Every quantitative survey question in your results now comes with a Response Summary - a compact statistics panel directly beneath each chart, plus interactive overlays you can drop onto the chart itself. Together they turn a bar chart into a decision-ready read on your testers: not just what they answered, but how strong, how consistent, and how certain that signal is.
You don't need a statistics background to use it. The headline numbers tell you the story at a glance, and the deeper measures (confidence intervals, interquartile range, standard deviation) are there for when you need to defend a decision or quantify how reliable a result is.
Where to find chart statistics
Open any test's Survey Results and scroll to a quantitative question (multiple choice, rating scale, ranking, etc.). Below each chart you'll see the Response Summary strip. It opens collapsed, showing only the headline stats. Click Show more to expand the full statistics panel, and Show less to collapse it again.
Expanded example:
Data type detected automatically
Different questions call for different statistics: an agreement scale isn't summarized the same way as a "pick your favorite" question. BetaTesting automatically classifies each question into a measurement type and shows only the statistics that are meaningful for it. The data type appears as a small chip next to Response Summary (e.g. Likert, Numeric Rating, Nominal · single-select). Hover the chip for a one-line description.
Type | What it is | Headline stats you'll see |
Likert | An agreement scale (Strongly Disagree → Strongly Agree) | Average (signed), Mode, Median, plus full spread & confidence measures |
Numeric Rating | A numeric scale (e.g. rate 1–10). 0–10 recommendation scales also show NPS | Average, Median, Range, Net Promoter Score |
Ordinal | Ranked categories without equal spacing (Low / Medium / High) | Mode, Median, per-option distribution |
Nominal | Unordered categories (favorite color). Single- or multi-select | Mode, per-option shares (multi-select shows "% of testers") |
Binary | A two-way choice (Yes / No) | Mode plus a decisiveness verdict |
Ranking | Testers drag options into order of preference | Top-ranked option, average rank, ranked-first / ranked-last, weighted score |
Classification runs automatically once a test has enough responses. If a question is genuinely ambiguous, BetaTesting leaves it unclassified rather than guessing - you'll simply see the standard chart without a summary.
The headline stats (collapsed view)
At a glance, the collapsed strip answers "what happened?" without any clicking:
Responses: how many testers answered this question.
Average: the mean response. On agreement scales it's shown signed and centered on neutral (see "How Likert averages are scored" below), so
+1.61reads instantly as net-positive.Mode (most common): the single most-chosen answer, with how many testers (or what share) picked it.
Median: the middle answer. For category questions this is shown as a label (e.g. "Strongly Agree"), not a raw number.
Range / Net Promoter Score: shown for numeric questions where they apply.
Show more: the full statistics panel
Click Show more to reveal the complete breakdown. Depending on the question type, this includes:
95% Confidence Interval for Average: the plausible range for the "true" average if you surveyed everyone, not just this sample.
Range: the lowest and highest answers given.
Interquartile Range (IQR): the spread of the middle 50% of responses (25th to 75th percentile). A tighter range means testers clustered around the median.
Standard Deviation: on average, how far each answer sits from the mean. Small = testers agreed; large = answers varied widely.
Variance: the standard deviation squared.
Per-option table: for every answer option: the Count, its Count CI (confidence interval), the Share (%), and its Share CI.
Counts vs. percentages
Use the # / % toggle in the chart's toolbar to switch every bar (and the labels) between raw response counts and percentages. The underlying statistics don't change, only how the distribution is displayed. Counts are best for understanding sample size; percentages are best for comparing across questions or reporting to stakeholders.
Chart overlays: Stats visualization on a chart
In the Response Summary, click a statistic to show the related overlay directly on the chart; click it again to clear it. In the expanded per-option table, hover any column or cell to preview that option's value and margin of error on the chart. Overlays appear in violet so they stand out against the bars.
Click / hover this | You'll see on the chart |
Average | A vertical line at the mean, labeled with its value. |
Confidence Interval for Average | A shaded band around the average line showing its plausible range. |
Interquartile Range | A shaded |
Range | Two markers at the lowest and highest answers given. |
Median / Mode | An outline around the corresponding bar. |
A Count CI / Share CI cell or column header | Each option's value marker plus its confidence band (its margin of error). |
Confidence intervals, explained
A confidence interval answers the question every researcher gets asked: "How sure are we?" Your testers are a sample, so each result is an estimate of how all users would respond. The confidence interval is the range that estimate plausibly falls in.
95% confidence interval explained: If you reran the test with new testers, a range like this would capture the true average about 95% of the time - so it's a safe bet the real value for all users (not just those who responded) sits inside it.
BetaTesting computes two kinds, using methods that hold up at the smaller sample sizes typical of beta tests:
For averages (the mean of a rating/Likert question): a Student's t-interval, which is appropriate when you're estimating a mean from a sample.
For proportions (each option's count and share): a Wilson score interval, which stays well-behaved at small samples and extreme rates (e.g. 0% or 100%) where the simple textbook formula breaks down.
Reading tip: Interval width reflects certainty. Tight intervals come from larger samples and can be taken at face value; wide intervals signal sparse data, so the estimate could shift meaningfully with more responses. Wider intervals are enough to show the general trend, not the precise number.
Choosing your confidence interval level
Confidence intervals default to 95%, the standard for most research. When the panel is expanded, an α menu appears in the top-right of the strip. Use it to switch between:
90% - narrower intervals; useful for quick, exploratory reads.
95% - the default and most widely reported standard.
99% - wider intervals; use when you want extra certainty before a high-stakes call.
Changing the level instantly updates every interval in the panel and any confidence overlay on the chart - no reload, no refetch.
Net Promoter Score (NPS)
For 0–10 "How likely are you to recommend…" questions, the Response Summary surfaces a Net Promoter Score as the headline metric.
NPS = the percentage of promoters (rated 9–10) minus the percentage of detractors (rated 0–6). Passives (7–8) are ignored. The score runs from −100 to +100.
NPS only appears when the question is actually a 0–10 recommendation scale, so it won't be misapplied to other rating questions.
How Likert averages are scored
On agreement scales, the answers are coded symmetrically around a neutral midpoint. For a 5-point scale:
−2 (Strongly Disagree)
−1 (Disagree)
0 (Neutral)
+1 (Agree)
+2 (Strongly Agree).
The chart notes this coding beneath the axis.
Because the scale is centered on zero, the sign of the average is its meaning: a positive average is net agreement, a negative average is net disagreement, and the further from zero, the stronger the consensus. To reinforce this, agreement values are tinted - green above neutral, red below - so a confidence interval that straddles neutral (one green bound, one red) instantly flags a divided result.
Exporting chart statistics
You can export any individual chart to a spreadsheet by clicking the menu icon ellipses to the right side of a chart and choosing Export the data to Excel.
The export includes:
Summary chart data (e.g. count per answer option)
Statistics data (like confidence interval, average, etc)
All the raw answer data (each participants answer)
Quick FAQ
Why don't I see a Response Summary on some questions?
Summaries appear on closed-ended questions once a test has enough responses to classify and summarize. Open-text and file-upload questions don't get statistical summaries, and genuinely ambiguous questions are left unclassified rather than mislabeled.
Do filters affect the statistics?
Yes. The summary and overlays recalculate against whatever response filters are currently applied, so you can compare segments.
Which average should I report: the signed Likert average or the raw scale value?
For agreement scales, the signed average (e.g. +1.61) is the clearest way to communicate net sentiment. The per-option table always shows the underlying counts and shares if you need to report the raw distribution.
My confidence interval looks wide. Is something wrong?
It usually just means there are a small number of responses. Collect more responses to tighten it, or switch the α menu to 90% for a narrower (less conservative) read.