Data Visualization is a powerful tool for communicating information to a broader audience. This means apart from using the right color scheme, font size, and font type; using a proper chart type to depict the data is also important.
The focus of this article, as clear from the title, is time series. As a general convention, we visualize time series as line plots. But, if I say this is not always true, we can even use a bar chart to visualize time series. Will you say I am misrepresenting the data, or is it fine to use bar charts as alternates for line charts? Or is there an exception when using bar charts over line plots?
To understand this paradigm, first, let us look at the Time Series definition as stated on Wikipedia:
In mathematics, a time series is a series of data points indexed (or listed or graphed) in time order. Most commonly, a time series is a sequence taken at successive equally spaced points in time. Thus it is a sequence of discrete-time data. Examples of time series are heights of ocean tides, counts of sunspots, and the daily closing value of the Dow Jones Industrial Average.
It means that the data points in a time series are discrete values but continuous if a relationship exists with the preceding points. If we look at the stock market, previous events heavily influence the current prices. Therefore, we can safely say the historical data influences the current data and is thus continuous. We use line plots for visualizing these types of time series. As the slope between two data points can provide information or we can interpolate the data to extract new information.
Another type of time series would be where the data points are uncorrelated. These types of time series are like competitions, world records, or, say, Olympic medals won by a country. For this type of time series, calculating the slope between adjacent data points or interpolating the data will not add any value. As the association is very weak between the historical data. In situations like these, it is best to represent the data using bar charts.
Now the question arises which chart type to choose when working with time-series data. Try answering the following questions. The answers to these questions will lead you to the chart type that should be the best representation of the time series you are after.
— What type of time series are you dealing with?
— Does the present data affected by historical data?
— Will interpolating the data make any sense?
— Is the time interval between the data points small or large?
Answering the above questions will help identify how your data behaves over time and will guide you in choosing the correct plot to visualize the time series.
Time series with related observations
We will use the EuStockMarkets dataset from the inbuilt data provided by R and plot the time series as bar chart and line plot.
The first plot is a bar chart where each bar is a discrete value and this type of plot does not provide any insights about the stock performance between any two consecutive indexes. This is because we cannot interpolate the data, as there is no continuity between the running index.
If we re-plot the above data as a line plot, then the plot gives much more insight in-between two adjacent indexes. Since the index is basically a time unit, so interpolating the data is easier to get hidden insights from the data, which is not possible when working with bar charts.
Time series with unrelated observations
Now we create the same plot types using another data set. Here, we will use the USA Olympics total medal tally (data source: Wikipedia).
We plot the Olympics medal data as a line chart with linear interpolation (red region) to calculate the missing value for the medals won in the year 1980. The reason this chart type is an incorrect choice is the data presented here are unrelated to the previous and future performances of the USA teams. One way to put this is that the contingents for two different Olympics might not have been the same, venues were different, and so did the weather, which would add to the athlete’s performance.
Again, we plot total medals won by Team USA in the Olympics with the help of a bar chart. For the year 1980, we plot no data as the USA did not take part in the Olympics held in Russia. For the following plot, we are looking at how the team performed in each Olympics, so it would be unwise to interpolate the missing value to predict the team performance for the year 1980 as the data is independent of past or future performances.
Concluding remarks
We saw from the above examples that when working with time series with correlated data points; it is a good practice to plot the data as a line plot. If the data is uncorrelated, then bar charts are a better option over line plots.
The best way to identify the chart type to implement is by questioning the data. Understanding the data, what it implements or depicts, can open doors to the correct representation of data with no misrepresentation. It is not just valid for the current scenario, but also applicable to other chart types. Make sure the data presented via visualizations sends the correct message to the audience.
Other relevant articles for data visualization by me:
Six Guidelines for Good Visualizations
Complete guide to animating bar charts
Master data visualization with ggplot2: pie charts, spider plots, and bar plots
Data visualization using ggplot2: 5 features worth knowing
Creating your own color palette for data visualization-Part 2
_You can connect with me on LinkedIn and Twitter to follow my data science and data visualization journey. Or you can follow me on Medium to stay updated on new articles on data visualization with R and ggplot2 package._
