Power method polynomial transformations are commonly used for simulating continuous nonnormal distributions with specified moments. However, conventional moment-based estimators can (a) be substantially biased, (b) have high variance, or (c) be influenced by outliers. In view of these concerns, a characterization of power method transformations by L-moments is introduced. Specifically, systems of equations are derived for determining coefficients for specified L-moment ratios, which are associated with standard normal and standard logistic-based polynomials of order five and three. Boundaries for L-moment ratios are also derived, and closed-formed formulae are provided for determining if a power method distribution has a valid probability density function. It is demonstrated that L-moment estimators are nearly unbiased and have relatively small variance in the context of the power method. Examples of fitting power method distributions to theoretical and empirical distributions based on the method of L-moments are also provided.
1. Introduction
The power method (see Fleishman [1], Headrick [2], and [3–7]) is a traditional procedure used for simulating continuous nonnormal distributions. Some applications of the power method have included such topics as ANOVA [8–10], asset pricing theories [11], business-cycle features [12], cluster analysis [13], item parameter estimation [14], item response theory [15, 16], factor analysis [17–19], price risk [20], structural equation models [21–24], and toxicology [25].
In the context of univariate data generation and for the purposes considered herein, the power method can be generally summarized by the polynomial transformation as in Headrick [2, equation (2.7)]
p(W)=∑i=1mciWi-1,
where W can be either a standard normal (Z) or a standard logistic (L) random variable with probability density function (pdf) and cumulative distribution function (cdf)
fW(w)={fW≔Z(w≔z)=ϕ(z)=(2π)-(1/2)exp{-z22},fW≔L(w≔l)=(π/3)exp{-(π/3)l}(1+exp{-(π/3)l})2,FW(w)={FW≔Z(w≔z)=Φ(z)=∫-∞zϕ(u)du,-∞<z<+∞,FW≔L(w≔l)=11+exp{-(π/3)l},-∞<l<+∞.
Setting m=4 (or m=6) in (1.1) gives the Fleishman [1] (or Headrick [3]) class of distributions associated with standard normal-based polynomials. The shape of p(W) in (1.1) is contingent on the coefficients ci, which are determined by moment-matching techniques. For example, see Headrick [2, equations (2.18)–(2.21); (2.22)–(2.25)] for determining the coefficients associated with standard normal orstandard logistic-based polynomials for m=4 in (1.1).
The pdf and cdf associated with p(W) in (1.1) are given in parametric form (ℝ2) as in Headrick [2, equations (2.12) and (2.13)]
fp(W)(p(w))=fp(W)(p(x,y))=fp(W)(p(w),fW(w)p′(w)),Fp(W)(p(w))=Fp(W)(p(x,y))=Fp(W)(p(w),FW(w)),
where the derivative p′(w)>0 in (1.3), that is, the polynomial p is a strictly increasing monotonic function and where fW(w) and FW(w) are the pdf and cdf associated with the random variable W in (1.2). To demonstrate, Figure 1 gives the graphs of a pdf and cdf associated with a standard normal-based power method distribution used in a Monte Carlo study by Berkovits et al. [8] (and, similarly, Enders [21] and Olsson et al. [24]).
Standard normal-based power method pdf and cdf. The coefficients used in (1.3) and (1.4) are c1=-0.252299, c2=0.418610, c3=0.252299, and c4=0.147593 and are associated with the parameters in Tables 1 and 2.
The graphs in Figure 1 and numerical solutions for the coefficients ci, which are based on the shape parameters in Table 1, were obtained using (1.3), (1.4), and the software package developed by Headrick et al. [26]. Note that the parameters in Table 1, γ3 (skew), γ4 (kurtosis), γ5, and γ6, are standardized cumulants and are scaled such that the normal distribution would have values of γ3=γ4=γ5=γ6=0.
Conventional moment-based parameters (γ) and their estimates (g) for the pdf in Figure 1. Each estimate (Est.), and bootstrap confidence interval (C.I.) was based on resampling 25,000 statistics. Each sample statistic was based on a sample size of n=200.
Parameter
Est. (95% C.I.)
Standard error
skew: γ3=3
g3: 2.51 (2.50, 2.52)
0.007522
kurtosis: γ4=21
g4: 13.4 (13.2, 13.5)
0.069400
γ5=182
g5: 69.8 (68.4, 71.3)
0.723200
γ6=2329
g6: 498 (485, 514)
7.289000
Conventional moment-based estimators, such as g3,…,g6 in Table 1, have unfavorable attributes to the extent that they can be substantially biased, have high variance, or can be influenced by outliers and thus may not be representative of the true population parameters (e.g., see [27, 28]). Some of these attributes are exemplified in Table 1 as the estimates of g3,…,g6 and their respective bootstrap confidence intervals attenuate below the population parameters of γ3,…,γ6 with increased bias and variance as the order of the estimate increases. That is, on average, the estimates of g3,…,g6 are only 83.67%, 63.81%, 38.35%, and 21.38% of their associated population parameters, respectively.
The estimates of g3,…,g6 in Table 1 were calculated based on samples of size n=200 and Fisher’s k-statistics (see, e.g., [29, pages 299-300]), that is, the formulae currently used by most commercial software packages such as SAS, SPSS (PASW), Minitab, and so forth, for computing values of skew and kurtosis. Thus, it should also be pointed out that these estimates (g3,…,g6) have another undesirable property of being algebraically bounded based on the sample size, that is, |g3|≤n1/2, g4≤n, |g5|≤n3/2, g6≤n2. As such, if a researcher was using a value of kurtosis in a Monte Carlo study, such as γ4=21 in Table 1, and drawing samples of size n=15, then the largest sample estimate of kurtosis possible is g4=15 or 71.43% of the parameter.
The method of L-moments is an attractive alternative to conventional method of moments in terms of describing theoretical or empirical probability distributions, estimating parameters, and hypothesis testing (see [27, 28, 30]). More specifically, the first four L-moments are analogous to conventional moments as they describe the location, scale, skew, and kurtosis of a data set. However, L-moments have demonstrated to be superior to conventional moments to the extent that they (a) exist whenever the mean of the distribution exists, (b) only require that a distribution have finite variance for their standard errors to be finite, (c) are nearly unbiased for all sample sizes and distributions, (d) do not suffer from the deleterious effects of sampling variability, (e) are more robust in the presence of outliers, and (f) are not algebraically bounded based on sample size (see [27, 28, 30–33]). Further, it has been demonstrated that there are conditions where the method of L-moments can also yield more accurate and efficient parameter estimates than the method of maximum likelihood when sample sizes are small to moderate (see [27, 34–37]). Other advances have also been made. For example, Elamir and Seheult (see [38, 39]) introduced trimmed L-moments and derived expressions for the exact variances and covariances of sample L-moments. Further, Necir and Meraghni [40] demonstrated that L-moments and trimmed L-moments are useful for estimating L-functionals in the context of heavy-tailed distributions.
Estimates of L-moments and L-moment ratios are based on linear combinations of order statistics unlike conventional moments that are based on raising the data to successive powers which in turn gives greater weight to data points located farther away from the mean and thus may result with estimates (gj) having substantial bias and (or) variance. For example, the L-moment ratio estimates (tj) in Table 2 are relatively much closer to their respective population parameters (τj) with smaller variance than their corresponding conventional moment-based analogs (gj) in Table 1. Specifically, the ratios (tj) are, on average, 99.03%, 99.55%, 98.35%, and 99.29% of their respective parameters.
L-moment ratios (τ) and their estimates (t) for the pdf in Figure 1. Each estimate and bootstrap confidence interval (C.I.) was based on resampling 25,000 statistics. Each sample statistic was based on a sample size of n=200.
Parameter
Est. (95% C.I.)
Standard error
L-skew: τ3=.313041
t3: 0.310 (.309,.311)
0.000407
L-kurtosis: τ4=.333498
t4: 0.332 (.332,.333)
0.000268
τ5=.0986303
t5: 0.097 (.097,.098)
0.000320
τ6=.157110
t6: 0.156 (.156,.157)
0.000252
In view of the above, the present aim is to obviate the problems associated with conventional moments in the context of power method transformations of the form in (1.1) by characterizing these transformations through L-moments. The focus is on standard normal and standard logistic-based polynomials. In Section 2, the essential requisite information associated with L-moments for theoretical and empirical distributions is provided as well as the derivations for the systems of equations for computing polynomial coefficients. Closed-formed formulae are provided for determining if any particular polynomial has a valid pdf. Further, the boundary regions for valid pdfs are derived and graphed for polynomials of order three and for the symmetric case associated with polynomials of order five. In Section 3, conventional moments and L-moments are compared in terms of estimation and distribution fitting to demonstrate the superior characteristics that L-moments have in the context of the power method. In Section 4, third- and fifth-order standard normal and logistic-based polynomials are compared in terms of their upper and lower L-moment ratio boundary points.
2. Methodology2.1. Preliminaries
L-moments are defined as linear combinations of probability weighted moments (PWMs). For a continuous theoretical distribution with a cdf denoted as F(x), the PWMs can be generally defined as in Hosking [27]
βr=∫x{F(x)}rdF(x),
where r=0,1,2,…. The L-moments can be determined by summing the PWMs as
λr+1=∑i=0rp(r,i)βi,
where
p(r,i)=(-1)r-i(r+i)!(i!)2(r-i)!
are coefficients from shifted orthogonal Legendre polynomials. Specifically, the first six L-moments based on (2.1)–(2.3) are expressed as
λ1=β0,λ2=2β1-β0,λ3=6β2-6β1+β0,λ4=20β3-30β2+12β1-β0,λ5=70β4-140β3+90β2-20β1+β0,λ6=252β5-630β4+560β3-210β2+30β1-β0.
Analogous to conventional moment theory, the values of λ1 and λ2 are parameters associated with the location and scale of the distribution. More specifically, the L-mean (λ1) is the usual arithmetic mean, and L-scale (λ2≥0) is one-half of Gini’s coefficient of mean difference (see, e.g., [29, pages 47-48]). Higher-order L-moments are transformed to dimensionless indices referred to as L-moment ratios defined as τr+1=λr+1/λ2 for r≥2. In general, L-moment ratios are bounded such that |τr+1|<1 as is the index of L-skew (τ3) where a symmetric distribution implies that τ2r-1=0. Smaller boundaries can be found for specific cases. For example, in the context of continuous distributions, L-kurtosis (τ4) and τ6 have boundaries of (see Jones [41])
5τ32-14<τ4<1,125(42τ42-14τ4-3)<τ6<1,
which indicate that τ4 and τ6 have lower bounds of -1/4 and -1/6, respectively. An example of a set of computed L-moments (λ1,λ2) and L-moment ratios (τ3,…,τ6) based on (2.1) through (2.4) is provided in the first column of Table 3 for a Beta distribution.
L-moment parameters (λ,τ) and percentiles for the Beta (α=5,β=4) distribution and power method approximations in Figure 3.
Percentiles
Parameters
%
Beta (Exact)
Third order
Fifth order
λ1=0.555556
1
0.1982
0.1679
0.1983
λ2=0.0895836
5
0.2892
0.2874
0.2892
τ3=-.0251716
25
0.4445
0.4507
0.4445
τ4=.0921653
75
0.6709
0.6648
0.6709
τ5=-.0066496
95
0.8071
0.8097
0.8071
τ6=.0284846
99
0.8791
0.9072
0.8789
Empirical L-moments for a set of data of size n are linear combinations of the sample order statistics X1:n≤X2:n≤⋯≤Xn:n. The unbiased estimates of the PWMs are
br=1n∑i=r+1n(i-1)(i-2)⋯(i-r)(n-1)(n-2)⋯(n-r)Xi:n
for r=0,1,2,… and where b0 is the sample mean. The sample L-moments are obtained by substituting br in place of βr in (2.4). The notations used for sample L-moments and L-moment ratios are tr+1=ℓr+1/ℓ2 for r≥2.
2.2. L-Moments for Standard Normal Polynomial Transformations
Using (1.1)-(1.2) and (2.1), the PWMs for power method polynomials based on the standard normal distribution are
βr=∫-∞∞∑i=1mcizi-1{Φ(z)}rϕ(z)dz.
Integrating (2.7) for βr=0,…,5 and subsequently substituting these PWMs into (2.4) (and after several tedious manipulations) and simplifying yields the following system of equations for fifth-order polynomials (m=6):
λ1=c1+c3+3c5,λ2=Δ2=(4c2+10c4+43c6)/(2π)2,τ3=6c3+26c53Δπ,τ4=202(δ1c2+δ2c4+δ3c6)Δπ3/2-32,τ5=845(δ4c3+δ5c5)-5π3(3c3+13c5)3Δπ2,τ6=15Δπ5/2-560π2(δ1c2+δ2c4+δ3c6)+13443(δ6c2+δ7c4+δ8c6)8Δπ5/2,
where Δ=(4c2+10c4+43c6)/(2π) is the coefficient of mean difference. The analytical expressions for the constants δi=1,…,8 in (2.11)–(2.13) are given in Appendix A.
The derivations above yield a system of six equations (2.8)–(2.13) expressed in terms of six real variables c1,…,c6. The first two equations of this system can be standardized by setting (2.8) to zero and (2.9) to 1/π. The next four equations, (2.10)–(2.13), are set to the desired values of τ3,…,τ6. Simultaneously solving this system of equations yields the solutions of c1,…,c6. The coefficients are then substituted into (1.1) to generate p(W) which has zero mean (λ1=0), one-half of the coefficient of mean difference for the unit-normal distribution (λ2=1/π ), and the specified values of τ3,…,τ6. If the negatives of τ3 and τ5 are desired, then inspection of (2.8)–(2.13) indicates that all that is needed are simultaneous sign reversals between c1,c3, and c5. These sign reversals will have no effect on τ2,τ4, or τ6. It is worthy to point out that it is not necessary to numerically solve the system of equations in (2.8)–(2.13) as the coefficients c1,…,c6 have unique solutions which can be determined by evaluating equations (A.9)–(A.14) in Appendix A. See Figures 2(a) and 2(b) for examples of standard normal-based fifth-order power method pdfs.
Standard normal and standard logistic-based power method pdfs.
Normal-based pdf
Normal-based pdf
Logistic-based pdf
Setting c5=c6=0 in (2.8)–(2.11) and simplifying yields the system of equations for the smaller class of distributions associated with third-order polynomials (m=4) in (1.1) as
λ1=c1+c3,λ2=Δ2=(4c2+10c4)/(2π)2,τ3=23c3Δπ,τ4=202(δ1c2+δ2c4)Δπ3/2-32,
where the solutions for the coefficients in (2.14)–(2.17) are
c1=-c3=-τ3π3,c2=-16δ2+2(3+2τ4)π8(5δ1-2δ2),c4=40δ1-2(3+2τ4)π20(5δ1-2δ2).
Examples of a fifth-order (c1,…,c6) and a third-order (c1,…,c4) power method distribution are given in Figure 3, both of which provide an approximation of the Beta (α=5,β=4) distribution. The coefficients associated with the two polynomial approximations in Figure 3 were used in (1.4) to compute the percentiles in Table 3 for the power method pdfs. Inspection of the graphs in Figure 3 and the percentiles in Table 3 indicate that the fifth-order polynomial provides a much more accurate approximation of the Beta distribution than the third-order polynomial. The reason for this is because the fifth-order system could produce a valid pdf that is based on an exact match with the Beta distribution’s L-moment ratios τ3,…,τ6, whereas the third-order system was unable to produce a valid pdf that has an exact match with τ4,τ5, and τ6. Thus, it is important to consider the boundary conditions for valid power method pdfs.
Third- and fifth-order standard normal-based polynomial approximations (dashed lines) of a Beta (α=5, β=4) distribution. The L-moment ratios for this Beta distribution are given in Table 3. The fifth-order polynomial coefficients are c1=.02697, c2=1.071, c3=-.0297, c4=-.0312, c5=.0009, and c6=.00065. The third-order polynomial coefficients are c1=.02576, c2=0.9993, c3=-.02576, and c4=.00027.
2.3. L-Moment Boundaries for Standard Normal Polynomial Transformations
The restriction that p′(w)>0 in (1.3) implies that a set of solved coefficients may not necessarily produce a valid pdf. To determine if a third-order polynomial produces a valid pdf, we first set the quadratic equation p′(w)=0 and subsequently solve for w as
w=-c3±(c32-3c2c4)1/23c4.
A set of solved coefficients will produce a valid pdf if the discriminant c32-3c2c4 in (2.19) is negative. That is, the complex solutions for w must have nonzero imaginary parts. As such, setting c32=3c2c4 will yield the point where the discriminant vanishes and thus real-valued solutions exist to p′(w)=0.
Standardizing (2.15), by setting the coefficient of mean difference to 1/π, and solving for c4 give
c4=25-2c25.
Substituting the right-hand side of (2.20) into (2.16) and (2.17) and setting c3=(3c2c4)1/2, because we only need to consider positive values of L-skew, gives
τ3=325πc2(1-c2),τ4=22(5δ1c2+2δ2-2δ2c2)π-32.
Inspection of (2.21) indicates that for real values of τ3 to exist then we must have c2∈[0,1] and thus from (2.20) c4∈[0,0.4]. Using (2.21) and (2.22), the graph of the region for valid third-order power method pdfs is given in Figure 4 along with the minimum and maximum values of τ3 and τ4. In summary, a valid standardized nonnormal third-order pdf will have the properties of (a) 0<c2<1, (b) 0<c4<2/5, and (c) c32-3c2c4<0.
Boundary region and values of L-skew |τ3| and L-kurtosis τ4 for third-order standard normal power method pdfs. For asymmetric distributions, valid pdfs exist in the region inside the graphed boundary. For symmetric distributions, valid pdfs exist for τ4 between a4<τ4<b4. The lower boundary point (a3,4) denotes the normal distribution. Note that a3 and a4 are lower limits and b3 and b4 are upper limits.
In terms of fifth-order polynomials, the formulae that solve the (quartic) equation p′(w)=0 must also have nonzero imaginary parts to ensure a valid pdf. Specifically, the closed-formed expressions for evaluating pdfs associated with symmetric distributions are
w=±[±(9c42-20c2c6)1/2-3c410c6]1/2,w=∓[±(9c42-20c2c6)1/2-3c410c6]1/2,
and for asymmetric distributions the expressions are
w=±12A1/2±12(B±C)1/2-c55c6,w=±12A1/2∓12(B±C)1/2-c55c6.
The expressions for A, B, and C in (2.24) are given in Appendix B.
The boundary for the larger class of symmetric fifth-order pdfs (i.e., τ3=τ5=0;c1=c3=c5=0) can be viewed by making use of (2.23). Specifically, the graph of the region for valid symmetric pdfs is given in Figure 5 along with the minimum and maximum values of τ4 and τ6. The elliptical graph in Figure 5 consists of four separate segments where τ4 in (2.11) and τ6 in (2.13) are expressed solely as functions of c2. See Appendix C for further details.
Boundary region and values of L-kurtosis τ4 and τ6 for symmetric fifth-order standard normal-based power method pdfs. Valid pdfs exist inside the elliptical graph. The point at the origin represents the regular uniform distribution (τ4=τ6=0). Note that a4 and a6 are lower limits and b4 and b6 are upper limits.
2.4. L-Moments and Boundaries for Standard Logistic Polynomial Transformations
The method for deriving the system of equations for power method transformations based on the standard logistic distribution is similar to the derivation of the system associated with (2.8)–(2.13). However, the derivation is more straight forward because the cdf for the logistic distribution is available in closed form. As such, the system of equations for fifth-order polynomials in (1.1) is
λ1=c1+c3+21c55,λ2=Δ2=23(c2+3c4+21c6)/π2,τ3=12c3+72c5Δπ2,τ4=23(90c4+900c6+π2(c2+3c4+21c6))6Δπ3,τ5=5π2c3+252c5+30π2c5Δπ4,τ6=2Δπ5+23(315π2c4+11340c6+3150π2c6)30Δπ5,
where Δ=23(c2+3c4+21c6)/π is the coefficient of mean difference and for standardized distributions the solutions for the coefficients c1,…,c6 are given in Appendix D. The system of equations for the smaller family of distributions based on third-order polynomials can be obtained by setting c5=c6=0 in (2.25)–(2.28), and for standardized distributions, the solutions for the coefficients are
c1=-c3=-τ3π23,c2=30+π2(1-6τ4)30,c4=π2(6τ4-1)90.
The derivation of the L-skew (τ3) and L-kurtosis (τ4) boundary for third-order polynomials is also analogous to the steps taken in Section 2.3. As such, the graph of the boundary for valid third-order power method pdfs is given in Figure 6 along with the minimum and maximum values of τ3 and τ4. Similarly, a valid standardized nonlogistic third-order pdf will have the properties of (a) 0<c2<1, (b) 0<c4<1/3, and (c) c32-3c2c4<0. In terms of fifth-order polynomials, the boundary region for symmetric pdfs and minimum and maximum values of τ4 and τ6 are given in Figure 7. The elliptical graph in Figure 7 also consists of four segments. See Appendix E for further details. Note that the expressions in (2.23) and (2.24) are also used to determine whether fifth-order pdfs are valid or not. An example of a logistic-based power method pdf is given in Figure 2(c).
Boundary region and values of L-skew |τ3| and L-kurtosis τ4 for third-order standard logistic power method pdfs. For asymmetric distributions, valid pdfs exist in the region inside the graphed boundary. For symmetric distributions, valid pdfs exist for τ4 between a4<τ4<b4. The lower boundary point (a3,4) represents the standard logistic distribution. Note that a3 and a4 are lower limits and b3 and b4 are upper limits.
Boundary region and values of L-kurtosis τ4 and τ6 for symmetric fifth-order standard logistic-based power method pdfs. Valid pdfs exist inside the elliptical graph. Note that a4 and a6 are lower limits and b4 and b6 are upper limits.
3. A Comparison of Conventional Moments and L-Moments3.1. Estimation
One of the advantages that sample L-moment ratios (t3,…,6) have overconventional moment-based estimates, such as skew (g3) and kurtosis (g4), is that t3,…,6 are less biased (e.g., [28]). This advantage can be demonstrated in the context of the power method by considering the simulation results associated with the indices for the percentage of relative bias (RBias%) and standard error (SE) reported in Table 4.
Overall average estimates (Est.), bootstrap confidence intervals (C.I.), standard errors (SE), and percentage of relative bias (RBias%) for the conventional moment-based statistics (g) and L-moment ratios (t).
Parameters (χdf=82) and coefficients
γ1=0
c1=-0.163968
λ1=0
c1=-0.169160
γ2=1
c2=0.950794
λ2=1/π
c2=0.980897
γ3=1
c3=0.165391
τ3=0.164666
c3=0.170627
γ4=1.5
c4=0.007345
τ4=0.131237
c4=0.007577
γ5=3
c5=-0.000474
τ5=0.051194
c5=-0.000489
γ6=7.5
c6=0.000014
τ6=0.048334
c6=0.000015
Est. (95% C.I.)
SE
RBias%
Est. (95% C.I.)
SE
RBias%
n=50
g3: .8860 (.8808,.8917)
.0028
−11.40
t3: .1626 (.1618,.1634)
.0004
−1.25
g4: 1.021 (.9975,1.043)
.0116
−31.93
t4: .1314 (.1308,.1321)
.0003
0.12
g5: 1.302 (1.218,1.389)
.0439
−56.60
t5: .0510 (.0505,.0516)
.0003
−0.38
g6: 1.623 (1.313,1.973)
.1673
−78.36
t6: .0484 (.0480,.0489)
.0002
0.14
n=100
g3: .9395 (.9350,.9436)
.0022
−6.05
t3: .1635 (.1629,.1640)
.0003
−0.71
g4: 1.225 (1.204,1.245)
.0104
−18.33
t4: .1314 (.1309,.1318)
.0002
0.09
g5: 1.930 (1.847,2.032)
.0466
−35.67
t5: .0511 (.0507,.0514)
.0002
−0.18
g6: 3.385 (3.004,3.862)
.2160
−54.87
t6: .0483 (.0479,.0486)
.0002
−0.07
n=1000
g3: .9941 (.9926,.9957)
.0008
−0.59
t3: .1645 (.1643,.1647)
.0001
−0.10
g4: 1.471 (1.462,1.480)
.0046
−1.93
t4: .1313 (.1311,.1314)
.0001
0.05
g5: 2.869 (2.822,2.921)
.0252
−4.37
t5: .0512 (.0510,.0512)
.0001
0.01
g6: 6.889 (6.586,7.186)
.1544
−8.15
t6: .0483 (.0482,.0484)
.0001
−0.01
Specifically, a Fortran algorithm was coded to generate twenty-five thousand independent samples of size n=50,100,1000, and the estimates of g3,…,6 and t3,…,6 were computed for each of the (3 × 25000) samples based on the parameters and coefficients listed in Table 4. The estimates of g3,…,6 were computed based on Fisher’s k-statistics, and the estimates of t3,…,6 were based on (2.6). Bootstrapped average estimates, confidence intervals, and SEs were obtained for g3,…,6 and t3,…,6 using twenty-five thousand resamples via the commercial software package Spotfire S+ [42]. The percentage of relative bias for each estimate was computed as RBias%,(gj)=100×(gj-γj)/γj and RBias%,(tj)=100×(tj-τj)/τj.
The results in Table 4 demonstrate the substantial advantage that L-moment ratios have overconventional estimates in terms of both bias and error. These advantages are most pronounced in the context of smaller sample sizes and higher-order moments. For example, given a sample size of n=50, the conventional estimates of g3 (g5) and g4 (g6) generated in the simulation were, on average, 11.40% (56.50%) and 31.93% (78.36%) less than their respective parameters. On the other hand, the amounts of relative bias associated with the L-moment ratios were essentially negligible. Further, the SEs associated with t3,…,6 are relatively much smaller than the corresponding SEs for g3,…,6.
3.2. Distribution Fitting
Presented in Figure 8 are conventional moment and L-moment-based power method pdfs superimposed on a histogram of data from the Project Match Research Group studies (see [43, 44]). The data are associated with participants who were assigned to a Twelve-Step Facilitation (TSF) treatment condition for alcoholism. Specifically, these data are the total number of drinks per ninety-day period which were determined by each participant’s reports of daily standard ethanol drinks (one drink was equal to one-half of an ounce). Drinking was assessed approximately every three months, beginning at pretreatment (baseline), immediately following treatment, and at three-month intervals over a twelve-month posttreatment follow-up period (i.e., six, nine, twelve, and fifteen months posttreatment).
Histograms and fifth-order power method approximations for the TSF group. aThe value of g6 had to be increased to −0.28898 to ensure a valid pdf.
The sample estimates (g3,…6; t3,…6) associated with Figure 8 were based on a sample size of n=311 participants. The estimates were also used to solve for the two sets of coefficients, which produced the power method pdfs based on (1.3). Note that the two polynomials were linearly transformed using the location and scale estimates (m,s;ℓ1,ℓ2) from the data. Visual inspection of the approximations in Figure 8 and the goodness of fit statistics given in Table 5 indicate that the L-moment pdf provides a more accurate fit to the actual data. The reason for this is partially attributed to the fact that the conventional moment-based power method pdf does not have an exact match with g6, whereas the L-moment pdf is based on an exact match with all of the sample estimates. Note also that the asymptotic P values in Table 5 are based on a chi-square distribution with degrees of freedom: df = 13(classes) − 6(estimates) − 1(sample size) = 6.
Chi-square goodness of fit statistics for the conventional (C) moment and L-moment approximations for the TSF group data (n=311) in Figure 8.
%
Expected
Obs (C)
Obs (L)
Total number of drinks (C)
Total number of drinks (L)
10
31.10
38
32
<207.51
<176.05
20
31.10
26
32
207.51–286.14
176.05–287.51
30
31.10
22
25
286.14–380.05
287.51–397.76
40
31.10
31
36
380.05–490.40
397.76–514.61
50
31.10
37
33
490.40–620.50
514.61–644.07
60
31.10
29
25
620.50–777.13
644.07–794.99
70
31.10
33
35
777.13–973.41
794.99–977.85
75
15.55
17
15
973.41–1094.27
977.85–1090.33
80
15.55
18
16
1094.27–1238.41
1090.33–1224.72
85
15.55
15
16
1238.41–1418.18
1224.72–1393.72
90
15.55
16
15
1418.18–1660.33
1393.72–1625.71
95
15.55
15
16
1660.33–2045.66
1625.71–2010.65
100
15.55
14
15
2045.66 or more
2010.65 or more
C: χ2=7.135
L: χ2=3.920
P=.309
P=.688
4. Discussion and Conclusion
This paper introduced L-moments in the context of standard normal and standard logistic-based power method polynomials. The boundaries were analytically derived for polynomials of order three and for symmetric pdfs associated with polynomials of order five. The lower boundary point associated with the standard normal third-order system is the standard normal distribution (i.e., τ4≈0.1226 and τ6≈0.0437). As such, it is worthy to point out that the larger family of pdfs for the standard normal fifth-order system includes the L-moment ratios associated with the regular uniform or Beta (α=1,β=1) distribution (i.e., τ4=0 and τ6=0). Further, the upper boundary of L-kurtosis for the third-order system of τ4≈.5728 extends up to a boundary of τ4≈.8233 for the fifth-order system which is a substantial increase. That is, in terms of conventional moments, this is equivalent to increasing kurtosis from γ4≈43 to γ4≈833.
In terms of the third-order logistic-based family of pdfs, it is noted that maximum L-skew is τ3≈.5513 whereas the maximum value of L-skew for the standard normal-based system is τ3≈.5352. Again, this is actually a considerable difference—as these values correspond to values of conventional skew of γ3≈8.4 and γ3≈4, respectively. Similarly, the maximum values of L-kurtosis for the standard logistic-based and normal-based systems are τ4≈.6733 and τ4≈.5728 and are equivalent to values of conventional kurtosis of γ4≈472 and γ4≈43, respectively.
Finally, we note that Mathematica ([45], Version 7.0) software for computing polynomial coefficients, cumulative probabilities, and fitting (graphing) power method pdfs to data are available on request.
AppendicesA. Analytical Expressions and Polynomial Coefficients (Normal)
The expressions for δi=1,…,8 in (2.11)–(2.13) and the coefficients ci=1,…,6 in (2.8)–(2.13) are
δ1=3tan-1[2]2-3π42=0.360451474758732546420622…,δ2=15tan-1[2]22-15π82+14=1.151128686896831366051556…,δ3=129tan-1[2]42-129π162+7748=5.479020020323041540688357…,δ4=15tan-1[53]-π154=0.489313210040355820119964…,δ5=-15tan-1[53]-853tan-1[15]+19π453+14=2.370357243508208553853177…,δ6=5πtan-1[2]43+583sin-1[13]2-5πsin-1[1/3]83-5π2163+532ε1=0.261016893769534900409394…,δ7=25πtan-1[2]83+2543sin-1[13]2-25πsin-1[1/3]43-25π2643+5π86-546+2534ε2=0.960807335426376575971986…,δ8=215πtan-1[2]163+21583sin-1[13]2-215πsin-1[1/3]83-215π21283+385π966-385486+21538ε3=4.908966006122127514487203…,c2=403202(δ3δ7-δ2δ8)-504(43δ7-10δ8)(3+2τ4)π+56(43δ2-10δ3)π2(27+28τ4+8τ6)100802(43δ2δ6-10δ3δ6-43δ1δ7+4δ3δ7+10δ1δ8-4δ2δ8),c3=-140τ3δ53π+13π3/25(2τ5+5τ3)140(13δ4-3δ5),c4=-403202(δ3δ6-δ1δ8)+504(43δ6-4δ8)(3+2τ4)π+56(-43δ1+4δ3)π2(27+28τ4+8τ6)100802(43δ2δ6-10δ3δ6-43δ1δ7+4δ3δ7+10δ1δ8-4δ2δ8),c5=28τ3δ43π-(3π3/2(2τ5+5τ3))/5364δ4-84δ5,c6=20160(δ2δ6-δ1δ7)-2522(5δ6-2δ7)(3+2τ4)π+53(5δ1-2δ2)π2(27+28τ4+8τ6)5040(43δ2δ6-10δ3δ6-43δ1δ7+4δ3δ7+10δ1δ8-4δ2δ8),c1=-c3-3c5.
Equations (A.6)–(A.8) were derived by adapting the techniques used by Renner [46] and where ε1, ε2, and ε3 are based on the standard recurrence relation for ∫secpudu. The coefficients in (A.9)–(A.14) are for standardized distributions (i.e., λ1=0,λ2=1/π).
B. Expressions for (2.24)
The expressions for A, B, and C in (2.24) are
A=4c5225c62+(E1/2+D)1/315c621/3+21/3F5c6(E1/2+D)1/3-2c45c6,B=8c5225c62-(E1/2+D)1/315c621/3-21/3F5c6(E1/2+D)1/3-4c45c6,C=A-(1/2)(48c4c5100c62-16c320c6-64c53500c63),D=54c43-216c3c4c5+432c2c52+540c32c6-1080c2c4c6,E=D2-4(9c42-24c3c5+60c2c6)3,F=3c42-8c3c5+20c2c6.
C. Symmetric Distributions Boundary (Normal)
The elliptical graph in Figure 5 is based on (2.11) and (2.13) and setting c4=(4-4c2-43c6)/10 from (2.9) and then setting c6 as follows:
Segment 1:0≤c2≤1,c6=-4(137c2-387)16641+405387c2-262c2216641,Segment 2:1≤c2≤387262,c6=-4(137c2-387)16641+405387c2-262c2216641,Segment 3:0≤c2≤1,c6=-4(137c2-387)16641-405387c2-262c2216641,Segment 4:1≤c2≤387262,c6=-4(137c2-387)16641-405387c2-262c2216641.
D. Polynomial Coefficients (Logistic)
The coefficients for the standard logistic-based system in (2.25)–(2.30) are
c2=840+28π2(1-6τ4)+π4(1-14τ4+20τ6)840,c3=42τ3π+π3(5τ3-12τ5)843,c4=π2(42(6τ4-1)+5π2(14τ4-1-20τ6))3780,c5=π3(12τ5-5τ3)5043,c6=π4(1-14τ4+20τ6)7560,c1=-c3-21c55.
E. Symmetric Distributions Boundary (Logistic)
The elliptical graph in Figure 7 is based on (2.28) and (2.30) and setting c4=(1-c2-21c6)/3 from (2.26) and then setting c6 as follows:
Segment 1:0≤c2≤1,c6=21-11c2441+2521c2-16c22441,Segment 2:1≤c2≤2116,c6=21-11c2441+2521c2-16c22441,Segment 3:0≤c2≤1,c6=21-11c2441-2521c2-16c22441,Segment 4:1≤c2≤2116,c6=21-11c2441-2521c2-16c22441.
FleishmanA. I.A method for simulating non-normal distributions19784345215322-s2.0-3425028390210.1007/BF02293811HeadrickT. C.2010Boca Raton, Fla, USAChapman & Hall/CRCHeadrickT. C.Fast fifth-order polynomial transforms for generating univariate and multivariate nonnormal distributions2002404685711193300510.1016/S0167-9473(02)00072-5ZBL1103.65302HeadrickT. C.SawilowskyS. S.Simulating correlated multivariate nonnormal distributions: extending the Fleishman power method1999641253510.1007/BF022943171685337HøylandK.KautM.WallaceS. W.A heuristic for moment-matching scenario generation2003242-3169185196915110.1023/A:1021853807313TadikamallaP. R.On simulating non-normal distributions19804522732792-s2.0-000136796610.1007/BF02294081ValeC. D.MaurelliV. A.Simulating multivariate nonnormal distributions19834834654712-s2.0-084227214410.1007/BF02293687BerkovitsI.HancockG. R.NevittJ.Bootstrap resampling approaches for repeated measure designs: relative robustness to sphericity and normality violations20006068778922-s2.0-0034339642KeselmanH. J.WilcoxR. R.AlginaJ.OthmanA. R.FradetteK.A comparative study of robust tests for spread: asymmetric trimming strategies2008612235253264903510.1348/000711008X299742LixL. M.FouladiR. T.Robust step-down tests for multivariate independent group designs20076022452652-s2.0-3654899973810.1348/000711006X117853Affleck-GravesJ.MacDonaldB.Nonnormalities and tests of asset pricing theories1989444889908HessG. D.IwataS.Measuring and comparing business-cycle features19971544324442-s2.0-0001660088SteinleyD.HensonR.OCLUS: an analytic method for generating clusters with known overlap2005222221250223117310.1007/s00357-005-0015-6KirisciL.HsuT.-c.YuL.Robustness of item parameter estimation programs to assumptions of unidimensionality and normality2001252146162182452810.1177/01466210122031975HarwellM.StoneC. A.HsuT. C.KirisciL.Monte Carlo studies in item response theory19962021011252-s2.0-0030494344StoneC. A.Empirical power and type I error rates for an IRT fit statistic that considers the precision of ability estimates200363456658310.1177/00131644022510342011165BensonJ.FleishmanJ. A.The robustness of maximum likelihood and distribution-free estimators to non-normality in confirmatory factor analysis19942821171362-s2.0-2134448005910.1007/BF01102757FloraD. B.CurranP. J.An empirical evaluation of alternative methods of estimation for confirmatory factor analysis with ordinal data2004944664912-s2.0-1084424549910.1037/1082-989X.9.4.466NasserF.WisenbakerJ.A Monte Carlo study investigating the impact of item parceling on measures of fit in confirmatory factor analysis200363572975710.1177/00131644032582282005483MahulO.Hedging price risk in the presence of crop yield and revenue insurance20033022172392-s2.0-0344585364EndersC. K.The impact of nonnormality on full information maximum-likelihood estimation for structural equation models with missing data2001633523702-s2.0-0035756118HauK.-T.MarshH. W.The use of item parcels in structural equation modelling: non-normal data and small sample sizes2004572327351219612510.1111/j.2044-8317.2004.tb00142.xHensonJ. M.ReiseS. P.KimK. H.Detecting mixtures from structural model differences using latent variable mixture modeling: a comparison of relative model fit statistics20071422022262368424OlssonU. H.FossT.TroyeS. V.Does the ADF fit function decrease when the kurtosis increases?2003562289303210177110.1348/000711003770480057HothornL.LehmacherW.A simple testing procedure “control versus k treatments” for one-sided ordered alternatives, with application in toxicology2007332179189HeadrickT. C.ShengY.HodisF. A.Numerical computing and graphics for the power method transformation using mathematica20071931172-s2.0-34247520620HoskingJ. R. M.L-moments: analysis and estimation of distributions using linear combinations of order statistics19905211051241049304ZBL0703.62018HoskingJ. R. M.WallisJ. R.1997Cambridge, UKCambridge University PressKendallM.StuartA.19774thNew York, NY, USAMacmillanHoskingJ. R. M.Moments or L-moments? An example comparing two measures of distributional shape1992463186189HoskingJ. R. M.Some theory and practical uses of trimmed L-moments2007137930243039232380910.1016/j.jspi.2006.12.002RoystonP.Which measures of skewness and kurtosis are best?19921133333432-s2.0-0026553826VogelR. M.FennesseyN. M.L moment diagrams should replace product moment diagrams1993296174517522-s2.0-002779619410.1029/93WR00341DelicadoP.GoriaM. N.A small sample comparison of maximum likelihood, moments and L-moments methods for the asymmetric exponential power distribution200852316611673242276210.1016/j.csda.2007.05.021HoskingJ. R. M.WallisJ. R.WoodE. F.Estimation of the generalized extreme-value distribution by the method of probability-weighted moments1985273251261797563HoskingJ. R. M.WallisJ. R.Parameter and quantile estimation for the generalized Pareto distribution1987293339349906643ZBL0628.62019HusseinA.LiuJ.Parametric estimation of mixtures of two uniform distributions2009793-4395410252235810.1080/00949650701810406ZBL1169.62011ElamirE. A. H.SeheultA. H.Trimmed L-moments2003433299314199681410.1016/S0167-9473(02)00250-5ElamirE. A. H.SeheultA. H.Exact variance structure of sample L-moments20041242337359208036910.1016/S0378-3758(03)00213-1NecirA.MeraghniD.Estimating L-functionals for heavy-tailed distributions and application2010201034707146260289910.1155/2010/707146JonesM. C.On some expressions for variance, covariance, skewness and L-moments2004126197106209068810.1016/j.jspi.2003.09.0012008Palo Alto, Calif, USATIBCO SoftwareMattsonM. E.AllenJ. P.MillerW. R.HesterR. K.ConnorsG. J.RychtarikR. G.RandallC. L.AntonR. F.KaddenR. M.CooneyN. L.DiClementeC. C.CarbonariJ.ZwebenA.LongabaughR.StoutR.WalkerR. D.DonovanD.BaborT. F.DelBocaF. K.Project MATCH: Rationale and methods for a multisite clinical trial matching patients to alcoholism treatment1993176113011452-s2.0-0027817306AllenJ. P.MattsonM. E.MillerW. R.ToniganJ. S.ConnorsG. J.RychtarikR. G.RandallC. L.AntonR. F.KaddenR. M.LittM.CooneyN. L.DiClementeC. C.CarbonariJ.ZwebenA.LongabaughR. H.StoutR. L.DonovanD.BaborT. F.Del BocaF. K.RounsavilleB. J.CarrollK. M.WirtzP. W.BaileyS.BradyK.CislerR.HesterR. K.KivlahanD. R.NirenbergT. D.PateL. A.SturgisE.Matching alcoholism treatments to client heterogeneity: project MATCH posttreatment drinking outcomes19975817292-s2.0-8044253999WolframS.19995thChampaign, Ill, USAWolfram Mediaxxvi+1470RennerR. M.Evaluation by power series of the means of normal order statistics of samples of sizes six and seven197642-31411470408078ZBL0356.62041