1cScientificInvestigation

1cScientificInvestigation - Exercise 1C Scientific...

Info iconThis preview shows page 1. Sign up to view the full content.

View Full Document Right Arrow Icon
This is the end of the preview. Sign up to access the rest of the document.

Unformatted text preview: Exercise 1C Scientific Investigation: Statistical Analysis Parts of this lab adapted from General Ecology Labs, Dr. Chris Brown, Tennessee Technological University and Ecology on Campus, Dr. Robert Kingsolver, Bellarmine University. In part C of our Scientific Investigation labs, we will use the measurement data from part B to ask new questions and apply some basic statistics. Ecology is the ambitious attempt to understand life on a grand scale. We know that the mechanics of the living world are too vast to see from a single vantage point, too gradual to observe in a single lifetime, and too complex to capture in a single narrative. This is why ecology has always been a quantitative discipline. Measurement empowers ecologists because our measuring instruments extend our senses, and numerical records extend our capacity for observation. With measurement data, we can compare the growth rates of trees across a continent, through a series of environmental conditions, or over a period of years. Imagine trying to compare from memory the water clarity of two streams located on different continents visited in separate years, and you can easily appreciate the value of measurement. Numerical data extend our capacity for judgment too. Since a stream runs muddier after a rain and clearer in periods of drought, how could you possibly wade into two streams in different seasons and hope to develop a valid comparison? In a world characterized by change, data sets provide reliability unrealized by single observations. Quantitative concepts such as averages, ratios, variances, and probabilities reveal ecological patterns that would otherwise remain unseen and unknowable. Mathematics, more than any cleverly crafted lens or detector, has opened our window on the universe. It is not the intention of this lab to showcase math for its own sake, but we will take measurements and make calculations because this is the simplest and most powerful way to examine populations, communities, and ecosystems. Sampling To demonstrate the power of quantitative description in ecology, you will use a series of measurements and calculations to characterize a population. In biology, a population is defined as a group of individuals of the same species living in the same place and time. Statisticians have a more general definition of a population, that is, all of the members of any group of people, organisms, or things under investigation. For the ecologist, the biological population is frequently the subject of investigation, so our biological population can be a statistical population as well. Think about a population of red-ear sunfish in a freshwater lake. Since the population's members may vary in age, physical condition, or genetic characteristics, we must observe more than one representative before we can say much about the sunfish population as a group. When the population is too large to catch every fish in the lake, we must settle for a sample of individuals to represent the whole. This poses an interesting challenge for the ecologist: how many individuals must we observe to ensure that we have adequately addressed the variation that exists in the entire population? How can this sample be collected as a fair representation of the whole? Biology 6C 23 Ecologists try to avoid bias, or sampling flaws that over- represent individuals of one type and under-represent others. If we caught our sample of sunfish with baited hooks, for example, we might selectively capture individuals large enough to take the bait, while leaving out smaller fish. Any estimates of size or age we made from this biased sample would poorly represent the population we are trying to study. After collecting our sample, we can measure each individual and then use these measurements to develop an idea about the population. If we are interested in fish size, we could measure each of our captured sunfish from snout to tail. Reporting every single measurement in a data table would be truthful, but not very useful, because the human mind cannot easily take in long lists of numbers. A more fruitful approach is to take all the measurements of our fish and systematically construct composite numerical descriptions, or descriptive statistics, which convey information about the population in a more concise form. Note that we did not measure each and every fish in the population, if for no other reason than we likely could never catch them all. Instead, we took a sample of fish from the population, and calculated our mean from this sample. Statistical values (the mean, variance, etc.) are called statistics because they represent one estimate of the true values of the mean and variance. So that’s what statistics really are…estimates of an unmeasured true value for a population. You can probably guess that, if we went and got a second sample of sunfish, the mean and variance would likely differ from what we got for the first sample. The true values for the mean, etc., found by measuring all individuals in our population, are called parameters , or parametric values. We almost never know these, but we assume that our statistics come reasonably close. In general, as long as we randomly sample our populations, and have a large enough sample size, this assumption will hold. However, it’s always possible that we just happen to sample uncommonly long (or uncommonly short) sunfish; if so, then our statistics will likely differ from the parametric values. Part A: Descriptive Statistics As the name implies, descriptive statistics describe our data. One common descriptive statistic is the mean (or average, as it’s more popularly called). The mean, or x , represents the “middle” (or, in mathematical terms, the central tendency) of the data, and is given by the formula: X= ÂX i n where the xi’s are your individual data points (for example, the body length measurements), and n equals sample size (the total number of sunfish measured). The symbol S indicates summation, so for the mean we need to add together all the xi’s and then divide this by n. We might find, for instance, that the mean length of sunfish in this lake is 12.07 centimeters, based on a sample of 80 netted fish. Note: the symbol µ is used for the mean of all fish in the population, which we are trying to estimate in our study. The symbol x is used for the mean of our sample, which we hope to be close to µ. 24 Exercise 1.2. Scientific Investigation: Statistical Analysis Means are useful, but they can be misleading. If a population were made up of small one-year-old fish and much larger two-year-old fish, the mean we calculate may fall somewhere between the large and small size classes- above any of the small fish, but below any of the large ones. A mean evokes a concept of the "typical" fish, but the "typical" fish may not actually exist in the population (Figure 1.7). For this reason, it is often helpful to use more than one statistic in our description of the typical member of a population. One useful alternative is the median, which is the individual ranked at the 50th percentile when all data are arranged in numerical order. Another is the mode, which is the most commonly observed length of all fish in the sample. Figure 1.7 The calculated mean describes a “typical” fish that does not actually exist in a population composed of two size classes. Biology 6C 25 Application In our exercise, the data you collected in your groups last lab period are samples and the population is defined as the entire Biology 6C class. Use the data in Table 1.2 from the previous lab to calculate the mean height and mean arm length for your group. Show your calculations below: Mean height = Mean arm length = Complete Table 1.5 below using the class data spreadsheet. Table 1.5 Group Mean Heights Groups Mean Height (cm) Group 1 Group 2 Group 3 Group 4 Group 5 Group 6 Group 7 Group 8 Group 9 Group 10 Entire Class Using the data table distributed in class, enter the sample (group) means in Table 1.5 and compare them to the population (class) mean. Are there any sample means that do not seem to represent the population mean? How could sample size affect how well the sample mean represents the population mean? How could method of choosing a sample from the population affect how well the sample mean represents the population mean? 26 Exercise 1.2. Scientific Investigation: Statistical Analysis Picturing Variation After calculating statistics to represent the typical individual, it is still necessary to consider variation among members of the population. A frequency histogram is a simple graphic representation of the way individuals in the population vary. To produce a frequency histogram: 1. 2. Choose a measurement variable, such as length in our red-ear sunfish. Assume we have collected 80 sunfish and measured each fish to the nearest millimeter. On a number line, mark the longest and shortest measurements taken from the population (Figure 1.8). The distance on the number line between these points, determined by subtracting the smallest from the largest, is called the range. In our example, the longest fish measures 16.3 cm, and the shortest fish measures 8.5 cm, so the range is 7.8 cm. Figure 1.8 3. Next, divide the range into evenly spaced divisions (Figure 1.9). In our example, each division of the number line represents a size class. It is customary to use between 10 and 20 size classes in a histogram. For our example, we will divide the range of sunfish sizes into 16 units of 0.5 cm each. The first size class includes fish of sizes 8.5 through 8.9 cm. The next size class includes fish of sizes 9.0 through 9.4 cm, and so forth to the last size class, which includes sizes 16.0-16.4. Figure 1.9 4. Having established these size classes, it is possible to look back at our measurement data and count how many fish fall into each class. The number of individuals falling within a class is the frequency of that class in the population. By representing each measurement with an X, as shown in Figure 1.10, we can illustrate frequencies on the number line. 5. On the completed histogram illustrated in Figure 1.11, the stack of X-marks is filled in as a vertical bar. The height of each bar represents the proportion of the entire population that falls within a given size class. Biology 6C 27 Figure 1.10 Example 1 of Histogram Illustrating Sunfish Length. Figure 1.11 Example 2 of Histogram Illustrating Sunfish Length. ?Check your progress In the sample described by the histograms (Figures 1.10 & 1.11), how many fish measured between 10.0 and 10.4 cm? Application Using your group’s height data from table 1.2 in part 1, create an appropriate scale for the X-axis and complete the following frequency histogram (Figure 1.6). Number of individuals Figure 1.6: Frequency Histogram of Height for group _____. 6 5 4 3 2 1 Height (cm) 28 Exercise 1.2. Scientific Investigation: Statistical Analysis If we define our Biology 6C lecture class as our population, the data you collected in your groups last lab period is a sample. The whole point of using a sample is that a smaller, more attainable subset of values can represent the entire population’s values, which in some cases would be impossible to collect. In order for a sample to best represent a population, it must be collected randomly from the population. How were our samples chosen? Were they chosen randomly? Explain. How would you choose a random sample of the class? Examine the data and frequency histogram distributed in class. Does the frequency histogram for the class (the population) look like Figure 1.6, the frequency histogram for your group (a sample)? Do you expect it to? Explain. Describing the Pattern of Variation Notice that in our sample histogram, the most common size classes are near the middle of the distribution (Figure 1.11). Extremely large and extremely small fish are rare, while intermediate sizes are more common. The "skyline" of the histogram fits under a bell-shaped curve that is symmetrical, and has characteristic rounded "shoulders" and "tails" that taper to the ends of the range in a predictable way. Statisticians call this shape a normal distribution (Figure 1.12). Figure 1.12 This pattern of variation is common in nature, and is encountered q uite o ften w hen o ne e ffect i s influenced by many independently acting causes. Since size in fish is influenced by temperature, diet, water purity, and many other factors, it would not be surprising to find that sizes of fish in a mixed-age population are normally distributed. Because the normal distribution is encountered so commonly, many of the statistical tools ecologists use to test hypotheses assume that variations in their data are distributed in this bell-shaped form. Models and tests based o n t his k ind o f d istribution a re c alled parametric statistics. If the histogram of variation is lopsided, has more than one peak, or is too broad or too narrow, then parametric tests should not be used (Figures 1.13 and 1.14). Non-parametric tests have been developed for these kinds of data. Because the nature of variation in your measurements is critical to further analysis, it is always a good idea to draw a histogram and compare your data to a normal distribution before taking your analysis any farther. Biology 6C 29 Figure 1.13 Bimodally distributed data Figure 1.14 Skewed data Measuring Variation How trustworthy is the mean that we calculate from our sample of fish? Two factors come into play. First, the sample size is critical. If fish in the lake vary a lot in size, a mean calculated from a small sample (say 10 fish) might be significantly off the mark. By chance, the 10 fish you caught might be larger or smaller than the average population size you are trying to describe. If the sample is expanded to 1000, it is much more likely that your calculated mean will accurately reflect the population average. A fundamental principle of data collection is that the sample size must be large enough to eliminate sampling errors due to chance departures from the population mean. To keep our thoughts straight, we use n = size of the sample, and N = size of the entire population. N is usually unknown, but can be estimated in a number of ways, as you will see in a later lab. How large, then, must a sample be? This depends on the amount of variation in the population. Samples taken from a fish farm where all the fish are nearly the same size will give reliable estimates, even if the sample is small. In a natural population with a great range of sizes, the sample has to be expanded to ensure that the larger variation is accounted for. Thus, the more variable the population, the larger the sample must be to achieve the same level of reliability. It becomes obvious that we need a statistic to measure variation. In the data curves below (Figure 1.15), you can easily see the relationship between the mean and the amount of variation around the mean. Curves A and C have different mean values, but their shape (amount of spread) are the same. On the other hand, curves A and B have the same mean, but very different shapes. The three most commonly used descriptive statistics to quantify the amount of variation around the mean are: standard deviation, variance and standard error. 30 Exercise 1.2. Scientific Investigation: Statistical Analysis Figure 1.15 Examples of variation around the mean. Standard Deviation Standard Deviation is a measure of variation within a collection of values. It essentially describes the distance each data point is from the mean of those data. Standard Deviation is the average deviation of values from the mean. The larger the standard deviation, the more variation there is in the collection of values. In other words, the larger the standard deviation, the wider the bellshaped curve. Standard Deviation = 1 Standard Deviation = 2 Standard Deviation = 3 Figure 1.16 Illustration of Standard Deviations; from Children’s Mercy Hospital & Clinics: Statistics Biology 6C 31 If you look at the formula for standard deviation closely, you’ll note it essentially describes the distance of each data point from the mean. s = where: s = standard deviation for a sample x = each individual value x = sample mean 2  (x – x ) n–1  = sum of values n = number of values in the sample Application Calculate (manually) the standard deviation for height from your group’s data in Table 1.2. Use these steps and Table 1.6 below to work through the formula: 1. compute x 2. subtract x from each individual value: (x - x ) 3. square each of the differences from step 2: (x - x )2 4. add all squares from step 3:  (x - x )2 5. divide the sum from step 4 by n-1 6. find the square root of the result of step 5 Table 1.6 Computation of Standard Deviation for Height for group ______. x (x - x ) (x – x )2 s= x= 32 Â= Exercise 1.2. Scientific Investigation: Statistical Analysis Frequency (actual number or %) The Empirical Rule states that population or sample frequency data having a bell shaped distribution will have about 68% of values fall within 1 standard deviation from the mean, about 95% of values within 2 standard deviations from the mean, and about 99.7% of values within 3 standard deviations from the mean Where: x = sample mean SD = standard deviation Values Figure 1.17 Illustration of Empirical Rule; from British Medical Journal Another way to look at it: Where: µ = population mean s = SD of population # of standard deviations from the mean Figure 1.18 Illustration 2 of Empirical Rule; from Statistics @ SUNY Oswego Think about it: Students always ask if I “curve” the grades. A “curve” refers to this bell shaped distribution. If I were to curve grades, what percentage of students would likely receive A’s? Do you really want me to “curve”? Biology 6C 33 Our sunfish data: Figure 1.19 Sunfish data illustrating Empirical Rule Variance Variance, s2, is another measure of the data scatter around the mean. The formula is: s 2 Â(X = i - X )2 n -1 Note that variance is just the standard deviation squared. Variance is more difficult to visualize because the units are squared and therefore differ from the units of the original data set. (This is especially true if the units are meaningless when squared. What is a minute squared?). For example, if we measure length in cm, then the sample variance is reported in cm2. To calculate the standard deviation, we take the square root of the variance, which returns us to our original measurement units, which is length in cm. 34 Exercise 1.2. Scientific Investigation: Statistical Analysis Application: Calculate the variance of the heights for your group data from Table 1.2 using the standard deviation you calculated above. Standard Error and Confidence Intervals There is one more statistic that you will find useful when characterizing the typical sunfish in your population with a mean. Incorporating sample size and variation, you can develop a measure of reliability of the mean called the standard error (S.E.). Assume there are 30 people in your ecology class. Each of you goes to the same pond sometime this week, nets a sample of sunfish in the same way, and measures 100 randomly selected individuals. Releasing the fish unharmed, you return to the lab and calculate a mean and standard deviation from your data. Everyone else does the same. Would your 30 sample means be identical? No, but the variation in means would be considerably smaller than the total variation among the fish in the pond. Repeating a sampling program 30 times is usually impractical. Fortunately, statistics gives us a way to measure reliability when we have only one mean developed from one sample. The variation among all possible sample means can be predicted from the sample size and the variation in the pond's sunfish with the following formula: Looking a t t he f ormula, y ou c an s ee t he relationship between error in our estimate, the variability of sunfish, and the sample size. The smaller the S.E. is, the more trustworthy your calculated mean. Note that the sample standard deviation is in the numerator of the calculation, so the more variable the size of the fish, the less accurate the estimate you made from a random sample. Sample size, on the other hand, is in the denominator. This implies that a large sample makes your mean more reliable. The formula shows that the more variable the population, the larger our sample must be to hold S.E. to an acceptably small margin of error. Biology 6C 35 Since standard errors tend to be normally distributed, it is a safe assumption that 95% of the variation in all possible means will fall within 1.96 S.E. of the actual mean. This fact can be used to calculate a 95% confidence interval as follows: To g o b ack t o o ur h istogram, t he confidence intervals can be represented by brackets around the sample mean. A 95% confidence interval implies that the actual population mean (µ) will fall within the brackets you have placed around your estimate ( x ) 95% of the time under these experimental conditions (Figure 1.20). Figure 1.14 Figure 1.20 95% Confidence Intervals around the Mean for Sunfish Data The importance of confidence limits cannot be overstated. Scientifically, it is dishonest to report a sample mean by itself. Without sharing with your readers the sampling methods, sample size, and the variance of the population, there is no way for them to know how accurately the sample mean represents the population. Always report sample size with a mean, and add some measure of variation. Any of the statistics representing variation (s, s2, or S.E.) can be reported, since anyone of these can be used to calculate other two. Application Calculate the standard error and the 95% confidence interval, including upper and lower limits, for the mean of heights of your group using the standard deviation and mean you calculated above. 36 Exercise 1.2. Scientific Investigation: Statistical Analysis Part B: Inferential Statistics and the Logic of Experimental Design Ecologists often collect data from two or more populations and want to compare these groups to see if they differ. For instance, we might ask whether body mass varies between a northern and a southern population of cotton rats (perhaps related to temperature differences). To conduct our experiment, we would go to each population, capture a number of cotton rats, and weigh each one. How can we determine whether mass differs between the two populations? To do this, we use inferential statistics. Before we collect data, we formulate our hypotheses. The null hypothesis in most studies says that no differences exist between our populations (that is, mass is the same at each latitude). As researchers, we typically hope to prove this hypothesis false, and so we propose a lternative hypotheses. One possible alternative hypothesis for our example would be that mass of cotton rats does differ between latitudes, but we don’t know beforehand which population will be larger. If we knew the parametric values, then it would be easy to determine if our populations differed: for example, if rats in the southern population averaged 45 g and the rats in the northern population averaged 75 g, then we know these differ because we’ve weighed every single rat, and 45 is not equal to 75!! However, we haven’t weighed every rat, so we have uncertainty in how close our statistics are to the parametric values. This is where inferential statistics come in; they help us make objective decisions about observed differences between two or more sets of data. To put it another way, they help answer the question, "Are the differences we see real, or have they occurred simply by chance?". Inferential statistical tests come in a wide variety of forms, but nearly all have the same conceptual framework. We begin with our null hypothesis (body mass does not differ between northern and southern populations), then collect our body mass data by weighing cotton rats from each population. We then choose an appropriate statistical test; this may or may not be a simple matter, but we’ll ignore those types of details for now. The test will calculate a test statistic based on your data, for example, a t-value for the t-test. You (or a computer program) will then compare your calculated test statistic to an expected, or “cut-off,” value, and this will generate something called the probability, or P-value. Now …what does all of the above mean!? To answer that, let’s first do a thought experiment. Suppose we take two different samples from just the southern rats. We would expect these samples to have similar (although not necessarily identical) means and variances. However, what if we happened to capture extremely large rats in one sample and extremely small rats in the other. These means and variances would differ greatly from each other, even though they both come from the same population with a single true mean and variance, simply because we caught very different rats by chance. Inferential statistics, and their associated P -values, attempt to quantify the likelihood that this is going on. In essence, they answer the following: if two populations really have the same mean, what is the probability that samples taken from each population will differ by as much as we calculate they do by random chance alone? A slightly different way to ask this is, what is the probability that our two samples come from populations with the exact same parametric mean value? Biology 6C 37 For example, suppose you perform a t-test on body mass and get a P-value of 0.25. This indicates that there is a 25% chance that these samples come from populations with the same parametric mean; that is, a 25% chance that mass does not differ between populations (or, a 75% chance that mass does differ between populations). The question then becomes, at what probability level do we say that two samples are different enough to be considered, well, different? The answer is actually very subjective (see below), but in general biologists have agreed that a test giving a Pvalue of 0.05 or less indicates “real,” or significant, differences between groups, and thus rejects the null hypothesis. On the other hand, a nonsignificant P-value of greater than 0.05 indicates no difference between groups and does not reject the null hypothesis. Another way of saying this is that, with P £ 0.05, we have at least a 95% probability that our two populations are really different. Why 0.05? Well, once a test has been done, there are four possible outcomes: (1) you “accept” a true null hypothesis; (2) you reject a false null hypothesis; (3) you reject a true null hypothesis (a Type I error); or (4) you “accept” a false null hypothesis (a Type II error). The last two conclusions are incorrect! We can minimize the chance of making a Type I error by choosing a low probability level (symbolized p) for rejection of the null hypothesis. Thus, we can choose our p = 0.001 (instead of 0.05) if we want; this ensures that only if there is a 99.9% chance that differences did not occur by chance will we reject the null hypothesis. However, by doing so we greatly increase the probability of incorrectly accepting the null hypothesis when it is in fact wrong, a Type II error. Biologists have therefore chosen p = 0.05 as a kind of compromise between these two types of error. The t-Test The t-test is a parametric test for comparing means of two sets of continuous data. For example, we may wish to know whether lengths of razor clam shells collected from two beaches are the same or different. Of course, two shell collections from the same population could have slightly different means, just because of random variation and limited sample size. The question we are asking is, "is the difference in means too large to explain by chance alone?" The null hypothesis in a t-test is that the actual means of the two populations are equal. In addition to the sizes of the samples, there are two things we need to know to determine significance with a t-test. First, of course, is the difference between the two means. If we show the samples as histograms, with one collection of clam shells represented by solid bars and the other by crosshatched bars, we can agree from inspection that the two samples on the left (Fig. 1.21a) are significantly different, but the two samples on the right (Fig. 1.21b) are harder to distinguish. The shapes of the distributions are equivalent, but the pair of samples on the left has a greater distance between the means (barbell line), and will generate a more highly significant t-test result. Figure 1.21a Comparison of two populations in which the means are very different. 38 Figure 1.21b Comparison of two populations in which the means are very close. Exercise 1.2. Scientific Investigation: Statistical Analysis Secondly, we need to consider the sample variances. The following sample pairs have equal differences between their means, but the two samples on the left (Fig. 1.22a) are more clearly distinguishable than those on the right (Fig. 1.22b) because of their smaller variances. Distinctions between means must always be viewed in the context of variances, because more scatter around the population mean provides a greater possibility for large-scale sampling errors. Figure 1 .22a Comparison of two p opulations w ith l ittle variation around the means. Figure 1.22b Comparison of two populations with a lot variation around the means. In summary, any difference between two means can be found significant if the sample size is large enough, and the variances small enough, to rule out sampling error as an alternative cause. To perform a t-test of the null hypothesis that two means are equal, use the following formula. Note that increasing the difference between means or increasing the sample sizes will increase the value of t, but increasing the variance will decrease t. Biology 6C 39 Once we have calculated our t, we compare it to values of t in the critical values table, using the appropriate degrees of freedom. For want of a better description, degrees of freedom describes the “strength” of your data. Larger sample sizes provide more “strength,” and thus a greater df, because they provide more accurate estimates of the true mean and variance. For the t-test, df = n1+n2-2. Recall that statistical significance occurs when our P-value is less than or equal to 0.05. This will occur when tstat, with n1+n2-2 df, is greater than the t-value listed in the table at the junction of the n1+n2-2 df row and the a = 0.05 column. So, if tstat > tcritical at the a = 0.05 level, we can say that our two populations are really different from each other. To consult the critical values table (Table 1.8): 1. First look down the left-hand column to find the degrees of freedom (df) that matches the one you calculated. Then reading across that row, compare your t value with the number in the second column. This number is the critical value of t for p = 0.05. 2. If your value of t is smaller than the p = 0.05 t value, then your two samples are not significantly different. You must accept the null hypothesis that the two samples come from the same population. There is no real difference between these means. 3. If your value of t is larger than the p = 0.05 value, then the difference between your means is significant. You can reject the null hypothesis with no more than a 5% rate of error for the conclusion. 4. Look at the right-hand column to compare your value of t with the p = 0.01 critical value. If your value of t is larger than this number, then the difference between your means is highly significant. You can reject the null hypothesis with a 1% rate of error for the conclusion. 40 Exercise 1.2. Scientific Investigation: Statistical Analysis An Example: Body Mass of Cotton Rats To give you an example of a t-test, I’ll do one “by hand.” I’ll use some made-up data on body mass of cotton rats. Ignoring the raw data, we have the following table: Table 1.7 Body Mass of Two Populations of Cotton Rats Mean ( X ) Variance (s2) Sample Size (n) Northern 41 90.43 15 Southern 22.4 15.11 15 These two means look pretty different, but we need to check this to be certain. Our null hypothesis, symbolized H0, states that the two population means do not differ: H0: m northern = m southern We’ll use the t-test to determine whether our data rejects this null hypothesis. t= (41-22.4) (90.43) + (15.11) 15 15 = 18.6 2.65 = 7.02 Now that we have tstat, we need two more bits of information, the degrees of freedom and the table t-value. The degrees of freedom equal df = 15 + 15 – 2 = 28 We can use the critical value table to get the table t-value. Simply follow the Degrees of Freedom column down until you get to 28, then go across that row until you get to the Probability = 0.05 column. You should find the value 2.05 here. Since tstat (= 7.01) > tcritical (= 2.05), we can reject our null hypothesis! Thus, cotton rats from different latitudes have different masses. (NOTE: If you look across the degrees of freedom = 28 row, you’ll notice that tcritical = 3.67 in the probability = 0.001 column. Our tstat value is greater than this as well, so we can actually say that we have at least a 99.9% likelihood that our two populations really do differ from each other!) Biology 6C 41 Table1.8 Table of Critical t-values Critical t-Values Degrees of Freedom 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 40 60 120 42 Probability = 0.1 6.31 2.92 2.35 2.13 2.02 1.94 1.89 1.86 1.83 1.81 1.80 1.78 1.77 1.76 1.75 1.75 1.74 1.73 1.73 1.72 1.72 1.72 1.71 1.71 1.71 1.71 1.70 1.70 1.70 1.70 1.68 1.67 1.66 0.05 12.71 4.30 3.18 2.78 2.57 2.45 2.37 2.31 2.26 2.23 2.20 2.18 2.16 2.14 2.13 2.12 2.11 2.10 2.09 2.09 2.08 2.07 2.07 2.06 2.06 2.06 2.05 2.05 2.05 2.04 2.02 2.00 1.98 0.01 63.66 9.93 5.84 4.60 4.03 3.71 3.50 3.36 3.25 3.17 3.11 3.06 3.01 2.98 2.95 2.92 2.90 2.88 2.86 2.85 2.83 2.82 2.82 2.80 2.79 2.78 2.77 2.76 2.76 2.75 2.70 2.66 2.62 0.001 636.62 31.60 12.92 8.61 6.87 5.96 5.41 5.04 4.78 4.59 4.44 4.32 4.22 4.14 4.07 4.02 3.97 3.92 3.88 3.85 3.82 3.79 3.77 3.75 3.73 3.71 3.69 3.67 3.66 3.65 3.55 3.46 3.37 Exercise 1.2. Scientific Investigation: Statistical Analysis Application We will use the height data collected last lab period to perform a simple t-test. The question we will ask is: "Is the difference in mean heights of men and women too large to explain by chance alone?" The null hypothesis is that the actual means of the two populations (men and women) are equal. Using class data, complete the following table to assist in your calculation of a t-value. Show your calculations below the table. Table 1.9 Calculation of t-Value for heights of Biology 6C Students Males Females Mean ( x ) Variance (s2) Sample Size (n) Degrees of Freedom (= n1 + n2 -2) t value (calculated) critical t-value (from the table) Explain your t-value in terms of what it means. Are our two population means (men and women) different? Biology 6C 43 Part C: Correlation Analysis A correlation analysis is used to determine if there is a relationship between two variables. To perform a correlation analysis, we will use a scatterplot and calculate a correlation coefficient. A scatterplot is a type of graph in which paired data are plotted such that the value of the x axis is matched with the value of the y axis. One point therefore represents two separate data sets. A correlation coefficient is a measure of the association between the two data sets. The closer the data points on a scatterplot are to forming a line, the stronger the association, or correlation, between the two variables (x and y). If the points are spread out, forming no line, there is no correlation. A positive correlation means that as the value along the x axis is increasing, so is the value along the y axis. A negative correlation means that as one variable increases, the other decreases. Application Complete the scatterplot below (Figure 1.23) using data from Table 1.2. Place the appropriate units on each axis (with equal intervals along each axis) and plot each student’s arm length and height as one dot on the graph below. Arm Length (cm) Figure 1.23 Comparison of Student Height to Student Arm Length for group ____. Height (cm) 44 Exercise 1.2. Scientific Investigation: Statistical Analysis Calculate the correlation coefficient for height vs. arm length for your group. The formula to calculate a correlation coefficient (r) is: n  xy – (  x) ( y) r= n (  x2) – (  x)2 n (  y2) – (  y)2 Using Table 1.9, complete the table below to facilitate the computation of your r value. Table 1.10 Computation of Correlation Coefficient for Height vs. Arm Length for group_____. x (height) Âx = y (arm length) Ây= xy  xy = x2  x2= y2  y2 = r= Biology 6C 45 Interpreting r The correlation coefficient (r) will fall between -1 and +1 where -1 is a perfect negative correlation, +1 is a perfect positive correlation and 0 is no correlation. If your r value is not -1, 0, or +1, how do you decide if it represents a correlation or not? You will use a Table of Critical Values of the Correlation Coefficient to interpret the meaning of your r value. Figure 1.24 Comparison of r values; from NASA: Solar System Exploration Division In order to evaluate the correlation coefficient (r) for your sample you must determine what the likelihood is that the r value you found could have occurred by chance alone. In other words, does the relationship you found in your sample really exist in the population or are your results a fluke (due to chance)? Like the t-test, a significance level of 0.05 is standard for correlation analysis. To recap what this means, if you recollected your data 100 times, with different samples from the same population and calculated r each time, the value for r would not be the same for each of the 100 samples that you collected. When you set your significance level to .05, you are saying that you are willing to be wrong (say there was a relationship in your sample when there was not one in your population) 5 times out of 100. In other words, a maximum of 5 of those 100 samples might show a relationship (r < > 0) when there really was no relationship in the population (r = 0). 46 Exercise 1.2. Scientific Investigation: Statistical Analysis Table 1.11 Critical Values of the Pearson Correlation of Coefficient (r) n 4 5 6 7 8 9 10 15 20 25 30 35 40 45 50 60 70 80 90 100 % = .05 % = .01 .950 .878 .811 .754 .707 .666 .632 .514 .444 .396 .361 .335 .312 .294 .279 .254 .236 .220 .207 .196 .999 .959 .917 .875 .834 .798 .765 .641 .561 .505 .463 .430 .402 .378 .361 .330 .305 .286 .269 .256 Where: n = the sample size % = significance level table values in % columns = critical values To use this table: Find your sample size in the n c olumn, choose your level of significance (usually % = . 05) a nd l ook d own t he a ppropriate significance column to the critical value listed across from your sample size. If your r value is greater than or equal to the value listed, your result (r value) is significant (there is a correlation) with at least 95% confidence (or 99% if you chose % = .01). What is the critical value for the correlation coefficient (r) calculated for your group’s data? Is the correlation between height and arm length significant for your group? Explain. What is the critical value for the correlation coefficient (r) of the class data? Is the correlation between height and arm length significant for the class? Explain. Which correlation (group or class) has the highest critical value? Explain why this is so. Biology 6C 47 Part D: Using Excel Lab Write Up Now that you have performed all of the calculations by hand, you will learn to use Excel to calculate statistics and generate graphs. Part 1: class data: male vs. female heights ß calculate the mean, standard deviation, and variance for both males and females ß make a simple bar graph showing the mean values for the two groups ß perform a t-test to determine if height varies between the two groups. Part 2: class data: arm length vs. height ß make a scatterplot of arm length vs. height ß calculate a correlation coefficient All of this is explained further on the next three pages, if you are unfamiliar with Excel. Part 3: explain your results: ß Explain what the t-test results tell you: are the two groups different or not in height? Compare your group results to the results for the class data. Are the results different? Why is the critical value for your group t-test higher than the value for the class data? ß Explain what the correlation coefficient results tell you: are the two characteristics correlated? Compare your group results to the results for the class data. Are the results different? Why is the critical value for your group correlation coefficient higher than the value for the class data? For the lab write up, turn in a printed copy of your Excel spreadsheet with descriptive statistics, your bar graph, and your t-test results, scatterplot and correlation coefficient. Please fit these all on one or two printed pages. Part three should be in paragraph form, typed and stapled to the spreadsheet. 48 Exercise 1.2. Scientific Investigation: Statistical Analysis Some Basic Excel Instructions for Novices If you’ve never (or rarely) used Excel before, the following should help you get some of the basics down. Excel is available in all computer labs on campus, and many home computers come with it preloaded. In addition to my outline, many good Excel tutorials exist online. For a biology related example, try http://ecology.botany.ufl.edu/2006introecology/Downloads/Labs/Lab2ExcelTutorial.pdf. 1. Double click on the Microsoft Excel icon. You should be presented with a blank worksheet. Cells locations are given by their corresponding letter and number; for example, the first cell is called cell A1. 2. Name the columns by clicking in cells A1 and B1 and typing in the name you want to use. For example, the first column could include the data from men and the second column could include data from women. Be sure to use an informative name so that you (or anyone else) know from where the data came. 3. Starting in cell A2, enter the data from the appropriate group for column A. Pressing return or the down arrow after each entry will move you down one space in the column. After you enter the last number for the first group, repeat the process for group 2, this time in column B. 4. To calculate the mean of the values in column A, click on the first empty cell in that column (probably cell A42). Type the following: =AVERAGE(A2:A41). This tells the computer to calculate the mean for all data from cell A2 up to and including cell A41 (NOTE: if your data start or end in some other cell number, be sure to alter the formula to reflect this). Hit enter or the down arrow, and the mean value will appear! To get the mean for column B, you can either type the formula as above, just replacing the A’s with B’s, or…place the cursor over the bottom right corner of cell A42 (a cross should appear). Left click the mouse, and drag the cross over to cell B42. This will copy the formula to cell B42, and at the same time automatically change the A’s to B’s. NOTE 2: You might want to type the word “MEAN” or “AVG” into cell C42, as a reminder that those are the values in that row. 5. To calculate the standard deviation in column A, click on the cell below the mean and enter the following: =STDEV(A2:A41). Repeat this formula for column B, either by re-entering it or copying from column A. Again, label this row in column C as a reminder. 6. To calculate the variance in column A, click on the cell below the standard deviation and enter the following: =VAR(A2:A41). Repeat this formula for column B, either by re-entering it or copying from column A. Again, label this row in column C as a reminder. 7. To create a bar graph of the two means, click on the middle of the cell with the column A mean (a large cross should appear) and, while holding the mouse button, drag the cursor over to the cell with the mean of column B and release the mouse button. If done correctly, both cells should be outlined in black. Next, click the Chart Wizard icon (the little bar graph) on the main toolbar. Alternatively, you can click on Insert, then Chart. Under Chart type: select Column, and then highlight the uppermost left box under Chart sub-type:. Click Next> and a preview of your chart should appear. Nothing needs to be changed here, so press Next> again. Here you can add a title to your graph, name the X and Y axes, change the axes or gridlines, remove or move the legend, and so on. The only thing you need to do is label your axes; the Y axis is “height (cm)” and the X axis is tree number or needle location. You can play around with the other options if you’d like, then click Next> when you are done. On this screen, I usually leave it as is (with As Object in: Sheet 1 highlighted) and click Finish. Your graph should appear on screen! You can move it where you want (off your data) by clicking anywhere inside the chart and dragging it to a new spot. You can make it smaller or larger by clicking any of the black squares and dragging these. 8. To calculate the t-test, click and hold on Tools from the upper menu, then drag the cursor to Data Analysis before releasing the mouse button. Scroll down and click on t-test: Two-Sample Assuming Equal Variance, then click on OK. NOTE: If Data Analysis does not show up under Tools, then you need to do one of two things. First, make sure you do not have the chart still highlighted. If you do, click on any cell outside the chart. If Data Analysis still is not there, click on Add-ins… under Tools, then check the box beside Analysis ToolPak and click OK. You should be ready to go! 9. A menu box will pop up. Under the heading Input you will see two boxes marked Variable 1 range and Variable 2 range. These two ranges will describe the two sets of data you want to compare (height from your two samples) and both are entered the same Biology 6C 49 way. For the Variable 1 range box, type in $A$2:$A$41, and for the variable 2 range box type in $B$2:$B$41. Notice all the dollar signs!! This tells Excel that you want to compare all the data in column A to all the data in column B (not including the mean, variance, and standard deviation). Alternatively, you can click in the empty Variable 1 range box, then click in cell A2 and drag down to cell A41. After you do this, you should see the above formula in the box. Repeat this for Variable 2 range, only highlighting the data in column B. 10. In the Output Options section, you can choose to have your results go to a new worksheet, a new workbook, or to your data page (the best option). To do the latter, click the button by Output range, then click inside the empty box by Output range, and then click in some empty cell where you want the test results to show up (somewhere below or to the right of your data!). Finally, click OK. NOTE: Be careful here! When you click the button beside Output range, Excel automatically bumps the cursor up to the Variable 1 range box for some reason. So, if you aren’t careful, you end up changing the data you want for variable 1. 8. Your results will magically appear! You’ll notice that this gives you both the mean and variance of the two groups, which should match what you calculated earlier. It also gives a lot of other results, only three of which have importance for us. First look at the row labeled t Stat. This is the statistic calculated by the t-test. We can then compare it to the value in the row labeled t Critical two-tail (the very bottom row). If t Stat is greater than t Critical, ignoring any negative sign on t Stat, then we say our two groups are significantly different from each other. That is, we reject the null hypothesis of no difference! We can verify this by looking at the row labeled P(T<=t) two tail; this will have a value less than 0.05 if t Stat is greater than t Critical. On the other hand, if t Stat is less than t Critical, again ignoring any negative sign on t Stat, then we say our two groups are NOT significantly different from each other. In this case we “accept” the null hypothesis of no difference! We can verify this by looking at the row labeled P(T<=t) two tail; this will have a value greater than 0.05 if t Stat is less than t Critical. 50 Exercise 1.2. Scientific Investigation: Statistical Analysis References: British Medical Journal, http://bmj.bmjjournals.com/collections/statsbk/index.shtml Children’s Mercy Hospital & Clinics: Statistics, http://www.childrensmercy.org/stats/definitions.asp Cox, George W., 2002, General Ecology Lab Manual, 8th ed., McGraw Hill, Boston, USA. Dolphin, Warren D., 2002, Biological Investigations, 6th ed., McGraw Hill, Boston, USA. NASA: Solar System Exploration Division, http://ssed.gsfc.nasa.gov/lepedu/IA-CorrCoeff.html. Oswego State University of New York: Statistics @ SUNY Oswego, http://www.oswego.edu/~srp/stats/index.htm Triola, Marc M., and Triola, Mario F., 2006, Biostatistics for the Biological and Health Sciences, 1st ed., Pearson, Boston, USA. Biology 6C 51 52 Exercise 1.2. Scientific Investigation: Statistical Analysis ...
View Full Document

This note was uploaded on 09/02/2011 for the course BIOL 6C taught by Professor Sundram during the Spring '09 term at DeAnza College.

Ask a homework question - tutors are online