Talk:Inverse Gaussian distribution

From Wikipedia, the free encyclopedia
Jump to: navigation, search
WikiProject Statistics (Rated Start-class, Low-importance)
WikiProject icon

This article is within the scope of the WikiProject Statistics, a collaborative effort to improve the coverage of statistics on Wikipedia. If you would like to participate, please visit the project page or join the discussion.

Start-Class article Start  This article has been rated as Start-Class on the quality scale.
 Low  This article has been rated as Low-importance on the importance scale.


it would be illustrative to have a graphical represaentation of the distribution

The last change (November 4) to the Brownian motion section was made by me... reference is "The Inverse Gaussian Distribution: Theory, methodology and applications" p29.

Deavik 23:39, 4 November 2007 (UTC)


There are several parameterizations of the inverse Gaussian distribution, including one which makes the relationship between it and a Brownian motion with drift more explicit (in terms of the drift parameter v and the variance parameter \sigma^2). These other parameterizations should be at least mentioned. I am not aware of any good reason to present the given one as "canonical"; though if such a reason exists, it, too, should be presented. Cheers, Eliezg 05:04, 6 November 2007 (UTC)

Error in likelihood function.[edit]

I think the likelihood function should be changed to:

L(\mu, \lambda)=
\left(      \frac\lambda{2\pi}   \right)^\frac n 2  
\left(      \prod^n_{i=1} \frac{w_i}{X_i^3}    \right)^{\frac12} 
\exp\left(\frac\lambda{\mu} -\frac\lambda{2\mu^2}\sum_{i=1}^n w_i X_i - \frac\lambda 2 \sum_{i=1}^n w_i \frac1{X_i} \right).

Without this additional term, you can not solve the first order condition for the MLE's, as they are given in this article.

Ryantg (talk) 23:14, 15 March 2009 (UTC)

I agree - fixed Batman50 (talk) 15:17, 4 May 2010 (UTC)

the misleading part[edit]

the article mentioned the name is misleading, and the inverse gaussian distribution is not the distribution of Y=\frac{1}{X} (X is normal), then what is the name of the distribution of Y???? It will be great if we say something about that. Jackzhp (talk) 14:11, 28 July 2009 (UTC)

If X is normal, the distribution of 1/X doesn't seem to have a special name, but its pdf is straightforward to work out from probability density function#Dependent variables and change of variables. I can't find any references but from a quick bit of experimental mathematics i think it doesn't have any moments if the mean of X is zero and none apart from the mean otherwise, and it's oddly-behaved around zero in general, although if the coefficient of variation of X is large you may not notice in practice. See also Box-Cox distribution for the case when X is truncated at zero. As i don't have any references, it's rather messy and doesn't seem to have much theoretical or practical importance, i'm not going to add anything to the article myself. Qwfp (talk) 18:01, 25 May 2010 (UTC)

Error in the labels of the figure[edit]

Apart from formatting differences, a figure identical to the one shown can be generated in R using

plot(x, dinvgauss(x, 1, 1), type="l", xaxs="i", yaxs="i", xlab="", ylab="", col=1); lines(x,dinvgauss(x,1,0.2),col=2); lines(x,dinvgauss(x,1,3),col=3); lines(x,dinvgauss(x,3,1),col=4); lines(x,dinvgauss(x,3,0.2),col=5)

However, according to the R manual (e.g., [1]), the second parameter of the function dinvgauss corresponds to 1/lambda, instead of lambda. Hence, the figure's labels should be replaced. Alternatively, use the following code to generate an interesting sequence of densities:

x<-seq(0,3,0.01); plot(x, dinvgauss(x, 1, 1/4), type="l", xaxs="i", yaxs="i", xlab="", ylab="", col=1,ylim=c(0,2.5)); lines(x,dinvgauss(x, 1, 1/2), type="l", col=2); lines(x,dinvgauss(x, 1, 1), type="l", col=3); lines(x,dinvgauss(x, 1, 2), type="l", col=4); lines(x,dinvgauss(x, 1, 4), type="l", col=5); lines(x,dinvgauss(x, 1, 8), type="l", col=6); lines(x,dinvgauss(x, 1, 16), type="l", col=7); lines(x,dinvgauss(x, 1, 32), type="l", col=8)

Btw, it would be useful to add code snippets that generate the graphs. —Preceding unsigned comment added by Szepi (talkcontribs) 21:16, 6 February 2010 (UTC)

Clarifying the initial description[edit]

I think one of the introductory paragraphs is likely to be misconstrued, specifically the clause, "while the Gaussian describes the distribution of distance at fixed time in Brownian motion . . . ." It is so commonplace to generalize Brownian motion to higher dimensions that I think it is confusing to claim the process's distance is Gaussian (since in higher dimensions, it isn't). I think this article should not assume that readers be familiar with Brownian motion. If we ditched the word "distance" and replaced it with "level," I think we would not only clear up the ambiguity, but the wording would nicely complement the following sentence: Gaussian describes level at a certain time; inverse Gaussian describes time to hit a certain level. Disadvantage: "level" isn't very precise, although my gut tells me it is fairly intuitive. LandruBek (talk) 22:07, 6 April 2010 (UTC)

If there is no drift...[edit]

The article should, I think, also mention the special case of this distribution as μ tends to infinity. This is how I was first introduced to the IG distribution -- it describes the hitting time of a drift-less Wiener process at level λ². The derivation is an elementary one from the CDF of a Gaussian using a simple change of variables, and thus I think readers are likely to expect to see this case mentioned explicitly. LandruBek (talk) 22:07, 6 April 2010 (UTC)


What are the w_i terms occurring in some of the equations? Weights?

Any information on the quantiles? -- (talk) 15:39, 17 August 2013 (UTC)