PRB: CALCULATE STD--Standard Deviation--Gives Numeric Overflow

ID: Q172084

The information in this article applies to:

Microsoft Visual FoxPro for Windows, versions 3.0, 3.0b, 5.0, 5.0a
on the following platforms: MACINTOSH, NT, Win95
Pro Edition of Microsoft Visual FoxPro for Mac, version 3.0

SYMPTOMS

When calculating the standard deviation using the CALCULATE STD() function, FoxPro can return results with a numeric overflow on certain numerical values over various ranges of records.

RESOLUTION

Use a custom routine that calculates the standard deviation. The following program accepts two parameters for the table and the field. Then it calculates the various sums and products of the fields before computing the standard deviation. Create a program called std and enter the following code:


   * Call this program by ?std("TableName","FieldName") or
   * by assigning the result to a variable, y= std("TableName","FieldName")
   *
      PARAMETER mtable mfield
      SELECT SUM(&mfield^2) FROM (mtable) INTO ARRAY yy
      SELECT SUM(&mfield)^2 FROM (mtable) INTO ARRAY zz
      SELECT COUNT(&mfield) FROM (mtable) INTO ARRAY xx
      stdev=((xx*yy-zz)/xx^2)^.5
      RETURN ROUND(stdev,12)

If a table named values contains the field num, enter this command in the Command Window:


   ?std("values","num")

STATUS

Microsoft is researching this problem and will post new information here in the Microsoft Knowledge Base as it becomes available.

MORE INFORMATION

The numeric overflow is generally observed when a table contains the same repeated real values over a broad range of records. For example, if a table contains 10 records and each record's value is 0.031, the CALCULATE STD() will produce an overflow. This can vary depending on the value and the number of records.

One way to detect that there was a numerical overflow is to display the result of the CALCULATE STD() to the desktop or in a WAIT WINDOW. If SET TALK is ON, the result will be displayed on the desktop automatically. If there is an overflow, all asterisks will be displayed. Also, if the result of the CALCULATE STD() is appended or replaced to a numerical field in a table, the following error may occur:

Numeric overflow. Data was lost.

Steps to Reproduce Behavior

Enter the following code in a program and run it:


   CLEAR
   SET SAFETY OFF
   SET DECIMALS TO 12
   CREATE TABLE stdresult (num N(15,8)) &&Table for result of calculation
   APPEND BLANK
   CREATE TABLE stdtest (num N(15,8))   &&Table of values for calculation
   FOR i = 1 TO 10
      INSERT INTO stdtest(num) VALUES (0.031)
   ENDFOR

   FOR i = 1 TO RECCOUNT()
      ?ALLTRIM(STR(i))+" "
      SET TALK OFF
      CALCULATE STD(num) FOR RECNO()<i TO nresult && Comment this line &
                                                  && the next line once
                                                  && you uncomment the
                                                  && commented lines below.

      ??nresult

      *nresult=custstd('stdresult','num') && Uncomment this line & the next
                                          && line to compute & display the
                                          && standard deviation with a
                                          && user-defined function.

      *?nresult
      SELECT stdresult
      REPLACE num WITH nresult   &&Overflow error on 10th iteration
      SELECT stdtest
   ENDFOR

   SET SAFETY ON
   SET TALK ON
   SET DECIMALS TO 2

   PROCEDURE custstd
   PARAMETER mtable mfield
   SELECT SUM(&mfield^2) FROM (mtable) INTO ARRAY yy
   SELECT SUM(&mfield)^2 FROM (mtable) INTO ARRAY zz
   SELECT COUNT(&mfield) FROM (mtable) INTO ARRAY xx
   stdev=((xx*yy-zz)/xx^2)^.5
   RETURN ROUND(stdev,12)

After running this code, the first nine iterations should be all zeroes. The calculation will be all asterisks and will produce the overflow error message on the REPLACE command.

Additional query words:

Keywords : kbcode kbVFp FxprgGeneral
Version : MACINTOSH:3.0; WINDOWS:3.0,3.0b,5.0,5.0a
Platform : MACINTOSH WINDOWS
Issue type : kbprb