Picture a stack of survey cards, one per student. Each card has two answers: grade (7th or 8th) and after-school choice (sports or music). Sort the cards into four piles, one for each combination: 7th-sports, 7th-music, 8th-sports, 8th-music. Write the size of each pile in a grid, grades down the side and activities across the top, and you have a two-way table. Add across each row and down each column, and the margins tell you how many 7th graders there are, how many chose music, and how many cards there are in all.
The grid exists so you can compare. Suppose 30 of 50 seventh graders chose sports and 35 of 70 eighth graders did. The 8th-grade count is bigger, but the rate is not: 30/50=60% and 35/70=50%. To compare rows fairly, divide each cell by its own row total. The question decides the total: "what share of 7th graders chose sports" uses the 7th-grade row total, while "what share of sports players are in 7th grade" uses the sports column total, 30/65≈46%.
Rows and columns alone do not make a two-way table. A table listing each student's height and arm span also has rows and columns, but each row is one person and each cell is a measurement, not a count of people. Adding a height to an arm span gives a meaningless total. If the row labels and the column labels are both categories and the cells are counts, it is a two-way table; if each row is one individual with two numbers, it is paired data for a scatter plot.
Think of it like this. A two-way table works like a sorting tray with one compartment for every pair of answers: each card drops into exactly one slot, so the compartments add up to the whole stack. The tray shows how many cards landed where, but not why they landed there.
Core idea. Sort the same individuals by two category questions at once, count each combination, and compare the groups using percentages of each group's own total rather than raw counts.