Anal. Chem. 1087, 59, 2818-2822
2818
and
b / m = C,V, So now the MOSA extrapolation does not yield C,, but C,V,, the analyte amount run through the test. Equation 6 shows that in this MOSA variation the slope m does not depend upon V,, but b / m does, as shown by eq 8. The C,V, here are the Q's in Figure 3C of ref 2. The fact that C,V, changes with TP (Cardone's sample size, W,,", mass meaning) does not impair the usefulness of this constant-slope MOSA variation, because C,V,, Le., Q,is not the analytical result C, (Cardone's C) but rather the mass of analyte involved in the test. The constant-slope MOSA plot is accurately described in ref 12.
FINAL REMARKS We can trace now the source of the confusion about Figure 3B in ref 2. According to the Glossary, the W's stand for either mass or concentration. This is only possible if there is a proportionality between both quantities. In that case the change of mass for concentration and vice versa, in terms of the W's, will have the same effect as the introduction of a scale factor. This means that the slope of the MOSA graph will be altered, but not so the shape of the whole plot. In the case under consideration the plot is necessarily of the constantslope type, because one of the two equivalent meanings given by Cardone to the Ws is the mass of added standard. On the other hand, mass and concentration are proportional only if the related volume is constant. The only possible constant volume in the MOSA technique, as it is sketched by Cardone ( I ) , is the final specified volume reached by diluting after spiking and, perhaps, treating with reagents, Le., the TS volume. Therefore, the concentration which is equivalent to the mass of added standard is not that in TP but that in TS, so the slope of the straight lines remains unaltered when the
Sir: It is gratifying to note that Ferrus ( I ) accepts the Youden one-sample regression plot (2) (a good convention by Ferrus since it differentiates from the Youden Plot (a twosample technique) that is discussed in ref 2) as the only procedure for the correct determination of the constant error, i.e., method blank and the method of standard additions (MOSA) as the best procedure for the in situ normalization of the proportional error ( 3 , 4 ) . However, Ferrus raises some questions about the corrigible error correction (CEC) technique. ABSCISSA UNITS OF RESPONSE FUNCTIONS One of the major underlying misconceptions throughout Ferrus's commentary concerns the abscissa units for the symbols W,, W,, W,, and Wz,uthat are used in presenting the CEC technique (3-5)in the various response functions as either a concentration or mass term, a choice that is actually normally dictated by the type of measurement system involved in the method. There is no problem with my symbolization since it represents accepted common practice. For example, Figure 5A of ref 4 is a typical generic standard response curve where S, could be the absorbances of a series of standard solutions of corresponding concentrations W, in any unit of concentration such as milligrams per liter. If one preferred, one could use the corresponding mass of the standard taken thereby plotting absorbance, SI,vs milligrams, W,. In such a case, the only 0003-2700/87/0359-2818$0 1.50/0
TP (sample size, W,+, mass meaning) varies. Putting W, (read as either added standard mass or added standard concentration in TS) as the abscissa label is not compatible with the focused MOSA plot depicted in Figure 3B of ref 2. The focused plot holds only if the label of the abscissas is the added standard concentration referred to TP. As an example of MOSA plot with added standard concentration in TS on abscissas we have ref 13. ACKNOWLEDGMENT The author thanks Mario J. Cardone for indicating to him ref 4 and 6 and for sending a copy of the former. LITERATURE CITED (1) Cardone, M. J. Anal. Chem. lB86, 58, 433-438. (2) Cardone. M. J. Anal. Chem. 1966, 58. 438-445. (3) Cardone, M. J. J. ASSOC.Off. Anal. Chem. lS83, 66, 1283-1294. g Anelytfccel Chemlsby; IUPAG (4) Horwitz. W. Nomendehve of S ~ n p / hh Comissbn, Vol. 3, Analytical Nomenclature. Provisional Proposal, 14th Draft, 1986, 16 May. (5) Ferrlis, R.; Torrades, F., unpublished work, Terrassa, 1986. (6) Bader, M. J . Chsm. €duc. 1980, 57. 703-706. (7) Foster, J. S.; Langstroth. G. 0.; McRae. D. R. Roc. R . SOC.London, A 1935, 153, 141-152. (8) DIN 38406, Besflmmung von Cadmium; Beuth: Berlin, 1980; Teil 19. (9) DIN 38406, BesUmmung yon Blel; Beuth: Berlin, f08l; Teil 8. 10) Caulcutt, R.; Boddy, R. Sfet/st/cs for Analytlcel C h m l s b ; Chapman and Hall: London, 1983; Chapter 13. 11) Foster, J. S.; Horton, C. A. Roc. R . SOC. London B 1937, 123, 422-430. 12) Kolthoff. I . M.; Sandell, E. B.; Meehan, E. J.; Bruckensteln, S. Ouantltatlve Chedcal Anaksls, 4th ed.;Macmlllan: London, 1969; Chapter 15. 13) DIN 38406, Bestlmmung von Chrom; Beuth: Berlin, 1085; Tell 10.
Ricard Ferrtis Escola dEnginyers Universitat PoliGcnica de Catalunya 08222 Terrassa, Spain
RECEIVED for review June 9,1986. Accepted July 22,1987.
effect would be a magnitude change in the value of the slope that results. Since, for example, in spectrophotometry, one reads an absorbance, the functionality A = abc makes it simpler to plot such standard curves with the concentration term, c, rather than a mass unit as the abscissa. Now looking at the MOSA part of Figure 5B from ref 4, any units of concentration or mass or a dimensionless term may be used for W, (added). If a concentration term were selected for the W, units in the standard curve, it would be simplest to select the same W , units in the MOSA; however, it is not mandatory but would eliminate the need for an appropriate scale factor. Obviously, it is most convenient to be consistent in the units selected for the standard curve, Youden one-sample curve, and the MOSA curve, either a concentration, mass, or dimensionless term throughout. No one in the model problems survey had any difficulties with these basic fundamentals (6). It was also uniformly understood by the respondents that the abscissa units were those for the aliquot (treated sample, TS as defined by Horwitz (7)). It was quite clear that the concentration of the analyte, W, in the original sample (test portion, TP (7)) when read (calculated) from the standardization plots (in W, units) becomes W,/ W , for the standard curve and single-point-ratio (SPR) techniques or W,/ Wz,ufor the MOSA technique. However, an explicit example using data from the literature follows ( 3 ) . The data for CPM (from Tables 5, 6, and 7 of ref 3) used a concentration term for the respective response 0 1987 American Chemical Soclety
ANALYTICAL CHEMISTRY, VOL. 59, NO. 23, DECEMBER 1, 1987
functions. Therefore for this illustration the data is converted to a mass term throughout. In the HPLC procedure, the standard weights, the sample weights, and the MOSA sample weights were diluted to 200 mL and a 10-pL injection volume was used. Regression of the recalculated X,Y data for the respective response curves yields (step 3) standard curve Y(PH) = 55.623X (ng) - 543
MOSA plot varies with the unspiked sample weight, W,,u, whereas in the MOSA plot with W, units, either concentration or mass, the slope does not vary with Wz,u. See Figure 5B in ref 4.) Again, using the actual example data (3) as above, from the regression of the recalculated X,Y data (Table 7 ) ,we obtain the MOSA response curves 50% level Y(PH) = 11342184X (mg/g) 4210 75% level
Youden curve Y(PH) = 552.8X (pg) - 874 MOSA curves 50% level Y(PH) = 59.694X (ng)
+ 4210
75% level Y(PH) = 58.172X (ng)
+ 6981
In changing the units from mg of CPM/mL to nanograms (ng) injected, the factor change simply results in a corresponding smaller slope, as is also seen in changing the units from milligrams of sample per milliliter to micrograms bg) injected. The intercepts are unchanged; hence the Youden blank (YB) is unchanged. The proportional error factor (PEF), P is the ratio, 58.933 ng-' (av)/55.623 ng-' = 1.06 (step 4b), the same as before. For the slope ratio assay calculation (step 51, we have mg of CPM/g of sample = 552.8 pg-'/58.933 ng-' and the value of 9.38 ng/pg (mg/g) is the same as before. For the first sample assay calculation listed in Table 6, the standard curve calculation becomes mg of CPM/g, corr = [[(2145 543)/55.623](200) X (lOOO)]/(lo)( lo6)(1.06)(0.0968)
+
where the term 200 is the milliliter sample dilution, 1000 converts milliliters to microliters, 10 is the microliter injection volume, lo6 converts nanograms to grams and 1.06 is the P fador. The result, 9.42, is the same as obtained in the previous illustrative example (steps 6 and 7). In the procedure for such a calculation, all of the above fadors would be combined into a single scale factor (SF), (200)(1000)/107 = 0.02. (More convenient units other than nanograms and micrograms could have been chosen but these were chosen deliberately to demonstrate that the choices only affect the factors that result. None of these factor and unit selection options are of any concern to the CEC technique and the experienced analyst chooses units such that the simplest factors result.) For the MOSA assay value, we use (step 9) the corrected MOSA intercept value, a (= A - TYB), so that mg of CPM/g of sample = [(4210+ 874)/58.933]/(0.02)(0.190) and the value for the 50% spike level, 9.08, is the same as before. The 75% spike level MOSA calculation follows accordingly with the same result as before, 9.35. I have now demonstrated that the response functions, standard, Youden, and MOSA, can be configured with W, units representing either a concentration or mass term for the aliquot or, generally, the treated sample, TS. These are the conventional practices. However, Ferrus insists that for the MOSA calculation, the units for the abscissa must be the spiked standard concentration of the analyte in the original sample or test portion, TP, i.e., W,/ WZ+.This process is merely a mathematical option since one simply divides all spike additions by the sample weight and, thence, one obtains the concentration of the analyte in the original sample directly from the MOSA plot. Again, only the slope of the MOSA curve is changed; the intercept, A, remains the same. (I will deal later with the fact that the slope of such a configured
2810
+ Y(PH) = 16578887X (mg/g) + 6981
The MOSA calculation then is mg of CPM/g of sample = [(4210
+ 875)/11342184](20 000)
or 8.96 for the 50% level and, correspondingly, 9.48 for the 75% level. The 20000 term is the scale factor, [(200)(1000)]/10 where 200 is the milliliter sample volume, lo00 converts milliliters to microliters, and the 10 is the microliter injection volume. We note that the values 8.96 and 9.48 mg/g are not exactly the same as those from the W, units calculations but this is because in those the average slope for the two levels was used. It is a simple matter to recalculate those values with the actual slopes for the levels and then the values match exactly.
PROPORTIONAL ERROR FACTOR The outcome of Ferrus' discussion of the proportional error factor, P, is that he would give it up. In doing so, Ferrus' comprehension of the CEC technique and its purposes as detailed in ref 3-5 is somewhat lacking. In ref 7, a key question asked was "...why do analysis results obtained from a standard response curve too often not agree with the results from the MOSA technique...". Ferrus agrees that the MOSA is the best technique for the in situ normalization of the proportional error. Normalization occurs because the same procedural operation is performed on the unspiked and the spiked samples so that a constant (reproducible) percentage bias is introduced on the sample across the dynamic range. If there is an interaction effect between the matrix and the analyte, the slope of the MOSA curve will be different, either higher or lower than the slope of the standard curve when both response functions are in the same units. Thus, the ratio of the slope of the MOSA curve, mMto that of the slope of the standard curve, m,is the determination of the PEF, which is necessary to correct standard curve results since, in this process, these are interpolations from a matrixless reference curve. We must assume, since there is no other recourse, that the variables remain the same in the response curve measurements. This is a major reason why the CEC technique demands randomized data sets for the response curves with a method that has sufficient ruggedness so that the procedural variables do not significantly influence the slopes and, hence, the final results. Under these defined conditions, the only source of proportional error is indeed, the matrix. Measured as described, the PEF may be greater or lesser than unity (P = 1,no matrix effect). (We must remember that these slope measurements are subject to random errors and, as discussed in ref 4, also to the vagaries of the negative correlation effect, the interplay between the slope and intercept.) Values of P between, say, 0.98 and 1.02 would be considered as a constant, Le., a mathematical parameter for the data set and the deviation from unity would not warrant a conclusion that proportional error existed. The exact value of the P constant, however, would be carried throughout all calculations just as we do with calibration constants in conventional practice. Since proportional and constant error sources interact with each other, rigorous comparison of the determined P value against a stoichiometric value as Ferrus suggests is hazardous,
2820
ANALYTICAL CHEMISTRY, VOL. 59, NO. 23, DECEMBER 1, 1987
especially the assumption that the application of the P value in any one case deteriorates a presumed analytical result. Proportional and constant errors can reinforce each other, both in either a plus or minus direction, or they may cancel each other. Agreement with expected values, even stoichiometric values does not prove a result is correct and such an expectancy frequently would demand the favorable cancellation of errors. It is well known that fortuitous cancellation of undelineated errors is bad analytical practice, whereas reliance on the CEC technique would be good scientific practice that consistently utilizes all the observed data irrespective of preformed conclusions.
MOSA PLOT A MOSA plot is a MOSA plot regardless of what the units for the abscissa are, as has been shown above, or whether the response (ordinate) is (1)the observed uncorrected response, (2) the response partially corrected for the constant error (SB) (but erroneously as explained in ref 4 when the partial correction is a directly measured method blank, as is the case in the work of Foster et al. (ref 7 in ref 1) cited by Ferrus), or (3) the totally corrected response obtained by using the total Youden blank (TYB) value. In each case, whether the abscka units are W , (in TS) or W,/ W,,” (in TP), the calculations utilize the MOSA intercept value as A - TYB for case 1, A - YB for case 2, and A for case 3. These considerations all follow from the disclosures in ref 3 and 4, although only case 1 was formally treated. FUNCTIONALITY OF THE MOSA Incorrect Model. In ref 4, p 441, the statement was made referring to the MOSA “Although it can be and is used as a freestanding technique, ita functionality is not freestanding.” The functionality referred to is the current, universally accepted presentation of the MOSA by all pre-CEC writers, which is represented in Figure 3A from ref 4. The writer, because of the challenge by Ferrus has resurveyed all the available current analytical textbooks, some older texts, reference works, and the MOSA literature and without exception, the graphical presentation is that in Figure 3A, noting that the choices in units, either of the ordinate or of the abscissa is irrelevant as has thus far been shown. Two citations suffice, the textbook by Kolthoff et al. (ref 12 in ref 1)and the paper by Bader (ref 6 in ref 1).The functionality depicted in Figure 3A is wrong because the model in no way provides for a blank correction. Ferrus protests that the A’s must be substituted with a’s and I agree, but such a correction does not derive from the model-it derives from the CEC (specifically the Youden one-sample plot) that Ferrus accepts. Ferrus quotes German Standards (refs 8,9,and 13 in ref 1) (note: these are Figure 3A models) as examples of the application of a blank correction. In the German Standards model, the extrapolation is not to the ordinate = zero but to the ordinate = blank, as Ferrus has pointed out. However, there is no functional relationship involved in the locus of the blank point so that its position is arbitrary and subjective. Its numerical value stems from a separately, directly measured experiment and is not, as discussed in ref 4, the true blank although the operation is quite an innovative means of depicting a method blank or system blank correction. A few authors ( 4 9 ) have called, from subjective considerations, for a system blank correction to the ordinate response but to this writer’s knowledge, no writer prior to the CEC articles has called for a true blank correction, Le., a blank determined in the presence of both the analyte and actual sample matrix. We have shown that calculations based on the Figure 3A model, if not corrected for the true blank, TYB, yield incorrect results, as the 31 correspondents to the model problems found (6) and as can easily be shown on the actual problem used above.
The nine MOSA cases discussed by Bader are all applications of the Figure 3A model and will all produce correct results only if the ordinate response R, (Bader’s notation) is corrected for the true blank, TYB. Bader’s case 1 is the one used in the CEC for reasons already discussed. All of the Bader’s cases are explicitly for the purpose of obtaining an assay result with the MOSA as a free-standing technique despite Ferrus’ protest otherwise. Bader’s innovative application of a dimensionless abscissa unit is a further illustration that the units used in any specific MOSA are irrelevant. All Bader’s cases apply to either a neat liquid sample, TP, or to a treated sample, TS, but this again is of no concern. As a free-standingtechnique, Bader’s cases are interesting and valid but have no other pertinency to the CEC presentation. Ferrus makes reference to a single addition of standard as does Bader (case 4). It is quite important to distinguish between two single standard addition techniques, one, where the single addition is simply a single point of the usual multiple additions (for situations of proven linearity) as against, two, the single addition technique of Dean (ref 59 in ref 2). The former was used by Foster et al. (ref 7 in ref 1). Only the latter is a true single addition technique that has value as a diagnostic device for the detection of proportional error (2). The discussion in ref 4 relative to the model leading to a negative and physically meaningless result still stands. The innovative but unnecessary mathematical “negative”standard addition device that Ferrus uses still leads to his eq 5, a negative value despite the rationale. This negative addition concept and the use of Bader’s equations are unnecessary since all of the concepts and derivations desired flow directly from the three response functions, with the most important derivations already reported in ref 5. Other derivations, to be demonstrated in the next section are also fundamental and not circuitous, i.e., based on identities between Bader’s expressions and those in ref 3 and 4. Correct Model. In the correct model, shown in ref 4, Figure 4 and 5B, the slope of the functionality is constant; hence the slope is independent of the size sample taken for spiking and the observed intercept A varies with sample size. Mathematically, these relationships follow directly from the MOSA true functionality as follows: but
S,’ = (W,)(mM) + TYB S, = S,’ = A
and
A-TYB=a
hence
a = (Wr)(mM)
(1)
(2)
or mM = a / W, (3) Equation 3 shows that the MOSA slope is a constant for a MOSA with W, units (in TS) (Ferrus calls this the constant-slope MOSA) because a is the response corrected for constant error that is proportional to W,, so the term a / W , is a constant (cf. Ferrus’s eq 6). Further, rearranging eq 3
W, = a/mM and
(4)
w, = w, dividing by Wz,u we get W,/ W,+ = a / ( m ~ ) ( W ~ , ~ )
(5)
which is the correct MOSA calculation expression (ref 5, eq 17 and expression IV, Figure 1, ref 4). By definition wx/wz,u
=
c
substituting into eq 5 and rearranging a = (C)(m,)( WZJ
ANALYTICAL CHEMISTRY, VOL. 59, NO. 23, DECEMBER 1, 1987
2821
Table I. MOSA Data for Problem 1'
1 sx
I
Wrl wz,
s,'
Youden One-Sample
x
Y
Si
slope
Set lAb 0.8000 1.0000 1.4870 1.6200
0 2.540 8.724 10.413
0.0000 0.0407 0.1396 0.1667
mMA
4.920 .
Set lBc 0
1.0000
0.0000
3.1429 5.7143 7.8730
1.2475 1.4500 1.6200
0.0380 0.0690 0.0951
'Reference 6,Table I. Figure 1.
ref 4-with
Overall MOSA configuration as shown in Figure
58,
from
w,/w,,, units.
Equation 6 shows that the intercept A , or corrected intercept a, varies with the analyte concentration C in the sample, TP, the slope of the function, mM, and the sample size Wz,utaken for spiking, as is evident by inspection of Figure 5B and as Ferrus concludes from his identical eq 8. Ferrus misinterprets the meaning of the CxVx(his eq 8) term, equivalent to the (C)(WZJ term in eq 6. Either term is the amount in W, units (either concentration or mass) of the analyte in the sample size taken, i.e., the W , in the W, sample, the portion of the functionality represented by the overall abscissa scale, W , (total) in Figure 5B. The MOSA plot is still one with a W, abscissa scale (in TS) and eq 5 for the assay calculation still obtains. Ferrus inconsistently rejects Figure 5B as the correct model. Inconsistently, because one cannot reject the whole, i.e., Figure 5B and accept a central part, the Youden one-sample plot as the correct procedure for the determination of the constant error (true sample blank), the procedure that is the basis for the substitution of the A's with a's, both at the same time. It is the left-hand side of Figure 5B, the Youden one-sample plot, that is the portion in question where the analyte-tomatrix ratio is constant, contrary to the situation with real samples, whereas the analyte-to-matrix ratio increases in the righ-hand, MOSA portion. Ferrus believes this difference invalidates Figure 5B. This was mentioned earlier in the discussion of the abscissa units section. The writer has been unable, since before the manuscripts for ref 2 and 3 were being prepared, to show, fundamentally, that the difference mentioned above is immaterial to the validity of the Youden one-sample plot. However, by the use of a model configured with W,/ W,+ units (Ferrus calls this a focused MOSA plot), a fundamental proof can now be presented. We return to the model data for problem 1that included MOSA data sets for spiked samples at two levels (6). With this configuration, shown in Figure 1,the slopes of the MOSA do vary with sample size (as we saw with the actual data example earlier) and as was pointed out by Ferrus. The data, calculated from the data in Table I, ref 6 is shown in Table I. The slopes for the data sets 1A and 1B are obtained from linear regression analysis. From the overall functionality (line A or line B) we have
S,' = ( ~ M ) ( W ~ / W+~TYB ,~)
(7)
and as before (see eq 2)) we get
a = (mM)(wr/wz,u)
(8)
m~ = (a/Wr)(W2,J
(9)
rearranging
mMB
6.521
= 62.4746. W,," = 82.7915.
and as before (see eq 3))the term a / W, is a constant, so the slope of the MOSA configured with W,/W,,u units is dependent upon the sample size as the data in Table I and Figure 1 show and as was demonstrated with the actual data example earlier. Ferrus, of course, demonstrates the same point circuitously with his eq 3. Also, from eq 8, we see that the intercept A , or the corrected intercept a , is proportional to the slope of the MOSA function and to the concentration of the analyte in the sample, TP, but is independent of the size sample, Wz,u,taken for spiking. Ferrus observes the latter point from his eq 2. TYB Intersection Point. If now we solve the fundamental expressions for both MOSA sample levels, lines A and B, simultaneously, we can calculate the coordinates for the intersection point line B
s,'
line A
S,' = (mid(W,/w,,J
+AB + AA
(11)
S,' = 6.520(Wr/W,,u)
+ 1.000
(12)
S,' = 4.920( W,/ W,,J
+ 0.800
(13)
= (mMB)(wr/wz,u)
(10)
which are, from the data in Table I
and
respectively, solving, the X coordinate is Wr/ W2,u= -0.125 substituting the X value into eq 12 or 13, we get the Y value
S,' = 0.185 We recognize the coordinates for the X,Y intersection point, -0.125 W,/ WZ+ and 0.185s: (or S,) as the correct assay value (negative in sign because of the extrapolation as required by the model) and the TYB value. Thus, this extrapolation through the Youden one-sample linear dynamic range results in exactly the same total Youden blank, TYB, value as does the direct Youden one-sample regression analysis of the raw sample data. Therefore, the Youden one-sample procedure is a valid procedure for the determination of the TYB. It is evident that if the TYB were subtracted from the S,' responses beforehand, the intersection point would fall on the abscissa (Y = 0), resulting in the correct assay value of -0.125. However, if only a single level MOSA function, line A or B, is extrapolated to the abscissa, as the model Figure 3A requires the erroneous values of p = -0.1626 and q = -0.1534 result, exactly the incorrect answers found by the survey respondents in ref 6, Table IV, Code A5.2. The correct Figure 5B MOSA configuration is evident in Figure 1and the correct calculation
2822
Anal. Chem. 1987, 59,2822-2827
for a single level has already been given. It should be noted that a working technique based on the spiking of two sample levels and solving the resulting MOSA functions configured with W,/ units simultaneously could be used as an assay procedure but reflection will indicate that there would be no real time or effort advantages.
SUMMATION The articles on the CEC technique have all utilized the MOSA configured with W , units and it has been shown that the W , abscissa units can be a concentration, mass, or dimensionless term. Also, the ordinate response can be uncorrected for constant error, partially corrected, or fully corrected. Proper calculation based on the published concepts can then be made for each respective situation. Ferrus makes no mistakes in his circuitous and unnecesary mathematical treatments but he unfortunately erroneously concludes that the only correct MOSA must be configured with W,/Wz,u units. Also, he loses sight of the CEC objectives and again erroneously concludes that the P factor is either questionable or of no value. It has been shown that a MOSA configured with W,/ Wz,u units is just another MOSA that could be useful in specific situations. Any stand-alone MOSA techniques such as those discussed by Bader, properly corrected for the TYB constant error, are acceptable and valid. However, for the purposes of the CEC technique, the treatment in the CEC articles stands. Using a MOSA configured with W,/ Wz,uunits, the writer has developed a fundamental proof that the Youden one-sample curve is a valid procedure for the determination of the true blank, defined in the CEC as the total Youden
blank, TYB. This proof was an omission in the CEC presentations, as pointed out by Ferrus, that the writer had heretofore been unable to resolve. With this proof in place, the true functionality of the MOSA, as shown in either Figure 5B of ref 4 or Figure 1, is firmly established.
ACKNOWLEDGMENT I am grateful to Ricard Ferrfis for his dedicated analysis of the CEC articles and for this communication which has required a strengthening and extension of the CEC concepts. In particular, for the idea that permitted the proof of a vexing problem. LITERATURE CITED (1) (2) (3) (4) (5) (6) (7) (8) (9)
Ferriis, Ricard Anal. Chem., preceding paper In this issue. Cardone, M. J. J . Assoc. Off. Anal. Chem. 1983, 66, 1257-1282. Cardone, M. J. J . Assoc. Off. Anal. Chem. 1983, 66, 1283-1294. Cardone, M. J. Anal. Chem. 1988, 58, 438-445. Cardone, M. J.; Lehman, J. G. J . Assoc. Off. Anal. Chem. 1985, 68, 199-202. Cardone, M. J. Anal. Chem. 1988, 58, 433-438. Horwitz, W. Nomenclature of Sampling in Analytical Chemistry; IUPAC Commission, Vol. 3, Analytical Nomenclature, Provisional Proposal, 14th Draft, 1986, 16 May. Larsen, I. L.: Hartmann, N. A.; Wagner, J. J. Anal. Chem. 1973, 45. 1511-1513. Rothe, C. F.; Sapirstein, L. A. Am. J . Clin. Pathol. 1955, 25, 1076- 1089.
Mario J. Cardone Chemistry Department State University of New York Binghamton, New York 13901 RECEIVED for review December 1, 1986. Accepted July 22, 1987.
Room-Temperature Phosphorimetry of Carbaryl in Low-Background Paper Sir: Carbaryl (1-naphthyl N-methylcarbamate) is widely used as an insecticide throughout the world and is frequently determined by use of a molecular absorption spectrophotometric method, based on the reaction of 1-naphthol-the carbaryl hydrolysis product when in alkaline medium-with p-nitrobenzenediazonium tetrafluoroborate (1). Room-temperature spectrofuorometry has also been suggested and applied to "real-life" samples by several authors (2-9). Lowtemperature (77 K) phosphorimetry (LTP) was first proposed by Moye and Winefordner (IO),and very recently solid-surface room-temperature phosphorimetry (SSRTP) using either common filter paper (11, 12) or ion-exchange filter papers (Whatman DE-81 and P-81) was also examined (13). Kirkbright and Shaw (11)examined the SSRTP of carbaryl in Titer paper (Whatman No. 30) using carbaryl solutions in 1 M NaI-1 M NaOH, in aqueous ethanol (1:l). A 3.8-fold enhancement was observed after the heavy atom treatment. There is no information in the paper if the addition of NaOH produced modification in the spectrum as a result of carbaryl hydrolysis. The limit of detection and the lifetime of carbaryl were not determined. More recently, Vanelli and Schulman (12) investigated the SSRTP of several pesticides, including carbaryl, using different solvents (such as acetonitrile, 0.1 M NaOH in aqueous methanol, and 0.1 M HC1 in aqueous methanol). Also, the presence of sodium acetate and heavy atom salts (NaI, lead acetate, thallium fluoride) in Whatman No. 42 filter paper 0003-2700/87/0359-2822$01.50/0
was examined. The best result was found with acetonitrile and 1M NaI + 1 M sodium acetate. A limit of quantifkation (defined as the spotted mass equivalent to a signal which is twice the blank signal) of 1000 pg was found (12). Su and co-workers (13) determined the SSRTP analytical figures of merit of six pesticides, including carbaryl, using ion-exchange filter papers. The combination of DE-81 (an anionic exchange type) and iodide produced the best limit of detection (800 pg) and a linear dynamic range larger than 200. SSRTP is an analytical technique that has been growing in the recent years as a promising and simple tool for the trace level detection and determination of several organic compounds such as pharmaceutical products, pesticides, polycyclic aromatic hydrocarbons, etc. Several reviews and books have been recently written dealing with the theoretical and practical aspects of this interesting analytical technique (14-1 7). Although much simpler to work with than low-temperature phosphorimetry,the broad background signal, which is usually present in the substrates (in the wavelength range of ca. 400-600 nm), has been one of the limitations of the SSRTP technique for trace analysis application. To avoid this problem to a certain degree, a series of paper substrates with low background signal was examined by Paynter, Wellons, and Winefordner (18). Lignin, hemicellulose, or even transition metals traces have been suggested as possible sources of the background signal present in cellulose substrates (19). Several attempts have been made to decrease this signal, to some 0 1987 American Chemical Society