Hacker News Reader

Comment by Gormo | Hacker News Reader

Interesting photo, bad statistical analysis.

The probability of a bird flying in front of the camera exactly as it takes a photo is in no way dependent on how long the camera has been running. Every image capture is an independent event.

There are also many, many webcams in existence, and while there is a low probability of a bird appearing in a photo captured by any specific camera, the probability of some bird appearing in front of some camera at some time would be much higher, and would likely have resulted in a similar amount of attention from bloggers.

View on HN

Every image is an independent event so the likelihood of a bird appearing in a specific image is unrelated to how long the camera is up. However the likelihood of the bird appearing in any of the images the camera has taken depends on the total number of images. So the analysis is sound.

However I do agree with the second half of you post, the probability of some bird clearly appearing in some shot from some webcam somewhere in the world is very likely near 1.

Good point; I initially interpreted the article's analysis as an attempt to calculate the prior probability of a bird appearing in a specific photo. But then I switched to thinking about the probability of the bird appearing within a sample of photos, which of course depends on the sample size.

Of course, you need to know the prior probabilities in order to calculate the posterior probability distribution, and in this scenario, there doesn't appear to be a reliable way to obtain the prior probabilities, especially considering that there's only a single instance of the event we're looking at in the sample data.

The probability of an event occuring once you know it occured is, surprisingly, "very likely near 1".

Sure, if you're looking at the probability of that specific event occurring.

But if you want to generalize about kinds of events, rather than a single specific one, it's useful to use past events as sample data to estimate probabilities of similar events occurring in the future (provided, of course, that you have defined meaningful categories for entities and events.)

Reply on news.ycombinator.com