, a publication of Strategic Directions International, Inc. (SDI, Los Angeles, CA), a consulting and market intelligence firm specializing in this market, the total worldwide market for analytical and life science instrumentation and related aftermarket and service revenues slightly exceeded $26 billion in 2004, increasing revenues about 6.6% from levels in 2003. As with the situation in 2004, some uncertainty still persists. Fears of terrorism and the adverse affects of high oil prices delayed some business spending in 2004. However, the impact of these factors appears to be lessening. What could be challenging this year, especially for the instrument industry, is Europe's slower growth, the pharmaceutical industry's regulatory and pipeline problems, and the expected slowdown in growth of the Chinese market. New product introductions continue to stimulate sales, and while many instrument makers did well in 2004, some major companies stumbled. Given these countervailing factors, SDI predicts a slightly lower overall industry revenue increase in 2005 of about 6.2%.  

 Table I. Spectroscopy product growth prospects for 2005			

Last year was characterized by a number of important trends and developments that affected the instrumentation marketplace. Perhaps most significant was the realization of the importance of China as both a major market and a manufacturing base. China's growth was evident particularly in 2004. As Chinese factories pumped out products of every type for its growing domestic market and for export, especially to North America and Europe, instrument demand increased proportionately. Local manufacturing of instruments also grew significantly, mostly the result of foreign direct investments. The year also saw the return of heavy industry investment, especially for metals production, as well as for chemicals and petroleum refining. Only in Europe was the climate somewhat muted, but even there, US-based firms did reasonably well because of favorable exchange rates, putting European-based manufacturers at a disadvantage. 

SDI disaggregates the spectroscopy market into three distinct sectors: molecular spectroscopy, atomic spectroscopy, and mass spectrometry. The applications of the techniques included in each sector and their prospects differ significantly, even within each sector. Overall worldwide revenues in the spectroscopy area totaled about $5.7 billion in 2004, or about 22% of the total laboratory analytical and life science instrument market. The overall spectroscopy market is estimated to have grown about 5.9% in 2004, an improvement over the previous year's growth. SDI predicts that the market will expand slightly faster in 2005, growing at about 6.1%, which is a continuation of a three-year trend.  

Figure 1. Spectroscopy market by product segment.			

Molecular spectroscopy techniques account for about 45% of the spectroscopy market, although many diverse techniques are represented that address a wide range of applications in every industry category. Atomic spectroscopy is the second largest sector at almost 35%, while mass spectrometry represents 20% of spectroscopy revenues. Please note, however, that general purpose liquid chromatography-mass spectrometry (LC-MS) and gas chromatography-mass spectrometry (GC-MS) — both major mass spectrometry techniques — are not included here because they are considered to be part of the chromatography market. Revenues of these hyphenated techniques would increase the significance of the mass spectrometry figures by about 50% if the MS portion of those systems were included. Mass spectrometry would then approach the size of the atomic spectroscopy sector. 

The market for molecular spectroscopy is estimated at well over $2.5 billion for 2004. There was, however, considerable variance in growth by product technique, led by the hard charging nuclear magnetic resonance (NMR) segment. Overall, the total revenues for all categories grew 5.4% in 2004, a somewhat slower expansion than seen in 2003. No single technique dominates this market, but three segments — NMR, UV/Vis, and infrared (IR) — account for almost 75% of the overall molecular spectroscopy market.  

Figure 2. 2004 spectroscopy market by sector ($5.7 billion).						

The primary demand for molecular spectroscopy comes from the vibrant life science industries, including the pharmaceutical, biotechnology, and the agriculture, and food and beverage industries, which account for about 35% of the market. Organic chemicals, academia, and government sectors account for another 30% of revenues, much of which also are life science-oriented. But molecular spectroscopy techniques have an enormous range of applications, so they can be found in every other industry, some of which continue to be depressed, such as environmental testing and semiconductor manufacturing, which are experiencing lower demand. 

As in recent years, NMR spectrometry represents the most dynamic molecular spectroscopy technique, with growth for 2004 estimated at almost 10%. NMR is an important technique that provides information on the structural analysis of organic molecules, particularly proteins. Recent growth trends could be described as an "arms race" as vendors develop, and researchers demand, ever more powerful superconducting magnets to improve upon already very high analytical performance. The availability of newly designed probes and enhanced software also stimulates the large aftermarket segment. Electron paramagnetic resonance (EPR) instruments also are included in this category and are experiencing renewed growth because of their usefulness in analyzing organic compounds with free radicals. Based upon its performance in 2004, the NMR market has overtaken UV/Vis to become the largest segment of molecular spectroscopy. We expect NMR to do equally well in 2005 although delayed purchases from government labs may depress growth slightly.  

Figure 3. Estimated growth for spectroscopy sectors in 2005.			

UV/Vis spectrophotometers comprise the second largest portion of the molecular spectroscopy market, but also one of the slowest growing segments. Unit sales of visible-only and nonscanning instruments continue to decline, except for environmental testing where such instruments are tied to prepackaged reagents and kits. Scanning UV/Vis and diode instruments, which are faster and more powerful instruments with marginally higher prices, offer the best prospects and frequently replace older lower performing instruments as they are retired. 

The Raman spectroscopy market continues to experience strong demand, with 2005 growth expected to reach 10%. Dispersive Raman systems with microscope interfaces currently are receiving the most attention from scientists. Likewise, dispersive systems significantly exceed the demand for Fourier transform (FT)-Raman systems. Raman often is used to inspect for purity and composition of plastics, rubbers, and resins. Raman also is an important technique for the characterization of pharmaceutical excipients. Fiber-optic probes give dispersive Raman systems a degree of portability, which enhances at-line use. 

Infrared and near-infrared (NIR) instruments represent about 27% of the molecular spectroscopy market, and both are experiencing moderate growth in the 5-6% range. Pharmaceutical and biotech applications represent growth opportunities, but food analysis, especially milk and grains, forensics, plastic recycling, and cellular analysis are the most important application areas. Color measurement instrument demand improved in 2004 and should grow at around 5% in 2005. Major industrial users such as paints, textiles, and printing have rebounded, while prospects in Asia, especially China, are burgeoning. 

Spectrofluorometers and luminometers are expected to grow only marginally (about 3%) because they compete unsuccessfully with microplate readers, which can handle much higher sample throughputs. Some manufacturers have tried to counter this trend by introducing microplate reader accessories for users with occasional large sample requirements, but this is seen only as a delaying action. Filter fluorometers are faring even worse and are experiencing unit declines as lower-cost spectrofluorometers offer a reasonably priced better alternative. Biosensors and calorimetry instruments also are competing successfully with fluorescence in applications dealing with molecular interactions, an important growth segment. 

The atomic spectroscopy market faired reasonably well in 2004, and continues to improve as compared to the situation a few years back. In fact, if it were not for the minimal prospects for atomic absorption (AA) and inductively coupled plasma (ICP), which account for about 35% of the atomic spectroscopy market, and combined are growing only 2% per year, atomic spectroscopy might be considered a genuine growth market. The important semiconductor industry revived somewhat in 2004, but still cannot match its demand for instrumentation of the boom years from 1999 to 2001. Many observers expect steady improvements during the next few years, but no real significant jump in capital investment. The story in the metals industry is sharply more upbeat. Led by China, the metals industry experienced a fantastic year in terms of production and profits, and instrument purchases have increased, with arc/spark, elemental analyzers and X-ray techniques being the greatest benefactors. 			

Prospects for atomic spectroscopy have brightened, and growth rates are expected to hold steady or slightly increase during the next few years. In 2004, the overall growth rate for atomic spectroscopy was 4.8%, more than a full percentage point better than in the previous two years. However, it is expected that this market will expand at more than 5% this year, and in 2006. These gains are due to a combination of factors, including the generally good economic conditions around the world, especially in Asia. Increased environmental concerns also are encouraging moderate investments for instrumentation for environmental monitoring and testing worldwide, in industrialized nations as well as developing countries. The life sciences area, including food and beverage processing, also can be expected to offer increased opportunities for manufacturers of atomic spectroscopy instruments. 

The X-ray category of instruments, X-ray diffraction (XRD) and X-ray fluorescence (XRF) represent about one-third of the worldwide market for atomic spectroscopy and are expected to exhibit good growth in 2005, each increasing in the 6-7% range. Both powder and single-crystal X-ray techniques should be in strong demand in 2005. The single-crystal segment is now showing renewed strength as the pharmaceutical industry attempts to improve its success rate in developing target compounds into viable drug prospects. Still, XRD is indispensable to proteomics research, and its continued importance is assured, although NMR and MS are strong alternative techniques for structural analysis. Powder XRD, the larger segment, has broader industrial applications, including some in life sciences, especially for pharmaceutical quality control. The market for XRF instruments is benefiting from increased capital investments in the metals, petroleum, and semiconductor industries, which account for more than 60% of worldwide applications. 

Increasing semiconductor, pharmaceutical, and environmental applications will drive ICP-MS and elemental analyzer demand with increases of 8.8% and 6.8% respectively. Total organic carbon (TOC) and mercury analyzers will outpace all other elemental analyzer products because of the improving prospects in the semiconductor industry, continued growth in pharmaceutical and food processing, and demand for pure water and ultrapure water testing. The ICP market has slowed somewhat because of state and local governmental budget constraints, especially in the United States. This is offset somewhat by purchases of ICP-MS instruments that offer increased sensitivity and fast analysis. 

Arc-spark optical emission spectroscopy is enjoying a renaissance of popularity as the metals industry copes with high demand worldwide. This is true especially for the rapidly expanding markets in China and India. North America and European prospects also are peaking as primary metals producers reap high profits at maximum capacity production levels. This, coupled with global interest in recycling, bodes well for a very mature technology although growth still likely will be only around 4%. AA spectroscopy, an even more mature technology, likely will see only minimal growth (less than 2%), with system sales declining slightly. AA benefits minimally from the resurgence in metals production and is much more dependent upon the environmental analysis market where other techniques such as ICP vie for the modest opportunities. Aftermarket purchases of accessories, consumables, and software, as well as service, really provide most of the growth that is expected. 

Mass spectrometry continues to be the growth engine of spectroscopy. If it were not for the declining sector MS market, MS would be growing overall at almost 9% per year. Even so, the market still is expected to turn in growth at about 7.5% for 2005. Although the important pharmaceutical and biotechnology industries have re-evaluated their appetite for capital investment, including sophisticated instrumentation, mass spectroscopists cannot seem to resist buying the latest products with enhanced sensitivity. Sensing this, MS suppliers have introduced a host of new high-resolution mass spectrometers to meet the pent-up demand. However, interest is not promotion driven, rather the technical challenges facing many scientists are daunting and mass spectrometry appears to offer near term solutions. 			

Quadrupole time-of-flight (Q-TOF) MS is one of the strongest performing techniques within the field, and for that matter the entire analytical instrument industry. Q-TOF sales increased at more than 10% in 2004 and should do about the same in 2005. Pharmaceutical and biotech users cannot have enough of these expensive but ultra-performance instruments. Attributes including exact mass, and high mass range and sensitivity make it a strong competitor to FT-MS instruments. New instrument offerings and a lifting of the cloud of patent infringement litigation further set the stage for a robust purchasing environment. Q-TOF's now are available in a wide variety of configurations, all with improved capabilities, and some are available at more affordable prices. Q-TOF mass spectrometry is well suited for applications such as structural analysis — thus competing with NMR and XRD techniques — metabolism studies, and protein-protein and protein-ligand interaction studies. The technique's high sensitivity and mass resolution are ideal qualities for research and development laboratories. 

New product introductions of a wide variety of high-resolution LC-MS instruments continue to generate excitement in this large and important market. However, the market is steadily reaching saturation, and must also compete with other mass spectrometry techniques such as Q-TOF and FT-MS. The largest category, triple-quadrupole systems, is expected to exhibit the slowest growth — about 5% for 2005. Ion trap instruments likely will grow at twice that rate. Likewise, both Q-trap LC-MS and LC-TOF instruments represent the most dynamic segments of this instrument group and are expected to expand at mid-double digit levels.  So the entire high-resolution LC-MS segment should grow at more than 7% in 2005 on the heels of an 11% increase in 2004. Drug discovery and protein research in pharmaceutical and contract research labs is the main driving force behind this growth.  

The market for matrix-assisted laser desorption ionization time-of-flight (MALDI-TOF) has been reinvigorated during the last year or two with the introduction of a number of new high-resolution instruments, including several tandem mass spec units. It is expected that the segment will grow slightly faster than the high-resolution LC-MS segment at around 7.5% in 2005. Product innovation is a great "cure all" for slumping markets. Some users, especially European researchers, prefer electrophoresis sample preparation to LC, so MALDI-TOF is a natural choice, although fully integrated proteomic systems combining electrophoresis, sample excising, and MS analysis are not necessarily the preferred approach at big pharmaceutical and biotech labs. 

Without question the FT-MS segment was the "hot" market in 2004, and is slated to continue into 2005, despite the very high prices approaching $1 million. There are now a number of product types available to researchers, some at "down to earth" prices, for example, under $500,000. Technology innovation and more attention to user interfacing has brought down some of the barriers to purchase. The superlative sensitivity and mass resolution of these instruments is the envy of many scientists as a solution to their most difficult analytical challenges. Suppliers continue to focus on higher performance at high cost, but buyers are still lining up. It is expected that growth for 2005 will be 20% or better. Because 2004 sales were slightly lower than anticipated due to manufacturing delays, a banner year is likely. 

The success of FT-MS is one more "nail in the coffin" of magnetic sector mass spectrometry, which continues to slowly decline. Fortunately, its primary proponents in government labs around the world still prefer the flexibility of these very complex systems, so the technique will be around for many more years.  

Yes, spectroscopy technology has been with us for many years, and some say it is a very mature market. Yet the suppliers of spectroscopy products do not all feel that way, as they continue to innovate and introduce new products every year. One has only to look at the AA field to appreciate this trend. A dozen or more new products have debuted in the last few years, some based upon unique and cost-effective approaches. The result has been much more dynamism in a well established market that also has resulted in better performance and lower prices for scientists. The AA suppliers are not philanthropists; rather they recognize the market opportunity that exists to replace an aging installed base of instruments, in addition to new demand from emerging economies. 			

At the same time, this marketplace is filled with relatively new techniques, which offer high performance and address the growing challenges that scientists and researchers face every day. From drug testing for designer drugs at the Olympics, to achieving the benefits of personalized medicine, spectroscopy instruments play a very important role. We can only hope that scientists continue to demand that suppliers develop new instruments to meet these needs and that suppliers are up to the challenge. It is a good bet that the trend will continue because so many segments of spectroscopy exhibit genuine high growth potential.

 is president and chief executive officer of Strategic Directions International, Inc. (Los Angeles, CA). Phone: (310) 641-4982; e-mail:

Articles

Overall, 2004 was a good year for the laboratory analytical and life science instrument industry. A rebound in industrial demand, steady increases in life science spending, and strong growth in Asia and North America drove instrument systems and aftermarket sales. Similar to last year, 2005 should bring healthy sales growth as many of the end-market and regional drivers of last year continue to fuel demand.

Economic Prosperity Drives Spectroscopy Prospects in 2005

During a year that has been saturated with talk of the national economy, a fledgling job market, and an unpredictable unemployment rate, spectroscopists can take solace in the fact that according to the 2005 

 Salary and Employment Survey, while their industry did not experience superlative growth during the previous 12 months, it did undergo a steady, albeit moderate, period of progress.  

A total of 405 spectroscopists responded to this year's survey. As in past years, a majority of these were male —  298, or 73.6% — while females totaled 107, or 26.4%. A little more than a third of all respondents were between the ages of 40 and 55 — the largest age group.  

While the most recent national unemployment rate (as of December 2004), according to the U.S. Labor Department, dipped to 5.2%, only 2.4% of those responding spectroscopists said they currently are not employed, compared with last year's 3.7%. Of those working, 90.7% are full-time employees, a slight increase from last year. 

The average annual salary for all respondents is $72,920, a slight increase over last year's $72,140. This marked the lowest salary increase in the last five years, however, at just $780, and is far below the average increase of $1646 between 2001 and 2005 (see Table 1).  

 Table I. 2001-2005 average salaries						

To coincide with the marginal across-the-board increase, the largest group of responders (8.7%) earn between $60,000 and $65,000. The largest single group in last year's survey, also 8.7% of the total respondents, earned between $75,000 and $80,000.  

This year, 4.5% of respondents report they earn less than $25,000 per year — more than double the 2.1% in this category from 2004. On the other hand, 3% are earning more than $150,000 per year, a marginal increase from last year's 2.1%. 

Industrial sector continues to strive, academia falters.

 Industrial spectroscopists continue to be the highest paid group at an average salary of $78,170, up slightly from last year's mark of $77,420. While the one-year increase of $750 is modest, at best, industrial spectroscopists have seen their yearly average salaries climb $6400 in just two years ($71,770 in 2003). 

While their industrial counterparts have enjoyed escalating pay, academic spectroscopists continue to see their yearly salaries dwindle at a rapid pace. This year academic scientists, on average, will earn $54,640, a significant decrease from 2004's $59,850 average, and a far cry from the $62,130 they averaged just two years ago.  

"Academia typically pays less than industry, and that's always been the case. However, many academic salaries are  9- or 10-month salaries, based on the idea that many academics get the summer off," says David Ball, a professor of Chemistry at Cleveland State University, and a member of 

's Editorial Advisory Board.  "Many academics have the opportunity to supplement their salaries with external grants or summer school teaching. When that additional salary is taken into account, in many cases, a full-year salary can be commensurate with an industrial salary. Academics also can consult, while industry frowns upon or forbids moonlighting, or can serve as expert witnesses or write textbooks for royalties. An academic scientist has other opportunities to earn income than many industrial workers do." 

Government spectroscopists rebound after a modest decline in their 2003-2004 average salaries, $68,430 and $66,610, respectively, and will earn an average of $68,700 in 2005 (see Figure 1).  

Figure 1. Salary summary by job sector, 2003-2005.			

 As it has been in the past, spectroscopists on the country's bookends continue to enjoy the highest financial satisfaction.  

The Pacific/Pacific Northwest is once again the highest paying region. On average, those working on the West Coast earn $81,590 a year. This marks a $2920 increase from last year's data, and more importantly, a $9140 increase in just two years. 

The Northeast/Middle Atlantic region also continues to be a prosperous area to work. At an average of $76,830, scientists in this region have enjoyed a $5450 increase in two years. The Midwest/Great Lakes region is once again the third-highest paying region with an average yearly salary of $72,540 — up from last year's $71,480.  

After experiencing a sudden spike in its average salary from 2003 to 2004 ($65,000 to $69,790) the Southern/South Atlantic region of the country declined to $68,940. On the other hand, the Western-Central region's freefall continued with an average 2005 salary of $64,590. Last year, the region's spectroscopists averaged $66,760, and in 2003, they averaged $68,240 — a $3650 decline in two years (see Figure 2).  

Figure 2. Salary summary by geographic location, 2003-2005.			

"Cost of living is a major factor," Ball says. "[My family and I] moved to Cleveland from the San Francisco Bay Area. I had a 12-month job offer from Henckel in Sonoma County and an academic position here in Cleveland. My annualized salary from the academic position was about the same as the industrial position, and given the cost of living in the Midwest, the money went much farther. It was a no-brainer what decision to make." 

 As can be expected, those with Ph.D.s (42.2% of all respondents) earn the most at an average of $85,400 per year, up from last year's $83,690 average and 2003's $81,900. M.S. holders (19.9%) also experienced a two-year increase. This year those with a Master's will earn an average of $71,450, a $1770 increase from last year and a $4170 jump from 2003. While B.S. holders enjoyed an unexpected hike in 2004 ($62,130 from 2003's $58,410 average), they saw a disappointing return to an average of $59,030 for this year's survey — a $3100 decrease. 

Males hold top spot, but women make greater strides.

 No matter which way you examine the numbers, males are consistently ahead of females when it concerns salary. 

According to the survey, a male spectroscopist will earn an average of $75,640 in 2005. This total represents more than $10,000 of what a female spectroscopist will earn, $65,450. While troubling, a closer look at the numbers shows that females closed a significant part of that gap that has separated the genders in the past.  

From 2003 to 2004, the pay for a male spectroscopist increased by $2660, from $72,690 to $75,350. From 2004 to this year, however, the increase was a miniscule $290. 

On the other hand, a female spectroscopist saw a decrease in her 2003-2004 salary by $220 — $58,900 to $58,680. 2005's figure ($65,450) represents a $6770 increase (see Figure 3).  

Figure 3. Male vs. Female Salaries, 2003-2005.			

"I think employers have been made very aware of the need to avoid discrimination based upon gender, and that may account for the higher annual increase," says Eileen Skelly Frame, of Full Spectrum Analytical Consultants, and a member of 

's Editorial Advisory Board. "Employers are either readjusting salaries to be 'equitable' or are promoting more women to higher paying management positions to avoid lawsuits." 

Two factors that always have favored males are time on the job and degree held. And while males still hold a significant edge in these departments, this year's female respondents have made substantial gains, hence the reason for their sudden salary surge. 

A little more than 46% of males hold a Ph.D. — the same as last year's numbers. In 2004, male respondents had an average of 21.75 years of experience. This year, however, that number dipped to 18.66 years.  

This year, 30.8% of female spectroscopists hold a Ph.D. Whether it's the random selection process of the survey, or a trend of females returning to school, this represents a 10% jump over last year's numbers. Also increasing was years of experience for the female respondents — 14.43 in 2003 to this year's average of 15.46. 

"There are still fewer high paid female managers than male managers in industry. There are also many more female B.S. level, or lower, scientists/technicians in spectroscopy labs than male B.S. scientists, in my experience, " Frame says. "I have had labs with all females reporting to me, labs with from 2:1 to 4:1 female/male ratios, and in most of these cases, the males had M.S. or Ph.D. degrees, while the females had high school diplomas or B.S. degrees. There is no question that M.S. and Ph.D. degrees equate to higher salaries in industry. 

"The increase in female Ph.D.s is going to increase the competition for jobs. It will certainly increase the pressure on older, more experienced job-seekers to keep updating their skills as employers try to hold down costs by hiring recent graduates," Frame continues. "For the industry, it means that employers must address the issues of family leave, on-site daycare, flexible working hours, etc. Some employers are ahead of the curve with respect to these issues."  

Table II. Job Titles for all Respondents			

Money might make the world go around, but can it make you happy? According to our survey, yes, there is a direct correlation between money and job satisfaction.  							

On whole, 24% of all respondents are "very satisfied" with their professional choices; 46% are "satisfied;" 23.5% are "somewhat satisfied;" and 6.4% (5.4% of males, 8.6% of females) are "dissatisfied." 

Not surprisingly, those in the Pacific/Pacific Northwest, who make the most on average, reported the lowest dissatisfaction level, just 2.3%. A little more than 11% of government spectroscopists are dissatisfied, followed by an 8.2% rate from those in academia.  

Figure 4. Respondents by Field (2005).			

To give a more direct link between satisfaction and salary, those who are either very satisfied or satisfied earn an average of $75,740 a year, or $2820 more than the average salary respondent. Those somewhat satisfied are making an average of $65,450 — $7470 less than an average respondent — and just $750 more than the dissatisfied portion of respondents. 

"Does money equal happiness? No, but it's a contributing factor, just as in any other job," Ball says. "I have had local working environments that have severely eroded my professional morale despite the fact that my salary was untouched by the circumstances. On the other hand, many times I feel like I have the best job in the world despite the fact that I'm underpaid relative to other scientists." 

Less than half (45%) of this year's respondents are eligible for supplemental income. As expected, those in the Pacific/Pacific Northwest region were the most eligible group at 54.5%, and nearly half of those who were most satisfied with their jobs were collecting bonus income as well. The dissatisfied population is the least eligible. 

The average yearly amount of supplemental income was $6,790, with the highest percent of respondents (15.6%) reporting they earn up to or more than $15,000 a year in supplemental income. The highest percentage of eligible male respondents (17.5%) fall into that category, while the highest percentage of female respondents (15%) fall into one of two categories: less than $1000 per year or $1000-$1999 per year. 

On average, males earn $7230 in supplemental income per year, and females earn $5350. Ph.D. holders, while the lowest percent of eligible respondents, earn the most at an average of $8880 per year. B.S. holders earn almost half of that at $4640. Academia earned the highest average of supplemental income at $7210, slightly ahead of private industry's $7,030 average. Government respondents' average is nearly three times lower, at $2360. 

A little less than one-third of all respondents (31.6%) are considering a job change within the next year. The highest percent among degree holders was Ph.D.s at 34.7%. Females are slightly more likely to make a job change than males (34.3% compared to 30.5% of males). Those in academia responded "yes" the most at 37.7%. Government workers had the highest "no" percentage at 71.7%. Those dissatisfied and with less than 10 years experience also are the most likely to look for changes in the next year.  			

For those looking for a change, the biggest factors are income (53.1%) and professional advancement (50%). While already enjoying the highest average salary, those in the Pacific/Pacific Northwest are, by percentage, now looking for advancement above all other factors (66.7%). 61.1% of the Northeast/Middle Atlantic congregation is seeking income above all else, 8% above the average. 

Ph.D. and B.S. holders are seeking a change of income, while 81% M.S. holders will seek advancement. Males and females equally cite income as a factor, but females are seeking advancement by a little more than 8%. Males, on the other hand, see company downsizing as a reason for change by a 2-to-1 ratio (16.7% to 8.3%). 

Nearly 74% of academic spectroscopists cite income as a factor in seeking change, while 53.8% of those in government report that their job environment is the cause. Spectroscopists with less than 10 years and 10-20 years of experience are seeking income and advancement above all other factors, while those with 21-30 years seek advancement (44.4%) over income and job environment (40.7%). More than 30 years in the industry puts a desire for more money and intellectual challenge on the same level (54.5%). 

On average those with less than 10 years experience earn $55,490 (up from $49,000 in 2004); those with from 10 to 20 years on the job pull in $80,040, or more than $10,000 from last year's average; from 20 to 30 years equals $83,700 a year ($79,710 a year ago); and finally, those with more than three decades under their belt, who made the most on average last year ($82,910), suddenly fell to an average of $75,640. Just three years ago, this group averaged $86,290 a year. 

So where is everyone headed? A large majority (76.4%) of all respondents who intend to seek a job change would prefer to work in private industry. Nearly 83% of those in the private sector intend to stand pat, while 70% of government employees are seeking entrance into private industry. Academics, on the other hand, are split between preferring to stay in their current field and switching into private industry. 

So what do all these numbers mean? To better understand where spectroscopists stand with regard to the broader scientific community one can refer to the August 2004 salary survey of members of the American Chemical Society (ACS), which appeared in 

The ACS survey was based on 11,600 responses. Close to 10,700 of those respondents were chemists; 427 were chemical engineers; and 400 held nonchemical degrees. 500 respondents were disqualified as they were retired or not seeking employment, for a total of 10,200 qualified respondents. Job fields of the ACS respondents included analytical chemistry, biotechnology, pharmaceutical chemistry, and materials science. 

Spectroscopy's annual look at practiitioners' incomes and attitudes about their jobs finds that both have improved modestly in the face of challenging economic and employment conditions.

2005 Salary Survey: Careers, Compensation, and Contentment in Today's Spectroscopy Community

Recently I attended the International Forum for Process Analysis and Control (IFPAC 2005) in Arlington, VA (January10-13, 2005). As usual, it was hectic with several simultaneous sessions. The forum, originally started as a showcase for chemical analyses, has become a showcase for pharmaceutical (process) analyses, as well. A mere five years ago, there was a single session in the morning and one in the afternoon on pharmaceutical analysis. With the introduction of FDA's Process Analysis Technologies (PAT) initiative, the number of sessions has exploded.			

Among the co-sponsors of the event were the Center for Process Analytical Chemistry (CPAC), Measurement and Control Engineering Center (MCEC), the Center for Pharmaceutical Process Research (CPPR) (along with the Particle Technology and Crystallization Consortium [PTCC]), Duquesne Center Pharmaceutical Technology (DCPT), the Consortium for the Advancement of Manufacturing of Pharmaceuticals (CAMP), the National Institute of Standards and Technology (NIST), and FDA's Center for Drug Evaluation & Research (CDER). All participated and contributed to the overall success of the meeting. 

Of course, the once "roaming" meeting, known for its warm-weather locations each January (Florida, Arizona, Nevada, and Southern California), has found a home in Virginia. It has eschewed the golf tournament for proximity to the Food and Drug Administration. This has paid off handsomely: the participation of FDA personnel has made the conference a "must-attend" for pharmaceutical scientists with any interest in doing process analyses. 

After a Tuesday morning plenary session, there were five afternoon sessions: "PAT — A Systems Approach," "NESSi Update — 2nd Generation," "Advanced Instrumental Concepts for Process Analysis," "Mass Spectroscopy — Novel Innovations," and "Homeland Security — Chemical & Biological Agent Detection." The conference never slowed down after that. 

One way to look at the papers is by "type" of talk; there were talks on the legalities of PAT, the "philosophy" of PAT, the hardware and software of PAT, and the "systematic" approach. This was a good mix, because all these need to be addressed for PAT to be a success in any company. As Ajaz Hussain has been saying from the beginning: "PAT is not about the hardware." I will adjust that to, "PAT is not 

 about the hardware." The general categories of the meeting now are presented in no specific order (actually, I'm a hardware person, so I will lead off there). There is, of course, no way to discuss all or even most of the papers given throughout the meeting, so the papers mentioned will highlight the diversity of topics addressed. 

As the "grandfather" of multiple pharmaceutical analyses (raw materials ID since 1984), NIR had a major portion of the talks at the meeting. Used for liquids, powders and solids, NIR has the attractive benefit of no sample preparation. This makes it ideas for process measurements. The line between a specific piece of hardware and a specific application is sometimes difficult to draw, so I will not attempt it.  			

 Ke Hong and Martin Warman (Pfizer, Morris Plains, NJ) discussed some interesting work with a Corona (wireless) system attached to a moving blender. They compared a "large spot" measurement for blend uniformity and a "small spot" attachment for troubleshooting purposes. Pfizer obtained the first patent for measuring mixing of powders in a moving blender and continue to lead the way in this field. 

Dry blend measurements of direct compression mixtures.

 Nathan Pixley of Merck (West Point, PA) addressed this topic. Because there is no second step of liquid addition and granulation to assist in the distribution of materials, the uniformity of the blend can even be more critical in a direct compression formulation. The methodology, endpoint criteria, and validation also were covered in the talk. 

In-line monitoring of moisture in fluid bed dryers.

 This was the topic of a talk by a group from Merck, the key point of which was that most similar previous applications have been at the university level. In this talk, several commercial-level successes were reported. Both 65- and 300-liter pilot-scale and 600-liter production-scale models with data were shown. The presenters also discussed the effects of sampling on-line. Indeed, when performing any on-line measurement, how and where samples are taken for reference will make or break the method. 

 Seetha Ananthaiah (Bristol Meyers Squibb, New Brunswick, NJ) discussed the use of NIR for this application. The ability to continuously monitor the drug content is especially valuable in narrow therapeutic index (NTI) products. This is where the efficacy of the active is defined within small parameters: too high and toxicity is a concern; too low and it might be ineffective. Such products are not well served by the "traditional" plus or minus 15% window of USP standards. Technology transfer from lab to process plant was also discussed. 

 Bruker Optics (Bellerica, MA) showed some of its work using a system that combines transmission content uniformity analyses with tablet weight, thickness, hardness, and diameter. This approach does more to determine the "goodness" of tablets than merely assaying them for active ingredient. 

 Larry McDermott and Richard Crocombe (Axsun, Bellerica, MA) presented a couple of papers about their work with these devices. They described the new superluminescent light emitting diodes (SLEDs) used as sources for their equipment. The crux of the hardware talk was that most other types of equipment use full-range NIR, because that is the nature of their sources. The Axsun claim is that, because only a shorter portion of the spectrum actually is used, the instrument provides maximum illumination over the needed range, using far lower power requirements and with a longer source life expectancy. 

The applications covered were simple, previously demonstrated assays. The difference was that Axsun used their quite small instrument to generate similar results. If this proves to be a rugged instrument, it will go a long way to bringing about the next generation of smaller, more durable instruments for PAT. 

 This was described by Ingrid Maes (Siemens, AG). The company's bioreactor not only monitors the individual components, but is set to refresh materials through feedback mechanisms. This allows the batch to be kept within "optimal" conditions, assuring optimal yield each time. 

 Fred Long (Spectroscopic Solutions, Randolph, NJ) shared the work he has done with these patches, explaining that the work needed to prepare the instrument and software for compliance before proceeding to the actual analysis mode. While early work on patches dates to the late 1980s, this is the first report on process analyses of them. 

The Mid-IR has not gone out of style yet. While (apparently) lacking the versatility of NIR, it still remains much better at "cleaner" spectra, allowing specific interpretations. 			

David Peters (Hamilton Sundstrand Systems-AIT Group) used FT-IR to determine "key" constituents in a fermentation process. The group measured glucose, ethanol, sucrose, lactic acid, and citric acid in a simulated fermentation system. 

One of the fastest growing techniques for PAT is Raman spectroscopy. Shih-Ying Chang, representing a group from Bristol-Meyers Squibb's (BMS) Pharmaceutical Research Institute presented the group's work on monitoring solid dosage forms via Raman. 			

Polymorph transformation was the subject of a talk by Dimuthu Jayawickrama (BMS, New Brunswick, NJ). The BMS work was used to describe polymorph transition from a hydrate to neat form in a water/glycol slurry. The model was transferred to a second instrument and has even been used to assess solid forms. The details of model building and method transfer were discussed. 

Raman was combined with energy dispersive X-ray fluorescence (EDXRF) microscopy in a talk by Fiona Clarke (Pfizer, Kent, UK) in consort with Fran Adar (Horiba Jobin Yvon, Edison, NJ). They explained that He Raman allows for high spatial resolution (1 Âµm) hyperspectral images for particle size and chemical distribution in a granulation. EDXRF allows for rapid hyperspectral images with 10 Âµm spatial resolution. A number of images were used for illustration. 

Gas monitoring of bioprocesses was the topic of a talk by Ronald Rich (Atmosphere Recovery, Inc., Plymouth, MN). The set-up consisted of a 45-cm detector module, using a low-power laser with chemically specific optical filters. The system was capable of performing a gas composition analysis every 50 milliseconds. It was seen that one instrument could monitor several bioprocessors. 

Of course, chemical imaging (CI) is not, 

, a single technique. The session consisted of (the above mentioned) Raman/EDXRF, NIR-CI, Terahertz Pulsed Imaging, diffuse reflection techniques, and enlisting a tunable laser light source. The contributors were from instrument companies (Spectral Dimensions, Olney, MD; OPOTEK, Inc., Carlsbad, CA; TeraView Ltd., Cambridge, UK), academia (Duquesne University, Pittsburgh, PA; Insitute fur Angewandte Forschung, Germany), government (FDA), and industry (Pfizer, US and UK). This subject is large and important enough to warrant an entire issue and is merely mentioned here. 			

It is inconceivable that the massive amount of data generated by PAT could be interpreted 

 chemometrics. (The mass of data once was referred to as "an insurmountable opportunity.") Several speakers addressed various aspects of the concepts of chemometrics.  			

 Charles Miller (DuPont, Houston, TX) gave a picture of the changing philosophy of process measurements. He showed that in the "good old days," process measurements were univariate: pH, pressure, temperature, flow rate, sample level, etc. Now, samples are recognized to be the result of complex interactions among all the potential influences within a system. In a real process, "ideal" ideas such as holding all parameters constant and only allowing one to change do not truly monitor a "real" system. Instead, all the parameters are monitored throughout one or more runs and the product is then evaluated in light of all the parameters using chemometrics. Miller explained how formerly lab-based instruments can be used in a process environment when statistical and chemometric (multivariate) methods are employed. 

Brian Rohrback (Infometrix, Inc., Bothell, WA) spoke about a new Microsoft-driven software, called ActiveX. It supposedly allows any system to access a chemometrics computational engine without reading or writing files. Its basic purpose is to seamlessly combine software from various sources and allow data to flow uninterrupted. 

Not to be outdone, Chris Ambrozic (Umetrics, Kinnelon, NJ) discussed a real-time chemometrics package wherein real-time data can be processed from multiple analyzers in a straightforward approach. 

William Doyle (Axiom Analytical Inc., Irvine, CA) discussed a computer architecture, compliant with 21 CFR Part 11, based upon an SQL database. It was designed to work with multiple instruments and integrate the resultant data into coherent files. 

Since the concept of PAT was put forth by the FDA (even prior to the Guidance), there were a number of approaches espoused by different people. A number of people believe that PAT is the total understanding and control of an entire system for each drug product. They think that a company should place sensors along an entire production line from the beginning. At the other extreme, a number of more conservative voices suggest that individual processes be understood and controlled first: raw materials, bending, drying, compression, etc.  			

Several FDA and industry speakers spoke about PAT as a "systems approach." These speakers included Rebecca Rodriguez (DFI/ORO/ORA/FDA), Gerd Fischer (Aventis, Kansas City, MO), Gordon Muirhead (GlaxoSmithKline, Wirral, UK), and Ron Miller (BMS, New Brunswick, NJ). All gave their own perspectives of PAT as an integrated system of instruments, software, new data treatment and storage approaches, and as a means of understanding the pharmaceutical process, prior to controlling the process. 

Pat Picariello, of the American Society for Testing and Materials (ASTM), gave an update on Committee E55 (Pharmaceutical Application of Process Analytical Technology). The committee has been in existence since January 2004. Its members include representatives from industrial, government, and academia involved in pharmaceutical manufacturing. The talk covered some of the work toward standardization to date; the main focus was on the scope and structure of the committee (ASTM is not known for hasty decisions. It is a consensus organization, by its very nature.) It also coordinates with ASTM International, Committee F38, and FDA to streamline standards for the industry. 

One particular study was quite detailed was presented by Tim Stevens (BMS). His presentation covered the major steps of PAT: 

c. Chemometric modeling and model building 

This model was presented for Loss on Drying (LOD) in a production setting across a period of over six months. The methodology used and results obtained were presented. This is one of the most complete models presented at a PAT meeting. While the chemistry of drying is trivial, the model presented was a working model for PAT, in general, for 

The explosion of chemometric methods and the flood of data generated by PAT has proved a trial for IT personnel. The acquisition, evaluation, storage, and retrieval of PAT data have proven to be major projects for "computer-jockeys." Performing logical data handling procedures in light of 21 CFR Part 11 is no trivial task. Some of the papers involved both software and approaches for data storage, so there were few clear-cut "management" talks.  			

Gawayne Mahboubian-Jones (Optimal Industrial Automation, Yale, Bristol, UK), did address "knowledge management" per se. Several problems and some solutions were offered. 

James Duckworth (Thermo Electron Corp., Minneapolis, MN) gave a nice talk about the data file standards needed for successful data storage and management. Some of the topics covered were  

a. Business drivers for common instrument data formats 

b. Historical efforts, i.e., JCAMP, AnDI, and SPC 

Columns

Attendees from the pharmaceutical industry and FDA discussed a range of topics, including new spectroscopic analysis techniques, chemical imaging, calibration, and current efforts at ASTM.

Molecular Spectroscopy Workbench - Report from the 19th IFPAC/PAT Initiative Conference

As we left off in our last column, we had proposed a definition of linearity. Now let's start by delving into the ins and outs of the Durbin-Watson statistic (1-6) and looking at how to use it to test for nonlinearity.  			

In fact, we've talked about the Durbin-Watson statistic previously in our columns, although a long time ago and under a different name. Quite a while ago we published a column titled "Alternative Ways to Calculate Standard Deviation" (7). One of the alternative ways described was the calculation by Successive Differences. As we shall see, that calculation is very closely related to the Durbin-Watson statistic. More recently we described this statistic (more directly named) in a sidebar to an article in the 

To relate the Durbin-Watson statistic to our current concerns, we go back to the basics of statistical analysis and remind ourselves how statisticians think about statistics. Here we get into the deep thickets of statistical theory, and meaning and philosophy. We will try to keep it as simple as possible, though.			

Let us start with two of the formulas for standard deviation presented in our earlier column (7). One of the formulas is the "ordinary" formula for standard deviation:  

The other formula is the formula for calculating standard deviation by Successive Differences:  

Now we ask ourselves the question: "If we calculate the standard deviation for a set of data (or errors) from these two formulas, will they give us the same answer?" And the answer to that question is that they will, if (that's a very big "if") the data and the errors have the characteristics that statisticians consider "good" statistical properties: random, independent (uncorrelated), constant variance, and in this case, a Normal distribution, and for errors, a mean (μ) of zero, as well. For a set of data that meets all these criteria, we can expect the two computations to produce the same answer (within the limits of what sometimes loosely is called "statistical variability"). 

So under conditions where we expect the same answer from both computations, we expect the ratio of the computations to equal 1 (unity). Basically, this is a general description of how statisticians think about problems: First, compare the results of two computations of what is nominally the same quantity when all conditions meet the specified assumptions. Then if the comparison fails, this constitutes evidence that 

 about the data is not conforming to the expected characteristic (that is, is not random, is correlated, is heteroscedastic, is not Normal, and so forth). The Durbin-Watson statistic is that type of computation, stripped to its barest essentials. Dividing equation 2 by equation 1 above, canceling similar terms, noting that the mean error is zero and ignoring the constant factor (2) we arrive at:  

Because of the way it is calculated, particularly the way the constant factor is ignored, the expected value of DW is 2, when the data do in fact meet all the specified criteria: random, independent errors, and so forth. Nonlinearity will cause the computed value of DW to be statistically significantly less than 2. (Homework assignment for the reader: what characteristic will make DW be statistically significantly 

Figure 1. An illustration of the behavior of linear data. (a) Linear data spread out around a straight line. (b) The residuals are spread evenly around zero.			

Figures 1 and 2 illustrate graphically what happens when you inspect the residuals from a calibration. When you plot linear data, the data are spread out evenly around the calibration line as shown in Figure 1a. When plotting the residuals, the line representing the calibration line is brought into coincidence with the 

 axis, so that the residuals are spread out evenly around the 

 axis, as shown in Figure 1b. For nonlinear data, shown in Figure 2a, a plot of the residuals shows that although the calibration line still coincides with the 

 axis, the data do not follow that line. Therefore, although the residuals still have equal positive and negative values, they are no longer spread out evenly around the zero line because the actual function is no longer a straight line. Instead, the residuals are spread out evenly around some hypothetical curved line (shown) representing the actual (nonlinear) function describing the data.  

Figure 2. A illustration of the behavior of non-linear data. (a) Nonlinear data does not surround a straight line evenly. (b) The residuals from nonlinear data are not spread out around zero.			

In both the linear and nonlinear cases the total variation of the residuals is the sum of the random error, plus the departure from linearity. When the data is linear, the variance due to the departure from nonlinearity effectively is zero. For a nonlinear set of data, because the 

-difference between adjacent data points is small, the nonlinearity of the function makes minimal contribution to the total difference between adjacent residuals; most of that difference contributing to the successive differences in the numerator of the DW calculation is due to the random noise of the data. The denominator term, on the other hand, is dependent almost entirely upon the systematic variation due to the curvature, and for nonlinear data this is much larger than the random noise contribution. Therefore the denominator variance of the residuals is much larger than the numerator variance when nonlinearity is present, and the Durbin-Watson statistic reflects this by assuming a value less than 2. 

			The problem we all have is that we want answers to be in clear, unambiguous terms: yes/no, black/white, is/isn't linear, while statistics deals in probabilities. It is certainly true that there is no single statistic — not SEE, not 

, not DW, nor any other — that is going to answer the question of whether a given set of data, or residuals, has a linear relation. If we wanted to be really ornery, we could even argue that "linearity" is, as with most mathematical concepts, an idealization of a property that never exists in real data. But that is not productive, and doesn't address the real-world issues that confront us. 							

What are some of these real-world issues? Well, you might want to check out the paper "Graphs in Statistical Analysis" by F.J. Anscombe (9). We'll describe his results again, but it really is worth getting hold of and reading the original paper anyway; it's quite an eye-opener. What Anscombe presents are four sets of synthetic data, representing four simple (single 

-variable) regression situations. One of the data sets represents a reasonably well-behaved set of data: uniform distribution of data along the 

 axis; errors are random, independent and Normally distributed; and in all respects has all the properties that statisticians consider "good." The other three sets show very gross departures of varying kinds (including one that is severely nonlinear), from this well-behaved data set. So what's the big deal about that? By design, all four sets of data have identical values of all the common regression statistics: coefficients, SEE, 

, and so forth. The intent is, of course, to show that no set of statistics can diagnose unambiguously all possible problems in all situations. It is immediately clear, when you look at the graphs of the four data sets on the other hand, which is the "good" one, which have the problems, and what the problems are. Any statistician worth his salt will tell you that if you are doing calibration work, you should examine the residual plots, and any others that might be informative.  

But the FDA/ICH guidelines do not promote that approach even though they are mentioned. To the contrary, they emphasize calculating and submitting the numerical results from the line-fitting process. Under ordinary circumstances, that really is not too bad, as long as you understand what you are doing, which usually means going back to basic statistical theory. This theory states that if data meet certain criteria, criteria that (always) include the fact that the errors that are random and independent, and (usually) Normally distributed, then certain calculations can be done and probabilistic statements made about the results of those calculations. 

If you make the calculation and the value turns out to be one of low probability, then that is taken as evidence that your data fail to meet one or more of the criteria that they are assumed to meet. Note that the calculation alone does not tell you which criterion is not met; the criterion that it does not meet might or might not be the one you are concerned with. The converse, however, is, strictly speaking, not true. If your calculated result turns out to be a high-probability value, it does not "prove" that the data meet the criteria. That is what Anscombe's paper is demonstrating, because there is a (natural) tendency to forget that point, and assume that a "good" statistic means "good" data.  

So where does that leave us? Does it mean that statistics are useless, or that FDA is clueless? No, but it means that all these things have to be done with an eye to knowing what can go wrong. We strongly suspect that FDA has taken the position it has because it has found that, even though numerical statistics are not perfect, they provide an objective measure of calibration performance, and they have found through hard experience that the subjective interpretation of graphs is fraught even more with problems than is the use of admittedly imperfect statistics. 			

For similar reasons, the statement, "If the Durbin-Watson test demonstrates a correlation, then the relationship between the two assays is not linear," is not exactly correct, either. Under some circumstances, a linear correlation also can give rise to a statistically significant value of DW. In fact, for any statistic, it always is possible to construct a data set that gives a high-probability value for the statistic, yet the data clearly and obviously fail to meet the pertinent criteria (again, Anscombe is a good example of this for a few common statistics). So what should we do? Well, different statistics show different sensitivities to particular departures from the ideal, and this is where DW comes in. 

The key to calculating the Durbin-Watson statistic is that prior to performing the calculation, the data must be put into a suitable order. The Durbin-Watson statistic then is sensitive to serial correlations of the ordered data. While the serial correlation often is thought of in connection with time series, that is only one of its applications. 

Draper and Smith (1) discuss the application of DW to the analysis of residuals from a calibration; their discussion is based upon the fundamental work of Durbin et al., in the references listed at the beginning of this column. While we cannot reproduce their entire discussion here, at the heart of it is the fact that there are many kinds of serial correlation, including linear, quadratic, and higher order. As Draper and Smith show, the linear correlation between the residuals from the calibration data and the predicted values from that calibration model is zero. Therefore, if the sample data are ordered according to the analyte values predicted from the calibration model, a statistically-significant value of the Durbin-Watson statistic for the residuals in indicative of high-order serial correlation, that is, nonlinearity. 

Draper and Smith point out that you need a minimum of 15 samples in order to get meaningful results from the calculation of the Durbin-Watson statistic (1). Because the Anscombe data set contains only 11 readings, statistically meaningful statements cannot be made. Nevertheless, it is interesting to see the results of the Durbin-Watson statistic applied to the nonlinear set of Anscombe data; the value of the statistic is 1.5073. For comparison, the Durbin-Watson statistic for the data set representing normal "good" data is 2.4816. 

Is DW perfect? Not at all. The way it is calculated, the highest-probability value (the "expected" value) for DW is, as we saw above, 2. Yet it is possible to construct a data set that has a DW value of 2, and clearly and obviously is not linear, as well as being non-random. That data set is as follows: 

				0, 1, 0, -1, 0, 1, 0, -1, 0, 1, 0, -1, 0, 1, 0, -1, 0, ...

But for ordinary data, we would not expect such a sequence to happen. This is the reason most statistics work as general indicators of data performance: the special cases that cause them to fail are themselves low-probability occurrences. In this case the problem is not whether the data are nonlinear — the problem is that they are nonrandom. This is a perfect example of the data failing to meet a criterion other than the one you are concerned with. Therefore the Durbin-Watson test fails, as would any statistical test fail for such data; they are simply not amenable to meaningful statistical calculations. Nevertheless, a "blind" computation of the Durbin-Watson statistic would give an apparently satisfactory value. But this is a warning that other characteristics of the data can cause it to appear to meet the criteria. 

But the mechanics of calculating DW for testing linearity is relatively simple, once you've gone through all the above: sort the data set according to the values predicted from the calibration model, then perform the calculation specified in equation 3. Note that, while the sorting is done using the predicted values from the model, the DW calculations are done using the residuals. But anyone doing calibration work should read Draper and Smith anyway — it's the "bible" of regression analysis (see reference 1). 

The discussions of DW appear on pages 69 and 181-192 of Draper and Smith, third edition (the second edition contains a similar but somewhat less extensive discussion). They also include an algorithm and tables of critical values for deciding whether the correlation is statistically significant. You might also want to check out page 64 for the proof that the linear correlation between residuals and predicted values from the calibration is zero. 

 test different things. As a specific test for nonlinearity, what is the relative utility of DW versus 

 for that purpose? Basically, the answer is that when done according to the way Draper and Smith (and we) described, DW then is sensitive specifically to nonlinearity in the predictions. So, for example, in the case of the Anscombe data, all the other statistics (including 

) might be considered satisfactory, and because they are the same for all four sets of data then all four sets would be considered satisfactory. But if you do the DW test on the data showing nonlinearity, it will flag it as having a low value of the statistic. Anscombe did not provide enough samples' worth of synthetic data in his sets, however, for the calculated statistics to be statistically meaningful. 			

We also note that as a practical matter, meaningful calculation of the Durbin-Watson Statistic requires many samples' worth of data. We noted above that for fewer than 15 samples critical values for this statistic are not listed in the tables. The reason for requiring so many samples is that essentially we are comparing two variances (or, at least, two measures of the same variance). Because variances are distributed as 

, for small numbers of samples this statistic has a very wide range of values indeed, so that comparisons virtually become meaningless because almost anything will fall within the confidence interval, giving this test low statistical power. 

 as a general measure of how good the fit is doesn't make us flinch, either; it is one of the standard statistics for doing that evaluation. Quite the contrary, when we saw 

 being specified as the way to test linearity, we wondered why it was chosen by FDA and ICH, because it is so 

specific. We still don't know why, except for the obvious guess that they weren't aware of DW. We are in favor of keeping the other statistics as measures of the general "goodness of fit" of the model to the data, but in the specific context of trying to assess linearity, still we have to promote DW over 

 as being more suited for that special purpose. (We eventually will discuss in our next few columns an even better method for assessing linearity.) 

As for testing other characteristics of a univariate calibration, there also are ways to test for statistical significance of the slope, to see whether unity slope adequately describes the relationship between test results and analyte concentration. These are described in the book 

Principles and Practice of Spectroscopic Calibration

 (10). The statistics described there are called the "Data Significance 

" (DST) test and the "Slope Significance 

" (SST) test. Unless the DST is significant statistically, though, the SST is meaningless. 			

In principle, there also is a test for the intercept. But because the expected value for the intercept depends upon the slope, it gets a bit hairy. It also makes the confidence interval so large that the test is nigh-on useless — few statisticians recommend it. 

But let's add this coda to the discussion of DW: The fact that DW specifically is sensitive to nonlinearity does not mean that it is perfect. There might be cases of nonlinearity that will not be detected (especially if it's a marginal amount), linear data occasionally will be flagged as nonlinear (α percent of the time, in the long run), and other types of defects in the data can show up by giving a statistically significant value to DW. But all this is true for any and all statistics. The existence of at least one data set that is known to fool the calculation is a warning that the Durbin-Watson statistic, while a (large) step in the right direction, is not the ultimate answer. 

Some further comments here: there does seem to be some confusion between the usage of the statistics recommended by the guidelines, which are excellent for their intended purpose of testing the general "goodness of fit" of a model, and the specific testing of a particular model characteristic, such as linearity. A good deal of this confusion probably is due to the fact that the guidelines recommend those general statistics for the specific task of testing linearity. As Anscombe shows, however, and as we referred to previously, those generalized statistics are not up to the task. 

In our next column we will discuss other methods of testing for linearity that have appeared in the literature. We then will turn our attention to a new test that has been devised. In fact, it turns out that while DW has much in its favor, it is not the final or best answer. The new method is much more direct and specific even than DW. It is the correct way to test for linearity. We will discuss it all in due course, in a future installment of "Chemometrics in Spectroscopy."  

A discussion of how DW can be a powerful tool when different statistical approaches show different sensitivities to particular departures from the ideal.

Spectroscopy-03-01-2005

Economic Prosperity Drives Spectroscopy Prospects in 2005

2005 Salary Survey: Careers, Compensation, and Contentment in Today's Spectroscopy Community

Molecular Spectroscopy Workbench - Report from the 19th IFPAC/PAT Initiative Conference

Chemometrics in Spectroscopy - Linearity in Calibration: The Durbin-Watson Statistic