EDUCBA.com provides the following distinction between “Descriptive Analytics” and “Predictive Analytics”:
Descriptive Analytics: This type of analytics is used to summarize or turn data into relevant information. In other words, it summarized what has occurred. This type of analytics has some meaningful impact but won’t be much helpful in forecasting.
Predictive Analytics: – Predictive analytics involves advanced statistical, modeling, data mining and one or more machine learning techniques to dig into data and allows analysts to make predictions. Predictive analytics is used to forecast what will happen in future.[i]
When I read this description, it reminded me of newly-minted managers, with the memories of their BBA or MBA graduation ceremonies fresh in their minds, arriving at their first assignments. It’s not that they’re ill-informed or anything – it’s just that what they think is new and innovative has actually been around for some time. In this case, the Project Management community has been on top of this epistemological mountain for so long, it’s easy to forget that we’re here, and the
whippersnappers newcomers need to be reminded of that.
Let’s look at a couple of key sentences from the above, shall we? According to this author, Descriptive Analytics describes “what has occurred,” and “…won’t be much helpful (sic) in forecasting.”[ii] Well, I disagree. But before I prove that I’m right, let’s take in what this post says about Predictive Analytics, that it “allows analysts to make predictions” (as if such analysts need permission to predict anything). Also, it’s “used to forecast what will happen in the future.”[iii]
Lots of stuff to unpack here. I’ll start with the distinction between these two types of Analytics provided, that “Descriptive” refers to “what has occurred,” which veteran members of GTIM Nation call “hard data,” and “Prescriptive,” which “involves advanced statistical, modeling, data mining and one or more machine learning techniques to dig into data” – in other words, some subjective data is being injected into the number-crunching fray. But is historical, or “Descriptive” data of little use(ful) in forecasting? Consider this blog from a few weeks back, where I proposed a competition between risk management and Earned Value, to identify which tasks or projects were likely to overrun or come in late. I proposed that the risk managers create a list of such tasks, and the EV specialists do the same. After tossing out those tasks that both analysts believe will overrun or come in late, we would have a hard basis of comparison, and that my money would be on the EV peeps, who can reliably predict which tasks will get in trouble, and on a consistent basis. Not only that, but they DO NOT use ”advanced statistical, modeling, data mining and one or more machine learning techniques.” Quite the contrary, they need only the size of the original budget, cumulative actuals, the tasks’ start date, today’s date, and an estimate of the tasks’ percent complete. And that’s it. It’s accurate (to within 10 points on a consistent basis), simple, and it’s been around since the 1960s. No, I’m not kidding.
Well, what of the
whippersnappers’ more recent techniques? They essentially fall into three categories:
- Risk management involves estimating the odds and impact of alternate scenarios,
- Regression bases their predictions on patterns that have occurred historically, and assume that similar outcomes will unfold when analogous situations or conditions are re-encountered, and
- Models, in which certain types of work or business activities are expected to unfold along the lines of a familiar template – Game Theory falls into this category.
Referencing the first bullet, the tried-and-true PM techniques of Earned Value and Critical Path Methodologies will always out-perform the risk analysts’ tactics. For this reason, I’m very confident that no serious fact-based rebuttal to my Variance at Completion challenge will be forthcoming.
As for the second bullet, it is the basis for much of the poison pixel-ink I’ve spilled on our friends, the Accountants. You see, when an accountant projects how much a specific task will end up costing when it’s complete, they typically use regression to make their estimates. They calculate how much the task has been spending, and assume it will continue to spend at that rate (or an average, or adjusted average, or whatever) for its duration. This calculation, while not as irrelevant as the risk managers’ output, still isn’t nearly as accurate at the EV technique.
About the third bullet – isn’t there a ProjectManagement.com blog about that?
So, it’s pretty plain to me that:
- Contrary to the citation, using data that covers things that have already occurred is, in fact, extremely valuable in predicting outcomes, given the right methodology;
- Statistics and modelling provide inaccurate and irrelevant information (“You have a 39% chance of overrunning”);
- Regression-based analysis, while relevant, isn’t nearly as accurate as the 1960s-era technique of calculating the Estimate/Duration at Completion, and
- Game Theory, with its payoff grids and Nash Equilibrium, can provide some insights in very specific, controlled circumstances; however, it has only a fraction of the predictive analytic power of even the most basic Earned Value or Critical Path Methodology-based system. (I’m comfortable making this statement a priori due to the massive amount of research I did while writing my second book, Game Theory in Management).
I don’t mean to come across like some curmudgeon (I should probably knock off the references to whippersnappers). It’s just that, when it comes to reliably predicting key Project Management data points, well, those techniques have been around for some time.
Before many of the whippersnappers were born, even.
[i] Retrieved from https://www.educba.com/data-analytics-vs-predictive-analytics/ on June 29, 2019, 21:38 MDT.