Nonparametric Interval Estimators for the Coefficient of Variation

Dongliang Wang; Margaret K. Formica; Song Liu

doi:10.1515/ijb-2017-0041

Artikel

Nonparametric Interval Estimators for the Coefficient of Variation

Dongliang Wang , Margaret K. Formica und Song Liu

Veröffentlicht/Copyright: 19. April 2018

Veröffentlicht von

Veröffentlichen auch Sie bei De Gruyter Brill

Manuskript einreichen Informationen für Autor*innen Erkunden Sie dieses Fachgebiet

Aus der Zeitschrift The International Journal of Biostatistics Band 14 Heft 1

Abstract

The coefficient of variation (CV) is a widely used scaleless measure of variability in many disciplines. However the inference for the CV is limited to parametric methods or standard bootstrap. In this paper we propose two nonparametric methods aiming to construct confidence intervals for the coefficient of variation. The first one is to apply the empirical likelihood after transforming the original data. The second one is a modified jackknife empirical likelihood method. We also propose bootstrap procedures for calibrating the test statistics. Results from our simulation studies suggest that the proposed methods, particularly the empirical likelihood method with bootstrap calibration, are comparable to existing methods for normal data and yield better coverage probabilities for nonnormal data. We illustrate our methods by applying them to two real-life datasets.

Keywords: coefficient of variation; empirical likelihood; Jackknife empirical likelihood; bootstrap; Wilks’ theorem

Acknowledgements

We are grateful to the constructive comments from the Associated Editor and the three anonymous Referees, which notably improved the quality of our manuscript.

Appendix

5.1 Proof of Theorem 2.2

Proof

Firstly we show that at τ=τ0, Un(τ) can be written as a U-statistic

Un(τ)=n2−1∑1≤i1\lti2≤nh(Xi1,Xi2;τ),

where h(X1,X2;τ)=12τ2(X12+X22)−12(τ2+1)(X1−X2)2, since

s2=1n−1∑i=1n(Xi−Xˉ)2=n2−1∑1≤i1\lti2≤n12(Xi1−Xi2)2.

We then show that Eh(X1,X2;τ)=0, which can be easily derived by noting the facts that

τ2=σ2μ2=σ2EX2−σ2,σ2=12E(X1−X2)2 and

EX2=12E(X12+X22)

Thus Theorem 2.2 follows directly from Theorem 2.1 in Jing et al. (2009).

5.2 R script for the proposed methods

[baselinestretch=0.75]if (F){ install.packages(“emplik”)}library(emplik)######################## EL functions######################cvhat4el.f = function(x){## CV estimate for EL n = length(x) m = floor(n/2) y = z = rep(NA,m) for (i in 1:m){ y[i] = (x[i]-x[m+i])^2/2 z[i] = (x[i]^2+x[m+i]^2)/2 } cvhat = sqrt(mean(y)/mean(z-y)) out = list(cvhat=cvhat,x=x, y=y, z=z) return(out)}el.f = function(y,z,tau){ zvals = y-tau^2*(z-y) tt = el.test(x=zvals,mu=0) ll = tt$“-2LLR” if(abs(sum(tt$wts)-length(y))>1) ll = 300 out = list(“-2LLR” = ll,zvals = zvals,tau = tau,n = n) return(out)}ci.el.f = function(x,avals = c(0.10,0.05),B = 1000,step = 0.01){ nalpha = length(avals) ci.alpha = array(NA,dim = c(nalpha,3,2)) cx.alpha = round(qchisq(1-avals,1),2) ## cut-off for chi-square n = length(x) m = floor(n/2) ttt = cvhat$el.f(x) cvhat = ttt$cvhat y = ttt$y z = ttt$z llboot = rep(NA,B) for (i in 1:B){ idx = sample(1:m,replace=T) ystar = y[idx] zstar = z[idx] llboot[i] = el.f(tau=cvhat, y=ystar, z=zstar)$“-2LLR” } cboot.alpha = round(quantile(llboot,prob=1-avals,na.rm=T),2) for (i in 1:nalpha){ if (el.f(tau = 1000,y = y, z = z)$“-2LLR” >= cx.alpha[i]){ ci = findUL(step = step, fun = el.f, MLE = cvhat, y = y, z = z, level = cx.alpha[i]) ci.alpha[i,1,] = c(ci$Low,ci$Up) } ci.alpha[i,3,] = ci.alpha[i,1,] if (el.f(tau = 1000,y = y, z = z)$“-2LLR” >=cboot.alpha[i]){ ci = findUL(step = step, fun = el.f, MLE = cvhat, y = y, z = z,level = cboot.alpha[i]) ci.alpha[i,3,] = c(ci$Low,ci$Up) } } out = list(ci.alpha = ci.alpha,x = x,avals = avals, B = B,cx.alpha = cx.alpha, cboot.alpha = cboot.alpha) return(out)}######################## JEL######################cvhat4jel.f=function(x){## CV estimate for JEL n=length(x) y=x^2 tt1 = mean(y) tt2 = sd(x)^2 cvhat = sqrt(tt2/(tt1-tt2)) out=list(cvhat=cvhat,x=x) return(out)}U_n.f=function(x,tau){ n=length(x) y=x^2 tt1 = mean(y) tt2 = sd(x)^2 tt=tt1*tau^2-(tau^2+1)*tt2 out=list(U=tt, tt1=tt1, tt2=tt2,x=x) return(out)}jkkf.f=function(x,tau){ n=length(x) Un=U_n.f(x,tau)$U Un1=vv=rep(NA,n) for (i in 1:n){ Un1[i]=U_n.f(x[-i],tau)$U } vjack=n*Un-(n-1)*Un1 out=list(vjack=vjack,n=n,tau=tau,x=x,Un=Un) return(out)}jel.f=function(tau,x){ n=length(x) vjack=jkkf.f(x=x,tau=tau)$vjack tt = el.test(x=vjack,mu=0) ll=tt$“-2LLR” if(abs(sum(tt$wts)-length(x))>1) ll=300 out=list(“-2LLR”=ll,vjack=vjack,tau=tau,n=n) return(out)}ci.jel.f=function(x,avals=c(0.10,0.05),B=1000,step=0.01){ n=length(x) cvhat=cvhat4jel.f(x)$cvhat nalpha=length(avals) ci.alpha=array(NA,dim=c(nalpha,3,2)) llboot=rep(NA,B) for (i in 1:B){ xstar=sample(x,n,replace=T) llboot[i]=jel.f(tau=cvhat,x=xstar)$“-2LLR” } cx.alpha=qchisq(1-avals,1) ## cut-off for chi-square cboot.alpha=round(quantile(llboot,prob=1-avals,na.rm=T),2) ## bootstrap cutoff for (i in 1:nalpha){ ci.alpha[i,1,]=rep(NA,2) if (jel.f(tau=1000,x=x)$“-2LLR”>=cx.alpha[i]){ ci=findUL(step=step, fun=jel.f, MLE=cvhat, x=x,level=cx.alpha[i]) ci.alpha[i,1,]=c(ci$Low,ci$Up) } ci.alpha[i,3,] = ci.alpha[i,1,] if (jel.f(tau=1000,x=x)$“-2LLR”>=cboot.alpha[i]){ ci=findUL(step=step, fun=jel.f, MLE=cvhat, x=x,level=cboot.alpha[i]) ci.alpha[i,3,]=c(ci$Low,ci$Up) } } out=list(ci.alpha=ci.alpha,x=x,avals=avals,B=B,cx.alpha=cx.alpha, cboot.alpha=cboot.alpha) return(out)}

References

[1] Pearson K. Mathematical contributions to the theory of evolution? III. Regression, heredity and panmixia. Philos Trans R Soc A. 1896;187:253–318.10.1098/rsta.1896.0007Suche in Google Scholar

[2] Reed GF, Lynn F, Meade BD. Use of coefficient of variation in assessing variability of quantitative Assays. Clin Diagn Lab Immunol. 2002;9(6);1235–1239.10.1128/CDLI.9.6.1235-1239.2002Suche in Google Scholar

[3] Chow SC, Wang H. On sample size calculation in bioequivalence trials. J Pharmacok Pharmacod. 2001;28:155–169.10.1023/A:1011503032353Suche in Google Scholar PubMed

[4] Lehmann EL. Testing statistical hypothesis, 2nd ed. New York: Wiley, 1996.Suche in Google Scholar

[5] McKay AT. Distribution of the coefficient of variation and the extended t distribution. J Roy Statist Soc B. 1932;95:695–698.10.2307/2342041Suche in Google Scholar

[6] David FN. Note on the application of Fisher’s k-statistics. Biometrika. 1949;36:383–393.10.1093/biomet/36.3-4.383Suche in Google Scholar

[7] Reh W, Scheffler B. Significance tests and confidence intervals for coefficients of variation. Comput Stat Data Anal. 1996;22(4):449–452.10.1016/0167-9473(96)83707-8Suche in Google Scholar

[8] Vangel MG. Confidence interval for a normal coefficient of variation. The Am Stat. 1996;50:21–26.10.1080/00031305.1996.10473537Suche in Google Scholar

[9] Wong ACM, Wu J. Small sample asymptotic inference for the coefficient of variation: normal and nonnormal models. J Stat Plann Inference. 2002;104:73–82.10.1016/S0378-3758(01)00241-5Suche in Google Scholar

[10] Verrill S, Johnson RA. Confidence bounds and hypothesis tests for normal distribution coefficients of variation. Commun Stat Theory Methods. 2007;36(12):2187–2206.10.2737/FPL-RP-638Suche in Google Scholar

[11] Mahmoudvand R, Hassani H. Two new confidence intervals for the coefficient of variation in a normal distribution. J Appl Stat. 2009;36(4):429–442.10.1080/02664760802474249Suche in Google Scholar

[12] Panichkitkosolkul W. Confidence intervals for the coefficient of variation in a normal distribution with a known population mean. Probab Stat J. 2013;Article ID 324940. DOI: 10.1155/2013/324940.Suche in Google Scholar

[13] Sharma KK, Krishna H. Asymptotic sampling distribution of inverse coefficient of variation and its applications. IEEE Trans Reliab. 1994;43(4):630–633.10.1109/24.370217Suche in Google Scholar

[14] Banik S, Kibria BMG. Estimating the population coefficient of variation by confidence intervals. Commun Stat Simul Comput. 2011;40:1236–1261.10.1080/03610918.2011.568151Suche in Google Scholar

[15] Monika G, Kibria BMG, Albatineh AN, Ahmed NU. A comparison of some confidence intervals for estimating the population coefficient of variation: a simulation study. SORT 2012;36:45–68.Suche in Google Scholar

[16] Albatineh AN, Boubakari I, Kibria BMG. New confidence interval estimator of the signal-to-noise ratio based on asymptotic sampling distribution. Commun Stat Theory Meth. 2015. DOI: 10.1080/03610926.2014.1000498.Suche in Google Scholar

[17] Albatineh AN, Kibria BMG, Wilcox ML, Zogheib B. Confidence interval estimation for the population coefficient of variation using ranked set sampling: a simulation study. J Appl Stat. 2014;41:733–751.10.1080/02664763.2013.847405Suche in Google Scholar

[18] Owen A. Empirical likelihood ratio confidences for single functional. Biometrika. 1988;75:237–249.10.1093/biomet/75.2.237Suche in Google Scholar

[19] Owen AB. Empirical likelihood. Boca Raton, FL: Chapman and Hall/CRC Press, 2001.Suche in Google Scholar

[20] Jing B, Yuan, J, Zhou W. Jackknife empirical likelihood. Journal of the American Statistical Association. 2009;104(487):1224–1232.10.1198/jasa.2009.tm08260Suche in Google Scholar

[21] Peng L, Qi Y. Smoothed jackknife empirical likelihood method for tail copulas. TEST. 2010;19(3):514–536.10.1007/s11749-010-0184-4Suche in Google Scholar

[22] Adimari G, Chiogna M. Jackknife empirical likelihood based confidence intervals for partial areas under ROC curves. Stat Sin. 2012;22:1457–1477.10.5705/ss.2011.088Suche in Google Scholar

[23] Yang H, Zhao Y. Smoothed jackknife empirical likelihood inference for the difference of ROC curves. J Multivariate Anal. 2013;115:270–284.10.1016/j.jmva.2012.10.010Suche in Google Scholar

[24] Yang H, Zhao Y. Smoothed jackknife empirical likelihood inference for ROC curves with missing data. J Multivariate Anal. 2015;140:123–138.10.1016/j.jmva.2015.05.002Suche in Google Scholar

[25] Wang D, Zhao Y, Gilmore DW. Jackknife empirical likelihood confidence interval for the Gini index. Stat Probab Lett. 2016;110:289–295.10.1016/j.spl.2015.09.026Suche in Google Scholar

[26] Wang D, Zhao Y. Jackknife empirical likelihood for comparing two Gini indices. Canadian J Stat. 2016;44(1):102–119.10.1002/cjs.11275Suche in Google Scholar

[27] Canty A, Ripley B. Boot: Bootstrap R (S-Plus) Functions. R package version 1.3-18, 2016.10.1002/9781118445112.stat06177.pub2Suche in Google Scholar

[28] Davison AC, Hinkley DV. Bootstrap methods and their applications. Cambridge: Cambridge University Press, 1997.10.1017/CBO9780511802843Suche in Google Scholar

[29] Zhou M. emplik: Empirical likelihood ratio for censored/truncated data. R package version 1.0-3, 2016.Suche in Google Scholar

[30] Wood RJ, Durham TM. Reproducibility of serological titers. J Clin Microbiol 1980;11:541–545.10.1128/jcm.11.6.541-545.1980Suche in Google Scholar PubMed PubMed Central

[31] Proschan F. Theoretical explanation of observed decreasing failure rate. Technometrics. 1963;5:375–383.10.1080/00401706.1963.10490105Suche in Google Scholar

[32] Gail MH, Gastwirth JL. A scale-free goodness-of-fit test for the exponential distribution based on the Gini statistic. J R Stat Soc Ser B. 1978;40:350–357.10.1111/j.2517-6161.1978.tb01048.xSuche in Google Scholar

[33] Eisenberg DT. Telomere length measurement validity: the coefficient of variation is invalid and cannot be used to compare quantitative polymerase chain reaction and Southern blot telomere length measurement techniques. Int J Epidemiol. 2016;45:1295–1298.10.1093/ije/dyw191Suche in Google Scholar PubMed

Received: 2017-06-05

Revised: 2018-02-15

Accepted: 2018-03-16

Published Online: 2018-04-19

Sie haben derzeit keinen Zugang zu diesem Inhalt.

Artikel in diesem Heft

https://doi.org/10.1515/ijb-2017-0041

Schlagwörter für diesen Artikel

coefficient of variation; empirical likelihood; Jackknife empirical likelihood; bootstrap; Wilks’ theorem