Percentiles and quartiles in the same graph

Click For Summary
SUMMARY

The discussion centers on the feasibility of depicting percentiles and quartiles in the same graph using Gnuplot. The dataset includes execution durations with input sizes ranging from 1.5GB to 6GB, featuring the 5th and 95th percentiles, as well as the 1st and 3rd quartiles. While it is acceptable to combine these metrics in a single visualization, the standard box-whisker plot may lead to confusion due to differing interpretations of whiskers. Alternative representations, such as line graphs or error bars, are recommended for clarity.

PREREQUISITES
  • Understanding of Gnuplot for data visualization
  • Familiarity with statistical concepts of percentiles and quartiles
  • Knowledge of box-whisker plot construction and interpretation
  • Basic data analysis skills to interpret execution duration datasets
NEXT STEPS
  • Research Gnuplot candlestick chart creation techniques
  • Learn about alternative data visualization methods for statistical data
  • Explore the implications of using box-whisker plots for non-standard data
  • Investigate how to effectively present execution duration data in graphs
USEFUL FOR

This discussion is beneficial for data analysts, statisticians, and software developers who need to visualize execution performance metrics and understand the relationship between percentiles and quartiles in their datasets.

xeon123
Messages
90
Reaction score
0
I don't know where to put this question. So, I am trying in this forum.

I have this dataset [1] that contains the duration of an execution with an input size from 1.5GB to 6GB. This dataset also contains the percentile 5%, 95%, and the 1st and 3rd quartile.

I am thinking in creating a gnuplot graph with candlesticks that will contain the data of the percentiles in the whiskersbars, and the boxes will contain the 1st and 3rd quartile. My goal of gathering percentiles and quartiles is to see how much the results are similar.

Is it ok to depict in the same graph the percentiles and the quartiles?

[1] My dataset

Code: 
size    duration        percentile5        1stquartile        3rdquartile        percentile95
1500    1177.31        1156.13            1090.47            1239.19            1424.83
3000    2021.48        1867.93            1440                 2317.33            3662.88
4500    3349.63        3885.14            2336.83            4376.49            4606.48
6000    3892.96        3785.14            3112.24            4790.24            5812.19
 
Physics news on Phys.org
xeon123 said:
Is it ok to depict in the same graph the percentiles and the quartiles?

Yes, of course.

Note however that if you make a box-whisker plot then you are not exactly following the usual practice of constructing it. The whiskers mean something different in a standard box-whisker plot than what you intend. So if you're presenting this to people, then perhaps using the box-whisker format is not a good idea since it is confusing. I'm sure you can come up with many other representations, like a graph of size vs duration where the different percentiles are all connected by the same line. Or using error bars...

For example:

logarithmic-data-with-outliers-with-regression-quantiles-and-least-squares.png


But yes, giving percentiles and quartiles in one plot is definitely ok.