In the past week, my fellow cohort and I have been hard at work reviewing and expanding our knowledge and proficiency of Tableau. Besides covering the fundamentals of the blues and greens, we explored several different chart types and their various use cases. All in all, it was a wonderful week of chart making (even though my ears still ring of mouse clicks!)

In this article, I want to talk about scatter plots, dot plots and jitter plots. In other words, I just want to talk about dots.

At a glance, scatter plots are no different from jitter plots. After all, a chart with little dots is all they are, right? Well, not quite. Scatter plots and jitter plots are more different than they are similar. More still, dot plots are the third wheel of the bunch!

For our purposes, we’ll be using the Superstore dataset. An important consideration to stress at the outset is that all of these graphs use the “product” hierarchy in its entirety as details. That is, every graph featured here uses the “category”, “sub-category”, “manufacturer” and “product name” variables. Simply put, this level of granularity is necessary so that there can be many dots in each of our graphs. If we didn’t have this level of detail, then the graphs would have few dots and we wouldn’t be able to perceive major visual differences between our graphs.

So, how are we to make sense of all of these little dots? Let’s take a look at each of them and work out how they are created, what they are meant to convey and how they differ as visualizations.

The above graph is a scatter plot that is made from having the sum of sales pill on the column shelf and the sum of profit pill in the row shelf. The order date pill, in the form of a discrete year, is also in the row shelf; however this is not strictly necessary.

At its core, scatter plots are graphs that are made by at least two numeric (usually continuous) variables that are being compared against one another. It is important to stress that the vertical component and horizontal component of a scatter plot are important. They both mean something and refer to an actual variable from the dataset.

Scatter plots are used to help us ascertain if there may be any relationships between two or more variables. Generally speaking, we are looking for any noticeable positive or negative correlations, which refer to the general direction of the placement of the dots of the scatter plot.

Moving on, let’s take a look at a jitter plot. At first glance, the jitter plot appears functionally identical to the scatter plot. However, this couldn’t be further from the truth.

In this case, this jitter plot was made by having the sum of sales pill in the column shelf and the order date and an interesting pill in the row self. That interesting pill is based on the random() function. This is incredibly important when creating jitter plots. You see, the “jitter” effect of a jitter plot is nothing more than a random vertical impression placed upon the data that is actually just horizontal. That “random vertical impression” is the result of placing the random() function in the row shelf.

So, what gives? Does this mean that the vertical component of a jitter plot actually signifies nothing important? Yes, exactly. The vertical component could be a range between any random set of values. Whether it is 0 to 1 or 10 to 1,000, there is fundamentally no meaning of the vertical component of a jitter plot.

In that case, what’s the point? Why use a graph that could easily be misunderstood? Well, the purpose of a jitter plot is to convey frequency and distribution in clearer detail. That is, whereas a scatter plot is generally used to convey relationships between at least two variables, a jitter plot is used to help visualize frequencies of a single variable that overlap. This will become immediately clear when we take a look at dot plots.

Finally, we have the dot plot. Right away, we notice two important differences between the dot plot versus the scatter plot and jitter plot. Firstly, there is absolutely no vertical component. Secondly, only a single numeric variable is being conveyed (sum of sales, which is in the column shelf).

With no vertical component, the dot plot conveys only a basic impression of the distribution of values. To be clear, dot plots can be vertical (just switch the pills), but in that case, they would be without a horizontal component. The point here is that dot plots convey frequency and distribution only with one axis whereas jitter plots do this with both axes.

In fact, if you remove the vertical component of a jitter plot, you are left with a dot plot. Quite the shock, I’m sure. In that case, why not just have a dot plot? After all, dot plots are easier to make and may be less prone to confusion or misinterpretation, right?

Well, as mentioned earlier, jitter plots are very helpful when trying to clearly see the spread of many overlapping dots. With a dot plot, you just can’t do that.

At a high level, these three graphs help us to interpret relationships (scatter plots) or frequencies and distributions (jitter plots and dot plots) of granular data. When we want to see highly detailed dynamics of our data, these three graphs are effective and interpretable options. Even more, these graphs can help us to quickly spot any possible outliers, which may call for us to do further analysis on those specific points or ask for clarification from our stakeholders who are responsible for the data that we are analyzing.