Comprehensive Guide to Describing Data through Distributions and Graphs

Distinguishing Between Qualitative and Quantitative Variables

Variables are fundamentally classified into two distinct categories: qualitative and quantitative. Qualitative variables, which are also frequently referred to as categorical variables, represent data that is nominal in nature. These variables categorize observations into groups based on shared characteristics rather than numerical value. In contrast, quantitative variables are those measured on a numeric scale, providing data that represents quantities or amounts that can be mathematically manipulated and compared.

Frequency Distributions and Tabular Representations

A frequency distribution is a descriptive tool that characterizes the pattern of a set of numbers by displaying either a count or a proportion for every possible value of a variable. It serves as a foundational method for understanding how data points are distributed across a range of values. One common visual depiction of this data is a frequency table, which lists each specific value along with its frequency of occurrence and its relative frequency. For example, a frequency table for test scores might show that a score of $A$ occurred with a frequency of $5$ and a relative frequency of $.24$, while a score of $B$ had a frequency of $4$ and a relative frequency of $.19$. A score of $C$ appeared $8$ times with a relative frequency of $.38$, and scores of $D$ and $E$ each had a frequency of $2$ and a relative frequency of $.10$. This tabular format allows researchers to quickly identify the most and least common outcomes within a dataset.

When a researcher deals with a wide range of values, a grouped frequency table is often more effective. This method involves depicting data visually by reporting frequencies within a specified interval rather than for every individual value. An example of this is a distribution of reaction times where frequencies are tallied into ten-millisecond blocks. In such a table, the interval of $41-50\,ms$ might have a frequency of $4$, the $31-40\,ms$ interval a frequency of $11$, $21-30\,ms$ a frequency of $10$, $11-20\,ms$ a frequency of $2$, and the $0-10\,ms$ interval a frequency of $1$. This grouping helps to manage large datasets and provides a clearer picture of the overall data trends.

Graphical Methods for Displaying Categorical and Continuous Data

Pie charts are circular graphs divided into slices, where each slice represents a specific category. The physical size of each slice is proportional to the percentage or proportion of the category it represents. A critical rule for pie charts is that the sum of all slices must always equal $100\%$ or a total proportion of $1.00$. For instance, a pie chart might illustrate student enrollment with $71\%$ part-time students and $29\%$ full-time students. However, it is advised that researchers use pie charts sparingly because they can be difficult to interpret compared to other graphical forms. A bar chart is a more standard alternative where categories are laid out on the x-axis and the frequency of those categories is measured on the y-axis. These are particularly useful for comparing distinct groups, such as frequencies of different college majors like Math, Psych, or Music.

Histograms serve a similar purpose to bar charts but are specifically used to depict continuous data for a single variable. In a histogram, the values of the variable are positioned on the x-axis and the frequencies are on the y-axis. Unlike bar charts representing discrete categories, the bars in a histogram represent intervals of a continuous scale. A stem-and-leaf display provides another way to visualize distribution, particularly when the dataset is not too numerous. In this display, the "stem" on the left represents the tens digits of the data points, while the "leaves" on the right represent the ones digits, allowing the viewer to see every individual data point while simultaneously observing the shape of the distribution.

Specialized Data Visualizations: Box Plots and Line Graphs

A box plot is a specialized graph used to provide a snapshot of the overall distribution of a dataset. The box itself is defined by the first quartile at the lower end and the third quartile at the upper end. A horizontal line drawn through the middle of the box represents the median of the data. Extending from the box are "whiskers," which reach out to the minimum and maximum scores recorded in the dataset. Box plots are also essential for identifying outliers. An outlier is defined as any data point that is more than $1.5$ times the interquartile range ($IQR$) higher than the third quartile or lower than the first quartile. In such cases, these points are represented as individual dots and are not included as part of the whisker range.

Line graphs are utilized primarily to illustrate the relationship between two continuous variables. This is common when researchers want to see how one variable changes in response to another. For example, a line graph can demonstrate the relationship between time spent studying, measured in hours, and test scores, measured as a percentage. The graph would plot study times (such as $0, 1, 2, 3, 4, 5, 6,$ and $7$ hours) on the x-axis against corresponding test scores (ranging from $0\%$ to $100\%$) on the y-axis, allowing for the visualization of trends such as whether more study time correlates with higher performance.

Guidelines for Ethical and Effective Graph Construction

Designing a graph requires a systematic two-step process to ensure clarity and accuracy. Step one involves examining the variables to decide which is the predictor variable, assigned to the x-axis, and which is the outcome variable, assigned to the y-axis. It is also necessary to identify the data type as nominal, ordinal, interval, or ratio. Step two involves selecting the correct graph type based on the data: use a histogram for one continuous variable with frequencies; a scatterplot or line graph for one continuous predictor and one continuous outcome; and a bar graph when the predictor is a nominal or ordinal variable paired with a continuous outcome.

Researchers must remain vigilant to avoid graphical mistakes that can mislead the audience. One major error is the use of 3D bars, which can distort the viewer's perception of the actual data values. Another significant concern is the Lie factor, which is defined as the ratio of the size of the effect shown in the graph to the actual size of the effect present in the numerical data. Furthermore, starting a vertical axis at a value other than zero can exaggerate differences between groups. If starting at zero is not possible or practical, the graph must include explicit "cut marks" to signal the broken axis to the reader.

Geometric Shapes of Data Distributions

Distributions often take on specific shapes that provide insight into the nature of the data. A normal distribution is characterized as a symmetric, unimodal, bell-shaped curve where most scores cluster around the center. Conversely, skewed distributions occur when one of the tails of the distribution is pulled away from the center. A distribution is described as positively skewed when the tail extends toward the right, in a positive direction. This often happens due to a floor effect, which is a constraint that prevents values from falling below a certain minimum. An example is giving a $6^{th}$ grade reading comprehension test to $3^{rd}$ graders; because the test is too difficult, many students will score at or near zero. While variation exists, the test cannot differentiate between low knowledge levels because of the lower boundary.

Negatively skewed distributions occur when the tail of the distribution extends to the left, in a negative direction. This is frequently the result of a ceiling effect, where a constraint prevents values from exceeding a specific maximum. A common example of a ceiling effect is a physical fitness test with a very low difficulty level administered to military recruits. In this scenario, most recruits will achieve the maximum possible score. Although there is actual variation in fitness among the recruits, the test is too easy to differentiate between high levels of fitness, causing a cluster of scores at the top end of the scale.