2nd September 2013, 09:06 PM
(This post was last modified: 2nd September 2013, 11:08 PM by Doug.)
Jack Wrote:Doug, I don't believe you. I may be misunderstanding, but if you could predict where archaeological 'sites' were or aren't with such a high accuracy and precision none of us would have a job and you'd be a multimillionaire.
Besides, what do you define as a 'site?'. We dig landscapes.
Yeah, I am guessing you are misunderstanding me. Accuracy and precision are tricky. As precision goes up accuracy goes down and other way around. Lets say you label everything in your area, lets say Scotland, as containing a site (yes, agree about landscapes but that is a whole other discussion). Well you are 100% accurate. You guessed where all the sites are- everywhere. Not very useful for most.
Now lets say you label only 10% of Scotland as 'likely' to contain sites, the precision part of the equation. It turns out that in that 10% you capture 70% of all sites. That means 30% of sites are in the so called "not likely" areas which might get bulldozed.
This is all hypothetical that will vary from area to area and definition to definition of 'site'. It will vary by how one cuts up the area you use. Some of the best predictive models will capture 85% of known sites in less than 20% of landscape in high probability area. That still means that you will find 15% of sites in the other 80% of the landscape. These are also called red flag sites or gross error.
Now it all depends what you want to do with this information. You could alter you model so that lets say 30% of landscape is labeled high potential and you capture 98% of all sites. That is still 2% of sites that would get destroyed if you only investigate that 30% of the landscape.
Is 2% loss ok? Is 5%? Is 10%? What is acceptable is really a tight rope to be walking. Now most predictive modellers have come to the conclusion that any loss is unacceptable. Also, many managers and to be fair most laws (outside UK included) agree with this assessment.
So when I say high accuracy and precision the definition of each is relative and highly highly subjective. This is why most modellers use PM as a method to estimate potential costs (e.g. there is a very low likelyhood of finding something there so build there) , NOT to decided where to dig.
Hope my condensed version of 40 years of pm was understandable.
So yes I can predict 95% of sites on 10% of the landscape but monetizing that is another problem. I can predict but the local gov. archaeologists can overide me (and should for those other 5% of sites), so what is the point.
However, that really does not matter. Lets say I predict where lots of sites are going to be were you develop your housing estate. What the fuck do you care? Archaeology is what .02% of your budget. You bought the land already. It would cost you 10x? 20x? 50x? as much money to buy a new plot of land and get all the permissions. Your going to build no matter what. You're also going to try as hard as you can to get out of spending that .02% on archaeology or be pissed because you have to spend it.
Now if it was a government road and you can take whatever land you want, yes predictive modelling works well because you can move your project, private development not as easy to move. Also, it helps if your afraid of idelling machines on site but some half decent planning can take care of that.
So yes- I can predict sites but it is not going to make me a millionaire or put people out of jobs.
If you don't believe me, how many people can point to the majority of their desk based assessments causing stuff to not be built. The answer will be zero. As much as people bitch archaeology rarely (it still happens) stops things from being built.