Review: Tableau makes sophisticated analysis a snap

Data science is best left to Ph.D. statisticians who can program in R and Python, compose complex SQL and MDX queries in their sleep, and leap tall Hadoop data sets in a single bound. Right?

Not according to Tableau. The company claims its products make analytics easy not only for analysts but for “executives, IT, everyone.” While my own training (a doctorate in physics with lots of statistics, SQL, and programming experience) is more than adequate for conventional data science, I tested Tableau 9.0, hot off the presses, with a “beginner’s mind.”

Tableau is considered the market leader in the BI and analytics space, having usurped Qlik, which in turn displaced Cognos (now IBM Cognos) and the other first-generation enterprise BI tools. Tableau is a prime exemplar of the business-user-driven data discovery and interactive analysis trend in BI that has largely taken over from traditional IT-driven reporting and analytics.

New in Tableau 9.0

If you’re already familiar with Tableau, you might want to know what’s new and different in the latest version. The two big areas of improvement:

1. The tool is smarter about what you are doing.
2. It’s faster to process data and show you analyses.

In the area of “smarter,” Tableau 9.0 has a much better start experience, both on the desktop and on the server, with easy access to your workbooks, data connections, training, and shared visualizations (Figure 1). It offers stories, which I will discuss later on, and it preview thumbnails for your sheets, so you can insert the right sheet into a dashboard or story. It adds ways to easily create analytics, zoom into your data, and create level of detail expressions. Smart maps include geographic search and census data. Data preparation does more with whatever format you happen to have in your data source, without requiring you to reformat your spreadsheets.

In the area of “faster,” Tableau 9.0 consolidates queries and aggregates in parallel; sends the queries for all independent views to the database in parallel; fuses multiple queries at the same level of detail into a single query; and caches query results to avoid rerunning them in the same session. The difference is night and day: 20 seconds versus five minutes for a moderate-sized project.

Choose your data source

Tableau Professional can connect to a wide assortment of file (Figure 2) and server data sources, including Excel workbooks, character- and tab-delimited files, statistical files, and upward of 40 server types, although 19 of those are only available from Windows. Tableau Personal is restricted to six kinds of data source; the free Tableau Public can only use four kinds of data source.

You can connect multiple data sources to a worksheet and create joins between tables and/or files. If you know the joined data has referential integrity, you can improve performance by telling Tableau to assume referential integrity.

It’s very common for raw data to be full of nulls, to have fields (especially name, date, and geographical fields) that aren’t quite in the right format for analysis, and (along the lines of the toast always falling jelly side down) to have the rows and columns reversed from where you need them. It’s also common for there to be a big title and subtitle in a spreadsheet that has been used for a presentation, which can mess up the usual assumption that the table starts with a row of column titles, then a block of actual data.

Tableau 9 handles all of that easily. It’s no problem to turn addresses into a hierarchy of country, state, city, and street address, and Tableau can infer latitude and longitude from addresses. It’s also easy to skip over any irrelevant material before the real table without going back to Excel and to pivot the rows and columns at any time during the analysis in Tableau.

Powerful analytics

Analysis, specifically ad-hoc analysis, is an area where Tableau shines. Once you have imported data, you can explore it using as many views into the data as you like.

Tableau analysis is a drag-and-drop process with property sheets, kind of like a Visual Basic for data scientists. As we can see in Figure 3, the data dimensions (fields used for classification) and measures (data values both primary and calculated) appear in the tab to the left. You can drag them into rows and columns, attributes and filters. By the way, this chart shows Case-Shiller home price index data; the geocoding and the U.S. map background were all generated by Tableau without any help from me.

The floating Show Me palette seen at the lower right has hints about what measures and dimensions each kind of display requires. If you add more rows and columns than required for individual charts, Tableau will automatically create a grid of smaller charts for you.

If you’d prefer letting the viewer select one or more parameters for exploratory purposes — for example, to see the evolution of a trend over time — you can add a quick filter with a user control, such as the Year slider at the upper right. To experience the power of this, look at the “See how your home town compares” sheet of the CNBC Recovery Watch story, drag the date slider back to January 2011, and click the right arrow to step through the housing recovery month by month.

Every feature of Tableau has additional options, though the default settings are often pretty good. For example, the size of bubbles can be controlled in the Edit Sizes dialog that comes up when you double-click the bubble card, below the marks and colors cards to the left of the chart in Figure 3. In this particular case, the default was too small and uniform for the data and my taste, so I enlarged the bubbles and widened the range of sizes to be more visible. I couldn’t decide whether the mean HPI (home price index) or the median HPI was more meaningful for this particular chart, so I assigned one to size and one to color.

Color, size, and shape give you the ability to represent extra dimensions and measures on a chart in addition to the row and column measures. You can also do a lot with actions and tool tips.

Tableau has had the ability to do calculations for a long time; version 9 adds ad-hoc calculations, which make it easy to add and edit calculated fields in your analysis. Tableau expressions look a lot like Excel formulas, except they use field names rather than cell names. Tableau formulas will conveniently autocomplete as you type them.

The new Analytics pane lets you easily drag summary, model, and reference lines onto a worksheet. Exactly what analytics you can use depends on the chart type.

Flexible presentation

Tableau organizes your analyses into worksheets, dashboards, and stories. Dashboards can contain an arrangement of multiple worksheets, and you can create actions for one worksheet that affect other worksheets in the dashboard, making it easy to, for example, see the data for a particular city over time, as in the “See what’s happening in your home town” dashboard of the CNBC Recovery Watch story I mentioned earlier.

Stories, as you saw, combine multiple dashboards and/or worksheets into a narrative that illustrates a point of analysis. Another good example of a story is Mapping the 1854 Cholera Outbreak, which reproduces in Tableau how John Snow manually analyzed the Soho, London, outbreak; isolated a particular source of bad water as the source of the disease; proved that cholera is water-borne; and ended the epidemic by shutting down one contaminated water pump.

Sharing analyses

You’ve been viewing Tableau stories on Tableau Public, which as you’ve seen displays the analyses in a browser. Anyone can sign up for Tableau Public for free and download the Tableau Public app for Windows or Mac to create other analyses. That version of the app can only save analyses to the Tableau Public server. (Publishing Tableau Public analyses used to be restricted to press.)

Tableau Server is a private version of the system you used. It runs on Windows, has an administration system to control sharing, and can work with all the same data sources as Tableau Desktop. Tableau Online is a hosted version of Tableau Server.

If you want to share analyses privately but don’t have access to Tableau Server or Tableau Online, you can export a “packaged” workbook from Tableau Desktop, send it to your colleagues, and have them open it with the free Tableau Reader. When creating the packaged workbook, you can control how much can be done with it from Tableau Reader. For example, you might not want Tableau Reader users to be able to drill down to personal identities.

Learning Tableau

Tableau supplies sample data, videos, quick starts, live classes, and webinars to help people get up to speed on the product. All versions of Tableau are available as free trials.

While it’s very easy to learn the superficial stuff in Tableau and many people can become productive with it in half an hour, it’s harder to learn features that are beneath the surface unless you happen to have seen them in a training video or know their names so that you can search the help. When those don’t apply, poking around Tableau can feel a little like playing a Nancy Drew game.

Case in point: It took me what felt like forever to discover that the interactive control over which subset of the data you display in a worksheet is called a Quick Filter. Once I knew that, I found at least three ways to create a Quick Filter, including the context menu item you can see in Figure 3.

Tableau 9.0 has an excellent assortment of data sources ranging from Excel, text, and statistical files to database servers, cloud data warehouses, various flavors of Hadoop, and systems of record such as Salesforce. Importing data is easy, as are data cleaning, inline data transformations, and construction of data joins.

Tableau’s selection of chart types is very good, and it provides excellent control over chart appearance. It also gives the analyst easy ways to display multiple dimensions and measures. Tableau’s maps, dashboards, and stories help the analyst explain the logic leading to a conclusion, and parameterized displays controlled by widgets allow the viewer to play along.

Tableau makes deep statistics available without writing code, though you can do even better if you do write code, especially R code. Overall it’s not hard to learn Tableau at a basic level. Learning the finer points will take some time and patience, however.

Reporting is an area where some organizations find Tableau lacking. Tableau does a great job of self-service reporting and analysis, but doesn’t always meet the needs of IT departments for production line-of-business reports. Some organizations that have adopted Tableau find they have to buy and maintain a second BI product for these needs, either because they find Tableau’s data preparation inadequate for scheduled automatic reports, or because they want a product that can do more sophisticated statistical analyses on a routine basis, and alert management appropriately when guidelines have not been met.

Price can also be an issue for Tableau customers. With competitors such as Microsoft and Qlik offering free options, it can be difficult to justify buying Tableau analysis capabilities for a whole company, although some companies have done exactly that. That leaves the company with a trade-off: Give Tableau only to the people who absolutely need it to stay within budget, or buy a slightly less capable and much less expensive product for everyone.