Abstract
Let \(N_\lambda \) and U be two independent random variables respectively distributed as a Poisson distribution with parameter \(\lambda >0\) and a uniform distribution on (0, 1). This paper establishes that the median, say M, of \(N_\lambda +U\) is close to \(\lambda +1/3\) and more precisely that \(M-\lambda -1/3=o(\lambda ^{-1})\) as \(\lambda \rightarrow \infty \). This result is used to construct a very simple robust estimator of \(\lambda \) which is consistent and asymptotically normal. Compared to known robust estimates, this one can still be used with large datasets (\(n\simeq 10^9\)).
Similar content being viewed by others
References
Adell J, Jodrá P (2005) The median of the Poisson distribution. Metrika 61(3):337–346
Chen J, Rubin H (1986) Bounds for the difference between median and mean of Gamma and Poisson distributions. Stat Probab Lett 4(6):281–283
Choi K (1994) On the medians of Gamma distributions and an equation of Ramanujan. Proc Am Math Soc 121(1):245–251
Coeurjolly JF (2017) Median-based estimation of the intensity of a spatial point process. Ann Inst Stat Math 69(2):303–331
Elsaied H, Fried R (2016) Tukey’s M-estimator of the Poisson parameter with a special focus on small means. Stat Methods Appl 25(2):191–209
Machado J, Silva J (2005) Quantiles for counts. J Am Stat Assoc 100(472):1226–1237
Serfling RJ (2009) Approximation theorems of mathematical statistics, vol 162. Wiley, Hoboken
Stevens W (1950) Fiducial limits of the parameter of a discontinuous distribution. Biometrika 37(1/2):117–129
van de Ven R, Weber N (1993) Bounds for the median of the negative binomial distribution. Metrika 40(1):185–189
Van der Vaart A (1998) Asymptotic statistics, vol 3. Cambridge University Press, Cambridge
Acknowledgements
The authors are grateful to reviewers for their suggestions and comments which improved a previous version of this paper. In particular, the authors thank the reviewers for reexpressing a previous version of Remark 1. They are also sincerely grateful to H. Elsaied and R. Fried for discussions and for sharing the R code implementing the Tukey’s modified M-estimator. The research of J.-F. Coeurjollly and J. Rousseau Trépanier is supported by the Natural Sciences and Engineering Research Council of Canada.
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
Appendix gathers technical lemmas, used in the proof of Proposition 1.
Lemma 1
Let \(k\in \mathbb {R}\), \(x\in [0,1)\) and \(r_n(x,k) = k/(n+x)\). Then the sequence \(w_n(x,k)\) reads as follows:
-
(i)
If \(x+r_n(x,k) \in [-1/3,2/3)\)
$$\begin{aligned} w_n(x,k)= \mathrm {P}\left( N_{n+x}\le n \right) +\left( x-\frac{2}{3}+\frac{k}{n+x}\right) \mathrm {P}\left( N_{n+x}=n\right) . \end{aligned}$$(14) -
(ii)
If \(x+r_n(x,k)\in [2/3,5/3)\)
$$\begin{aligned} w_n(x,k)= \mathrm {P}\left( N_{n+x}\le n\right) +\left( x-\frac{2}{3}+\frac{k}{n+x}\right) \mathrm {P}(N_{n+x}=n+1). \end{aligned}$$(15)
Let \(k\in \mathbb {R}\), then for n sufficiently large, if \(x\in [0,2/3)\) or \(x=2/3\) and \(k<0\), then \(w_n(x,k)\) reads as in (14). In the same way, if \(x\in (2/3,1)\) or \(x=2/3\) and \(k\ge 0\), then \(w_n(x,k)\) reads as in (15). The proof of Lemma 1 is omitted as it derives easily from (5).
Lemma 2
Let \(k\in \mathbb {R}\), \(x\in [0,1)\) and \(r_n(x,k) = k/(n+x)\). Let \(\varDelta _n(x,k)\) be given by
There exists \(n_0 \in \mathbb {N}\) such that for all \(n\ge n_0\), we have the two following cases:
-
(i)
If \(x+r_{n}(x,k) \in [-1/3,2/3)\)
$$\begin{aligned} \varDelta _n(x,k) =&\;c_n(0,x)-\int _0^1 c_n(v,x)\mathrm {d}v + \left( x-\frac{2}{3} \right) \left( 1-c_n(0,x) \frac{n+1}{n+x}\right) \nonumber \\&+ r_{n+1}(x,k)- \frac{n+1}{n+x} c_n(0,x)r_n(x,k) \end{aligned}$$(16)where \(c_n(\cdot ,x):[0,1]\rightarrow \mathbb {R}\) is defined by
$$\begin{aligned} c_n(v,x)= \left( \frac{n+v+x}{n+1+x} \right) ^{n+1} \exp (1-v). \end{aligned}$$ -
(ii)
If \(x+r_{n}(x,k) \in [2/3,5/3)\)
$$\begin{aligned} \varDelta _n(x,k) =&\,c_n(0,x) -\int _0^1 c_n(v,x)\mathrm {d}v +\left( x-\frac{2}{3} \right) \left( \frac{n+1+x}{n+2}-c_n(0,x) \right) \nonumber \\&+ \frac{n+1+x}{n+2}r_{n+1}(x,k)- c_n(0,x)r_n(x,k). \end{aligned}$$(17)
Proof
-
(i)
Using the Poisson-Gamma relation \(\mathrm {P}(N_\lambda \le n)=\frac{1}{n!}\int _\lambda ^\infty g_n(u)\mathrm {d}u\) with \(g_n(u)=\mathrm {e}^{-u}u^n\) and Lemma 1 (i), we can rearrange the difference \(w_{n+1}(x,k)-w_n(x,k)\) as
$$\begin{aligned} w_{n+1}(x,k)-w_n(x,k)&=\frac{1}{(n+1)!}\int _{n+x}^{n+1+x}-g_{n+1}(u)du+\frac{g_{n+1}(n+x)}{(n+1)!}\\&\qquad +\left( x-\frac{2}{3}+r_{n+1}(x,k)\right) \mathrm {P}\left( N_{n+1+x}=n+1\right) \\&\qquad -\left( x-\frac{2}{3}+r_n(x,k)\right) \mathrm {P}\left( N_{n+x}=n\right) , \end{aligned}$$which leads to the result after little algebra by noticing that
$$\begin{aligned} \mathrm {P}(N_{n+x}=n)=\frac{(n+1)(n+x)^n\mathrm {e}^{-(n+x)}}{(n+1)!}=\frac{(n+1)g_n(n+x)}{(n+1)!}. \end{aligned}$$ -
(ii)
Using the Poisson-Gamma relation and Lemma 1 (ii), we can rearrange the difference \(w_{n+1}(x,k)-w_n(x,k)\) as
$$\begin{aligned} w_{n+1}(x,k)-w_n(x,k) =&\,\frac{1}{(n+1)!}\int _{n+1+x}^{\infty }\mathrm {e}^{-u}u^{n+1}du\\&-\frac{1}{(n+1)!}\int _{n+x}^{\infty }\mathrm {e}^{-u}u^{n+1}du\\&+\left( x-\frac{2}{3}+r_{n+1}(x,k)\right) \mathrm {P}\left( N_{n+1+x}=n+2\right) \\&-\left( x-\frac{2}{3}+r_n(x,k)\right) \mathrm {P}\left( N_{n+x}=n+1\right) , \end{aligned}$$which leads to the result after little algebra by noticing that
$$\begin{aligned} \mathrm {P}(N_{n+1+x}=n+2)=\frac{1}{(n+1)!}g_{n+1}(n+1+x)\frac{n+1+x}{(n+2)} \end{aligned}$$and
$$\begin{aligned} \mathrm {P}(N_{n+x}=n+1)=\frac{1}{(n+1)!}g_{n+1}(n+1+x)c_n(0,x). \end{aligned}$$
\(\square \)
Lemma 3
Let \(k\in \mathbb {R}\), \(x\in [0,1)\) and \(r_n(x,k) = k/(n+x)\), for any \(k \in \mathbb {R}\) and \(x\in [0,1)\), then for n sufficiently large we have
where \({\mathcal {H}}\) is the function given by (7).
Proof
Let \(k\in \mathbb {R}\), then for any \(x\in [0,1)\), there exists \(n_0\in \mathbb {N}\) such that for \(n\ge n_0\), either (i) \(x+r_{n}(x,k)\in [-1/3,2/3)\) or (ii) \(x+r_n(x,k)\in [2/3,5/3)\). In the sequel, we consider both cases and expand the expression of \(\varDelta _n(x,k)\) given by Lemma 2. The following expansions extensively make use of auxiliary results gathered in Lemma 4.
-
(i)
Case \(x+r_{n}(x,k)\in [-1/3,2/3)\).
$$\begin{aligned} \varDelta _n(x,k)&=\frac{1}{n+1+x}\left( -\frac{x}{2}+\frac{1}{6}\right) +\frac{1}{(n+1+x)^2}\left( -\frac{x^2}{6}-\frac{x}{24}+\frac{7}{120}\right) \\&\qquad +\frac{1}{n+1+x}\left( x-\frac{1}{2}\right) +\frac{1}{(n+1+x)^2}\left( \frac{x^2}{2}-\frac{5}{24}\right) \\&\qquad +\left( x-\frac{2}{3}\right) \left[ \frac{-1/2}{n+1+x}+\frac{1}{(n+1+x)^2}\left( \frac{x^2}{2}-\frac{x}{2}-\frac{7}{24}\right) \right] \\&\qquad -\frac{3k}{2}\frac{1}{(n+1+x)^2}+o\left( \frac{1}{n^2}\right) \\&=\frac{3}{2(n+1+x)^2}\left( \frac{x^2(x-1)}{3}+\frac{4}{135}-k\right) +o\left( \frac{1}{n^2}\right) \\&=\frac{3}{2(n+1+x)^2} \big ( {\mathcal {H}}(x) - k \big ) \; + \; o\left( \frac{1}{n^2}\right) . \end{aligned}$$ -
(ii)
Case \(x+r_{n}(x,k)\in [2/3,5/3)\).
$$\begin{aligned} \varDelta _n(x,k)&=\frac{1}{n+1+x}\left( -\frac{x}{2}+\frac{1}{6}\right) +\frac{1}{(n+1+x)^2}\left( -\frac{x^2}{6}-\frac{x}{24}+\frac{7}{120}\right) \\&\qquad +\frac{1}{n+1+x}\left( x-\frac{1}{2}\right) +\frac{1}{(n+1+x)^2}\left( \frac{x^2}{2}-\frac{5}{24}\right) \\&\qquad +\left( x-\frac{2}{3}+r_{n+1}(x,k)\right) \left( \frac{n+1+x}{n+2}\right) \\&\qquad -\left( x-\frac{2}{3}+r_n(x,k)\right) c_n(0,x)\\&=\frac{1}{(n+1+x)^2}\left( \frac{x^3}{2}-2x^2+\frac{5x}{2}-\frac{43}{45}-\frac{3k}{2}\right) +o\left( \frac{1}{n^2}\right) \\&=\frac{3}{2(n+1+x)^2}\left( \frac{x^3}{3}-\frac{4x^2}{3}+\frac{5x}{3}-\frac{86}{135}-k\right) +o\left( \frac{1}{n^2}\right) \\&=\frac{3}{2(n+1+x)^2} \big ( {\mathcal {H}}(x) - k \big ) \; + \; o\left( \frac{1}{n^2}\right) . \end{aligned}$$
\(\square \)
Lemma 4
Let \(v,x \in [0,1]\), then we have the following expansions as \(n \rightarrow \infty \):
-
(i)
$$\begin{aligned} c_n(v,x)=&\,1+\frac{1}{n+1+x}\left( x(1-v)-\frac{(1-v)^2}{2}\right) \nonumber \\&+\frac{1}{(n+1+x)^2}\left( \frac{(1-v)^2}{2}(x^2+x)\right. \nonumber \\&\left. -(1-v)^3\left( \frac{x}{2}+\frac{1}{3}\right) +\frac{(1-v)^4}{8}\right) \nonumber \\&+o\left( \frac{1}{n^2}\right) . \end{aligned}$$(18)
-
(ii)
$$\begin{aligned}&c_n(0,x) =1+\frac{1}{n+1+x}\left( x-\frac{1}{2}\right) +\frac{1}{(n+1+x)^2}\left( \frac{x^2}{2}-\frac{5}{24}\right) +o\left( \frac{1}{n^2}\right) . \end{aligned}$$(19)
-
(iii)
$$\begin{aligned} \int _0^1c_n(v,x)dv=&\,1+\frac{1}{n+1+x}\left( \frac{x}{2}-\frac{1}{6}\right) \nonumber \\&+\frac{1}{(n+1+x)^2}\left( \frac{x^2}{6}+\frac{x}{24}-\frac{7}{120}\right) +o\left( \frac{1}{n^2}\right) . \end{aligned}$$(20)
-
(iv)
$$\begin{aligned} 1-c_n(0,x)\frac{n+1}{n+x}= & {} \frac{-1/2}{n+1+x}+\frac{1}{(n+1+x)^2}\left[ \frac{x^2}{2}-\frac{x}{2}-\frac{7}{24}\right] \nonumber \\&+o\left( \frac{1}{n^2}\right) . \end{aligned}$$(21)
-
(v)
$$\begin{aligned} r_{n+1}(x,k)- \frac{n+1}{n+x} c_n(0,x)r_n(x,k)=\frac{-3k}{2}\frac{1}{(n+1+x)^2}+o\left( \frac{1}{n^2}\right) . \end{aligned}$$(22)
-
(vi)
$$\begin{aligned} \frac{n+1+x}{n+2}-c_n(0,x)= & {} \frac{1/2}{n+1+x} + \frac{1}{(n+1+x)^2} \left( \frac{x^2}{2}-2x -\frac{5}{24}\right) \nonumber \\&+o\left( \frac{1}{n^2} \right) . \end{aligned}$$(23)
-
(vii)
$$\begin{aligned} \frac{n+1+x}{n+2}r_{n+1}(x,k)- c_n(0,x)r_n(x,k) = \frac{-3k}{2}\frac{1}{(n+1+x)^2}+o\left( \frac{1}{n^2}\right) .\nonumber \\ \end{aligned}$$(24)
The proof which relies upon simple Taylor expansions, is omitted to save space. Details can be provided upon request from the authors.
Rights and permissions
About this article
Cite this article
Coeurjolly, JF., Rousseau Trépanier, J. The median of a jittered Poisson distribution. Metrika 83, 837–851 (2020). https://doi.org/10.1007/s00184-020-00765-3
Received:
Published:
Issue Date:
DOI: https://doi.org/10.1007/s00184-020-00765-3