{"id":95,"date":"2021-01-25T00:58:39","date_gmt":"2021-01-25T00:58:39","guid":{"rendered":"https:\/\/courses.lumenlearning.com\/suny-hccc-generalscience\/?post_type=chapter&#038;p=95"},"modified":"2021-01-25T00:58:39","modified_gmt":"2021-01-25T00:58:39","slug":"9-data-uncertainty-error-and-confidence","status":"publish","type":"chapter","link":"https:\/\/courses.lumenlearning.com\/suny-hccc-generalscience\/chapter\/9-data-uncertainty-error-and-confidence\/","title":{"raw":"9. Data: Uncertainty, Error, and Confidence","rendered":"9. Data: Uncertainty, Error, and Confidence"},"content":{"raw":"<div class=\"article-introduction\">\r\n\r\nThe Olympic sport of biathlon (Figure 1) is a cross-country ski race of 20 km in which the athletes stop on four occasions to shoot 0.57 cm diameter bullets from a .22 caliber rifle at targets. The sport requires not only great endurance, but exceptional <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/accuracy\/pop\">accuracy<\/a> as the athletes shoot on two occasions from the prone position (lying down) and on two occasions while standing. The targets the athletes aim for are all 50 m away, but the size varies to match the <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/precision\/pop\">precision<\/a> expected of them; those targeted while shooting in the prone position are 4.5 cm in diameter while those targeted from the more difficult standing position are 11.5 cm in diameter. In both cases, however, the diameter of the target is many times larger than the diameter of the bullet itself \u2013 why?\r\n<figure><a title=\"&lt;strong&gt;Figure 1:&lt;\/strong&gt; Biathletes in the shooting area of a competition.\" href=\"https:\/\/www.visionlearning.com\/images\/figure-images\/157-a-2x.jpg\"> <img src=\"https:\/\/www.visionlearning.com\/images\/figure-images\/157-a.jpg\" alt=\"Biathletes in the shooting area of a competition\" \/> <\/a><figcaption><strong>Figure 1:<\/strong> Biathletes in the shooting area of a competition. <span class=\"credit\">image \u00a9 Marc Schuelper<\/span><\/figcaption><\/figure>\r\nWhile the legend of Robin Hood splitting one arrow with another is well-known, it is also unrealistic. Biathlon targets are purposely sized many times larger than the bullets the athletes shoot to account for the <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/inherent\/pop\">inherent<\/a> error and <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/uncertainty\/pop\">uncertainty<\/a> involved in long distance riflery. Even the most skilled marksman cannot account for every <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/variable\/pop\">variable<\/a> affecting the path of the bullet, like sudden gusts of wind or variations in air pressure. Shooting from the standing position involves even greater uncertainty, as indicated by the larger targets used, because even the simple rise and fall of an athlete's chest as they breathe can affect the aim of their rifle.\r\n\r\n<\/div>\r\n<section id=\"toc_1\" class=\"article-section\">\r\n<h2>Categorizing uncertainty: Accuracy vs. precision<\/h2>\r\nScientific measurements also incorporate variability, and scientists report this as <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/uncertainty\/pop\">uncertainty<\/a> in an effort to share with others the level of error that they found acceptable in their measurements. But uncertainty in science does not imply doubt as it does in everyday use. Scientific uncertainty is a quantitative measurement of variability in the <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/data\/pop\">data<\/a>. In other words, uncertainty in science refers to the idea that all data have a range of expected <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/value\/pop\">values<\/a> as opposed to a precise point value. This uncertainty can be categorized in two ways: <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/accuracy\/pop\">accuracy<\/a> and <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/precision\/pop\">precision<\/a>.\r\n<ul class=\"bulleted\">\r\n \t<li><em>Accuracy<\/em> is a term that describes how well a measurement approximates the theoretically correct <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/value\/pop\">value<\/a> of that measurement, for example, how close the arrow is to the bullseye (Figure 2).<\/li>\r\n \t<li>The term <em>precision<\/em>, by comparison, describes the degree to which individual measurements vary around a central value. Measurements with high <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/precision\/pop\">precision<\/a> are highly reproducible because repeated measurement will reliably give a similar result; however, they may or may not be accurate (Figure 2).<\/li>\r\n<\/ul>\r\n<figure><img src=\"https:\/\/www.visionlearning.com\/img\/library\/modules\/mid157\/Image\/VLObject-4142-081105111152.jpg\" alt=\"targets\" \/><figcaption>Figure 2: A representation of accuracy and precision as hits on a target. The target at left depicts good accuracy as the marks are close to the bullseye, but poor precision; in contrast, the target at right depicts good precision as the marks are grouped closely, but poor accuracy.<\/figcaption><\/figure>\r\n<\/section><section id=\"toc_2\" class=\"article-section\">\r\n<h2>Uncertainty in nature<\/h2>\r\nKarl Pearson, the English statistician and geneticist, is commonly credited with first describing the concept of <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/uncertainty\/pop\">uncertainty<\/a> as a measure of <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/data\/pop\">data<\/a> variability in the late 1800s (Salsburg, 2001). Before Pearson, scientists realized that their measurements incorporated variability, but they assumed that this variability was simply due to error. For example, measurement of the orbits of planets around the sun taken by different scientists at different times varied, and this variability was thought to be due to errors caused by inadequate instrumentation. The French mathematician <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/Laplace%2C+Pierre~Simon\/pop\">Pierre-Simon Laplace<\/a> discussed a <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/method\/pop\">method<\/a> for quantifying error distributions of astronomical measurements caused by small errors associated with instrument shortcomings as early as 1820. As technology improved through the 1800s, astronomers realized that they could reduce, but not eliminate this error in their measurements.\r\n\r\nPearson put forward a revolutionary idea: <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/uncertainty\/pop\">Uncertainty<\/a>, he proposed, was not simply due to the limits of technology in measuring certain events \u2013 it was <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/inherent\/pop\">inherent<\/a> in nature. Even the most careful and rigorous scientific investigation (or any type of investigation for that matter) could not yield an exact measurement. Rather, repeating an investigation would yield a scatter of measurements that are distributed around some central <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/value\/pop\">value<\/a>. This scatter would be caused not only by error, but also by natural variability. In other words, measurements themselves, independent of any human or instrument inaccuracies, exhibit scatter.\r\n\r\nWhether it is the flight path of an arrow, the resting heart rate of an adult male, or the age of a historical artifact, measurements do not have exact <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/value\/pop\">values<\/a>, but instead always exhibit a range of values, and that range can be quantified as <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/uncertainty\/pop\">uncertainty<\/a>. This uncertainty can be expressed as a plot of the <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/probability\/pop\">probability<\/a> of obtaining a certain value, and the probabilities are distributed about some central, or <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/mean\/pop\">mean<\/a>, value.\r\n<div class=\"comprehension-checkpoint\">\r\n<p class=\"leader\">Comprehension Checkpoint<\/p>\r\n<p class=\"question\">Certain large scale scientific measurements, such as the orbit of planets, have no uncertainty associated with them.<\/p>\r\n\r\n<form class=\"question\" name=\"cc5836\">\r\n<ul class=\"quiz-options\">\r\n \t<li class=\"option-a\"><label class=\"choice\" for=\"q1-5836-0-option-a\">True.<\/label><\/li>\r\n \t<li class=\"option-b\"><label class=\"choice\" for=\"q1-5836-1-option-b\">False.<\/label><\/li>\r\n<\/ul>\r\n<\/form><\/div>\r\n<\/section><section id=\"toc_3\" class=\"article-section\">\r\n<h2>Uncertainty and error in practice: Carbon-14 dating<\/h2>\r\nArchaeologists, paleontologists, and other researchers have long been interested in dating objects and artifacts in an effort to understand their history and use. Unfortunately, written <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/record\/pop\">records<\/a> are a relatively recent human invention, and few historical artifacts are accompanied by precise written histories.\r\n\r\nIn the first half of the 20th century, an American nuclear chemist by the name of Willard F. Libby became interested in using the radioactive <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/isotope\/pop\">isotope<\/a> <sup>14<\/sup>C to date certain objects. The <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/theory\/pop\">theory<\/a> of radiocarbon dating is relatively simple. Most carbon in the Earth's <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/atmosphere\/pop\">atmosphere<\/a> is in the form of <sup>12<\/sup>C, but a small amount of the isotope <sup>14<\/sup>C is produced naturally through the bombardment of <sup>14<\/sup>N with cosmic rays (W. F. Libby, 1946). As plants take up carbon from the atmosphere through <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/respiration\/pop\">respiration<\/a>, they incorporate both <sup>14<\/sup>C as well as the more abundant <sup>12<\/sup>C into their tissues. Animals also take up both carbon isotopes through the foods that they eat. Thus, all living <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/organism\/pop\">organisms<\/a> have the same <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/ratio\/pop\">ratio<\/a> of <sup>14<\/sup>C and <sup>12<\/sup>C isotopes in their body as the atmosphere.\r\n\r\nUnlike <sup>12<\/sup>C, <sup>14<\/sup>C is a radioactive <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/isotope\/pop\">isotope<\/a> that is constantly undergoing <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/decay\/pop\">decay<\/a> to its <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/daughter\/pop\">daughter<\/a> <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/product\/pop\">product<\/a> <sup>14<\/sup>N at a known rate. While an <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/organism\/pop\">organism<\/a> is alive, it is taking up new <sup>14<\/sup>C from the <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/environment\/pop\">environment<\/a> and thus remains in <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/equilibrium\/pop\">equilibrium<\/a> with it. When that organism dies, however, the carbon in its tissues is no longer replaced, and the amount of <sup>14<\/sup>C slowly decreases in time as it decays to <sup>14<\/sup>N. Thus, the amount of radioactive <sup>14<\/sup>C remaining in a piece of wood or an animal bone can be used to determine when that organism died. In essence, the longer the organism has been dead, the lower the <sup>14<\/sup>C levels.\r\n\r\nThe amount of radioactive material (such as <sup>14<\/sup>C) in a sample can be quantified by counting the number of <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/decay\/pop\">decays<\/a> that the material undergoes in a specific amount of time, usually reported in counts per minute (cpm). When Libby began his radiocarbon work in the 1940s, the technology available was still quite new. A simple Geiger counter was only first invented in 1908 by the German scientist Hans Wilhelm Geiger, a student of <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/Rutherford%2C+Ernest\/pop\">Ernest Rutherford<\/a>'s, and it was not perfected until 1928 when Walther M\u00fcller, a student of Geiger's, improved on the design, allowing it to detect all types of <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/radiation\/pop\">radiation<\/a>. Libby himself is credited with building the first Geiger counter in the United States in the 1930s.\r\n\r\nLibby, however, faced a major hurdle with using the instrument to measure <sup>14<\/sup>C. The problem was that naturally-occurring background <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/radiation\/pop\">radiation<\/a> from cosmic rays and Earth, along with the variability associated with that background signal, would overwhelm the small <sup>14<\/sup>C signal he expected to see. In 1949, he reported on a <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/method\/pop\">method<\/a> for reducing the background signal and variability: He placed the entire sample and the detector inside of a tube shielded by 2 inches of lead and 4 inches of iron (W. F. Libby, Anderson, &amp; Arnold, 1949). In this way, Libby and his colleagues reduced the background signal from 150 cpm to 10 cpm and minimized the variability associated with the signal to \"about 5-10% error,\" or less than 1 cpm.\r\n\r\nLibby and colleagues do not use the word <em>error<\/em> as we do in common language, where it refers to a mistake such as a typographical error or a baseball error. The Latin origin of the word error (<em>errorem<\/em>) means <em>wandering<\/em> or <em>straying<\/em>, and the scientific use of the word is closer to this original meaning. In science, <em>error<\/em> is the difference between the true <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/value\/pop\">value<\/a> and the measured value, and that difference can have many different causes. Libby calculated the error associated with his measurements by counting the number of <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/decay\/pop\">decay<\/a> events in the sample in a known amount of time, repeating the measurement over multiple periods, and then using statistical techniques to <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/quantify\/pop\">quantify<\/a> the error (see our <a href=\"https:\/\/www.visionlearning.com\/library\/module_viewer.php?mid=155\">Statistics in Science<\/a> module).\r\n\r\nIn 1949, Libby, working with his post-doctoral student James Arnold, reported the first use of radiocarbon dating for determining the age of wood fragments from archaeological sites around the world (Arnold &amp; Libby, 1949). Because the <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/method\/pop\">method<\/a> was new, Arnold and Libby were careful to replicate their measurements to provide a detailed estimate of different types of error, and they compared the results of their method with samples of a known age as a <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/control\/pop\">control<\/a> (Table 1).\r\n<div class=\"table-container\">\r\n<div class=\"table-container\">\r\n<table class=\"generic-data size-80\">\r\n<thead>\r\n<tr>\r\n<th rowspan=\"2\">Sample<\/th>\r\n<th>Specific activity (cpm\/g of carbon)<\/th>\r\n<th colspan=\"2\">Age (years)<\/th>\r\n<\/tr>\r\n<\/thead>\r\n<tbody>\r\n<tr>\r\n<td><\/td>\r\n<td>Found<\/td>\r\n<td>Found<\/td>\r\n<td>Expected<\/td>\r\n<\/tr>\r\n<tr>\r\n<td>Tree Ring<\/td>\r\n<td>11.10 \u00b1 0.31<\/td>\r\n<td>1100 \u00b1 150<\/td>\r\n<td>1372 \u00b1 50<\/td>\r\n<\/tr>\r\n<tr>\r\n<td><\/td>\r\n<td>11.52 \u00b1 0.35<\/td>\r\n<td><\/td>\r\n<td><\/td>\r\n<\/tr>\r\n<tr>\r\n<td><\/td>\r\n<td>11.34 \u00b1 0.25<\/td>\r\n<td><\/td>\r\n<td><\/td>\r\n<\/tr>\r\n<tr>\r\n<td><\/td>\r\n<td>10.15 \u00b1 0.44<\/td>\r\n<td><\/td>\r\n<td><\/td>\r\n<\/tr>\r\n<tr>\r\n<td><\/td>\r\n<td>11.08 \u00b1 0.31<\/td>\r\n<td><\/td>\r\n<td><\/td>\r\n<\/tr>\r\n<tr>\r\n<td><\/td>\r\n<td>Average: 10.99 \u00b1 0.15<\/td>\r\n<td><\/td>\r\n<td><\/td>\r\n<\/tr>\r\n<tr>\r\n<td colspan=\"4\">\r\n<figure><strong>Table 1<\/strong>: Age determinations on samples of known age from Arnold &amp; Libby (1949). The specific activities for five different replicates of a sample of wood from a Douglas fir excavated from the Red Rock Valley are shown in the second column of Table 1. Each individual measurement has an error shown to the right of it, indicated by the \u00b1 sign. Arnold and Libby describe these measurements in their paper, stating, \"The errors quoted for the specific activity measurements are standard deviations as computed from the Poisson statistics of counting random events.\" In other words, the individual error is calculated from the expected uncertainties associated with radioactive decay for each sample.<\/figure>\r\n<\/td>\r\n<\/tr>\r\n<\/tbody>\r\n<\/table>\r\n<\/div>\r\n<\/div>\r\nThe specific activities for five different replicates of a sample of wood from a Douglas fir excavated from the Red Rock Valley are shown in the second column of Table 1. Each individual measurement has an error shown to the right of it, indicated by the \u00b1 sign. Arnold and Libby describe these measurements in their paper, stating, \"The errors quoted for the specific activity measurements are <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/standard+deviation\/pop\">standard deviations<\/a> as computed from the Poisson <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/statistic\/pop\">statistics<\/a> of counting random events.\" In other words, the individual error is calculated from the expected uncertainties associated with radioactive <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/decay\/pop\">decay<\/a> for each sample.\r\n\r\n<\/section><section id=\"toc_4\" class=\"article-section\">\r\n<h2>Statistical vs. systematic error<\/h2>\r\nAs seen in Table 1, an <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/average\/pop\">average<\/a> specific activity <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/value\/pop\">value<\/a> (10.99) is provided at the bottom with an overall error. The overall error (0.15) is smaller than the individual error reported with each measurement. This is an important feature of the statistical calculation of error associated with scientific <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/data\/pop\">data<\/a> \u2013 as you increase the number of measurements of a value, you decrease the <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/uncertainty\/pop\">uncertainty<\/a> and increase the confidence associated with the approximation of the value. The error reported alongside of the specific activity provides a measure of the <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/precision\/pop\">precision<\/a> of the value, and is commonly referred to as <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/statistical+error\/pop\">statistical error<\/a>. Statistical error is what Pearson described as the <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/inherent\/pop\">inherent<\/a> uncertainty of a measurement. It is caused by random fluctuations within a <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/system\/pop\">system<\/a>, such as the random fluctuation of radioactive <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/decay\/pop\">decay<\/a>, and is sometimes referred to as <em>random error<\/em> as the researcher has little control over it. Statistical error cannot be eliminated, as Pearson described, but it can be measured and reduced by conducting repeated <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/observation\/pop\">observations<\/a> of a specific event.\r\n\r\nIn column 3 of Table 1, Arnold and Libby estimate the age of the Douglas fir sample based on the <sup>14<\/sup>C activity as 1100 years old (placing its first season of growth in the year 849 CE). In column 4 of Table 1, they report the actual age of the Douglas fir as calculated by counting tree rings in the sample as 1372 years old (placing its first season in the year 577 CE). By comparing the <sup>14<\/sup>C age to a theoretically correct <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/value\/pop\">value<\/a> as determined by the tree ring count, Arnold and Libby allow the reader to gauge the <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/accuracy\/pop\">accuracy<\/a> of their <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/method\/pop\">method<\/a>, and this provides a measure of a second type of error encountered in science: <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/systematic+error\/pop\">systematic error<\/a>. <em>Statistical error<\/em> cannot be eliminated, but it can be measured and reduced by conducting repeated <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/observation\/pop\">observations<\/a> of a specific event. <em>Systematic error<\/em>, in contrast, can be corrected for \u2013 like if you know your oven is 50\u00b0 too hot, you set it for 300\u00b0 rather than 350\u00b0.\r\n\r\nBased on their <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/data\/pop\">data<\/a>, Arnold and Libby state that the \"agreement between prediction and <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/observation\/pop\">observation<\/a> is seen to be satisfactory.\" However, as he continued to do <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/research\/pop\">research<\/a> to establish the <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/method\/pop\">method<\/a> of <sup>14<\/sup>C dating, Libby began to recognize that the discrepancy between radiocarbon dating and other methods was even larger for older objects, especially those greater than 4,000 years old (W.F. Libby, 1963). Where theoretically correct dates on very old objects could be established by other means, such as in samples from the temples of Egypt, where a calendar <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/system\/pop\">system<\/a> was well-established, the ages obtained through the radiocarbon dating method (the \"found\" ages in Table 1) were consistently older than the \"expected\" dates, often by as much as 500 years.\r\n\r\nLibby knew that there was bound to be <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/statistical+error\/pop\">statistical error<\/a> in these measurements and had anticipated using <sup>14<\/sup>C dating to calculate a range of dates for objects. But the problem he encountered was different: <sup>14<\/sup>C-dating systematically calculated ages that differed by as much as 500 years from the actual ages of older objects. <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/systematic+error\/pop\">Systematic error<\/a>, like Libby encountered, is due to an unknown but non-random fluctuation, like instrumental bias or a faulty assumption. The radiocarbon dating <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/method\/pop\">method<\/a> had achieved good <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/precision\/pop\">precision<\/a>, replicate analyses gave dates within 150 years of one another as seen in Table 1; but initially it showed poor <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/accuracy\/pop\">accuracy<\/a> \u2013 the \"found\" <sup>14<\/sup>C age of the Douglas fir was almost 300 years different than the \"expected\" age, and other objects were off by some 500 years.\r\n\r\nUnlike <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/statistical+error\/pop\">statistical error<\/a>, <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/systematic+error\/pop\">systematic error<\/a> can be compensated for, or sometimes even eliminated if its source can be identified. In the case of <sup>14<\/sup>C-dating, it was later discovered that the reason for the systematic error was a faulty assumption: Libby and many other scientists had assumed that the production rate for <sup>14<\/sup>C in the <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/atmosphere\/pop\">atmosphere<\/a> was constant over time, but it is not. Instead, it fluctuates with changes in Earth's magnetic field, the uptake of carbon by plants, and other factors. In addition, levels of radioactive <sup>14<\/sup>C increased through the 20<sup>th<\/sup> century because nuclear weapons testing released high levels of <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/radiation\/pop\">radiation<\/a> to the atmosphere.\r\n<figure><a title=\"&lt;strong&gt;Figure 3:&lt;\/strong&gt; Tree-ring dates have been used to recalibrate the radiocarbon dating method.\" href=\"https:\/\/www.visionlearning.com\/images\/figure-images\/157-c-2x.jpg\"> <img src=\"https:\/\/www.visionlearning.com\/images\/figure-images\/157-c.jpg\" alt=\"Tree-ring dates have been used to recalibrate the radiocarbon dating method\" \/> <\/a><figcaption><strong>Figure 3:<\/strong> Tree-ring dates have been used to recalibrate the radiocarbon dating method. <span class=\"credit\">image \u00a9 Hannes Grobe<\/span><\/figcaption><\/figure>\r\nIn the decades since Libby first published his <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/method\/pop\">method<\/a>, researchers have recalibrated the radiocarbon dating method with tree-ring dates from bristlecone pine trees (Damon et al., 1974) and corals (Fairbanks et al., 2005) to correct for the fluctuations in the production of <sup>14<\/sup>C in the <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/atmosphere\/pop\">atmosphere<\/a>. As a result, both the <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/precision\/pop\">precision<\/a> and <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/accuracy\/pop\">accuracy<\/a> of radiocarbon dates have increased dramatically. For example, in 2000, Xiaohong Wu and colleagues at Peking University in Beijing used radiocarbon dating on bones of the Marquises (lords) of Jin recovered from a cemetery in Shanxi Province in China (see Table 2) (Wu et al., 2000). As seen in Table 2, not only is the precision of the estimates (ranging from 18 to 44 years) much tighter than Libby's reported 150 year error range for the Douglas fir samples, but the radiocarbon dates are highly accurate, with the reported deaths dates of the Jin (the theoretically correct values) falling within the <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/statistical+error\/pop\">statistical error<\/a> ranges reported in all three cases.\r\n<div class=\"table-container\">\r\n<div class=\"table-container\">\r\n<table class=\"generic-data size-80\">\r\n<thead>\r\n<tr>\r\n<th>Name of Jin Marquis<\/th>\r\n<th>Radiocarbon Date (BCE)<\/th>\r\n<th>Documented Death Date (BCE)<\/th>\r\n<\/tr>\r\n<\/thead>\r\n<tbody>\r\n<tr>\r\n<td>Jing<\/td>\r\n<td>860-816<\/td>\r\n<td>841<\/td>\r\n<\/tr>\r\n<tr>\r\n<td>Li<\/td>\r\n<td>834-804<\/td>\r\n<td>823<\/td>\r\n<\/tr>\r\n<tr>\r\n<td>Xian<\/td>\r\n<td>814-796<\/td>\r\n<td>812<\/td>\r\n<\/tr>\r\n<tr>\r\n<td colspan=\"3\">\r\n<figure><strong>Table 2<\/strong>: Radiocarbon estimates and documented death dates of three of the Marquises of Jin from Wu et al., 2000).<\/figure>\r\n<\/td>\r\n<\/tr>\r\n<\/tbody>\r\n<\/table>\r\n<\/div>\r\n<\/div>\r\n<div class=\"comprehension-checkpoint\">\r\n<p class=\"leader\">Comprehension Checkpoint<\/p>\r\n<p class=\"question\">Which type of error is not random and can be compensated for?<\/p>\r\n\r\n<form class=\"question\" name=\"cc5845\">\r\n<ul class=\"quiz-options\">\r\n \t<li class=\"option-a\"><label class=\"choice\" for=\"q1-5845-0-option-a\">statistical error<\/label><\/li>\r\n \t<li class=\"option-b\"><label class=\"choice\" for=\"q1-5845-1-option-b\">systematic error<\/label><\/li>\r\n<\/ul>\r\n<\/form><\/div>\r\n<\/section><section id=\"toc_5\" class=\"article-section\">\r\n<h2>Confidence: Reporting uncertainty and error<\/h2>\r\nAs a result of error, scientific measurements are not reported as single <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/value\/pop\">values<\/a>, but rather as ranges or <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/average\/pop\">averages<\/a> with error bars in a graph or \u00b1 sign in a table. <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/Pearson%2C+Karl\/pop\">Karl Pearson<\/a> first described mathematical <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/method\/pop\">methods<\/a> for determining the <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/probability\/pop\">probability<\/a> distributions of scientific measurements, and these methods form the basis of statistical applications in scientific <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/research\/pop\">research<\/a> (see our <a href=\"https:\/\/www.visionlearning.com\/library\/module_viewer.php?mid=155\">Data: Statistics<\/a> module). Statistical techniques allow us to estimate and report the error surrounding a value after repeated measurement of that value. For example, both Libby and Wu reported their estimates as ranges of one <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/standard+deviation\/pop\">standard deviation<\/a> around the <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/mean\/pop\">mean<\/a>, or average, measurement. The standard deviation provides a measure of the range of variability of individual measurements, and specifically, defines a range that encompasses 34.1% of individual measurements above the mean value and 34.1% of those below the mean. The standard deviation of a range of measurements can be used to compute a <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/confidence+interval\/pop\">confidence interval<\/a> around the value.\r\n\r\nConfidence statements do not, as some people believe, provide a measure of how \"correct\" a measurement is. Instead, a confidence statement describes the <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/probability\/pop\">probability<\/a> that a measurement range will overlap the <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/mean\/pop\">mean<\/a> <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/value\/pop\">value<\/a> of a measurement when a study is repeated. This may sound a bit confusing, but consider a study by Yoshikata Morimoto and colleagues, who examined the <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/average\/pop\">average<\/a> pitch speed of eight college baseball players (Morimoto et al., 2003). Each of the pitchers was required to throw six pitches, and the average pitch speed was found to be 34.6 m\/s (77.4 mph) with a 95% <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/confidence+interval\/pop\">confidence interval<\/a> of 34.6 \u00b1 0.2 m\/s (34.4 m\/s to 34.8 m\/s). When he later repeated this study requiring that each of the eight pitchers throw 18 pitches, the average speed was found to be 34.7 m\/s, exactly within the confidence interval obtained during the first study.\r\n\r\nIn this case, there is no \"theoretically correct\" <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/value\/pop\">value<\/a>, but the <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/confidence+interval\/pop\">confidence interval<\/a> provides an estimate of the <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/probability\/pop\">probability<\/a> that a similar result will be found if the study is repeated. Given that Morimoto determined a 95% confidence interval, if he repeated his study 100 times (without exhausting his baseball pitchers), his confidence interval would overlap the <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/mean\/pop\">mean<\/a> pitch speed 95 times, and the other five studies would likely yield pitch speeds that fall outside of his confidence interval.\r\n\r\nIn science, an important indication of confidence within a measurement is the number of significant figures reported. Morimoto reported his measurement to one decimal place (34.6 m\/s) because his instrumentation supported this level of <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/precision\/pop\">precision<\/a>. He was able to distinguish differences in pitches that were 34.6 m\/s and 34.7 m\/s. Had he just rounded his measurements to 35 m\/s, he would have lost a significant amount of detail contained within his <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/data\/pop\">data<\/a>. Further, his instrumentation did not support the precision needed to report additional significant figures (for example, 34.62 m\/s). Incorrectly reporting significant figures can introduce substantial error into a data set.\r\n<div class=\"comprehension-checkpoint\">\r\n<p class=\"leader\">Comprehension Checkpoint<\/p>\r\n<p class=\"question\">Scientific measurements are reported as ranges or with the +\/- sign rather than as single values because<\/p>\r\n\r\n<form class=\"question\" name=\"cc5848\">\r\n<ul class=\"quiz-options\">\r\n \t<li class=\"option-a\"><label class=\"choice\" for=\"q1-5848-0-option-a\">every measurement has some degree of error.<\/label><\/li>\r\n \t<li class=\"option-b\"><label class=\"choice\" for=\"q1-5848-1-option-b\">some scientists are not sure that their calculations are correct.<\/label><\/li>\r\n<\/ul>\r\n<\/form><\/div>\r\n<\/section><section id=\"toc_6\" class=\"article-section\">\r\n<h2>Error propagation<\/h2>\r\nAs Pearson recognized, <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/uncertainty\/pop\">uncertainty<\/a> is <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/inherent\/pop\">inherent<\/a> in scientific <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/research\/pop\">research<\/a>, and for that reason it is critically important for scientists to recognize and account for the errors within a <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/dataset\/pop\">dataset<\/a>. Disregarding the source of an error can result in the propagation and magnification of that error. For example, in 1960 the American mathematician and meteorologist Edward Norton Lorenz was working on a mathematical <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/model\/pop\">model<\/a> for predicting the weather (see our <a href=\"https:\/\/www.visionlearning.com\/library\/module_viewer.php?mid=153\">Modeling in Scientific Research<\/a> module) (Gleick, 1987; Lorenz, 1993). Lorenz was using a Royal McBee computer to iteratively solve 12 equations that expressed relationships such as that between atmospheric pressure and wind speed. Lorenz would input starting <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/value\/pop\">values<\/a> for several <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/variable\/pop\">variables<\/a> into his computer, such as temperature, wind speed, and barometric pressure on a given day at a series of locations. The model would then calculate weather changes over a defined <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/period\/pop\">period<\/a> of time. The model recalculated a single day's worth of weather changes in single minute increments and printed out the new <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/parameter\/pop\">parameters<\/a>.\r\n\r\nOn one occasion, Lorenz decided to rerun a particular <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/model\/pop\">model<\/a> scenario. Instead of starting from the beginning, which would have taken many hours, he decided to pick up in the middle of the run, consulting the printout of <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/parameter\/pop\">parameters<\/a> and re-entering these into his computer. He then left his computer for the hour it would take to recalculate the model, expecting to return and find a weather pattern similar to the one predicted previously.\r\n\r\nUnexpectedly, Lorenz found that the resulting weather prediction was completely different from the original pattern he observed. What Lorenz did not realize at the time was that while his computer stored the numerical <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/value\/pop\">values<\/a> of the <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/model\/pop\">model<\/a> <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/parameter\/pop\">parameters<\/a> to six significant figures (for example 0.639172), his printout, and thus the numbers he inputted when restarting the model, were rounded to three significant figures (0.639). The difference between the two numbers is minute, representing a margin of <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/systematic+error\/pop\">systematic error<\/a> less than 0.1% \u2013 less than one thousandth of the value of each parameter. However, with each <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/iteration\/pop\">iteration<\/a> of his model (and there were thousands of iterations), this error was compounded, multiplying many times over so that his end result was completely different from the first run of the model. As can be seen in Figure 4, the error appears to remain small, but after a few hundred iterations it grows exponentially until reaching a <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/magnitude\/pop\">magnitude<\/a> equivalent to the value of the measurement itself (~0.6).\r\n<figure><img src=\"https:\/\/www.visionlearning.com\/img\/library\/modules\/mid157\/Image\/VLObject-4144-081105111159.jpg\" alt=\"graph - Representations of error propagation\" \/><figcaption>Figure 4: Representation of error propagation in an iterative, dynamic system. After ~1,000 iterations, the error is equivalent to the value of the measurement itself (~0.6), making the calculation fluctuate wildly. Adapted from IMO (2007).<\/figcaption><\/figure>\r\nLorenz published his <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/observation\/pop\">observations<\/a> in the now classic work Deterministic Nonperiodic Flow (Lorenz, 1963). His observations led him to conclude that accurate weather prediction over a period of more than a few weeks was extremely difficult \u2013 perhaps impossible \u2013 because even infinitesimally small errors in the measurement of natural conditions were compounded and quickly reached levels equal to the measurements themselves.\r\n\r\nThe work motivated other researchers to begin looking at other dynamic <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/system\/pop\">systems<\/a> that are similarly sensitive to initial starting conditions, such as the flow of water in a stream or the dynamics of <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/population\/pop\">population<\/a> change. In 1975, the American mathematician and physicist James Yorke and his collaborator, the Chinese-born mathematician Tien-Yien Li, coined the term <em>chaos<\/em> to describe these systems (Li &amp; Yorke, 1975). Again, unlike the common use of the term chaos, which implies randomness or a state of disarray, the science of chaos is not about randomness. Rather, as Lorenz was the first to do, chaos researchers work to understand underlying patterns of behavior in complex systems toward understanding and quantifying this <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/uncertainty\/pop\">uncertainty<\/a>.\r\n<div class=\"comprehension-checkpoint\">\r\n<p class=\"leader\">Comprehension Checkpoint<\/p>\r\n<p class=\"question\">Scientists should look for the source of error within a dataset<\/p>\r\n\r\n<form class=\"question\" name=\"cc5851\">\r\n<ul class=\"quiz-options\">\r\n \t<li class=\"option-a\"><label class=\"choice\" for=\"q1-5851-0-option-a\"><strong>only when<\/strong> the error is very large.<\/label><\/li>\r\n \t<li class=\"option-b\"><label class=\"choice\" for=\"q1-5851-1-option-b\"><strong>even when<\/strong> the error is very small.<\/label><\/li>\r\n<\/ul>\r\n<\/form><\/div>\r\n<\/section><section id=\"toc_7\" class=\"article-section\">\r\n<h2>Recognizing and reducing error<\/h2>\r\nError propagation is not limited to <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/mathematical+modeling\/pop\">mathematical modeling<\/a>. It is always a concern in scientific <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/research\/pop\">research<\/a>, especially in studies that proceed stepwise in multiple increments because error in one step can easily be compounded in the next step. As a result, scientists have developed a number of techniques to help <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/quantify\/pop\">quantify<\/a> error. Here are two examples:\r\n<blockquote><strong>Controls:<\/strong> The use of controls in scientific experiments (see our <a href=\"https:\/\/www.visionlearning.com\/library\/module_viewer.php?mid=150\">Experimentation in Scientific Research<\/a> module) helps quantify statistical error within an experiment and identify systematic error in order to either measure or eliminate it.\r\n\r\n<strong>Blind trials:<\/strong> In research that involves human judgment, such as studies that try to quantify the perception of pain relief following administration of a pain-relieving drug, scientists often work to minimize error by using \"blinds.\" In blind trials, the treatment (i.e. the drug) will be compared to a control (i.e. another drug or a placebo); neither the patient nor the researcher will know if the patient is receiving the treatment or the control. In this way, systematic error due to preconceptions about the utility of a treatment is avoided.<\/blockquote>\r\nError reduction and measurement efforts in scientific <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/research\/pop\">research<\/a> are sometimes referred to as <em>quality assurance<\/em> and <em>quality control<\/em>. Quality assurance generally refers to the plans that a researcher has for minimizing and measuring error in his or her research; quality <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/control\/pop\">control<\/a> refers to the actual procedures implemented in the research. The terms are most commonly used interchangeably and in unison, as in \"quality assurance\/quality control\" (QA\/QC). QA\/QC includes steps such as calibrating instruments or measurements against known standards, reporting all instrument detection limits, implementing standardized procedures to minimize human error, thoroughly documenting research <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/method\/pop\">methods<\/a>, replicating measurements to determine <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/precision\/pop\">precision<\/a>, and a host of other techniques, often specific to the type of research being conducted, and reported in the <em>Materials and Methods<\/em> section of a scientific paper (see our <a href=\"https:\/\/www.visionlearning.com\/library\/module_viewer.php?mid=158\">Understanding Scientific Journals and Articles<\/a> module).\r\n\r\nReduction of <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/statistical+error\/pop\">statistical error<\/a> is often as simple as repeating a <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/research\/pop\">research<\/a> measurement or <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/observation\/pop\">observation<\/a> many times to reduce the <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/uncertainty\/pop\">uncertainty<\/a> in the range of <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/value\/pop\">values<\/a> obtained. <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/systematic+error\/pop\">Systematic error<\/a> can be more difficult to pin down, creeping up in research due to instrumental bias, human mistakes, poor research design, or incorrect assumptions about the behavior of <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/variable\/pop\">variables<\/a> in a <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/system\/pop\">system<\/a>. From this standpoint, identifying and quantifying the source of systematic error in research can help scientists better understand the behavior of the system itself.\r\n\r\n<\/section><section id=\"toc_8\" class=\"article-section\">\r\n<h2>Uncertainty as a state of nature<\/h2>\r\nWhile <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/Pearson%2C+Karl\/pop\">Karl Pearson<\/a> proposed that individual measurements could not yield exact <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/value\/pop\">values<\/a>, he felt that careful and repeated scientific investigation coupled with statistical <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/analysis\/pop\">analysis<\/a> could allow one to determine the true value of a measurement. A younger contemporary of Pearson's, the English statistician Ronald Aylmer Fisher, extended and, at the same time, contradicted this concept. Fisher felt that because all measurements contained <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/inherent\/pop\">inherent<\/a> error, one could never identify the exact or \"correct\" value of a measurement. According to Fisher, the true distribution of a measurement is unattainable; statistical techniques therefore do not estimate the \"true\" value of a measurement, but rather they are used to minimize error and develop range estimates that approximate the theoretically correct value of the measurement. A natural consequence of his idea is that occasionally the approximation may be incorrect.\r\n\r\nIn the first half of the 20<sup>th<\/sup> century, the concept of <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/uncertainty\/pop\">uncertainty<\/a> reached new heights with the discovery of <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/quantum+mechanics\/pop\">quantum mechanics<\/a>. In the quantum world, uncertainty is not an inconvenience; it is a state of being. For example, the <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/decay\/pop\">decay<\/a> of a radioactive <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/element\/pop\">element<\/a> is inherently an uncertain event. We can predict the <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/probability\/pop\">probability<\/a> of the decay profile of a <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/mass\/pop\">mass<\/a> of radioactive <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/atom\/pop\">atoms<\/a>, but we can never predict the exact time that an individual radioactive atom will undergo decay. Or consider the <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/Heisenberg+Uncertainty+Principle\/pop\">Heisenberg Uncertainty Principle<\/a> in quantum physics, which states that measuring the position of a <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/particle\/pop\">particle<\/a> makes the momentum of the particle inherently uncertain, and, conversely, measuring the particle's momentum makes its position inherently uncertain.\r\n\r\nOnce we understand the concept of <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/uncertainty\/pop\">uncertainty<\/a> as it applies to science, we can begin to see that the purpose of scientific <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/data\/pop\">data<\/a> <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/analysis\/pop\">analysis<\/a> is to identify and <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/quantify\/pop\">quantify<\/a> error and variability toward uncovering the relationships, patterns, and behaviors that occur in nature. Scientific knowledge itself continues to evolve as new data and new studies help us understand and quantify uncertainty in the natural world.\r\n\r\n<\/section><section id=\"toc-999\" class=\"article-section\">\r\n<h3>Summary<\/h3>\r\nThere is uncertainty in all scientific data, and even the best scientists find some degree of error in their measurements. This module uses familiar topics - playing baseball, shooting targets, and calculating the age of an object - to show how scientists identify and measure error and uncertainty, which are reported in terms of confidence.\r\n<h3>Key Concepts<\/h3>\r\n<ul class=\"bulleted\">\r\n \t<li>Uncertainty is the quantitative estimation of error present in data; all measurements contain some uncertainty generated through systematic error and\/or random error.<\/li>\r\n \t<li>Acknowledging the uncertainty of data is an important component of reporting the results of scientific investigation.<\/li>\r\n \t<li>Uncertainty is commonly misunderstood to mean that scientists are not certain of their results, but the term specifies the degree to which scientists are confident in their data.<\/li>\r\n \t<li>Careful methodology can reduce uncertainty by correcting for systematic error and minimizing random error. However, uncertainty can never be reduced to zero.<\/li>\r\n<\/ul>\r\n<\/section><footer>\r\n<ul class=\"indented links\">\r\n \t<li>\r\n<h5>Further Reading<\/h5>\r\n<\/li>\r\n \t<li><a href=\"https:\/\/www.visionlearning.com\/en\/library\/Process-of-Science\/49\/Using-Graphs-and-Visual-Data-in-Science\/156\">Using Graphs and Visual Data in Science<\/a><\/li>\r\n \t<li><a href=\"https:\/\/www.visionlearning.com\/en\/library\/Process-of-Science\/49\/Statistics-in-Science\/155\">Statistics in Science<\/a><\/li>\r\n \t<li><a href=\"https:\/\/www.visionlearning.com\/en\/library\/Process-of-Science\/49\/Data-Analysis-and-Interpretation\/154\">Data Analysis and Interpretation<\/a><\/li>\r\n<\/ul>\r\n<a name=\"refs\"><\/a>\r\n<ul class=\"indented list\">\r\n \t<li>\r\n<h5>References<\/h5>\r\n<\/li>\r\n \t<li>Arnold, J. R., &amp; Libby, W. F. (1949). Age determinations by radiocarbon content: Checks with samples of known age. <em>Science, 110<\/em>, 678-680.<\/li>\r\n \t<li>Damon, P. E., Ferguson, C. W., Long, A., &amp; Wallick, E. I. (1974). Dendrochronologic calibration of the radiocarbon time scale. <em>American Antiquity, 39<\/em>(2), 350-366.<\/li>\r\n \t<li>Fairbanks, R. G., Mortlock, R. A., Chiu, T.-C., Cao, L., Kaplan, A., Guilderson, T. P., . . . Nadeau, M. (2005). Radiocarbon calibration curve spanning 0 to 50,000 years BP based on paired 230Th\/ 234U\/ 238U and 14C dates on pristine corals. <em>Quaternary Science Reviews, 24,<\/em> 1781-1796.<\/li>\r\n \t<li>Gleick, J. (1987) <em>Chaos: Making a new science.<\/em> New York: Penguin Books.<\/li>\r\n \t<li>IMO. (2007). Long range weather prediction.<em> The Icelandic Meteorological Office.<\/em> Retrieved December 18, 2007, from http:\/\/andvari.vedur.is\/~halldor\/HB\/Met210old\/pred.html<\/li>\r\n \t<li>Li, T. Y., &amp; Yorke, J. A. (1975). Period three implies chaos. <em>American Mathematical Monthly, 82,<\/em> 985.<\/li>\r\n \t<li>Libby, W. F. (1946). Atmospheric helium three and radiocarbon from cosmic radiation. <em>Physical Review, 69<\/em>(11-12), 671-672.<\/li>\r\n \t<li>Libby, W. F. (1963). Accuracy of radiocarbon dates. <em>Science, 140,<\/em> 278-280.<\/li>\r\n \t<li>Libby, W. F., Anderson, E. C., &amp; Arnold, J. R. (1949). Age determination by radiocarbon content: World-wide assay of natural radiocarbon. <em>Science, 109<\/em>(2827), 227-228.<\/li>\r\n \t<li>Lorenz, E. (1963). Deterministic nonperiodic flow. <em>Journal of the Atmospheric Sciences, 20,<\/em> 130-141.<\/li>\r\n \t<li>Lorenz, E. (1993). <em>The essence of chaos<\/em>. The University of Washington Press.<\/li>\r\n \t<li>Morimoto, Y., Ito, K., Kawamura, T., &amp; Muraki, Y. (2003). Immediate effect of assisted and resisted training using different weight balls on ball speed and accuracy in baseball pitching. <em>International Journal of Sport and Health Science, 1<\/em>(2), 238-246.<\/li>\r\n \t<li>Peat, F. D. (2002). <em>From certainty to uncertainty: The story of science and ideas in the twentieth century<\/em>. Joseph Henry Press, National Academies Press.<\/li>\r\n \t<li>Salsburg, D. (2001). <em>The lady tasting tea: How statistics revolutionized science in the twentieth century<\/em>. New York: W. H. Freeman &amp; Company.<\/li>\r\n \t<li>Wagner, C. H. (1983). Uncertainty in science and statistics. <em>The Two-Year College Mathematics Journal, 14<\/em>(4), 360-363.<\/li>\r\n \t<li>Wu, X., Yuan, S., Wang, J., Guo, Z., Liu, K., Lu, X., . . . Cai, L. (2000). AMS radiocarbon dating of cemetery of Jin Marquises in China. <em>Nuclear Instruments and Methods in Physics Research, B, 172<\/em>(1-4), 732-735.<\/li>\r\n<\/ul>\r\n<\/footer>","rendered":"<div class=\"article-introduction\">\n<p>The Olympic sport of biathlon (Figure 1) is a cross-country ski race of 20 km in which the athletes stop on four occasions to shoot 0.57 cm diameter bullets from a .22 caliber rifle at targets. The sport requires not only great endurance, but exceptional <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/accuracy\/pop\">accuracy<\/a> as the athletes shoot on two occasions from the prone position (lying down) and on two occasions while standing. The targets the athletes aim for are all 50 m away, but the size varies to match the <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/precision\/pop\">precision<\/a> expected of them; those targeted while shooting in the prone position are 4.5 cm in diameter while those targeted from the more difficult standing position are 11.5 cm in diameter. In both cases, however, the diameter of the target is many times larger than the diameter of the bullet itself \u2013 why?<\/p>\n<figure><a title=\"&lt;strong&gt;Figure 1:&lt;\/strong&gt; Biathletes in the shooting area of a competition.\" href=\"https:\/\/www.visionlearning.com\/images\/figure-images\/157-a-2x.jpg\"> <img decoding=\"async\" src=\"https:\/\/www.visionlearning.com\/images\/figure-images\/157-a.jpg\" alt=\"Biathletes in the shooting area of a competition\" \/> <\/a><figcaption><strong>Figure 1:<\/strong> Biathletes in the shooting area of a competition. <span class=\"credit\">image \u00a9 Marc Schuelper<\/span><\/figcaption><\/figure>\n<p>While the legend of Robin Hood splitting one arrow with another is well-known, it is also unrealistic. Biathlon targets are purposely sized many times larger than the bullets the athletes shoot to account for the <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/inherent\/pop\">inherent<\/a> error and <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/uncertainty\/pop\">uncertainty<\/a> involved in long distance riflery. Even the most skilled marksman cannot account for every <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/variable\/pop\">variable<\/a> affecting the path of the bullet, like sudden gusts of wind or variations in air pressure. Shooting from the standing position involves even greater uncertainty, as indicated by the larger targets used, because even the simple rise and fall of an athlete&#8217;s chest as they breathe can affect the aim of their rifle.<\/p>\n<\/div>\n<section id=\"toc_1\" class=\"article-section\">\n<h2>Categorizing uncertainty: Accuracy vs. precision<\/h2>\n<p>Scientific measurements also incorporate variability, and scientists report this as <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/uncertainty\/pop\">uncertainty<\/a> in an effort to share with others the level of error that they found acceptable in their measurements. But uncertainty in science does not imply doubt as it does in everyday use. Scientific uncertainty is a quantitative measurement of variability in the <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/data\/pop\">data<\/a>. In other words, uncertainty in science refers to the idea that all data have a range of expected <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/value\/pop\">values<\/a> as opposed to a precise point value. This uncertainty can be categorized in two ways: <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/accuracy\/pop\">accuracy<\/a> and <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/precision\/pop\">precision<\/a>.<\/p>\n<ul class=\"bulleted\">\n<li><em>Accuracy<\/em> is a term that describes how well a measurement approximates the theoretically correct <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/value\/pop\">value<\/a> of that measurement, for example, how close the arrow is to the bullseye (Figure 2).<\/li>\n<li>The term <em>precision<\/em>, by comparison, describes the degree to which individual measurements vary around a central value. Measurements with high <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/precision\/pop\">precision<\/a> are highly reproducible because repeated measurement will reliably give a similar result; however, they may or may not be accurate (Figure 2).<\/li>\n<\/ul>\n<figure><img decoding=\"async\" src=\"https:\/\/www.visionlearning.com\/img\/library\/modules\/mid157\/Image\/VLObject-4142-081105111152.jpg\" alt=\"targets\" \/><figcaption>Figure 2: A representation of accuracy and precision as hits on a target. The target at left depicts good accuracy as the marks are close to the bullseye, but poor precision; in contrast, the target at right depicts good precision as the marks are grouped closely, but poor accuracy.<\/figcaption><\/figure>\n<\/section>\n<section id=\"toc_2\" class=\"article-section\">\n<h2>Uncertainty in nature<\/h2>\n<p>Karl Pearson, the English statistician and geneticist, is commonly credited with first describing the concept of <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/uncertainty\/pop\">uncertainty<\/a> as a measure of <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/data\/pop\">data<\/a> variability in the late 1800s (Salsburg, 2001). Before Pearson, scientists realized that their measurements incorporated variability, but they assumed that this variability was simply due to error. For example, measurement of the orbits of planets around the sun taken by different scientists at different times varied, and this variability was thought to be due to errors caused by inadequate instrumentation. The French mathematician <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/Laplace%2C+Pierre~Simon\/pop\">Pierre-Simon Laplace<\/a> discussed a <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/method\/pop\">method<\/a> for quantifying error distributions of astronomical measurements caused by small errors associated with instrument shortcomings as early as 1820. As technology improved through the 1800s, astronomers realized that they could reduce, but not eliminate this error in their measurements.<\/p>\n<p>Pearson put forward a revolutionary idea: <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/uncertainty\/pop\">Uncertainty<\/a>, he proposed, was not simply due to the limits of technology in measuring certain events \u2013 it was <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/inherent\/pop\">inherent<\/a> in nature. Even the most careful and rigorous scientific investigation (or any type of investigation for that matter) could not yield an exact measurement. Rather, repeating an investigation would yield a scatter of measurements that are distributed around some central <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/value\/pop\">value<\/a>. This scatter would be caused not only by error, but also by natural variability. In other words, measurements themselves, independent of any human or instrument inaccuracies, exhibit scatter.<\/p>\n<p>Whether it is the flight path of an arrow, the resting heart rate of an adult male, or the age of a historical artifact, measurements do not have exact <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/value\/pop\">values<\/a>, but instead always exhibit a range of values, and that range can be quantified as <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/uncertainty\/pop\">uncertainty<\/a>. This uncertainty can be expressed as a plot of the <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/probability\/pop\">probability<\/a> of obtaining a certain value, and the probabilities are distributed about some central, or <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/mean\/pop\">mean<\/a>, value.<\/p>\n<div class=\"comprehension-checkpoint\">\n<p class=\"leader\">Comprehension Checkpoint<\/p>\n<p class=\"question\">Certain large scale scientific measurements, such as the orbit of planets, have no uncertainty associated with them.<\/p>\n<form class=\"question\" action=\"action\" id=\"cc5836\">\n<ul class=\"quiz-options\">\n<li class=\"option-a\"><label class=\"choice\" for=\"q1-5836-0-option-a\">True.<\/label><\/li>\n<li class=\"option-b\"><label class=\"choice\" for=\"q1-5836-1-option-b\">False.<\/label><\/li>\n<\/ul>\n<\/form>\n<\/div>\n<\/section>\n<section id=\"toc_3\" class=\"article-section\">\n<h2>Uncertainty and error in practice: Carbon-14 dating<\/h2>\n<p>Archaeologists, paleontologists, and other researchers have long been interested in dating objects and artifacts in an effort to understand their history and use. Unfortunately, written <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/record\/pop\">records<\/a> are a relatively recent human invention, and few historical artifacts are accompanied by precise written histories.<\/p>\n<p>In the first half of the 20th century, an American nuclear chemist by the name of Willard F. Libby became interested in using the radioactive <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/isotope\/pop\">isotope<\/a> <sup>14<\/sup>C to date certain objects. The <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/theory\/pop\">theory<\/a> of radiocarbon dating is relatively simple. Most carbon in the Earth&#8217;s <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/atmosphere\/pop\">atmosphere<\/a> is in the form of <sup>12<\/sup>C, but a small amount of the isotope <sup>14<\/sup>C is produced naturally through the bombardment of <sup>14<\/sup>N with cosmic rays (W. F. Libby, 1946). As plants take up carbon from the atmosphere through <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/respiration\/pop\">respiration<\/a>, they incorporate both <sup>14<\/sup>C as well as the more abundant <sup>12<\/sup>C into their tissues. Animals also take up both carbon isotopes through the foods that they eat. Thus, all living <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/organism\/pop\">organisms<\/a> have the same <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/ratio\/pop\">ratio<\/a> of <sup>14<\/sup>C and <sup>12<\/sup>C isotopes in their body as the atmosphere.<\/p>\n<p>Unlike <sup>12<\/sup>C, <sup>14<\/sup>C is a radioactive <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/isotope\/pop\">isotope<\/a> that is constantly undergoing <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/decay\/pop\">decay<\/a> to its <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/daughter\/pop\">daughter<\/a> <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/product\/pop\">product<\/a> <sup>14<\/sup>N at a known rate. While an <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/organism\/pop\">organism<\/a> is alive, it is taking up new <sup>14<\/sup>C from the <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/environment\/pop\">environment<\/a> and thus remains in <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/equilibrium\/pop\">equilibrium<\/a> with it. When that organism dies, however, the carbon in its tissues is no longer replaced, and the amount of <sup>14<\/sup>C slowly decreases in time as it decays to <sup>14<\/sup>N. Thus, the amount of radioactive <sup>14<\/sup>C remaining in a piece of wood or an animal bone can be used to determine when that organism died. In essence, the longer the organism has been dead, the lower the <sup>14<\/sup>C levels.<\/p>\n<p>The amount of radioactive material (such as <sup>14<\/sup>C) in a sample can be quantified by counting the number of <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/decay\/pop\">decays<\/a> that the material undergoes in a specific amount of time, usually reported in counts per minute (cpm). When Libby began his radiocarbon work in the 1940s, the technology available was still quite new. A simple Geiger counter was only first invented in 1908 by the German scientist Hans Wilhelm Geiger, a student of <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/Rutherford%2C+Ernest\/pop\">Ernest Rutherford<\/a>&#8216;s, and it was not perfected until 1928 when Walther M\u00fcller, a student of Geiger&#8217;s, improved on the design, allowing it to detect all types of <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/radiation\/pop\">radiation<\/a>. Libby himself is credited with building the first Geiger counter in the United States in the 1930s.<\/p>\n<p>Libby, however, faced a major hurdle with using the instrument to measure <sup>14<\/sup>C. The problem was that naturally-occurring background <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/radiation\/pop\">radiation<\/a> from cosmic rays and Earth, along with the variability associated with that background signal, would overwhelm the small <sup>14<\/sup>C signal he expected to see. In 1949, he reported on a <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/method\/pop\">method<\/a> for reducing the background signal and variability: He placed the entire sample and the detector inside of a tube shielded by 2 inches of lead and 4 inches of iron (W. F. Libby, Anderson, &amp; Arnold, 1949). In this way, Libby and his colleagues reduced the background signal from 150 cpm to 10 cpm and minimized the variability associated with the signal to &#8220;about 5-10% error,&#8221; or less than 1 cpm.<\/p>\n<p>Libby and colleagues do not use the word <em>error<\/em> as we do in common language, where it refers to a mistake such as a typographical error or a baseball error. The Latin origin of the word error (<em>errorem<\/em>) means <em>wandering<\/em> or <em>straying<\/em>, and the scientific use of the word is closer to this original meaning. In science, <em>error<\/em> is the difference between the true <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/value\/pop\">value<\/a> and the measured value, and that difference can have many different causes. Libby calculated the error associated with his measurements by counting the number of <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/decay\/pop\">decay<\/a> events in the sample in a known amount of time, repeating the measurement over multiple periods, and then using statistical techniques to <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/quantify\/pop\">quantify<\/a> the error (see our <a href=\"https:\/\/www.visionlearning.com\/library\/module_viewer.php?mid=155\">Statistics in Science<\/a> module).<\/p>\n<p>In 1949, Libby, working with his post-doctoral student James Arnold, reported the first use of radiocarbon dating for determining the age of wood fragments from archaeological sites around the world (Arnold &amp; Libby, 1949). Because the <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/method\/pop\">method<\/a> was new, Arnold and Libby were careful to replicate their measurements to provide a detailed estimate of different types of error, and they compared the results of their method with samples of a known age as a <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/control\/pop\">control<\/a> (Table 1).<\/p>\n<div class=\"table-container\">\n<div class=\"table-container\">\n<table class=\"generic-data size-80\">\n<thead>\n<tr>\n<th rowspan=\"2\">Sample<\/th>\n<th>Specific activity (cpm\/g of carbon)<\/th>\n<th colspan=\"2\">Age (years)<\/th>\n<\/tr>\n<\/thead>\n<tbody>\n<tr>\n<td><\/td>\n<td>Found<\/td>\n<td>Found<\/td>\n<td>Expected<\/td>\n<\/tr>\n<tr>\n<td>Tree Ring<\/td>\n<td>11.10 \u00b1 0.31<\/td>\n<td>1100 \u00b1 150<\/td>\n<td>1372 \u00b1 50<\/td>\n<\/tr>\n<tr>\n<td><\/td>\n<td>11.52 \u00b1 0.35<\/td>\n<td><\/td>\n<td><\/td>\n<\/tr>\n<tr>\n<td><\/td>\n<td>11.34 \u00b1 0.25<\/td>\n<td><\/td>\n<td><\/td>\n<\/tr>\n<tr>\n<td><\/td>\n<td>10.15 \u00b1 0.44<\/td>\n<td><\/td>\n<td><\/td>\n<\/tr>\n<tr>\n<td><\/td>\n<td>11.08 \u00b1 0.31<\/td>\n<td><\/td>\n<td><\/td>\n<\/tr>\n<tr>\n<td><\/td>\n<td>Average: 10.99 \u00b1 0.15<\/td>\n<td><\/td>\n<td><\/td>\n<\/tr>\n<tr>\n<td colspan=\"4\">\n<figure><strong>Table 1<\/strong>: Age determinations on samples of known age from Arnold &amp; Libby (1949). The specific activities for five different replicates of a sample of wood from a Douglas fir excavated from the Red Rock Valley are shown in the second column of Table 1. Each individual measurement has an error shown to the right of it, indicated by the \u00b1 sign. Arnold and Libby describe these measurements in their paper, stating, &#8220;The errors quoted for the specific activity measurements are standard deviations as computed from the Poisson statistics of counting random events.&#8221; In other words, the individual error is calculated from the expected uncertainties associated with radioactive decay for each sample.<\/figure>\n<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n<\/div>\n<\/div>\n<p>The specific activities for five different replicates of a sample of wood from a Douglas fir excavated from the Red Rock Valley are shown in the second column of Table 1. Each individual measurement has an error shown to the right of it, indicated by the \u00b1 sign. Arnold and Libby describe these measurements in their paper, stating, &#8220;The errors quoted for the specific activity measurements are <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/standard+deviation\/pop\">standard deviations<\/a> as computed from the Poisson <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/statistic\/pop\">statistics<\/a> of counting random events.&#8221; In other words, the individual error is calculated from the expected uncertainties associated with radioactive <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/decay\/pop\">decay<\/a> for each sample.<\/p>\n<\/section>\n<section id=\"toc_4\" class=\"article-section\">\n<h2>Statistical vs. systematic error<\/h2>\n<p>As seen in Table 1, an <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/average\/pop\">average<\/a> specific activity <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/value\/pop\">value<\/a> (10.99) is provided at the bottom with an overall error. The overall error (0.15) is smaller than the individual error reported with each measurement. This is an important feature of the statistical calculation of error associated with scientific <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/data\/pop\">data<\/a> \u2013 as you increase the number of measurements of a value, you decrease the <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/uncertainty\/pop\">uncertainty<\/a> and increase the confidence associated with the approximation of the value. The error reported alongside of the specific activity provides a measure of the <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/precision\/pop\">precision<\/a> of the value, and is commonly referred to as <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/statistical+error\/pop\">statistical error<\/a>. Statistical error is what Pearson described as the <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/inherent\/pop\">inherent<\/a> uncertainty of a measurement. It is caused by random fluctuations within a <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/system\/pop\">system<\/a>, such as the random fluctuation of radioactive <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/decay\/pop\">decay<\/a>, and is sometimes referred to as <em>random error<\/em> as the researcher has little control over it. Statistical error cannot be eliminated, as Pearson described, but it can be measured and reduced by conducting repeated <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/observation\/pop\">observations<\/a> of a specific event.<\/p>\n<p>In column 3 of Table 1, Arnold and Libby estimate the age of the Douglas fir sample based on the <sup>14<\/sup>C activity as 1100 years old (placing its first season of growth in the year 849 CE). In column 4 of Table 1, they report the actual age of the Douglas fir as calculated by counting tree rings in the sample as 1372 years old (placing its first season in the year 577 CE). By comparing the <sup>14<\/sup>C age to a theoretically correct <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/value\/pop\">value<\/a> as determined by the tree ring count, Arnold and Libby allow the reader to gauge the <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/accuracy\/pop\">accuracy<\/a> of their <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/method\/pop\">method<\/a>, and this provides a measure of a second type of error encountered in science: <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/systematic+error\/pop\">systematic error<\/a>. <em>Statistical error<\/em> cannot be eliminated, but it can be measured and reduced by conducting repeated <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/observation\/pop\">observations<\/a> of a specific event. <em>Systematic error<\/em>, in contrast, can be corrected for \u2013 like if you know your oven is 50\u00b0 too hot, you set it for 300\u00b0 rather than 350\u00b0.<\/p>\n<p>Based on their <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/data\/pop\">data<\/a>, Arnold and Libby state that the &#8220;agreement between prediction and <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/observation\/pop\">observation<\/a> is seen to be satisfactory.&#8221; However, as he continued to do <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/research\/pop\">research<\/a> to establish the <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/method\/pop\">method<\/a> of <sup>14<\/sup>C dating, Libby began to recognize that the discrepancy between radiocarbon dating and other methods was even larger for older objects, especially those greater than 4,000 years old (W.F. Libby, 1963). Where theoretically correct dates on very old objects could be established by other means, such as in samples from the temples of Egypt, where a calendar <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/system\/pop\">system<\/a> was well-established, the ages obtained through the radiocarbon dating method (the &#8220;found&#8221; ages in Table 1) were consistently older than the &#8220;expected&#8221; dates, often by as much as 500 years.<\/p>\n<p>Libby knew that there was bound to be <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/statistical+error\/pop\">statistical error<\/a> in these measurements and had anticipated using <sup>14<\/sup>C dating to calculate a range of dates for objects. But the problem he encountered was different: <sup>14<\/sup>C-dating systematically calculated ages that differed by as much as 500 years from the actual ages of older objects. <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/systematic+error\/pop\">Systematic error<\/a>, like Libby encountered, is due to an unknown but non-random fluctuation, like instrumental bias or a faulty assumption. The radiocarbon dating <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/method\/pop\">method<\/a> had achieved good <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/precision\/pop\">precision<\/a>, replicate analyses gave dates within 150 years of one another as seen in Table 1; but initially it showed poor <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/accuracy\/pop\">accuracy<\/a> \u2013 the &#8220;found&#8221; <sup>14<\/sup>C age of the Douglas fir was almost 300 years different than the &#8220;expected&#8221; age, and other objects were off by some 500 years.<\/p>\n<p>Unlike <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/statistical+error\/pop\">statistical error<\/a>, <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/systematic+error\/pop\">systematic error<\/a> can be compensated for, or sometimes even eliminated if its source can be identified. In the case of <sup>14<\/sup>C-dating, it was later discovered that the reason for the systematic error was a faulty assumption: Libby and many other scientists had assumed that the production rate for <sup>14<\/sup>C in the <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/atmosphere\/pop\">atmosphere<\/a> was constant over time, but it is not. Instead, it fluctuates with changes in Earth&#8217;s magnetic field, the uptake of carbon by plants, and other factors. In addition, levels of radioactive <sup>14<\/sup>C increased through the 20<sup>th<\/sup> century because nuclear weapons testing released high levels of <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/radiation\/pop\">radiation<\/a> to the atmosphere.<\/p>\n<figure><a title=\"&lt;strong&gt;Figure 3:&lt;\/strong&gt; Tree-ring dates have been used to recalibrate the radiocarbon dating method.\" href=\"https:\/\/www.visionlearning.com\/images\/figure-images\/157-c-2x.jpg\"> <img decoding=\"async\" src=\"https:\/\/www.visionlearning.com\/images\/figure-images\/157-c.jpg\" alt=\"Tree-ring dates have been used to recalibrate the radiocarbon dating method\" \/> <\/a><figcaption><strong>Figure 3:<\/strong> Tree-ring dates have been used to recalibrate the radiocarbon dating method. <span class=\"credit\">image \u00a9 Hannes Grobe<\/span><\/figcaption><\/figure>\n<p>In the decades since Libby first published his <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/method\/pop\">method<\/a>, researchers have recalibrated the radiocarbon dating method with tree-ring dates from bristlecone pine trees (Damon et al., 1974) and corals (Fairbanks et al., 2005) to correct for the fluctuations in the production of <sup>14<\/sup>C in the <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/atmosphere\/pop\">atmosphere<\/a>. As a result, both the <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/precision\/pop\">precision<\/a> and <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/accuracy\/pop\">accuracy<\/a> of radiocarbon dates have increased dramatically. For example, in 2000, Xiaohong Wu and colleagues at Peking University in Beijing used radiocarbon dating on bones of the Marquises (lords) of Jin recovered from a cemetery in Shanxi Province in China (see Table 2) (Wu et al., 2000). As seen in Table 2, not only is the precision of the estimates (ranging from 18 to 44 years) much tighter than Libby&#8217;s reported 150 year error range for the Douglas fir samples, but the radiocarbon dates are highly accurate, with the reported deaths dates of the Jin (the theoretically correct values) falling within the <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/statistical+error\/pop\">statistical error<\/a> ranges reported in all three cases.<\/p>\n<div class=\"table-container\">\n<div class=\"table-container\">\n<table class=\"generic-data size-80\">\n<thead>\n<tr>\n<th>Name of Jin Marquis<\/th>\n<th>Radiocarbon Date (BCE)<\/th>\n<th>Documented Death Date (BCE)<\/th>\n<\/tr>\n<\/thead>\n<tbody>\n<tr>\n<td>Jing<\/td>\n<td>860-816<\/td>\n<td>841<\/td>\n<\/tr>\n<tr>\n<td>Li<\/td>\n<td>834-804<\/td>\n<td>823<\/td>\n<\/tr>\n<tr>\n<td>Xian<\/td>\n<td>814-796<\/td>\n<td>812<\/td>\n<\/tr>\n<tr>\n<td colspan=\"3\">\n<figure><strong>Table 2<\/strong>: Radiocarbon estimates and documented death dates of three of the Marquises of Jin from Wu et al., 2000).<\/figure>\n<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n<\/div>\n<\/div>\n<div class=\"comprehension-checkpoint\">\n<p class=\"leader\">Comprehension Checkpoint<\/p>\n<p class=\"question\">Which type of error is not random and can be compensated for?<\/p>\n<form class=\"question\" action=\"action\" id=\"cc5845\">\n<ul class=\"quiz-options\">\n<li class=\"option-a\"><label class=\"choice\" for=\"q1-5845-0-option-a\">statistical error<\/label><\/li>\n<li class=\"option-b\"><label class=\"choice\" for=\"q1-5845-1-option-b\">systematic error<\/label><\/li>\n<\/ul>\n<\/form>\n<\/div>\n<\/section>\n<section id=\"toc_5\" class=\"article-section\">\n<h2>Confidence: Reporting uncertainty and error<\/h2>\n<p>As a result of error, scientific measurements are not reported as single <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/value\/pop\">values<\/a>, but rather as ranges or <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/average\/pop\">averages<\/a> with error bars in a graph or \u00b1 sign in a table. <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/Pearson%2C+Karl\/pop\">Karl Pearson<\/a> first described mathematical <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/method\/pop\">methods<\/a> for determining the <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/probability\/pop\">probability<\/a> distributions of scientific measurements, and these methods form the basis of statistical applications in scientific <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/research\/pop\">research<\/a> (see our <a href=\"https:\/\/www.visionlearning.com\/library\/module_viewer.php?mid=155\">Data: Statistics<\/a> module). Statistical techniques allow us to estimate and report the error surrounding a value after repeated measurement of that value. For example, both Libby and Wu reported their estimates as ranges of one <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/standard+deviation\/pop\">standard deviation<\/a> around the <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/mean\/pop\">mean<\/a>, or average, measurement. The standard deviation provides a measure of the range of variability of individual measurements, and specifically, defines a range that encompasses 34.1% of individual measurements above the mean value and 34.1% of those below the mean. The standard deviation of a range of measurements can be used to compute a <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/confidence+interval\/pop\">confidence interval<\/a> around the value.<\/p>\n<p>Confidence statements do not, as some people believe, provide a measure of how &#8220;correct&#8221; a measurement is. Instead, a confidence statement describes the <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/probability\/pop\">probability<\/a> that a measurement range will overlap the <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/mean\/pop\">mean<\/a> <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/value\/pop\">value<\/a> of a measurement when a study is repeated. This may sound a bit confusing, but consider a study by Yoshikata Morimoto and colleagues, who examined the <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/average\/pop\">average<\/a> pitch speed of eight college baseball players (Morimoto et al., 2003). Each of the pitchers was required to throw six pitches, and the average pitch speed was found to be 34.6 m\/s (77.4 mph) with a 95% <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/confidence+interval\/pop\">confidence interval<\/a> of 34.6 \u00b1 0.2 m\/s (34.4 m\/s to 34.8 m\/s). When he later repeated this study requiring that each of the eight pitchers throw 18 pitches, the average speed was found to be 34.7 m\/s, exactly within the confidence interval obtained during the first study.<\/p>\n<p>In this case, there is no &#8220;theoretically correct&#8221; <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/value\/pop\">value<\/a>, but the <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/confidence+interval\/pop\">confidence interval<\/a> provides an estimate of the <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/probability\/pop\">probability<\/a> that a similar result will be found if the study is repeated. Given that Morimoto determined a 95% confidence interval, if he repeated his study 100 times (without exhausting his baseball pitchers), his confidence interval would overlap the <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/mean\/pop\">mean<\/a> pitch speed 95 times, and the other five studies would likely yield pitch speeds that fall outside of his confidence interval.<\/p>\n<p>In science, an important indication of confidence within a measurement is the number of significant figures reported. Morimoto reported his measurement to one decimal place (34.6 m\/s) because his instrumentation supported this level of <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/precision\/pop\">precision<\/a>. He was able to distinguish differences in pitches that were 34.6 m\/s and 34.7 m\/s. Had he just rounded his measurements to 35 m\/s, he would have lost a significant amount of detail contained within his <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/data\/pop\">data<\/a>. Further, his instrumentation did not support the precision needed to report additional significant figures (for example, 34.62 m\/s). Incorrectly reporting significant figures can introduce substantial error into a data set.<\/p>\n<div class=\"comprehension-checkpoint\">\n<p class=\"leader\">Comprehension Checkpoint<\/p>\n<p class=\"question\">Scientific measurements are reported as ranges or with the +\/- sign rather than as single values because<\/p>\n<form class=\"question\" action=\"action\" id=\"cc5848\">\n<ul class=\"quiz-options\">\n<li class=\"option-a\"><label class=\"choice\" for=\"q1-5848-0-option-a\">every measurement has some degree of error.<\/label><\/li>\n<li class=\"option-b\"><label class=\"choice\" for=\"q1-5848-1-option-b\">some scientists are not sure that their calculations are correct.<\/label><\/li>\n<\/ul>\n<\/form>\n<\/div>\n<\/section>\n<section id=\"toc_6\" class=\"article-section\">\n<h2>Error propagation<\/h2>\n<p>As Pearson recognized, <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/uncertainty\/pop\">uncertainty<\/a> is <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/inherent\/pop\">inherent<\/a> in scientific <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/research\/pop\">research<\/a>, and for that reason it is critically important for scientists to recognize and account for the errors within a <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/dataset\/pop\">dataset<\/a>. Disregarding the source of an error can result in the propagation and magnification of that error. For example, in 1960 the American mathematician and meteorologist Edward Norton Lorenz was working on a mathematical <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/model\/pop\">model<\/a> for predicting the weather (see our <a href=\"https:\/\/www.visionlearning.com\/library\/module_viewer.php?mid=153\">Modeling in Scientific Research<\/a> module) (Gleick, 1987; Lorenz, 1993). Lorenz was using a Royal McBee computer to iteratively solve 12 equations that expressed relationships such as that between atmospheric pressure and wind speed. Lorenz would input starting <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/value\/pop\">values<\/a> for several <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/variable\/pop\">variables<\/a> into his computer, such as temperature, wind speed, and barometric pressure on a given day at a series of locations. The model would then calculate weather changes over a defined <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/period\/pop\">period<\/a> of time. The model recalculated a single day&#8217;s worth of weather changes in single minute increments and printed out the new <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/parameter\/pop\">parameters<\/a>.<\/p>\n<p>On one occasion, Lorenz decided to rerun a particular <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/model\/pop\">model<\/a> scenario. Instead of starting from the beginning, which would have taken many hours, he decided to pick up in the middle of the run, consulting the printout of <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/parameter\/pop\">parameters<\/a> and re-entering these into his computer. He then left his computer for the hour it would take to recalculate the model, expecting to return and find a weather pattern similar to the one predicted previously.<\/p>\n<p>Unexpectedly, Lorenz found that the resulting weather prediction was completely different from the original pattern he observed. What Lorenz did not realize at the time was that while his computer stored the numerical <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/value\/pop\">values<\/a> of the <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/model\/pop\">model<\/a> <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/parameter\/pop\">parameters<\/a> to six significant figures (for example 0.639172), his printout, and thus the numbers he inputted when restarting the model, were rounded to three significant figures (0.639). The difference between the two numbers is minute, representing a margin of <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/systematic+error\/pop\">systematic error<\/a> less than 0.1% \u2013 less than one thousandth of the value of each parameter. However, with each <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/iteration\/pop\">iteration<\/a> of his model (and there were thousands of iterations), this error was compounded, multiplying many times over so that his end result was completely different from the first run of the model. As can be seen in Figure 4, the error appears to remain small, but after a few hundred iterations it grows exponentially until reaching a <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/magnitude\/pop\">magnitude<\/a> equivalent to the value of the measurement itself (~0.6).<\/p>\n<figure><img decoding=\"async\" src=\"https:\/\/www.visionlearning.com\/img\/library\/modules\/mid157\/Image\/VLObject-4144-081105111159.jpg\" alt=\"graph - Representations of error propagation\" \/><figcaption>Figure 4: Representation of error propagation in an iterative, dynamic system. After ~1,000 iterations, the error is equivalent to the value of the measurement itself (~0.6), making the calculation fluctuate wildly. Adapted from IMO (2007).<\/figcaption><\/figure>\n<p>Lorenz published his <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/observation\/pop\">observations<\/a> in the now classic work Deterministic Nonperiodic Flow (Lorenz, 1963). His observations led him to conclude that accurate weather prediction over a period of more than a few weeks was extremely difficult \u2013 perhaps impossible \u2013 because even infinitesimally small errors in the measurement of natural conditions were compounded and quickly reached levels equal to the measurements themselves.<\/p>\n<p>The work motivated other researchers to begin looking at other dynamic <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/system\/pop\">systems<\/a> that are similarly sensitive to initial starting conditions, such as the flow of water in a stream or the dynamics of <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/population\/pop\">population<\/a> change. In 1975, the American mathematician and physicist James Yorke and his collaborator, the Chinese-born mathematician Tien-Yien Li, coined the term <em>chaos<\/em> to describe these systems (Li &amp; Yorke, 1975). Again, unlike the common use of the term chaos, which implies randomness or a state of disarray, the science of chaos is not about randomness. Rather, as Lorenz was the first to do, chaos researchers work to understand underlying patterns of behavior in complex systems toward understanding and quantifying this <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/uncertainty\/pop\">uncertainty<\/a>.<\/p>\n<div class=\"comprehension-checkpoint\">\n<p class=\"leader\">Comprehension Checkpoint<\/p>\n<p class=\"question\">Scientists should look for the source of error within a dataset<\/p>\n<form class=\"question\" action=\"action\" id=\"cc5851\">\n<ul class=\"quiz-options\">\n<li class=\"option-a\"><label class=\"choice\" for=\"q1-5851-0-option-a\"><strong>only when<\/strong> the error is very large.<\/label><\/li>\n<li class=\"option-b\"><label class=\"choice\" for=\"q1-5851-1-option-b\"><strong>even when<\/strong> the error is very small.<\/label><\/li>\n<\/ul>\n<\/form>\n<\/div>\n<\/section>\n<section id=\"toc_7\" class=\"article-section\">\n<h2>Recognizing and reducing error<\/h2>\n<p>Error propagation is not limited to <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/mathematical+modeling\/pop\">mathematical modeling<\/a>. It is always a concern in scientific <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/research\/pop\">research<\/a>, especially in studies that proceed stepwise in multiple increments because error in one step can easily be compounded in the next step. As a result, scientists have developed a number of techniques to help <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/quantify\/pop\">quantify<\/a> error. Here are two examples:<\/p>\n<blockquote><p><strong>Controls:<\/strong> The use of controls in scientific experiments (see our <a href=\"https:\/\/www.visionlearning.com\/library\/module_viewer.php?mid=150\">Experimentation in Scientific Research<\/a> module) helps quantify statistical error within an experiment and identify systematic error in order to either measure or eliminate it.<\/p>\n<p><strong>Blind trials:<\/strong> In research that involves human judgment, such as studies that try to quantify the perception of pain relief following administration of a pain-relieving drug, scientists often work to minimize error by using &#8220;blinds.&#8221; In blind trials, the treatment (i.e. the drug) will be compared to a control (i.e. another drug or a placebo); neither the patient nor the researcher will know if the patient is receiving the treatment or the control. In this way, systematic error due to preconceptions about the utility of a treatment is avoided.<\/p><\/blockquote>\n<p>Error reduction and measurement efforts in scientific <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/research\/pop\">research<\/a> are sometimes referred to as <em>quality assurance<\/em> and <em>quality control<\/em>. Quality assurance generally refers to the plans that a researcher has for minimizing and measuring error in his or her research; quality <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/control\/pop\">control<\/a> refers to the actual procedures implemented in the research. The terms are most commonly used interchangeably and in unison, as in &#8220;quality assurance\/quality control&#8221; (QA\/QC). QA\/QC includes steps such as calibrating instruments or measurements against known standards, reporting all instrument detection limits, implementing standardized procedures to minimize human error, thoroughly documenting research <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/method\/pop\">methods<\/a>, replicating measurements to determine <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/precision\/pop\">precision<\/a>, and a host of other techniques, often specific to the type of research being conducted, and reported in the <em>Materials and Methods<\/em> section of a scientific paper (see our <a href=\"https:\/\/www.visionlearning.com\/library\/module_viewer.php?mid=158\">Understanding Scientific Journals and Articles<\/a> module).<\/p>\n<p>Reduction of <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/statistical+error\/pop\">statistical error<\/a> is often as simple as repeating a <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/research\/pop\">research<\/a> measurement or <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/observation\/pop\">observation<\/a> many times to reduce the <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/uncertainty\/pop\">uncertainty<\/a> in the range of <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/value\/pop\">values<\/a> obtained. <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/systematic+error\/pop\">Systematic error<\/a> can be more difficult to pin down, creeping up in research due to instrumental bias, human mistakes, poor research design, or incorrect assumptions about the behavior of <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/variable\/pop\">variables<\/a> in a <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/system\/pop\">system<\/a>. From this standpoint, identifying and quantifying the source of systematic error in research can help scientists better understand the behavior of the system itself.<\/p>\n<\/section>\n<section id=\"toc_8\" class=\"article-section\">\n<h2>Uncertainty as a state of nature<\/h2>\n<p>While <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/Pearson%2C+Karl\/pop\">Karl Pearson<\/a> proposed that individual measurements could not yield exact <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/value\/pop\">values<\/a>, he felt that careful and repeated scientific investigation coupled with statistical <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/analysis\/pop\">analysis<\/a> could allow one to determine the true value of a measurement. A younger contemporary of Pearson&#8217;s, the English statistician Ronald Aylmer Fisher, extended and, at the same time, contradicted this concept. Fisher felt that because all measurements contained <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/inherent\/pop\">inherent<\/a> error, one could never identify the exact or &#8220;correct&#8221; value of a measurement. According to Fisher, the true distribution of a measurement is unattainable; statistical techniques therefore do not estimate the &#8220;true&#8221; value of a measurement, but rather they are used to minimize error and develop range estimates that approximate the theoretically correct value of the measurement. A natural consequence of his idea is that occasionally the approximation may be incorrect.<\/p>\n<p>In the first half of the 20<sup>th<\/sup> century, the concept of <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/uncertainty\/pop\">uncertainty<\/a> reached new heights with the discovery of <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/quantum+mechanics\/pop\">quantum mechanics<\/a>. In the quantum world, uncertainty is not an inconvenience; it is a state of being. For example, the <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/decay\/pop\">decay<\/a> of a radioactive <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/element\/pop\">element<\/a> is inherently an uncertain event. We can predict the <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/probability\/pop\">probability<\/a> of the decay profile of a <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/mass\/pop\">mass<\/a> of radioactive <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/atom\/pop\">atoms<\/a>, but we can never predict the exact time that an individual radioactive atom will undergo decay. Or consider the <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/Heisenberg+Uncertainty+Principle\/pop\">Heisenberg Uncertainty Principle<\/a> in quantum physics, which states that measuring the position of a <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/particle\/pop\">particle<\/a> makes the momentum of the particle inherently uncertain, and, conversely, measuring the particle&#8217;s momentum makes its position inherently uncertain.<\/p>\n<p>Once we understand the concept of <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/uncertainty\/pop\">uncertainty<\/a> as it applies to science, we can begin to see that the purpose of scientific <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/data\/pop\">data<\/a> <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/analysis\/pop\">analysis<\/a> is to identify and <a class=\"term\" title=\"\" href=\"https:\/\/www.visionlearning.com\/en\/glossary\/view\/quantify\/pop\">quantify<\/a> error and variability toward uncovering the relationships, patterns, and behaviors that occur in nature. Scientific knowledge itself continues to evolve as new data and new studies help us understand and quantify uncertainty in the natural world.<\/p>\n<\/section>\n<section id=\"toc-999\" class=\"article-section\">\n<h3>Summary<\/h3>\n<p>There is uncertainty in all scientific data, and even the best scientists find some degree of error in their measurements. This module uses familiar topics &#8211; playing baseball, shooting targets, and calculating the age of an object &#8211; to show how scientists identify and measure error and uncertainty, which are reported in terms of confidence.<\/p>\n<h3>Key Concepts<\/h3>\n<ul class=\"bulleted\">\n<li>Uncertainty is the quantitative estimation of error present in data; all measurements contain some uncertainty generated through systematic error and\/or random error.<\/li>\n<li>Acknowledging the uncertainty of data is an important component of reporting the results of scientific investigation.<\/li>\n<li>Uncertainty is commonly misunderstood to mean that scientists are not certain of their results, but the term specifies the degree to which scientists are confident in their data.<\/li>\n<li>Careful methodology can reduce uncertainty by correcting for systematic error and minimizing random error. However, uncertainty can never be reduced to zero.<\/li>\n<\/ul>\n<\/section>\n<footer>\n<ul class=\"indented links\">\n<li>\n<h5>Further Reading<\/h5>\n<\/li>\n<li><a href=\"https:\/\/www.visionlearning.com\/en\/library\/Process-of-Science\/49\/Using-Graphs-and-Visual-Data-in-Science\/156\">Using Graphs and Visual Data in Science<\/a><\/li>\n<li><a href=\"https:\/\/www.visionlearning.com\/en\/library\/Process-of-Science\/49\/Statistics-in-Science\/155\">Statistics in Science<\/a><\/li>\n<li><a href=\"https:\/\/www.visionlearning.com\/en\/library\/Process-of-Science\/49\/Data-Analysis-and-Interpretation\/154\">Data Analysis and Interpretation<\/a><\/li>\n<\/ul>\n<p><a name=\"refs\" id=\"refs\"><\/a><\/p>\n<ul class=\"indented list\">\n<li>\n<h5>References<\/h5>\n<\/li>\n<li>Arnold, J. R., &amp; Libby, W. F. (1949). Age determinations by radiocarbon content: Checks with samples of known age. <em>Science, 110<\/em>, 678-680.<\/li>\n<li>Damon, P. E., Ferguson, C. W., Long, A., &amp; Wallick, E. I. (1974). Dendrochronologic calibration of the radiocarbon time scale. <em>American Antiquity, 39<\/em>(2), 350-366.<\/li>\n<li>Fairbanks, R. G., Mortlock, R. A., Chiu, T.-C., Cao, L., Kaplan, A., Guilderson, T. P., . . . Nadeau, M. (2005). Radiocarbon calibration curve spanning 0 to 50,000 years BP based on paired 230Th\/ 234U\/ 238U and 14C dates on pristine corals. <em>Quaternary Science Reviews, 24,<\/em> 1781-1796.<\/li>\n<li>Gleick, J. (1987) <em>Chaos: Making a new science.<\/em> New York: Penguin Books.<\/li>\n<li>IMO. (2007). Long range weather prediction.<em> The Icelandic Meteorological Office.<\/em> Retrieved December 18, 2007, from http:\/\/andvari.vedur.is\/~halldor\/HB\/Met210old\/pred.html<\/li>\n<li>Li, T. Y., &amp; Yorke, J. A. (1975). Period three implies chaos. <em>American Mathematical Monthly, 82,<\/em> 985.<\/li>\n<li>Libby, W. F. (1946). Atmospheric helium three and radiocarbon from cosmic radiation. <em>Physical Review, 69<\/em>(11-12), 671-672.<\/li>\n<li>Libby, W. F. (1963). Accuracy of radiocarbon dates. <em>Science, 140,<\/em> 278-280.<\/li>\n<li>Libby, W. F., Anderson, E. C., &amp; Arnold, J. R. (1949). Age determination by radiocarbon content: World-wide assay of natural radiocarbon. <em>Science, 109<\/em>(2827), 227-228.<\/li>\n<li>Lorenz, E. (1963). Deterministic nonperiodic flow. <em>Journal of the Atmospheric Sciences, 20,<\/em> 130-141.<\/li>\n<li>Lorenz, E. (1993). <em>The essence of chaos<\/em>. The University of Washington Press.<\/li>\n<li>Morimoto, Y., Ito, K., Kawamura, T., &amp; Muraki, Y. (2003). Immediate effect of assisted and resisted training using different weight balls on ball speed and accuracy in baseball pitching. <em>International Journal of Sport and Health Science, 1<\/em>(2), 238-246.<\/li>\n<li>Peat, F. D. (2002). <em>From certainty to uncertainty: The story of science and ideas in the twentieth century<\/em>. Joseph Henry Press, National Academies Press.<\/li>\n<li>Salsburg, D. (2001). <em>The lady tasting tea: How statistics revolutionized science in the twentieth century<\/em>. New York: W. H. Freeman &amp; Company.<\/li>\n<li>Wagner, C. H. (1983). Uncertainty in science and statistics. <em>The Two-Year College Mathematics Journal, 14<\/em>(4), 360-363.<\/li>\n<li>Wu, X., Yuan, S., Wang, J., Guo, Z., Liu, K., Lu, X., . . . Cai, L. (2000). AMS radiocarbon dating of cemetery of Jin Marquises in China. <em>Nuclear Instruments and Methods in Physics Research, B, 172<\/em>(1-4), 732-735.<\/li>\n<\/ul>\n<\/footer>\n","protected":false},"author":51812,"menu_order":11,"template":"","meta":{"_candela_citation":"[]","CANDELA_OUTCOMES_GUID":"","pb_show_title":"on","pb_short_title":"","pb_subtitle":"","pb_authors":[],"pb_section_license":""},"chapter-type":[],"contributor":[],"license":[],"class_list":["post-95","chapter","type-chapter","status-publish","hentry"],"part":49,"_links":{"self":[{"href":"https:\/\/courses.lumenlearning.com\/suny-hccc-generalscience\/wp-json\/pressbooks\/v2\/chapters\/95","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/courses.lumenlearning.com\/suny-hccc-generalscience\/wp-json\/pressbooks\/v2\/chapters"}],"about":[{"href":"https:\/\/courses.lumenlearning.com\/suny-hccc-generalscience\/wp-json\/wp\/v2\/types\/chapter"}],"author":[{"embeddable":true,"href":"https:\/\/courses.lumenlearning.com\/suny-hccc-generalscience\/wp-json\/wp\/v2\/users\/51812"}],"version-history":[{"count":1,"href":"https:\/\/courses.lumenlearning.com\/suny-hccc-generalscience\/wp-json\/pressbooks\/v2\/chapters\/95\/revisions"}],"predecessor-version":[{"id":96,"href":"https:\/\/courses.lumenlearning.com\/suny-hccc-generalscience\/wp-json\/pressbooks\/v2\/chapters\/95\/revisions\/96"}],"part":[{"href":"https:\/\/courses.lumenlearning.com\/suny-hccc-generalscience\/wp-json\/pressbooks\/v2\/parts\/49"}],"metadata":[{"href":"https:\/\/courses.lumenlearning.com\/suny-hccc-generalscience\/wp-json\/pressbooks\/v2\/chapters\/95\/metadata\/"}],"wp:attachment":[{"href":"https:\/\/courses.lumenlearning.com\/suny-hccc-generalscience\/wp-json\/wp\/v2\/media?parent=95"}],"wp:term":[{"taxonomy":"chapter-type","embeddable":true,"href":"https:\/\/courses.lumenlearning.com\/suny-hccc-generalscience\/wp-json\/pressbooks\/v2\/chapter-type?post=95"},{"taxonomy":"contributor","embeddable":true,"href":"https:\/\/courses.lumenlearning.com\/suny-hccc-generalscience\/wp-json\/wp\/v2\/contributor?post=95"},{"taxonomy":"license","embeddable":true,"href":"https:\/\/courses.lumenlearning.com\/suny-hccc-generalscience\/wp-json\/wp\/v2\/license?post=95"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}