Introduction: Categories with an Order
In Categorical Versus Quantitative Variables, you learned that a categorical variable places individuals into groups. Some categories have a meaningful order. For example, a survey might ask a student to rate a school service as “very poor,” “poor,” “fair,” “good,” or “excellent.” These responses are categories, but their sequence carries information: “good” indicates a more favorable rating than “fair.”
A rating category is not automatically a measurement just because its choices can be arranged from low to high. The words tell us the order, but they do not tell us that the difference from “poor” to “fair” is the same size as the difference from “good” to “excellent.” In AP Statistics, treat such responses as categorical data and preserve their order when displaying and summarizing them.
What Rating Categories Tell You
A common rating question offers ordered choices such as “strongly disagree,” “disagree,” “neither agree nor disagree,” “agree,” and “strongly agree.” This is an example of a Likert-type item. The responses have an order from less agreement to more agreement, so the variable is ordinal categorical.
The order provides useful information. We can say that a response of “agree” is more favorable than “neither agree nor disagree.” But the response categories do not measure how much more favorable it is. A person’s move from “disagree” to “neither” need not represent the same change in opinion as a move from “agree” to “strongly agree.”
This differs from a quantitative measurement such as time in minutes. For a quantitative variable, numerical differences have meaning in the units of measurement. For ordinal responses, the labels indicate rank, not a known amount of change. This is why a rating scale should not be treated as quantitative merely because its categories are printed in order or stored as numbers.
Displaying Ordinal Responses
An ordered bar chart is a natural display for one ordinal categorical variable. Each bar represents a response category, and its height represents the count or proportion of responses in that category. Put the categories along the horizontal axis in their meaningful order, from lowest to highest or from least to most favorable. The vertical axis should identify whether it shows counts or proportions.
The order of the bars matters. If “strongly disagree” through “strongly agree” are rearranged alphabetically, the graph hides the scale’s progression. An ordered bar chart keeps that progression visible while still treating each response as a category. A bar chart for a nominal variable, by contrast, can order its categories in any convenient way because there is no inherent sequence.
A frequency table can also show the response categories in order, with a count and proportion for each. Counts answer “how many?”; proportions answer “what fraction of responses?” For a sample of \(n\) responses, a category’s proportion is its count divided by \(n\). If you report percentages, make clear that they are percentages of the responses in that sample.
Worked Example: Rating a New Study-Space Policy
A school asks 40 students whether they support a proposed study-space policy. The ordered response counts are shown below. Make an ordered frequency and relative-frequency summary, identify the most common response, and describe a useful cumulative proportion.
| Response | Count | Proportion |
|---|---|---|
| Strongly disagree | 3 | 3/40 = 0.075 |
| Disagree | 5 | 5/40 = 0.125 |
| Neither agree nor disagree | 8 | 8/40 = 0.200 |
| Agree | 16 | 16/40 = 0.400 |
| Strongly agree | 8 | 8/40 = 0.200 |
| Total | 40 | 1.000 |
Check the totals: The counts add to \(3+5+8+16+8=40\). The proportions add to \(0.075+0.125+0.200+0.400+0.200=1.000\), allowing for rounding if proportions are displayed with fewer decimal places.
Display: An ordered bar chart would place the categories from “strongly disagree” through “strongly agree,” with bar heights 3, 5, 8, 16, and 8 if the vertical axis shows counts. If it shows proportions, the heights would be 0.075, 0.125, 0.200, 0.400, and 0.200.
Most common response: “Agree” has the greatest count, 16, so it is the mode. It accounts for \(16/40=0.400\), or 40% of these responses.
Cumulative proportion: The proportion who selected “agree” or a category above it is \((16+8)/40=24/40=0.600\). Thus, 60% of the surveyed students selected “agree” or “strongly agree.” This combines two explicitly named categories; it does not imply that the ratings are equally spaced.
Answer: The responses are ordinal categorical data. An ordered bar chart, category proportions, the mode, and clearly defined cumulative proportions summarize their distribution without treating the response labels as measured amounts.
Summarizing the Order
For ordinal data, start with the distribution across the original categories. Counts and proportions show where responses fall, and the mode identifies the most frequent category. A cumulative count or proportion can answer a question such as “What fraction selected this category or a more favorable one?” State exactly which categories are included so the summary is clear.
The position of the responses can also be described. For instance, one can identify a middle response category by locating the middle observation or observations in the ordered list. With an odd number of responses, there is one middle position; with an even number, there are two. If the two middle responses fall in different categories, report that they straddle those categories rather than forcing them into a numeric average. For a beginner’s AP summary, the full ordered distribution and mode are often more informative and less ambiguous.
A tempting shortcut is to assign numbers such as 1 to “strongly disagree,” 2 to “disagree,” and so on, then calculate a mean. Those codes are labels chosen to represent order. The mean would depend on treating the steps between codes as equal, an assumption the response choices do not establish. A mean of 3.7 does not mean that a respondent selected a category called 3.7, either. For a single ordinal item, report the category distribution instead.
Worked Example: Customer Ratings for a Repair Desk
Thirty customers rate a repair desk as “poor,” “fair,” “good,” “very good,” or “excellent.” The counts, in that order, are 2, 4, 9, 10, and 5. Describe the distribution, including its mode and the middle positions.
Check the total and proportions: The counts sum to \(2+4+9+10+5=30\). Dividing each count by 30 gives proportions \(2/30=0.067\), \(4/30=0.133\), \(9/30=0.300\), \(10/30=0.333\), and \(5/30=0.167\), rounded to three decimal places. The rounded proportions sum to 1.000.
Mode: “Very good” is the mode, with 10 of 30 responses. Its proportion is \(10/30=0.333\), or about 33.3%.
Middle positions: With 30 responses, the middle positions are the 15th and 16th when responses are ordered from “poor” to “excellent.” The first 2 positions are “poor”; positions 3 through 6 are “fair”; and positions 7 through 15 are “good.” Positions 16 through 25 are “very good.” Therefore, the two middle responses are “good” and “very good.”
Interpretation: The central responses straddle “good” and “very good.” This describes their place in the ordered categories; it does not establish a halfway rating between them. The category counts and proportions give the fuller picture of all 30 customers’ responses.
Comparing Ordered Rating Distributions
When comparing groups, use the same ordered categories for each group and compare their category proportions, not only their counts, if group sizes differ. A table or side-by-side bar chart can display each group’s distribution. The categories should remain in the same order for both groups, and the graph should make clear whether bar heights are counts or proportions.
You may also compare the proportions at or above a stated category, such as the proportion selecting “agree” or “strongly agree.” This can be a useful summary, but it combines categories and loses some detail. Always state the cutoff and, when the full distribution matters, show the original category proportions as well.
Worked Example: Comparing Two App-Usability Ratings
Two groups of 20 users each rate a redesigned app as “very difficult,” “difficult,” “neither,” “easy,” or “very easy.” Group A has counts 2, 3, 5, 7, and 3. Group B has counts 1, 2, 4, 8, and 5. Compare the distributions using proportions and the proportion rating the app “easy” or “very easy.”
| Rating | Group A count | Group A proportion | Group B count | Group B proportion |
|---|---|---|---|---|
| Very difficult | 2 | 2/20 = 0.10 | 1 | 1/20 = 0.05 |
| Difficult | 3 | 3/20 = 0.15 | 2 | 2/20 = 0.10 |
| Neither | 5 | 5/20 = 0.25 | 4 | 4/20 = 0.20 |
| Easy | 7 | 7/20 = 0.35 | 8 | 8/20 = 0.40 |
| Very easy | 3 | 3/20 = 0.15 | 5 | 5/20 = 0.25 |
Check the group totals: Group A’s counts sum to \(2+3+5+7+3=20\); Group B’s sum to \(1+2+4+8+5=20\). Each group’s proportions sum to 1.00.
Compare the full distributions: Both groups’ largest category is “easy.” Group B has a larger proportion in “very easy” (0.25 compared with 0.15), while Group A has a larger proportion in “neither” (0.25 compared with 0.20) and in both difficult categories combined.
Compare the top two categories: In Group A, \(7+3=10\) users rated the app “easy” or “very easy,” so the proportion is \(10/20=0.50\). In Group B, \(8+5=13\) users did so, giving \(13/20=0.65\). In these groups, the proportion giving one of the two most favorable ratings is 15 percentage points higher in Group B.
Answer: The distributions suggest that Group B’s ratings are more favorable overall, particularly because a greater proportion chose “very easy” and because 65% chose “easy” or “very easy,” compared with 50% in Group A. This is a descriptive comparison of these responses, not a conclusion that the redesign caused the difference.
Common Mistakes and AP Exam Tips
- Calling an ordered rating quantitative. Having a clear low-to-high order does not make the gaps measured amounts. Identify the response as ordinal categorical unless the variable is separately defined as a numerical measurement.
- Rearranging the categories arbitrarily. Alphabetical order can hide the progression in a rating scale. Put categories in their natural order and label that order clearly.
- Taking the mean of category codes without justification. Coding responses 1 through 5 does not establish equal spacing. For one rating item, summarize the category counts and proportions rather than presenting the coded mean as if it were a measured rating.
- Comparing counts when group sizes differ. Larger groups tend to have larger counts. Compare proportions when the question concerns how each group’s responses are distributed.
- Reporting a combined category without naming it. “Positive responses” is vague unless you define which categories count as positive. Say, for example, “the proportion selecting ‘easy’ or ‘very easy.’”
- Describing only the mode. The most common response does not reveal the rest of the distribution. A full-credit description can mention the mode and use counts or proportions to describe other important features or differences.
For clear AP communication, name the variable and its ordered categories, state whether a graph shows counts or proportions, and describe the pattern in context. If you combine categories, specify the cutoff. Avoid claiming that category steps are equal or that a descriptive difference proves a cause.
Check Your Understanding
For each question, focus on preserving the distinction between an ordered category and a quantitative measurement.
- A survey asks residents to rate neighborhood noise as “not a problem,” “small problem,” “moderate problem,” or “major problem.” Is this variable ordinal, nominal, or quantitative? Explain what its order does and does not tell you.
- A five-category rating question has counts 4, 6, 12, 10, and 8, listed from least favorable to most favorable. What is the total number of responses, and which category is the mode?
- For the counts in Question 2, calculate the proportion in the most favorable category. Give your answer as a decimal rounded to three places.
- Two groups have different sample sizes. Why might relative frequencies be more useful than raw counts for comparing their ordered rating distributions?
- A student codes “poor,” “fair,” “good,” and “excellent” as 1, 2, 3, and 4, then reports a mean rating. What assumption does this calculation make about the scale, and why is a category distribution safer for a single ordinal item?