Read a rating scale without losing the distribution

Keep the labels attached to the values

Our meal-kit demonstration uses five options: strongly agree, somewhat agree, neither agree nor disagree, somewhat disagree, and strongly disagree. The statement is “This meal kit would make weeknight dinners easier.” The downloadable CSV contains 250 constructed answers for showing the report format, not evidence from a research run.

Write a single proposition that can reasonably be agreed or disagreed with. “This meal kit is affordable and convenient” combines two judgments. A reader could agree about convenience and disagree about affordability, leaving the answer ambiguous. Ask separately if both judgments matter.

Show all five categories first

• Strongly agree: 65 of 250, or 26%.
• Somewhat agree: 90 of 250, or 36%.
• Neither agree nor disagree: 40 of 250, or 16%.
• Somewhat disagree: 32 of 250, or 12.8%.
• Strongly disagree: 23 of 250, or 9.2%.

These counts sum to 250, and the percentages sum to 100%. Agreement is concentrated more in “somewhat” than “strongly.” That distinction disappears if the headline only says “62% agree.” The combined negative share is 22%, not the 38% left after subtracting agreement: that remainder also includes neutral answers.

Define a top-two-box calculation explicitly

For this item, top-two-box agreement means adding strongly and somewhat agree: (65 + 90) ÷ 250 × 100 = 62%. Top-box agreement is just the strongest positive category: 65 ÷ 250 × 100 = 26%. Neither is universally the right success metric. Pick the measure before looking for a favorable headline, and report the categories it includes.

In a reversed numerical coding scheme, the positive categories may have the lowest numbers. Select categories by meaning, not simply by taking the two largest stored values. When comparing studies, keep question wording, labels, exposure and calculation consistent.

Neutral is not the same as not applicable

“Neither agree nor disagree” is a position on the agreement scale. “Not applicable” or “I do not have enough information” describes a different reason for not giving a directional judgment. Decide whether those options belong before running the study. Do not recode them into neutral afterward without explaining the change.

For a benefit a person cannot assess from the concept, adding detail to the stimulus may be more useful than forcing a choice. In a simulation, including an uncertainty option does not by itself prove the system accurately recognizes its knowledge limits.

Treat averages as a disclosed coding choice

If strongly disagree is coded 1 through strongly agree as 5, this example has a coded mean of (23 × 1 + 32 × 2 + 40 × 3 + 90 × 4 + 65 × 5) ÷ 250 = 3.568. That arithmetic is reproducible, but interpreting the spacing between labels as equal is an assumption. A mean also hides how answers are distributed.

Keep the labeled distribution alongside any average. Do not compare a 1–5 mean with a 1–7 mean as if they used the same scale. Avoid treating one agreement item as a validated multi-item measurement instrument.

Tie the result to the question actually asked

Agreement that dinners would be easier is not purchase intent, willingness to pay or observed buying. A next question might present a complete offer and price; a next real-world test might ask eligible people to try the meal kit. The constructed result provides neither a conversion forecast nor proof of audience demand.

When designing your own scale, use clear wording and meaningful answer options. Pew Research Center’s questionnaire guidance explains why wording, options and ordering deserve deliberate review. Our crosstab guide carries this same dataset into age-group calculations.