4,941 reputation
1323
bio website quantdec.com
location Northeastern US
age 14
visits member for 4 years
seen yesterday

Consultant (environmental and spatial stats a specialty), expert witness, and teacher. I can be reached through (outdated but still valid) links posted on my web site.

Twitter: @WilliamAHuber // ASA-P website: http://amstatphilly.org/


Why waste time learning, when ignorance is instantaneous?

--T(iger) Hobbes.

For any complex problem there is a simple solution. And it's always wrong.

--[Mis?]attributed to H.L. Mencken by Dava Sobel, Longitude.


Sep
9
comment Mean of squared “sum of squared errors”
See Advanced Theory of Statistics, Volume I, chapters 3 ("Moments and Cumulants") and 12 ("Cumulants of Sampling Distributions--(2)").
Sep
9
comment Mean of squared “sum of squared errors”
Jonas, the very first comment to your question indicates how such derivations can be done. Your example, when fully expanded, is a quartic form in the data and therefore the expectation (because it's a linear operator) becomes a homogeneous polynomial (in a suitable sense) of the first four moments of $X_1$. Mathematica merely is doing that routine algebra under the hood. An algebraic theory has been developed; it is explained in great detail in Kendall & Stuart (5th Ed.).
Sep
8
comment How to find a mapping function from n dimensional space to m dimensional space
Could you perhaps add some information to this question to show readers why it might be of interest on this site?
Aug
4
comment Does affine equivariance implies shape unbiasedness?
(Due to lack of answers on CV, this question has been migrated to Math at the OP's request.)
Jul
5
comment Combinatorics: Mean and Variance of an indicator function of items arranged in a circle.
You have hit on the crux of the matter: the question, to be unambiguous, must somehow include enough information to enable calculation of all probabilities. In some settings those probabilities are given explicitly, but usually some particular process whereby the randomization is achieved is described. When the process consists of placing each blue ball within one of the spaces between red balls, all with equal probability (of $1/15$), and doing so in a way that the placement of one blue ball affects no other blue balls, you get the multinomial probabilities.
Jul
5
comment Combinatorics: Mean and Variance of an indicator function of items arranged in a circle.
Perhaps an illustration would help. In the case $n=k=2$ there are two distinguishable configurations modulo symmetries of the circle: RBRB (red-blue-red-blue in counterclockwise order) and RBBR. In the multinomial scenario both are assumed to have probability $1/2$, while in the permutation scenario they are assumed to have probabilities $1/3$ and $2/3$, respectively. Some people have argued that absent any additional information, one should assume the maximum-entropy distribution. That's the multinomial one in this case--not the permutation one!
Jul
5
comment Combinatorics: Mean and Variance of an indicator function of items arranged in a circle.
Yes I can still come up with two (and even more) solutions. In this post I have explained two ways of looking at the situation that are perfectly consistent with the problem. If you disagree, the burden is on you to demonstrate how one (or even both!) violate some explicitly-stated criterion in the problem statement. I would even go so far as to accept that there are some conventional implicit criteria in questions like this one: the obvious one is that the distribution ought to be invariant under symmetries of the configuration. That is the case in both scenarios.
Jul
5
comment Combinatorics: Mean and Variance of an indicator function of items arranged in a circle.
No, that's not right. "Random" is not a quantitative characterization: you have to specify the process involved. The problem statement does not give enough information to "assess" a unique distribution. Both my answers are correct given their assumptions about how the balls are distributed randomly. One of them happens to agree with yours, but that only means it likely adopted the same assumptions you did.
Jul
1
comment How to understand Demartines theorem
This is a minor variant of the Central Limit Theorem, which asserts the standardized version of $||X||_2^2$ is asymptotically Normal. It requires that $X$ have finite variance.
Jun
26
comment Transforming vector elements to element indices
Because matrix operations represent linear transformations and the relationship between Y and X is not linear, no such Q exists.
Jun
18
comment Does affine equivariance implies shape unbiasedness?
Some things to consider: when $\sigma(X)=0$ (the $p\times p$ zero matrix) the procedure is trivially equivariant but obviously not unbiased. More generally, when $\sigma$ is equivariant unbiased and $p\gt 1$, the procedure $\tau(X)=\lambda\sigma(X)$ is also equivariant but when $\lambda^p\ne \lambda$ it must be biased. This suggests that dividing $\sigma(X)$ by its determinant might not be the right definition of "unbiased" to be using.
Apr
7
comment Holder's inequality
Your second statement is not well defined until you assume both $X$ and $Y$ are strictly positive.
Feb
26
comment Lower bound for non-negative definite matrix
The matrix $$\left( \begin{array}{cc} 0 & 1 \\ -1 & 2 \\ \end{array} \right)$$ is non-negative definite but your expression equals $-2$. Perhaps you would like to stipulate that $A$ also be symmetric?
Dec
16
comment Determine the number of revolutions the normal to a curve makes as it moves along a curve in three dimensions?
That is correct, which is why I claimed your question is not well-defined! It only becomes so once you fix a pair of points on the sphere once and for all.
Dec
6
comment If $X$ is independent of $Z$ and $Y$ is dependent with $Z$ is it possible for $X + Y$ to be independent of $Z$?
@Did Thank you for pointing that out!
Dec
6
comment If $X$ is independent of $Z$ and $Y$ is dependent with $Z$ is it possible for $X + Y$ to be independent of $Z$?
If $X+Y$ is independent of $Z$ and $X$ is independent of $Z$, then $(X+Y)-X = Y$ must also be independent of $Z$.
Dec
6
comment Determine the number of revolutions the normal to a curve makes as it moves along a curve in three dimensions?
You can make this question well-defined by first restricting to curves whose normals are never zero, allowing you to define a unit normal everywhere, and assuming this unit normal is a continuous map into the unit sphere. Select two points on the unit sphere not in the image of that map. Use (say) a stereographic projection based at one of them to project the curve into the plane. Define the number of "revolutions" to be the winding number of the (projected) unit normal around the second point.
Dec
6
comment Mapping CDF's to each other
The result is false for discrete distributions. A simple example is given by two Bernoulli variables with parameters $p$ and $p'$ such that $p\ne p'$ and $p\ne 1-p'$.
Oct
25
comment How do you calculate an upper-confidence bound on a problem with 2 means?
That's right--and leave out the left endpoint of the interval altogether.
Oct
25
comment How do you calculate an upper-confidence bound on a problem with 2 means?
This is a nice exposition but it addresses a slightly different problem: the question asks for a UCL of the mean, not a symmetric CI for the mean.