QUESTION: Based on the given sample size, what is the margin of error?
ANSWER: Let's calculate the margin of error.
Disclaimer: I'm not a statistics guy by any stretch, indeed, needed to google the formula to get the answer.
Source: http://www.dummies.com/how-to/content/how-to-calculate-the-margin-of-error-for-a-sample0.html
Assumptions: For this argument we will assume 3 fires per 19000 Model S. We will assume confidence interval of 95% is desired as per BillHamp's post, and as it seems to be a rather common confidence interval.
[...]
you can conclude with 95% confidence that somewhere between -0.0021% and 0.0337% of Model S cars catch fire.
My take: Looks to me that we still have a pretty wide margin of error (so as everyone has been suggesting, we need a larger sample size before we really know what's going on). Based on this calculation, for the next 19000 cars we should expect anywhere between approximately 0 and 6 fires with 95% confidence.
Thanks for digging to the ground here, Clay. But I think it was the wrong hole. The linked math page talks about standard normal (Z-) distribution which is applicable to values strewing left and right of an average value. As Mario Kadastik already pointed out, with 3 fires in 19.000 cars we have a distribution that is near the limit of 0. You applied the error margin and got a negative value for the left hand border value of the 95% confidence interval. This indicates trouble for the validity of your results.
For values strewing near a border value (e.g. zero here), the Poisson distribution must be applied. I don't grok the math there ATM.
But any calculation for the 95% confidence interval should result in positive error margin values for the 3 fires. Further, the probability for ZERO fires for one year while the car population grew from 0 to 19.000 should lie in the 95% confidence interval, too. If these two cannot be fulfilled, we don't have a statistical (=random) series of events and other factors are in play.