|
|
 | | From: | David Fickbohm | | Subject: | Reason to use missing instead of zero. | | Date: | 22 Jan 05 00:44:59 GMT |
|
|
 | People,
A while ago I asked about how to replace missing values with zeroes. People
of all over SAS-L asked, told, begged me not to replace missing values with
zeroes. I did anyway and fortunately it was after all mathematics, all
logic had been done. I did this just prior to exporting the data into a
spreadsheet.
I wondered why everyone was so excited. This afternoon I spent about an
hour trying to determine why, when I had a column containing only one and
zero, proc means was counting incorrectly, acting strangly, etc. I finally
read the (fabulous) documentation and realized I needed missing values
instead of zeroes. When I had missing values proc means counted properly
and acted properly.
So to all those who previously told me not to change missing values to
zeros. YOU ARE TOTALLY CORRECT.
Dave
Dave Fickbohm
Data Mining Analyst
Homegain+
2450 45th St.
Emeryville, CA, 94608
Phone 510 655 0800 ext 4151
|
|
| | From: | David L. Cassell | | Subject: | Re: Reason to use missing instead of zero. | | Date: | 22 Jan 05 00:56:11 GMT |
|
|
| David Fickbohm wrote [in part]:
> I wondered why everyone was so excited. This afternoon I spent about
an
> hour trying to determine why, when I had a column containing only one
and
> zero, proc means was counting incorrectly, acting strangly, etc. I
finally
> read the (fabulous) documentation and realized I needed missing values
> instead of zeroes. When I had missing values proc means counted
properly
> and acted properly.
>
> So to all those who previously told me not to change missing values to
> zeros. YOU ARE TOTALLY CORRECT.
It's often noticeable in PROC MEANS and PROC UNIVARIATE and PROC PLOT.
It can be much harder to find in the statistical procs. And when you
get to the neural nets in Enterprise Miner, you'll never notice the
problem, much less be able to back out what's weird about your results.
So it's A Bad Thing.
David, who isn't saying "I told you so".. very much. :-)
--
David Cassell, CSC
Cassell.David@epa.gov
Senior computing specialist
mathematical statistician
|
|
|
