visual analytics best practices

45
Visual Analytics Best Practices Sasha Pasulka Senior Manager, Product Marketing [email protected] #tableau8

Post on 17-Oct-2014

11.135 views

Category:

Technology


0 download

DESCRIPTION

This slideshow provides an overview for best practices for visual analysis within Tableau. This is intended for anyone who wants to tell more compelling stories with their data.

TRANSCRIPT

Slide 1

Visual Analytics Best Practices

Sasha PasulkaSenior Manager, Product [email protected] #tableau8

I like to start these sessions off with a game I call count the nines. So, count the nines. Raise your hand when you think you know how many nines there are on here. [pause] In fact, just go ahead and shout it out. If you know how many nines are on here, shout out the answer. [pause]2

NOW count the nines. How many nines do you see? Raise your hand when you know. 3What is Visual Analytics?So what is visual analytics? THATs visual analytics.4Visual analytics is the representation and presentation of data that exploits our visual perception abilities in order to amplify cognition. - Andy Kirk, author of Data Visualization: a successful design process

Those are really fancy words for what you just experienced. Humans can see visual patterns very well, but only when the patterns really play to a humans strengths. 5Lets Look at Some DataIIIIIIIVxyxyxyxy108.04109.14107.4686.5886.9588.1486.7785.76137.58138.741312.7487.7198.8198.7797.1188.84118.33119.26117.8188.47149.96148.1148.8487.0467.2466.1366.0885.2544.2643.145.391912.51210.84129.13128.1585.5674.8277.2676.4287.9155.6854.7455.7386.89Lets see another example. Heres some data. The are 4 sets of data here, each with 11 sets of x-y coordinates.

For the purposes of this exercise, lets assume the x data represents, in millions, the net sales of a single retail store over the course of a month. Lets say the y data represents, in millions, the total profit from that store. So were looking here at a set of points that represent profit by sales, where each point is a single store. The four data sets represent regions, say, West, Central, South and East.

Lets say youre a manager responsible for maximizing profit at these stores.

Whats your move? [pause for 10 seconds, let people try to say smart things ]6IIIIIIIVxyxyxyxy108.04109.14107.4686.5886.9588.1486.7785.76137.58138.741312.7487.7198.8198.7797.1188.84118.33119.26117.8188.47149.96148.1148.8487.0467.2466.1366.0885.2544.2643.145.391912.51210.84129.13128.1585.5674.8277.2676.4287.9155.6854.7455.7386.89Lets Look at Some DataPropertyValueMeanofxin each case9(exact)Varianceofxin each case11(exact)Mean ofyin each case7.50(to 2 decimal places)Variance ofyin each case4.122 or 4.127(to 3 decimal places)Correlationbetweenxandyin each case0.816(to 3 decimal places)Linear regressionline in each casey=3.00+0.500x(to 2 and 3 decimal places, respectively)OK, OK, so youd typically have a bit more information than that when youre making a decision. So now lets look at some more information about these data sets. Maybe we can learn something about them from their means, or their variances. When were crunching numbers, we rely a lot on things like means and variances. And probably looking at correlation or doing a linear regression would help, too. It turns out that these four data sets all have the same means, the same variances, the same x-y correlations, and even boil down to an identical linear regression.

So whats your move? [Pause for 10 seconds or so]

7Lets Look at Some Data Visually

Anscombes QuartetSource: WikipediaHere are these same four data sets, plotted visually, with trend lines.

Now, whats your move? [Let the audience make some suggestions. You can chime in with things like, Yeah, you might want to talk to the manager of the outlier in set 3 and see what shes doing right or You might want to talk to the managers of some of the stores in set 4 and see why their profits are underperforming compared to stores with similar sales.]

What other pieces of information might you want? [Let them make suggestions, and if necessary you can chime in with things like You might want to see how many orders each store is producing, or what categories of product theyre selling most, or how frequently they offer discounts.]

It would be nice to be able to encode some of that information on these graphs, like maybe have a larger circle for stores that offer, on average, larger discounts, or to be able to quickly split this data up to show sales by product category by store. Like, maybe with just one click. And then it would be nice to be able to share this view with your individual store managers, with just a couple of clicks. And it would be nice to have that view you shared update with real-time data, so those store managers could see day-by-day how their stores were performing compared to their peers, and interact with that live data to understand why their stores are succeeding or lagging. So that they are each empowered to explore the information they need to meet and exceed their profit goals.

Thats what Tableau does.

8AgendaHuman Perception and CognitionVisual Analysis CycleVisualization Best PracticesHuman Perception & CognitionHumans Are Slow at Mental Math 34 X 72------------------Humans are slow at mental math. Were not designed to manipulate complex numbers in our heads. Go ahead and try to solve this multiplication problem in your head. 11Were Faster When We Use the World 34 X 72------------------ 6823180------------------2448If we give you a pencil and paper, all of a sudden this problem becomes a lot easier to solve. How much easier? 12Much Faster 34 X 72------------------ 6823180------------------2448

About 5 times easier. It takes educated people an average of 50 seconds to solve this problem in their head. Give them the right tools, and suddenly its solved in just under 10 seconds. 13Were Faster When We Can See Data

Which product subcategory is the most unprofitable? 14Were Faster When We Can See Data

Which product subcategory is the most unprofitable? 15Were Faster When We Can See Data

Now which product subcategory is the most unprofitable? 16

Preattentive Visual AttributesPreattentive attributes are information we can process visually almost immediately, before sending the information to the attention processing parts of our brain. This is information we process and understand almost unconsciously. These are generally the best ways to present data, because we can see these patterns without thinking too hard.

If you have time/Internet/sound, show this video: http://www.youtube.com/watch?feature=player_detailpage&v=wnvoZxe95bo17Visual Interruptions Make People Slow

Show Jocks interruption video18Visual Interruptions Make People Slow

Show Jocks interruption video. If you have time and an internet connection, this YouTube video is great, too: http://www.youtube.com/watch?feature=player_detailpage&v=vBPG_OBgTWg19The Cycle of Visual AnalysisThe Cycle of Visual Analysis

Visual analysis isnt just looking at a chart, or using colors its an entire lifecycle that includes identifying and getting your data, establishing the structure of that data, choosing the best way to visualize that data, drawing conclusions or insight from those visualizations, and then getting buy-in around any conclusions supported by that data, which means you have to be able to tell a compelling story succinctly. And to help a team benefit from visual analysis, you need to support the whole cycle of visual analysis. 21Supporting the CycleIncremental: allow people to easily and incrementally change the data and how they are looking at it

Expressive: there is no single view for all tasks and all data

Unified: leverage the revolutionary changes in database technology

Direct: make the tool disappear so the user can directly interact with the data

clickclick22Visualization Best PracticesBest Practices OverviewRepresenting data for humansColorMapsCreating dashboards

24Types of DataQualitative (nominal / categorical)Arizona, New York, TexasSarah, John, MariaCoors, Bud Light, Stella ArtoisQualitative (ordinal)Gold, silver, bronzeExcellent health, good health, poor healthLove it, like it, hate it QuantitativeWeight (10 lbs, 20 lbs, 5000 lbs)Cost ($50, $100, $0.05)Discount (5%, 10%, 12.8%)

25How Do Humans Like Their Data?

26How Do Humans Like Their Data? PositionColorSizeShapeMore importantLessimportant27How Do Humans Like Their Data?

Time: on an x-axisLocation: on a mapComparing values: bar chartExploring relationships: scatter plotRelative proportions: treemap

28How Do Humans Like Their Data? Orient data so people can read it easily

BetterGood29Color Me ImpressedColor perception is relative, not absolute

30Color Me ImpressedProvide a consistent background

31Color Me ImpressedHumans can only distinguish ~8 colors

This is not helpful.32Color Me ImpressedHumans can only distinguish ~8 colorsThis is helpful.

33Color Me ImpressedFor quantitative data, color intensity and diverging color palettes work well

34Mapping to InsightUse maps when location is relevant

35Mapping to InsightUse filled maps (cloropleths) for defined areas and only ONE measure

36Mapping to InsightFilled maps wont work for multiple measures

37Mapping to InsightDont use maps just because you can

38Mapping to InsightMaps dont have to be geographic

39Mapping to InsightMaps dont have to be geographic

40DashboardsDashboards bring together multiple views

41DashboardsDashboards should pass the 5-second test

42Dashboarding for the 5-second TestMost important view goes on top or top-leftLegends go near their viewsAvoid using multiple color schemes on a single dashboardUse 5 views or fewer in dashboardsProvide interactivity

43Dashboarding for the 5-second TestUse your words!TitlesAxesKey facts and figuresUnitsRemove extra digits in numbersGreat tooltips

44

The key words are are see, understand, and people. Tableau builds software for people, not specialists. We believe anyone should be able to harness the power of data. Thats our mission. 45