Having trouble interpreting this experiment on the Poisson distribution

84 Views Asked by At

von Bortkiewicz considered the frequency of deaths from kicks in the Prussian army corps. From the study of 14 corps over a 20-year period, he obtained the data shown in the table below. Fit a Poisson distribution to this data.

$$\begin{array}{|c|c|} \hline & \text{Number of deaths} & \text{Number of corps with x deaths in a given year} \\ \hline &0 & 144 \\ \hline &1 & 91\\ \hline &2 & 32 \\ \hline &3 & 11 \\ \hline &4 & 2 \\ \hline \end{array}$$ The solution computes $\lambda = 1\frac{91}{280}+2\frac{32}{280}+3\frac{11}{280}+4\frac{2}{280} = 0.7$ and the expected number of corps with x death as $280P(X = j)$ where P is a Poisson distribution with $\lambda = 0.7$ and X representing the number of death.

I see $\lambda$ as the expected number of death per corp-year (as it appears to be total death divided by total corps-year). If so, since $\lambda = np$, qualitatively what would $n$ and $p$ represent? I initially thought n is a corps-year and p is the probability of death, but this doesn't make sense as if each corps-year is a trial then X would be number of corps-year with a death, not number of death in a corps-year. To me $P(X =j)$ should give the probability of $j$ death in a corps-year.

1

There are 1 best solutions below

0
On BEST ANSWER

You are correct that $\lambda$ is the Poisson rate of deaths per corps-years of exposure. So for a single corps${}^*$, the expected number of deaths in a single year is $0.7$.

Consequently, if $X$ is a random variable denoting the random number of deaths observed in a single corps in a year, then $X \sim \operatorname{Poisson}(\lambda)$ and $$\Pr[X = x] = e^{-\lambda} \frac{\lambda^x}{x!}, \quad x = 0, 1, 2, \ldots.$$ Now, among a sample of $n = 280$ corps, let $Y$ represent the number of corps with exactly $j$ deaths. More specifically, suppose we have $X_1, X_2, \ldots, X_{280} \sim X$ be independent and identically distributed as $X$, where $X_i$ is the random number of deaths in corps $i$ in one year, and $$Y_i = \mathbb{1} (X_i = j)$$ be an indicator random variable that equals $1$ if corps $i$ has exactly $j$ deaths that year. Then $$Y = Y_1 + Y_2 + \cdots + Y_{280}.$$ It is clear that $$\Pr[Y_i = 1] = \Pr[X_i = j] \overset{\text{iid}}{=} \Pr[X = j].$$ So $$Y \sim \operatorname{Binomial}(n = 280, p = \Pr[X = j])$$ and the expectation of $Y$ is $$\operatorname{E}[Y] = np = 280\Pr[X = j]$$ as claimed. Of course, it bears reminding that this is not to say that the expected number of deaths among the $280$ corps is equal to the expected number of corps with $j$ deaths.


${}^*$ The singular and plural of "corps" are both "corps."