SPSS stands for Statistical Package for the Social Sciences, but the cool thing is that IBM will let anyone from any business use it. Pretty slick, eh? What is it? Thought you’d never ask.
But Watson Studio doesn’t do everything, and some additional functionality may be required for you to get fully invested into your AI project. You can get some of what you need from the IBM SPSS Statistics package.
SPSS Statistics: What Is It?
The SPSS “line” actually consists of two entities. The first is the SPSS Statistical package, which we will discuss in this article. The second is the SPSS Modeler, which is used in close conjunction with Watson Studio to help set up and evaluate the model your machine-learning project is using. We’ll talk about that software next month.
Modeler is very interesting, but in terms of raw excitement what could possibly compare with a statistical package? And SPSS Statistics is no exception. As IBM says on its own website, “Quickly dig deeper into your data, with a much more effective tool than spreadsheets, databases, or other standard multi-dimensional tools. Use a wide range of advanced statistical analysis, 130+ extensions that offer seamless integration with RStudio, Python and more.”
Or, in normal English, AI is not just about setting up a model, training it on a set of data, and then running it on another set of data to see what the predictions are. Maybe the most important step is to analyze the predictions that we create and see how close they come to reality. This is typically done via some sort of statistical package.
And that is what SPSS Statistical provides.
But these are programming-oriented packages. You use them with Jupyter Notebook or another IDE, and they’re generally wrapped into a Python or R script. They’re pretty easy to get used to, but they are, after all, basically a language.
Plus, you sort of have to know where the various tools that you want to use live. For example, want to create a data frame with your data? That’s in pandas. And Support Vector Machines is in scikit-learn, but did you know that? Not a big deal; you can always Google to see what library a given statistical technique is in, but it still does have kind of a technical spin to it.
The SPSS Statistics package, however, like many IBM products, is point-and-click oriented. The basic format looks more like Excel than a Jupyter Notebook, and there are many built-in functions within that tool that let you better define your data and even run simulations on it.
But if it is point-and-click, then it must be kind of lightweight, right? Not necessarily.
SPSS Statistics Highlights
On the surface, yes, you might be tempted to consider SPSS Statistics to be somewhat lightweight. You know, the IBM equivalent to Microsoft’s Power BI product. But that’s not really fair to Statistics. It gives you, in a point-and-click environment, the ability to do data access, manage data from a variety of sources (including clusters), easily perform linear regression (a basic ML analysis tool), and other statistical “things”… all with integrated charts and graphs.
Unlike pandas or scikit-learn, the visualization features (graphs, pie charts, etc.) are directly incorporated into the model, so there is no need to write scripts to perform those functions. Those visualizations can then be moved over into a professional-looking document for publication to your team. Or the results can be downloaded into Word, PowerPoint, Excel 2016, or even IBM Cognos.
In addition, you can add in specialty routines in Python, R, .Net, or Java to make the results more specific to your data and your project. SPSS Statistics provides pivot table–like functionality, and you can even perform Monte Carlo simulations on the data.
Is That All There Is?
Why unfortunately? Aren’t business decisions already hard enough to make? Just once I want to see a choice between something really, really good and something dreadfully, dreadfully awful.
But the base Statistics package is just the starting point. To it, you can add three add-on packages.
The first is the Custom Tables and Advanced Statistics enhancement. The big takeaway here is that you will be able to do more-advanced statistical testing including 2 stage Least Squares Regression, Bayesian statistics, mixed logical models, logistic regression, loglinear and multivariate analysis, nonlinear and quantile regression, survival analysis, and other methods.
The second add-on is the Complex Sampling and Testing enhancement. This adds a number of abilities, including being able to find missing values (columns in a row that are not filled in), Decision Trees, Conjoint Analysis, Time Series Analysis, and that sort of thing. Most important is that this provides the ability to do Neural Networks. Now, does this mean that you need the optional SPSS Statistics Package to work with Neural Networks? The answer is, I don’t know. I couldn’t get a suitable answer from IBM. My guess is no, you don’t need to get the SPSS package to use Neural Networks in Watson Studio, but it probably makes it easier. But that’s just a guess (my current correct guess percentage is almost 40 percent, so you make the call there).
The third add-on is Forecasting and Decision Trees. Strangely enough, though, when you look at the details for what this add-on does, there’s no mention of Decision Trees at all. There is a lot of stuff there: ARIMA, CHAID, QUEST, spectral analysis (which I assume has something to do with the paranormal), direct marketing analysis, seasonal decomposition, and others.
For a full list of what you get with the base package and the add-ons, see this.
Pricing is always interesting, but for some reason IBM has made pricing on this item very straightforward. The base package is $99 per user per month. Each add-on is an additional $79 per user per month.
So that is the Statistical Package part of SPSS. But what about the Modeler part?
Patience. That is another story for another time. Like next month.