Measures of Variability Author(s) David M. Lane Prerequisites Percentiles, Distributions, Measures of Central Tendency Learning Objectives 1. 2. 3. 4. 5. 6.

Determine the relative variability of two distributions Compute the range Compute the inter-quartile range Compute the variance in the population Estimate the variance from a sample Compute the standard deviation from the variance

Quiz 1 Quiz 2

WHAT IS VARIABILITY? Variability refers to how "spread out" a group of scores is. To see what we mean by spread out, consider graphs in Figure 1. These graphs represent the scores on two quizzes. The mean score for each quiz is 7.0. Despite the equality of means, you can see that the distributions are quite different. Specifically, the scores on Quiz 1 are more densely packed and those on Quiz 2 are more spread out. The differences among students were much greater on Quiz 2 than on Quiz 1.

Figure 1. Bar charts of two quizzes. The terms variability, spread, and dispersion are synonyms, and refer to how spread out a distribution is. Just as in the section on central tendency where we discussed measures of the center of a distribution of scores, in this chapter we will discuss measures of the variability of a distribution. There are four frequently used measures of variability: the range, interquartile range, variance, and standard deviation. In the next few paragraphs, we will look at each of these four measures of variability in more detail. RANGE The range is the simplest measure of variability to calculate, and one you have probably encountered many times in your life. The range is simply the highest score minus the lowest score. Lets take a few examples. What is the range of the following group of numbers: 10, 2, 5, 6, 7, 3, 4? Well, the highest number is 10, and the lowest number is 2, so 10 - 2 = 8. The range is 8. Lets take another example. Heres a dataset with 10 numbers: 99, 45, 23, 67, 45, 91, 82, 78, 62, 51. What is the range? The highest number is 99 and the lowest number is 23, so 99 23 equals 76; the range is 76. Now consider the two quizzes shown in Figure 1. On Quiz 1, the lowest score is 5 and the highest score is 9. Therefore, the range is 4. The range on Quiz 2 was larger: the lowest score was 4 and the highest score was 10. Therefore the range is 6. INTERQUARTILE RANGE

The interquartile range (IQR) is the range of the middle 50% of the scores in a distribution. It is computed as follows: IQR = 75th percentile - 25th percentile For Quiz 1, the 75th percentile is 8 and the 25th percentile is 6. The interquartile range is therefore 2. For Quiz 2, which has greater spread, the 75th percentile is 9, the 25th percentile is 5, and the interquartile range is 4. Recall that in the discussion of box plots, the 75th percentile was called the upper hinge and the 25th percentile was called the lower hinge. Using this terminology, the interquartile range is referred to as the H-spread. A related measure of variability is called the semiinterquartile range. The semi-interquartile range is defined simply as the interquartile range divided by 2. If a distribution is symmetric, the median plus or minus the semi-interquartile range contains half the scores in the distribution. VARIANCE Variability can also be defined in terms of how close the scores in the distribution are to the middle of the distribution. Using the mean as the measure of the middle of the distribution, the variance is defined as the average squared difference of the scores from the mean. The data from Quiz 1 are shown in Table 1. The mean score is 7.0. Therefore, the column "Deviation from Mean" contains the

score minus 7. The column "Squared Deviation" is simply the previous column squared. Table 1. Calculation of Variance for Quiz 1 scores.
Scores 9 9 9 8 8 8 8 7 7 7 7 7 6 6 6 6 Deviation from Mean 2 2 2 1 1 1 1 0 0 0 0 0 -1 -1 -1 -1 Squared Deviation 4 4 4 1 1 1 1 0 0 0 0 0 1 1 1 1

6 6 5 5

-1 -1 -2 -2 Means

1 1 4 4

1.5

One thing that is important to notice is that the mean deviation from the mean is 0. This will always be the case. The mean of the squared deviations is 1.5. Therefore, the variance is 1.5. Analogous calculations with Quiz 2 show that its variance is 6.7. The formula for the variance is:

where 2 is the variance, is the mean, and N is the number of numbers. For Quiz 1, = 7 and N = 20. If the variance in a sample is used to estimate the variance in a population, then the previous formula underestimates the variance and the following formula should be used:

where s2 is the estimate of the variance and M is the sample mean. Note that M is the mean of a sample taken from a population with a mean of . Since, in practice, the

variance is usually computed in a sample, this formula is most often used. The simulation "estimating variance" illustrates the bias in the formula with N in the denominator. Let's take a concrete example. Assume the scores 1, 2, 4, and 5 were sampled from a larger population. To estimate the variance in the population you would compute s2 as follows: M = (1 + 2 + 4 + 5)/4 = 12/4 = 3. s2 = [(1-3)2 + (2-3)2 + (4-3)2 + (5-3)2]/(4-1) = (4 + 1 + 1 + 4)/3 = 10/3 = 3.333 There are alternate formulas that can be easier to use if you are doing your calculations with a hand calculator. You should note that these formulas are subject to rounding error if your values are very large and/or you have an extremely large number of observations.

For this example,

and

STANDARD DEVIATION The standard deviation is simply the square root of the variance. This makes the standard deviations of the two quiz distributions 1.225 and 2.588. The standard deviation is an especially useful measure of variability when the distribution is normal or approximately normal (see Chapter on Normal Distributions) because the proportion of the distribution within a given number of standard deviations from the mean can be calculated. For example, 68% of the distribution is within one standard deviation of the mean and approximately 95% of the distribution is within two standard deviations of the mean. Therefore, if you had a normal distribution with a mean of 50 and a standard deviation of 10, then 68% of the distribution would be between 50 - 10 = 40 and 50 +10 =60. Similarly, about 95% of the distribution would be between 50 - 2 x 10 = 30 and 50 + 2 x 10 = 70. The symbol for the population standard deviation is ; the symbol for an estimate computed in a sample is s. Figure 2 shows two normal distributions. The red distribution has a mean of 40 and a standard deviation of 5; the blue distribution has a mean of

60 and a standard deviation of 10. For the red distribution, 68% of the distribution is between 35 and 45; for the blue distribution, 68% is between 50 and 70.

Example: Male Female Jewish Muslim Other

Catholic Protestant

Ordinal Level (Grouping and Ranking) Ordinal level data can be grouped and and ranked. With Ordinal data, you can say that a measure is higher or lower than another measure. But, you may not say how much higher or lower. Figure 2. Normal distributions with standard deviations of 5 and 10. Example: Preferred Flavors of Ice Cream 1. Vanilla 2. Chocolate 3. Strawberry 4. Cherry Interval Level (Grouping, ranking, and includes the exact distance between measures) Interval level data can be grouped, ranked, and include the exact distance between measures. Note: Interval measures never contain a zero ( 0 ) as a starting point. Example: Sam is 2" taller than Bill and 3" taller than Steve.

Levels of Measurement There are four levels of measurement. They are:

Nominal Ordinal Interval, and Ratio

Associated with each level are acceptable statistical methods. The nominal level is simplest, while Ratio measures are the most sophisticated. Nominal Level (Grouping) Nominal level data is generally preferred to as the "lowest" level of measure. Data is limited to groups and categories. No numerical data is ever provided.

Bill is 2" shorter than Sam and 1" taller than Steve. Steve is 3" shorter than Sam and 1" shorter than Bill. Note: We do not know how tall Sam, Bill or Steve are, we only know exactly the difference in their heights when compared to one another. Ratio Level (Grouping, ranking, exact distance between measurement, and contains an absolute "0") Ratio level data are said to be at the highest level and can be grouped, ranked, and the exact distance between measures determined. Also, Ratio level measures contain an absolute "0". By having an absolute "0" in your measurement "scale", you are able to describe data in terms of ratios. Example: You could say that Jack, who weighs 200 lbs., is twice as heavy as Mary who weighs 100 lbs. (twice as heavy is a ratio statement). It should be noted that with social science data, there are rarely any outside standard scales, such as a yardstick to measure height. Therefore, social research rarely generates data that goes beyond Interval level measures.

Rule of Thumb #1 Use sample groups larger than 30 for interval level measures. Rule of Thumb #2 If the total population that you are examining is less than 30. Use all of them. Rule of Thumb #3 You should have a sample size of 30 for every relationship you measure. 30+ people ----------------> compared against "X" OK 15 Women ----------------> compared against "X" Not OK 15 Men -------------------> compared against "X" Not OK 30+ Women --------------> compared against "X" OK 30+ Men -----------------> compared against "X" OK Rule of Thumb #4 Consult this table Selecting Samples To select a random sample, use a table of Random Numbers or use a computerized random number generator. Note: For a more detailed discussion of sample size, see pages 385-387 in your textbook.

Sample Size The question ""how large should my sample be?" is a common one. And, one with no simple answer. While there are a number of elegant approaches to answer this question, for our purposes, several "rules of thumb" will serve us better.

Measures of Central Tendency

If you want to find a Yak-Yak bird, the first question you might ask yourself is, "Where do most Yak-Yak birds live?" In other words, "where would I have to go to have the greatest chance of finding a Yak-Yak bird?" Measures of Central Tendency tell you where most of whatever you are measuring can be found. Mean = Average All scores are added up and divided by the number of scores.

Mode = Most common score The mode is the score that occurs most often.

Range Median = Middle score Count the total number of scores. The one in the middle is the median. Range is the difference between the highest and lowest scores. You should only use the range to describe interval or ratio level data. To calculate the range, subtract the lowest score from the highest score Example:

Note: If there are an even number of scores, select the middle two and average them. This will give you the median. Note: In some statistics books, they will define range as the High Score minus the Low Score, Plus one (1). This is an inclusive measure of range, rather than a measure of the difference between two scores. For example: the inclusive range for data ranging from 6 to10 would be 5.

We should note at this point that a normal distribution (Bell Curve) is an important concept for statisticians because it gives them a "theoretical standard" by which to compare data that may not form a perfect bell curve. If you have data where the mean, median and mode are quite different, the scores are said to be skewed. For our purposes, we will define the range as the difference between the highest and lowest scores. Example: Scores: 7, 8, 9, 10, 11, 11, 12, 12, 12, 13, 13 Mean: 10.7 Media: 11 Mode: 12 Range: 6

Skewness When the mean, median and mode are equal, you will have a normal or bell shaped distribution of scores. Example: Scores: 7, 8, 9, 9, 10, 10, 10, 11, 11, 12, 13 Mean: 10 Median: 10 Mode: 10 Range: 6

Scores that are "bunched" at the right or high end of the scale are said to have a negative skew.

In a positive skew, scores are bunched near the left or low end of a scale.

Bi-modal Distribution A bi-modal distribution occurs when the data forms two clumps Standard Deviation Standard Deviation is a measure of dispersion, i.e., the extent to which data (scores) are "spread out" from the average or Mean. Other measures of dispersion are the range and variance. Since variance is not often used, and range has already been discussed, let's focus upon standard deviation. Knowing the standard deviation for a set of scores is important because it is a good indicator for judging the Mean as a representation of the "average" response. Let's look at this concept graphically. If I have a large standard deviation, it indicates that my scores are widely dispersed, and therefore, my Mean doesn't really tell me that much about the average score.

When this happens, it is a good idea to look for common characteristics within the two data clumps, and for differences between the two data clumps. For example you might generate a bi-modal distribution if you asked people, "On a scale from 1 to 10 do you like romantic movies that explore feelings and relationships?" Men would probably provide low scores, while women would provide scores that were high. Kurtosis Kurtosis refers to how tall or flat your curve is as compared to a normal curve. Curves taller than a normal curve are call Leptokurtic. Curves that are flatter than a normal curve are call Platykurtic.

On the other hand, if my standard deviation is small, it indicates that my scores are close to the Mean, and therefore, the Mean is a good indicator of the "average" score.

To calculate the standard deviation of a set of scores, use the following formula. Scores (n) = 10 Mean = 400/10 = 40 Median = 40.5 Mode = 41 Range = 16

The following should help. Let's say the data below represents the test scores for 10 trainees. (The top score anyone could make was 50).

Let me try it on the Internet So what does the standard deviation tell you? It tells you that most trainees made 40, give or take 5 points. And, that in this case, you can say with confidence that the Mean is a very good indicator of the "average" score.

When we know the standard deviation for a set of scores, it is possible to compare our data at a glance with a Normal Distribution to determine the degree of dispersion.

By establishing the standard deviation for a set of scores, you will be able to describe accurately how various scores differs from one another and from the Mean (average). Note: In a Normal Distribution:

approximately 68% of all scores will fall within one (+/-) 1 standard deviation from the Mean; approximately 95% of all scores will fall within one (+/-) 2 standard deviation from the Mean; approximately 99% of all scores will fall within one (+/-) 3 standard deviation from the Mean.

In the above example, we can see that all of our scores fell within two standard deviations of the Mean. Which again reaffirms that our Mean is a very good indicator of the "average" score.

A Quick and Dirty Way for Determining the Standard Deviation for a Set of Scores If your Mean, Median and Mode are very close to one another, you can calculated the Standard Deviation by: 1. Determine the Range 2. Divide the Range by four 3. The resulting number will approximately equal the true standard deviation Example:

Scores: 7, 8, 9, 10, 10, 10, 11, 11, 12, 13 Mean: 10 Media: 10 Mode: 10 Range: 6 True Standard Deviation: 1.67 Q&D Standard Deviation: 6/4 = 1.5 CAUTION: Only use this technique IF your Mean, Media and Mode are very similar.

Measures of Association Pearson's r Interval Level Measure There will be many times when you will want to know if two variables are "related," and to what extent. For example, you may want to find out the relationship between Yearly Income and Education. To do this, you would randomly select 10 individuals . To help you "sort out" your data, you would construct the following table. With this data, you can now use the Pearson's r or product moment correlation coefficient formula.

Measures of Association Spearman's Rank Order Ordinal Level Measure Once in a while, it is useful to compare rankings between two people or groups. Fox example, let's say a group of employees and a group of managers want to find out if there is a difference in the workplace value held by each group. In this example, each group ranks 10 workplace values.

Let me try it on the Internet Using the following table, you can see that there is a very high correlation between Income and Years of Education.

In order to calculate a Spearman Rank Order, you must first construct the following table.

Let me try it on the Internet Again, using the following table, you can see that this time there is no correlation between the two rankings.

Step 2 From this data, you can construct the following cross-tabs table.

Chi-Square Test of Significance Nominal Level Measure Let's say you wanted to know if there was any significant difference between the production rates for departments which had trained supervisors versus those departments whose supervisors were not trained. Step 1 Lets begin by collecting some data.

Note: We have included the totals for trows and columns in our table. Step 3 We must now construct a table which describes what would have happened if training did not impact production. (We call this table an Expectancy Table).

9 departments produced above standard 11 departments produced at standard 10 departments produced below standard 17 departments are supervised by trained supervisors 13 departments are supervised by untrained supervisors

A = 9 departments with above standard production B = 11 departments with standard production

C = 10 departments with below standard production M = Number of departments with trained supervisors (17) N = Number of departments with untrained supervisors (13) T = Total number of supervisors Step 4 We are now ready to calculate the Chi-square using the following formula.

Let me try it on the Internet v.1 - Chi Square Let me try it on the Internet v.2 - Contingency Table Step 5 A Chi-square by itself is not of much value. You have to use a Chi-square table that you can find in most statistics books. However, before you can use the table, you must first determine the degree of freedom of your table. To do this, blot out one row and one column. The remaining number of cells will be your degrees of freedom.

Where O = the actual production records for each cell in the table Where E = the expected production record for each cell in the table In our example table, we can see that we have 2 cells that are not filled in, which means we have 2 degrees of freedom (df) Now, using a Chi-square table, find where the 8.416 falls on the line of numbers for df 2. (Online Chi Square Table)

What this tells you is that with 98%+* confidence (probability) that supervisory training is responsible for increase production. Note: You can use Chi-square with cross-tab tables that have up to 30 degrees of freedom. (The maximum for most Chi-square tables). __________________ * .02 = 2% 100% - 2% = 98%