| |
 |
|
|
Science Forum Index » Statistics - Education Forum » Date density diagram
Page 1 of 1
|
| Author |
Message |
| Tobias Weber |
 Posted: Mon Mar 10, 2008 4:51 am |
|
|
|
Guest
|
Hi
I have a lot of data like "list of dates it did not rain in London" and
want to do diagrams that show when something occurs more often than
usual.
I tried assigning a value of 1 to each date and doing subtotals by week.
Unfortunately this kind of histogram will smooth over a significant rise
in density if e.g. Saturday and Tuesday were sunny.
So I calculated in a column how many days ago the last event occured,
which projected over time and inverted kind of works, but breaks down
when the event occurs every third day on average but every day at high
times. The differences are just too small.
What I'd need is something that grows when dates cluster.
And something less annoying than Excel. I looked into R, but dates seem
to be second class citizens there.
--
Tobias Weber |
|
|
| Back to top |
|
| John Kane |
| Posted: Mon Mar 10, 2008 1:49 pm |
|
|
|
Guest
|
Tobias Weber wrote:
Quote: Hi
I have a lot of data like "list of dates it did not rain in London" and
want to do diagrams that show when something occurs more often than
usual.
I tried assigning a value of 1 to each date and doing subtotals by week.
Unfortunately this kind of histogram will smooth over a significant rise
in density if e.g. Saturday and Tuesday were sunny.
So I calculated in a column how many days ago the last event occured,
which projected over time and inverted kind of works, but breaks down
when the event occurs every third day on average but every day at high
times. The differences are just too small.
What I'd need is something that grows when dates cluster.
And something less annoying than Excel. I looked into R, but dates seem
to be second class citizens there.
Without an example it is rather difficult to see what you want.
Re R and dates have a look at the chron and zoo packages.
--
John Kane, Kingston ON Canada |
|
|
| Back to top |
|
| Richard Ulrich |
| Posted: Mon Mar 10, 2008 2:46 pm |
|
|
|
Guest
|
On Mon, 10 Mar 2008 10:51:52 +0100, Tobias Weber <towb@gmx.net> wrote:
Quote: Hi
I have a lot of data like "list of dates it did not rain in London" and
want to do diagrams that show when something occurs more often than
usual.
I tried assigning a value of 1 to each date and doing subtotals by week.
Unfortunately this kind of histogram will smooth over a significant rise
in density if e.g. Saturday and Tuesday were sunny.
So I calculated in a column how many days ago the last event occured,
which projected over time and inverted kind of works, but breaks down
when the event occurs every third day on average but every day at high
times. The differences are just too small.
If "inverting" is not a problem, then "transforming" like
by exponentiating, or whatever, should not be a problem.
But - the differences are "just too small" for what purpose?
It sounds like you have a "ceiling effect" if there is "rain
every day." (Use "dry days", and then transform; or use
reciprocals to invert and spread scores at one end.)
SHOULD they be distinguishable?
If "expert opinion" would take a certain configuration at
say that This is worth 100, while THAT is worth 90, and THESE
are worth 80, 70, 60,... respectively -- Then one last-ditch
solution is to use Expert Opinion to form scores for the
specific, observed configurations of rainfall.
In the end, you then use Expert Scoring, or take the
set of expert scores and find a numerical expression for
them from polynomial regression, or whatever, on the raw
data.
Quote:
What I'd need is something that grows when dates cluster.
And something less annoying than Excel. I looked into R, but dates seem
to be second class citizens there.
--
Rich Ulrich
http://www.pitt.edu/~wpilib/index.html |
|
|
| Back to top |
|
| |
|
Page 1 of 1
All times are GMT - 5 Hours
The time now is Tue Oct 07, 2008 2:22 pm
|
|