1
$\begingroup$

I have been playing a mobile game where the listed drop rate for a loot box prize is 1/3. I have opened 48 loot boxes and received 6 prizes (1/8). It has been so long since stats class, but can't I use a t-test with a 95% confidence interval to find if my H0 should be rejected? I know my sample size is small but RNG can't be that bad, can it? It seems that the drop rate is far less than 1/3.

I'd like to know how to set it up so that I can do it on my own in the future. I'm having trouble understanding my sample SD and using a chart.

$\endgroup$
3
  • $\begingroup$ When did you decide you wanted to test? After seeing the data? see en.wikipedia.org/wiki/Testing_hypotheses_suggested_by_the_data $\endgroup$
    – Glen_b
    Sep 23, 2018 at 8:36
  • $\begingroup$ I started to keep track when it became apparent that I was not getting a 1/3 return. I'm aware of type 1 errors and plan on gathering more samples going forward. I did not want to invest money into a chance game where the return was no where near the posted rate. $\endgroup$
    – James
    Sep 23, 2018 at 16:00
  • $\begingroup$ How did you decide when to test it? Edit: sorry, that's unclear. How did you decide that now was the time to stop collecting data and try to test the hypothesis? $\endgroup$
    – Glen_b
    Sep 23, 2018 at 16:27

1 Answer 1

1
$\begingroup$

What you are looking at is a Bernoulli variable, that is outcome 0 or 1 for one repetition. When considering many such repetitions you are looking at a binomial distribution. There you can calculate the 95 % CI for the null hypothesis (proportion is equal to 1/3) with

qbinom(c(0.025,0.975),48,1/3)

which get you [10,23], so your value of 6 is outside the 95 % CI. The probability of this event (6 out of 48 with true probability 1/3) is

pbinom(6,48,1/3)

which is 0.0009170562.

Edit: if you are not ready to use R then have a look at Wikipedia page where you have an approximate formula using normal distribution https://en.wikipedia.org/wiki/Binomial_proportion_confidence_interval

$\endgroup$
3
  • $\begingroup$ Given OP doesn't study stats, I don't think he will know how to use R, nor any other statistical programming language $\endgroup$
    – Patty
    Sep 23, 2018 at 8:59
  • $\begingroup$ I do not know how to use R. If I use the equation from the wiki page (which I can do), do I recieve a z score that I can look up on a z table? I wasn't sure what [10,23] represented or what I use that with. $\endgroup$
    – James
    Sep 23, 2018 at 16:08
  • $\begingroup$ Yes, the z score can be obtained from a z table, although if you are going to use the 95 % CI you can always use the predefined value of 1.96 for z. [10,23] represents the 95 % Confidence Interval (CI), that is where you can expect 95 % of all values to lie, if your null hypothesis is true. $\endgroup$ Sep 24, 2018 at 6:09

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

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