sas whitepaper data visualization techniques

8/11/2019 SAS Whitepaper Data Visualization Techniques

1/20


2/20

ContentsIntroduction ........................................................................1

Generating the Best Visualizations for Your Data ....... 1

The Basics: Charting 101 ................................................. 2

Line Graphs .............................................................................2

Bar Charts ................................................................................3

Scatter Plots .............................................................................4

Bubble Plots A Scatter Plot Variation ...............................4

Pie Charts .................................................................................6

Visualizing Big Data ...............................................................7

Large Data Volumes ..............................................................8

Different Varieties of Data (Semistructuredand Unstructured) ................................................................10

Visualization Velocity ...........................................................10

Filtering Big Data .................................................................12

Data Visualization Made Easy with Autocharting ..........12

Decision Trees .....................................................................14

Can You See Into the Future? .......................................15

Mobile ...............................................................................17

Conclusion ........................................................................17


3/20

1

IntroductionA picture is worth a thousand words especially when you aretrying to understand and gain insights from data. It is particularlyrelevant when you are trying to nd relationships amonghundreds, or even thousands, of variables to determinetheir relative importance.

Organizations of all types and sizes generate data each minute,hour and day. Everyone including executives, departmentaldecision makers, call center workers and employees onproduction lines hopes to learn things from collected datathat can help them make better decisions, take smarter actionsand operate more efficiently.

Regardless of how much data you have, one of the best ways

to discern important relationships is through advanced analysisand high-performance data visualization. If sophisticatedanalyses can be performed quickly, even immediately, andresults presented in ways that showcase patterns and allowquerying and exploration, people across all levels in yourorganization can make faster, more effective decisions.

To create meaningful visuals of your data, there are some basicsyou should consider. Data size and column composition play animportant role when selecting graphs to represent your data.This paper discusses some of the basic issues concerning datavisualization and provides suggestions for addressing thoseissues. In addition, big data brings a unique set of challengesfor creating visualizations. This paper covers some of thosechallenges and potential solutions as well.

If you are working with massive amounts of data, one challengeis how to display results of data exploration and analysis in away that is not overwhelming. You may need a new way to lookat the data one that collapses and condenses the results in anintuitive fashion but still displays graphs and charts that decisionmakers are accustomed to seeing. And, in todays on-the-gosociety, you may also need to make the results available quickly

via mobile devices, and provide users with the ability to easilyexplore data on their own in real time.

SAS Visual Analytics is a business intelligence solutionthat uses intelligent autocharting to help business analysts andnontechnical users visualize data. It creates the best possiblevisual based on the data that is selected. The visualizations makeit easy to see patterns and trends and identify opportunities forfurther analysis.

The heart and soul of SAS Visual Analytics is the SAS LASR Analytic Server, which can execute and accelerate analyticcomputations through in-memory processing. Thecombination of high-performance analytics and an easy-to-usedata exploration interface enables different types of usersto create and interact with graphs so they can understandand derive value from their data faster than ever. This createsan unprecedented ability to solve difficult problems,improve business performance and mitigate risk rapidlyand condently.

Generating the BestVisualizations for Your DataThere are a few basic concepts that can help you generate the

best visuals for displaying your data: Understand the data you are trying to visualize, including

its size and cardinality. Determine what you are trying to visualize and what kind

of information you want to communicate. Know your audience and understand how it processes

visual information. Use a visual that conveys the information in the best and

simplest form for your audience.

What Is Data Cardinality?

Cardinality is the uniqueness of data valuescontained in a column. High cardinalitymeans there is a large percentage of uniquevalues (e.g., bank account numbers,because each item should be unique). Lowcardinality means a column of data contains

a large percentage of repeat values (such asa gender column).


4/20

2

The Basics: Charting 101Here is a quick guide to help you decide which chart type(or graph) to use for your data.

Line GraphsA line graph, or line chart, shows the relationship of onevariable to another. They are most often used to track changesor trends over time (see Figure 1). Line charts are also usefulwhen comparing multiple items over the same time period(see Figure 2). The stacking lines are used to compare thetrend or individual values for several variables.

You may want to use line graphs when the change in a variableor variables clearly needs to be displayed and/or when trendingor rate-of-change information is of value. It is also important tonote that you shouldnt pick a line chart merely because youhave data points. Rather, the number of data points that youare working with may dictate the best visual to use. For example,if you only have 10 data points to display, the easiest way tounderstand those 10 points might be to simply list them in aparticular order using a table.

When deciding to use a line chart, you should consider whetherthe relationship between data points needs to be conveyed. If itdoes, and the values on the X axis are continuous, a simple linechart may be what you need.

Figure 1: Line graphs show the relationship of one variable to another.


5/20


6/20

4

Scatter PlotsA scatter plot (or X-Y plot) is a two-dimensional plot that showsthe joint variation of two data items. In a scatter plot, eachmarker (symbols such as dots, squares and plus signs)represents an observation. The marker position indicates thevalue for each observation. Scatter plots also support grouping.When you assign more than two measures, a scatter plot matrix

is produced. A scatter plot matrix is a series of scatter plots thatdisplays every possible pairing of the measures that areassigned to the visualization.

Scatter plots are useful for examining the relationship, orcorrelations, between X and Y variables. Variables are said tobe correlated if they have a dependency on, or are somehowinuenced by, each other. For example, prot is often relatedto revenue and the relationship that exists might be that asrevenue increases, prot also increases (a positive correlation).A scatter plot is a good way to visualize these relationshipsin data.

In a scatter plot, you can also apply statistical analysis withcorrelation and regression. Correlation identies the degreeof statistical correlation between the variables in the plot.Regression plots a model of the relationship between thevariables in the plot.

Once you have plotted all of the data points using a scatter plot,you are able to visually determine whether data points arerelated. Scatter plots can help you gain a sense of how spreadout the data might be or how closely related the data points are,as well as quickly identify patterns present in the distribution ofthe data (see Figure 4). Scatter plots are helpful when you havemany data points. If you are working with a small set of datapoints, a bar chart or table may be a more effective way todisplay the information.

} Scatter plots can help you gain a senseof how spread out the data might be orhow closely related the data points are.They can also quickly identify patternspresent in the distribution of the data.

Bubble Plots A Scatter Plot VariationA bubble plot is a variation of a scatter plot in which the markersare replaced with bubbles. In a bubble plot (see Figure 5), eachbubble represents an observation. The location of the bubblerepresents the value for two measured axes; the size of thebubble represents the value for a third measure. These plotsare useful for data sets with dozens to hundreds of values orwhen the values differ by several orders of magnitude. You canalso use a bubble plot when you want specic values to berepresented by different bubble sizes. Animated bubble plotsare a good way to display changing data over time.

Figure 3: This bar graph a waterfall chart is used to represent the relative contribution of each category to the total.


7/20

5

Figure 4: A scatter plot is a good way to visualize relationships in data.

Figure 5: A bubble plot animated by time.


8/20

6

Pie Charts There is much debate around the value of pie charts, whichare used to compare the parts of a whole. However, they canbe difficult to interpret because the human eye has a hard time

estimating areas and comparing visual angles. Anotherchallenge with using a pie chart for analysis is that it is difficultto compare slices of the pie that are similar in size but notlocated next to each other. If you do use pie charts, they aremost effective when there are limited components and whentext and percentages are included to describe the content. Byproviding additional information, information consumers do nothave to guess the meaning and value of each slice. If youchoose to use a pie chart, the slices should be a percentage ofthe whole (see Figure 6). When designing reports or

dashboards, another consideration for the efficacy of a piechart is the amount of space the pie chart requires in the sizingof the report. Because of the round shape, pie charts requireextra real estate, so they may be less than ideal whendeveloping dashboards for small screens or mobile devices.Other charts may provide a better way to represent the sameinformation in less space (see Figure 7).

} Pie charts are most effective when thereare limited components and when textand percentages are included todescribe the content.

Figure 6: A pie chart helps you compare the percentages of different components..


9/20

7

Of course, there are many other chart types you can use topresent data and analytical results. The selection of chartsusually will depend upon the number of categories andmeasures (or dimensions) you want to visualize. By followingthe tips outlined here and understanding the examples, youmay need to try different types of visuals and test them withyour audience to make sure the correct information isbeing conveyed.

Visualizing Big DataBig data brings new challenges to visualization because ofthe speed, size and diversity of data that must be taken intoaccount. The cardinality of the columns you are trying tovisualize should also be considered. One of the mostcommon denitions of big data is data that is of such volume,variety and velocity that an organization must move beyondits comfort zone technologically to derive intelligence foreffective decisions.

Volume refers to the size of the data. Variety describes whether the data is structured,

semistructured or unstructured. Velocity is the speed at which data pours in and how

frequently it changes.

Building upon basic graphing and visualization techniques,SAS Visual Analytics has taken an innovative approach toaddressing the challenges associated with visualizing data.Using innovative, in-memory capabilities combined with SASAnalytics and data discovery, SAS provides new techniquesbased on core fundamentals of data analysis and thepresentation of results.

Figure 7: Alternatives to pie charts include line charts and bar charts.


10/20

8

Large Data VolumesOne challenge when working with big data is how to displayresults of data exploration and analysis in a way that ismeaningful and not overwhelming. You may need a new way

to look at the data that collapses and condenses the results inan intuitive fashion but still displays graphs and charts thatdecision makers are accustomed to seeing. You may also needto make the results available quickly via mobile devices, andprovide users with the ability to easily explore data on theirown in real time.

} One challenge when working with bigdata is how to display results of data

exploration and analysis in a way that ismeaningful and not overwhelming.

When working with massive amounts of data, it can be difficultto immediately grasp what visual might be the best to use. Theautocharting capability in SAS Visual Analytics takes a look atthe data you wish to examine and then, based on the amountof data and the type of data, it presents the most appropriatevisualization. This intelligent autocharting helps businessanalysts and nontechnical users easily visualize their data. Theycan build hierarchies on the y, interactively explore data and

display the data in different ways to answer specic questionsor solve new problems without having to rely on constantassistance from IT to provide changing views of information.

In addition, what does it mean explanations in SAS VisualAnalytics display information about the analysis that has beenperformed, and identify and explain the relationships betweenthe variables that are displayed (see Figure 8). This makesanalytics and the creation of data visualizations easy, eventhose with nontechnical or limited analytic backgrounds.

} The autocharting capability in SAS Visual Analytics takes a look at all of thedata you wish to examine and then,based on the amount of data and thetype of data, it presents the mostappropriate visualization.

Data volume can become an issue because traditionalarchitectures and software may not be able to process hugeamounts of data in a timely manner, thus requiring you to makecompromises and aggregate the details you want to visualize.Even the most common descriptive statistics calculations canbecome complicated when you are dealing with big data anddont want to be restricted by column limits, storage constraintsand limited support for different data types. One SAS solution tothese issues is an in-memory engine that speeds the task of dataexploration and a visual interface that clearly displays the resultsin a simple visualization (as provided by SAS Visual Analytics).

For example, what if you have a billion rows in a data set andwant to create a scatter plot on two measures? The user trying toview a billion points in a scatter plot will have a hard time seeingso many data points. And the application creating the visual

may not be able to plot a billion points in a timely or effectivemanner. One potential solution is to use binning (the groupingtogether of data) on both axes so that you can effectivelyvisualize the big data (see Figure 8).

Box plots are another example of how the volume of data canaffect how a visual is shown. A box plot is a graphical display ofve statistics (the minimum, lower quartile, median, upperquartile and maximum) that summarize the distribution of a setof data. The lower quartile (25th percentile) is represented bythe lower edge of the box, and the upper quartile (75thpercentile) is represented by the upper edge of the box. Themedian (50th percentile) is represented by a central line thatdivides the box into sections. Extreme values are representedby whiskers that extend out from the edges of the box. Usually,these display well when using big data (see Figure 9).

Often, box plots are used to understand the outliers in the data.Generally speaking, the number of outliers in the data can berepresented by 1 percent to 5 percent of the data. Withtraditionally sized data sets, viewing this proportion of the datais not necessarily hard to do. However, when you are workingwith massive amounts of data, viewing 1 percent to 5 percent of

the data is rather challenging.

For example, if you were working with a billion rows of data, theoutliers would represent 10 million data points and visualizing10 million data points would be difficult. In Figure 9, we haveone category and two measures. To visualize outliers couldmean plotting 20 million outlier points. If you bin the results andshow a box plot with whiskers, you can view the distribution ofthe data and see the outliers all calculated on big data.


11/20

9

Figure 8: SAS Visual Analytics provides autocharting and what does it mean pop-ups to help nontechnical users create andunderstand data visualizations. The what does it mean pop-up (bottom) explains that the correlation shown in this binned boxplot indicates a strong linear relationship between Sales Rep Rating and Vendor Satisfaction.

Figure 9: This box plot compares the distribution of data points within a category.


12/20

10

Different Varieties of Data (Semistructured andUnstructured)Data variety brings challenges because semistructured andunstructured data require new visualization techniques. A word

cloud visual (where the size of the word represents its frequencywithin a body of text) can be used on unstructured data as a wayto display high- or low-frequency words (see Figure 10).

SAS Visual Analytics takes the concept of word clouds a stepfurther. It takes advantage of taxonomies and ontologies tomake associations and then organizes words into topics basedon how the words are being used. SAS Visual Analytics wordclouds display the hot topics of the day gleaned from thistext analysis. End users can drill down by clicking on theindividual topic to see exactly what words or phrases comprise

a particular topic.

Figure 10: A word cloud shows the words or phrasesassociated with a topic.

For example, you could use the topic cloud to categorizecustomer comments on Twitter about your products orservices and then click on a topic to drill down to see theactual comments.

} While visualizing structured data isfairly simple, semistructured orunstructured data requires newvisualization techniques, such asword clouds or network diagrams.

Another visualization technique that can be used forsemistructured or unstructured data is the network diagram.Network diagrams view relationships in terms of nodes(representing individual actors within the network) and ties(which represent relationships between the individuals, such asfriendship, kinship, organizations, business relationships, etc.).These networks are often depicted in a diagram where nodesare represented as points and ties are represented as lines.Network diagrams can be used in many applications anddisciplines. For example, businesses analyze social networks tounderstand their interactions with customers, while counter-intelligence and law enforcement might map a clandestine orcovert organization such as an espionage ring, an organizedcrime family or a street gang. You can also superimpose thenetwork diagram on a map, for example, to show therelationship or product sales across geographic areas (see

Figure 11). Word clouds and network diagrams are currentlyavailable in solutions such as SAS Text Miner and SAS SocialMedia Analytics.

Visualization VelocityVelocity is all about the speed at which data is coming into theorganization. The ability to access and process varying velocitiesof data quickly is critical. A correlation matrix combines big dataand fast response times to quickly identify which variables arerelated. It also shows how strong the relationships are betweenvariables. SAS Visual Analytics makes it easy to assess the

relationships. Simply select a group of variables and drop theminto a visualization pane. The intelligent autocharting functiondisplays a color-coded correlation matrix that quickly identiesstrong and weak relationships between the variables. Darkerboxes indicate a stronger correlation; lighter boxes indicate aweaker correlation. If you hover over a box, a summary of therelationship is shown. You can double-click on a box in thematrix for further details.

Figure 12 displays 45 correlation calculations on slightlymore than 1.1 billion rows of data. This graph shows thecorrelation values, and returns results in two to six secondsusing the SAS LASR Analytic Server. Previously, this type ofcalculation would take hours. Now it can be done in seconds.By using box plots and correlation matrices, SAS VisualAnalytics can help speed up your analytics life cycle becauseanalytical modelers can perform variable reductions morequickly and efficiently.


13/20

11

Figure 11: Network diagrams explore relationships within a data set, including connections across geographic areas.

Figure 12: In this correlation matrix, darker boxes indicate a stronger correlation; lighter boxes indicate a weaker correlation.You can double-click on a box for further details.


14/20

12

} A correlation matrix combines big dataand fast response times to quickly

identify which variables among themillions or billions are related. It alsoshows how strong the relationship isbetween the variables.

Figure 13: High cardinality in a bar chart with big data may bedifficult to understand.

Cardinality becomes a concern in big data because the datamay have many unique values per column. The example inFigure 13 shows only 128 unique cities. Because you cannotsee the labels for each bar, the graph becomes less meaningful.Imagine if you had a million bars! It would be impossible tosee them.

SAS has adopted a method for dealing with high cardinality inSAS Visual Analytics bar charts that provide an overview barthat zooms into the bar chart and enable information consumers

to scroll through the entire chart. The level of zoom can also becontrolled. If you compare Figure 13 to Figure 14, it is easy tosee that Figure 14 presents the information more clearly.

Figure 14: This overview axis bar chart shows the highcardinality in this big data more clearly. You can easily scrollthrough the entire chart.

Filtering Big DataWhen working with large amounts of data, being able to quicklyand easily lter your data is important. What if you only want toview data for a certain region, product line or some othervariable? SAS Visual Analytics has ltering capabilities that makeit easy to rene the information you see. Simply add a measureto the lter pane or select one thats already there, and thenselect or deselect the items on which to lter.

But what if the lter isnt meaningful or it skews the data inundesirable ways? One way to better understand thecomposition of your data is through the use of histograms.Histograms provide a visual distribution of the data along withcues for how the data will change if you lter on a particularmeasure. Histograms save time by giving you an idea of theeffect the lter will have on the data before you apply it. Ratherthan relying on trial and error or instinct, you can use the

histogram to help you decide what to focus on.

Data Visualization Made Easy WithAutochartingIn SAS Visual Analytics, intelligent autocharting produces thebest visual based on what data you drag and drop onto thevisual palette. It is important to note that autocharting may notalways create the exact visualization you had in mind. In thatcase, you also can select a specic visual to build. However,when you are rst exploring a new data set, autocharts areuseful because they provide a quick view of the data. You then


15/20

13

have the ability to switch to another specic visual as desired.For example, with autocharting, when a single measure isselected, distribution of that measure is shown (Figure 15).

Figure 15: Autocharting in SAS Visual Analytics produces abar chart to show the distribution of a single measure

The addition of a second measure results in either anautocharted heat map (Figure 16) or a scatter plot (Figure 4).

Figure 16: With autocharting, two measures can either result

in a heat map (above) or a scatter plot.

A category of data can be one of three types: standard, date orgeographic. When the category type is standard, the visual willshow a frequency count of data (see Figure 17). If the categoryis a date, then a measure is also required and the visual will be aline graph (see Figures 1 and 2). If the category is geographic,then a map will be displayed (see Figure 18).

Figure 17: When the category type of data is standard, SASVisual Analytics displays a frequency count.

Figure 18: If SAS Visual Analytics determines that the datacategory is geographic, it uses a map frequency chart.

} In SAS Visual Analytics, intelligentautocharting produces the best visualbased on what data you drag and droponto the visual palette. When you arerst exploring a new data set, autochartsare useful because they provide a quickview of the data.


16/20

14

Say you wanted to better understand what drives your vendorssatisfaction and maybe even how to inuence it in the future. Byselecting the decision tree visualization, you simply drag anddrop Vendor Satisfaction, and then add whatever othervariables you believe might inuence it, such as Sales RepRating, Marketing Investment and Vendor Rating, as shown inFigures 20 and 21.

Figure 20: Part of a decision tree. At the top, we can see thatSales Rep Rating seems to be the best explanatory variableto describe impact to Vendors Satisfaction.

The decision tree algorithm will immediately indicate whichvariable has the most inuence on vendor satisfaction (in thiscase Sales Rep Rating), and at what value it has the greatestimpact (57 percent). The next branching point (shown in Figure21) shows the second most important factor (MarketingInvestment) and so on down the tree.

Each branch in the tree increasingly renes the various elementsthat affect our analysis, and we segment our data according tothe various branching out points, until we nd the next potentialsegment that we want to investigate further (see Figure 21).

While business users with no background in advanced analyticscan use this powerful tool, advanced users also have access tomany parameters so they can further ne-tune this analysis (seeFigure 21, on the right side).

The autocharting in SAS Visual Analytics takes into accountthe cardinality of the data and adjusts the visuals accordingly.Using the visual in Figure 19 as an example, the cardinality ofthe column Product Style was 355. Autocharting checkedthe cardinality of the selected column and automaticallyprovided the overview bar axis because the cardinality wasdeemed high. The overview axis is an option that can beturned on and off as required.

Figure 19: Autocharting checked the cardinality of theselected column and automatically provided the overviewaxis, an option that can be turned on or off as desired.

Decision TreesDecision trees, also known as classication or regression trees,allow you to analyze problems (or even predict behaviors) thatare inuenced by multiple factors in a manner thats easy tounderstand. Decision trees present inuencing factorsincrementally as a series of one-cause, one-effect relationshipsthat are easier to understand than more complex, multiple-variable techniques.

Decision trees attempt to nd a strong relationship betweeninput values and target values in a group of observations thatform a data set. When the analytics identies a set of inputvalues as having a strong relationship to a target value, then allof these values are grouped in a bin that becomes a branch ofthe decision tree. A strong relationship is dened as one whereknowledge of the value of an input improves the ability topredict the value of the target.


17/20

15

Can You See Into the Future?Forecasting estimates future values for your data based onstatistical trends. As such, it is an extremely important tool fororganizational planning. Fortunately, SAS Visual Analytics canhelp you expand the culture of forecasting in your organization.Easy-to-use capabilities take the complexity out of forecasting,so that users of all skill levels can see for themselves what mighthappen in the future.

A simple menu guides users through the process of generatingforecasting results. Select the date, time or date-time data itemsyou want to use for the forecast. The software automaticallychooses the most appropriate forecasting algorithm for thedata chosen. You also have the option to select the forecastingintervals. When you click OK, a line chart is created, along with aclear explanation of the forecasting results in the what does itmean section at the bottom of the screen, as shown in Figure22. This is just another way SAS Visual Analytics brings advancedanalytics to nontechnical users in an approachable format.

When additional measures are added to the forecast (as shownin Figure 23), three things happen in SAS Visual Analytics:

1 Each variable is evaluated to determine whether itinuences the forecast. Variables deemed to beinuencers are added to the bottom of the screen forsimulation purposes.

2 When inuencers are found, the forecast is recalculated andrened. As you can see, the condence interval (light bluebars) around the forecast in Figure 23 is much tighter than inFigure 22.

3 Users can manipulate the values of the inuencing variablesto see the potential impact on the forecast, in effect byperforming simulations.

Figure 21: Part of a decision tree. At the bottom, we can see the data segmented according to the various branching points,and on the left side we can see the various ne-tuning parameters available in Expert Mode.


18/20

16

Figure 22: With automated forecasting capabilities, SAS Visual Analytics chooses the most appropriate forecasting algorithmfor the selected data. What does it mean? (bottom) provides explanations of analytic functions and data correlations,so even nontechnical users can understand what the data means.

Figure 23: By adding additional measures, these underlying factors are evaluated as to their potential impact on the forecast,the forecast is recalculated accordingly, and users can use these additional values to perform simulations.


19/20


20/20

To contact your local SAS office, please visit: sas.com/offices

SAS and all other SAS Institute Inc. product or service names are registered trademarks or trademarks of SASInstitute Inc. in the USA and other countries. indicates USA registration. Other brand and product names aretrademarks of their respective companies. Copyright 2014, SAS Institute Inc. All rights reserved.
http://www.sas.com/officeshttp://www.sas.com/offices

sas whitepaper data visualization techniques

Documents