Defining a Lattice with Additional Features

Overview

The following sections explain how to generate Stock Plot, which requires the following tasks:
  • transforming the input data
  • using external axes instead of internal cell axes
  • adding sidebars that display descriptive text
  • using column headers
  • sizing rows.
Stock Plot
Stock Plot

Transforming the Input Data

A common use for a lattice is to create a graph that shows different subsets of the same input data. In some cases, those subsets are already defined in the input data. However, you will frequently have to transform the input data to make it suitable for the graph you are trying to create. This might require any or all of the following:
  • summarizing the data
  • transposing the data
  • scaling the data values
  • creating new variables that represent subsets of the data.
The graph that is shown in Stock Plot is based on data from SASHELP.STOCKS, which contains several years of monthly stock information for three companies. The data set contains columns for STOCK, DATE , VOLUME, and ADJCLOSE (Adjusted Closing Price). However, it does not have the volume and price information in the form that is needed for the graph. The LATTICE layout does not support subsets of the input data on a per-cell basis. So, in order to make the cell content different, unique variables must be created for each cell to provide the appropriate date, volume, and price information. The following DATA step performs the necessary input data transformations:
data stock;
  set sashelp.stocks;
  where stock eq "Microsoft" and year(date) in (2004 2005);
  format Date2004 Date2005 date.
         Price2004 Price2005 dollar6.;
  label Date2004="2004" Date2005="2005";
  if year(date) = 2004 then do;
     Date2004=date;
     Vol2004=volume*10**-6;
     Price2004=adjclose;
  end;
  else if year(date)=2005 then do;
     Date2005=date;
     Vol2005=volume*10**-6;
     Price2005=adjclose;
  end;
  keep Date2004 Date2005 Vol2004 
       Vol2005 Price2004 Price2005;
run;
The data is filtered for Microsoft and for the years 2004 and 2005. Next, new variables are created for each year and the Volume and Stock Price within each year. Because the volumes are large, they are scaled to millions. This scaling will be noted in the graph. This coding results in a "sparse" data set, but it is the correct organization for the lattice because observations with missing X or Y values are not plotted.
Obs    Date2004    Date2005    Price2004    Price2005    Vol2004    Vol2005

  1          .     01DEC05           .          $26        .        62.8924
  2          .     01NOV05           .          $27        .        71.4692
  3          .     03OCT05           .          $25        .        72.1325
  4          .     01SEP05           .          $25        .        66.9765
  5          .     01AUG05           .          $27        .        65.5300
  6          .     01JUL05           .          $25        .        69.0466
  7          .     01JUN05           .          $25        .        62.9567
  8          .     02MAY05           .          $25        .        62.6998
  9          .     01APR05           .          $25        .        77.0902
 10          .     01MAR05           .          $24        .        72.8997
 11          .     01FEB05           .          $25        .        75.9923
 12          .     03JAN05           .          $26        .        79.6428
 13    01DEC04           .         $26            .      84.4881      .
 14    01NOV04           .         $26            .      86.4461      .
 15    01OCT04           .         $25            .      65.7429      .
 16    01SEP04           .         $24            .      57.7253      .
 17    02AUG04           .         $24            .      52.1046      .
 18    01JUL04           .         $25            .      76.6667      .
 19    01JUN04           .         $25            .      77.0683      .
 20    03MAY04           .         $23            .      58.9425      .
 21    01APR04           .         $23            .      77.3867      .
 22    01MAR04           .         $22            .      77.1119      .
 23    02FEB04           .         $23            .      57.3859      .
 24    02JAN04           .         $24            .      63.6359      .
The key point to be aware of is that every plot in every cell must use variables that contain just the information appropriate for that cell. You cannot use WHERE clauses within the template definition to form subsets of the data.
The following initial template defines the lattice:
proc template;
define statgraph lattice1;
 begingraph;
   entrytitle "Microsoft Stock Performance";
   layout lattice / columns=2 rows=2;
     /* define row 1 */
     seriesplot y=price2004 x=date2004 / lineattrs=GraphData1;
     seriesplot y=price2005 x=date2005 / lineattrs=GraphData1;

     /* define row 2 */
     needleplot  y=vol2004 x=date2004 /
       lineattrs=GraphData2(thickness=2px pattern=solid);
     needleplot  y=vol2005 x=date2005 /
       lineattrs= GraphData2(thickness=2px pattern=solid);

   endlayout;
 endgraph;
end;
run;

proc sgrender data=stock template=lattice1;
run;
Note that because Date2004 and Date2005 have an associated SAS date format that a TIME axis is used and the variable labels are used for X-axis labels.
Initial Lattice for the Graph
Initial Lattice

Using External Axes

Initial Lattice for the Graph would benefit from externalizing the X and Y axes because the external axes will reduce the redundant X axis information and unify the data ranges in the Y axes. We would also like to add grid lines to all axes. To conserve space along the X axes, the automatic formatting of each TIME axis is turned off in the following template code. The TICKVALUEFORMAT=MONNAME1. setting indicates how to format the time axis tick values.
proc template;
define statgraph lattice2;
 begingraph / designwidth=495px designheight=370px;
   entrytitle "Microsoft Stock Performance";
   layout lattice / columns=2 rows=2
                    rowdatarange=union columndatarange=union
                    rowgutter=3px columngutter=3px ;
     /* define row 1 */
     seriesplot x=date2004  y=price2004 / lineattrs=GraphData1;
     seriesplot x=date2005  y=price2005 / lineattrs=GraphData1;

     /* define row 2 */
     needleplot  x=date2004  y=vol2004 /	
         lineattrs=GraphData2(thickness=2px pattern=solid);

     needleplot   x=date2005 y=vol2005 /
         lineattrs= GraphData2(thickness=2px pattern=solid);

     rowaxes;
       rowaxis / griddisplay=on display=(label tickvalues)
                 label="Price"  labelattrs=(weight=bold);
       rowaxis / griddisplay=on display=(label tickvalues)
                 label="Volume" labelattrs=(weight=bold);
     endrowaxes;

     columnaxes;
       columnaxis / griddisplay=on display=(label tickvalues)
                    labelattrs=(weight=bold)
                    timeopts=(tickvalueformat=monname1.);
       columnaxis / griddisplay=on display=(label tickvalues)
                    labelattrs=(weight=bold)
                    timeopts=(tickvalueformat=monname1.);
     endcolumnaxes;
   endlayout;
 endgraph;
end;
run;

proc sgrender data=stock template=lattice2;
run;
Lattice with External Axes
Lattice with External Axes

Using Cell Axes

In most cases externalizing axes improves graph appearance and streamlines coding. However, if there are some axis options that do not apply uniformly to all axes in a column or row, you need to use the standard axis options on a cell basis instead of external axes.
For example, if you wanted X-axis grid lines to appear on the top row of plots but not on the second row of plots, you could not use external axes. Instead, you would enclose the cell contents in an overlay-type layout block and add XAXISOPTS= options on the layout statements. as shown in the following layout blocks:
/* overlay blocks define X-axis options for row 1 */
layout overlay / xaxisopts=(display=none  griddisplay=on);
  seriesplot x=date2004  y=price2004 / lineattrs=GraphData1;
endlayout;

layout overlay /  xaxisopts=(display=none griddisplay=on);
  seriesplot x=date2005  y=price2005 / lineattrs=GraphData1;
endlayout;

/* overlay blocks define X-axis options for row 2 */
layout overlay / xaxisopts=(display=(label tickvalues)
                            timeopts=(tickvalueformat=monname1.));
  needleplot  x=date2004  y=vol2004 /	
    lineattrs=GraphData2(thickness=2px pattern=solid);
endlayout;

layout overlay / xaxisopts=(display=(label tickvalues)
                 timeopts=(tickvalueformat=monname1.));
  needleplot   x=date2005 y=vol2005 /
    lineattrs= GraphData2(thickness=2px pattern=solid);
endlayout;

Adding Sidebars

The graph in Lattice with External Axes is progressing well, but the ENTRYTITLE is centered on the entire graph. It would look better if it were centered on the grid area. This can be accomplished by removing the ENTRYTITLE statement and replacing it with a SIDEBAR block. Four sidebar areas are available: two that span all columns (one on the TOP and one on the BOTTOM), and two that span all rows (one on the RIGHT and one on the LEFT).
sidebar / align=top;
  entry "Microsoft Stock Performance" /
    textattrs=GraphTitleText pad=(bottom=5px);
endsidebar;
Finally, we need a way of explaining that the prices in the first row represent an adjusted close value. We also need to explain that the axis scaling for the second row is in millions of shares. Two strategies are available for providing this information.
The first strategy is to create an external legend. For this strategy, we must define legend text on two of the plot statements, and add a DISCRETELEGEND statement to the BOTTOM sidebar.
seriesplot x=date2004  y=price2004 /
  lineattrs=GraphData2(thickness=2px pattern=solid)
  name="series" legendlabel="Adjusted Close";

needleplot  x=date2004  y=vol2004 /
  lineattrs=GraphData2(thickness=2px pattern=solid)
  name="needle" legendlabel="Millions of Shares";

sidebar / align=bottom;
  discretelegend "series" "needle" / border=off pad=(top=10px);
endsidebar;
The following graph shows what this modification looks like:
Modified Lattice
The other strategy is to add to the row information. At first glance it would seem that you could do this very simply by extending the axis label text:
rowaxes;
  rowaxis / griddisplay=on display=(tickvalues)
            label="Volume (Millions of Shares)" ;
  rowaxis / griddisplay=on display=(tickvalues)
            label="Price (Adjusted Close)" ;
endrowaxes;
The problem here is that the extra axis label text might not fit; depending on the text size and the graph size, the text might be truncated. The axis option SHORTLABEL="string" is available to handle truncation, but we want more text, not alternate text, and there is no way to wrap the axis label to two lines. The solution is use row headers instead of specifying axis labels.

Using Column or Row Headers

For the graph that is shown in Lattice with External Axes, we want to explain that the axis scaling in the first row is in millions of shares, and that the prices in the second row represent an adjusted close value. The strategy that we used in Adding Sidebars was to create an external legend that displays that information. Another strategy we can use is to remove the label information from the row axes and introduce a ROWHEADERS block, as shown in the following code:
rowaxes;
  rowaxis / griddisplay=on display=(tickvalues);
  rowaxis / griddisplay=on display=(tickvalues);
endrowaxes;

rowheaders;
  layout gridded / columns=1;
    entry "Volume" / textattrs=GraphLabelText;
    entry "(Millions of Shares)" / textattrs=GraphValueText;
  endlayout;
  layout gridded / columns=1;
    entry "Price" / textattrs=GraphLabelText;
    entry "(Adjusted Close)" / textattrs=GraphValueText;
  endlayout;
endrowheaders;
By nesting the ENTRY statements in the GRIDDED layouts, we can have multiple lines of text split exactly where we want and in any text style we desire. Without the GRIDDED layouts, only one ENTRY statement could be used per row.
Nearly Final Lattice
To allow more space for the plots, we can rotate the row header text to make it appear to be a row axis label. Notice that we must specify COLUMNS=2 for the GRIDDED layouts.
rowaxes;
  rowaxis / griddisplay=on display=(tickvalues);
  rowaxis / griddisplay=on display=(tickvalues);
endrowaxes;

rowheaders;
  layout gridded / columns=2 ;
    entry "Price" / textattrs=GraphLabelText rotate=90 ;
    entry "(Adjusted Close)" / textattrs=GraphValueText rotate=90 ;
  endlayout;
  layout gridded / columns=2 ;
    entry "Volume" / textattrs=GraphLabelText rotate=90 ;
    entry "(Millions of Shares)" / textattrs=GraphValueText rotate=90 ;
  endlayout;
endrowheaders;
Final Lattice
The clean look of the graph is achieved by removing redundant cell axis information and moving it to external column and row locations. In this example, the use of row headers provided the desired flexibility over row axis labels.

Adjusting the Sizes of Rows and Columns

By default, the rows and columns of the lattice are of the same depth and width. You can use the ROWWEIGHTS= and COLUMNWEIGHTS= options on the LAYOUT LATTICE statement to designate different row depths and/or column widths. Consider the following settings:
LAYOUT LATTICE /  ROW=2 COLUMNS=2
  ROWWEIGHTS=(.6 .4) COLUMNWEIGHTS=(.45 .65) ;
LAYOUT LATTICE with different Row and Column Sizes uses these settings. The ROWWEIGHTS= setting specifies that the first row gets 60% of available row space, and the second row gets 40%. The COLUMNWEIGHTS= setting specifies that the first column gets 45% of available column space, and the second column gets 65%. Potentially, the settings on these options affect the space that is allocated to cell headers and to row and column headers.
LAYOUT LATTICE with different Row and Column Sizes
LAYOUT LATTICE with different Row and Column Sizes
In a traditional stock plot, the area devoted to price information is larger than the area devoted to the volume information. Here is the adjustment made to the row depths:
layout lattice / columns=2 rows=2 rowweights=(.6 .4)
                 rowdatarange=union columndatarange=union
                 rowgutter=3px columngutter=3px;
LATTICE with Adjusted Row Depths
LATTICE with Adjusted Row Depths
This next example shows another way that the ROWWEIGHTS= and COLUMNWEIGHTS= options can be used. Graph with ROWEIGHTS=(.9 .1) shows a two row by one column lattice. The first row is an overlay of a histogram, a density plot, a fringe plot (the short vertical lines below the histogram) representing each observation, and a legend. The second row contains a box plot. The X axes have a uniform scale to ensure that the box plot aligns correctly with the histogram. Because the space that is required to show the second row (box plot) is so much less than the space that is required for the first row, the option ROWEIGHTS=(.9 .1) has been used to reapportion the row space.
Graph with ROWEIGHTS=(.9 .1)
Graph with ROWEIGHTS=(.9 .1)
proc template;
  define statgraph distribution;
    begingraph;
      entrytitle "Distribution of Cholesterol";
      entryfootnote halign=left
                   "From Framingham Heart Study (SASHELP.HEART)";
      layout lattice / rowweights=(.9 .1)
                       columndatarange=union rowgutter=2px;
        columnaxes;
          columnaxis / display=(ticks tickvalues);
        endcolumnaxes;
        layout overlay / yaxisopts=(offsetmin=.04 griddisplay=auto_on);
          discretelegend "Normal" / location=inside
            autoalign=(topright topleft) opaque=true;
          histogram Cholesterol / scale=percent binaxis=false;
          densityplot Cholesterol / normal( ) name="Normal";
          fringeplot Cholesterol / datatransparency=.7;
        endlayout;
        boxplot y=Cholesterol / orient=horizontal  boxwidth=.9;
      endlayout;
    endgraph;
  end; 
run;

proc sgrender data=sashelp.heart template=distribution;
run;
For a generic version of this template, which can be used to show the distribution for any continuous variable without redefining the template, see Using Dynamics and Macro Variables to Make Flexible Templates.