• Status: Solved
  • Priority: Medium
  • Security: Public
  • Views: 367
  • Last Modified:

find distribution for strange data


I have data which represent the time for process. I have spent more than two weeks to find a good distribution for it, but I coudln't. Actually I have sued exponentiol with Lambda 142.43 but when I insert that in my simulation I faced a problem due to the varionce between sample huge and that value for Lambda would genertate wrong values. So, please I need any hints or help in this point

The data attached to this question
1 Solution
Ken ButtersCommented:
Not sure if this is exactly what you are looking for or not.  

What I did..

insert a chart to map the points, then inserted a trendline using the option for automatic polynomial.  

selected to show the polynomial in the chart.   It seems to capture the distribution of points rather well.

Attaching updated spreadsheet with chart.
You can fit any set of data points to a polynomial curve, but the real question is , Does the curve mean anything?"

What is the precision of your data?  (It looks pretty good)
What is the relation you expect between y- axis and x- axis?
(mu guess is you expect it to be flat.
Butters give you a nice graph.
My guess is that something major happened in your process between points 1392 and 1393.
And that an equation reproducing your points will be worthless (except for pinpointing the fact that something drastic happented
Here is one way to analyze the data in Excel.

1.  Calculate the time interval between successive events.
2.  Calculate the log of the time differences.
3.  Assign an Index to each log value.
4.  Copy the values of the Log and Index columns to two new columns.
5.  Sort the new columns by log value.
6.  Plot the sorted Log (Top) and Index (Bottom) columns.

There are a few outliers, but most of the Log values are uniformly distributed between -3 and +1.  This corresponds to time intervals between 0.001 and 1.0 seconds (or days or time units).

The Index plot suggests that small interval values may occur anywhere in the sequence.
As the values get larger, they are more likely to appear later.

I still don't know what the data means or what questions you are trying to answer.
Ejgil HedegaardCommented:
Another approach is to find the number of values in a series of intervals, to see how the values are distributed, and find a matching curve, to give the probability for any time value to occur.

First step is to plot the data to get the frequency distribution, and see what it looks like.
I used a step of 10 up to 1060, counting how many values are in each interval, using the frequency function, divided with the number of datapoints (1396), giving the distribution.
The result for 10 is 0.2048, meaning that the probability is 20 % for a value between 0 and 10. For a value between 10 and 20 it is 11 % etc.

It looks exponential, but a check with the logarithmic function shows that it is only the case for the lower values, the upper values follows a more linear function, and a constant will be ok.
See the comments on the sheet.

I end up with a formula like this P(X) = a*Exp(-b*X) + c

The best approximation is then found by calculating the sum of P(X)-Data(X) squared, finding the minimum. For that I used the add-in Solver function.
It gives a correlation of 0.97 witch is ok with a dataset as this.
And the chart shows that the approximation is ok.
obad62Author Commented:
Thanks, That what I am looking for exactly.

Featured Post

Free Tool: Site Down Detector

Helpful to verify reports of your own downtime, or to double check a downed website you are trying to access.

One of a set of tools we are providing to everyone as a way of saying thank you for being a part of the community.

Tackle projects and never again get stuck behind a technical roadblock.
Join Now