had not yet been observed. Finally, it was noted that gender plays an independent role, with women consistently having higher cholesterol levels for equivalent smoking status.