Hacker News Reader

Comment by kqr | Hacker News Reader

This is what statistical process control charts were invented to tell us: https://static.loop54.com/XmR.html?baseline=7.5,7.9,8.5,8.3,...

Nope, last number is well within the limits set by the variation of the previous numbers. Depending on how sensitive you feel like being to false alarms, I wouldn't even say there's been a clear trend this year.

If you detrend it by successive differences we see that the slope is very close to zero compared to variation: https://static.loop54.com/XmR.html?baseline=0.4,0.6,-0.2,0.3...

However, the last number is close to a control limit, so if this continues next month we may have a signal!

View on HN

I'm not familiar with this, does this method require that each value is measured independently of the others? Because in that case, you'd want to use monthly inflation rates, rather than monthly estimates of inflation relative to 12 months prior, right?

Ideally, yes. I read too quickly and assumed that these were annualised monthly measurements.

However -- the technique might work anyway: the signal would be smaller, but so should the variation. Yes, one decreases as the square root and the other does not, but over just 12 months that might not make a big difference.

Only backtesting or a more thorough theoretical investigation can tell!

Edit: actually it doesn't have to be very thorough at all, it's a factor of 3. Probably important!

Deming statistics are a cargo cult. Try running some (Monte Carlo style) simulations; I did these many moons ago and they don't work at all.

I assure you Deming knew what he was doing, and my statistics knowledge is not limited to cargo culting the more popular SPC tools. However, they are useful for teaching the general public about accounting for the internal variation of data from the same population.

That said, as I've admitted in a more constructive parent comment, this tool was misapplied here due to my misreading of the GGP.

Reply on news.ycombinator.com