Exponential binning


by atillaqurd
Tags: binning, exponential
atillaqurd
atillaqurd is offline
#1
May15-13, 05:16 AM
P: 2
Hi,

I need to find out how to plot my data with exponential binning.
To better see the exponent of f(x) = x ^ \alpha, where x and f(x) are given, I am asked to do exponential binning the data.

Would appreciate you help.

Yours
Atilla
Phys.Org News Partner Physics news on Phys.org
Researchers find tin selenide shows promise for efficiently converting waste heat into electrical energy
After 13 years, progress in pitch-drop experiment (w/ video)
Global scientific team 'visualizes' a new crystallization process (w/ video)
mathman
mathman is offline
#2
May15-13, 02:31 PM
Sci Advisor
P: 5,935
Use semilog graph paper.
Mute
Mute is offline
#3
Jun9-13, 12:54 PM
HW Helper
P: 1,391
It seems the OP hasn't replied, but there are some important issues that need to be addressed here, so I will comment on them for any future posters who stumble across this thread.

If you are generating histograms of something which you expect to follow a power law ##y(x) \sim x^\alpha##, you need to use logarithmic binning, not exponential binning.

That is, you want your bins to be equally spaced on a log scale, which means you want the edge of the $k$th bin, B(k), to be given by

$$\log_{10}(B(k)) = m \log_{10} (k) + c,$$
where m is the slope and c is the intercept, which are determined by your bin range and your number of bins. For example, if you want 10 bins between 10-6 and 100, then ##B(0) = 10^{-6}## and ##B(10) = 10^0##, and you can solve for m and c.

Now, this next point is extremely important: when using logarithmic binning, you must divide your y-data by the width of the bin. If you do not do this, the power of ##x^\alpha## that you measure will be wrong.

Furthermore, when estimating power laws from data, if you need anything more than a rough estimate, a linear regression is a terrible way to find the exponent. It is very prone to systematic errors. Maximum likelihood fits are a much better method. See this preprint for a discussion of properly calculating power laws from data (as well as using hypothesis testing to see if you can rule out other behaviors like log-normal distributions).

atillaqurd
atillaqurd is offline
#4
Jun10-13, 07:59 AM
P: 2

Exponential binning


Thanks indeed!


Register to reply

Related Discussions
Proving a matrix exponential determinant is a exponential trace Calculus & Beyond Homework 13
Mathematica: Binning 2D data Math & Science Software 8
binning a unit sphere Calculus & Beyond Homework 5
Data binning Set Theory, Logic, Probability, Statistics 4
Silicon binning techniques Electrical Engineering 0