Group order in GTL

This post could be titled something like "Everything you wanted to know about Group Order in GTL - and more."   The group ordering shows up in three different ways in your graph.

  1. Assignment of attributes (color, marker symbol) to group values.
  2. Position of group values in the graph.
  3. Display of the group values in the legend.

Unique group values are assigned their visual attributes (color, marker symbol) from the GraphData1 - GraphDataN elements defined in the active style.  Most SAS shipped styles like LISTING or HTMLBLUE have 12 group elements.

Starting with SAS 9.3M1, assignment of attributes to group values is based only on the order in which these group values are present in the data  So, if the group value "B" is encountered first in the data, it gets the attributes from GraphData1 style element, and so on.  This happens even if an observation cannot actually be drawn in the graph due to other other reasons.  Also, missing group values get the "GraphMissing" element, and do not impact assignment of the other non-missing values.

I have created all graphs for this article using the GTL code shown below to keep the discussion simple.  The SGPLOT procedure uses the SAS summary object to compute some statistics under the covers.  This can change the order of the data received by the graph renderer.  Here is the GTL program.

proc template;
  define statgraph bar;
    dynamic _resp _footnote _order;
      entrytitle 'Mileage by Origin and Type';
      entryfootnote halign=left "Data Set = " _footnote;
      layout overlay / xaxisopts=(display=(tickvalues));
	barchart category=Origin response=_resp / group=Type groupdisplay=cluster 
                 stat=mean outlineattrs=graphdatadefault name='a' ;
        discretelegend 'a' / sortorder=_order;

Note, _resp, _footnote and _order are dynamics so we can use the same GTL template to produce all the graphs in this article.  These dynamics are set in the SGRENDER procedure.  If a dynamic value for an option is not set, that option is ignored.

The template itself uses the following:

  • A BARCHART statement, with Category, Response and Group options, using cluster groups and stat=mean.
  • A DiscreteLegend, with the sort option.

CarsIn the full data set (minus Hybrids), the first group value is SUV, but the first group value for the category Europe is Sedan.  We use this data set to create the graph using the SGRENDER step shown below.

proc sgrender data=cars template=bar;
dynamic _resp="mpg_city"

The  graph is shown below.   The group colors are assigned in the order the unique group values are encountered in the data.  "SUV" is the first value for TYPE in the data set, and hence gets GraphData1 as the style element, with the blue fill color.  "Sedan" is the second group value, so gets GraphData2, with the red fill color.  All of the group colors are displayed in the legend in the order they are encountered in the data.  Click on the graph for a higher resolution image.

BarCarsThe order of displaying each group value within each category is unique to each category.  So, for the category "Europe", Sedan is the first group value, and hence is displayed first.  The color used for Sedan is consistent in the whole graph.

Now, instead of letting the GTL BarChart summarize the data, let us summarize the data ourselves.  We will use the MEANS procedure to get the MeanMpg by Origin and Type as follows:

CarMeansproc means data=cars noprint;
class origin type;
var mpg_city;
output out=carmeans

Note, in the data shown above, the order of the type values has changed.  Now the first type value for Europe is also SUV.  This is different from the order of the original data.

Now, we use this data to plot the MeanMpg by Origin and Type using the same template and the SGRENDER code below.

BarCarMeansproc sgrender data=carmeans template=bar;
  dynamic _resp="meanMpg" _footnote="CarMeans"; 

Note in the graph above, the order for assigning the group colors has now changed.  Wagon and Truck have now swapped positions and colors.  Also, the position of each type value within each category has changed.  The colors are correct within the graph, but are no longer consistent with the first graph.

BarCarMeansByMpgThe purpose of this exercise was to order the group values within each category by descending value of the response.  Since the order within each category is retained as was in the data, we can now sort the data however we want, and display the values in our custom order.

The graph above shows the car types by origin sorted by the mean mileage within each category.  Now, since Sedan is first in the data, it gets the first color, and so on.  But, every time you do this kind of a custom sort (say by car counts), the incoming data order changes, and so does the color assignment.  How can we retain consistent group colors across all graphs?

AttrMapTo ensure the colors are consistent we can use a Discrete Attr Map.  Extract the original order of the group values using the MEANS procedure, and construct an attr map data set so that the colors are specified using the order.   The attr map data set is shown on the right.

BarCarMeansByMpgMapThe graph created using this attr map is shown on the right, with the legend entries sorted alphabetically.  The colors of each type are exactly as the original graph, though the positions in the legend are now in alphabetical order.

BarCarMeansByMpgMergedIf we are really picky, and want the colors assigned as per the original order AND get positions in the legend in the same data order, we have to play a little trick.   Instead of building an attr map, we use the extracted the original order, and prepend those values into the sorted data, with missing values for the Origin column.   See code in the attached program.

Remember, we said at the top of the article that colors are assigned based on the order of the group values, even if the observations cannot be drawn due to other reasons.  So, prepending the unique group values in the order you want, with other missing values will do this trick.  Now, the color values are assigned in the original data order (or, whatever order we want), and the observations are drawn in the order they are in rest of the data.

SAS 9.4 Code:  Group_Order_94


Post a Comment


Spirals are cool.  And useful.  We use them every day without thinking about it.  Every time the road turns from a straight line to a curve, we go through a transition spiral.  Spirals allow us to change curvature in a steady increasing or decreasing fashion.   Without a spiral, this transition would be abrupt.

Weather_SeriesFor our purpose, spirals can also be useful for visualizing data that is cyclical in nature.  If you are visualizing daily high temperature over a 2 year period, you could plot it on a straight X axis as shown on the right.

The cyclical nature of the data is evident in the graph.  But, this same cyclical behavior may be  easier to understand on a spiral with one cycle per year.   So, that is the plan for today.

A visit to Wikipedia page on spirals reveals there are many kinds of spirals including logarithmic, hyperbolic and many more.  Let us start with the simple Archimedean Spiral.  This has the simple formula:  R = A + B * theta.

Experimenting with this equation yields these different spirals for various values of A and number of cycles.

Spiral_N1A=0, Cycles=1

The x and y points are computed using the spiral equation, for theta values up to N * 360, where N is the number  of cycles.  The curve is plotted using the series statement of the SGPLOT procedure.

proc sgplot data=spiral nowall noborder pad=0;
  series x=x y=y / smoothconnect;
  xaxis min=-&max max=&max grid display=none;
  yaxis min=-&max max=&max grid display=none;

A=0.5, Cycles=2

The first spiral starts from the center (A=0) and turns through 360 degree cycle.  The second spiral starts at an offset of 0.5 the radius and turns through 720 degrees.

Once the spiral is drawn, now we need to map the data on to the spiral so that the time axis is along the spiral and the response values are drawn normal to the spiral, towards the center.

Spiral_N3_VThe graph on the right shows the basic idea.  At each point along the time axis, we compute the theta and then find the (x, y) point on the spiral.  The direction vector (cx, cy) for the response (the arrows) is towards the center of the spiral (0, 0).   In the example on the right, all arrow heights are half the spiral spacing.  So, we can compute the (x2, y2) location of the arrow heads from (x, y) and (cx, cy) as shown in the program.

I have used some SAS 9.4 features to draw smooth curves and remove background wall and border.   A SAS 9.3 version program is also included.

Spiral_N3_VS5With real time series data, we would normalize the response over the entire range, and plot the data one side of the spiral by using the abs() value and a color to signify sign.  Then, scale the vector by response and plot.  A simulated example is shown on the right.

We will cover the mapping real world time series data (as shown in the first graph on top) on to the spirals in next article on Spirals.

SAS 9.4 Program:  Spiral_Macro_94

SAS 9.3 Program:  Spiral_Macro_93

Post a Comment

Proportional Euler Diagram

The topic of VENN diagrams had come up a while ago.  At that time, I thought it may be interesting to build a proportional VENN diagram.  But, reading up on VENN Diagrams, I learned that VENN diagrams represent all intersections of N sets, regardless of whether there are actually any observations in one of the regions.  So, there did not seem any purpose to make a proportional VENN diagram, and maybe the term itself is an oxymoron.

Euler_30_20_0_SheenI was interested in a graphical representation of the number of different types of subjects in a study, say subjects with Diabetes, or Hypertension or both.   It turns out, Euler Diagrams do represent the real world data, and not all theoretical combination.  So, it would make sense to draw a Proportional Euler Diagram.

I started with the simple 2-Set case, as it seems achievable.  The results are shown on the right.  The values for N1, N2 and NI are also Euler_30_0_20_sheenshown in the footnote, along with the value of the convergence error.  The two special cases are shown on the right, and are straightforward.  Click on the graphs for a higher resolution image.

The two cases with intersecting circles are shown below.  For the first one, the numbers are such that the intersection point of the two circles lies in-between the centers of the two circles.   For the second case, the intersection lies to the right of the smaller circle.

Euler_30_20_10_NoneIn all cases, the radius of the larger circle is set to 10 (arbitrary), and I compute the area of the smaller circle proportional to the number of observations in the circles.

Here are the details of my program:

  • N1, N2 and NI are the number of observations in Set 1, Set 2 and intersection ONLY.
  • So, N1+NI is first circle, and N2+NI is the 2nd circle, and NI is the intersection.
  • Euler_30_10_30_PressN1 >= N2.
  • Special case #1 -> NI=0.  This means the two circles are non-overlapping.
  • Special case #2 -> N2=0.  This means the circle 2 is fully inside circle 1.
  • Case #3 -> the intersecting vertical line is between center 1 and center 2.
  • Case #4 -> the intersection vertical line is to the right of centers #2.

Here is the algorithm:

  • First, I assign v - height of the intersection above centerline = 1.
  • Compute the three different areas.
  • Compute the area per observation in each section.
  • Then, based on the ratio of ANI / AN1, I adjust v by the error ratio.  V is kept < r2.
  • I repeat this while the error is > 0.001 and number of iterations < N.
  • Now, if the error is still > 0.001, convergence is not reached and the intersection is to the right of the center 2.
  • Now, set v=0.99999*r2 and repeat the same computations above, with reducing v.

I assume convergence is reached, and based on this value of v, I compute the horizontal distance from center of each circle to the intersection, d1 and d2 and other numbers needed to plot the details.

I can use the ELLIPSEPARM or BUBBLE (RelativeScale=False)  statement to draw the plot.  However, SGPLOT procedure does not support these statements (not in the 80-20 range for simple plots).  So, I used GTL, with the BubblePlot because I wanted to use skins.

I made it into a macro, with three parameters N1, N2 and NI.  Skin is optional.  If you have a need for Proportional Euler Diagrams in your work, please chime in and let me know if this is useful to you.  Maybe you have made one of your own and I would love to hear how you went about solving for the intersection areas.

VENN diagram shapes for 2, 3, 4 and more sets are available on the web, would be possible to make these using EllipseParm statement for both circles and ellipses.

I plan to tackle the case of the 3 set Proportional Euler Diagram.  This same algorithm may not extend to this case.  I would love to hear your ideas.

Full GTL Macro program:  Euler_Bubble_Macro

Post a Comment

Graphs are easy in SAS University Edition

By now you have heard all about the SAS(R) STUDIO software that provides access to the power of SAS analytics in a Web browser.  The SAS(R) University Edition is also available free for higher education teaching, learning and  research.

This software includes ODS Graphics software for creating graphs.  You can use the familiar program window to write your own SAS data step and procedures.  An example of running your own SGPLOT program is shown in Robert's recent article on How to create a Histogram using the SAS University Edition.

Studio_UI_2Making graphs gets even easier in SAS Studio by using the graph tasks that are included with the software.  When you first launch the software, you will see the user interface shown on the right.  Click on the "Tasks" button on the left, and you will see a list of tasks by category.

Here I have highlighted the Tasks folder and the Graph subfolder under it.

Multiple graph tasks are available including Bar Chart, Bar-Line Chart, Box Plot, Histogram, Line Chart, Pie Chart, Scatter Plot, Series Plot and more.

Histogram_TaskEach of these tasks presents you with a form to set the data and various options as shown on the right.  Here, I have launched the "Histogram" task as shown highlighted in blue.  This starts the Histogram task.

Each task presents you with an easy to use visual interface to set the parameters and options necessary to make the graph.  These are all collected under two tabs - The Data tab and the Options Tab.  In the image on the right, the Data tab for the Histogram task is highlighted in yellow.

Each graph task allows you to provide the name of the data set and the required variables to create the graph.  In case of Histogram, you need to provide only one numeric variable for the Analysis Role.  In the example above, we have selected the SASHELP.CARS data set and the MPG_CITY column for graphing.

Histogram_ResultsOnce the required parameters are provided, you can submit the task by pressing the run button or 'F3'.  The task will render the histogram using the default settings for styles and present the graph to you in the Results window as shown on the right.

Each task also supports optional settings which are included under the "Options" tab.  These options can be used to customize your graph, including setting of titles, footnotes and  graph size.  In this case, I have set the graph size to 4" x 3" to fit the small region.

Each graph task generates the required SGPLOT code needed to render the graph.  This code is available in the Code window under the "Code / Results" tab.  This code is built and updated as you apply the settings in the Data and Options panels.  So, this is a good way to get started with learning the SGPLOT procedure.

The tasks cover many of the features available in the SGPLOT procedure, but not all.  So, you can cut and paste the code into the program window and customize it to your own needs.

You can learn more about creating graphs using SG procedures right here in this blog.  Learn all about the procedures themselves in the book on Statistical Graphics Procedures by Example.




Post a Comment

Swimmer plot

At PharmaSUG 2014 in San Diego, I had the pleasure of attending "Swimmer Plot: Tell a Graphical Story of Your Time to Response Data Using PROC SGPLOT", by Stacey Phillips.  In this paper, Stacey presented an interesting graph showing the effects of a study drug on patients' tumor size.

Swimmer_StaceyStacey says in her paper that often "investigators prefer to dig deeper and look at an individual subject’s pattern of response. A swimmer plot is a graphical way of showing multiple pieces of a subject’s response 'story' in one glance."    The final graph includes a bar showing the length of treatment duration for each patient, classified by the disease stage at baseline, one for each patient in the study.  Graph also includes indicators for the start and end of each response episode, classified by complete or partial response, and an indicator showing whether the patient is a "Durable responder".

Stacey uses a combination of HBarParm, Scatter and annotations to create this graph.  The annotation is used for adding the "Continued response" arrow, and for the display of the inner legend for decoding of the various symbols in the graph.

Along with many of the attendees of the presentation, I was impressed and intrigued by this visual.  I was curious if its creation could be simplified using some of the new features released with SAS 9.3.  In particular, I wanted to see if I could make this graph without any annotation.

DataSAS 9.3 includes some versatile plot statements and features to create graphs.  Two of these are the HIGHLOW plot and the Discrete Attributes Map for controlling the color of the group values.

The data used to create the graph is is eyeballed from Stacey's graph and shown above.  Updated graph is shown below. Click on the graph for a higher resolution image.

Swimmer_93Here are the features of this graph:

  1. This graph uses the High Low plot to draw the bar representing the duration of the response for each subject.
  2. The bar has a arrow on the right side to indicate a continued response.  This is explained in the 1st footnote.
  3. Each response episode is represented by start and end events joined by a line classified by the type of response - Complete or Partial.  Connecting the start and end event and using a common classification color groups these together as one event, and is easier for the eye to consume.  Continuing response does not have an end event on the right.
  4. All event lines and markers are included in the inner legend.

Swimmer2_94It is also possible to place the indicator for continued event into the key legend using a "TriangleRightFilled" marker in the graph.  This marker is drawn outside the plot region, but is included in the legend.  Some items in the legend are shown in grey, to indicate the meaning of the shape since the actual marker will have different colors in the graph based on other criteria.

The graph on the right uses SAS 9.4 with a few aesthetic features for bar skins and filled, outlined markers.  Note the shorter line segments in the legend.

Note, the marker for the right arrow in intentionally made bigger to match the right arrows of the HighCap of the HighLow plot.

SAS 9.3 Code:

footnote  J=l h=0.8 'Each bar represents one subject in the study.';
footnote2 J=l h=0.8 'A durable responder is a subject who has confirmed response for at least 183 days (6 months).';
proc sgplot data= swimmer dattrmap=attrmap nocycleattrs;
  highlow y=item low=low high=high / highcap=highcap type=bar group=stage fill nooutline
          lineattrs=(color=black) name='stage' nomissinggroup transparency=0.3;
  highlow y=item low=startline high=endline / group=status lineattrs=(thickness=2 pattern=solid) 
          name='status' nomissinggroup attrid=status;
  scatter y=item x=start / markerattrs=(symbol=trianglefilled size=8 color=darkgray) name='s' legendlabel='Response start';
  scatter y=item x=end / markerattrs=(symbol=circlefilled size=8 color=darkgray) name='e' legendlabel='Response end';
  scatter y=ymin x=low / markerattrs=(symbol=trianglerightfilled size=14 color=darkgray) name='x' legendlabel='Continued response ';
  scatter y=item x=durable / markerattrs=(symbol=squarefilled size=6 color=black) name='d' legendlabel='Durable responder';
  scatter y=item x=start / markerattrs=(symbol=trianglefilled size=8) group=status attrid=status;
  scatter y=item x=end / markerattrs=(symbol=circlefilled size=8) group=status attrid=status;
  xaxis label='Months' values=(0 to 20 by 1) valueshint;
  yaxis reverse display=(noticks novalues noline) label='Subjects Received Study Drug' min=1;
  keylegend 'stage' / title='Disease Stage';
  keylegend 'status' 'd' 's' 'e'  'x' / noborder location=inside position=bottomright across=1;

The part that I believe makes this version easier to consume is the continuity of the response events.  Joining the start and end events with a line segment, all having the same color as per the event classification allows the eye to see each event and its duration clearly.

The part I like best is the graph uses no annotation.

Full SAS 9.3 Code:Swimmer_93




Post a Comment

Grouped Timeline

Recently, a user posed a question on how to plot stacked frequencies on a time axis.  The data included frequencies of different viruses by week.  The data is modified to preserve confidentiality and is shown below.

DataThe user's first instinct was to use a bar chart with stacked groups.  This works for automatically computing frequencies by week and group and also stacked the group values.  Except, the x axis is made discrete and the bars are only drawn where data exists.  However, the user wants to see all weeks positioned correctly the x axis, with gaps where there is no data for some weeks.  The data starts in April 2013 and goes to March 2014, so plotting by week displays the data out of order.

Here is the graph, created using the bar chart.  The graph shows the frequencies for the two viruses by week, using stacked groups.  The data for week numbers 1-14 are listed first even though these actually for 2014.  The weeks are drawn as discrete values, and there are no gaps for weeks that are missing because the bar chart treats the Category axis as discrete.  However, the VBAR statement makes it easy to see the stacked frequencies.


To get this kind of graph on a scaled time axis, one would need to use a Needle plot or a HighLow plot.  However, neither of these will automatically compute the frequencies by date and group for a stacked display.

HighLow_DataSo, I used the MEANS procedure to compute the frequencies by week and virus.  Then, I ran a data step by year and week to compute the low and high values for each virus in a given week.  I also compute a "date" value for each week of the year.  Here is the data set:

Now, I use the HighLow plot to draw the bar segments for each virus value by date.  The low and high values for each group segment are already computed.


proc sgplot data=stacked dattrmap=attrmap;
  format week 2.; 
  highlow x=dateOfWeek low=low high=high / group=virus name='a' type=bar
          lineattrs=graphdatadefault attrid=virus; 
  yaxis display=(nolabel) offsetmin=0 grid;
  xaxis display=(nolabel);
  keylegend 'a' / title='Virus' location=inside position=topright across=1;

As you can see, the SGPLOT code is very simple:

  • We use a HighLow plot by dateOfWeek and GROUP=VIRUS.
  • We used the previously defined discrete attributes map for each virus name.
  • We set other details like legend and axis properties.

The user wanted to see the week values displayed, which can be easily done using the LOWLABEL option of the HighLow plot.


The full SAS code is snown below, however, I cannot share the data as it is confidential.  You can see the structure of the data above and if you simulate similar data, you can run the code.

Full SAS 9.3 program (not including data): HighLow_Timeline

Post a Comment

Lab Values Panel

It was almost two weeks ago that I got started making a display for lab tests for a subject, based on a graph I saw on the web for an article on this blog.  KPI_Panel_Crop

This graph is a part of a larger panel display of the lab values for a subject.  The panel includes the display of multiple lab values, including a gradient range of the percentile values for the general population.  The lab value for the subject is shown in the box on the left and also in the gradient range.  The graph is shown on the left.

Cruise_Crop_SmallWhile working on this article, I ran into a few issues including the minor issue of a long planned vacation to Hawaii that included a cruise around the islands.  Suffice it to say, the the islands are fabulous, and the cruise lived up to all the expectations one can imagine.  Here is a picture I took of the boat, when anchored off Kona on the Big Island.

Then, it was time for PharmaSUG in San Diego.  The conference was a resounding success, and I had the opportunity to meet with SAS users interested in creating graphs using ODS Graphics.  The presentations were excellent, with users much more likely to be persuaded by the experiences of fellow SAS users rather than hearing from SAS staff.

Back from these two diversions, I finally got back to this project.  Here is the step by step progression to making this graph.

Data_PanelFirst, on the right is the data I gleaned from the web image, and with Rick's help, created this data set of the values in the graph.  Now, the expectation is that when you make such a graph, you have all the pertinent data in hand.  Note that each value V1, V2, etc. are for the 0, 25th, 50th, 75th and 100th percentile of the data.  Note, for all tests, the "better" numbers are on the left, and the "worse" numbers are on the right.  I use the column "Rev" to indicate the ranges are reversed, with higher numbers to the left.

Lipid_DashboardThis graph uses SAS 9.4 but all significant feature of the graph can be created using SAS 9.3.  Here is the simple graph showing the test, values and the percentile ranges.

For each test, the percentile values for the larger population are shown on the right, with the percentile values above the box, and actual test values inside the box.  The actual test value for the subject is also shown at the correct percentile location on each bar.


title 'Lipid Panel for Subject XXX-XX-XXXX';
proc sgplot data=Lipid noautolegend nowall noborder;
  highlow y=test low=low high=high / type=bar outline nofill barwidth=0.5 ;
  hbarparm category=test response=vn / barwidth=0.2  dataskin=gloss 
           fillattrs=(color=gray) nooutline baselineattrs=(thickness=0);

  scatter y=test x=vn2 / markerchar=v2 markercharattrs=(color=lightgray);
  scatter y=test x=vn3 / markerchar=v3 markercharattrs=(color=lightgray);
  scatter y=test x=vn4 / markerchar=v4 markercharattrs=(color=lightgray);

  scatter y=test x=vnl / markerchar=lvn1 markercharattrs=(color=gray) discreteoffset=-0.35;
  scatter y=test x=vn2 / markerchar=lvn2 markercharattrs=(color=gray) discreteoffset=-0.35;
  scatter y=test x=vn3 / markerchar=lvn3 markercharattrs=(color=gray) discreteoffset=-0.35;
  scatter y=test x=vn4 / markerchar=lvn4 markercharattrs=(color=gray) discreteoffset=-0.35;
  scatter y=test x=vnh / markerchar=lvn5 markercharattrs=(color=gray) discreteoffset=-0.35;

  scatter y=test x=vn / markerattrs=(symbol=trianglefilled size=12) discreteoffset=0.2
          filledoutlinedmarkers markerfillattrs=(color=white) dataskin=gloss;
  scatter y=test x=vn / markerchar=value discreteoffset=0.4 
          markercharattrs=(size=8 weight=bold);

  xaxis display=none offsetmin=0 offsetmax=0;
  yaxis display=(nolabel noticks noline);

The program above uses a HighLow plot to draw the box of ranges, and a scatter plot with markerchar option to display the percentile values above the box and the actual values in the middle.  An offset triangle marker is used to denote the percentile location of the actual value, and the value itself is displayed below the marker.

Lipid_Dashboard_Box_NameThe test names in the original graph are left aligned, and the values are displayed in a box next to the test name along with the units of the values.  I added this information using additional HighLow plots with HighLabel option to display the test name, the test value and the units.

The only unit that needs improvement is the "muMol/L", where it would be better to use the greek symbol for "mu".

title 'Lipid Panel for Subject XXX-XX-XXXX';
proc sgplot data=Lipid noautolegend nowall noborder;
  highlow y=test low=boxL high=boxH / type=bar nofill outline lineattrs=(color=black) barwidth=0.6;
  scatter y=test x=boxM / markerchar=value discreteoffset=0 markercharattrs=(size=8 weight=bold);
  scatter y=test x=boxM / markerchar=units discreteoffset=-0.4 markercharattrs=(size=7 color=gray);
  highlow y=test low=nameL high=nameH / type=bar nooutline barwidth=0.6 fillattrs=(transparency=1);
  scatter y=test x=nameL / datalabel=test datalabelattrs=(size=8 weight=normal) datalabelpos=right
  highlow y=test low=low high=high / type=bar outline nofill barwidth=0.5 ;
  hbarparm category=test response=vn / barwidth=0.2  dataskin=gloss 
           fillattrs=(color=gray) nooutline baselineattrs=(thickness=0);
  scatter y=test x=vn2 / markerchar=v2 markercharattrs=(color=lightgray size=7);
  scatter y=test x=vn3 / markerchar=v3 markercharattrs=(color=lightgray size=7);
  scatter y=test x=vn4 / markerchar=v4 markercharattrs=(color=lightgray size=7);
  scatter y=test x=vnl / markerchar=lvn1 markercharattrs=(size=7 color=gray) discreteoffset=-0.35;
  scatter y=test x=vn2 / markerchar=lvn2 markercharattrs=(size=7 color=gray) discreteoffset=-0.35;
  scatter y=test x=vn3 / markerchar=lvn3 markercharattrs=(size=7 color=gray) discreteoffset=-0.35;
  scatter y=test x=vn4 / markerchar=lvn4 markercharattrs=(size=7 color=gray) discreteoffset=-0.35;
  scatter y=test x=vnh / markerchar=lvn5 markercharattrs=(size=7 color=gray) discreteoffset=-0.35;
  scatter y=test x=vn / markerattrs=(symbol=trianglefilled size=12) discreteoffset=0.2
          filledoutlinedmarkers markerfillattrs=(color=white) dataskin=gloss;
  scatter y=test x=vn / markerchar=value discreteoffset=0.4 markercharattrs=(size=8 weight=bold);
  xaxis display=none offsetmin=0 offsetmax=0;
  yaxis display=none;

Now, let us get to the display of the gradient green-yellow-red ranges in the display.  There is no plot statement in SG or GTL that can draw a gradient color across three colors.  Some plot statements support a Color Response option, but essentially the entire entity is rendered with the color derived from the color gradient.

Lipid_Dashboard_Grad_Name_ValueOnce again, we resort to using the versatile  HighLow plot to draw the gradient.  HighLow plot does not support a color gradient option, but does support a GROUP option that colors each segment with the group color from the style, or a Discrete Attributes Map.  Here, we will use the DAttrMap option of the SGPLOT procedure to draw the ramp.

We create Low and High columns for 100 HighLow segments for each test name.  Each segment is 1 unit, in a do loop from 0 to 99 by 1.  Each segment has an id - the loop variable.

We also create a DAttrMap data set, such that each value 1-99 has a corresponding color that gradiates from green to yellow to red.  See the code in the full program attached at the bottom.  The result is the gradient ranges as shown in the graph above.

Lipid_Dashboard_Grad_AnnoFinally, we use some simple annotations to add the information at the top of the graph.   Five observations in the SGANNO data set describe the way to draw the four text strings and the arrow object.

Once again, this exercise has exposed the need for some more features that will make this task easier such as support of ColorResponse for bar charts and Highlow plot.  We will look into adding such options in a future release.

The technique to creating such non standard and complex graphs using SG or GTL is to analyze the graph, and break it down in to its component parts.  Then use the appropriate plot statement "creatively" to build the graph l layer at a time.   Some details that cannot be done using plot statement can be handled by annotate.

Full SAS9.4 Code without the Gradients:  Lipid_Dashboard

Full SAS9.4 Code with Gradients: Lipid_Dashboard_Gradient

Full SAS9.3 Code with Gradients:  Lipid_Dashboard_Gradient_93

Post a Comment

Report from PharmaSUG 2014

Just getting back from PharmaSUG 2014 in San Diego.  The conference was great, both inside and outside.  The organizers ordered up some great weather for the Padres game and also for dinner on the flight deck of the Midway Carrier.

DG01_Time_To_Event_PanelOur focus here being on graphics, we were all extremely gratified by the presentations in the Data Presentation section.  Amos Shu got us started with graphs for Adverse Event timeline graphs and panels in his paper Techniques of Preparing Datasets for Visualizing Clinical Adverse Events.

Wu, Dai and Gau presented a Graphical Representation of Patient Profile for Efficacy Analyses in Oncology  with Efficacy Patient Profile graphs using GPLOT and ANNOTATE:


Mayur Uttarwar and Murali Kanakenahalli proposed Developing Graphical Standards: A Collaborative, Cross-Functional Approach to ensure the correct list of Symbols and colors for the plots in the graph.

DG07_SwimmerStacey Philips presented a Swimmer Plot: Tell a Graphical Story of Your Time to Response Data Using PROC SGPLOT, displaying disease stages for each subject with additional information on the events.

Kriss Harris presented Napoleon Plot for PharmaSUG and I Am Legend for PharmaSUG , presenting displays for assessing treatment safety, and ways to create just a legend, when the number of entries in the legend are too many to be included in one graph.

Jeffery Meyers presented Kaplan-Meier Survival Plotting Macro %NEWSURV which used the GTL layouts in creative ways to display loads of information in one plot or panel.


SP14_SurvivalWarren Kuhfeld presented ways to customize the popular Survival Plot graph created by the LIFETEST Procedure for SAS 13.1 using a combination of %ProvideSurvivalMacros, Customization macros, %CompileSurvivalTemplates to create the customized templates, and then run the LIFETEST procedure to produce the customized graph output.


DG14_GTL_LayoutsFinally, I presented my paper from SGF 2014 -Up Your Game with Graph Template Language Layouts using GTL layouts to create complex custom graphs.  This paper will get you started using the GTL layouts to go beyond the graphs you can create using the SGPLOT procedure.

As usual, PharmaSUG lived up to its reputation of taking care of its attendees by providing fabulous food for breakfast, lunch and dinners.  In addition to all the knowledge, I feel like I also gained 5 pounds.

For me, the highlight is always meeting and interacting with SAS users, who bring so much enthusiasm to the conference.  One quote that I took back to my team from a presentation was "Making graphs with SAS is FUN".  It is nice to get validation of our efforts to provide you the tools you need to easily create beautiful and effective graphs with SAS.

Post a Comment


DataArrowSG Procedures and GTL provide you with a large set of plot statements, such as BarChart, ScatterPlot, BoxPlot and more.  You can use them for the intended purpose, and all is well and good.  However, the real fun starts when you leverage a plot to do something that was not obvious.  One such plot that is designed to be used in creative and flexible ways is the HIGHLOW plot statement.

Let us start with the MEANS procedure to compute mean mileage values by Origin and Type.  We have added a few columns  as shown in the table above.

proc means ne 'Hybrid'));
  class origin type;
  var mpg_city mpg_highway;
  output out=carmeans(where=(_type_ &gt; 2))
    mean(mpg_city mpg_highway) = City Highway;

BarWe now have what we call as "Chart Ready" data.  We can create a bar chart using the SGPLOT procedure.  I used the VBARPARM statement since the data is already summarized.  The graph shows the mean highway mileage by Origin and Type.  The mileage value of each bar is displayed on top.  Click on the graph for a high resolution image.



proc sgplot data=carmeans;
  vbarparm category=origin response=highway / group=type datalabel 
           outlineattrs=graphoutlines dataskin=matte;
  xaxis display=(nolabel noticks);

HighLowBarVertZeroExactly the same graph can be created using the HIGHLOW plot statement, introduced in both SGPLOT and GTL in SAS 9.3.  This statement can be used with one of these two syntax specifications for the SGPLOT procedure.

HIGHLOW X=<var> High=<num-var> Low=<num-var> / <options>;

HIGHLOW Y=<var> High=<num-var> Low=<num-var> / <options>;

In the first case, vertical bar segments are drawn from Low value to the High value for each value of X  .  In the second case horizontal bar segments are drawn from Low value to the High value for each value of Y .


proc sgplot data=carmeans;
  highlow x=origin high=highway low=zero / group=type type=bar 
          groupdisplay=cluster highlabel=highway lineattrs=graphoutlines
  xaxis display=(nolabel noticks);
  yaxis offsetmin=0;

Note in the program above, we have used the following required parameters:

HighLow X=Origin High=Highway Low=Zero / Type=Bar GroupDisplay=Cluster 

HighLowBarVertSetting LOW=zero makes sure all bar segments are drawn to the baseline in the graph shown above.  Using Low=City and High=Highway allows us to draw floating bar segments depicting the mileage range by Origin and Type.  We can use both the LowLabel and HighLabel to display both the low (City) and high (Highway) value for each bar.





Using the parameter Y instead of X, allows us to create horizontal bar segments as shown on the right.  In this case, we have enabled the drawing of horizontal bands to help visually cluster the groups within each category.

Another useful feature of the HighLow plot is the display of caps at the end of each bar.  This can be useful to indicate certain characteristics for each bar such as direction (increasing or decreasing) or a continuation of an event in either direction.


The graph on the right displays a horizontal HighLow plot with caps.  The columns LOWCAP and HIGHCAP from the data set are used to display the caps.  The values are set in the columns when certain conditions are met.

In this case, a low cap is drawn for observations where City mileage is < 18 and a high cap is drawn when Highway mileage > 27.



Another interesting feature is drawing of a "Clip Cap".  This feature automatically draws a cap to indicate the bar is clipped when the bar value exceeds the min or max value of the axis.

In the example on the right, x axis Max is set to 28.  We have used the option CLIPCAP which draws a special cap at the end of any bar segment that is clipped by the axis min or max value.  Here, we have drawn a reference line at x=28 to display the max setting.

SGPLOT code with Clip Caps:

proc sgplot data=carmeans;
  highlow y=origin high=highway low=city / group=type type=bar groupdisplay=cluster
          lowlabel=city highlabel=highway clipcap
          barwidth=1 clusterwidth=0.8 lineattrs=graphoutlines dataskin=matte;
  refline 28 / axis=x lineattrs=(pattern=dash);
  xaxis display=(nolabel) max=28 values=(12 to 30 by 4);
  yaxis display=(nolabel noticks) colorbands=even colorbandsattrs=(transparency=0.5);


HighLowLinePlotAllSo far we have seen the use of the HIGHLOW plot with TYPE=BAR.  This plot statement also supports TYPE=LINE (default).  This plot type is useful to display a stock plot and can also be used to display four response values per line.  The example on the right displays monthly stock values, showing the high, low, open and close values.


The HighLow plot can be used where ever you want to display some events of certain duration, such as a Schedule Plot or an Adverse Event Plot.  These examples are shown below.



Full SAS 9.4 Program for High Low Plots:  HighLow

Post a Comment

Multi-Group Series Plots

The series plot is a popular way to visualize response data over a continuous axis like date with a group variable like treatment.   Here is some data I made up of a response value by date, treatment, classification and company that makes the drug.  The data is simulated as shown in the attached program (see bottom of article).

DataThe data includes the columns VALUE, DATE, DRUG, CLASS and COMPANY.  The columns LABEL and VALUEL are computed at every 5th observation per drug for labeling.


Series_94We can use the GTL SERIESPLOT to display Value by Date and Drug as shown on the right.  Click on the graph to see a higher resolution graph.  The drug name for each curve is displayed at the right end of the curve, and also in the legend below.  We could turn off the legend if needed.

If a GROUP variable is not provided, the entire data is plotted as one series.  When a GROUP variable is provided, the data is plotted as one curve for each group value.  Each curve gets the display attributes such as color and line pattern from one of the GraphData01 - 12 style elements, in the order the group values are encountered in the data.  Alternatively, one can also use a Discrete Attributes Map to assign specific color and line pattern values by group value.

Sample code shown here is for SAS 9.4.  While some new options may not work, the basic ideas discussed below also works at SAS 9.3 or earlier.

SAS 9.4 GTL code for series with group:

proc template;
  define statgraph Series;
    begingraph / subpixel=on;
      entrytitle 'Values by Date and Treatment';
      layout overlay / xaxisopts=(display=(ticks tickvalues))
	 seriesplot x=date y=value / group=drug name='a' lineattrs=(thickness=2) 
        discretelegend 'a' / title='Drug:';
proc sgrender data=SeriesGroup template=Series;

SeriesLabel_94Note, the curve labels drawn at the end can get cluttered, as happened above for groups B and C.  To improve this situation, we can label each curve along its length at frequent intervals.  We do this by using the columns LABEL and VALUEL, which have non missing values at every 5th observation per group.

We can use these columns to overlaying a scatter plot with the marker character option.  To reduce clutter of overlaid text, we add a white marker behind each letter.  We discussed such ideas in Labeled Curves.

SAS 9.4 GTL code for curves with inline labels.

proc template;
  define statgraph SeriesLabel;
    begingraph / subpixel=on;
      entrytitle 'Values by Date and Treatment';
      layout overlay / xaxisopts=(display=(ticks tickvalues))
	seriesplot x=date y=value / group=drug name='a' lineattrs=(thickness=2) 
	scatterplot x=date y=valueL / group=drug 
                     markerattrs=(symbol=circlefilled color=white size=10);
        scatterplot x=date y=valueL / group=drug markercharacter=label; 
        discretelegend 'a' / title='Drug:' itemsize=(linelength=15px) 
                    location=inside across=1 halign=right valign=top;

In the program above, we have also used ATTRPRIORITY=COLOR on the ODS GRAPHICS statement to delay the use of patterns till after all colors are exhausted.  See attached full program.  This option makes all regular styles behave like the HTMLBLUE style.  Each group is rendered by a different color from the style, using a total of four colors.

SeriesLineColorGroup_94Now, we want to be able to group the curves for each drug by another grouping variable like the drug class.  I assigned two classes "NSAID" and "Opioid".  Since each curve is labeled by the name of the drug, we want to use the color to depict the class of the drug.  We can do this by using a secondary group role called LINECOLORGROUP.  The graph is shown on the right where each curve is now colored either blue or red based on the drug class.  The legend contains a color swatch with its value.

SAS 9.4 GTL code for line color by a group:

proc template;
  define statgraph SeriesLineColorGroup;
    begingraph / subpixel=on;
      entrytitle 'Values by Date, Treatment and Class';
      layout overlay / xaxisopts=(display=(ticks tickvalues))
        seriesplot x=date y=value / group=drug name='a' lineattrs=(thickness=2) 
                   smoothconnect=true linecolorgroup=class;
        scatterplot x=date y=valueL / group=drug 
                  markerattrs=(symbol=circlefilled color=white size=10);
        scatterplot x=date y=valueL / group=drug markercharacter=label
        discretelegend 'a' / title='Drug Class:' type=linecolor location=inside 
                  across=1 halign=right valign=top;

Note the features of the graph above:

  • We have labeled each treatment curve by its own label, so no need for a legend for this case.
  • We have assigned the color for each curve by a secondary group variable CLASS.
  • We have used a Discrete Legend of TYPE=LINECOLOR.  This displays only color swatchs.
  • The only requirement here is that the GROUP variable must be the lowest grouping factor for each curve.  The LINECOLORGROUP value must remain the same for all obs with same GROUP value.

The good news here is that LINECOLORGROUP has been available in GTL SERIESPLOT all along since SAS 9.2.  It is used by the POWER procedures, but the feature was tested only for the POWER procedures' use cases.  Hence, we did not feel confident we could document this feature as ready for general use.  Now, after hearing multiple users express the need for such use cases, we felt it was necessary to release this as production.  Now this feature has been well tested, and no problems have been found.  So, we feel the risk-to-reward ratio is in favor of exposing this feature to you.

In addition to LINECOLORGROUP, you can also use LINEPATTERNGROUP, MARKERCOLORGROUP and MARKERSYMBOLGROUP.  Each one can be used with the group variable and this value should not change withing a GROUP value.


In the graph on the right, I have used COMPANY as the LINEPATTERNGROUP.  Now, each drug is colored by its CLASS and patterned by the COMPANY.  I have also added a discrete legend of TYPE=LINEPATTERN.  Both these legends are wrapped inside a LAYOUT GRIDDED and placed at the top right of the cell.

SAS 9.4 GTL code for series with line color and line pattern groups:

proc template;
  define statgraph SeriesLineColorPatternGroup;
    begingraph / subpixel=on;
      entrytitle 'Values by Date, Treatment, Class and Company';
      layout overlay / xaxisopts=(display=(ticks tickvalues))
        seriesplot x=date y=value / group=drug name='a' lineattrs=(thickness=2) 
                  smoothconnect=true linecolorgroup=class linepatterngroup=company;
	scatterplot x=date y=valueL / group=drug 
                  markerattrs=(symbol=circlefilled color=white size=10);
        scatterplot x=date y=valueL / group=drug markercharacter=label
        layout gridded / halign=right valign=top columns=2 columngutter=5;
          discretelegend 'a' / title='Drug Class' type=linecolor location=inside 
                        across=1 halign=right valign=top;
          discretelegend 'a' / title='Company' type=linepattern location=inside 
                        across=1 halign=right valign=top itemsize=(linelength=30);

Note the features of the graph above:

  • We have labeled each treatment curve by its own label, so no need for a legend for this case.
  • We have assigned the color for each curve by a secondary group variable CLASS.
  • We have assigned the pattern for each curve by a secondary group variable COMPANY.
  • We have used a Discrete Legend of TYPE=LINECOLOR.  This displays only color swatches.
  • We have used a Discrete Legend of TYPE=LINEPATTERN.  This displays patterns without color.
  • The only requirement here is that the GROUP variable must be the lowest grouping factor for each curve.  The LINECOLORGROUP and LINEPATTERNGROUP variables must remain the same for all obs with same GROUP value.

While you can display many different classifications in the graph at the same time, the graph can become complex very quickly.  You can  turn on the display of the markers for the series plot, and then control the visual attribute of the markers using MARKERCOLORGROUP and MARKERSYMBOLGROUP.

In the process of making the graphs for this article I noticed the lack of a way to make the scatter markercharacter color by group, to match the color of the drug names to the line when using LINECOLORGROUP.  There is no matching MARKERCOLORGROUP in the SCATTERPLOT.  I will see what we can do about that.  Please chime in with your comments and observations.

I certainly look forward to see the ways in which you can leverage these features.

Full SAS 9.4 Code:  MultiGroup_94

Full SAS 9.3 Code:  MultiGroup_93

Post a Comment