Statistical Theory and Methods Statistical Theory and Methods

2 Measures of Central Tendency (Averages)

Besides using descriptive statistics for charts and graphs, trainers can perform several
Statistical Theory and Methods numeric calculations on them. The most common are called measures of central tendency,
or averages, of which there are three: mean, median, and mode. Each type of average
Statistics allow workplace learning and performance (WLP) professionals to quantitatively serves a unique purpose.
describe and draw inferences about people, things, or events. In other words, statistics allow
data to be organized and summarized and make it possible to draw generalizations and Mean
inferences. Statistics enable WLP professionals to document current levels of performance The mean score is considered the most robust, or least affected by the presence of extreme
(individual, group, or organizational), measure the impact of their programs, and offer well- values (outliers), of the three types of central tendency measures, because each number in
grounded feedback for change. the data set has an impact on its (mean) value.
For many WLP professionals, the use of statistics is an onerous task—but it shouldn’t be. The mean is represented by the following formula:
Several software applications can do the number crunching for trainers; however,
practitioners must understand how to use statistics. The selection and interpretation of Mean = Sum of all numbers divided by the number of values that make up the sum
statistics still rests in their hands. To use statistics properly, a statistics consumer needs to The mean is a good measure of central tendency for roughly symmetric distributions but can
understand some essential concepts and principles. Although this chapter is designed to be misleading in skewed, or nonsymmetric, distributions because it can be influenced a great
serve as a primer for statistics, occasionally calculations are used to help understanding. deal by extreme scores. Therefore, other statistics, such as the median, may be more
WLP professionals must have a broad understanding of how data falls into distributions (for informative and appropriate for distributions that are often quite skewed, such as reaction
example, variance and normal distribution) and how data relates to other data (for example, time or family income.
correlation and regression). In addition, from an inferential standpoint, WLP professionals Median
must understand concepts related to hypothesis testing, such as effect sizes and confidence
intervals. It’s easy to misuse or misinterpret statistics. Having a real understanding of The median is the middle of a distribution arranged by magnitude: Half the scores are
statistics means that people can apply them correctly, represent findings accurately, and draw above the median, and half are below the median. The median is less sensitive to extreme
appropriate inferences. scores than the mean, which makes it a better measure than the mean for highly skewed
distributions. The median income is usually more informative than the mean income, for
Learning Objectives: example.
; Define and illustrate the three measures of central tendency. The median in a distribution of odd- or even-numbered values, as noted in Table 1-1, is the
; Define and compare the various types of frequency distributions. calculated average of the two numbers of the high and low side of the nth number in the
; Express how measurement scales and statistical implications are used in the formula.
collection of measurement data. Which of these two measures—the sample mean or median—should trainers use? It
; Explain how the measures of variance are used in statistics. depends. Usually medians are a better measure for skewed distributions than means, but this
; Describe how distributions are used with standard scores. guideline must be tempered by common sense. In a dispute between the American Medical
; Identify the correct usage of correlation versus causation in data. Association and the American Bar Association about the rising costs of malpractice
; List the five steps in the hypothesis-testing process. insurance for doctors, the doctors used means to show a sharp rise in costs in the period
; Demonstrate knowledge related to effect sizes and confidence intervals. 1980 to 1984, and the lawyers used medians to show that there was no rise at all (Schwarz
; Recognize the appropriate use of statistical information. 1998).
Mode
The mode, the most frequently occurring score in a distribution, is also used as a measure of
central tendency. The advantage of the mode as a measure of central tendency is that its
meaning is obvious. Further, it’s the only measure of central tendency that can be used with
nominal data.
 Statistical Theory and Methods Statistical Theory and Methods

The mode is greatly subject to sample fluctuations, so it’s not recommended for use as the Frequency Distributions
only measure of central tendency. A further disadvantage of the mode is that many
distributions have more than one mode; these distributions are called multimodal. A set of numbers may be summarized in two major ways: using pictures and using summary
numbers. Each method has advantages and disadvantages, and the use of one method need
The mean score is considered the most robust of the three types of central tendency because not exclude the use of the other. This section describes drawing pictures of data called
each number in the data set has an impact on the value of the mean. The median and the frequency distributions.
mode can be unaffected by individual numbers.
A frequency distribution can show the actual number of observations falling in each range or
Table 2-1 shows a sample set of data collected by a trainer. These scores represent the percentage of observations. With percentage of observations, the distribution is called a
pretest scores on a knowledge test a trainer administered before the start of a training relative frequency distribution.
program. This data is used to show the three types of measures of central tendency as well as
variance. Some conditions that a frequency distribution might illustrate include dispersion, clusters,
skewness, outliers, and normal distribution. The next section explores these concepts in
Table 2-1. Pretest Scores (Arranged From High to Low) more depth.
Scores (x) Mode (mo) Median (mdn)a Definition of Terms
29 The mode is the mdn=(n+1) ÷2 Confounding variable is an unknown or uncontrolled variable that produces an effect
35 most frequently in an experimental setting. A confounding variable is an “independent variable” that the
occurring number evaluator didn’t somehow recognize or control. It becomes a variable that confounds the
37 mdn=(12+1) ÷2
37 experiment.
mo=37
46 mdn=6.5th number Continuous variable is a variable whose quantification can be broken down into
52 extremely small units (for example, time, speed, distance).
56 mdn=54 Control group is a group of participants in an experiment that’s equal in all ways to the
59 experimental group except that it didn’t receive the experimental treatment.
61 Covariates are the multiple dependent variables in a study with multiple independent
73 variables.
77
Dependent variable is frequently thought of as the “outcome,” or treatment variable.
82
The dependent variable’s outcome depends on the independent variable and covariates.
Dichotomous variable is a variable that falls into one of two possible classifications
∑x=644
(for example, gender [male or female]). An artificially dichotomous variable is imposed
for classification purposes (for example, age classified as retired [>65] or not retired
The number of values, n, equals [<65]).
12
Discrete variable is a variable in which the units are in whole numbers, or “discrete”
The mean = ∑x÷n
units (for example, number of children, number of defects).
The mean = 644 ÷12=53.67
Experimental group is the treatment group; those participants who receive the
∑x = sum of values in the list “treatment,” for example, the training program.
n = number of values in the list
Independent variable is the variable that influences the dependent variable. Age,
a Note that the formula in the example applies only to a case with an even
number of values. For the odd number of values it is not necessary to average seniority, gender, shift, level of education, and so on may all be factors (independent
the two middle values to come up with the “virtual median.” In the case of an variables) that influence a person’s performance (the dependent variable).
odd number, the appropriate formula is (n+1÷2) ÷2.
 Statistical Theory and Methods Statistical Theory and Methods

random error. The test can’t consider biases resulting from nonrandom error (for example, a Appropriate Use of Statistical Information and Data
badly selected sample).
As Mark Twain said, “Collecting data is like collecting garbage. Pretty soon, we have to do
These are some key concepts about statistical significance: something with it.” If not used properly, statistical information and evaluation data are
• In statistical terms, significant does not necessarily mean important. useless. Improper use of evaluation data can lead to four major problems:
• Probability values should be read in reverse. • Too many organizations don’t use evaluation data at all. In these situations, data is
collected, tabulated, catalogued, filed, and never used by any particular group other
• Too many significance tests turn up some falsely significant relationships.
than the person who initially collected the data.
• It’s important to check the sampling procedure to avoid bias.
• Data is not provided to the appropriate groups. Different groups need different
Effect Sizes types of data and often in very different formats. Analyzing target audiences and
determining the specific data needed for each group are important for
Effect size is a way of quantifying the difference between two groups. For example, if one communicating data.
group (the treatment group) has had an experimental treatment and the other (the control
group) has not, the effect size is a measure of the effectiveness between the two groups. • Data isn’t used to drive improvement. Most evaluation data uncovers process
Effect size uses standard deviation to contextualize the difference between the two groups. improvement opportunities and identifies features that could be adjusted or changes
that should be made to make the program more effective. If it’s not part of the
Confidence Intervals feedback cycle, evaluation falls short of what it’s intended to do.
The confidence interval is the range where something is expected to be. Saying “expected” • Data is used for the wrong reasons—to take action against a person or a group or to
leaves open the possibility of being wrong. The degree of confidence measures the withhold funds rather than improve processes. Sometimes the data is used in
probability of that expectation to be true. political ways to gain power or advantage over another person.
The degree of confidence is linked with the width of the confidence interval. It’s easy to be These problems represent dysfunctional activities that can destroy evaluation processes.
very confident that something will be within a very wide range, and vice versa. Also, the They must be addressed if evaluation is to add value.
amount of information (typically related to the sample size) has an influence on the degree
of confidence and the width of the confidence interval. With more information, there can be
more confidence that what’s being measured will be within a given interval. Also, with more
information and keeping a given degree of confidence, the interval can be narrowed.
For example, say a survey is conducted in Alexandria, Virginia. The question is “Do you
prefer Coca-Cola or Pepsi?” Of the responses, 60 percent answer Coca-Cola, and 40 percent
answer Pepsi. So the estimation is that, in this city, 60 percent prefer Coca-Cola. This doesn’t
mean that 60 percent of the population in this city prefers Coca-Cola—unless everyone in
the population answered the survey. However, there’s some “confidence” that the actual
proportion of people choosing Coca-Cola will be within some interval around the 60 percent
found in the sample. The amount of confidence depends on how wide the interval is. If the
survey is based on a sample of 100 people, there can be 90 percent confidence that the actual
proportion of those preferring Coca-Cola will be between 52 percent and 68 percent. Also,
there can be 99 percent confidence that the actual proportion will be between 48 percent
and 72 percent (for the same sample size, with more confidence and a wider interval). If the
survey had been on a sample of 1,000 people instead of 100, there could be 90 percent
confidence that the actual proportion is between 57.5 percent and 62.5 percent (compared
with 52 percent and 68 percent for the same confidence with a sample of 100). Keep in
mind that the larger the sample, the higher the degree of confidence.
 Statistical Theory and Methods Statistical Theory and Methods

9 Chapter 2 Knowledge Check 6. Which of the following types of data include the feature of identifying an absolute
zero point?
1. Which of the following best describes a situation where the mode < median <
mean? a. Nominal
a. Negative skewness b. Ordinal
b. Positive skewness c. Interval
c. Outlier d. Ratio
d. Normal distribution 7. Variance is defined as how spread out a distribution of data points is, whereas the
standard deviation is the measure of how spread out the data points are when the
2. Which of the following best describes normal distribution? mean is used to calculate central tendency.
a. An observation in a data set that’s far removed in value from others in the data a. True
set
b. False
b. The symmetry in the distribution of the same data values
8. The reason that practitioners convert raw scores to standard scores includes which
c. The way in which observations tend to pile up around the mean, also known as of the following?
the bell-shaped curve
a. To indicate the number of correct answers to allow scores to be compared
d. Variation in values that could be widely scattered or tightly clustered
b. To reflect where they fall with respect to the mean to allow scores to be
3. Which of the following best describes dispersion? compared and interpreted
a. An observation in a data set that’s far removed in value from the others in the c. To understand the cause-and-effect connections between variables
data set
d. Because they are always expressed as a number between –1.00 and +1.00
b. The symmetry in the distribution of the same data values
9. An example of a relational study statistic that measures the relationship between two
c. The way in which observations tend to pile up around the mean, also known as or more variables includes
the bell-shaped curve
a. Correlation coefficient
d. Variation in values that could be widely scattered or tightly clustered
b. Cause-and-effect connection
4. Which of the following best describes an outlier?
c. Normal distribution
a. An observation in a data set that’s far removed in value from the others in the
data set d. Skewness
b. The symmetry in the distribution of the same data values 10. The primary goal of hypothesis testing is to test a hypothesis and then accept or
reject the hypothesis based on the findings.
c. The way in which observations tend to pile up around the mean, also known as
the bell-shaped curve a. True
d. Variation in values that could be widely scattered or tightly clustered b. False
5. Which of the following types of data make it possible to rank order items measured
in terms of which has less or more of the quality represented?
a. Nominal
b. Ordinal
c. Interval
d. Ratio