How to read survey crosstabs

Start with a table you can check

The table below uses our 250-row meal-kit CSV. These are constructed demonstration data, not findings from a research run or a survey of customers. The statement is: “This meal kit would make weeknight dinners easier.” Agreement combines “Strongly agree” and “Somewhat agree.”

Ask what each percentage divides by

For ages 25–34, 90 agreeing answers divided by 125 valid answers gives 72%. For ages 35–44, 65 divided by 125 gives 52%. Each percentage describes answers within that age group. The two agreement percentages do not need to add to 100%, because they have different denominators.

A different question is: “Of all agreeing answers, what share belong to ages 25–34?” That is 90 ÷ 155, or about 58.1%. It describes the composition of the agreeing group, not agreement among younger respondents. Do not switch between these interpretations without changing the label.

Recompute the overall result from counts

There are 155 agreeing answers among 250 total answers: 155 ÷ 250 × 100 = 62%. Averaging 72% and 52% also works here only because the two age groups have equal bases. When bases differ, add the relevant counts and divide by the combined base instead of taking a simple average of group percentages.

If a report applies weights, use its documented weighted totals and keep the unweighted base visible. This demonstration is unweighted. Do not add weights merely to make a result look more representative.

Describe the difference before explaining it

The gap between 72% and 52% is 20 percentage points. It is not a 20% relative increase. A relative comparison would be (72 − 52) ÷ 52, approximately 38.5%, but that is usually a less direct way to describe this table.

Neither calculation explains why the groups differ. Age may travel with other characteristics, and this dataset was constructed for demonstration. A useful next question is whether the proposed convenience benefit deserves follow-up—not whether this table proves younger customers will buy.

Check exclusions and multiple-choice rules

Before calculating a percentage, distinguish missing answers, invalid answers, “not applicable,” and substantive choices. State which belong in the base. If 10 of 125 answers are missing and 60 of the remaining 115 agree, valid-answer agreement is 60 ÷ 115, about 52.2%; using all 125 would give 48%. This is a separate arithmetic illustration, not a change to the downloadable sample.

For a select-all-that-apply question, percentages of respondents can sum above 100%. For an exclusive single-choice question, valid category counts should add to the valid base. Check those rules before assuming a discrepancy is a rounding error.

Use audience differences as a question, not automatic proof

Choose relevant comparisons before examining every possible subgroup. Record the bases, exact question, observed difference and next action together. Small groups can produce visually large movements from a few answers. Generated answers also require evidence about the simulation itself; a larger generated count does not establish population accuracy.

A restrained example conclusion is: “The constructed table illustrates a stronger convenience response in the younger group. In an actual study, this would be a reason to examine the pattern and test the benefit with eligible consumers.” No significance test is implied. For disclosure principles, see AAPOR’s survey research best practices.