|WikiProject Measurement||(Rated Start-class, Low-importance)|
Request for references
Request for references: I suppose some of it is still classified, but nonetheless, are there any references for this article, particularly pertaining to Turing's and Good's work? I added a couple of references to the article, but they only pertain to the material I added about Hartley. 126.96.36.199 22:52, 25 November 2005 (UTC)
The material is now in the Ban (information ) page. This page can be deleted. —Comment added 19:33, 19 November 2006 (UTC) by User:Grgarza (talk) — Preceding unsigned comment added by SudoMonas (talk • contribs)
There was a  tag in the sentence "decibans corresponds to an odds ratio of 10:1; 20 decibans to 100:1 odds, etc". That is relatively straightforward math (see bellow) so I presume it refers to the use of it as in "The deciban is a particularly useful measure of information in odds-ratios or weights of evidence" and moved it there.
Given a event with odds 1:n, then the information is .
For n "large",
- If we use probability 1/n instead of 1:n odds, there's no need to stipulate "for large n". --Doradus (talk) 16:29, 20 July 2011 (UTC)
- One could be deduced, since it should be the same as for entropy. But that's not relevant for this article. —DAGwyn (talk) 21:00, 16 February 2015 (UTC)
Change page name to Ban_(unit)
Ban as a measurement unit is related to Nat, but Nat is indicated as Nat_(unit). To keep things consisitant and less confusing, this page name should be changed to Ban_(unit). Eyreland (talk) 01:44, 9 August 2014 (UTC)
- Done – Not controversial. —Quondum 04:16, 9 August 2014 (UTC)
Reference to 'probability' in table is misleading
The entry for "probability" in the table that (very informatively) translates decibans into odds equivalents is misleading, or at least invites a mistake. A transformation of the likelihood ratio, a deciban's characterization of the "weight" of a piece of evidence refers to how much more consistent the evidence is with one hypothesis than another. By itself, then, a deciban doesn't tell you how probable the hypothesis is; to figure that out, you would, in a manner consistent with Bayes's Theorem, need to combine the weight of the evidence measured in decibans, or transformed into a likelihood ratio, with your prior. The only time it makes sense to transform a deciban measure into a probability is when it refers to your posterior odds -- or basically your (provisional) resting point in assessing the probability of a hypothesis. If you characterize the weight of any particular piece of evidence measured in decibans as the "probability of the hypothesis being true," you will be committing what is known as the prosecutor's fallacy. This is pretty important to the exposition, since the whole point of the deciban is the contribution it makes to conveying information about the weight of pieces of evidence used in an iterative, Bayesian updating of the probability of a hypothesis; it is in the course of that sort of ongoing assessment of available pieces of evidence that that there is a practical need for a tractable unit of "evidentiary weight." There's not much practical need, in contrast, for an alternative way to characterize the weight of posterior odds; indeed, at that point, one just expresses the conclusion in terms of odds or odds transformed to probability. Dmk38 (talk) 15:14, 25 December 2014 (UTC)
- Here is a different point about the same table: It seems to make a fundamental confusion in the calculation. For example, an event with 1:1 odds has 50% probability of occurring, which should be −log10(1/2) ≈ 3.0 deciban. 0 deciban should correspond to the probability of an event being 100%, or odds of 1:0. Taking the logarithm of the ratio as expressed in odds seems to make no sense in any context. Presumably the table needs to be corrected? —Quondum 17:01, 15 February 2015 (UTC)