Next Article in Journal
Kumaraswamy Generalized Power Lomax Distributionand Its Applications
Previous Article in Journal
An Upper Bound of the Bias of Nadaraya-Watson Kernel Regression under Lipschitz Assumptions
 
 
Font Type:
Arial Georgia Verdana
Font Size:
Aa Aa Aa
Line Spacing:
Column Width:
Background:
Article

General Formulas for the Central and Non-Central Moments of the Multinomial Distribution

by
Frédéric Ouimet
Department of Physics, Mathematics and Astronomy, California Institute of Technology, Pasadena, CA 91125, USA
Stats 2021, 4(1), 18-27; https://doi.org/10.3390/stats4010002
Submission received: 10 November 2020 / Revised: 31 December 2020 / Accepted: 5 January 2021 / Published: 6 January 2021
(This article belongs to the Section Multivariate Analysis)

Abstract

:
We present the first general formulas for the central and non-central moments of the multinomial distribution, using a combinatorial argument and the factorial moments previously obtained in Mosimann (1962). We use the formulas to give explicit expressions for all the non-central moments up to order 8 and all the central moments up to order 4. These results expand significantly on those in Newcomer (2008) and Newcomer et al. (2008), where the non-central moments were calculated up to order 4.

1. The Multinomial Distribution

For any d N , the d-dimensional (unit) simplex is defined by S : = x [ 0 , 1 ] d : i = 1 d x i 1 , and the probability mass function k P k , m ( x ) for ξ : = ( ξ 1 , ξ 2 , , ξ d ) Multinomial ( m , x ) is defined by
P k , m ( x ) : = m ! ( m i = 1 d k i ) ! i = 1 d k i ! · ( 1 i = 1 d x i ) m i = 1 d k i i = 1 d x i k i , k N 0 d m S ,
where m N and x S . If x d + 1 : = 1 i = 1 d x i , then (1) is just a reparametrization of ( ξ , 1 i = 1 d ξ i ) Multinomial ( m , ( x , x d + 1 ) ) where i = 1 d + 1 x i = 1 . In this paper, our main goal is to give general formulas for the non-central and central moments of (1), namely
E i = 1 d ξ i p i and E i = 1 d ( ξ i E [ ξ i ] ) p i , p 1 , p 2 , , p d N 0 .
We obtain the formulas using a combinatorial argument and the general expression for the factorial moments found in Mosimann (1962) [1], which we register in the lemma below.
Lemma 1
(Factorial moments). Let ξ Multinomial ( m , x ) . Then, for all k 1 , k 2 , , k d N 0 ,
E i = 1 d ξ i ( k i ) = m ( i = 1 d k i ) i = 1 d x k i ,
where m ( k ) : = m ( m 1 ) ( m k + 1 ) denotes the k-th order falling factorial of m.
The formulas that we develop for the expectations in (2) will be used to compute explicitly all the non-central moments up to order 8 and all the central moments up to order 4, which expands on the third and fourth order non-central moments that were previously calculated in (Newcomer (2008) [2], Appendix A.1). We should also mention that explicit formulas for several lower-order (mixed) cumulants were presented in Wishart (1949) [3] (see also Johnson et al. (1997) [4], page 37), but not for the moments.

2. Motivation

To the best of our knowledge, general formulas for the central and non-central moments of the multinomial distribution have never been derived in the literature. The central moments can arise naturally, for example, when studying asymptotic properties, via Taylor expansions of statistical estimators involving the multinomial distribution. For a given sequence of i.i.d. observations X 1 , X 2 , , X n , two examples of such estimators are the Bernstein estimator for the cumulative distribution function
F n , m ( x ) : = k N 0 d m S 1 n i = 1 n 𝟙 ( , k m ] ( X i ) P k , m ( x ) , x S , m , n N ,
and the Bernstein estimator for the density function (also called smoothed histogram)
f ^ n , m ( x ) : = k N 0 d ( m 1 ) S m d n i = 1 n 𝟙 ( k m , k + 1 m ] ( X i ) P k , m 1 ( x ) , x S , m , n N ,
over the d-dimensional simplex. Some of their asymptotic properties were investigated by Vitale (1975), Stadtmüller (1986), Tenbusch (1997), Petrone (1999), Ghosal (2001), Petrone and Wasserman (2002), Babu et al. (2002), Kakizawa (2004), Bouezmarni and Rolin (2007), Bouezmarni et al. (2007), Leblanc (2009, 2010, 2012), Curtis and Ghosh (2011), Igarashi and Kakizawa (2014), Turnbull and Ghosh (2014), Lu (2015), Guan (2016, 2017) and Belalia et al. (2017, 2019) [5,6,7,8,9,10,11,12,13,14,15,16,17,18,19,20,21,22,23,24,25,26,27,28,29,30] when d = 1 , by Tenbusch (1994) [31] when d = 2 , and by Ouimet (2020) [32,33] for all d 1 , using a local limit theorem from Ouimet (2020) [34] for the multinomial distribution (see also Arenbaev (1976) [35]). The estimator (5) is a discrete analogue of the Dirichlet kernel estimator introduced by Aitchison and Lauder (1985) [36] and studied theoretically in Brown and Chen (1999), Chen (1999, 2000) and Bouezmarni and Rolin (2003) [37,38,39,40] when d = 1 (among others), and in Ouimet (2020) [41] for all d 1 .

3. Results

First, we give a general formula of the non-central moments of the multinomial distribution in (1).
Theorem 1
(Non-central moments). Let ξ Multinomial ( m , x ) . For all p 1 , p 2 , , p d N 0 ,
E i = 1 d ξ i p i = k 1 = 0 p 1 k d = 0 p d m ( i = 1 d k i ) i = 1 d p i k i x i k i ,
where p k denotes a Stirling number of the second kind (i.e., the number of ways to partition a set of p objects into k non-empty subsets).
Proof. 
We have the following well-known relation between the power p N 0 of a number x R and the falling factorials of x:
x p = k = 0 p p k x ( k ) .
See, e.g., (Graham et al. (1994) [42], page 262). Apply this relation to every ξ i p i and use the linearity of the expectation to get
E i = 1 d ξ i p i = k 1 = 0 p 1 k d = 0 p d p 1 k 1 p d k d E i = 1 d ξ i ( k i ) .
The conclusion follows from Lemma 1.    □
We deduce a general formula for the central moments of the multinomial distribution.
Theorem 2
(Central moments). Let ξ Multinomial ( m , x ) . For all p 1 , p 2 , , p d N 0 ,
E i = 1 d ( ξ i E [ ξ i ] ) p i = 1 = 0 p 1 d = 0 p d k 1 = 0 1 k d = 0 d m ( i = 1 d k i ) ( m ) i = 1 d ( p i i ) i = 1 d p i i i k i x i p i i + k i ,
where p denotes the binomial coefficient p ! ! ( p ) ! .
Proof. 
By applying the binomial formula to each factor ( ξ i E [ ξ i ] ) p i and using the fact that E [ ξ i ] = m x i for all i { 1 , 2 , , d } , note that
E i = 1 d ( ξ i E [ ξ i ] ) p i = 1 = 0 p 1 d = 0 p d E i = 1 d ξ i i · i = 1 d p i i ( m x i ) p i i .
The conclusion follows from Theorem 1.    □

4. Numerical Implementation

The formulas in Theorems 1 and  2 can be implemented in Mathematica as follows:
  NonCentral[m_, x_, p_, d_] :=
    Sum[FactorialPower[m, Sum[k[i], {i,1,d}]] *
    Product[StirlingS2[p[[i]], k[i]] * x[[i]] ^ k[i], {i,1,d}], ##] & @@
    ({k[#], 0, p[[#]]} & /@ Range[d]);
  Central[m_, x_, p_, d_] :=
    Sum[Sum[FactorialPower[m, Sum[k[i], {i,1,d}]] *
    (-m) ^ Sum[p[[i]] - ell[i], {i,1,d}] *
    Product[Binomial[p[[i]], ell[i]] * StirlingS2[ell[i], k[i]] *
    x[[i]] ^ (p[[i]] - ell[i] + k[i]), {i,1,d}], ##] & @@
    ({k[#], 0, ell[#]} & /@ Range[d]), ##] & @@
    ({ell[#], 0, p[[#]]} & /@ Range[d]);

5. Explicit Formulas

In Newcomer (2008) [2], explicit expressions for the non-central moments of order 3 and 4 were obtained for the multinomial distribution, see also Newcomer et al. (2008) and Ouimet (2020) [43,44]. To expand on those results, we use the formula from Theorem 1 in the two subsections below to calculate (explicitly) all the non-central moments up to order 8 and all the central moments up to order 4.
Here is a table of the Stirling numbers of the second kind that we will use in our calculations:
0 0 = 1 , 1 0 = 0 , 1 1 = 1 , 2 0 = 0 , 2 1 = 1 , 2 2 = 1 , 3 0 = 0 , 3 1 = 1 , 3 2 = 3 , 3 3 = 1 , 4 0 = 0 , 4 1 = 1 , 4 2 = 7 , 4 3 = 6 , 4 4 = 1 , 5 0 = 0 , 5 1 = 1 , 5 2 = 15 , 5 3 = 25 , 5 4 = 10 , 5 5 = 1 , 6 0 = 0 , 6 1 = 1 , 6 2 = 31 , 6 3 = 90 , 6 4 = 65 , 6 5 = 15 , 6 6 = 1 , 7 0 = 0 , 7 1 = 1 , 7 2 = 63 , 7 3 = 301 , 7 4 = 350 , 7 5 = 140 , 7 6 = 21 , 7 7 = 1 , 8 0 = 0 , 8 1 = 1 , 8 2 = 127 , 8 3 = 966 , 8 4 = 1701 , 8 5 = 1050 , 8 6 = 266 , 8 7 = 28 , 8 8 = 1 .

5.1. Computation of the Non-Central Moments Up to Order 8

By applying the general expression in Theorem 1 and by removing the Stirling numbers p i k i that are equal to 0, we get the following results directly.
Order 1: For any j 1 { 1 , 2 , , d } ,
E [ ξ j 1 ] = x j 1 m .
Order 2: For any distinct j 1 , j 2 { 1 , 2 , , d } ,
E [ ξ j 1 2 ] = x j 1 [ m + m ( 2 ) x j 1 ] ,
E [ ξ j 1 ξ j 2 ] = x j 1 x j 2 m ( 2 ) .
Order 3: For any distinct j 1 , j 2 , j 3 { 1 , 2 , , d } ,
E [ ξ j 1 3 ] = x j 1 [ m + 3 m ( 2 ) x j 1 + m ( 3 ) x j 1 2 ] ,
E [ ξ j 1 2 ξ j 2 ] = x j 1 x j 2 [ m ( 2 ) + m ( 3 ) x j 1 ] ,
E [ ξ j 1 ξ j 2 ξ j 3 ] = x j 1 x j 2 x j 3 m ( 3 ) .
Order 4: For any distinct j 1 , j 2 , j 3 , j 4 { 1 , 2 , , d } ,
E [ ξ j 1 4 ] = x j 1 [ m + 7 m ( 2 ) x j 1 + 6 m ( 3 ) x j 1 2 + m ( 4 ) x j 1 3 ] ,
E [ ξ j 1 3 ξ j 2 ] = x j 1 x j 2 [ m ( 2 ) + 3 m ( 3 ) x j 1 + m ( 4 ) x j 1 2 ] ,
E [ ξ j 1 2 ξ j 2 2 ] = x j 1 x j 2 [ m ( 2 ) + m ( 3 ) ( x j 1 + x j 2 ) + m ( 4 ) x j 1 x j 2 ] ,
E [ ξ j 1 2 ξ j 2 ξ j 3 ] = x j 1 x j 2 x j 3 ] m ( 3 ) + m ( 4 ) x j 1 ] ,
E [ ξ j 1 ξ j 2 ξ j 3 ξ j 4 ] = x j 1 x j 2 x j 3 x j 4 m ( 4 ) .
Order 5: For any distinct j 1 , j 2 , j 3 , j 4 , j 5 { 1 , 2 , , d } ,
E [ ξ j 1 5 ] = x j 1 [ m + 15 m ( 2 ) x j 1 + 25 m ( 3 ) x j 1 2 + 10 m ( 4 ) x j 1 3 + m ( 5 ) x j 1 4 ] ,
E [ ξ j 1 4 ξ j 2 ] = x j 1 x j 2 [ m ( 2 ) + 7 m ( 3 ) x j 1 + 6 m ( 4 ) x j 1 2 + m ( 5 ) x j 1 3 ] ,
E [ ξ j 1 3 ξ j 2 2 ] = x j 1 x j 2 [ m ( 2 ) + m ( 3 ) ( 3 x j 1 + x j 2 ) + m ( 4 ) ( x j 1 2 + 3 x j 1 x j 2 ) + m ( 5 ) x j 1 2 x j 2 ] ,
E [ ξ j 1 3 ξ j 2 ξ j 3 ] = x j 1 x j 2 x j 3 [ m ( 3 ) + 3 m ( 4 ) x j 1 + m ( 5 ) x j 1 2 ] ,
E [ ξ j 1 2 ξ j 2 2 ξ j 3 ] = x j 1 x j 2 x j 3 [ m ( 3 ) + m ( 4 ) ( x j 1 + x j 2 ) + m ( 5 ) x j 1 x j 2 ] ,
E [ ξ j 1 2 ξ j 2 ξ j 3 ξ j 4 ] = x j 1 x j 2 x j 3 x j 4 [ m ( 4 ) + m ( 5 ) x j 1 ] ,
E [ ξ j 1 ξ j 2 ξ j 3 ξ j 4 ξ j 5 ] = x j 1 x j 2 x j 3 x j 4 x j 5 m ( 5 ) .
Order 6: For any distinct j 1 , j 2 , j 3 , j 4 , j 5 , j 6 { 1 , 2 , , d } ,
E [ ξ j 1 6 ] = x j 1 [ m + 31 m ( 2 ) x j 1 + 90 m ( 3 ) x j 1 2 + 65 m ( 4 ) x j 1 3 + 15 m ( 5 ) x j 1 4 + m ( 6 ) x j 1 5 ] ,
E [ ξ j 1 5 ξ j 2 ] = x j 1 x j 2 [ m ( 2 ) + 15 m ( 3 ) x j 1 + 25 m ( 4 ) x j 1 2 + 10 m ( 5 ) x j 1 3 + m ( 6 ) x j 1 4 ] ,
E [ ξ j 1 4 ξ j 2 2 ] = x j 1 x j 2 m ( 2 ) + m ( 3 ) ( 7 x j 1 + x j 2 ) + m ( 4 ) ( 6 x j 1 2 + 7 x j 1 x j 2 ) + m ( 5 ) ( x j 1 3 + 6 x j 1 2 x j 2 ) + m ( 6 ) x j 1 3 x j 2 ,
E [ ξ j 1 4 ξ j 2 ξ j 3 ] = x j 1 x j 2 x j 3 [ m ( 3 ) + 7 m ( 4 ) x j 1 + 6 m ( 5 ) x j 1 2 + m ( 6 ) x j 1 3 ] ,
E [ ξ j 1 3 ξ j 2 3 ] = x j 1 x j 2 m ( 2 ) + m ( 3 ) ( 3 x j 1 + 3 x j 2 ) + m ( 4 ) ( x j 1 2 + 9 x j 1 x j 2 + x j 2 2 ) + m ( 5 ) ( 3 x j 1 2 x j 2 + 3 x j 1 x j 2 2 ) + m ( 6 ) x j 1 2 x j 2 2 ,
E [ ξ j 1 3 ξ j 2 2 ξ j 3 ] = x j 1 x j 2 x j 3 [ m ( 3 ) + m ( 4 ) ( 3 x j 1 + x j 2 ) + m ( 5 ) ( x j 1 2 + 3 x j 1 x j 2 ) + m ( 6 ) x j 1 2 x j 2 ] ,
E [ ξ j 1 3 ξ j 2 ξ j 3 ξ j 4 ] = x j 1 x j 2 x j 3 x j 4 [ m ( 4 ) + 3 m ( 5 ) x j 1 + m ( 6 ) x j 1 2 ] ,
E [ ξ j 1 2 ξ j 2 2 ξ j 3 2 ] = x j 1 x j 2 x j 3 m ( 3 ) + m ( 4 ) ( x j 1 + x j 2 + x j 3 ) + m ( 5 ) ( x j 1 x j 2 + x j 1 x j 3 + x j 2 x j 3 ) + m ( 6 ) x j 1 x j 2 x j 3 ,
E [ ξ j 1 2 ξ j 2 2 ξ j 3 ξ j 4 ] = x j 1 x j 2 x j 3 x j 4 [ m ( 4 ) + m ( 5 ) ( x j 1 + x j 2 ) + m ( 6 ) x j 1 x j 2 ] ,
E [ ξ j 1 2 ξ j 2 ξ j 3 ξ j 4 ξ j 5 ] = x j 1 x j 2 x j 3 x j 4 x j 5 [ m ( 5 ) + m ( 6 ) x j 1 ] ,
E [ ξ j 1 ξ j 2 ξ j 3 ξ j 4 ξ j 5 ξ j 6 ] = x j 1 x j 2 x j 3 x j 4 x j 5 x j 6 m ( 6 ) .
Order 7: For any distinct j 1 , j 2 , j 3 , j 4 , j 5 , j 6 , j 7 { 1 , 2 , , d } ,
E [ ξ j 1 7 ] = x j 1 m + 63 m ( 2 ) x j 1 + 301 m ( 3 ) x j 1 2 + 350 m ( 4 ) x j 1 3 + 140 m ( 5 ) x j 1 4 + 21 m ( 6 ) x j 1 5 + m ( 7 ) x j 1 6 ,
E [ ξ j 1 6 ξ j 2 ] = x j 1 x j 2 m ( 2 ) + 31 m ( 3 ) x j 1 + 90 m ( 4 ) x j 1 2 + 65 m ( 5 ) x j 1 3 + 15 m ( 6 ) x j 1 4 + m ( 7 ) x j 1 5 ,
E [ ξ j 1 5 ξ j 2 2 ] = x j 1 x j 2 m ( 2 ) + m ( 3 ) ( 15 x j 1 + x j 2 ) + m ( 4 ) ( 25 x j 1 2 + 15 x j 1 x j 2 ) + m ( 5 ) ( 10 x j 1 3 + 25 x j 1 2 x j 2 ) + m ( 6 ) ( x j 1 4 + 10 x j 1 3 x j 2 ) + m ( 7 ) x j 1 4 x j 2 ,
E [ ξ j 1 5 ξ j 2 ξ j 3 ] = x j 1 x j 2 x j 3 [ m ( 3 ) + 15 m ( 4 ) x j 1 + 25 m ( 5 ) x j 1 2 + 10 m ( 6 ) x j 1 3 + m ( 7 ) x j 1 4 ] ,
E [ ξ j 1 4 ξ j 2 3 ] = x j 1 x j 2 m ( 2 ) + m ( 3 ) ( 7 x j 1 + 3 x j 2 ) + m ( 4 ) ( 6 x j 1 2 + 21 x j 1 x j 2 + x j 2 2 ) + m ( 5 ) ( x j 1 3 + 18 x j 1 2 x j 2 + 7 x j 1 x j 2 2 ) + m ( 6 ) ( 3 x j 1 3 x j 2 + 6 x j 1 2 x j 2 2 ) + m ( 7 ) x j 1 3 x j 2 2 ,
E [ ξ j 1 4 ξ j 2 2 ξ j 3 ] = x j 1 x j 2 x j 3 m ( 3 ) + m ( 4 ) ( 7 x j 1 + x j 2 ) + m ( 5 ) ( 6 x j 1 2 + 7 x j 1 x j 2 ) + m ( 6 ) ( x j 1 3 + 6 x j 1 2 x j 2 ) + m ( 7 ) x j 1 3 x j 2 ,
E [ ξ j 1 4 ξ j 2 ξ j 3 ξ j 4 ] = x j 1 x j 2 x j 3 x j 4 [ m ( 4 ) + 7 m ( 5 ) x j 1 + 6 m ( 6 ) x j 1 2 + m ( 7 ) x j 1 3 ] ,
E [ ξ j 1 3 ξ j 2 3 ξ j 3 ] = x j 1 x j 2 x j 3 m ( 3 ) + m ( 4 ) ( 3 x j 1 + 3 x j 2 ) + m ( 5 ) ( x j 1 2 + 9 x j 1 x j 2 + x j 2 2 ) + m ( 6 ) ( 3 x j 1 2 x j 2 + 3 x j 1 x j 2 2 ) + m ( 7 ) x j 1 2 x j 2 2 ,
E [ ξ j 1 3 ξ j 2 2 ξ j 3 2 ] = x j 1 x j 2 x j 3 m ( 3 ) + m ( 4 ) ( 3 x j 1 + x j 2 + x j 3 ) + m ( 5 ) ( x j 1 2 + 3 x j 1 x j 2 + 3 x j 1 x j 3 + x j 2 x j 3 ) + m ( 6 ) ( x j 1 2 x j 2 + x j 1 2 x j 3 + 3 x j 1 x j 2 x j 3 ) + m ( 7 ) x j 1 2 x j 2 x j 3 ,
E [ ξ j 1 3 ξ j 2 2 ξ j 3 ξ j 4 ] = x j 1 x j 2 x j 3 x j 4 m ( 4 ) + m ( 5 ) ( 3 x j 1 + x j 2 ) + m ( 6 ) ( x j 1 2 + 3 x j 1 x j 2 ) + m ( 7 ) x j 1 2 x j 2 ,
E [ ξ j 1 3 ξ j 2 ξ j 3 ξ j 4 ξ j 5 ] = x j 1 x j 2 x j 3 x j 4 x j 5 [ m ( 5 ) + 3 m ( 6 ) x j 1 + m ( 7 ) x j 1 2 ] ,
E [ ξ j 1 2 ξ j 2 2 ξ j 3 2 ξ j 4 ] = x j 1 x j 2 x j 3 x j 4 m ( 4 ) + m ( 5 ) ( x j 1 + x j 2 + x j 3 ) + m ( 6 ) ( x j 1 x j 2 + x j 1 x j 3 + x j 2 x j 3 ) + m ( 7 ) x j 1 x j 2 x j 3 ,
E [ ξ j 1 2 ξ j 2 2 ξ j 3 ξ j 4 ξ j 5 ] = x j 1 x j 2 x j 3 x j 4 x j 5 m ( 5 ) + m ( 6 ) ( x j 1 + x j 2 ) + m ( 7 ) x j 1 x j 2 ,
E [ ξ j 1 2 ξ j 2 ξ j 3 ξ j 4 ξ j 5 ξ j 6 ] = x j 1 x j 2 x j 3 x j 4 x j 5 x j 6 [ m ( 6 ) + m ( 7 ) x j 1 ] ,
E [ ξ j 1 ξ j 2 ξ j 3 ξ j 4 ξ j 5 ξ j 6 ξ j 7 ] = x j 1 x j 2 x j 3 x j 4 x j 5 x j 6 x j 7 m ( 7 ) .
Order 8: For any distinct j 1 , j 2 , j 3 , j 4 , j 5 , j 6 , j 7 , j 8 { 1 , 2 , , d } ,
E [ ξ j 1 8 ] = x j 1 m + 127 m ( 2 ) x j 1 + 966 m ( 3 ) x j 2 2 + 1701 m ( 4 ) x j 1 3 + 1050 m ( 5 ) x j 1 4 + 266 m ( 6 ) x j 1 5 + 28 m ( 7 ) x j 1 6 + m ( 8 ) x j 1 7 ,
E [ ξ j 1 7 ξ j 2 ] = x j 1 x j 2 m + 63 m ( 3 ) x j 1 + 301 m ( 4 ) x j 1 2 + 350 m ( 5 ) x j 1 3 + 140 m ( 6 ) x j 1 4 + 21 m ( 7 ) x j 1 5 + m ( 8 ) x j 1 6 ,
E [ ξ j 1 6 ξ j 2 2 ] = x j 1 x j 2 m ( 2 ) + m ( 3 ) ( 31 x j 1 + x j 2 ) + m ( 4 ) ( 90 x j 1 2 + 31 x j 1 x j 2 ) + m ( 5 ) ( 65 x j 1 3 + 90 x j 1 2 x j 2 ) + m ( 6 ) ( 15 x j 1 4 + 65 x j 1 3 x j 2 ) + m ( 7 ) ( x j 1 5 + 15 x j 1 4 x j 2 ) + m ( 8 ) x j 1 5 x j 2 ,
E [ ξ j 1 6 ξ j 2 ξ j 3 ] = x j 1 x j 2 x j 3 m ( 3 ) + 31 m ( 4 ) x j 1 + 90 m ( 5 ) x j 1 2 + 65 m ( 6 ) x j 1 3 + 15 m ( 7 ) x j 1 4 + m ( 8 ) x j 1 5 ,
E [ ξ j 1 5 ξ j 2 3 ] = x j 1 x j 2 m ( 2 ) + m ( 3 ) ( 15 x j 1 + 3 x j 2 ) + m ( 4 ) ( 25 x j 1 2 + 45 x j 1 x j 2 + x j 2 2 ) + m ( 5 ) ( 10 x j 1 3 + 75 x j 1 2 x j 2 + 15 x j 1 x j 2 2 ) + m ( 6 ) ( x j 4 4 + 30 x j 1 3 x j 2 + 25 x j 1 2 x j 2 2 ) + m ( 7 ) ( 3 x j 1 4 x j 2 + 10 x j 1 3 x j 2 2 ) + m ( 8 ) x j 1 4 x j 2 2 ,
E [ ξ j 1 5 ξ j 2 2 ξ j 3 ] = x j 1 x j 2 x j 3 m ( 3 ) + m ( 4 ) ( 15 x j 1 + x j 2 ) + m ( 5 ) ( 25 x j 1 2 + 15 x j 1 x j 2 ) + m ( 6 ) ( 10 x j 1 3 + 25 x j 1 2 x j 2 ) + m ( 7 ) ( x j 1 4 + 10 x j 1 3 x j 2 ) + m ( 8 ) x j 1 4 x j 2 ,
E [ ξ j 1 5 ξ j 2 ξ j 3 ξ j 4 ] = x j 1 x j 2 x j 3 x j 4 [ m ( 4 ) + 15 m ( 5 ) x j 1 + 25 m ( 6 ) x j 1 2 + 10 m ( 7 ) x j 1 3 + m ( 8 ) x j 1 4 ] ,
E [ ξ j 1 4 ξ j 2 4 ] = x j 1 x j 2 m ( 2 ) + m ( 3 ) ( 7 x j 1 + 7 x j 2 ) + m ( 4 ) ( 6 x j 1 2 + 49 x j 1 x j 2 + 6 x j 2 2 ) + m ( 5 ) ( x j 1 3 + 42 x j 1 2 x j 2 + 42 x j 1 x j 2 2 + x j 2 3 ) + m ( 6 ) ( 7 x j 1 3 x j 2 + 36 x j 1 2 x j 2 2 + 7 x j 1 x j 2 3 ) + m ( 7 ) ( 6 x j 1 3 x j 2 2 + 6 x j 1 2 x j 2 3 ) + m ( 8 ) x j 1 3 x j 2 3 ,
E [ ξ j 1 4 ξ j 2 3 ξ j 3 ] = x j 1 x j 2 x j 3 m ( 3 ) + m ( 4 ) ( 7 x j 1 + 3 x j 2 ) + m ( 5 ) ( 6 x j 1 2 + 21 x j 1 x j 2 + x j 2 2 ) + m ( 6 ) ( x j 1 3 + 18 x j 1 2 x j 2 + 7 x j 1 x j 2 2 ) + m ( 7 ) ( 3 x j 1 3 x j 2 + 6 x j 1 2 x j 2 2 ) + m ( 8 ) x j 1 3 x j 2 2 ,
E [ ξ j 1 4 ξ j 2 2 ξ j 3 2 ] = x j 1 x j 2 x j 3 m ( 3 ) + m ( 4 ) ( 7 x j 1 + x j 2 + x j 3 ) + m ( 5 ) ( 6 x j 1 2 + 7 x j 1 x j 2 + 7 x j 1 x j 3 + x j 2 x j 3 ) + m ( 6 ) ( x j 1 3 + 6 x j 1 2 x j 2 + 6 x j 1 2 x j 3 + 7 x j 1 x j 2 x j 3 ) + m ( 7 ) ( x j 1 3 x j 2 + x j 1 3 x j 3 + 6 x j 1 2 x j 2 x j 3 ) + m ( 8 ) x j 1 3 x j 2 x j 3 ,
E [ ξ j 1 4 ξ j 2 2 ξ j 3 ξ j 4 ] = x j 1 x j 2 x j 3 x j 4 m ( 4 ) + m ( 5 ) ( 7 x j 1 + x j 2 ) + m ( 6 ) ( 6 x j 1 2 + 7 x j 1 x j 2 ) + m ( 7 ) ( x j 1 3 + 6 x j 1 2 x j 2 ) + m ( 8 ) x j 1 3 x j 2 ,
E [ ξ j 1 4 ξ j 2 ξ j 3 ξ j 4 ξ j 5 ] = x j 1 x j 2 x j 3 x j 4 x j 5 [ m ( 5 ) + 7 m ( 6 ) x j 1 + 6 m ( 7 ) x j 1 2 + m ( 8 ) x j 1 3 ] ,
E [ ξ j 1 3 ξ j 2 3 ξ j 3 2 ] = x j 1 x j 2 x j 3 m ( 3 ) + m ( 4 ) ( 3 x j 1 + 3 x j 2 + x j 3 ) + m ( 5 ) ( x j 1 2 + x j 2 2 + 3 x j 1 x j 3 + 3 x j 2 x j 3 + 9 x j 1 x j 2 ) + m ( 6 ) ( x j 1 2 x j 3 + x j 2 2 x j 3 + 3 x j 1 2 x j 2 + 3 x j 1 x j 2 2 + 9 x j 1 x j 2 x j 3 ) + m ( 7 ) ( x j 1 2 x j 2 2 + 3 x j 1 2 x j 2 x j 3 + 3 x j 1 x j 2 2 x j 3 ) + m ( 8 ) x j 1 2 x j 2 2 x j 3 ,
E [ ξ j 1 3 ξ j 2 3 ξ j 3 ξ j 4 ] = x j 1 x j 2 x j 3 x j 4 m ( 4 ) + m ( 5 ) ( 3 x j 1 + 3 x j 2 ) + m ( 6 ) ( x j 1 2 + 9 x j 1 x j 2 + x j 2 2 ) + m ( 7 ) ( 3 x j 1 2 x j 2 + 3 x j 1 x j 2 2 ) + m ( 8 ) x j 1 2 x j 2 2 ,
E [ ξ j 1 3 ξ j 2 2 ξ j 3 2 ξ j 4 ] = x j 1 x j 2 x j 3 x j 4 m ( 4 ) + m ( 5 ) ( 3 x j 1 + x j 2 + x j 3 ) + m ( 6 ) ( 3 x j 1 x j 2 + 3 x j 1 x j 3 + x j 2 x j 3 ) + m ( 7 ) ( x j 1 2 x j 2 + x j 1 2 x j 3 + 3 x j 1 x j 2 x j 3 ) + m ( 8 ) x j 1 2 x j 2 x j 3 ,
E [ ξ j 1 3 ξ j 2 2 ξ j 3 ξ j 4 ξ j 5 ] = x j 1 x j 2 x j 3 x j 4 x j 5 m ( 5 ) + m ( 6 ) ( 3 x j 1 + x j 2 ) + m ( 7 ) ( x j 1 2 + 3 x j 1 x j 2 ) + m ( 8 ) x j 1 2 x j 2 ,
E [ ξ j 1 3 ξ j 2 ξ j 3 ξ j 4 ξ j 5 ξ j 6 ] = x j 1 x j 2 x j 3 x j 4 x j 5 x j 6 [ m ( 6 ) + 3 m ( 7 ) x j 1 + m ( 8 ) x j 1 2 ] ,
E [ ξ j 1 2 ξ j 2 2 ξ j 3 2 ξ j 4 2 ] = x j 1 x j 2 x j 3 x j 4 m ( 4 ) + m ( 5 ) ( x j 1 + x j 2 + x j 3 + x j 4 ) + m ( 6 ) ( x j 1 x j 2 + x j 1 x j 3 + x j 1 x j 4 + x j 2 x j 3 + x j 2 x j 4 + x j 3 x j 4 ) + m ( 7 ) ( x j 1 x j 2 x j 3 + x j 1 x j 2 x j 4 + x j 1 x j 3 x j 4 + x j 2 x j 3 x j 4 ) + m ( 8 ) x j 1 x j 2 x j 3 x j 4 ,
E [ ξ j 1 2 ξ j 2 2 ξ j 3 2 ξ j 4 ξ j 5 ] = x j 1 x j 2 x j 3 x j 4 x j 5 m ( 5 ) + m ( 6 ) ( x j 1 + x j 2 + x j 3 ) + m ( 7 ) ( x j 1 x j 2 + x j 1 x j 3 + x j 2 x j 3 ) + m ( 8 ) x j 1 x j 2 x j 3 ,
E [ ξ j 1 2 ξ j 2 2 ξ j 3 ξ j 4 ξ j 5 ξ j 6 ] = x j 1 x j 2 x j 3 x j 4 x j 5 x j 6 [ m ( 6 ) + m ( 7 ) ( x j 1 + x j 2 ) + m ( 8 ) x j 1 x j 2 ] ,
E [ ξ j 1 2 ξ j 2 ξ j 3 ξ j 4 ξ j 5 ξ j 6 ξ j 7 ] = x j 1 x j 2 x j 3 x j 4 x j 5 x j 6 x j 7 [ m ( 7 ) + m ( 8 ) x j 1 ] ,
E [ ξ j 1 ξ j 2 ξ j 3 ξ j 4 ξ j 5 ξ j 6 ξ j 7 ξ j 8 ] = x j 1 x j 2 x j 3 x j 4 x j 5 x j 6 x j 7 x j 8 m ( 8 ) .

5.2. Computation of the Central Moments Up to Order 4

With the results of the previous subsection and some algebraic manipulations (or the formula in Theorem 2), we can now calculate the central moments explicitly. We could calculate them up to order 8, but it would be very tedious. Instead, we write them up to order 4 for the sake of brevity. The simplifications we make to obtain the boxed expressions below are done with Mathematica.
Order 2: For any distinct j 1 , j 2 { 1 , 2 , , d } ,
E [ ( ξ j 1 E [ ξ j 1 ] ) 2 ] = E [ ξ j 1 2 ] ( E [ ξ j 1 ] ) 2 = x j 1 [ m + m ( 2 ) x j 1 ] m 2 x j 1 2 = m x j 1 ( 1 x j 1 )
E [ ( ξ j 1 E [ ξ j 1 ] ) ( ξ j 2 E [ ξ j 2 ] ) ] = E [ ξ j 1 ξ j 2 ] E [ ξ j 1 ] E [ ξ j 2 ] = m ( 2 ) x j 1 x j 2 m x j 1 m x j 2 = m x j 1 x j 2 .
Order 3: For any distinct j 1 , j 2 , j 3 { 1 , 2 , , d } ,
E [ ( ξ j 1 E [ ξ j 1 ] ) 3 ] = E [ ξ j 1 3 ] 3 E [ ξ j 1 2 ] E [ ξ j 1 ] + 2 ( E [ ξ j 1 ] ) 3 = x j 1 [ m + 3 m ( 2 ) x j 1 + m ( 3 ) x j 1 2 ] 3 x j 1 [ m + m ( 2 ) x j 1 ] m x j 1 + 2 m 3 x j 1 3 = m x j 1 ( x j 1 1 ) ( 2 x j 1 1 )
E [ ( ξ j 1 E [ ξ j 1 ] ) 2 ( ξ j 2 E [ ξ j 2 ] ) ] = E [ ξ j 1 2 ξ j 2 ] E [ ξ j 1 2 ] E [ ξ j 2 ] 2 E [ ξ j 1 ξ j 2 ] E [ ξ j 1 ] + 2 ( E [ ξ j 1 ] ) 2 E [ ξ j 2 ] = x j 1 x j 2 [ m ( 2 ) + m ( 3 ) x j 1 ] x j 1 m + m ( 2 ) x j 1 m x j 2 2 m ( 2 ) x j 1 x j 2 m x j 1 + 2 m 2 x j 1 2 m x j 2 = m x j 1 x j 2 ( 2 x j 1 1 )
E [ ( ξ j 1 E [ ξ j 1 ] ) ( ξ j 2 E [ ξ j 2 ] ) ( ξ j 3 E [ ξ j 3 ] ) ] = E [ ξ j 1 ξ j 2 ξ j 3 ] E [ ξ j 1 ξ j 2 ] E [ ξ j 3 ] E [ ξ j 1 ξ j 3 ] E [ ξ j 2 ] E [ ξ j 2 ξ j 3 ] E [ ξ j 1 ] + 2 E [ ξ j 1 ] E [ ξ j 2 ] E [ ξ j 3 ] = m ( 3 ) x j 1 x j 2 x j 3 m ( 2 ) x j 1 x j 2 m x j 3 m ( 2 ) x j 1 x j 3 m x j 2 m ( 2 ) x j 2 x j 3 m x j 1 + 2 m 3 x j 1 x j 2 x j 3 = 2 m x j 1 x j 2 x j 3 .
E [ ( ξ j 1 E [ ξ j 1 ] ) 4 ] = E [ ξ j 1 4 ] 4 E [ ξ j 1 3 ] E [ ξ j 1 ] + 6 E [ ξ j 1 2 ] ( E [ ξ j 1 ] ) 2 3 ( E [ ξ j 1 ] ) 4 = x j 1 [ m + 7 m ( 2 ) x j 1 + 6 m ( 3 ) x j 1 2 + m ( 4 ) x j 1 3 ] 4 x j 1 [ m + 3 m ( 2 ) x j 1 + m ( 3 ) x j 1 2 ] m x j 1 + 6 x j 1 [ m + m ( 2 ) x j 1 ] ( m x j 1 ) 2 3 m 4 x j 1 4 = 3 m 2 x j 1 2 ( x j 1 1 ) 2 + m x j 1 ( 1 x j 1 ) ( 6 x j 1 2 6 x j 1 + 1 )
Order 4: For any distinct j 1 , j 2 , j 3 , j 4 { 1 , 2 , , d } ,
E [ ( ξ j 1 E [ ξ j 1 ] ) 3 ( ξ j 2 E [ ξ j 2 ] ) ] = E [ ξ j 1 3 ξ j 2 ] E [ ξ j 1 3 ] E [ ξ j 2 ] 3 E [ ξ j 1 2 ξ j 2 ] E [ ξ j 1 ] + 3 E [ ξ j 1 2 ] E [ ξ j 1 ] E [ ξ j 2 ] + 3 E [ ξ j 1 ξ j 2 ] ( E [ ξ j 1 ] ) 2 3 ( E [ ξ j 1 ] ) 3 E [ ξ j 2 ] = x j 1 x j 2 [ m ( 2 ) + 3 m ( 3 ) x j 1 + m ( 4 ) x j 1 2 ] x j 1 [ m + 3 m ( 2 ) x j 1 + m ( 3 ) x j 1 2 ] m x j 2 3 x j 1 x j 2 [ m ( 2 ) + m ( 3 ) x j 1 ] m x j 1 + 3 x j 1 [ m + m ( 2 ) x j 1 ] m x j 1 m x j 2 + 3 m ( 2 ) x j 1 x j 2 m 2 x j 1 2 3 m 3 x j 1 3 m x j 2 = m x j 1 x j 2 ( 3 ( m 2 ) x j 1 ( x j 1 1 ) 1 )
E [ ( ξ j 1 E [ ξ j 1 ] ) 2 ( ξ j 2 E [ ξ j 2 ] ) 2 ] = E [ ξ j 1 2 ξ j 2 2 ] 2 E [ ξ j 1 2 ξ j 2 ] E [ ξ j 2 ] 2 E [ ξ j 1 ξ j 2 2 ] E [ ξ j 1 ] + E [ ξ j 1 2 ] ( E [ ξ j 2 ] ) 2 + E [ ξ j 2 2 ] ( E [ ξ j 1 ] ) 2 + 4 E [ ξ j 1 ξ j 2 ] E [ ξ j 1 ] E [ ξ j 2 ] 3 ( E [ ξ j 1 ] ) 2 ( E [ ξ j 2 ] ) 2 = x j 1 x j 2 [ m ( 2 ) + m ( 3 ) ( x j 1 + x j 2 ) + m ( 4 ) x j 1 x j 2 ] 2 x j 1 x j 2 [ m ( 2 ) + m ( 3 ) x j 1 ] m x j 2 2 x j 1 x j 2 [ m ( 2 ) + m ( 3 ) x j 2 ] m x j 1 + x j 1 [ m + m ( 2 ) x j 1 ] m 2 x j 2 2 + x j 2 [ m + m ( 2 ) x j 2 ] m 2 x j 1 2 + 4 m ( 2 ) x j 1 x j 2 m x j 1 m x j 2 3 m 2 x j 1 2 m 2 x j 2 2 = m ( m 2 ) x j 1 x j 2 ( 3 x j 1 x j 2 ( x j 1 + x j 2 ) + 1 ) + m x j 1 x j 2
E [ ( ξ j 1 E [ ξ j 1 ] ) 2 ( ξ j 2 E [ ξ j 2 ] ) ( ξ j 3 E [ ξ j 3 ] ) ] = E [ ξ j 1 2 ξ j 2 ξ j 3 ] E [ ξ j 1 2 ξ j 2 ] E [ ξ j 3 ] E [ ξ j 1 2 ξ j 3 ] E [ ξ j 2 ] 2 E [ ξ j 1 ξ j 2 ξ j 3 ] E [ ξ j 1 ] + E [ ξ j 1 2 ] E [ ξ j 2 ] E [ ξ j 3 ] + 2 E [ ξ j 1 ξ j 2 ] E [ ξ j 1 ] E [ ξ j 3 ] + 2 E [ ξ j 1 ξ j 3 ] E [ ξ j 1 ] E [ ξ j 2 ] + E [ ξ j 2 ξ j 3 ] ( E [ ξ j 1 ] ) 2 3 ( E [ ξ j 1 ] ) 2 E [ ξ j 2 ] E [ ξ j 3 ] = x j 1 x j 2 x j 3 [ m ( 3 ) + m ( 4 ) x j 1 ] x j 1 x j 2 [ m ( 2 ) + m ( 3 ) x j 1 ] m x j 3 x j 1 x j 3 [ m ( 2 ) + m ( 3 ) x j 1 ] m x j 2 2 m ( 3 ) x j 1 x j 2 x j 3 m x j 1 + x j 1 [ m + m ( 2 ) x j 1 ] m x j 2 m x j 3 + 2 m ( 2 ) x j 1 x j 2 m x j 1 m x j 3 + 2 m ( 2 ) x j 1 x j 3 m x j 1 m x j 2 + m ( 2 ) x j 2 x j 3 m 2 x j 1 2 3 m 2 x j 1 2 m x j 2 m x j 3 = m ( m 2 ) x j 1 x j 2 x j 3 ( 3 x j 1 1 )
E [ ( ξ j 1 E [ ξ j 1 ] ) ( ξ j 2 E [ ξ j 2 ] ) ( ξ j 3 E [ ξ j 3 ] ) ( ξ j 4 E [ ξ j 4 ] ) ] = E [ ξ j 1 ξ j 2 ξ j 3 ξ j 4 ] E [ ξ j 1 ξ j 2 ξ j 3 ] E [ ξ j 4 ] E [ ξ j 1 ξ j 2 ξ j 4 ] E [ ξ j 3 ] E [ ξ j 1 ξ j 3 ξ j 4 ] E [ ξ j 2 ] E [ ξ j 2 ξ j 3 ξ j 4 ] E [ ξ j 1 ] + E [ ξ j 1 ξ j 2 ] E [ ξ j 3 ] E [ ξ j 4 ] + E [ ξ j 1 ξ j 3 ] E [ ξ j 2 ] E [ ξ j 4 ] + E [ ξ j 1 ξ j 4 ] E [ ξ j 2 ] E [ ξ j 3 ] + E [ ξ j 2 ξ j 3 ] E [ ξ j 1 ] E [ ξ j 4 ] + E [ ξ j 2 ξ j 4 ] E [ ξ j 1 ] E [ ξ j 3 ] + E [ ξ j 3 ξ j 4 ] E [ ξ j 1 ] E [ ξ j 2 ] 3 E [ ξ j 1 ] E [ ξ j 2 ] E [ ξ j 3 ] E [ ξ j 4 ] = m ( 4 ) x j 1 x j 2 x j 3 x j 4 m ( 3 ) x j 1 x j 2 x j 3 m x j 4 m ( 3 ) x j 1 x j 2 x j 4 m x j 3 m ( 3 ) x j 1 x j 3 x j 4 m x j 2 m ( 3 ) x j 2 x j 3 x j 4 m x j 1 + m ( 2 ) x j 1 x j 2 m x j 3 m x j 4 + m ( 2 ) x j 1 x j 3 m x j 2 m x j 4 + m ( 2 ) x j 1 x j 4 m x j 2 m x j 3 + m ( 2 ) x j 2 x j 3 m x j 1 m x j 4 + m ( 2 ) x j 2 x j 4 m x j 1 m x j 3 + m ( 2 ) x j 3 x j 4 m x j 1 m x j 2 3 m 4 x j 1 x j 2 x j 3 x j 4 = 3 m ( m 2 ) x j 1 x j 2 x j 3 x j 4 .

6. Conclusions

In this short paper, we found general formulas for the central and non-central moments of the multinomial distribution, as well as explicit formulas for all the non-central moments up to order 8 and all the central moments up to order 4. Our work expands on the results in [2], where the central moments were calculated up to order 4. It also complements the general formula for the (joint) factorial moments from [1] and the explicit formulas for some of the lower-order (mixed) cumulants that were presented in [3].

Author Contributions

All contributions were made by the sole author of the article, F.O. The author has read and agreed to the published version of the manuscript.

Funding

This research was funded by a postdoctoral fellowship from the NSERC (PDF) and a postdoctoral fellowship supplement from the FRQNT (B3X).

Acknowledgments

We thank the three referees for their comments.

Conflicts of Interest

The author declares no conflict of interest.

Abbreviations

The following abbreviations are used in this manuscript:
i.i.d.independent and identically distributed

References

  1. Mosimann, J.E. On the compound multinomial distribution, the multivariate β-distribution, and correlations among proportions. Biometrika 1962, 49, 65–82. [Google Scholar]
  2. Newcomer, J.T. Estimation Procedures for Multinomial Models with Overdispersion. Ph.D. Thesis, University of Maryland, College Park, MD, USA, 2008. [Google Scholar]
  3. Wishart, J. Cumulants of multivariate multinomial distributions. Biometrika 1949, 36, 47–58. [Google Scholar] [CrossRef]
  4. Johnson, N.L.; Kotz, S.; Balakrishnan, N. Discrete Multivariate Distributions; Wiley Series in Probability and Statistics: Applied Probability and Statistics; John Wiley & Sons, Inc.: New York, NY, USA, 1997; p. xxii+299. [Google Scholar]
  5. Vitale, R.A. Bernstein polynomial approach to density function estimation. In Statistical Inference and Related Topics; Academic Press: New York, NY, USA, 1975; pp. 87–99. [Google Scholar]
  6. Gawronski, W.; Stadtmüller, U. Smoothing histograms by means of lattice and continuous distributions. Metrika 1981, 28, 155–164. [Google Scholar] [CrossRef]
  7. Stadtmüller, U. Asymptotic distributions of smoothed histograms. Metrika 1983, 30, 145–158. [Google Scholar] [CrossRef]
  8. Gawronski, W. Strong laws for density estimators of Bernstein type. Period. Math. Hungar. 1985, 16, 23–43. [Google Scholar] [CrossRef]
  9. Stadtmüller, U. Asymptotic properties of nonparametric curve estimates. Period. Math. Hungar. 1986, 17, 83–108. [Google Scholar] [CrossRef]
  10. Tenbusch, A. Nonparametric curve estimation with Bernstein estimates. Metrika 1997, 45, 1–30. [Google Scholar] [CrossRef]
  11. Petrone, S. Bayesian density estimation using Bernstein polynomials. Can. J. Stat. 1999, 27, 105–126. [Google Scholar] [CrossRef]
  12. Petrone, S. Random Bernstein polynomials. Scand. J. Stat. 1999, 26, 373–393. [Google Scholar] [CrossRef]
  13. Ghosal, S. Convergence rates for density estimation with Bernstein polynomials. Ann. Stat. 2001, 29, 1264–1280. [Google Scholar] [CrossRef]
  14. Petrone, S.; Wasserman, L. Consistency of Bernstein polynomial posteriors. J. Roy. Stat. Soc. Ser. B 2002, 64, 79–100. [Google Scholar] [CrossRef]
  15. Babu, G.J.; Canty, A.J.; Chaubey, Y.P. Application of Bernstein polynomials for smooth estimation of a distribution and density function. J. Stat. Plan. Inference 2002, 105, 377–392. [Google Scholar] [CrossRef]
  16. Kakizawa, Y. Bernstein polynomial probability density estimation. J. Nonparametr. Stat. 2004, 16, 709–729. [Google Scholar] [CrossRef]
  17. Bouezmarni, T.; Rolin, J.M. Bernstein estimator for unbounded density function. J. Nonparametr. Stat. 2007, 19, 145–161. [Google Scholar] [CrossRef]
  18. Bouezmarni, T.; Mesfioui, M.; Rolin, J.M. L1-rate of convergence of smoothed histogram. Stat. Probab. Lett. 2007, 77, 1497–1504. [Google Scholar] [CrossRef]
  19. Leblanc, A. Chung-Smirnov property for Bernstein estimators of distribution functions. J. Nonparametr. Stat. 2009, 21, 133–142. [Google Scholar] [CrossRef]
  20. Leblanc, A. A bias-reduced approach to density estimation using Bernstein polynomials. J. Nonparametr. Stat. 2010, 22, 459–475. [Google Scholar] [CrossRef]
  21. Curtis, S.M.; Ghosh, S.K. A variable selection approach to monotonic regression with Bernstein polynomials. J. Appl. Stat. 2011, 38, 961–976. [Google Scholar] [CrossRef]
  22. Leblanc, A. On estimating distribution functions using Bernstein polynomials. Ann. Inst. Stat. Math. 2012, 64, 919–943. [Google Scholar] [CrossRef]
  23. Leblanc, A. On the boundary properties of Bernstein polynomial estimators of density and distribution functions. J. Stat. Plan. Inference 2012, 142, 2762–2778. [Google Scholar] [CrossRef]
  24. Igarashi, G.; Kakizawa, Y. On improving convergence rate of Bernstein polynomial density estimator. J. Nonparametr. Stat. 2014, 26, 61–84. [Google Scholar] [CrossRef]
  25. Turnbull, B.C.; Ghosh, S.K. Unimodal density estimation using Bernstein polynomials. Comput. Stat. Data Anal. 2014, 72, 13–29. [Google Scholar] [CrossRef] [Green Version]
  26. Lu, L. On the uniform consistency of the Bernstein density estimator. Stat. Probab. Lett. 2015, 107, 52–61. [Google Scholar] [CrossRef]
  27. Guan, Z. Efficient and robust density estimation using Bernstein type polynomials. J. Nonparametr. Stat. 2016, 28, 250–271. [Google Scholar] [CrossRef] [Green Version]
  28. Guan, Z. Bernstein polynomial model for grouped continuous data. J. Nonparametr. Stat. 2017, 29, 831–848. [Google Scholar] [CrossRef] [Green Version]
  29. Belalia, M.; Bouezmarni, T.; Leblanc, A. Smooth conditional distribution estimators using Bernstein polynomials. Comput. Stat. Data Anal. 2017, 111, 166–182. [Google Scholar] [CrossRef]
  30. Belalia, M.; Bouezmarni, T.; Leblanc, A. Bernstein conditional density estimation with application to conditional distribution and regression functions. J. Korean Stat. Soc. 2019, 48, 356–383. [Google Scholar] [CrossRef]
  31. Tenbusch, A. Two-dimensional Bernstein polynomial density estimators. Metrika 1994, 41, 233–253. [Google Scholar] [CrossRef]
  32. Ouimet, F. Asymptotic properties of Bernstein estimators on the simplex. arXiv 2020, arXiv:2002.07758. [Google Scholar]
  33. Ouimet, F. Asymptotic properties of Bernstein estimators on the simplex. Part 2: The boundary case. arXiv 2020, arXiv:2006.11756. [Google Scholar]
  34. Ouimet, F. A precise local limit theorem for the multinomial distribution and some applications. arXiv 2020, arXiv:2001.08512. [Google Scholar]
  35. Arenbaev, N.K. Asymptotic behavior of the multinomial distribution. Teor. Veroyatnost. Primenen. 1976, 21, 826–831. [Google Scholar]
  36. Aitchison, J.; Lauder, I.J. Kernel Density Estimation for Compositional Data. J. R. Stat. Soc. Ser. (Appl. Stat.) 1985, 34, 129–137. [Google Scholar] [CrossRef]
  37. Brown, B.M.; Chen, S.X. Beta-Bernstein smoothing for regression curves with compact support. Scand. J. Stat. 1999, 26, 47–59. [Google Scholar] [CrossRef]
  38. Chen, S.X. Beta kernel estimators for density functions. Comput. Stat. Data Anal. 1999, 31, 131–145. [Google Scholar] [CrossRef]
  39. Chen, S.X. Beta kernel smoothers for regression curves. Stat. Sin. 2000, 10, 73–91. [Google Scholar]
  40. Bouezmarni, T.; Rolin, J.M. Consistency of the beta kernel density function estimator. Can. J. Stat. 2003, 31, 89–98. [Google Scholar] [CrossRef]
  41. Ouimet, F. Density estimation using Dirichlet kernels. arXiv 2020, arXiv:2002.06956. [Google Scholar]
  42. Graham, R.L.; Knuth, D.E.; Patashnik, O. Concrete Mathematics, 2nd ed.; Addison-Wesley Publishing Company: Reading, MA, USA, 1994; p. xiv+657. [Google Scholar]
  43. Newcomer, J.T.; Neerchal, N.K.; Morel, J.G. Computation of Higher Order Moments from Two Multinomial Overdispersion Likelihood Models; Department of Mathematics and Statistics, University of Maryland: Baltimore, MD, USA, 2008; pp. 1–11. Available online: http://www.math.umbc.edu/~kogan/technical_papers/2008/Newcomer_Nagaraj_Morel.pdf (accessed on 10 November 2020).
  44. Ouimet, F. Explicit formulas for the joint third and fourth central moments of the multinomial distribution. arXiv 2020, arXiv:2006.09059. [Google Scholar]
Publisher’s Note: MDPI stays neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Share and Cite

MDPI and ACS Style

Ouimet, F. General Formulas for the Central and Non-Central Moments of the Multinomial Distribution. Stats 2021, 4, 18-27. https://doi.org/10.3390/stats4010002

AMA Style

Ouimet F. General Formulas for the Central and Non-Central Moments of the Multinomial Distribution. Stats. 2021; 4(1):18-27. https://doi.org/10.3390/stats4010002

Chicago/Turabian Style

Ouimet, Frédéric. 2021. "General Formulas for the Central and Non-Central Moments of the Multinomial Distribution" Stats 4, no. 1: 18-27. https://doi.org/10.3390/stats4010002

APA Style

Ouimet, F. (2021). General Formulas for the Central and Non-Central Moments of the Multinomial Distribution. Stats, 4(1), 18-27. https://doi.org/10.3390/stats4010002

Article Metrics

Back to TopTop