Let us start with a quiz! And before you leave: It is about the TV series Friends! The question is: “What was Chandler Muriel Bing’s first job in the first season of Friends?” I know, it is an easy question. For all of those who may not remember the exact wording, it is: “Statistical Analysis and Data Reconfiguration”. I have to admit that as a die hard data nerd, this description is not satisfactory… At all!
Is he creating models for predictive purposes? Does he do Data Cleansing? Is correlation and causation important? Do optimization procedures play a role? I have to ask again: Data Cleansing?! Many questions and I will probably never receive an answer…
Oh well, at least I can seize this opportunity to write a post about Data Analytics, the different types thereof, and why you should never enforce an Ascendancy Model upon your Data Experts. Unless you want your Data Experts to quit. Then just go right ahead.
You thought this post was about Friends? Ahm, well… Look over there!
What Is Data Analytics?
Let us begin with a definition of Data Analytics, in order to be on the same page. Or line… You know what I mean. John Tukey defines Data Analytics as the following:
“Procedures for analyzing data, techniques for interpreting the results of such procedures, ways of planning the gathering of data to make its analysis easier, more precise or more accurate, and all the machinery and results of (mathematical) statistics which apply to analyzing data.”
John Tukey
Okay, let us analyze his definition one by one. The first part – “Procedures for analyzing data…” – refers to concepts such as the mean, median, and standard deviation. The second part – “… techniques for interpreting the results of such procedures…” – uses the aforementioned concepts to interpret them by e.g. generating a Box Plot. With those two parts, we would already be able to do Data Analytics. How well you ask? Let us have a look at the other two parts.
The third part – “… ways of planning the gathering of data to make its analysis easier, more precise or more accurate …” – references towards Data Collection. The term Big Data should come to your mind immediately. The idea behind it is as simple as it is ingenious: the more data you have, the better you and/or a model, can determine/predict the underlying distribution of the data.
See it this way: Data follows a certain distribution. The more data you have, the more likely you and/or a model can determine this distribution. By doing so it becomes easier to understand whether an outlier really is an outlier, or if your current dataset is simply skewed and therefore your outlier could actually be the median of the real distribution of the data.
Of course there is more nuance to Data Collection and Distribution but for the time being… look over there!
Part Four! “… and all the machinery and results of (mathematical) statistics which apply to analyzing data.” is more or less a wild card. Anything that was and will ever be developed for data analysis, fits here. This basically allows us to use concepts such as Machine Learning, Decision Analytics, and Business Intelligence (that one again…) to improve our procedures and make more diverse analyses.
This leads us straight to our next part of the post:
What Types of Data Analytics Are There?
Many different forms of Data Analytics exist: Big Data Analysis, Exploratory Data Analysis, Text Data Analysis, Confirmatory Data Analysis, and many many more. But we will not look into these. Though each of the previously mentioned Analyses have a certain purpose, we will only cover the big four:
From left to right:
Descriptive Analytics
Descriptive Analytics can be described (pun intended) as the interpretation of historical data to (better) understand past events. It uses statistical tools such as mean, median, standard deviation, and distribution to add context to numbers and metrics. Possible examples are year-over-year pricing changes, fluctuation of the number of users, and total revenue per subscriber.
Diagnostic Analytics
The Sherlock Holmes of Data Analytics. This type is applied to understand why certain events happened and Correlation, as well as, Causation are the two heavy lifters here. Due to Causation, it (still?) requires human intelligence. For a short and easy introduction, have a look at this video.
Predictive Analytics
This is one of two proactive Data Analytics approaches. It takes historical and current data to look for patterns to determine the likelihood of recurring events. This is the part of this post, where Machine Learning enters. There are many well known and studied use-cases (Wine Prediction, Customer Churn Rate, etc.) which apply Predictive Analytics, and because one could fill entire books (this and that) about this topic, I will leave it at that.
Prescriptive Analytics
Now let us have a look at the crystal ball of Analysis. I really hate to say it and Investopia has a very nice definition of Prescriptive Analytics:
“Prescriptive Analytics factors information about possible situations or scenarios, available resources, past performance, and current performance, and suggests a course of action or strategy.”
Investopia
What the above quote states is basically what we discussed about Decision Intelligence. We use different (scientific) fields to determine and analyse different situations, in order to make decisions that are as beneficial as possible.
For the sake of completeness I must address two issues:
- Is it Fortune Telling?
- Do we really need Prescriptive Analytics? Is it not similar, if not even identical, to Predictive Analytics?
Issue number one: That is an easy one. If you show me someone who can clearly predict the future, I show you a liar.
Issue number two: Yes, this controversy exists and I highly recommend reading this post. My overall stance, which will be discussed in the third part of this post, would be that a strict separation of the Data Analytics types is futile and can potentially lead to worse outcomes. That being said here a short summary of the questions Data Analytics can answer, if used wisely:
- What happened? -> Descriptive Analytics
- Why did it happen? -> Diagnostic Analytics
- What is likely to happen? -> Predictive Analytics
- What do I need to do? -> Prescriptive Analytics
Next: why the difference should not matter!
Why Difference Should Not Matter?
Finally, the good stuff begins!
After reading the previously discussed definitions of the four types of Data Analytics, you might have a certain Data Analytics approach in mind: We start with Descriptive Analytics, work our way through Diagnostic and Predictive Analytics and finally reach Prescriptive Analytics aka the end boss.
For the ones who like graphs (yeah, me too), Gartner has something prepared (which I may have pimped up a “tiny” bit):
Let us have a look at Figure 2. The y-axis represents Value and the x-axis Difficulty. So far so good and unfortunately, not really helpful. The arrow in the middle, which should indicate a continuous growth rate (?), brings more “-sight” into the image description. From the bottom left to the top right side of the arrow we have the following words listed: Information, Hindsight, Insight, Foresight and Optimization. These words immediately let us think about the four types of Data Analytics and of course they are also mentioned (with their respective question).
Now the interpretation: Figure 2 allows us to make two main conclusions (at least these are the two I am aiming at):
The first conclusion would be that Descriptive Analytics is the least difficult type of analysis and also provides the least value; whereas Prescriptive Analytics is the most difficult one, which is also providing the most value.
The second conclusion is that each type of analysis builds upon the information gained from the one before. We use data/information to gain hindsight, which is then transformed into insight, processed into foresight to ultimately optimize our decision making process.
In theory, both conclusions make sense. Unfortunately, theory and the real world, are two different things. The criticism is that Ascendancy and Maturity Models indicate project advancements via levels, which, usually, is not the case. If you now think that this sounds awfully familiar like project management, you are absolutely right! The boundaries of the different analysis types are fuzzy and one might need different information at different points of the project.
With this being said one possible option would be to change the point of view of how we perceive Data Analytics. Entering the Venn Diagram:
So, what is going on in Figure 3? We have three bubbles: Data, Insight, and Forecasting. Descriptive Analytics only references Data, whereas Diagnostic Analytics is the combination of Data and Insight. Predictive Analytics utilizes Data and Forecasting, and the final type, Prescriptive Analytics, uses all three bubbles. Do not get me wrong, it is not the solution to all problems. It is simply a nice shift of perspective, allowing us to take a step back, not following a strict path, and simply thinking about what we currently want/need to know to take the next step in our project.
If you are interested in more information, please have a look at these two post:
- How Analytics Maturity Models are Stunting Data Science Teams
- 4 Stages of Data Analytics Maturity: Challenging Gartner’s Model
Now let us have a look at the final part of this post:
What Comes Next?
There are two reasons why I wrote this post. Firstly, to give an overview of the different Data Analytics capabilities. Secondly, to show that Data Analytics project do not have a “one size fits all”-formula; Otherwise, I so daring disclaim, a majority of Data Science practitioners would loose interest. Hard!
With this being said, what are potential next steps? If you are a project manager, or manager in general, who is interested in Data and Decision Science, get yourself emerged in how to best encourage and challenge your teammates. If you are a teammate, do the same thing. Without wanting to sound too cliche (I know, I failed that one already at the beginning of the post): read a book, or two.
This leads to the end of this post. If you liked this post, didn’t like it, and/or have a book recommendation, please leave a comment, subscribe to the blog, and share the post. I am always looking for ways to improve myself and learn new things.
DISCLAIMER: The visualizations presented in this graph do NOT represent quantitative data. They are merely used for argumentation purposes and therefore are not scientifically sound! For more information & a good laugh, have a look at the book Calling Bullshit by Bergstrom & West, Chapter 7: Data Visualization.
