Oh I'm going to have fun
fixitsan wrote:
I've seen the sort of variation we have now in the low accident rate. One year up a little, the next down a little.
Wrong! It’s actually: one year down a lot, the next year down a lot, the next year down a lot (this goes on for a few years, then) the next year down less, the next year down even less, the next year down hardly any, the next year no change.
fixitsan wrote:
Where this years change in accidents might be that there are slightly more, and last year there were slightly less it is easy to say there has been a large difference, but go back and count the actual number of deaths or the actual numbner of arguments and you see that there is no large difference. This is the noise I am talking asbout and the reason it wasn't seen befoe was that the signal was above the noise floor. The noise was there, but the signal was louder. I believe we have regressed to a point very close to the mean
You can't make such a claim without having a quantitive idea of what the mean
and noise floor actually is, so please do tell (and demonstrate how you came to those figures), otherwise the reader will have no choice but to conclude you are making up your arguments to suit.
What’s the significant difference between the noise floor and the fatality rate, given that the former is the latter? What is the cause of the noise floor in this case? (noise does not appear out of nowhere).
You really believe random noise is really going to be at all significant compared to the signal when averaged out at that kind of group size?
Let's put this into context: given you think there is an impassable ‘noise floor’ in this case, do you believe that beyond a point there's absolutely nothing we can do to reduce fatalities? Given that the fall has indeed levelled off (please check the 2006 figure and add it to your data), would you agree that imposing even more speed limit drops or additional cameras would be useless? What about better post crash care, better in-car protection, better education, continued road improvements? Shall we just stop wasting money on all these other methods now?
fixitsan wrote:
smeggy wrote:
Regardless, why would this 'noise' only affect the back end of the curve and no-where else? This inherently makes it a pattern.
Like I say, because the signal was previously louder than the noise.
Are you saying the noise is now bigger then the signal? What on earth do you base that on? Why is the transition now?
Can you explain why 'your' noise manifests itself as a steady level, why isn't there significant level fluctuating much between years at the disputed tail end of the curve? Why are the figures actually steady? Is this not the essence of a trend?
Paul, you really should update your graphs to include 2006.
Fixitsan, please factor the 2006 figure into your calculations.