You are on page 1of 10
used to estimate the uncertainty in the value of the dependent variable read from a graph corresponding to some chosen known value of independent variable. For a linear graph representing relation y = a + bx, the slope b and the worst case error Ab in its value are given by Ay tay, Auytde and ab = = | b | respectively. The worst case error evaluated above by type B method may be considered to give the limits of rectangular distribution. In that case the standard uncertainty would be given by u(b) =4b/V3. But it appears more appropriate to consider the distribution to be normal. In that case we may assume the worst case value as giving the 95% confidence level and take the standard uncertainty u(b) =b/2. The intercept of the linear graph is also interpreted as some result. To estimate the uncertainty in the value of the intercept we can use the relation of straight line viz., y= a+ bx. Then Aa = Ay + bAx + xAb. The intercept b is read on y-axis when x = 0. Therefore, Aa = Ay + bAx. This shows that there are two factors contributing to the uncertainty of the value of the intercept: the uncertainty in reading the position of the point on the y-axis and the uncertainty resulting due to the slope multiplied by the uncertainty in Ax. In the same way the worst case error in the intercept on x-axis is given by Aa’ = Ax + Ay/b. The expressions giving the respective standard uncertainties of y and x intercepts would be u(a) = (uly) + b’u(x)'}"” and u(a’) = (u(x)? + uly)?/ b’)°” respectively. From the above two expressions we see that the uncertainties in the y and x intercepts can be quite large depending upon the magnitudes of slopes even when the worst case errors Ax and Ay are small. When the slope b is large, the worst case error in y-intercept due to gets magnified to bx, but in the x-intercept that due to dy gets reduced to 4y/b. Converse is the case when the slope is small. A little thought shows that the smaller of the two (y and x) intercepts is with lesser uncertainty. These inferences should be of significance when results are determined from the intercepts of linear graphs. But if there are undetected systematic errors, they increase or decrease the magnitudes of the intercepts. Therefore, the results from slopes are preferred to those from intercepts. The magnitude of uncertainty of slope and intercepts determined using above formula would give the worst case limits. Normal distribution appears to be more appropriate to determine the standard uncertainty in these cases. The worst case limit values may be treated as values with 95% confidence level. That would give standard uncertainty equal to 18/2 for slope and 40/2 for intercept When the results of experiments are calculated using the values of graphs, the above methods may be used to determine the absolute uncertainty in the slope. Alternately, 9 methods based on regression can be used but the calculations become much more involved than those with the above simple method. Also, in some cases using regression method, the uncertainty in slope may turn out to be zero or negligibly small when random errors are of magnitudes less than the least counts of measured quantities and are dominated by the rounding errors in the last digits of the measured quantities. The conclusions in those cases, where all the points appear on the line, that there are no uncertainties in slope and intercepts are definitely incorrect, because the rounding errors remain uncorrected and hence the values of slope would be uncertain due to their presence. The above method treats the rounding errors as of type B and estimates the uncertainty using their extreme values to define the limits of the rectangular distribution and would give the correct estimation of uncertainties in slope and intercepts in those cases. ‘When the points on graph show large scatter, the slope and intercepts of the graph have large uncertainties. The mathematical method to determine the uncertainty is linear regression or the method of least squares and it based on the principles of statistics. So, the uncertainty analysis by the method of least squares is of type A. But with few points on the graph the statistical inferences do not enjoy high level of confidence. An alternate method of type B would be to draw the best fit line by visual judgment and then draw two more lines, one with larger slope and the other with smaller slope, with the condition that both appear reasonably good linear fits representing the data. The idea is to determine the range of uncertainty of the slope of the best fit. IF b; and by be the slopes of these lines, then (bs ~ bz) /2.can be taken as the worst case uncertainty Ab in b. The method depends upon the judgment in drawing the two lines and only with experience one learns how to choose those lines, One may leave out some of the points if there is ground to believe that the deviation is large enough due to some systematic error in the observations. The uncertainty estimated this way varies with individual for the same data. The uncertainty evaluated by regression methods, on the other hand, gives same value and gives rise to a false belief that regression method is the ‘correct’ method and the above method is only approximate. But the conditions for reliable statistical inferences are rarely satisfied in most cases due to limited data and so the above method need not be considered as inferior. Similarly the worst case uncertainty Aain the intercept a would be (a:~ a3)/2. 10 Though the uncertainties determined above are using type B method, normal distribution appears to be more appropriate to represent the probability. By treating the worst case uncertainties equivalent to 95% confidence level, the standard uncertainties would be given by Ab/2 and Aa/2 respectively, Some points to be noted for entering observation data and calculations 1. Always express the worst case error in measured quantity at the time of entering the reading in observation table on paper. From the magnitude of the worst case error the standard uncertainty may be calculated stating the procedure followed and expressed with the measured value. Be particular to mention whether the number you have entered is worst case error or standard uncertainty so that there would be no confusion. 2. The last digit in the reading should have the decimal place same as that of the least count of the measuring instrument. For example though we may say twenty four centimeters, if the length is measured using a tape with least count of 1 mm, our record should read 24.0 cm or 240 mm or 0.240 m. 3. If itis possible to make one of the readings on the scale coincide precisely with one end of the object then the uncertainty in the measurement would be equal to the uncertainty in the other reading only and it would be the uncertainty due to rounding at the reading, For example, suppose one end of a plate coincides with zero of the scale and the reading closest to the other end of the plate is 20.2 cm. The last. digit of the reading is, in fact, the rounded number representing any number between 0.15 and 0.25. The worst case rounding error is + 0.5 mm if the smallest division on the scale is 1 mm. The worst case rounding error is always equal to half the least count (resolution limit) of the measuring instrument. If current is measured on an ammeter with pointer moving over a scale marked in units in multiples of 2 mA, the worst case rounding error a reading 8 mA is +1mA and not + un 0.5 mA. The uncertainty introduced due to rounding is estimated by assuming that these errors have a rectangular distribution with limits set by the worst case values. If the quantity is obtained by adding or subtracting two readings, then the uncertainty in the measured quantity would have to be obtained by combining the uncertainties in the two readings. The method is to obtain the standard uncertainty for the difference. Since the upper limit of uncertainty at each reading is equal to half the least count, assuming the rectangular distribution for the rounding error, the standard uncertainty is equal to (least count/2}/V3.Square of this gives the variance. ‘Adding the variances of the two and taking the square root of the sum gives the standard uncertainty of the combined uncertainty. Thus the standard uncertainty of a measured quantity obtained as a sum or difference of two readings becomes equal to V{(2)(least count/2v3)" J = 0.408 x (least count). Tt of least count. Generally, a measurement consists of reading the scale of the measuring instrument at two points (often, one point is made to coincide with zero of the scale but that coincidence is only approximate) and so the uncertainty in the measurand due to rounding of the two readings should be evaluated by the above method. In the example given in JCGM 100_2008, the standard uncertainty in measured quantity due to uncertainty of limit of resolution is taken as equal to {least count/2)/v3 = 0.29 of the least count. / think, the student can take the uncertainty as equal to 0.4 of the least count stating the above reasoning, In the above example, with scale of least count 1 mm the readings would be 24.0 cm with standard uncertainty of 0.04 em or 240 mm with standard uncertainty of 0.4 mm or 0.240 m with standard uncertainty of .0004 m. If the markings on the tape are of 0.5 mm, then the standard uncertainty in the reading would be 0.2 mm. Sometimes, some other factors such as inability to monitor the measuring process may be approximated to 0.4 etc., may introduce larger errors in the measurements. So, when such factors are suspected, larger estimated uncertainties may be recorded. For example while measuring the distance between two points, if it is felt that the measuring tape cannot be held straight, then the estimated worst case error larger than the least count of the measuring tape may be recorded, but the supporting reason for that must be stated. While making measurements with digital meters always refer to the operating manuals supplied with them for the accuracy (or worst case error) evaluation. The error in the reading may often be larger than the least count of the instrument. Often it is stated as a combination of some percentage of the reading and some multiple of the least count. For example, the dc voltage of reading on a DMM is stated to have “an accuracy of 0.8% of the reading plus one digit"(i.e. the least count). The formula giving the accuracy is assumed to be giving worst case errors. This means when the reading is 15.00 V, the least count is 0.01 V but the worst case 2 error is 0.12 + 0.01 V = 0.13 V. If it is assumed that the error distribution is rectangular then the ‘standard uncertainty’ would be equal to 0.13/V3 V as explained earlier. While reporting the final result and the uncertainty the guiding instructions in JCGM should be followed. They are reproduced below. Reporting Uncertainty 7.1.4 Although in practice the amount of information necessary to document a measurement result depends on its intended use, the basic principle of what is required remains unchanged: when reporting the result of a measurement and its uncertainty, it 's preferable to err on the side of providing too much information rather than too little For example, one should a) describe clearly the methods used to calculate the measurement result and its uncertainty from the experimental observations and input data; b) list all uncertainty components and document fully how they were evaluated; ¢) Present the data analysis in such @ way that each of its important steps can be readily followed and the calculation of the reported result can be independently repeated if necessary; 4) give all corrections and constants used in the analysis and their sources. A test of the foregoing list is to ask oneself “Have | provided enough information in a new information sufficiently clear manner that my result can be updated in the future or data become available?” 7.2 Specific guidance 7.2.1 When reporting the result of a measurement, and when the measure of uncertainty is the combined standard uncertainty u(y), one should a) give a full description of how the measurand Y is defined; ) give the estimate y of the measurand ¥ and its combined standard uncertainty ucly); the units of y and uc(y] should always be given; ¢) include the relative combined standard uncertainty u.(y)/|y|, ly| #0, when appropriate. ‘Some points to remember while evaluating the combined uncertainty 1. JCGM 100:2008 clearly states that “Proper use of the pool of available information for a Type B evaluation of standard uncertainty calls for insight based on experience and general knowledge, and is 2 skill that can be learned with practice. It should be recognized that a Type B evaluation of standard uncertainty can be as reliable as a Type A evaluation, 2B especially in a measurement situation where a Type A evaluation is based on a comparatively small number of statistically independent observations. mA 2. For getting the variance of type B uncertainty, first determine the width ‘2o’or range “ry of the absolute uncertainty. Then for rectangular distribution the variance is either o2/3 or 7/12. Add all variances in the formula for propagation of uncertainty and take the square root of the sum to get ‘the standard combined uncertainty. Multiply that by two to get the expanded uncertainty with confidence level of 95%. ‘This is the generally accepted procedure. If there are some type A uncertainties, then type A variance is calculated from the set of observations with random errors. If ‘the set consists of n independent observations then the variance for the sample of n z) ‘observations is given by the formula x,” = The standard deviation of the population is given by s/vn which represents the standard deviation in the mean of n observations and so si?/n is the variance for type A errors. For combining Uncertainties of independent measurands, RSS procedure is to be followed irrespective of the type of uncertainty. 3, The uncertainties before being combined should be expressed in two or three significant digits so that rounding errors do not get compounded in the calculations. The rounding procedure should be the usual one: neglect the fraction less than half of the unit in the digit to be rounded and add one unit if the fraction is larger than half the unit. 4. The uncertainty in the final result should be expressed in either one or two, but not more than two significant digits. (According to the prevalent convention, it is expressed in two digits if the first digit happens to be either 1 or 2 and in single digit otherwise.) The last digit in the uncertainty so expressed is always rounded upwards, For example, 0.137... after rounding, should be written as 0.2 oF 0.14 and not as 0-1 and 0.14. The reason for rounding upwards is that when the last ‘tin the measurand is rounded up or down, the uncertainty is increased in both cases, For example, g = 9.748..NW/kg and expanded uncertainty is 0.344..N/kg. If we express the uncertainty in single digit the value of g should have its last digit after the decimal point and after rounding it should be,9.7. But the fraction neglected would increase the uncertainty. Similarly, if we express the uncertainty in two digits, the value of g would be expressed as 9.75 and the fraction by which it hhas been increased would increase the uncertainty. Thus in the two cases the results should be expressed as : = 9.7 £ 0.4 N/kg and g = 9.75 + 0.35 N/kg respectively. [see that in the first case the total uncertainty is 0.048 +0. 344 = 0.392 = In the second case it is 0.002 + 0.344 = 0.346 = 0.35 N/ke) 4. N/kg. Finally, itis worth giving a thought to the following statement from JCGM 100_2008, 14 “although this Guide provides a framework for assessing uncertainty, it cannot substitute for critical thinking, intellectual honesty and professional skill. The evaluation of uncertainty is neither a routine task nor a purely mathematical one; it depends on detailed knowledge of the nature of the measurand and of the measurement. The quality and utility of the uncertainty quoted for the result of @ measurement therefore ultimately depend on the understanding, critical analysis, and integrity of those who contribute to the assignment of its value.” Examples 1, Determine the density of a rectangular plate with mass 8.6 g measured with least count of 0.1 g, length 5.0 cm and breadth 4.0 cm with least count of Imm and eight readings for thickness at various places are 5.6 mm, 5.5 mm, 5.6 mm, 5.8 mm, 5.5 mm, 5.6 mm, 5.5 mm and 5.5 mm with least count of 0.1 mm. The uncertainty in each reading arises due to the rounding errors in readings. To measure the length of the plate by holding it against a scale marked with units of 1 mm, the ends of the plate may be observed to coincide to the 0.0 cm and 5.0 cm marks respectively on the scale. The difference between these readings gives the length equal to 5.0 cm and the standard uncertainty equal 0.4 of the least count gives the value of 0.04 cm. Similarly the standard uncertainty in mass is 0.04 g and in breadth 0.04 cm. The standard uncertainty in each measured value of thickness is 0.04 mm. But many readings are taken for determining the thickness of the plate because it is not same throughout. The variations in the readings arise due to ‘problem of definition’ of thickness. The sample does not represent repeated readings of the same quantity and so strictly speaking the uncertainty evaluation by statistical methods (type A) is not valid. But we can use type B method to estimate the uncertainty. The average of the eight values gives the average thickness as ta = 5.575 mm. The worst case error can be (5.8 -5.5)/2 = 0.15 mm. This is larger than the uncertainty in each reading of thickness and so this is to be considered as overriding the standard uncertainty in each measurement of thickness. The standard uncertainty in thickness, assuming uniform distribution, would then be 0.15/ v3 = 0.087 mm. We have y = 7 or The density d = 8.6 /( 5.0 x 4.0 x 0.558) = 0.7706093 g/cm’. 15 We do not know how many digits in the final result are significant from the calculation. We do not use the rules for significant figures for that purpose. We shall determine the uncertainty first and from that we come to know how many digits should be retained in the result. ‘The uncertainties in all readings are of type B and are assumed to have rectangular (or uniform) distribution within intervals defined by [-a, a] for which the standard deviation is given by o/V3. Relative standard uncertainty is the ratio of standard uncertainty divided by average value of measurand. Since all measurements are independent, the formula for propagation of uncertainty is » (sen (aan) (220) tee fe At Hence we get (uid)) 0.08)? (0.04)? _ (0.087 uid?) 9:04) if 0 ea rs 0 Ca 0) Cao) “A588 J So, w(d) = 0.7706093 x 00207056 0.015959 g/cm’, This corresponds to the standard combined uncertainty. The expanded uncertainty with 95% confidence level is therefore, 0.0319 g/cm’ or 0.04 g/cm’ (rounded upwards in the uncertainty) The value of density is now to be rounded to have the last digit in the same decimal place as that in the expanded uncertainty. “Therefore, the result should be expressed as d= 0.77 + 0.04 g/cm If the expanded uncertainty obtained from the product of the coverage factor with the standard uncertainty in three figures, then round it to retain only two digits, so that the final result will not have more than two non significant figures. More often it is rounded to retain single significant digit as illustrated in the above example. 2. Determine the value of g from the readings of a simple pendulum of length 60.0 cm. ‘Three readings for time interval of 20 oscillations were 31.58 s, 31.74 s and 31.10 s. We have g = 4n'(L/T") and 16 The three values of T are 1.579 s, 1.587 s and 1.555 s, The mean value is 1.574 s and the range © = 0.032 s, evaluating the uncertainty in T by type B procedure, the standard uncertainty is 0.032/2V3 s and the relative standard uncertainty is (0.032/2v3) /1.574. Therefore, yh ] b= a.0rr7s06 vi) 1) g = 4xn°x 0.60/(1.574)? = 9.5609607 N/kg Therefore standard combined uncertainty u(g) = 9.5609607 x 0.011756 = 0.112 N/kg. With expanded uncertainty rounded upwards 9 = 9.56 + 0.23 N/kg. 3. The ac voltage across a series combination of resistance f and capacitance Cis read as 8.40 V (on 20 V range of DMM 3800) and the current through as 2.48 mA (on 20 mA range). The frequency of the ac is 1000 Hz. The resistance R measured using ‘ohmmeter is read as 1.050 kohm. Determine the value of capacitance Cand estimate the uncertainty in the value. (For DMM 3800 the operating manual states the accuracy for readings on 20 V(ac) range as equal to + 0.8% of rdg + 3 digits, for readings on 20 mA(ac) range as equal to + 1.% of edg + 3 digits and for readings on 2 kohm range as equal to + 0.5% of rdg +1 digit.) L We have z = “andc = f oVz?-R Therefore, Z=8.4/2.48 kohm =3.387096 kohm and = art 3.387096 20233339 «10 = 0.019033 x10 0494233 uF. vy We treat the accuracy as stated in the operating manual as giving the worst case errors. SO, ‘AV 0.8% of 8.40 + 0.03 = 0.0672 + 0.03 = 0.10 Vis to be treated as the worst case error. So the standard relative uncertainty would be equal to = Similarly the worst case error in | would be 1.0% of 2.48 + 0.03 = 0.0248 + 0.03 =0.06 mA, To obtain the standard relative uncertainty in Z we add the squares of standard relative Lav a uncertainties in Vand J and take the square root of the sum. = 8.9308 «107? The standard uncertainty u(Z) is then obtained from 8.9308 x 10°x 3.387096 = 0.0302.. kchm = 0.03 kohm (= 30 ohm) and Z should be expressed as 3.39 + 0.06 kohm at 95% confidence level. ‘The worst case error AR is 0.5% Of 1.050 + 0.001 = 0.00625 kchm (= 6 chm) and so the standard uncertainty u(R) = 3.6 ohm The standard uncertainty in C from the formula is given by 1 [aza@* Corwen? | nce 20 (me In the above expression u(Z) and u(R) are standard uncertainties. Substituting the values calculated above we get Mcy= (2.39? 10% 1.08? 108) af (23.39 «10? 30) 24 (21.05 «10 3.6) 21000 = 4.7509 1078 «2.03505 10% = IGT ae X10 WE ‘Therefore, with expanded uncertainty at 95% confidence level C= 0.049 + 0.002 pF. 18

You might also like