cluded from the measured value through calibration with a similar standard. This aspect of pH measurement is treated in detail in [8].

Having established the magnitude of the influence of the various factors on measured pH, we combined the RT data with calibration data obtained by using the strong acid reference solution SA83 1000. These data are summarized in the table 6. The normalized pH results are in reasonable agreement. The deviations given in the last column of table 6 are generally small. The largest deviation of +0.063 pH is approximately equal to two standard deviations for the between the laboratory component of variability observed in our interlaboratory study [8].

be tolerated in accurate pH measurements, as these salts change not only the mean activity coefficients of solutions, but also unpredictably change the electrode behavior and hence the measured pH. The ruggedness tests have revealed, in addition to the above main effects, three 2FI. These 2FI could not be determined from an experiment which changed only one factor at a time.

References

[1] Paule, R. C.; G. Marinenko, M. Knoerdel, and W. F. Koch, Ruggedness Testing-Part I: Ignoring Interactions, J. Res. Natl. Bur. Stand. 91-1, 3-8 (1986).

[2] Paule, R. C.; G. Marinenko, M. Knoerdel, and W. F. Koch, Ruggedness Testing-Part II: Recognizing Interactions, J. Res. Natl. Bur. Stand. 91-1, 9-16 (1986).

[3] Hamer, W. J., and Y. C. Wu, Osmotic coefficients and mean activity coefficients of uni-univalent electrolytes in water at 25 °C, J. Phys. Chem. Ref. Data, 1(4), 1047-1099 (1972). [4] Koch, W. F.; G. Marinenko and J. W. Stolz, Simulated Precipitation Reference Materials IV, NBSIR 82-2581 (1982). [5] Koch, W. F., and G. Marinenko, Simulated Precipitation Reference Materials: Measurement of pH and Acidity, Sampling and Analysis of Rain, S. A. Campbell, ed., ASTM STP 823, 10-17 (1983).

[6] Covington, A. K., and P. D. Whalley, Improvements in the precision of pH measurements, Anal. Chim. Acta 169, 221-229 (1985).

[7] Davison, W., and C. Woof, Performance Tests for the Measurement of pH with Glass Electrodes in Low Ionic Strength Solutions Including Natural Waters, Anal. Chem. 57, No. 13, 2567-2570 (Nov. 1985).

[8] Koch, W. F.; G. Marinenko and R. C. Paule, An Interlaboratory Test of pH Measurements in Rainwater, J. Res. Natl. Bur. Stand. 91-1, 23–32 (1986).

The rationale behind the sample selection and the sequence of measurements was as follows: Solution A was intended as a check on the calibration of each participant's pH measurement system, since the composition of this solution was nearly identical to the one provided for calibration.

Solution B was prepared by dilution of high-purity hydrochloric acid. The pH of the solution was determined using hydrogen gas electrodes in cells without liquid junction. The apparatus and calculations were identical to those used in the certification of NBS buffers [3,4]. The pH value was confirmed through calculations based on independent measurements using highprecision coulometry and ion chromatography. Solution B served as the reference or normalizing solution for this test. Solution C also was prepared by dilution of hydrochloric acid and was intended as a low ionic strength acidic solution in a simple matrix. Solutions D and E were more complex acidic matrices composed of several anions and cations simulating the composition of rainwater. Solution F was a repeat of Solution B to check for instrument drift, and possible hysteresis of the

liquid junction. Solution G was a repeat of C to check instrument drift. Solution H was a repeat of A to check calibration drift and hysteresis. Participants were asked to repeat the sequence in order to establish the precision. of the measurements. Participants were also asked to supply information as to types of electrodes and standards used.

Results and Discussion

Because the samples were to be sent to participants via the U.S. Postal System during the winter months, it seemed advisable to test whether freezing and thawing the solutions would affect the pH values. A set of samples was frozen in a laboratory freezer for 24 hours and then thawed. No significant differences in pH values were observed with this set versus a control set which had not been frozen. Even after several freeze-thaw cycles, there were no significant differences noted.

Full cooperation was obtained from all participants with regard to quality, completeness, and timeliness of response. All of the measured pH values are shown in table 2, including the buffer standards used by the participants (STD 1 and STD 2). The average pH values for measurement trials 1 and 2 for each solution by each participant are shown in table 3. The laboratory number does not correspond to the alphabetical listing of laboratories in Appendix A. The data and general results will first be presented graphically since this is easier to assimilate, and will then be described in a more quantitative fashion through the use of statistical analyses of variance. As a general rule in routine pH measurements using combination electrodes, an uncertainty of ±0.02 pH units is to be expected. This permissable variability will be applied in the following discussion of the data.

With the exception of laboratory 3, all participants reproduced the value for solution A, the standard buffer solution of potassium acid phthalate. Subsequent to the test, laboratory 3 discovered that the commercial buffer that they were using was biased. They have since corrected this problem. This indicates that good calibration practices were in effect and that the instruments were in a state of control. It also suggests that if accurate results are required, then standardization should be done using quality reference buffers, such as Standard Reference Materials supplied by the National Bureau of Standards. The values for solution H (which is identical to A) scatter a little more, with three labs (7, 8, and 17) being out of compliance. However, this is most likely due to hysteresis at the liquid junction. When comparing the values for A and H, and C with G, no trends in instrument drift are apparent. Furthermore, hysteresis of the electrodes, when subjected to buffers and low ionic strength solutions, should not be a major problem if, as for this

exercise, a strict measurement protocol is established and followed. In fact, with the exceptions of laboratories 3, 7, and 8, excellent repeatability was observed for the duplicate solutions (A/H, B/F, and D/G). It can be concluded that, with few exceptions, the precision of pH measurements made by a single operator, using a single set of electrodes, is within the permissable variability of ±0.02 pH, and is not a matter of great concern. The data reported in table 2 for the duplicate measurements also shows good within-laboratory precision. However, this conclusion can be misleading because of the constraints of a single laboratory, a single operator, and a single set of electrodes, and because of the inattention to the matter of accuracy. Figure 1 shows the wide range of pH values obtained by the participants for solution B and their relationship to the true pH of this solution. The lower case letters, s through z, above each data point serve to categorize the electrodes used by the participants according to type and manufacturer. The range of values spans more than 0.3 pH units. Only two laboratories are within the permissable variability of 0.02 pH. Most laboratories are biased several hundredths of a unit high. A few laboratories are biased low and once again laboratories 3, 7, and 8 stand out in this regard. It is clear, that in spite of excellent withinlaboratory precision, between-laboratory precision is very poor, and the measurements show a great deal of systematic bias. The high degree of scatter is confirmed. by the values reported for solutions C, D, and E, as shown in figure 2. The solid line is the mean value of the 17 measurements and should be viewed as a point of reference only, not as the true value. Unlike B, the true pH values for these solutions have not been determined.

Exhibited in figure 3 are the results of normalizing the pH values of C, D, and E, with respect to B. In essence,

solution B is used as the calibration standard rather than the customary buffer solution. The vast improvement in terms of decreased scatter is obvious. Most values are now within ±0.02 pH units of the mean. (The extremely high value for E by laboratory 7 must be considered an outlier and is discussed in more detail below.)

As noted, the participants included in their reports the make, model, and type of electrodes used in their measurements. Eight distinct types of electrodes were used in the study. These have been identified in code on figures 1-3 with the lower case letters s through z. An association of the type of electrode with deviations from either the true value or the mean is exhibited by only two of the eight types, namely, "s" and "w." Type “s” is a combination electrode in which the liquid junction of the reference electrode is somewhat larger in area than the other electrodes used in the test. This apparently results in a larger variability in residual liquid junction potential. However, when normalized with solution. B, the values obtained by electrode "s" conform quite well. Type "w", used by laboratory 7, is a combination electrode which incorporates a gel-filled reference electrode. During this test, this electrode behaved quite erratically in low ionic strength solutions, as evidenced by its non-conformity even after the normalization process.

Electrode types "y" and "z" had open-junction reference electrodes; that is, the junction between the filling solution of the reference electrode and the sample solution was formed at a capillary tip, rather than across a ceramic or fiber frit as is customary in combination electrodes. In the low ionic strength solutions tested in this exercise there was no apparent advantage to this type of junction, although it has proven its worth in other types. of solutions. Correlations with the other requested information (Appendix D) were not readily apparent.