Error Analysis and Propagation Sources of Error and Uncertainty There are numerous sources of error and uncertainty

in any lab experiment. One important aspect of a lab class is to identify and estimate the errors and uncertainties in measurement, and to determine how those measurement errors affect results calculated from measured values. Errors arise from three major sources: Random fluctuation the idea that the actual quantity may vary slightly around some central value. It is assumed that the variations go in both directions equally. The way to deal with random fluctuation is to take several measurements of the same quantity and calculate the average. Systematic error the idea that a piece of equipment may not be calibrated or functioning correctly, or that conditions are not as you assume. All data points will vary in the same direction (all too high or too low). The way to deal with systematic error is to examine possible causes inherent in your equipment and/or procedures, and whether they may be corrected. Human error (aka mistakes) the idea that humans mess up. The way to deal with human error is to repeat the measurement. Uncertainty arises primarily because no piece of measuring equipment is infinitely precise (sensitive) or infinitely accurate (correct). At some point, the student must estimate how uncertain a data point may be, based on the limitations of the instruments and on the measurement process. When you make measurements in these labs, you need to estimate the uncertainty of each measurement you make, and record it in your lab report. You also need to take those estimated uncertainties and determine the uncertainty in the results you calculate. Below is a brief discussion of how to do this. Uncertainties Single Measurements Suppose that you must measure the length of a rod. For this you normally use a meter stick or ruler (depending on the length of the rod). You need to determine both the length of the rod and the uncertainty in that length. The meter sticks in most labs are marked in centimeters (the large numbered marks) and in millimeters (the small marks). We can say that this meter stick is calibrated in millimeters.

Suppose that you measure a rod and find that its length falls between 82.7 centimeters (827 millimeters) and 82.8 centimeters (828 millimeters). If you have reasonable eyesight, you can estimate whether it is closer to 82.7 or to 82.8, and you might even be able to make a good guess at another decimal place, say, 82.76. In this instance, you can read to a tenth of a centimeter, which is a millimeter. You can determine with confidence that the length is 82.7 centimeters plus some little bit, some number of hundredths of centimeters. You read to a tenth of a centimeter and estimate to a hundredth of a centimeter. So you may be several hundredths of a centimeter off. How many? You can probably estimate whether the length is between 82.70 and 82.75, or between 82.75 and 82.80 you can get the correct half of the interval. You also have a pretty good idea that the length is less than 82.80. So in this instance, you might estimate your uncertainty as 0.04 cm (the difference between 82.80 and 82.76). We generally denote the uncertainty of a quantity with the symbol (delta), so the uncertainty in the quantity L (length) would be called L. With good eyesight, you might report your result for the length as L L = 82.76 0.04 cm. When performing labs, estimate the uncertainty in each quantity you measure directly, and record this uncertainty in your lab report. Report final results in standard form: L L.

Uncertainties Multiple Measurements In many labs, you repeat a measurement several times and take an average value as your data point. (Remember, this is the way to deal with random fluctuation.) It is fervently hoped that you all remember how to take an average: add up all the numbers and divide by the number of data points you have. Average value of n measurements of x = xi /n How do we estimate an uncertainty for this average value? There are several ways, but for these labs we will choose a straightforward way. We will calculate the average deviation associated with these measurements. The average deviation represents how far, on the average, your individual measurements are from their average value. Heres how:

The deviation of any one measurement is the value of the individual data point minus the average value of all the data points: Deviation = xi x avg for each individual point xi. To find the average deviation, first take the absolute value of each deviation (make them all positive). The average deviation is just the average of all these positive numbers (the average of the magnitudes). Add up all the absolute values of the deviations, and divide by the number of measurements you took. Simple.

Example: You have performed an experiment in which you timed the fall of a mass. You got the following values, all in seconds: 12.24 12.32 12.25 12.11 The average value of these times is the sum of the values, divided by 4 (the number of measurements): 12.24 + 12.32 + 12.25 + 12.11 = 48.92 48.92 4 = 12.23 So 12.23 is the average value. Now, to find the deviation of each measurement, we subtract this average value: Value deviation 12.24 12.32 12.25 12.11 0.01 0.09 0.02 -0.12

And then we take the absolute values of each deviation, and average them: (0.01 + 0.09 + 0.02 + 0.12) 4 = 0.24 4 = 0.06

The average deviation for this set of measurements is therefore 0.06. The value should then be reported as 12.23 0.06 seconds.

Standard Deviation Students of statistics may object that there is a more scientific way to estimate errors. They are correct. There is a quantity known as the standard deviation, commonly denoted by the letter (sigma). Below are two common formulas for the standard deviation: = [(xi x avg)2/(N 1)] 2 = (x2) avg - (x avg)2 where N is the number of measurements taken. The significance of the standard deviation is that in general, about 68% of data points will fall within one sigma value of the average, about 95% will fall within two, and about 99% within three. Propagating Uncertainties Now suppose you go on with your lab, and later you are directed to calculate some quantity that involves the length you measured. How large of an uncertainty will your calculated quantity have, based on the uncertainty in your length? Suppose, for example, that you needed to calculate the sum of two lengths. L1 is the length of the rod you measured and L2 is the length of the clamp attached to it. The total length L is the sum of the two. Here are your data: L1 = 82.76 0.05 cm L2 = 14.52 0.05 cm The total length L is, of course, 97.28 cm (= 82.76 + 14.52). What is its uncertainty? The rule here is that if any two quantities are added or subtracted, you add the uncertainties to find the uncertainty of the result. In this instance, L = L1 + L2 = 0.05 + 0.05 cm = 0.10 cm So you would report your total length as 97.28 0.10 cm. Below is a summary of the rules on how to calculate uncertainties of some common calculations. In each case, C is a function of A and B.

If C equals:

Then:

A + B or A - B

C = A + B C/C = A/A + B/B

AB (A times B)or A/B (A divided by B)

C/C = m(A/A) + n(B/B) Am Bn (A to the m power times B to the n where m denotes the absolute power) value of m

A more advanced function, like sin A

C = C/AA + C/BB (NOTE: this is the general formula)

NOTE: if any of your variables is an angle, be sure to express both it and its uncertainty in radians. Reporting Results Measured Values However you estimate your uncertainty, you need to follow generally accepted practice when reporting your measured values. You should report all the known digits, plus one that is estimated or otherwise subject to uncertainty. In the first example above, where you measured the length of the rod, you would report the length as 82.76 cm. You know the 82.7, and the last digit - the 6 - is estimated.

Reporting Results Calculated Values The calculator is a wonderful thing. It can give you the result of any calculation to as many decimal places as it has memory to display. The problem is, most of those decimal places are meaningless. Think back to the meter stick in the first example. If you use that meter stick, and the best you can estimate a length is plus or minus 0.05 cm, then it is not only meaningless but actually dishonest to report some value calculated from that length with four or five or eight decimal places. That would imply a precision you just dont have. So we need to observe the rules of significant figures.

Significant Figures The significant figures in a number are all the digits, including the zeros, EXCEPT for those zeros that are used only to locate the decimal point.

Thus, 1.032 has four significant figures, as does 0.03765. In some cases the same number may be expressed differently to give an indication of the precision (sensitivity) of the measurement. For example, consider the number 9800. If this number was measured to the nearest hundred, then the zeros are used only to locate the decimal point, and thus it has only two significant figures. This number could be written as 9.8 x 103. If, however, the quantity was measured to the nearest ten, then the number could be written as 9.80 x 103. Since that last zero is not used to locate the decimal point, it is considered a significant figure. When you combine values that have different numbers of significant figures, you must round off your answer appropriately. Adding and Subtracting: Round the answer to the same number of decimal places as the LEAST precise quantity. Example: 703 + 7 + 0.66 = 710.66. Round off to 711. Multiplying and Dividing: Report the answer to the same number of significant figures as the LEAST precise number used. Example: 72.4 x 0.084 = 6.08. Round off to 6.1 (two significant figures, just like 0.084).

More on Rounding Off When rounding off a number, drop the digits to the right of the last significant figure. When the first one to be dropped is 5, round up. Example: 15.37, to the nearest tenth Example: 83.225, to the nearest hundredth Round up to 15.4 Round up to 83.23

Percent Error and Percent Difference Now suppose you have arrived at a calculated value for some quantity, and have found uncertainties and expressed it with the proper number of significant figures. Suppose further that you want to compare your result with a standard value, or with some other result. The difference between your experimental value and the accepted value is called the discrepancy between the two values. You then calculate and report the percent error:

Percent error = [(accepted value - your value) accepted value] x 100%

In general, we do not include uncertainties in calculations of percent error. The uncertainty is included in the final statement of your value in the form: L = 97.28 0.10 cm.

To Wrap Up The rules explained above are basic to all scientific and technical work. Following them in your lab reports will help ensure that your results are reported in a clear, consistent, and dependable manner. With practice, it will also help you develop the ability to read and interpret other peoples results always a useful skill. Please save these instructions for your future use. (This document was adapted by Lynda Williams from a document found online at Georgia Perimeter College)