I am struggling with a small part of the proof of the law of large numbers.

I understand from Markov's inequality: $$P(X\ge t) \le \frac{E(X)}{t} $$ and therefore if $ X = (\bar{Y} - E(Y))^2$

$$P(|\bar{Y} - E(Y)|\ge t) \le \frac{\sigma^2}{nt^2} $$

However, what i don't quite understand is what happens when we switch the inequality on the left from $\ge t$ to $\le t$.

I understand the fact that if $P(X>t)=c$ then $P(X<t)=1-c$, however I am finding it hard to think about why the correct answer is:

$$P(|\bar{Y} - E(Y)|\le t) \ge 1-\frac{\sigma^2}{nt^2} $$

and not:

$$P(|\bar{Y} - E(Y)|\le t) \le 1-\frac{\sigma^2}{nt^2} $$

  • $\begingroup$ It's a typo. Your answer is correct and your argument proves it. $\endgroup$ – Michael Feb 12 '17 at 20:28
  • $\begingroup$ Sorry I haven't worded it correctly, but I am not clear why either of the two options is correct. I don't fully understand why the less than sign is changed to greater than. $\endgroup$ – user3084100 Feb 12 '17 at 20:45
  • 2
    $\begingroup$ If you have $1-c \leq A$ then $-c \leq A-1$ then $ c\geq 1-A$. Does that help? For you $A=\sigma^2/n t^2$. $\endgroup$ – Michael Feb 12 '17 at 20:48
  • $\begingroup$ Regarding your first comment, I've changed the order of the equations so my original question makes sense. Thanks for your follow up comment, I was originally just converting the right-hand side to 1-c and being confused why the inequality doesn't match. I realise I should convert the left side to 1-P(...<t) then work forwards using your method to get the right answer! $\endgroup$ – user3084100 Feb 12 '17 at 21:25
  • $\begingroup$ Since you appear to have arrived at an answer based on Michael's comments, please post it (with proper attribution of the ideas contributed by Michael, naturally) $\endgroup$ – Glen_b Feb 13 '17 at 2:23

So we know that:

$$P(|\bar{Y} - E(Y)|\ge t) = 1 - P(|\bar{Y} - E(Y)|\le t) $$

and from the question:

$$P(|\bar{Y} - E(Y)|\ge t) \le \frac{\sigma^2}{nt^2} $$


$$1 - P(|\bar{Y} - E(Y)|\le t) \le \frac{\sigma^2}{nt^2} $$

Equal to :

$$1 \le \frac{\sigma^2}{nt^2} + P(|\bar{Y} - E(Y)|\le t) $$ $$1 - \frac{\sigma^2}{nt^2} \le P(|\bar{Y} - E(Y)|\le t) $$

So as n approaches infinity:

$$1 \le P(|\bar{Y} - E(Y)|\le t) $$

Thanks to Michael for clearing up the problem in the comments.

| cite | improve this answer | |

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.