Having $S_n$ being a sum of i.i.d. Bernoulli($p$) random variable, where does $S_n/\sqrt{n}$ converge to?

435 Views Asked by At

In our lecture, we learned about the CLT and we were shown the following application:

Let $X_1, X_2, \dots$ ~ $B(p)$, meaning that $P(X_i = 1) = p \in (0,1)$. Additionally, let $X_1, X_2, \dots$ be independent and identically distributed. Let $S_n = \sum_{i=1}^nX_i$, then (knowing the moment-conditions are fulfilled), the following convergence holds:

$\frac{S_n - np}{\sqrt{np(1-p)}} \longrightarrow N(0,1)$

Since $E[S_n] = np$ and $Var(S_n)=np(1-p)$ with $N(0,1)$ denoting a standard-normal.

Since also the following equation holds:

$\frac{S_n - np}{\sqrt{n}} \longrightarrow N(0,p(1-p))$

I was wondering if (or why?) $\frac{S_n}{\sqrt{n}} \longrightarrow N(p,p(1-p))$ is also a right expression.

Having $E[\frac{S_n}{\sqrt{n}}] = \sqrt{n}p$, I do not understand why the expectation of the asymptotic normal distribution should be simply p (which is what my lecturer told us).

Can anyone please help out?

Thanks.

1

There are 1 best solutions below

5
On

Since $\frac{S_n-np}{\sqrt{n}} =\frac{S_n}{\sqrt{n}} - \sqrt{n}p$ converges in distribution to $\mathcal N\left(0,p(1-p)\right)$ as $n$ increases,

and $\sqrt{n}p$ increases without bounds as $n$ increases,

you will have $\frac{S_n}{\sqrt{n}}$ diverging to $+\infty$.

What your lecturer may have said is that $\frac{S_n}{n} \to p$ in probability and almost surely. This is the law of large numbers; note there is no square root in the denominator.