Re: [Harp-L] RE: THE COMB DEBATE REVISITED: SPAH Test

[Michelle LeFree <mlefree@xxxxxxxxxxxxxxxxxxxxxx>]

Vern writes:
> Here is a brief description of the planned test protocol:
>    
Hey, Vern old Harpbuddy!
> The purpose is to see if players (not listeners in the audience) can attribute distinctive sounds to comb materials.
>    
Hmm. I thought there were going to be three listeners on the "other side 
of the curtain" and spectal analysis as well...

> One set of covers and reedplates will be used.  They have been slightly modified to facilitate rapid comb changes.
>
> The players will be blindfolded.
> The harp has an attached weight to mask the weight differences of brass, wood, and plastic combs.
> There will be an aromatic material inside the lower cover (e.g. Mentholatum) to mask odor differences in wooden combs.
> There will be a coating on the front of the comb that the player can touch with the tongue to mask taste and texture differences.
>
> Combs of several different materials will be placed in the harp in random order.
>
> All of the players will play each comb and rate them against a set of adjectives selected by Brendan. e.g. warm, bright, loud, etc.
> We have a plentiful supply of alcohol swabs for sanitizing the harp between players.
>    
Will all the players play the same notes/tunes? Same positions? 
Overbends allowed? Hand effects OK or played on a rack?
> There will be a total number of ratings for (number of players) x (number of combs) x (number of adjectives)
>    
No particular reason to arbitrarily limit the number of observations to 
1 per player or 1 per comb. That's where statistical analysis comes in. 
You can dramatically increase the accuracy of the results of your 
expersiment by increasing the number of observations. Here's a neat 
little web site that I found that is basically statistics by "Rules of 
Thumb."

http://www.internetraining.com/Statkit/StatKit.htm

As a Mechanical Engineer, I'm sure you are quite comfortable with rules 
of thumb. As stated under the "Sample Size" heading on that web page, a 
good rule of thumb is to use a "sample size of 30 for each relationship 
you measure." You can easily see how the number of samples grows very 
quickly with each relationship tested and the benefit of limiting the 
number of variables in your experiment.
> This is not necessarily the best design for rigorous statistical analysis, but it should be fun and entertaining.
>    
Nothing wrong with fun and entertaining, so long as everyone understands 
going in that the fact that the results are subjective puts some pretty 
severe restraints on the conclusions you could make.

methods.My point is that once you've assembled all the elements of your 
experiment (harmonica harware, players, listeners, spectographic 
equiment and so on, it seems worth it to go to a bit of extra time and 
effort to improve your ability to make conclusions based on accepted, 
objective, scientific You guys are so close to constructing a really 
bullet-proof way to put the many questions surrounding the "Great Comb 
Debate" to rest at long last. My question is: "How many times do you 
want to perform this basic experiment before you decide to answer its 
questions definitively?" IMHO, your time, the resources you will use to 
perform experiment and the person-hours involved (including the time 
you've already spent) are too valuable to ~not~ take the next step. If 
not now, when?
> If the adjective ratings are evenly distributed among the comb materials, this will indicate the players' inability to distinguish one comb material from another by the sound.
> If the ratings have a high correlation between adjectives and materials, this will indicate that the ability to distinguish among them may exist.
> For example, if the brass comb has a high total score for "bright" and a low score for "warm" and the pear wood comb has the opposite, that would signify that the players were able to distinguish between brass and wood.
>    
You may well find that simple statistical coefficients of correlation 
from "Paired-T" tests are not capable of removing bias in experiments 
like you describe and that more formal analysis such as Anaysis of 
Variance (ANOVA) may better characterize differences between the 
populations in your groups. Wouldn't it be nice to know that your data 
set holds statistically significant number of samples?

Another question -- what about the spectral analysis? You didn't really 
mention it in your above description, but I think the spectral analysis 
might be the most revealing and inherently quantitative of your data 
sets. There are many potential ways to extract quantitative information 
from the spectra. You could define ranges of frequencies that would 
represent the different adjectives, for example, "warm" frequencies 
would be in the lower requency range; "bright" frequencies would likely 
be found in the upper frequency ranges. The relative loudness parameter 
could be desribed by the amplitude of peak frequencies or the integral 
of the spectrum. You could look for overtone frequencies that might be 
present in one material and not another. Or one player's resonance and 
not another. Very fertile (and objective) ground, seems to me.

Vern (and Brendan), I hasten too avoid leaving the impression that I 
want to diminish in any way the experimental elements you have obviously 
worked hard to define. Quite the contrary, it's precisely because I 
think the long-term value of your experiment would be signifiantly 
enhanced with just a tad more scientific and statistical rigor that I 
open my big mouth in the first place.

Please know also that I want to avoid even the remote possibility of 
leaving the impression that I would seek to "horn-in" or take any credit 
at all for your guy's "baby." However, though I am certainly no "real" 
statistician, I have studied mathematical statistics and probability 
theory and have 35+ years of high-level medical research under my belt. 
If you could use another hand on deck, I will be going to SPAH this year 
and volunteer for any way I might be of assistance. Just leaving the 
door open...

Once again, this is all the way ~I~ see it. And the both of you seem to 
have navigated through life thus far OK without any help from me. ;-)

Vern, in any case I look very much forward to meeting you in person and 
shaking your hand after all these years! Seems like I probably owe you a 
beer from somewhere along the line, too. :-)

Michelle

PS: Oh, Dr. Antaki...