Top 2 Box Score: How to Calculate It, When It Beats the Mean and How Precise It Is

SurveyMonkey describes it as "a way of summarizing the positive responses from a Likert scale survey question" and gives a purchase-intent example on a 5-point scale: 15% "extremely likely" plus 25% "very likely" gives a top 2 box of 40% (SurveyMonkey). The same page explains that bottom 2 box simply adds the two lowest responses.

The logic does not change with scale length, but the numbers do. A 7-point scale spreads opinion over more points, so a T2B from a 5-point question and one from a 7-point question are different measures. Keep the scale, labels and order identical across waves and groups.

1. Count the respondents who picked the top two points.
2. Divide by the number of respondents who answered the question.
3. Multiply by 100.

The denominator is where mistakes happen. Effectory's help page says the score is calculated by dividing the count by "the total number of respondents", so neutral and negative answers stay in the base (Effectory). Decide in advance whether "don't know" and "not applicable" answers are excluded from the base, write the rule down, and apply it to every wave.

Variants exist. SurveyMonkey also describes a weighted top 2 box, which gives the top point more weight than the second, and calls the weights "the somewhat subjective piece of the equation". If you use one, publish the weights.

The counts are invented to show the arithmetic and are not survey results. Two brands each get 400 ratings on a 5-point satisfaction scale.

Both brands have a T2B of 65.0%. Brand A has the higher mean (3.85 against 3.80) and fewer unhappy customers. Brand B has more enthusiasts and more people at the bottom. SurveyMonkey makes the same point: do not rely on the aggregate T2B in isolation "when comparisons are close in magnitude".

Which brand is stronger depends on the question. If you worry about complaints and churn, read bottom 2 box. If you are after advocates, read top box.

Effectory contrasts the two: T2B shows the percentage choosing the top two answers and is "less sensitive to neutral or negative responses", while the average is sensitive to the full range of responses. The page recommends T2B for international organisations because it is "understandable for everyone".

Report both, plus the full distribution in a chart or appendix. The mean treats the gap between 1 and 2 as equal to the gap between 4 and 5, which is an assumption about how people use the scale. T2B makes no such assumption but throws away the detail inside each half.

T2B is a proportion, so the usual interval for a proportion applies. NIST's e-Handbook gives the common normal-approximation formula as p ± z × √(p(1 − p) ÷ n), notes that it can produce a lower limit below zero, and recommends the Wilson interval as the better general choice (NIST).

In the example, 260 of 400 gives 65.0%. The 95% normal interval runs from 60.3% to 69.7%, and the Wilson interval from 60.2% to 69.5%. The two nearly agree at this size. They diverge for small bases or scores near 0% or 100%.

The widest margin occurs at 50%. At 95% confidence it is:

These margins assume a simple random sample. Weighting, clustered samples and panels that are not random samples change them, so ask your analyst how they were calculated. To compare two independent groups, test the difference between the proportions (see our statistical significance guide); do not just check whether the two intervals overlap. If the same people rated both brands, the comparison must account for that pairing.

• Changing the base between waves. Excluding "don't know" in one wave and including it in the next moves the score.
• Comparing across scale lengths. A 5-point T2B and a 7-point T2B are not interchangeable.
• Reading a tie as equality. The example above shows why you also need bottom 2 box and the mean.
• Reporting to a decimal place on a small base. A 100-answer base carries a margin near ±10 points.
• Treating a T2B target as a result. "Reach 70% T2B" means little unless the question wording, scale and audience are fixed.

On Hercules, Poseidon selects its formulas by question type, and top-2-box is among them (panel methodology). Whichever tool computes your score, check the base and the scale before you read it. The Free plan includes 30 responses, enough to see how your rating question looks on a phone before a full study. For a related single-number measure, see our NPS methodology guide.