View Single Post
Old 09-15-2019, 03:51 PM
Sam Stone is offline
Charter Member
Join Date: Jun 1999
Posts: 28,297
Originally Posted by Buck Godot View Post
I wouldn't put much faith in a study set that small. p. The actual accuracy could be 75% and would still have 13% likelihood of appearing 100% accurate over 7 trials. even more so if you account for cherry picking of economic indeators and the inclusion of a delay that gives them a large amount of wiggle room. Recall the obviously coincidental Redskin Rule which had 100% accuracy until it didn't
Totally agree. Fudging the time window is a big giveaway. Historically, the average business cycle has lasted five and a half years. If you consider a recession unlikely in the first half of the cycle, you can simply 'predict' that a recession will hit sometime in the last two years of the business cycle and be correct most of the time. And of course, with only seven samples it's easy enough to be correct through sheer luck so long as you are even remotely in the ballpark.

The same is true for almost all political data. In this election year, you'll hear a lot of crap like, 'If he wins, this will only be the first time in 50 years that a person with X has won an election in that state while not carrying the capital' or some such rot. Then you look at the stats and discover that in the time mentioned there were only three or four elections where that could possibly have mattered, so it's essentially just random.