Abstract
The most often used distribution for modelling directional data has been the circular normal (CN) (a.k.a. von-Mises) distribution. Recently Kato and Jones (K–J) introduced a family of distribution which includes the CN distribution as a special case. We study the SB-robustness of the circular mean functional (CMF) and show that the CMF is not SB-robust at the family of all symmetric Kato–Jones distributions but is SB-robust at sub-families with bounded parameters. It is also found to be SB-robust for certain sub-families of wrapped-t (WT) distributions, mixtures of K–J distributions and mixtures of K–J and WT distributions. The SB-robustness of the circular trimmed mean functional (CTMF) is also studied and it is found that the CTMF is SB-robust for larger sub-families of symmetric Kato–Jones distributions compared to that of CMF. The SB-robustness of the CMF for asymmetric families of distributions is studied and it is shown that CMF is SB-robust at a sub-family of asymmetric Kato–Jones distributions. The performance of CTM is compared with that of circular mean (CM) through extensive simulation. It is seen that CTM has better robustness properties than the CM both theoretically and practically. Some guidelines for choice of trimming proportion for CTM is given.
Similar content being viewed by others
Change history
12 September 2019
Unfortunately, due to a technical error, the articles published in issues 60:2 and 60:3 received incorrect pagination. Please find here the corrected Tables of Contents. We apologize to the authors of the articles and the readers.
References
Abramowitz M, Stegun IA (1965) Handbook of mathematical functions. Dover, New York
Azzalini A, Genton MG (2008) Robust likelihood methods based on the skew-t and related distributions. Int Stat Rev 76(1):106–129
Bickel PJ (1976) Another look at robustness: a review of reviews and some new developments. Scand J Stat 3:145–168
Bickel PJ, Lehmann EL (1976) Descriptive statistics for nonparametric models. III. Dispersion. Ann Stat 4(6):1139–1158
Collet D (1980) Outliers in circular data. Appl Stat 29:50–57
Elton SD, Gray DA (1994) The application of circular statistics to specific radar pulse train detection, In Proc.EUSIPCO-94, vol 1, pp 284–287
Ferguson DE, Landreth HF, Mckeown JP (1967) Sun compass orientation of the northern cricket frog, Acris crepitans. Anim Behav 15:45–53
Fisher NI (1993) Statistical analysis of circular data. Cambridge University Press, Cambridge
Fisher NI, Lewis T (1983) Estimating the common mean direction of several circular or spherical distributions with differing dispersions. Biometrika 70:333–341
Gill J, Hangartner D (2010) Circular data in political science and how to handle it. Polit Anal 18:316–336
Gradshteyn IS, Ryzhik IM (1994) Table of integrals, series, and products, 5th edn. Academic Press, San Diego
Gradshteyn IS, Ryzhik IM (2000) Table of integrals, series, and products, 6th edn. Academic Press, San Diego
Hampel FR (1974) The Influence curve and its role in robust estimation. J Am Stat Assoc 69:383–393
Huber PJ (1964) Robust estimation of a location parameter. Ann Math Stat 35:73–101
Jammalamadaka SR, SenGupta A (2001) Topics in circular statistics. World Scientific, Singapore
Kato S, Jones MC (2010) A family of distributions on the circle with links to, and applications arising from mobius transformation. J Am Stat Assoc 105(489):249–262
Kato S, Eguchi S (2016) Robust estimation of Location and concentration parameters for the von Mises–Fisher distribution. Stat Pap 57(1):205–234
Kato S, Shimizu K (2004) A further study of t-distributions on spheres. Technical report, School of fundamental science and technology, Keio University, Yokohama
Kim S, SenGupta A (2013) A three-parameter generalized von Mises distribution. Stat Pap 54(3):685–693
Ko D, Chang T (1993) Robust M-estimators on spheres. J Multivar Anal 45:104–136
Ko D, Guttorp P (1988) Robustness of estimators for directional data. Ann Stat 16:609–618
Laha AK, Mahesh KC (2011) SB-robustness of directional mean for circular distributions. J Stat Plan Inference 141:1269–1276
Laha AK, Mahesh KC (2012) SB-robust estimator for the concentration parameter of circular normal distribution. Stat Pap 53(2):457–467
Laha AK, Mahesh KC, Ghosh DK (2013) SB-robust estimators of the parameters of the wrapped normal distribution. Commun Stat 42:660–672
Mardia KV, Jupp PE (2000) Directional statistics. Wiley, Chichester
Oldham K, Myland J, Spanier J (2009) An atlas of functions with equator, the atlas function calculator, 2nd edn. Springer, New York
Patricia L, Morellato C, Alberti LF, Hudson IL (2010) Applications of circular statistics in plant phenology; a case study approach. Methods for environmental and climate change analysis., Phenological research Springer, Dordrecht
Pewsey A (2000) The wrapped skew-normal distribution on the circle. Commun Stat 29:2459–2472
Pewsey A, Jones MC (2005) A family of symmetric distributions on the circle. J Am Stat Assoc 100(472):1422–1428
Pewsey A, Lewis T, Jones MC (2007) The wrapped t family of circular distributions. Aust N Z J Stat 49(1):79–91
Rudin W (1987) Real and complex analysis, 3rd edn. McGraw-Hill, New York
SenGupta A, Laha AK (2001) The slippage problem for the circular normal distribution. Aust N Z J Stat 43(4):461–471
Shimatani IK, Yoda K, Katsumata N, Sato K (2012) Toward the quantification of conceptual framework for movement ecology using circular statistical modeling. PLoS ONE 7(11):1–12
Stigler SM (1973) The asymptotic distribution of the trimmed mean. Ann Stat 1(3):472–477
Stigler SM (2010) The changing history of robustness. Am Stat 64(4):277–281
Umbach D, Jammalamadaka SR (2009) Building asymmetry into circular distributions. Stat Probab Lett 79:659–663
Wehrly ET, Shine EP (1981) Influence curves of estimators for directional data. Biometrika 68:334–335
Acknowledgements
The authors express gratitude to the anonymous referees and the editor for their valuable comments which led to substantial improvement of the article.
Author information
Authors and Affiliations
Corresponding author
Additional information
Publisher's Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Appendix
Appendix
Lemma 1
Consider the family of distributions\(\mathfrak {I}=\left\{ KJ\left( {0,\kappa ,r} \right) :\kappa >0,0\le r\le r_0 <1 \right\} \). Let \( S(F)=\sqrt{1-\rho _F }\)and \(S^{*}(F)=E_F (d(\Theta ,0))\)for \(\mathrm{F}\in \mathfrak {I}\). Then \(S\mathop \sim \limits ^\mathfrak {I}S^{*}\).
Proof
We note that
is a continuous function of \(\upkappa \) and r. Also,
is a continuous function of \(\kappa \) and r which is positive on the domain \(\kappa >0\) and \(0\le r\le r_0 <1\).
Hence, \(R(F)=\frac{S(F)}{S^{*}(F)}\) is a continuous function of \(\kappa \) and r on the domain \(\kappa >0\) and \(0\le r\le r_0 <1\). When \(r=0\) we have \(\mathop {\lim }\limits _{\kappa \rightarrow \infty } R(F)=\sqrt{\pi }\) and \(\mathop {\lim }\limits _{\kappa \rightarrow 0} R(F)=\frac{2}{\pi }\) (see proof of Theorem 5.2 of Laha and Mahesh 2011). When \(r\ne 0\) and \(\upkappa \rightarrow \infty \) using the fact that \(KJ\left( {0,\kappa ,r} \right) \) tends to the \(N\left( {0,\kappa ^{-1}\omega _r^2 } \right) \) where \(\omega _\mathrm{r} =\frac{1-r}{1+r}\), we have
Expanding and simplifying we get,
Straightforward calculations show that \(S^{*}\left( F \right) =\omega _r \sqrt{\frac{2}{\pi \kappa }}\) . Then we have \(\mathop {\lim }\limits _{\kappa \rightarrow \infty } R(F)=\frac{\sqrt{\pi }}{2}.\) Again, when \(r\ne 0\) and \(\kappa \rightarrow 0\), using the fact \(\mathrm{KJ}\left( {\mathrm{0,}\,\kappa \mathrm{,r}} \right) \) tends to \(WC\left( {0,\mathrm{r}} \right) \), we have \(S(F)=\sqrt{1-r}\) and \(S^{*}(F)=\frac{1-r^{2}}{2\pi }\int \limits _0^{2\pi } {\frac{\left| {\pi -|\pi -\theta |} \right| }{1+r^{2}-2r\cos \theta }d\theta } \). Using the series expansion of \(\frac{1-r^{2}}{1+r^{2}-2r\cos \theta }=1+2\sum \limits _{p=1}^\infty {(r)^{p}} \cos p\theta \) (see Gradshteyn and Ryzhik 2000), we get \(S^{*}(F)=\frac{\pi }{2}-\frac{4}{\pi }\sum \limits _{n=0}^\infty {\frac{r^{(2n+1)}}{\left( {2n+1} \right) ^{2}}} \).
Therefore, \(\mathop {\lim }\limits _{\kappa \rightarrow 0} R(F)=\sqrt{1-r}\left( {\frac{\pi }{2}-\frac{4}{\pi }\sum _{n=0}^\infty {\frac{r^{(2n+1)}}{\left( {2n+1} \right) ^{2}}} } \right) ^{-1}\).
From the above we can conclude that \(\mathop {\mathrm{sup}}\limits _\mathfrak {I}\mathrm{R}\left( \mathrm{F} \right) <\infty \). Also, it can be seen that \(\mathop {\mathrm{sup}}\limits _\mathfrak {I}\mathrm{R}^{-1}\left( \mathrm{F} \right) <\infty \). Hence \(S\mathop \sim \limits ^\mathfrak {I}S^{*}\).
Hence the lemma is established. \(\square \)
Lemma 2
Consider the family of distributions\(\mathfrak {I}_1^*=\left\{ \alpha \mathrm{KJ}(0,\kappa _1 ,r_1 )+ (1-\alpha )\mathrm{{KJ}}(0,\,\kappa _2 ,r_2 )\mathrm{; 0}<\mathrm{m}\le \kappa _1 ,\kappa _2 \le M,\mathrm{0}\le \mathrm{r}_1 \mathrm{,r}_2 \le \mathrm{r}_0<1,0\le \alpha <1 \right\} .\)Then \(S\mathop \sim \limits ^{\mathfrak {I}_1^*} S^{*}\).
Proof
Let \(D^{*}=\left\{ (\kappa _1 ,\kappa _2 \mathrm{,r}_1 \mathrm{,r}_2 \mathrm{,}\alpha \mathrm{):0}<\mathrm{m}\le \kappa _1 ,\kappa _2 \le M,\mathrm{0}\le \mathrm{r}_1 \mathrm{,r}_2 \le \mathrm{r}_0<1,0\le \alpha <1 \right\} \). Since the p.d.f of \(\alpha ~\mathrm{KJ}\,(0,\kappa _1 ,r_1 )+ (1-\alpha )~\mathrm{KJ}\,(0,\kappa _2 ,r_2 )\) is continuous in \(\kappa _1 ,\kappa _2 ,\mathrm{r}_1 \mathrm{,r}_2 \) and \(\alpha \) we can conclude that both \(S(F)=\sqrt{1-\left| {E_F \left( {\cos \Theta } \right) } \right| }\) and \(S^{*}(F)=E_F (d(\Theta ,0))\) are continuous functions of \(\kappa _1 ,\kappa _2 ,\mathrm{r}_1 \mathrm{,r}_2 \) and \(\alpha \) on the set \(D^{*}\) where \(\mathrm{F}=\alpha \mathrm{F}_1 +\left( {1-\alpha } \right) \mathrm{F}_2 \), \(\mathrm{F}_1 =\mathrm{KJ}\left( {0,\kappa _1 \mathrm{,r}_1 } \right) \) and \(\mathrm{F}_2 =\mathrm{KJ}\left( {0,\kappa _2 \mathrm{,r}_2 } \right) \). Since for all \(F\in \mathfrak {I}_1^*\) we have \(R(F)=S(F)S^{*-1}(F)\) is a continuous function of \(\kappa _1 ,\kappa _2 ,\mathrm{r}_1 \mathrm{,r}_2 \) and \(\alpha \). Therefore, \(\mathop {\mathrm{sup}}\limits _{\mathfrak {I}_1^*}\, \mathrm{R}\left( \mathrm{F} \right) <\infty \). Similarly we can prove \(\mathop {\mathrm {sup}}\limits _{\mathfrak {I}_1^{*} }\, {\mathrm R}^{-1}\left( {\mathrm F} \right) <\infty \). Hence \(S\mathop \sim \limits ^{\mathfrak {I}_1^*} S^{*}\). Hence the lemma is established. \(\square \)
Lemma 3
Consider the family of distributions
Then\(S\mathop \sim \limits ^{\mathfrak {I}_1 } S^{*}\).
Proof
We have \(S\left( F \right) =\sqrt{1+C\left( \nu \right) -\alpha \left( {C\left( \nu \right) +\int \limits _0^{2\pi } {\cos \theta \, \mathrm{f}_1 \left( \theta \right) \mathrm{d}\theta } } \right) }\) and
Note that \(I_1 \) is a continuous function of \(\kappa \) and r which is positive on the domain \(\left\{ {\kappa >0} \right\} \cap \{0\le r\le r_0 <1\}\). Also, following arguments given in the proof of theorem 3 it can be seen that, as \(\nu \rightarrow \infty \), \(I_2 \) is finite. Hence we can conclude that \(\mathop {\mathrm{sup}}\limits _{\mathfrak {I}_1 }\, \mathrm{R}\left( \mathrm{F} \right) <\infty \). Following arguments similar to above, it can be seen that \(\mathop {\mathrm{sup}}\limits _{\mathfrak {I}_1 }\, \mathrm{R}^{-1}\left( \mathrm{F} \right) <\infty \). Hence \(S\mathop \sim \limits ^{\mathfrak {I}_1 } S^{*}\).
Hence the lemma is established. \(\square \)
Lemma 4
Consider the family of distributions
\(\mathfrak {I}_2 =\left\{ {\mathrm{{WT}}\left( {0,\uplambda ,\upnu } \right) :0<\uplambda _0 \le \uplambda \le \uplambda _1 <\infty ,\upnu \ge \upnu _0 } \right\} .\)Then \(S\mathop \sim \limits ^{\mathfrak {I}_2 } S^{*}\).
Proof
Let \(S(F)=Q_1 \left( {\lambda ,\nu } \right) =\sqrt{1-C\left( {\lambda ,\nu } \right) }\) and \(S^{*}(F)=Q_2 (\lambda ,\nu )=\int \limits _0^{2\pi } {\left| {\pi -\left| {\pi -\theta } \right| } \right| } f\left( \theta \right) d\theta \) where \(f(\theta )\) is density of \(WT\left( {0,\lambda ,\nu } \right) \) distribution. Note that both are continuous functions of \(\lambda \) and \(\nu \), \(\mathop {\lim }\limits _{\nu \rightarrow \infty } Q_1 (\lambda ,\nu )<\infty \) and \(\mathop {\lim }\limits _{\nu \rightarrow \infty } Q_2 (\lambda ,\nu )<\infty \). Hence \(\mathop {\mathrm{sup}}\limits _{\mathfrak {I}_2 } \mathrm{R}\left( \mathrm{F} \right) <\infty \) and \(\mathop {\mathrm{sup}}\limits _{\mathfrak {I}_2 } \mathrm{R}^{\mathrm{-1}}\left( \mathrm{F} \right) <\infty .\) Thus \(S\mathop \sim \limits ^{\mathfrak {I}_2 } S^{*}\).
Hence the lemma is established. \(\square \)
Lemma 5
Let \(F\in \mathfrak {I}_1^{**} \). In the expansion \(\rho _{\gamma ,\mathrm{F}} =C_1 +\frac{C_2 }{\kappa }+\frac{C_3 }{\kappa ^{2}}+\cdots \), we have \(C_1 >0\).
Proof
Consider \(\rho _{\gamma ,F} =\frac{1}{1-2\gamma }\int \limits _{B_\gamma }^{A_\gamma } {\cos \theta ~~\mathrm{f}\left( \theta \right) } d\theta =\frac{2}{1-2\gamma }\int \limits _0^{{\omega _r }/{\sqrt{\kappa }}} {\cos \theta ~\mathrm{f}~\left( \theta \right) } d\theta \). Now, since \(\cos \theta ~~\mathrm{f}\left( \theta \right) \) is a monotonically decreasing function in the interval \(\left( {0,A_\gamma } \right) \) we have
Using the series expansion of \(\mathrm{cos }\left( {\frac{\omega _r }{\sqrt{\kappa }}} \right) \) and replacing \(f\left( \theta \right) \) by \(\mathrm{f}\left( {\frac{\omega _r }{\sqrt{\kappa }}} \right) \) we get, \(\rho _{\gamma ,F} \ge \frac{1}{\pi \left( {1-2\gamma } \right) }\mathop {\lim }\limits _{\kappa \rightarrow \infty } \frac{\exp \left( {\tau \left( {\kappa ,\mathrm{r}} \right) } \right) }{\sqrt{\kappa }I_0 \left( \kappa \right) }\) where \(\tau \left( {\kappa ,\mathrm{r}} \right) =\left\{ {\frac{\kappa \left\{ {\left( {1+r^{2}} \right) \left( {1-\frac{\omega _r^2 }{\kappa }} \right) -2r} \right\} }{1+r^{2}-2r\left( {1-\frac{\omega _r^2 }{\kappa }} \right) }-\kappa } \right\} \).
Using \(I_0 \left( \kappa \right) =\frac{\exp \left( \kappa \right) }{\sqrt{2\pi \kappa }}\) for large \(\kappa \)(see Oldham et al. 2009) and simplifying \(\exp \left( {\tau \left( {\kappa ,\mathrm{r}} \right) } \right) \) as \(\kappa \rightarrow \infty \) we get \(\rho _{\gamma ,F} \ge \frac{\sqrt{2}}{\sqrt{\pi }\left( {1-2\gamma } \right) }\frac{1}{\omega _r }\exp \left( {-\frac{1}{2\omega _r^2 }} \right) =\tau _1 \left( r \right) \). Since \(0\le \mathrm{r}_0 \le r\le r_1 <1\) we get \(\tau _1 \left( {r_1 } \right) <1\). Also we know that \(\tau _1 \left( r \right) >0\) and hence \(0<\tau _1 \left( {r_1 } \right) <1\). Thus, \(C_1 >0\).
Hence the lemma is established. \(\square \)
Lemma 6
Let \(\mathfrak {I}_1^{**} =\left\{ {KJ\left( {0,\kappa ,r} \right) \mathrm{; }\kappa >M,\mathrm{0}\le \mathrm{r}_0 \le r\le r_1 <1} \right\} \)and \(\mathfrak {I}_2^{**} =\left\{ {KJ\left( {0,\kappa ,r} \right) \mathrm{; }\kappa<m,\mathrm{0}\le \mathrm{r}_0 \le r\le r_1 <1} \right\} \). Then (a) \(S_1 \mathop \sim \limits ^{\mathfrak {I}_1^{**} } S_2 \)and (b) \(S_1 \mathop \sim \limits ^{\mathfrak {I}_2^{**} } S_2 \).
Proof
(a) Note that as \(\kappa \rightarrow \infty \), the \(\mathrm{{KJ}}\left( {0,\upkappa , \mathrm{r}} \right) \) tends to \(\mathrm{N}\left( {0, \upomega _\mathrm{r}^2 } \right) \) distribution where \(\omega _\mathrm{r} =\frac{1-\mathrm{r}}{1+\mathrm{r}}\). Using this we get \(\mathrm{A}_\upgamma =\frac{\upomega _r \Phi ^{\mathrm{-1}}\left( {1-\upgamma } \right) }{\sqrt{\upkappa }}\). Since \(\mathrm{{S_2}} \left( \mathrm{F} \right) =\sqrt{\mathrm{E}_{\upgamma ,\mathrm{F}} \left( {1-\cos \Theta } \right) }\), using Jensen’s inequality we get
Note that \({S}_1 ({F})=\frac{2}{\left( {1-2\gamma } \right) }\int \limits _0^{{ A}_\gamma } {\theta {f}\left( \theta \right) } {d}\theta \) and hence \(\mathop {\sup }\limits _{\mathfrak {I}_1^{**} } \frac{\mathrm{S}_1 (F)}{\mathrm{S}_2 (F)}\le \sqrt{2}<\infty \). Further,
Hence \(\frac{\mathrm{S}_2 \left( \mathrm{F} \right) }{\mathrm{S}_1 \left( \mathrm{F} \right) }\cong \frac{\left( {\Phi ^{-1}\left( {1-\upgamma } \right) } \right) ^{{-1}/2}\sqrt{\left( {1-2\upgamma } \right) }\left[ {\frac{1}{3}-\frac{\left( {\Phi ^{-1}\left( {1-\upgamma } \right) } \right) ^{2}}{10}} \right] ^{1/2}}{\left[ {1-\frac{\left( {\Phi ^{-1}\left( {1-\upgamma } \right) } \right) ^{2}}{4}} \right] }\). Therefore, \(\mathop {\sup }\limits _{\mathfrak {I}_1^{**} } \frac{\mathrm{S}_2 (\mathrm{F})}{\mathrm{S}_1 (\mathrm{F})}<\infty \).
Thus,\(S_1 \mathop \sim \limits ^{\mathfrak {I}_1^{**} } S_2 \).
(b) We have \(S_1 \left( F \right) =\frac{1}{2\pi \left( {1-2\gamma } \right) }\left\{ {\alpha ^{2}\left( r \right) +4\sum \limits _{p=1}^\infty {r^{p}} \left( {\frac{\alpha \left( r \right) S_{p,\mathrm{0}} }{\mathrm{p}}+\frac{C_{p,\mathrm{0}} -1}{\mathrm{p}^{2}}} \right) } \right\} \) where \(S_{p,\nu } =\sin \left[ {p\alpha \left( r \right) -\nu } \right] \) and \(C_{p,\nu } =\cos \left[ {p\alpha \left( r \right) -\nu } \right] \) and \(S_2 \left( F \right) =\sqrt{1-\rho _{\gamma ,\mathrm{F}} }\). Now a straight forward calculation yields \(\mathop {\sup }\limits _{\mathfrak {I}_2^{**} } R(F)<\infty \) and \(\mathop {\sup }\limits _{\mathfrak {I}_2^{**} } R^{-1}(F)<\infty \). Thus, \(S_1 \mathop \sim \limits ^{\mathfrak {I}_2^{**} } S_2 \).
Hence the lemma is established. \(\square \)
Lemma 7
Let \(H\left( r \right) =\int \limits _0^\pi {\frac{\left( {1-r^{2}} \right) \theta }{1+r^{2}-2r\cos \theta }} d\theta \), then \(\mathop {\lim }\limits _{r\rightarrow 1^{-}} H\left( r \right) =0\).
Proof
Let \(g\left( \theta \right) =\frac{\theta }{1+r^{2}-2r\cos \theta }\), \(g^{{\prime }}\left( \theta \right) =\frac{\left( {1+r^{2}-2r\cos \theta } \right) -\theta \left( {2r\sin \theta } \right) }{\left( {1+r^{2}-2r\cos \theta } \right) ^{2}}\)
Now let \(k\left( \theta \right) =\cos \theta +\theta \sin \theta \). Then \(k^{{\prime }}\left( \theta \right) =\theta \cos \theta \). Note that \(k^{{\prime }}\left( \theta \right) \ge 0\) for all \(\theta \in \left[ {0,\frac{\pi }{2}} \right] \) and \(k^{{\prime }}\left( \theta \right) <0 \) for all \(\theta \in \left[ {\frac{\pi }{2},\pi } \right] \). Therefore \(k\left( \theta \right) \) is increasing in \(\left[ {0,\frac{\pi }{2}} \right] \) and decreasing in \(\left( {\frac{\pi }{2},\pi } \right] \). Hence \(k\left( \theta \right) \) has a maximum value at \(\theta =\frac{\pi }{2}\).
Thus \(k\left( 0 \right) \le \theta \sin \theta +\cos \theta \le k\left( {\frac{\pi }{2}} \right) \Rightarrow \)
Now since \(\frac{1+r^{2}}{2r}>1\) for \(0<r<1\) and \(\frac{1+r^{2}}{2r}\rightarrow 1\)as \(r\rightarrow 1^{-}\). Using (9) we get \(\frac{1+\mathrm{r}^{2}}{2r}=\cos \theta +\theta \sin \theta \) has a solution \(\theta ^{*}\)for r close to 1. Observe that \(g^{{\prime }}\left( \theta \right) >0\) when \(\theta <\theta ^{*}\)and \(g^{{\prime }}\left( \theta \right) <0\) for \(\theta >\theta ^{*}\). Therefore \(g\left( \theta \right) \) has a maximum value at \(\theta ^{*}\).
Further note that \(\theta ^*=\theta _{r} \rightarrow 0\) as \(r\rightarrow 1^{-}\). Using the approximation \(\sin \theta \approx \theta \) for small \(\theta \) and \(\cos \theta \approx 1-\frac{\theta ^{2}}{2}\) for small \(\theta \), we have
Solving we get \(\theta ^{*}=\frac{\left( {1-r} \right) }{\sqrt{r}}\) for r close to 1.
Thus \(g\left( {\theta ^{*}} \right) =\frac{\theta ^{*}}{1+r^{2}-2r\cos \theta ^{*}}\approx \frac{1}{\sqrt{r}\left( {1-r} \right) }\) for r close to 1.
Now \(\int \limits _0^\pi {\frac{\left( {1-r^{2}} \right) \theta }{1+r^{2}-2r\cos \theta }d\theta } =\int \limits _0^\delta {\frac{\left( {1-r^{2}} \right) \theta }{1+r^{2}-2r\cos \theta }d\theta +} \int \limits _\delta ^\pi {\frac{\left( {1-r^{2}} \right) \theta }{1+r^{2}-2r\cos \theta }d\theta } \) for any small \(\delta >0\).
Now \(\mathop {\lim }\limits _{r\rightarrow 1^{-}} \int \limits _\delta ^\pi {\frac{\left( {1-r^{2}} \right) \theta }{1+r^{2}-2r\cos \theta }d\theta =\int \limits _\delta ^\pi {\mathop {\lim }\limits _{r\rightarrow 1^{-}} } } \frac{\left( {1-r^{2}} \right) \theta }{1+r^{2}-2r\cos \theta }d\theta =0\) as \(\frac{\left( {1-r^{2}} \right) \theta }{1+r^{2}-2r\cos \theta }\) is a continuous function of r. Again, \(\int \limits _0^\delta {\frac{\left( {1-r^{2}} \right) \theta }{1+r^{2}-2r\cos \theta }d\theta } =\left( {1-r^{2}} \right) \int \limits _0^\delta {g\left( \theta \right) } d\theta \le \left( {1-r^{2}} \right) \delta g\left( {\theta ^{*}} \right) \approx \frac{1-r^{2}}{\sqrt{r}\left( {1-r} \right) }\delta =\frac{\left( {1+r} \right) \delta }{\sqrt{r}}\) for r close to 1. Therefore \(\int \limits _0^\delta {\frac{\left( {1-r^{2}} \right) \theta }{1+r^{2}-2r\cos \theta }} d\theta \le \frac{\left( {1+r} \right) }{\sqrt{r}}\delta \le 3\delta \) for r close to 1. Since \(\delta \) is arbitrary, we conclude\(\mathop {\lim }\limits _{r\rightarrow 1^{-}} \int \limits _0^\pi \frac{\left( {1-r^{2}} \right) \theta }{1+r^{2}-2r\cos \theta }d\theta =0 \). \(\square \)
Rights and permissions
About this article
Cite this article
Laha, A.K., Pravida Raja, A.C. & Mahesh, K.C. SB-robust estimation of mean direction for some new circular distributions. Stat Papers 60, 877–902 (2019). https://doi.org/10.1007/s00362-016-0853-9
Received:
Revised:
Published:
Issue Date:
DOI: https://doi.org/10.1007/s00362-016-0853-9