21

If I have random variables $X_1,X_2,\ldots,X_n$ that are Poisson distributed with parameters $\lambda_1, \lambda_2,\ldots, \lambda_n$, what is the distribution of $Y=\left\lfloor\frac{\sum_{i=1}^n X_i}{n}\right\rfloor$ (i.e. the integer floor of the average)?

A sum of Poissons is also Poisson, but I am not confident enough in statistics to determine if it is the same for the case above.

Lubo Antonov
  • 335
  • 1
  • 2
  • 6
  • @amoeba I rolled back your edit of the title because this is not actually "rounding." Cardinal's previous edit, although not quite as precise, seems preferable because it's accurate. – whuber Nov 22 '17 at 19:51
  • @whuber Okay. I was hesitating when making this edit, but decided to include the word "rounding" because currently the title does not hint to the main difficulty here (and so is in a way misleading). The proper term should be "rounding down", so maybe "What is the distribution of an average of Poisson random variables, *rounded down*?" - even though I will admit it sounds a bit cumbersome. – amoeba Nov 22 '17 at 19:54
  • @amoeba Further edits are of course welcome! – whuber Nov 22 '17 at 19:56

4 Answers4

28

A generalization of the question asks for the distribution of $Y = \lfloor X/m \rfloor$ when the distribution of $X$ is known and supported on the natural numbers. (In the question, $X$ has a Poisson distribution of parameter $\lambda = \lambda_1 + \lambda_2 + \cdots + \lambda_n$ and $m=n$.)

The distribution of $Y$ is easily determined by the distribution of $mY$, whose probability generating function (pgf) can be determined in terms of the pgf of $X$. Here's an outline of the derivation.


Write $p(x) = p_0 + p_1 x + \cdots + p_n x^n + \cdots$ for the pgf of $X$, where (by definition) $p_n = \Pr(X=n)$. $mY$ is constructed from $X$ in such a way that its pgf, $q$, is

$$\eqalign{q(x) &=& \left(p_0 + p_1 + \cdots + p_{m-1}\right) + \left(p_m + p_{m+1} + \cdots + p_{2m-1}\right)x^m + \cdots + \\&&\left(p_{nm} + p_{nm+1} + \cdots + p_{(n+1)m-1}\right)x^{nm} + \cdots.}$$

Because this converges absolutely for $|x| \le 1$, we can rearrange the terms into a sum of pieces of the form

$$D_{m,t}p(x) = p_t + p_{t+m}x^m + \cdots + p_{t + nm}x^{nm} + \cdots$$

for $t=0, 1, \ldots, m-1$. The power series of the functions $x^t D_{m,t}p$ consist of every $m^\text{th}$ term of the series of $p$ starting with the $t^\text{th}$: this is sometimes called a decimation of $p$. Google searches presently don't turn up much useful information on decimations, so for completeness, here's a derivation of a formula.

Let $\omega$ be any primitive $m^\text{th}$ root of unity; for instance, take $\omega = \exp(2 i \pi / m)$. Then it follows from $\omega^m=1$ and $\sum_{j=0}^{m-1}\omega^j = 0$ that

$$x^t D_{m,t}p(x) = \frac{1}{m}\sum_{j=0}^{m-1} \omega^{t j} p(x/\omega^j).$$

To see this, note that the operator $x^t D_{m,t}$ is linear, so it suffices to check the formula on the basis $\{1, x, x^2, \ldots, x^n, \ldots \}$. Applying the right hand side to $x^n$ gives

$$x^t D_{m,t}[x^n] = \frac{1}{m}\sum_{j=0}^{m-1} \omega^{t j} x^n \omega^{-nj}= \frac{x^n}{m}\sum_{j=0}^{m-1} \omega^{(t-n) j.}$$

When $t$ and $n$ differ by a multiple of $m$, each term in the sum equals $1$ and we obtain $x^n$. Otherwise, the terms cycle through powers of $\omega^{t-n}$ and these sum to zero. Whence this operator preserves all powers of $x$ congruent to $t$ modulo $m$ and kills all the others: it is precisely the desired projection.

A formula for $q$ follows readily by changing the order of summation and recognizing one of the sums as geometric, thereby writing it in closed form:

$$\eqalign{ q(x) &= \sum_{t=0}^{m-1} (D_{m,t}[p])(x) \\ &= \sum_{t=0}^{m-1} x^{-t} \frac{1}{m} \sum_{j=0}^{m-1} \omega^{t j} p(\omega^{-j}x ) \\ &= \frac{1}{m} \sum_{j=0}^{m-1} p(\omega^{-j}x) \sum_{t=0}^{m-1} \left(\omega^j/x\right)^t \\ &= \frac{x(1-x^{-m})}{m} \sum_{j=0}^{m-1} \frac{p(\omega^{-j}x)}{x-\omega^j}. }$$

For example, the pgf of a Poisson distribution of parameter $\lambda$ is $p(x) = \exp(\lambda(x-1))$. With $m=2$, $\omega=-1$ and the pgf of $2Y$ will be

$$\eqalign{ q(x) &= \frac{x(1-x^{-2})}{2} \sum_{j=0}^{2-1} \frac{p((-1)^{-j}x)}{x-(-1)^j} \\ &= \frac{x-1/x}{2} \left(\frac{\exp(\lambda(x-1))}{x-1} + \frac{\exp(\lambda(-x-1))}{x+1}\right) \\ &= \exp(-\lambda) \left(\frac{\sinh (\lambda x)}{x}+\cosh (\lambda x)\right). }$$

One use of this approach is to compute moments of $X$ and $mY$. The value of the $k^\text{th}$ derivative of the pgf evaluated at $x=1$ is the $k^\text{th}$ factorial moment. The $k^\text{th}$ moment is a linear combination of the first $k$ factorial moments. Using these observations we find, for instance, that for a Poisson distributed $X$, its mean (which is the first factorial moment) equals $\lambda$, the mean of $2\lfloor(X/2)\rfloor$ equals $\lambda- \frac{1}{2} + \frac{1}{2} e^{-2\lambda}$, and the mean of $3\lfloor(X/3)\rfloor$ equals $\lambda -1+e^{-3 \lambda /2} \left(\frac{\sin \left(\frac{\sqrt{3} \lambda }{2}\right)}{\sqrt{3}}+\cos \left(\frac{\sqrt{3} \lambda}{2}\right)\right)$:

Means

The means for $m=1,2,3$ are shown in blue, red, and yellow, respectively, as functions of $\lambda$: asymptotically, the mean drops by $(m-1)/2$ compared to the original Poisson mean.

Similar formulas for the variances can be obtained. (They get messy as $m$ rises and so are omitted. One thing they definitively establish is that when $m \gt 1$ no multiple of $Y$ is Poisson: it does not have the characteristic equality of mean and variance) Here is a plot of the variances as a function of $\lambda$ for $m=1,2,3$:

Variances

It is interesting that for larger values of $\lambda$ the variances increase. Intuitively, this is due to two competing phenomena: the floor function is effectively binning groups of values that originally were distinct; this must cause the variance to decrease. At the same time, as we have seen, the means are changing, too (because each bin is represented by its smallest value); this must cause a term equal to the square of the difference of means to be added back. The increase in variance for large $\lambda$ becomes larger with larger values of $m$.

The behavior of the variance of $mY$ with $m$ is surprisingly complex. Let's end with a quick simulation (in R) showing what it can do. The plots show the difference between the variance of $m\lfloor X/m \rfloor$ and the variance of $X$ for Poisson distributed $X$ with various values of $\lambda$ ranging from $1$ through $5000$. In all cases the plots appear to have reached their asymptotic values at the right.

set.seed(17)
par(mfrow=c(3,4))
temp <- sapply(c(1,2,5,10,20,50,100,200,500,1000,2000,5000), function(lambda) {
  x <- rpois(20000, lambda)
  v <- sapply(1:floor(lambda + 4*sqrt(lambda)), 
              function(m) var(floor(x/m)*m) - var(x))
  plot(v, type="l", xlab="", ylab="Increased variance", 
       main=toString(lambda), cex.main=.85, col="Blue", lwd=2)
})

Plots

whuber
  • 281,159
  • 54
  • 637
  • 1,101
  • 1
    This is a great answer! It will probably take me some time to digest :) – Lubo Antonov Aug 27 '12 at 09:41
  • 1
    and that is why I said "Using the floor function ... affects the variance slightly too though in a more complicated manner." – Henry Aug 28 '12 at 08:34
  • 1
    +1 Thanks for the detailed answer. There certainly are complicated ways in which the floor function affects the variance. – Dilip Sarwate Aug 28 '12 at 10:39
  • 1
    @Henry Yes, indeed: although the last simulation to study how the variance is affected is a little tangential to the question, I was so struck by the complex pattern that I had to share it. Who would have predicted those logarithmically-spaced peaks in the plots? It would be fair to take my answer as a mere gloss on yours (responding to Dilip's request in a comment): I did not even begin writing it until reading yours with appreciation (and, of course, voting it up). – whuber Aug 28 '12 at 14:26
  • 1
    +1 for simulation in R with code --- this is a very nice example of using `sapply()` for simulation. Thanks. – Assad Ebrahim Aug 21 '13 at 15:05
  • The answer from @whuber is mostly correct. Except that the pgf of $X$ is $E(s^X)=p_0 + p_1s + \cdots p_ns^n + \cdots$ for any $s \in {\rm I\!R}$ for which the sum converges. That is, throughout the response $s$ should replace (all small caps) $x$. Without this change the entire answer falls apart. For example, "the value of the $kth$ derivative of the pgf evaluated at $x=1$ is the $kth$ factorial moment" makes no sense unless $x=1$ is replaced with $s=1$. Minus this issue, great job! – Roberto Rivera Dec 27 '17 at 20:06
  • 2
    @Roberto Thank you. However, the distinction between "$x$" and "$s$", being purely a matter of notation, is utterly trivial and of no mathematical or statistical import. – whuber Dec 27 '17 at 20:16
13

As Michael Chernick says, if the individual random variables are independent then the the sum is Poisson with parameter (mean and variance) $\sum_{i=1}^{n} \lambda_i$ which you might call $\lambda$.

Dividing by $n$ reduces the mean to $\lambda / n$ and variance $\lambda / n^2$ so the variance will be less than the equivalent Poisson distribution. As Michael says, not all values will be integers.

Using the floor function reduces the mean slightly, by about $\frac12 -\frac{1}{2n}$, and affects the variance slightly too though in a more complicated manner. Although you have integer values, the variance will still be substantially less than the mean and so you will have a narrower distribution than the Poisson.

Henry
  • 30,848
  • 1
  • 63
  • 107
  • thanks, not a result I can use, but at least I know now :) – Lubo Antonov Aug 25 '12 at 00:11
  • If the lambdas are not all equal, shouldn't the result be more like a negative binomial than a Poisson (ignoring the non-integer part for the moment)? What am I missing here? – gung - Reinstate Monica Aug 25 '12 at 16:06
  • 2
    @gung: You are missing the point that the individual $\lambda_i$ only affect the distribution through their sum and how many there are. It doesn't matter what particular values they take: $\lambda_1=1, \lambda_2=2, \lambda_3=9$ will give the same result as $\lambda_1=4, \lambda_2=4, \lambda_3=4$. – Henry Aug 25 '12 at 16:13
11

The probability mass function of the average of $n$ independent Poisson random variables can be written down explicitly, though the answer might not help you very much. As Michael Chernick noted in comments on his own answer, the sum $\sum_i X_i$ of independent Poisson random variables $X_i$ with respective parameters $\lambda_i$ is a Poisson random variable with parameter $\lambda = \sum_i \lambda_i$. Hence, $$P\left\{ \sum_{i=1}^n X_i= k\right\} = \exp(-\lambda)\frac{\lambda^k}{k!}, ~~ k = 0, 1, 2, \ldots,$$ Thus, $\hat{Y} = n^{-1} \sum_{i=1}^n X_i$ is a random variable taking on value $k/n$ with probability $\exp(-\lambda)\frac{\lambda^k}{k!}$. Note that $\hat{Y}$ is not an integer-valued random variable (though it does take on uniformly-spaced rational values). It follows easily that $Y = \lfloor \hat{Y} \rfloor$ is an integer-valued random variable taking on value $m$ with probability $$P\{Y = m\} = P\left\{\left\lfloor \frac{1}{n}\sum_{i=1}^n X_i \right\rfloor = m\right\} = \exp(-\lambda)\sum_{i=0}^{n-1}\frac{\lambda^{mn+i}}{(mn+i)!}, ~~ m = 0, 1, 2, \ldots,$$ This is not the probability mass function of a Poisson random variable. Formulas for the mean and variance can be written down using this probability mass function, but they don't obviously lead to nice simple answers in terms of $\lambda$ and $n$. Approximate values can be obtained as pointed out by Henry.

Dilip Sarwate
  • 41,202
  • 4
  • 94
  • 200
  • +1 There *are* closed formulas for the moments of $Y$, though. – whuber Aug 25 '12 at 04:21
  • Thanks for the rigorous formulation! Any chance you'd like to take a crack at the formulas for mean and variance? – Lubo Antonov Aug 25 '12 at 08:05
  • 2
    Perhaps @whuber will post a link (or a citation of a book or journal article) where the closed-form formulas for the moments can be found, or will write an answer giving the formulas themselves, with or without a detailed derivation. – Dilip Sarwate Aug 25 '12 at 13:22
  • @Dilip My claim about closed formulas was not based on anything published, so I have posted a separate reply indicating what I had in mind and how it might be used to understand this situation. – whuber Aug 26 '12 at 20:48
4

Y will not be Poisson. Note that Poisson random variables take on non negative integer values. Once you divide by a constant you create a random variable that can have non-integer values. It will still have the shape of the Poisson. It is just that the discrete probabilities may occur at non-integer points.

Michael R. Chernick
  • 39,640
  • 28
  • 74
  • 143
  • That makes sense, but what if $Y$ is actually discrete, for example the floor of the average? Would that make it Poisson? – Lubo Antonov Aug 24 '12 at 22:23
  • @lucas1024 I don't think so but I am not sure. – Michael R. Chernick Aug 24 '12 at 22:30
  • The shape of the sum $\sum X_i$ is definetevely Poisson, right? its mean and variance are identical as well. Isn't there something like an scaled Poisson ? Y is just a poisson variable (the sum) that is scaled by $n^{-1}$ – JDav Aug 24 '12 at 22:41
  • @JDav The sum is Poisson with the rate parameter equal to the sum of the individual rate parameters. But the OP scales by 1/n and then wants to truncate the the the integer just below Y. I don't know exactly what that does to the distribution. – Michael R. Chernick Aug 24 '12 at 23:08
  • My previous comment assumed independence. – Michael R. Chernick Aug 24 '12 at 23:22
  • The variable $Y$ in the question *does* take on only non-negative integer values. Thus, this post doesn't really seem applicable. It's also incorrect--although not obviously so--to write that $Y$ has "the shape of a Poisson." No version of $Y$, except for the trivial case $n=1$, is Poisson. – whuber Nov 22 '17 at 19:48