Editing 2236: Is it Christmas?

https://isitchristmas.com/ is a popular simplistic website that informs the visitor whether or not it's {{w|Christmas}}. Christmas is a holiday observed in many parts of the world on December 25 of each year. At the top on the tab of the site in the browser it says "Is it Christmas?" with a large '''NO''' printed if it is not December 25, and a '''YES''' if it is December 25. This website asks the user's browser for the date, and updates accordingly if it is indeed Christmas. In addition, isitchristmas.com gives the answer in the language of your region (i.e. for a visitor from Canada, the site gives the answer in English and French to account for Canada's bilingularity, and in most other countries just their word for No will be shown). Since the page uses the browsing computer's time setting, it is possible to easily check that the page works by changing the date on the computer used to access the page to see the text change to Yes if you are reading it on December 25. This also means that the page is only as correct as the time setting on the computer used to view the page (so in case of connection problems, you may check your computer's calendar instead).

Here [[Randall]] spoofs the website. He claims to have made a competitor to isitchristmas.com which nearly always correctly tells if it is Christmas. The joke is that the comic will always display a static image reading '''NO''', even on Christmas Day, and that the rare incorrect answer is rare enough to not cause any concern.

Randall lists a rounded calculation of 99.73% for the precision of his prediction of whether or not it is Christmas. This number is accurate with or without including leap year. An average year is 365.24 days, meaning that he is only wrong 1 out of 365.24 days. So only 1/365.24 ≈ 0.2738% of the days would the prediction be wrong, resulting in a correct reply rate of 99.726%, which he has rounded to 99.73%. Using or not using the leap year will give the same result to three decimal places.  

When building a model for rare events, a common mistake is to ignore the implicit cost function built into the standard prediction accuracy validity statistic for binary events. Prediction accuracy (# correct guesses/total guesses) assumes that false positives and false negatives are equally bad.  Given the implicit cost function of this performance statistic, the best-performing model is commonly a persistence forecast model--i.e., the optimal prediction model returns the most common value whatever the model inputs are. It's probably a better choice to optimize a model using a performance statistic which relies on a cost function that penalizes missing correct prediction of rare events more than it penalizes missing correct prediction of common events.

* [https://knowyourphrase.com/even-a-broken-clock-is-right-twice A stopped watch is accurate twice a day] while a running watch is almost never accurate (and oddly, is more frequently correct the faster/slower it runs). A watch that runs backwards is right 4 times a day.  If you make it spin at thousands of rpm it is right multiple times per second.  (A better metric would be something like the root mean square of the time error -- it's acceptable for a watch to be a little off, as long as it's not off by too much.)