E_utopia :
there is no problem with using multivariable statistics, as long as you can reasonably bound the variables
If you are using a multivariable analysis you don't need to bound the variables, that is the methods whole purpose. The only time you need to bound them is if you want to reduce the problem to a one dimensional one and need to make sure you are not overlooking an effect larger than your significance level.
I should clarify this as I am using bound in two different ways in the same thread. I you want to ignore a variation you must be able to determine that the amount that it will effect your result is less than your significance level. This is usually done just as an estimate as you are interested only in order of magnitude. Thus you bound the variable as being only to have between a small range of effect which is not significant.
If you are actually including the variable in the model you do not need to bound it in this manner and it can have any size of an effect. However you must then be able to determine its value to lower than the precision level that you want to be able to quote. In effect this is also bounding it but in a different manner.
Concering the force used and the angle of the cut and the grade of the material. As I have noted before I don't compare single slices/cuts. If I do 1000 cuts on cardboard with one blade and 1000 on another, even if the variance in force, angle and abrasion of the cardboard are really high (say 50%) each, then the mean scatter in the resulting wear produced will be really low, less than 5%.
This of course is also not just theory I have done verification work many times to make sure that the single run variation is not that high that would make the mean scatter larger than the significance level - repeated pieces of work with the same blade to verify that the results are consistent from trial to trial and also with duplicates of the same model at the same time.
[high stress testing]
you cannot extrapolate from that wear to wear resulting from lesser stress.
Fatigue is basically described by two aspects, the critical level needed to induce it and the number of cycles needed for failure. If a blade responds very well to high stress impact work then its critical level will be very high meaning that it will handle high cycle low stress work much better than a blade with a much lower critical value. This is not a law set in stone but a generalization which seems to hold for most blade steels I have used - otherwise I would not be using it obviously.
Again, this is not just theory, I have done high cycles of low impact (chopping wood) and low cycles of high impact (chopping bone / steel), the results are consistent with most blade steels. Of course you have to be fairly particular about interpreting the results as compression resistance, impact toughness and strength will all govern the behavior of high stress work differently than low stress work but you can draw meaning conclusions from them.
And of course as I have said before, this is not an all inclusive way to judge a blade. It is just a shortcut used to get an initial impression as well as how the blade will perform in accidently high stress cases, which is enough to want to do it anyway.
if you hit a piece of steel lightly with a hammer, how ever many times you want, you will not produce the same damage as giving it one good whack with a sledge.
You can generate gross failure due to high cycle fatigue, which I have done on blades due to impacts in exactly the manner you describe (hundreds of low impact hits), all that is important is that you are above the critical level for fatigue to happen.
As for the makers that I work with, it should be pretty obvious who they are. If you really want to know just email the people I have worked with and those that I am currently working on blades with.
The most amusing part about the whole "ITS NOT SCIENTIFIC THERE IS TOO MUCH VARIATION!!" argument is that as anyone who has actually discussed the results with me knows I am readily aware of several large variances that can skew the results (not the random variance discussed here as I exaplained above they are not significant). They are often commented on in detail in the reviews. For example when I compared the Boye blades to the Deerhunter to a Henckels paring knife on cutting help rope, I noted that the Boye blades had a significant geometrical advantage (they are at an angle to the handle) whereas the Henckels blade is at a disadvantage as the edge runs up towards the point so these would skew the cutting performance away from a purely steel comparision.
There are lots of other areas as well. For example I often use fabric to guage cutting performance. The problem is that a blade that cuts one fabric well many do badly on another so it is difficult to compare to past work. So I have been trying to narrow it down to a few stock materials instead of using just whatever I happen to find. The end goal is to be able to index blades so that when I want to compare a blade to one I have already used I don't have to go back and repeat the work with that one.
All kinds of variations can happen which can make even simple questions difficult "which blade chops better". Well on what kind of wood? What kind of technique are you using. Again this is something that is commented on in the reviews and how the performance can change between blades as you use different methods and cut different materials.
The reviews are a work in progress, they don't look much like they did a year ago, and I doubt that in a year they will look much like they do now as I am working on a serious change in how I will index blade performance (it will be a multivariable model for cutting performance). A lot of people have helped significantly in this regard makers and users by suggesting different ways to look at performance, methods I have overlooked, pointing out things that should be improved, things that need clarification etc. . There are also people that just complain to complain without basis who are not consistent in their viewpoint (as Gator pointed out for example). It is hardly something new, you will find people like that everywhere. They constantly get loud and personal as it is the only way to attract attention to what they say.
-Cliff
[This message has been edited by Cliff Stamp (edited 08-04-2000).]
