Introducing Quality Control in the Chemistry ... - ACS Publications

Sep 9, 2009 - In the 1920s, Walter A. Shewhart, working for. Bell labs manufacturing telephony equipment, pioneered QC charts to aid business efficien...
0 downloads 0 Views 544KB Size
In the Laboratory

Introducing Quality Control in the Chemistry Teaching Laboratory Using Control Charts Benjamin Schazmann,* Fiona Regan, Mary Ross, Dermot Diamond, and Brett Paull School of Chemistry, Dublin City University, Dublin 9, Ireland; *[email protected]

The concepts of quality assurance (QA) and quality control (QC) originate from a commercially driven manufacturing environment. In the 1920s, Walter A. Shewhart, working for Bell labs manufacturing telephony equipment, pioneered QC charts to aid business efficiency (1). Control charts remain one of the most fundamental and effective quality tools today. This also encompasses chemical analysis (2, 3). Control charts reveal bias or systematic errors in a process that generates numerical data. The center of a control chart has a predetermined mean value line. Based on convention, lines are placed above and below the mean line corresponding to 2 and 3 standard deviations (σ) of this mean. These lines represent warning and control limits, respectively. Certain tolerances are allowed within the ±3σ lines, but if values fall outside these lines, the analysis may be “out of control” and intervention becomes essential. By the late 1940s, control charts were still uncommon in the field of chemical analysis (4). However, inevitably, QC–QA and control charts were examined more seriously in chemical analysis when the economic benefits were considered. The theory of QC–QA and control charts is now widely taught at undergraduate level, but is not yet fully applied and incorporated into laboratory courses (5–8). The main reason may be a lack of economic consequences at undergraduate level; there is no direct accountability or responsibility for the quality of results presented in student reports. In the worst case some marks may be lost for not taking corrective action, but money, jobs, or lives are not endangered (8). The situation is compounded by a lack of time and teaching resources to expand laboratory courses. The simultaneous teaching of analytical instrumentation skills and the more theoretical qualities of QC is crucial to expose students to the intimate interdependence of the two in real-life scenarios. Several experiments have appeared in this Journal that emphasize the statistical analysis of data (9–12). We describe our initial steps to change an existing laboratory course to incorporate QC analysis. We discuss our findings from this initiative based on results from two semesters over a two-year period: 2006–2007 and 2007–2008. Theoretical Basis and Background We added QC analysis to the analytical chemistry laboratory course aimed at third-year undergraduate analytical science and environmental science and health students as well as master’s students in instrumental analysis. The master’s students have a degree in chemistry and typically over five years of experience in industry. Particularly through work experience, this group should be familiar with QA–QC matters. The undergraduate students have had little or no industrial experience. One of our aims was to correlate previous experience with performance on the QC charts. The course ran over 12 weeks (1 semester per year), consisting of 11 experiments with up to 9 hours allocated to each

experiment over 2 days per week. Students typically worked in pairs on each experiment. In the twelfth week, a presentation was given by the students, allowing for feedback and assessment of quality control skills acquired. As the QC chart is completed, each student has the opportunity to identify other nonrandom features from the overall chart, apart from a more pass–fail vision revolving around the control limits (±3σ) (13). In this article we focus on the QC analysis portion of two experiments: UV–vis determination of Fe2+ in water via colored complex formation with phenanthroline and HPLC determination of caffeine in beverages (14). The UV–vis QC point entailed the calculation of the molar extinction coefficient, ε, of a copper nitrate solution from a simple absorbance measurement. For the HPLC experiment, a caffeine control sample was run and the column efficiency, N, and retention factor, k, were used for the QC analysis. The experiments were chosen as the instruments are among the most useful and commonly encountered in the analytical chemistry work place and represent a broad range of experimental difficulty. In the first year, the initial chart mean and limits were established by the staff and all students used an aliquot of the same sample stock at subsequent QC points. The staff values of means and chart limits were based on 10 replicate measurements carried out each day over 2 weeks by our staff, prior to the commencement of the teaching activity. In the second year (and further years) of the study, the mean and limits could be established using previous year’s student-generated data. The QC part of the experiment was typically completed prior to continuing with calibration and the main analysis of each laboratory session. Experimental Procedure for QC Analysis UV–Vis The copper nitrate (0.01 M) solution is obtained from the instructor. The UV–vis instrument is set to single-read mode at a wavelength of 800 nm and is zeroed with distilled water. Three replicate absorbance measurements of the copper nitrate solution are recorded and saved. For each of the readings, the molar absorption coefficient at 800 nm, ε, is calculated from the Beer–Lambert equation A =  εcl, where A is absorbance, c is concentration of the sample, and l is sample path length. The mean and standard deviation of these calculations are reported to the instructor on the same day as the experiment is carried out. The data are entered on the appropriate QC control chart and briefly discussed with the instructor. HPLC The detector is set to 254 nm and a flow rate of 1.5 mL∙min is obtained. The system is equilibrated with 50∙50 methanol∙water mobile phase for at least 15 min and the baseline is observed to ensure that it does not drift. A small volume of an unknown caffeine sample is issued by the instructor. The instructor is consulted on how to use the injector. The sample is run in trip-

© Division of Chemical Education  •  www.JCE.DivCHED.org  •  Vol. 86  No. 9  September 2009  •  Journal of Chemical Education

1085

In the Laboratory

tR N = 5. 54 W1/ 2

2

k =

(t R

− t0 ) t0

where tR is the analyte retention time, t0 is the unretained mobile-phase peak, and W is the analyte peak width. The results are given to the instructor on the day of the experiment. Hazards Methanol is extremely flammable and toxic by inhalation, ingestion, and skin absorption. Potassium nitrate is an oxidizer and causes irritation to skin, eyes, and respiratory tract. Caffeine is harmful if swallowed and may affect the central nervous system. Results and Discussion Initially, we placed emphasis on minimal intervention and used student data as presented to us. This was to establish the unbiased status quo of the student’s ability to use each particular instrument with the associated equations and calculations and to gauge awareness of the correct magnitude of numbers and apply correct units (if applicable). We were surprised to find that more than 50% of students stated that they had difficulties supplying QC chart data within the time frame of the experiments. This appeared mainly to be due to difficulties with calculations and also a lack of experience with interpretation of raw data. To provide a pedagogic progress element, feedback was given after completion of each week’s laboratory QC analysis to discuss these problems together with laboratory errors encountered. Both staff and students perceived the HPLC experiment to be more complex than the UV­–vis experiment, requiring correctly making the mobile phase, degassing, system equilibration, and correct manual injection of sample into the system while avoiding contamination and the introduction of air bubbles. The software was also more complex than for the UV–vis experiment. Following some background research beyond the lab manual’s content, N and k were calculated manually from chromatograms. Chromatogram interpretation, more than one variable per equation, and calculations complicated the HPLC experiment. Background research by students should have shown that k is a small number and N should be of the order of thousands. In the charts individual or paired (lines link data from student pairs) data points represent the results of one experiment by the students. Where no error bars are present, no standard deviations were reported. For clarity, the mean, warning, and control lines are not always shown. UV–Vis The UV–vis results following completion of the 12week course by the two student categories in the first year, 2006–2007, are presented in Figure 1. Virtually all QC data submitted fell outside the staff-generated control limits but are generally of the correct order of magnitude. Two of the undergraduate students reported ε values close to 0 and another two 1086

A

ε / (L mol∙1 cm∙1)



students reported a value of 7300 L mol‒1 cm‒1 (not shown in Figure 1), indicative of poor knowledge regarding the expected magnitude of results. All master’s data were of the correct order of magnitude perhaps reflecting greater previous experience. Closer examination of the student reports revealed that about 10% of QC points contained calculation errors. The majority of errors was attributed to laboratory factors. When calculation errors were corrected by the staff, the results in Figure 2 were obtained. Data points are no longer paired as the staff correction of calculations reconciles the results of student pairs. With the corrected calculations, the frequency and magnitude of deviation from the mean improved marginally, but the high incidence of laboratory error remaining could not be resolved within the time frame of the class. The QC charts from the first year, 2006–2007, were constructively discussed at the end of semester and then with the second-year class, 2007–2008, to obtain feedback including means to improve quality and motivate the students to be more QC aware. Some observations generated by this discussion are as follows: Quality of results appeared to be better for the master’s class, perhaps reflecting their greater experience. The error patterns in all cases appear to be random, largely ruling out the occurrence of systematic error such as sample or instrument (lamp) degradation as major error sources. Factors contributing to random errors could include differing methods to zero the instrument and varying levels of contamination and sample handling. Despite great efforts to control these factors, error was continually encountered and week after week the set limits

B 18

18

16

16

14

14

12

12

10

10 8

8

Student

Student

Figure 1. QC charts for undergraduate (A) and master’s (B) UV–vis experiment in the first year, 2006–2007. The data were not corrected for calculation errors. The dashed lines are the 3σ control lines derived by the staff. Solid lines connect the data from student pairs.

A

ε / (L mol∙1 cm∙1)

licate. The column efficiency, N, and the retention factor, k, are calculated for each of the three chromatograms (based on the main peak). The equations used to calculate N and k are

B 18

18

16

16

14

14

12

12

10

10

8

8

Student

Student

Figure 2. QC charts for undergraduate (A) and master’s (B) UV–vis experiment in the first year, 2006–2007, with data corrected for calculation errors. The dashed lines are the 3σ control lines derived by the staff.

Journal of Chemical Education  •  Vol. 86  No. 9  September 2009  •  www.JCE.DivCHED.org  •  © Division of Chemical Education 

In the Laboratory

were exceeded. UV–vis based methods are potentially temperature sensitive and this may introduce random error. Laboratory doors opening and closing and a basic air-conditioning service could contribute to considerable temperature fluctuations and this is error that is not easily eliminated. Nevertheless, we were surprised to find this persistent high level of error throughout the semester for what we considered to be one of the more simple experimental procedures. The temperature dependence of experiments is often highlighted and examined in physical chemistry laboratory courses, but may be a secondary factor in simplified analytical experiments, which often assume room temperature conditions (typically 293 K). To remind students of this important point, temperature compensation charts or plots could be provided. Students, having recorded the actual lab temperature, can compensate for deviations from the assumed room temperature conditions by using the data in the temperature charts. By making these adjustments to observed UV–vis absorption coefficients, for example, an improvement in data quality is demonstrated. Besides learning to appreciate

B

ε / (L mol∙1 cm∙1)

A

16

14

14

13

12

12

10

11

8 6

10

Student

Student

Figure 3. QC charts for undergraduate UV–vis experiment in the second year, 2007–2008, with data corrected for calculation errors: (A) limits (dashed lines) are based on staff data and (B) limits based on student data from the first year. The lines are mean (bold), warning 2σ (…), and control 3σ (- - -).

A

B

12.5

12.5 10.0

N / 103

10.0 7.5

7.5

5.0

5.0

2.5

2.5

0.0

0.0

Student C

D

3.0

k

Student 3.0

2.5

2.5

2.0

2.0

1.5

1.5

1.0

1.0

0.5

0.5

0.0

0.0

∙0.5

∙0.5

Student

Student

Figure 4. QC charts for undergraduate (A and C) and master’s (B and D) HPLC experiment in 2006–2007. The calculation errors were not corrected. The staff-generated lines are mean (bold), warning 2σ (…), and control 3σ (- - -). Solid lines connect the data from student pairs.

the necessity to control the physical parameters of an experiment, this could also provoke thought to enrich the discussion or conclusions of a lab report where possible sources of error and suggested improvements should be listed. It must be acknowledged that the tight margins insisted on in a commercial setting may not be feasibly implemented in an undergraduate teaching situation. The instruments and facilities may be older and more basic and the students are clearly less experienced in laboratory skills. Thermostats are generally not used. Another strategy therefore, would be for the control and warning limits to be expanded for teaching purposes as other authors have suggested (6, 8). At any rate, the values obtained by students should, with reasonable care, fall within the control limit decided on, allowing satisfactory progress onto the remainder of the content of the laboratory session. For the 2007–2008 class, an alternative chart mean and limits were derived directly from the 2006–2007 student data set, to reflect student ability more closely.1 A more or less random error pattern was again obtained in 2007–2008 (Figure 3). Perhaps unsurprisingly, the QC limits obtained from previous year student data are considerably wider than staff-generated limits. It can be seen that again the majority of data points lie outside the 3σ staff-generated control limits (Figure 3A) as in 2006–2007; however, for 5 of 15 students the data fell within the limits, an improvement from 2006–2007 where no student achieved this. When the limits determined from the student data of 2006–2007 were applied, all student data complied with the QC limits comfortably (Figure 3B). In this case teaching progress was swift as troubleshooting was at a minimum. In the absence of troubleshooting necessity, students could be asked to comment in detail on possible reasons why their data failed the QC test when applied to staff limits and what actions could be taken to rectify this including a repeat of the QC analysis. Clearly the use of dual limits gives great scope for learning and discussion. Some instructors may feel that student-based limits are too generous, resulting in a lack of troubleshooting opportunities. The gathering of QC data is an ongoing process and the limits can be fine tuned further with time as more data become available resulting in a more realistic (and pedagogically useful) frequency of error. Where it is deemed too easy to pass a QC test in a learning environment, the instructor has the possibility of introducing spiked or altered QC samples, forcing more thought and action on behalf of the student, where QC tests do not pass. HPLC For the HPLC experiment, the column efficiency, N, and capacity factor, k, were examined. The HPLC control charts for 2006–2007 are shown in Figure 4. These data were not corrected for calculation errors. The first impression is that again a sizeable portion of data points are outside staff-generated control limits. For both k and N, typically half of the data points fell outside the staff-generated control limits for both the master’s and undergraduate classes. This compares to virtually 100% failure for the UV–vis experiment, which is a surprising result as we envisaged the UV–vis experiment to be among the more simple experiments. Literature data indicate that N should be in the thousands whereas k typically falls between 1 and 10. Virtually all values of k reported were of the correct order of magnitude; however, two master’s students reported negative values (impossible)

© Division of Chemical Education  •  www.JCE.DivCHED.org  •  Vol. 86  No. 9  September 2009  •  Journal of Chemical Education

1087

In the Laboratory

belying their supposedly greater experience. More than 50% of undergraduate students and about 40% of master’s students reported N values approaching 0, indicative of a lack of awareness of correct orders of magnitude. Our expectation that the master’s class would present the better results did materialize in this instance. Further inspection revealed that about 50% and 10% of all results contained calculation errors for N and k, respectively. Owing to the regular incidence of student calculation error (for all instruments), we again decided to correct these on

A

B

12.5

12.5

N / 103

10.0

10.0

7.5

7.5

5.0

5.0

2.5

2.5

0.0

0.0

Student C

D

3.0

k

Student 3.0

2.5

2.5

2.0

2.0

1.5

1.5

1.0

1.0

0.5

0.5

0.0

0.0

∙0.5

∙0.5

Student

Student

Figure 5. QC charts for undergraduate (A and C) and master’s (B and D) HPLC experiment 2006–2007. The calculation errors were corrected. The lines are mean (bold), warning 2σ (…), and control 3σ (- - -).

A

B

12.5

12.5 10.0

N / 103

10.0

7.5

7.5

5.0

5.0

2.5

2.5

0.0

0.0

∙2.5

Student

C

Student

3.0 2.5 2.0 1.5 1.0 0.5 0.0 ∙0.5 ∙1.0

3.0 2.5 2.0 1.5

k

D

1.0 0.5 0.0

∙0.5

Student

Student

Figure 6. QC charts for undergraduate HPLC experiment 2007–2008 with data corrected for calculation errors: (A and C) limits based on staff data and (B and D) limits based on student data from 2006–2007. The lines are mean (bold), warning 2σ (…), and control 3σ (- - -).

1088

behalf of students for further QC discussion. This enabled us to focus more on possible laboratory errors. When calculations were corrected the percentage of QC points for N outside the control limits dropped from over 50% for both master’s and undergraduate students to 20% and 30%, respectively (laboratory error remaining), representing a dramatic improvement (Figure 5). For k, there is a minimal improvement in results outside the control limits for undergraduates. Conversely, 100% compliance is now achieved by the master’s class, which may reflect more advanced laboratory skills. As with the UV–vis experiment, a discussion of the HPLC QC charts with students resulted in constructive observations and suggestions to improve the quality of results. Some points discussed for HPLC were as follows: In general, data points outside the control limits had high values for k and low values for N. This indicates some systematic error. These points were first discussed in the context of the equations used to generate results. The parameters affecting N and k are analyte retention time, tR ; unretained mobile-phase peak, t0; and analyte peak width, W. At elevated lab temperatures tR may decrease. The use of a column oven may reduce this type of error.2 If the proportion of organic solvent in the mobile phase is too high, tR may also be low. This can be caused by not measuring water and methanol separately. If methanol is added to water in the same graduated cylinder, more than 50% methanol may be present in the end as the overall volume shrinks upon mixing. Increased methanol content may also arise if one of the other mobile-phase mixtures required for the remainder of the laboratory is used accidentally (e.g., 60/40 methanol/water). These factors may affect tR, N, and k. The reverse of the above is true if not sufficient methanol is present. If the mobile phase is not filtered or degassed, artifacts due to air bubbles may appear on the chromatogram, which may affect the correct choice of tR and t0. The flow rate used for the QC component may not be the same as for the main analysis, therefore if a flow rate that is too fast is selected, tR and W may be lower affecting the value of N. The retention factor, k, is also affected. As several detector wavelengths are presented in the standard operating procedure, peak retention times may be similar if a wrong wavelength is unknowingly used, but peak area (and peak width W) may be greatly affected, which can have a dramatic influence on N. There may be a tendency among students not to use the recommended minimum of 15 minutes equilibration prior to sample injection. Differing values of k and N will ensue if the HPLC system is not fully equilibrated or if a drifting baseline is present. Column degradation with time tends to lower the observed N value. With experience and discussion, the above and other common sources of error can be reduced or eliminated. The QC charts obtained by undergraduates for the HPLC experiment in 2007–2008 are shown in Figure 6. As in the case of the UV–vis experiment, we made use of dual chart limits: those generated by staff and those generated from 2006–2007 student data. The data in Figure 6 show that there is total compliance with the stricter staff-generated limits for capacity factor k. For N there is good compliance with an occasional point outside the warning and control limits. This scenario represents a good balance between progressive teaching of laboratory skills and stopping for troubleshooting. On the other hand, studentgenerated limits for both N and k result in near total compliance and may be too generous. The students may be encouraged to

Journal of Chemical Education  •  Vol. 86  No. 9  September 2009  •  www.JCE.DivCHED.org  •  © Division of Chemical Education 

In the Laboratory

discuss which limit set is most indicative of true quality and contrast results with previous year’s charts. Discussion of previous years QC charts and interactive feedback in 2007–2008 contributed to the improved quality of results in 2007–2008 as compared to 2006–2007. In the 2007–2008 period, more time and emphasis were placed on the QC charts prior to the commencement of each laboratory session. We believe that it is primarily this improved training by the instructor that led to better quality results in 2007–2008. Undoubtedly, healthy competition and the students’ ambition to improve on the previous year’s performance also played a part. As more charts become available over the years this picture will become even clearer. Challenges Most authors are in agreement that among the most challenging aspects of QA–QC are calculation, statistics, and data interpretation. In theory, many of these errors could be eliminated using computer spreadsheets or laboratory instruments with appropriate in-built calculation tools. In a commercial setting, such calculations are heavily automated and are rarely left in the hands of operators. In 1990 Laquer concluded that the greatest student criticism of newly introduced QA–QC components in the lab lay with the complexity of calculations (6). Laquer suggested that computers be used to eliminate these concerns. This was an era before widespread computer use in educational facilities. Conversely, other authors believe that students should do the calculations themselves for better understanding of results obtained. The use of commercial templates was therefore not recommended by Libes (5). Interestingly, this article came a decade after the article by Laquer, well into the era of mass computer use. Time enough to realize that automation and computers may in some instances impede student learning. In our experience with QC charts, we found that calculation errors are prominent and that extra teaching resources for mathematics are a wise investment. We have found that this extra teaching of mathematical skills may be beyond the scope of a laboratory class and should be dealt with separately. For now we opt for computer-assisted calculations to eliminate this considerable source of error allowing a greater focus on laboratory errors. The demonstration and discussion of real laboratory error sources with students is perhaps more educationally useful (and cheaper!) than putting in place the means to eliminate errors. For example in a HPLC column thermostat may be essential for a pharmaceutical LC analysis, giving even and reproducible results over time, but in a teaching lab its presence might be counter productive. At the same time as the quality of data is improved through lectures and interactive student discussion, the chart limits can be fine tuned to achieve a better balance between learning progress while providing sufficient opportunity for troubleshooting. As in the real world, this dynamic process is never ending, requiring a growing archive of QC data to work from. Summary and Conclusions In subsequent years the QC chart archive will grow allowing for the optimization of the limits set at the start of each year. Limits set must not be too strict or too lenient but rather should

allow for a reasonable compromise between allowing progress with laboratory teaching, yet providing regular incidences where trouble shooting is required. The detection and rectification of errors must be possible with reasonable extra care, within the time constraints of a laboratory teaching session and respecting student ability. Because of high incidences of calculation errors where results were calculated manually, we have opted for computer-assisted calculations (templates) allowing the focus to be on laboratory error in future. In the first year we were surprised to find that virtually all QC data points were outside the control limits for the simpler UV–vis experiment for master’s and undergraduate students. We believe that feedback and student discussion prior to the start of the second year contributed to improved quality of results and better compliance for undergraduate students although most data points were still outside the staff-generated limits. Total compliance was achieved in the second year based on limits calculated using the first-year student data. For subsequent years we shall continue to use and fine tune student-generated limits for this experiment. For the HPLC experiment in the first year, total compliance for k and a 20% failure rate for N was achieved by master’s students. The undergraduate class had a 30% failure rate for N and a 35% failure rate for k. These dramatically higher failure rates perhaps reflects the less experience of the undergraduate students. A marked improvement in quality of results was again observed for the second year. Undergraduates achieved total compliance with staff-generated limits of k and improved compliance in the QC test for N, with about 7% results outside the control limits of N, requiring troubleshooting and repeat analysis. For the HPLC experiment we intend to continue using staff-generated QC chart limits. Unlike for the UV–vis experiment, the wider student data-based limits are perhaps too lenient, not providing enough incidences where student action is required. We have found that the quality of results does not necessarily follow the perceived simplicity of an experiment. Each experiment is clearly unique and the process of establishing an appropriate mean and limits varies. The goal is the same in all cases, where adequate progress learning laboratory skills must be balanced with the provision of enough realistic troubleshooting scenarios (occasional quality failures). With time and a growing archive of QC data, the instructor can get a better feel for the level of performance that can be reasonably expected from students. An additional step in the future may be that a student or a small group of students could take charge of a particular QC chart, collecting data from other groups over consecutive weeks. By communicating with teachers and colleagues, they could help manage the quality and troubleshooting for their experiment, thereby taking responsibility and developing a greater sense of involvement with the laboratory function. Interestingly the QC components in the experiments, besides motivating the individual, were found to nurture a healthy element of competition between students helping to achieve better quality results. Students were shown the real-world fact that improving quality control is a dynamic never-ending yet essential task. Both the QC tolerances and achieved quality continually vary with changing technology, market expectations, and caliber of laboratory personnel.

© Division of Chemical Education  •  www.JCE.DivCHED.org  •  Vol. 86  No. 9  September 2009  •  Journal of Chemical Education

1089

In the Laboratory

Acknowledgment The work in this article is dedicated to the memory of Deirdre Gallagher from the third-year analytical science class in Dublin City University who was among the students contributing data for this publication. Notes 1. No master’s data are shown as the course is only run every two years for master’s students. Such data will be included in future reports. 2. Thermostats are rarely used for HPLC experiments at the undergraduate level. Clearly, temperature-dependent thermodynamic and kinetic processes are also present in analytical processes. The distribution of solutes between mobile and stationary phases in chromatography, being an equilibrium situation, is a key influence on the separation of a mixture, causing variations in peak retention times and size.

Literature Cited 1. Deming, W. E. Am. Stat. 1975, 29, 146–152. 2. Analytical Chemistry; Kellner, R., Mermet, J.-M., Otto, M., Widmer, H. M., Eds.; Wiley-VCH: Weinheim, Germany, 1998.

1090

3. Christian, G. D. Analytical Chemistry; Wiley-VCH: New York, 1994. 4. Mitchell, J. A. Anal. Chem. 1947, 19, 961–967. 5. Libes, S. M. J. Chem. Educ. 1999, 76, 1642–1648. 6. Laquer, F. C. J. Chem. Educ. 1990, 67, 900–902. 7. Perone, S. P.; Englert, P.; Pesek, J.; Stone, C. J. Chem. Educ. 1993, 70, 846–846. 8. Bell, S. C.; Moore, J. J. Chem. Educ. 1998, 75, 874–877. 9. Vitha, M. F.; Carr, P. W.; Mabbott, G. A. J. Chem. Educ. 2005, 82, 901–902. 10. Carter, D. W. J. Chem. Educ. 1985, 62, 497–498. 11. Spencer, R. D. J. Chem. Educ. 1984, 61, 555–563. 12. Salzsieder, J. C. J. Chem. Educ. 1995, 72, 623. 13. Miller, J. C.; Miller, J. N. Statistics for Analytical Chemistry, 3rd ed.; Ellis Horwood Series: London, 1993. 14. Undergraduate Laboratory Manual: Analysis of Organic and Inorganic Species; Dublin City University: Dublin, 2006.

Supporting JCE Online Material

http://www.jce.divched.org/Journal/Issues/2009/Sep/abs1085.html Abstract and keywords Full text (PDF) Links to cited JCE articles

Journal of Chemical Education  •  Vol. 86  No. 9  September 2009  •  www.JCE.DivCHED.org  •  © Division of Chemical Education