animatronic
Senior Member
Some thoughts on the arrest predictions:
Normally you can improve predictions by averaging a number of individual predictions - this removes personal bias and takes out the extremes. However, you need to have decent information as an input. Election predictions are a good example - if you survey a bunch of people who have access to polling data and are following the news, you'll usually end up with a decent guess. In many cases the guesses will even fall into a normal distribution. If the histogram doesn't fall into a normal distribution, there's a good chance that consensus on a specific date is lacking. This is a wildly oversimplistic explanation but don't sweat it.
Now, shown below is a histogram showing the predictions to date (the list from this morning plus some of the stragglers). Note that most predictions came in last week, so "week 0" of the series is 23 Feb. Also note that the time series is reasonably logarithmic. The take-away from this is:
- Very few people think the arrest will happen in the current or following week
- Around a quarter to a third of people think the arrest will happen 2-3 weeks out, and half think it will happen in the next month
- There is an emotional tie to 1-year anniversaries of major events
- The guesses are almost evenly distributed by day of the week except for Sat/Sun, which are each half as frequent. There's a slight preference for Thursday/Friday
- The shape of the curve makes it likely that this will be the result regardless of when the survey is taken
- The above will only stay true in the absence of any major developments
Discuss.
View attachment 23645
Normally you can improve predictions by averaging a number of individual predictions - this removes personal bias and takes out the extremes. However, you need to have decent information as an input. Election predictions are a good example - if you survey a bunch of people who have access to polling data and are following the news, you'll usually end up with a decent guess. In many cases the guesses will even fall into a normal distribution. If the histogram doesn't fall into a normal distribution, there's a good chance that consensus on a specific date is lacking. This is a wildly oversimplistic explanation but don't sweat it.
Now, shown below is a histogram showing the predictions to date (the list from this morning plus some of the stragglers). Note that most predictions came in last week, so "week 0" of the series is 23 Feb. Also note that the time series is reasonably logarithmic. The take-away from this is:
- Very few people think the arrest will happen in the current or following week
- Around a quarter to a third of people think the arrest will happen 2-3 weeks out, and half think it will happen in the next month
- There is an emotional tie to 1-year anniversaries of major events
- The guesses are almost evenly distributed by day of the week except for Sat/Sun, which are each half as frequent. There's a slight preference for Thursday/Friday
- The shape of the curve makes it likely that this will be the result regardless of when the survey is taken
- The above will only stay true in the absence of any major developments
Discuss.
View attachment 23645
Attachments
Last edited: