fe Neor, UL Avy 4 amen i:tinoaeie amenteois — = — . B3 CIGNIFICANCE TESTING ET — 0820" Poalyfical Chemisty ‘These are statistical tests used to compare individual values or sets of values for significant differences. . AA measurement or result that appears to differ significantly fromvothers in the same set of replicates is described as an outlier. “The F-test enables the precisions of two sets of data to be compared using their variances. “The t-test is used to compare two experimental means, an experimental mean with a known value or sets of pairs of experimental values. 5 FEGUER] F-tests can be applied to several sets of data to assess and compare different sources of variability. _Reiated topic ‘Assessment of accuracy and precision (B2) Significance Significan.e tests involve a comparisun between @ calculated experimental 5 NTT TOME MEST ceo CC TE fader @ caluldled experimental data and a selected probability level that .> conclusion is correct. fpcior delerined oy They are used for sever purposes, suchas: hak. ae purpose of siqntiun® the umber of © to check individual values in a set of data for the presence of determinate “9”, vor, vous © Ye errors (bias); ek of © tocompare the precision of two or more sets of data using their variances; © “to compare the means of two or more sets of data with one another or with known values to establish levels of accuracy. ‘Tests are based on a null hypothesis ~ an assumption that there is no signifi- vant difference between the Values being compared, The hypothesis is accepted if the calculated experimental factor is less than the corresponding tabulated factor, otherwise it is rejected and there is said to be a significant-difference between the values at the selected probability level. The ‘conclusion: should always be stated clearly and unambiguously: Probability levels of 90%, 95% and 99% are generally considered appropriate sponding 10%, 5% or 1% probabilities, respectively, of the opposite conelusion ‘being valid. For example, if a test indicates that the mull hypothesis is cormsct and that there is no significant difference between two values at the 95% probe; bility level, it also allows the possibility that there is a significant difference at the 5% level Scanned. with CamScannerrificane Test oe Two oe tail Tests Tess Outliers Q-test ’- Significance testing 95 Separate tabular values for some significance test factors have been compiled for what are described as one-tailed and two-tailed tests. The exact purpose of the comparison that is to be made determines which table to use. © The one-tailed test is used EITHER to establish whether one experimental value is significantly greater than the other OR the other way around. © The two-tailed test is used to establish whether there is a significant differ- ence between the two values being compared, whether one is higher or lower than the other not being specified. “The two-tailed test is by farthe most widely used. Examples are given below. Inspection of a set of replicate measurements or results may reveal that one oF moreis considerably higher or lower than the remainder and appears tobeoutside the range expected from the inherent effects of indeterminate (random) errors alone. Such values are termed outliers, or suspect values, because it is possible that they may have a bias due to a determinate error. On occasions, the source of ‘error may already be known or itis discovered on investigation, ‘and the outlier(s) can be rejected without recourse to a statistical test. Frequently, however, this is ot the case, and a-test of significance such as'the Q-test should be applied to a puspedtvaluctoteterminzvriher chou berviected an in any further. domputations and ‘statistical assessments of the data, ‘Also know as(@ixon’s w-tes}, sis one of several that have been devised 10 test suspected pulls ih a 96 of replicates. It involves the calculation of a ratio, Qupur definédlas the'absohife “difference between a suspect value and the value Glocest-to it divided:by the spread [Link] values intheset: ~ =| suspect palue ~ nearest value|/(largest value — smallest value) Quay iS then. compared'witl'a tabuiaiee-value, Quy, at a selected level Or procs bility, usualy 90% or 95%; for a set of m values (Table 1). 1f Qnya is less than Quy, then the null [Link] there is: no. significant ifference between the suspect value and the other valties in the s .ecepiéd, and the suspect value is accepte is retained for further dlata,progessing, However, if Quy is greater than Quay then the*suspect-valu is:regarded:tas anioutlier and is rejected. A rejected value e should NOT be used in the rémaining calculations. Table 1. Critical values of Q at the 95% (P = 0.05) level for a two-tailed fest ‘Sample size Critical value 0.631 0.717 0.621 0.870 0.524 Example 1 Four replicate values were obtained for the,determination of a pesticide in river water 10.403, 0.410, 0.401, 0380 ng di? oo Scanned. with CamScannera Section B - Assessment of Inspection of the data suggests that 0.380 1g dm’ is a possible outlier. Quya =| 0.380 - 0.401]/(0.410 - 0.380) = 0.021 /0.03=0.70 Quy = 0.83 for four values at the 95% probability level oe = ~A8 Quy Teas than Qua 0380 jig dm is not an oullie? at the 95% level and “c~~should beretained.: =~=- a sz Example 2 If, in Example 1, three additional values of 0.400, 0.413 and 0.411 pg dm were incl 881g dein ctilbepossiblecutlicr eee Qeyu =| 0.380 — 0.400 /(0.413 - 0.380) = 0.020/0,083 = 0.61 ‘Qus = 0.57 for seven values at the 95% probability level Now, as Qa is greater than Quy, 0.380 pg dm” is an outlier at the 95% level and should be rejected. Note that because the three additional values are all around 0.4 ug dm”, the suspect value of 0.380 jig dm” appears even more anomalous. F-test This test is used to compare the precisions of two sets of data which may origi- nate from two analysts in theGamélaboratory, two different methods of analysis . for the same anaiyte or results from two aulterent laboratories. A statistic, F, is - can the ratio of the population variances(o,7/g,2,)or the sample variances, 5i2s,}, of the two sets of data where the larger VAF'=" Aad significantly different. eR Table 2. Critical values of F at the 95% (P = 0.05) level for a two-talled test % 5 7 8 Ye 5 7.148 6.853 66st 7 5.205 4.995 4823 9 4404 4.197 4026 = numberof degrees of freedom of the numerator. v,= number of dagrees of freedom of he senaminator Scanned. with CamScannerJ significance testing ‘The application of' a two-tailed Fest is demonstrated by the following example. Example 3. ‘A proposed new method for the determination ofits) an industrial waste ar eacheis compared with an existing method, giving the following results: Method Mean/g dm? No. of No. of degrees s/mg dev replicates ._ of frpedom oe Existing 72 . 8 7 3.98 New 7 8 7 1.50 Is there a significant difference between the precisions of the two methods? i Sia] 2 3% Font te, asor 7 °° 7 degrees of freedom for both the ‘The two-tailed tabular value for F wit numerator and the denominator is Fy, =5.00 at th€ 95% probability level greater than Fys, the sll hypothesis is relected; the two methods are Z significantly different precisions. =, test ‘This test is used to compare the experimental mepns of two sets of data or to compare the experimental mean of one set of dais: \witlf . knowin or reference . compar gealtslic fis defined, deperding on the sizcumstansss by one of three alternative equations Comparison of two experimental means, X4 and Goad, (NMA ; = : eT sow NEM) - wher¢ my, sumated standard ‘dcviaiion (Topic B2) for sets A and B,andN hors values insets A. and B respectively- IfN=M, to (N/2)", A simplified ‘version of equation (4), senrtive second term reduces Topic B2, can be used to calculate Sa 2S thereare onl smuelfov-vasor-na] [veara]} a @ the use of equation (1) may not be appropriate for the Jes of when this may be the case iy two sets of data. In some circumstances, comparison of two-experimental means. Exampl are if ‘e the amount of sample is 50 restricted as to allow only one determination by each of the two methods: ‘ « hemethods are to be compared for a series of samples containing different renels of analyte rather than replicating the-analysis at one level only: samples are tobe analyzed over a long period of time when the same expe mental conditions cannot be guaranteed, it may therefore be essential or convenient to pair the results (one from each method) and use a paired t-test where is defined by Scanned. with CamScanner: ___ Section B= Assessment ot dat" ae . fe oiixn @ % being the mean diterence between palred values and oy The estine standard deviation of the differences. Sexy ) Using the appropriate. equation, an experimental value, fay is calculated and compared with a tabulated value, ty, at a defined probability level, usually between 90 and 99%, and for N-1 degrees of freedom (equations (3) and (4)) or (N+M-2) degrees of freedom (equation (1)). If teprdS less than f,, then the null . hypothesis that there is no significant difference between the two experimental ‘means or between the experimental mean and a known value is accepted, ie. i there is no evidence of a bias. However, if tay is greater than fy,, there is a / significant difference indicating a bias. Both one-tailed and two-tailed t-tests can be used, depending on circum- REBT TST IE ET PEE TEST TRESS SP t-test equations is demonstrated by the following examples. Table 3. Critical values of tat the 95% and 99% (P = 0.05 and 0.01) levels for a two-talled test ~ Sa eee ees of freedom 95 percent level 9° percent level 3 430 ase = pa itera cg 4.03 ss ~ 3.10 2.88 % $i . — * Example 1 : #42 methods for the determination of polyaromatic hydrocarbons*in soils were z compares: by analyzing a starsia! "3 the following results: No. of degrees of fréedor UV spectrophotometry: Fluorimetry: Do the mean results for the two methods differ significantly? A Equation (2) is first used to calculate a pooled standard deviation: r ae Speed = {{o ~Ds+- vw +M- al|'- (9x03? +9x0.23)/18)% Spay = 0.267 mg kg" _ ‘Then equation (1) is used to evaluate Jays Haat) y ( aaa y = ((28.0 ~ 26.25)/0.267] x 5*=14.7 : For [Link] of freedom, the two-tailed value of fy at the 95% probability SEAT TS LTO, AME ATURE DINE TEV HED BE Scanned. with CamScanner