| Author | Message | ||
     Tony Davies (td) Moderator Username: td Post Number: 180 Registered: 1-2001 |
Hello Rahimi, This is an interesting topic but what we really need to know is: Why did you ask the question? Is this general interest or do you have a specific problem you are trying to solve? Fred McClure pioneered the use of Fourier transformation for NIR data compression in the 1980s. I was one of a very few people who were interested and I did did quite a bit of research with Fred and we demonstrated many uses. More recently I worked with Tom Fearn on the use of wavelets. We showed that for the majority of our test spectra, wavelets required fewer coefficients then FT for the same level of fit but for others FT was better. The differences were not great and we concluded that purely for data compression the added complications of wavelets was not really worth the effort. However, we do believe then when it comes to utilising coefficients then there may be a worthwhile advantage for wavelets. There is lots more to say if this is of interest to you. Best wishes, Tony | ||
| Howard Mark (hlmark) Senior Member Username: hlmark Post Number: 202 Registered: 9-2001 |
Rahami - I know of one "Table Curve", which is a software program to fit functions to data. In terms of spectroscopy, for example, it might be used to take a single spectrum and try to fit a polynomial to that spectrum, as, for example, a Taylor series might do. Like many other mathematical approaches, the goal would be to create a descriptor of the spectrum in terms of a relatively small number of other, presumably simpler, functions that can approximate the target spectrum. Examples of this general approach that we are all more-or-less familiar with are: 1) Taylor series, that uses polynomials 2) Fourier series, that uses sines & cosines The goals of these methods of approximating the spectrum under consideration is to balance off the number of functions used, with the degree of approximation, i.e., how closely the approximating functions fit the data spectrum. The program Table Curve (or maybe it's Tablecurve) has a very large repertoire of functions that it can use to make the approximation to the data set available. It can include not only polynomials and trigonometric functions, but also exponentials, logarithmic, and what we might consider "advanced" functions: Bessel functions, Hankel functions, etc., also combinations of all the above. The program has the capability of allowing the user to select or prevent the use of various classes of trial functions. After trying many functions, the program will tell you which one(s) fit the data "best", and what the parameters (coefficients) of the best-fitting function are. In a sense it could be used instead of PLS (or PCR) by using the values of the parameters as input to the calibration process instead of PLS/PCA scores. But only someone who is thoroughly familiar with both PLS, the concepts underlying the process of fitting functions to data, and the data set at hand should try it. It is very easy to do the equivalent of overfitting the approximation to the data at hand, much more easily and more flagrantly than with PLS. I've seen cases where a fairly simple function fit a data set fairly well over the range of the data, then the fitting function immediately went vertically off-scale just beyond the last data point. This was an obvious case of overfitting, but to fit an entire data set of spectra that way would be extremely risky. \o/ /_\ | ||
| rahimi (aarahimi) New member Username: aarahimi Post Number: 1 Registered: 2-2008 |
hi every body does any body know anything about "table curve" software using in analysis of NIR spectrum? could it be used instead of PLS? thanks | ||
| Howard Mark (hlmark) Senior Member Username: hlmark Post Number: 86 Registered: 9-2001 |
Ken - don't worry about it. I knew what you meant, I just figured I'd pull your leg a little over the funny wording. \o/ /_\ | ||
| Kenneth Gallaher (ken_g) Junior Member Username: ken_g Post Number: 9 Registered: 7-2006 |
Howard, Sorry if it was misinterpreted . NIR is not the problem, octane is... | ||
| Howard Mark (hlmark) Senior Member Username: hlmark Post Number: 85 Registered: 9-2001 |
Ken - you said "Actually the problem with NIR vs. octane is that �octane� is a stupid parameter." I certainly don't know enough about the petroleum industry to know whether octane is "stupid" or not, but if it is, then for heaven's sake, don't put the blame on NIR ("...the problem with NIR...")! \o/ /_\ | ||
| Kenneth Gallaher (ken_g) Junior Member Username: ken_g Post Number: 8 Registered: 7-2006 |
A lot of good discussions here which I will address in no particular order. In the case of fuels there are unfortunately desired analytes which NIR cannot detect. Two that come to mind are sulfur at low ppm levels (EPA regulated) and heavy metals which are catalyst poisons at ppm levels. You may have seen claims that NIR can do one of these � they are fraudulent. I believe that ALL chemometric based methods should include a calibration validation step with each analysis � ASTM writers take note. I continue to see NIR �Octane analyzers� at Pittcon that will gladly give you an octane for Pepsi. I like to think I am not an �instrument chauvinist� and agree with Martens� observation there. Actually there is a reason for such people � you want an expert who will do anything to make �their� method work�You just don�t want them to be the technical manager who decides the approach to start with � that person needs to be a generalist with knowledge of many methods and allegiance to none. I like to think I am the second type. Re �octane gloating� - I�m not, it�s just a successful NIR example on a difficult application. Quit frankly most Pharma applications are trivial by comparison � except for the sampling part�. Actually the problem with NIR vs. octane is that �octane� is a stupid parameter. One is supposed to degrade a lot of high quality NIR-based chemical information into two parameters RON and MON which are claimed to represent engine performance in any engine from Yugo to Porsche. The raw chemical compositional information would be lead to a far better estimate of engine performance. Regrettably octane is one of those historical measures that will probably never die. As far as the �giggles� test of putting impurities in tablets and not telling me � my validation methods would find those and not predict � assuming yes that they have a NIR spectrum. If you are spiking them with Polonium I will get tricked. One of the barriers to NIR in petroleum initially was this idea of ASTM regulated method. What the oil people figured out was that for routine (every minute or so) analysis NIR beat everything else. If regulation was an issue you did post testing on the final tank. Many might not admit it but NIR now has such credibility that the tank tests are not done and in fact if the octane engine and NIR disagree the engine will be rerun � you know you will get same NIR answer. I see the same situation now in Pharma � a fear to even try NIR because of certification issues. Try it � you will learn a lot - certification or not. And you will learn in due time what you can trust and what not. | ||
| Howard Mark (hlmark) Senior Member Username: hlmark Post Number: 84 Registered: 9-2001 |
Now I'd like to discuss some of Chuck's technical points, especially the one about an undetectable impurity. I don't think that in practice, that can happen, mainly because virtually everything has some absorbance in the NIR. Even something that has no absorbance (of which there are very few) will affect the optical properties of the sample because of its scattering properties and thereby give notice of its presence. The main reason something could be undetectable is if it is present in too small quantity (i.e., -> 0) to be detected. But then you have to ask the question of whether its presence matters; that will depend on whether it affects the product or the final usage of the product. For NIR that is generally a non-issue since NIR is normally used for measuring the major constitutents in materials, and leaves the minor ones to other analytical techniques. \o/ /_\ | ||
| Howard Mark (hlmark) Senior Member Username: hlmark Post Number: 83 Registered: 9-2001 |
Chuck makes some very good points about validation, so I'm going to use this forum to announce that ASTM has a committee (E13.11 on Chemometrics) developing recommendations, which will eventually become an ASTM standard practice, for validation of chemometric methods, and welcomes all recommendations as to what should be included in a validation exercise. I'll also take the opportunity to mention that ASTM practice E1655 already contains some recommendations, which I do so that redundant recommendations are unnecessary. The committee is hoping to be able to send the draft practice out for it's first vote by EAS, so this is an ideal time to get your thoughts in. Obviously, nobody can vote on a proposal that isn't in the draft. \o/ /_\ | ||
| Charles E. Miller (millerce) New member Username: millerce Post Number: 1 Registered: 10-2006 |
This is a very useful and interesting discussion! Don: With regard to the "tablet challenge" scenario you posed, it is perfectly acceptable (although not necessarily desireable) for a method to "refuse" to provide an answer at all. An advantage of multivariate methods such as NIR vs. univariate ones is the potential ability to detect abnormalities and "intercept" the method before providing a misleading result. This really ought to be done for all NIR deployments. This, however, leads me to a more "sinister" scenario: suppose that the presence of the impurities had NO effect on the NIR spectrum of the tablets, AND their presence greatly affected the quality of the product for it's intended purpose? Even multivariate abnormality detection based on NIR alone would fail to detect the impurity, and would allow the method to report a value that is likely to be very inaccurate. Of course, all analytical methods could, in principle, be defeated in a similar manner. My opinion- I think that the burden of NIR implementers really ought to be to ensure that their methods are validated with a rigor that is commensurate with the severity of the consequences of it's failure. As a "community", I agree that we could help define what is necessary and sufficient for different levels of risk. Educating regulators, and academics who influence them, on the virtues of NIR and chemometrics might be useful, but I don't think it should be the primary goal. I want to note that, for "high severity" cases, "sufficient" validation would likely require both testing of a method's performance for samples within it's intended range of performance, as well outside of this range. Such "outside" testing of a method is certainly an open-ended proposition- but it just requires some creativity to consider the ways in which someone/something might try to defeat it. Also, as an Analytical Chemist, I agree that the field often suffers from "instrument chauvinism" (I think Harald Martens came up with this one). This is rather unfortunate, because one possible way to guard against such method defeat is to augment/"fuse"/combine NIR with other analytical techniques. | ||
| Donald J Dahm (djdahm) Junior Member Username: djdahm Post Number: 8 Registered: 2-2007 |
A person who apparently wishes not to join in the discussion directly wrote me to point out that in some cases the "octane" situation and pharma applications are not so different. "There is no material called 'dissolution time' or 'hardness' or 'flowability' in pharmaceutical materials, either, but we correlate on them, don't we? The actual assay is easily done by HPLC; it's the parameters for which there are no comparisons where NIR (and Raman, etc.) shine." As I was saying, the QA procedures will probably need to be tailored to the analysis. [email protected] | ||
| Donald J Dahm (djdahm) Junior Member Username: djdahm Post Number: 7 Registered: 2-2007 |
Ken G: So you discussed this before. And now your're gloating becasue NIR turned out to be the best game in town. All of us can point to successes: times when we were able to show that NIR did the job required better (including points for speed) than anything else going. Further, it isn't like we are not skeptics ourselves. We have had to accumulate evidence, fuzzy as it might be, that we could trust what we were doing. To me, the issue is: When are we going to figure out a way to define just how "good" our analyses are. I'm sure that many of us have given this some thought, and maybe somebody can lay out a proper Quality Assurance protocol for NIR Spectroscopy for the general case. [We certainly can give advice on things that a proper analysis should embody. For example, make sure all the variation you will encounter when using the method is included in the calibration set.] However, I'm guessing it will have to be a QA protocol that changes for the analysis at hand. I think comparing the two cases, octane and Pharma, too directly is a mistake. There is no physical property of "octane". As this poor chemistry professor understands it, the octane rating is a correlation of the auto-ignition properties of a "random" mixture of hydrocarbons to that of a controlled mixture of 2,2,4-trimethylpentane and n-heptane. Of course, "random" here must have limits to be useful. It must embody conditions that are found in refined petroleum. Thus there are limits, whether defined or not, to the composition of mixtures that can be tested in the "standard" engine and give meaningful "octane" results. Further, the "random" mixture that keeps my automobile engine from "knocking" depends on things besides the compression ratio of my engine. [The main one that we are familiar about through real life is "load" (like driving at slow speed up a hill.)] All that NIR can claim here is that the rating it produces keeps everybody out of trouble just as effectively as the standard engine measurements, and is far more cost effective. While I am genuinely impressed by that accomplishment, that is not the case we have in pharmaceuticals. Suppose some kid in the FDA (or the equivalent in another country) presented the folks who want to use NIR in their submissions a challenge. They say: 1) We will produce a set of standard blends of pharmaceutical ingredients from the pure materials you provide us and make them into tablets. 3) You will analyze the tablets and turn in the results for evaluation. 4) And just for giggles some of the tablets will have impurities which we will add and not disclose to you. You show that you have the claimed accuracy and we will accept your method, as long as it has a proper QA scheme in place. What would the response be? (I can assure you that the EPA has done things similar to that. Perhaps the FDA already has.) [email protected] | ||
| Howard Mark (hlmark) Senior Member Username: hlmark Post Number: 82 Registered: 9-2001 |
Bruce - well, we know that NIR has been used for food analysis almost from the beginning. The General Foods', Pillsburys', Kelloggs', etc, of the world were customers of Technicon from as long back as I can remember. Exactly what they used them for, I don't know; it may well be that they felt them reliable enough and used them for routine analysis and QC, but used the compendial method for those analyses that were done to meet the regulatory requirements. I don't know. As for the standards for HPLC, GC, mass spec, etc - there are probably very few; most of the standards are not used to standardize the GC, or whatever, to some absolute standard. They are not instrumental standards, they are standards for measuring particular analytes. For example, if NIST provides a standard piece of steel, they will certify the amount of (say) chromium in it. Then if you analyze it you should get the same value for the chromium content, regardless of the method you use; that standardizes your analysis, not your instrument. Other samples are then analyzed by comparison with respect to the standard. In NIR we can't do that because we know a priori about the sensitivity of the measurement to the details of the sample condition, preparation, presentation, etc. - what the classical chemists call "matrix effects". For the classical chemists those are (usually) minor annoyances, for us they're major problems. Academics, of course, are in a separate class. The analysis is not their product. Their product is a person who knows how to titrate, weigh, pipette, etc, - in other words to teach the skills of an analytical chemist and also when, where and how to apply them. Granted that very few of them will teach chemometrics. That is a legitimate skill, but not taught partly because they don't know it themselves, and partly because, much as we are enamored of it, we have to remember that NIR is still only one of dozens of analytical techniques available, and not suited for every analysis (how would you measure the chromium in steel with NIR, for example?), and the academics have to teach them all. So we can't expect them to have the same reactions to NIR as an industrial analytical chemist. The idea of using "equivalent" methods is catching on with the FDA, I'd say, with the new PAT initiative. I think the main difference is that the FDA has much more stringent requirements for showing "equivalency". \o/ /_\ | ||
| Bruce H. Campbell (campclan) Moderator Username: campclan Post Number: 95 Registered: 4-2001 |
Howard et al., (Between the time I wrote this and posting it I found a response that echoes my first paragraph, but rather than change it, I will leave it as is.) The largest regulated industry that I know of uses NIR. This is the petroleum one. Many, if not all, state regulatory agencies in the USA checking octane number use NIR and so do the refineries. They do this because the precision of NIR is the best of the various techniques used routinely. It would be of interest to find out if the FDA also uses and allows NIR for foods (since the F in FDA is food). With respect to standards, I am familiar with many of them and the requirements they must meet with respect to purity, etc. How many are there for say, HPLC, GC, mass spec, etc.? And when can a company, say a pharmaceutical one, use a primary standard for calibration of their HPLC when used for a product? So in my mind, the use of standards is necessary but many of them are not primary standards; they are often said to be traceable to primary standards but sometimes the trace ability is tenuous. Thus my contention that NIR is weak because it is a secondary standard method is more of hot air than of substance. Whenever a user of NIR is confronted with this terminology the argument could be of such as I outlined above. Another place classical analysts are in the main are in academia. These teachers have their favorite methodology, which usually are the ones I outlined in my earlier missive; i.e., ones that require the sample to be pretreated in some fashion to obtain a response in the final step. Thus, these professors don�t teach a method, such as chemometrics that is contrary to their own training (although this neglect of chemometrics is declining, but slowly). I mentioned many if not all compendia have a section allowing the use of an equivalent method. One reason that I know of as to why this is done is that for some years I was a member of the committee that wrote, and revised, the Food Chemical Codex. Many times I heard committee members, when examining a new chemical and its analysis, express the need to make the analytical method as simple as possible to allow any supplier to be able to equip and staff a small laboratory without excessive outlay for equipment, etc. This has stifled the use of better and more economical analytical methods. The section of equivalence was to allow any supplier the ability to use a different, equivalent method that was better without having to go through the process of obtaining committee approval. I think this same approach in compendia can and do stifle the use of better methods, such as NIR. It is this stifling that I think compendia need to relax by being more definitive on what equivalence really is. These thoughts are helping to define the problem that Don stated. What we really need are solutions. I think having compendia take the actions I described helps. There must be many more solutions � does anyone else not yet �spoken� have ones they would like to proffer? | ||
| Kenneth Gallaher (ken_g) Junior Member Username: ken_g Post Number: 7 Registered: 7-2006 |
There was a lot of talk on this subject when NIR started into Octane and related fuel properties. Octane is regulated by state agencies and by sales contracts. In fact what happened is that users quickly found out that by comparison with the "primary method" - an octane engine for example - that NIR was far more reproducible, and cheaper, and if done properly more accurate as well. Given that they might keep an engine in the back room just to be sure but NIR has become the primary method. I expect the same thing will happen/is happening in pharma both for the reasons above and because the PAT initiavtive is far more open about alternative methods if they can be shown to do the job. | ||
| Howard Mark (hlmark) Senior Member Username: hlmark Post Number: 81 Registered: 9-2001 |
Bruce - a good deal of what you say is valid - - - historically, for users in non-regulated industries. But if NIR is to make inroads into those industries that are regulated, then it has to meet the requirements of the regulatory agencies. The analysts in those industries are, for the most part, "classical" analytical chemists, and more to the point, so are the analytical chemists at the regulatory agencies. Those are the ones whose acceptance is needed, in order for the industry to avail itself of the benefits of the technology. As for your other comments, I tend to agree with them, with some modifications. Yes, strictly speaking it's true that the only analytical methods that can be related to fundamental primary standards are those that are completely gravimetric, mass being one of the fundamental physical properties. But accepted "primary" standards exist, they are just not "fundamental" properties. NIST, for example, provides hundreds of materials with known and defined properties, carefully measured. Those are accepted, both practically and legally, as "primary standards". If impurities are present, they are specified and quantified. I don't think you can fault them unless you want all industry to grind to a halt. For those, "secondary" standards and practical usage of them are based on application of known and well-tested physical and chemical laws (law of constant proportions, etc.). So justifying NIR by denigrating other methods of analysis is not a useful or profitable approach (no matter how badly they might sometimes by applied in particular cases). NIR certainly has it's merits, as we all know (or we wouldn't be here!). Nevertheless, it has somewhat of an uphill battle to fight in order to get acceptance by those chemists who have to be convinced that the statistical methods as we use them are, in fact, as good as the "classical" methods. Don't forget that in some cases, they are required by law to do that. \o/ /_\ | ||
| Bruce H. Campbell (campclan) Moderator Username: campclan Post Number: 94 Registered: 4-2001 |
It appears to me that this exchange is on the topic of acceptance of NIR. In that case, I have to ask myself acceptance by whom? Those responsible for the usage of NIR, such as businesses when usage of NIR yields a profit, are accepting NIR. In my mind, the acceptance being discussed here is acceptance by other analytical chemists. When one looks at the backgrounds or educational training of NIR users, few are what I would call classical analytical chemists. A factor to consider in acceptance of NIR is the usage of the term �secondary method.� Since, classically, analytical methods involve steps to isolate the material or item of interest as intermediate steps, pure and stable standards can be used for calibration. Isolation of the material of interest is not done in NIR. Further, our experiences are such that we don�t think it can be done in the near future, if at all for most if not all cases. Thus users of NIR are in opposition to the education system that teaches that isolation, or previous steps, is the only way to perform analyses. Still another aspect is NIR is dependent upon standards whose purity is from �classical� methods of analyses. Thus NIR is called a secondary method of analyses. However, if we step back and examine almost all analytical methods of analyses, they also are essentially secondary methods. Almost all other methods depend on standards whose purity is often questionable, whose stability is not determined, and whose homogeneity has not been measured. At best, these other methods of analyses use standards that are often labeled as traceable to some governmental standard. Therefore standards in NIR can also be labeled as traceable. There is yet another difference between NIR and classical methods is that one uses a large number of standards to calibrate in NIR, many more than any other analytical technique that I am familiar with, except chemometric ones. In my mind, this is a superior approach. Although round robins are used in compendial endeavors, I have seen their usage more for determining repeatability and reproducibility. If one or more laboratories have results outside the statistical acceptance level, theses labs are so notified and given a chance to make corrections. I am more concerned about another compendial move and that is to state that the methods being published are the ones to use for the particular analyses, a laboratory/analyst can use equivalent methods. This means NIR users must show equivalence. Scientifically, that is not necessarily impossible, as we all know. However, some businesses will not allow such because they feel that if the case ever goes to a lawsuit or governmental action, they will be in a weak position. I think that to mitigate this barrier to acceptance of chemometric NIR compendia should undertake a more definitive explanation of what equivalence means in terms of both accuracy and precision. Perhaps even publishing acceptance levels, such as agreement at the 95% level, would be beneficial. Anyway, these are some my thoughts about acceptance, or lack of, when considering NIR. | ||
| Howard Mark (hlmark) Senior Member Username: hlmark Post Number: 80 Registered: 9-2001 |
Neil - to be sure, there are uses for round-robin tests besides developing and validating analytical methods. When I was at Technicon, we were part of a round-robin testing program wherein we received a monthly set of milk samples. We measured them and sent the results in to the central office and they would send back a report of how we did relative to all the other participating laboratories. The purpose of this round-robin was to allow all the participating laboratories to judge their performance in relation to other laboratores, and also to serve as a periodic QC procedure. The requirements for that round-robin were not stringent. A laboratory could use any analytical method they chose (we used NIR, of course). But for the purpose of developing and validating a new method the requirements are much more stringent, and it is the ability of laboratories to meet those more stringent requirements that give the general analytical community confidence in the method, as you and Jim described. \o/ /_\ | ||
| Neil J. Dominy (neil) Junior Member Username: neil Post Number: 8 Registered: 7-2003 |
Hi, Speaking from the simplistic world of just running an operating network these are questions which are raised the whole time. In short one is normally able to show that the NIR is more repeatable than the reference method but "less" accurate. The acceptability of this is dependant on the customers requirements. With respect to Round Robn tests we submit NIR predictions on the AAFCO ring tests but this is also controlled by whether the sample is represented in the databse. Neil | ||
| James Malone (jmalone) New member Username: jmalone Post Number: 5 Registered: 8-2006 |
Don - I agree with you and Howard, this is great topic and one that I hope gets a lot of responses. One thing to keep in mind is that the reference techniques you and Howard are referring to are "absolute" methods (as much as anything can be) but spectroscopy is a RELATIVE technique. So, I see a fundamental difference already in how these data are and can be used. As we all know, NIR calibrations, regardless of the software are done using concentration values obtained from these "reference" methods. Those of us who have been working in the industrial arena know that "reference" values often suffer from much poorer precision than the NIR spectra. (An informative exercise is to split a homogeneous sample into 5 aliquots; label each one independently, and submit for analysis by a laboratory using a documented reference method. The variation in concentration values will be illuminating.) Howard has already detailed a round robin experience but I think that even if one were done with NIR, it would be inconclusive because of the difficulties mentioned above. Your initial point, however, remains valid and a thorough discussion of it by the "gurus" on this site should be interesting. Thanks for bring it up. Jim | ||
| Howard Mark (hlmark) Senior Member Username: hlmark Post Number: 78 Registered: 9-2001 |
Don - I hope this works; you're right: we need something like this. Might need more than one thread to do it, too. I hope it doesn't become just you and me sniping at each other! There's a related issue, too: has anybody thought about how reference methods got to BE reference methods? With known precisions and accuracies? And validated? For major analytical methods in widespread use (e.g., Kjeldahl analysis for "protein") the answer is that an organization in the field (such as AOAC (for general analysis) or AACC (for agricultural analysis) organizes and runs a round-robin study. There are whole books written on how to do that, but the essence is that an analytical method is run according to a recipe that is prescribed in every detail; every laboratory taking part uses exactly the same procedures (supposedly, anyway: in practice there are inevitably departures from the recipe). If any laboratory's results are too far afield, then that laboratory's procedures are audited and if the departure from the recipe is due to a failure of the instructions to make some point sufficiently clear, then the instructions are revised and the whole thing repeated. As you might imagine, this is extremely time-consuming, resource-intensive and EXPENSIVE. So it's normally not done unless there is evidence that suffiecient use will be made of the procedure to justify the effort. In the NIR world, round-robin tests are essentially never done. There may have been one done, in the mid-1970's, before NIR was commercially promoted. The USDA sponsored an NIR forage analysis project. What documents I have related to that are all up on the IDRC website's History Project tab. Maybe someone can tell us more about the details of what was done and what was found. Otherwise, I know of no case where a true round-robin study was done for an NIR method. The nature of NIR analysis, and to some extent the mystique of it rather precludes that. The need (or at least perceived need) for every instrument to be individually calibrated sabotages the first rule of round robins: that every laboratory runs exactly the same method; this would require the instruments to use exactly the same calibrations, i.e. "transferred" calibrations, something the technology is still struggling with. But that's a different topic. The important point that's on the topic is that every instrument in the study would have to use the same calibration; instruments of different models, different manufacture and even using different technologies. For the general analytical community to accept an NIR method on the same basis as other methods, that NIR method would have to be subjected to a round-robin study the same way the other methods are. That's not likely to happen. Certainly no single user's lab has the motivation or resources to do such a study, and the technology is sold on the basis of paradigms that conflict, to some extent, with the requirements for round-robins. \o/ /_\ | ||
| Donald J Dahm (djdahm) New member Username: djdahm Post Number: 5 Registered: 2-2007 |
I'll just take a bit of a philosophical bend here, and open a discussion of what I think is a somewhere between a looming disaster and a barrier to acceptance. As many of you know, I have been the author of the "Theoretically Speaking...." column in NIRnews. One of the problems with that forum is that it is not so easy to "talk back" as this one. Here I am just a jerk on the computer, so I assume I will get some argument. (You will notice the catch phrase "looming disaster" in the first sentence. That's supposed to make you wade through the crap hoping that it winds up somewhere interesting in the end.) What I will pledge is that if we collectively come up with anything worthwhile, I will write it up for NIRnews. However, since I no longer have my "official" platform, we will have to make sure it meets the Editor's quality standards. I'll start with just one discussion point. We are not trusted by the field of Analytical Chemistry because we do not adhere to the "Quality Control" system that has evolved in that field. In the portion of my life in which I masqueraded as an Analytical Chemist, I was indoctrinated in the use of the three " S's ": Standards, Splits, and Spikes. The tradition of NIR Spectroscopy that is emerging ignores two of them. The best we do is some variant of Splits: repeat samples. Are you impressed with the "Quality Control" data that accompanies the papers in our Journals? Why on earth would anyone trust our analytical results when they are not presented with an honest rendition of how the method performed on samples completely unknown to the person who developed the method, and separated from him/her in time and space? Why would not an objective person look askance at our practices of master/slave instruments, the avowed purpose of which is to fudge the data so that two instruments give the same result. I'm not saying what we are doing is terrible. I'm just saying that we don't give much cause for the objective outsider to believe we know what we are doing. And, of course, there is the possibility that we are fooling ourselves. Don ([email protected]) |