Monotonicity properties of spatial depth
Introduction
Statistical data depth is a mapping that assigns to a given point in , , and a probability distribution on , a non-negative number that characterizes how much “centrally located” is with respect to . The point maximizing the depth generalizes the median to -valued data, and the loci of points of high depth form the central regions of the distribution . A low depth value indicates that is atypical, or far from this centre. A vast body of literature on data depth and its applications is available — for a general account see Zuo and Serfling (2000), for an overview of some applications see Liu et al. (1999).
The spatial depth is a depth function that builds upon the notion of spatial (also called geometric) quantiles for multivariate data, considered by Chaudhuri (1996) and Koltchinskii (1997). These quantiles, and the associated spatial signs and ranks, are indispensable in modern nonparametric statistics of multivariate data (see, e.g., Möttönen and Oja (1995), Möttönen et al. (1997), or the book by Oja 2010).
A depth associated with the spatial quantiles was first considered by Vardi and Zhang (2000). In its current form, the spatial depth was defined by Serfling (2002), and soon after, independently, also by Gao (2003). For and a probability distribution on the spatial depth is given by where , and stands for the Euclidean norm. In the definition and throughout this note, we use the convention .
The spatial depth function has a number of attractive properties. Its maximal value is attained at the spatial median, a robust location parameter well known in statistics. It is invariant with respect to translations and orthogonal rotations of the data. Unlike for many other depth functions, the computation of is extremely simple and fast, also in high-dimensional spaces. Finally, it is well applicable also to infinite-dimensional (functional) data Chakraborty and Chaudhuri (2014a), Chakraborty and Chaudhuri (2014b). The spatial depth and its variants have been successfully used in many practical tasks, see, e.g., Chen et al. (2009), Li et al. (2013), Sguera et al. (2014), Dutta et al. (2016), and Serfling and Wijesuriya (2017).
In the paragraph after Proposition 2 in Gao (2003), a conjecture regarding the behaviour of on rays emanating from the centre of symmetry of is formulated.
Conjecture. For any angularly symmetric1 around ,
For any angularly symmetric around , the spatial depth is maximized at (Gao, 2003 Proposition 2). Thus, condition (C) is in fact a weaker version of the following property, considered by Liu (1990, Theorem 3) and Zuo and Serfling (2000, property P3 and Definition 2.1) :
-
P3.
For any probability distribution on such that ,
This condition, frequently called Monotonicity relative to the deepest point, is standardly recognized as desirable for depth functions. Geometrically, it means that the upper level sets of the spatial depth for form a collection of nested sets, star-shaped relative to the spatial median2 . In particular, they are always connected, and the depth induces a reasonable centre-outwards ordering of the data, as required in applications. See also Liu (1990, Remark A).
In the present note we provide two examples of probability distributions that demonstrate that the conjecture ofGao (2003) does not hold, in general. As a consequence, the spatial depth does not satisfy property P3. In Section 2.1 we present an atomic angularly symmetric distribution on such that (C) is not satisfied for . In Section 2.2 we extend this result to show that also for absolutely continuous, (C) can be violated.
It is important to mention that these examples are not in conflict with the monotonicity property of the spatial distribution, as asserted by Koltchinskii (1997, Proposition 2.4), see also Chakraborty and Chaudhuri (2014b, Theorem 3.1). There, a different version of monotonicity is considered, in no direct relation with (C) or P3. Recall that the spatial distribution of with respect to a probability distribution on is given by This map is a special case of the -distribution, studied in detail by Koltchinskii (1997). The spatial depth can be written as . For the spatial distribution, the following monotonicity property is established in Koltchinskii (1997, Proposition 2.4).
Lemma 1 For any probability distribution
on
and any
it holds true that
.
Herein, stands for the inner product of the vectors .
Using Lemma 1, it is possible to devise a bound on the spatial depth, see Section 2.3. Exploiting this bound, we provide in Section 2.3 a discussion that illustrates the difference between the monotonicity properties of the spatial depth, and the monotonicity of the spatial distribution.
For the sake of clarity, some computational details and a gif animation related to the examples are provided in the online Supplementary Material accompanying this paper.
Section snippets
Examples and discussion
Write for the closed ball centred at with radius . For random vectors and , means that and are identically distributed.
Acknowledgements
The author gratefully acknowledges the helpful suggestions of an anonymous referee. This research is supported by the IAP research network no. P7/06 of the Federal Science Policy (Belgium). The author is a Research Assistant of the Research Foundation—Flanders, and acknowledges support from this foundation.
References (20)
- et al.
The deepest point for distributions in infinite dimensional spaces
Stat. Methodol.
(2014) Data depth based on spatial rank
Statist. Probab. Lett.
(2003)- et al.
Depth-based nonparametric description of functional data, with emphasis on use of spatial depth
Comput. Statist. Data Anal.
(2017) - et al.
The spatial distribution in infinite dimensional spaces and related quantiles and depths
Ann. Statist.
(2014) On a geometric notion of quantiles for multivariate data
J. Amer. Statist. Assoc.
(1996)- et al.
Outlier detection with the kernelized spatial depth function
IEEE Trans. Pattern Anal. Mach. Intell.
(2009) - et al.
Multi-scale classification using localized spatial depth
J. Mach. Learn. Res.
(2016) -estimation, convexity and quantiles
Ann. Statist.
(1997)- et al.
Nonparametric multivariate CUSUM control charts for location and scale changes
J. Nonparametr. Stat.
(2013) On a notion of simplicial depth
Proc. Natl. Acad. Sci. USA
(1988)
Cited by (5)
Model-based statistical depth with applications to functional data
2023, Journal of Nonparametric StatisticsChoosing Among Notions of Multivariate Depth Statistics
2022, Statistical ScienceNonparametric Imputation by Data Depth
2020, Journal of the American Statistical AssociationDepth and depth-based classification with R package ddalpha
2019, Journal of Statistical SoftwareSUPPLEMENTARY MATERIAL: ON A GENERAL DEFINITION OF DEPTH FOR FUNCTIONAL DATA
2017, Statistical Science