Eighth Grade Mathematics: Identifying Associations in Categorical Data

The central reason for collecting bivariate categorical data is to evaluate whether an association exists between the two variables. In medical research, scientists study whether a new vaccine is associated with a lower infection rate; in business, marketers analyze whether advertising campaigns are associated with product purchases. In statistics, an association exists between two categorical variables if the distribution of one variable differs depending on the category of the other variable. If the proportions remain identical across all categories, the variables are independent. In this chapter, you will master using conditional relative frequencies to prove or disprove the existence of statistical associations.

The Mathematical Definition of Categorical Association

Two categorical variables are said to have an association (or statistical dependency) if the conditional relative frequencies of one variable vary across the categories of the second variable:

The Criterion for Association:
- There IS an Association:
  If the conditional relative frequencies for a category are SIGNIFICANTLY DIFFERENT
  across different groups. Knowing an individual's category in Variable A changes the
  probability of their category in Variable B.

- There is NO Association (Independence):
  If the conditional relative frequencies for a category are ROUGHLY EQUAL across all groups.
  Knowing an individual's category in Variable A provides NO predictive power regarding Variable B.

Visualizing Association vs. Independence:

Scenario A: Strong Association (Music Preference by Age Group)
                Prefers Pop     Prefers Classical
Teens           85%             15%
Senior Citizens 20%             80%
(85% is vastly different from 20%! There is a strong association between age and music preference!)

Scenario B: No Association (Music Preference by Handedness)
                Prefers Pop     Prefers Classical
Right-Handed    60%             40%
Left-Handed     60%             40%
(60% equals 60%! Handedness has zero effect on music preference; the variables are independent!)

The Step-by-Step Association Evaluation Protocol

To prove whether an association exists between two categorical variables, execute this three-step protocol:

The Association Evaluation Protocol:
Step 1: Calculate the CONDITIONAL RELATIVE FREQUENCIES for one of the variables
        (convert raw row counts into row percentages, or column counts into column percentages).

Step 2: Compare the resulting conditional percentages across the categories.

Step 3: State a Formal Conclusion:
        - If the percentages show a clear, meaningful difference:
          Conclude that an association EXISTS, and describe the direction of the trend in context.
        - If the percentages are nearly identical:
          Conclude that there is NO EVIDENCE OF ASSOCIATION (the variables are independent).

Never compare raw frequencies to evaluate association! A group might have more raw counts simply because there were more total people surveyed in that group. Only conditional relative frequencies provide a fair, normalized percentage comparison.

Contextual Pitfall: Association Does Not Mean Causation

One of the most famous principles in modern science is: Association does not imply causation.

Finding a strong statistical association between two categorical variables simply proves that they occur together more frequently than expected by chance; it does NOT prove that one variable causes the other. A lurking variable (a third, unmeasured variable) may be driving both characteristics simultaneously. For example, a strong association exists between owning larger shoes and having higher reading levels among children; however, buying larger shoes does not cause a child to read better. The underlying lurking variable is age: older children have both larger feet and more advanced reading education.

Chapter Exercises

Exercise 1: A school surveys 200 students to evaluate whether there is an association between participating in extracurricular clubs and maintaining an A-grade average: of the 120 students in clubs, 72 have an A average; of the 80 students not in clubs, 24 have an A average. Calculate the conditional relative frequencies of having an A average for each group and determine whether an association exists. State your conclusion in paragraph form.

Exercise 2: A clinic tests whether drinking green tea is associated with catching a winter cold: among 100 tea drinkers, 20 caught a cold; among 100 non-tea drinkers, 21 caught a cold. Calculate the conditional relative frequencies and evaluate whether an association exists between green tea consumption and catching a cold.

Exercise 3: Explain in paragraph form why evaluating association requires comparing conditional relative frequencies rather than raw frequency counts, providing a numerical counterexample where raw numbers mislead.

Solutions and Explanations

Solution 1: To determine whether an association exists between participating in clubs and maintaining an A average, calculate the conditional relative frequency of having an A average within each group. For students who participate in clubs, 72 out of 120 students have an A average: 72 / 120 = 0.60, or 60 percent. For students who do not participate in clubs, 24 out of 80 students have an A average: 24 / 80 = 0.30, or 30 percent. Comparing the two conditional proportions: 60 percent of students in clubs maintain an A average, compared to only 30 percent of students not in clubs. Because these two conditional relative frequencies are substantially different (60% versus 30%), there is a clear statistical association between participating in clubs and maintaining an A average. Students who participate in extracurricular clubs are twice as likely to maintain an A average as those who do not.

Solution 2: To evaluate whether drinking green tea is associated with catching a winter cold, compute the conditional relative frequencies of catching a cold for both groups. For green tea drinkers, 20 out of 100 individuals caught a cold: 20 / 100 = 0.20, or 20 percent. For non-tea drinkers, 21 out of 100 individuals caught a cold: 21 / 100 = 0.21, or 21 percent. Comparing the two conditional relative frequencies: 20 percent versus 21 percent represents an essentially identical infection rate (the 1 percent difference is negligible and well within random sampling fluctuation). Because the probability of catching a winter cold is virtually identical regardless of whether an individual drinks green tea, there is no evidence of an association between green tea consumption and catching a winter cold in this study; the two variables appear to be statistically independent.

Solution 3: Evaluating association requires comparing conditional relative frequencies rather than raw counts because raw frequency counts fail to account for differing group sizes. If one group is substantially larger than another, its raw numbers will naturally be higher even if the underlying rate or tendency is identical. For example, suppose a survey examines 1,000 adults and 100 teenagers regarding bicycle ownership: 200 adults own bicycles, and 50 teenagers own bicycles. Looking strictly at raw counts, four times as many adults own bicycles as teenagers (200 versus 50), which might mislead an untrained observer to conclude that adults are more associated with bicycling. However, converting to conditional relative frequencies reveals the true reality: the proportion of adults who own bicycles is 200 / 1,000 = 20 percent, while the proportion of teenagers who own bicycles is 50 / 100 = 50 percent. Teenagers are actually two and a half times more likely to own a bicycle than adults (50% versus 20%). Relying on raw counts produces the completely backwards conclusion, whereas conditional relative frequencies normalize for sample size and reveal the true association.