Three Types of Percentages – MeasuringU

It’s common in UX research to encounter percentages:

• • 90% task completion rates
  • 10% reduction in time
  • 250% increase in mean scores
  • −20% Net Promoter Score
  • −10% new UX jobs

Despite their familiarity, their interpretation can be tricky because the same term (“percentage”) is associated with distinctly different concepts. The three key types are:

• Absolute percentages
• Relative percentages
• Net percentages

Three types of percentages … how are they the same, and more importantly, how are they different?

Absolute percentages are expressed as parts of a whole in terms of 100. “Percent” literally means “per one hundred.” For example, 50% means 50 out of every 100, and 75% means 75 out of 100. Percentages can be used on any sample size, so 80% can also come from eight parts out of ten. Examples from UX research might be eight successful task completions out of ten participants or a top box score in which eight of ten respondents selected the most favorable response.

An absolute percentage can also be thought of as the mean of an array of 0s and 1s expressed as a percentage. Any percentage can be expressed as a proportion, and any proportion can be expressed as a percentage (e.g., .271 is the same as 27.1%). Continuing with the example from the previous paragraph, if there were eight successes and two failures for ten participants attempting a task in a usability study, assign a 1 to the successes and a 0 to the failures. This sums to 8. Divided by the number of participants, the mean is 0.8; expressed as a percentage, it’s 80%.

Absolute percentages can never be less than 0% or more than 100%. Of the different types of percentages, these are the easiest to understand because they directly represent a fixed proportion of a whole without requiring comparison to any other reference point. Even so, interpreting absolute percentages as poor or good requires knowledge of the context in which they were collected. For example:

• A 90% task completion rate sounds good compared to a 50% completion rate, but both are unacceptable if the task is for accountants to accurately report income.
• Low percentages can be good. Click-through rates (CTR) for online advertisements (the number of clicks on an ad divided by its number of presentations) are never high, but they don’t have to be high to be worthwhile. Examples of CTRs considered good include 6–7% for Google ads, about 1% for Facebook ads, and about 0.5% for banner ads.

To really understand an absolute percentage, you must know the total sample size because that determines the precision of the estimated percentages. Using the width of the 95% adjusted-Wald binomial confidence interval as a measure of precision, when 80% is computed from 8/10, the lower and upper limits of the interval are 48 and 95% (a width of 47 points). When the ratio is 80/100, the limits are 71–86% (a width of 15 points); for 800/1000, the limits are 77–82% (a width of 5 points). We recommend reporting absolute percentages because they are easier for people to comprehend and compare than fractions, but we also recommend documenting the fractions (or at least the sample size) so readers can tell whether the percentage is precise.

A relative percentage is based on the ratio between two numbers, one of which is a reference point. For example, suppose you have a puppy that weighed 9 pounds one month ago and weighs 9.2 pounds today—an absolute change of .2 pounds. The change expressed as a relative percentage is the new value minus the reference value divided by the reference value. In this example, the relative percentage is (9.2 − 9) / 9 = .2 / 9 ≈ .022 or 2.2%. A weight of 9.2 pounds is an increase of 2.2% relative to 9 pounds. Suppose the puppy had lost .2 pounds instead of gaining it. In that case, the relative percentage would be (8.8 − 9) / 9 = −.2 / 9 ≈ −.022 or −2.2%.

While we don’t do much work with puppies in UX research, the numbers could also be means on a rating scale. For example, a mean of 9.2 on an eleven-point likelihood-to-recommend (LTR) scale collected this month versus a mean of 9 collected last month is also a 2.2% increase.

Unlike absolute percentages, relative percentages can be negative and can be larger than 100% (positive or negative). For example, consider a mouse that weighs 25 grams and an elephant that weighs 2,500 kilograms. Using the mouse as the reference point, the increase in weight from mouse to elephant is (2,500,000 − 25) / 25 = 99,999 or an increase of 9,999,900%. An elephant weighs almost 10 million percent more than a mouse.

For a more UX-based example, if a task takes 2.5 minutes to complete on your company’s website but 1 minute to complete on your competitor’s website, the task takes 150% (1.5 minutes) longer to complete on your website.

Or consider that the average predicted salary of a UX professional with ten years of experience living in Australia and working at a large company is $64,035 USD. The same UX professional living in California would be predicted to make $197,696. That CA-based pro would have a nominal salary that’s 209% higher than the Australia-based one. Before you decide to leave the southern hemisphere to move to the Golden State, keep in mind that the bulk of that difference is due to a higher cost of living.

To really understand a relative percentage, it’s important to know the reference point and, ideally, all the values that were used to compute the relative percentage including the sample sizes. Otherwise, small absolute values can be represented in a way that exaggerates their apparent importance.

For example, suppose you’re 39 years old and you see an advertisement for a new drug that, although it has the usual list of scary side effects, reduces the probability of having a heart attack in your 40s by 50%. This sounds great, but you don’t know its actual effect unless you know the reference probability that, in the U.S., is about 0.1% (97.6 heart attacks per 100,000 people in their 40s). That means that the absolute reduction in your probability of having a heart attack in your 40s would be about 0.05%.

When Jim worked on speech dictation products in the 1990s, a major competitive factor was recognition accuracy. There were different ways to express improvements in the percentage of correctly recognized words. If the reference recognition rate was 95% and a change in a recognition model increased that to 96%, you could say that the improvement in accuracy was 1/95 or about 1.1%. Or you could take a different approach and focus on error rates. If the previous error rate was 5% (95% accuracy) and the new error rate was 4% (96% accuracy), then the percentage reduction in error rate would be −1 / 5 = −20%. Both approaches had a 1% change, but you can imagine which relative percentage speech researchers preferred to report to their managers and speech product marketers preferred to put in their product specifications.

Relative percentages provide another way to quantify changes in values but beware of relative percentages that are presented without their reference values.

A net percentage is the difference between two absolute percentages. Because absolute percentages range from 0 to 100%, net percentages can range from −100 to +100%.

Probably the best-known net percentage in UX is the Net Promoter Score (NPS), which is the percentage of respondents to a 0–10-point LTR scale classified as Promoters (selected 9 or 10) minus the percentage classified as Detractors (selected 0 to 6). (Yes, the NPS can be expressed as a percentage, just like any other net percentage.)

One of the reasons for the popularity of the NPS in organizations is that most executives can understand what scores like an NPS of 50% (really good) or −10% (bad) mean. However, even for the NPS, very small net percentages are ambiguous without knowledge of the absolute percentages. For example, if the percentages of Promoters and Detractors are both 50%, the NPS is 0%; but if all respondents are Passives, the NPS is also 0%—the same value from very different response patterns.

Other net percentages are hard to interpret in isolation. For a single net percentage computed by subtracting the bottom-box percentage from the top-box percentage, is 50% good? That’s an open research question, but it clearly depends on factors such as the number of response options in the rating scale, the number of response options included in the box, item wording, and the research context.

As an example of interpretable net percentages, we analyzed an item from the 2024 UXPA salary survey for which respondents indicated whether their organization added staff, lost staff, or stayed the same. In 2024, the net percentage (35% added staff minus 35% lost staff) was 0%. That doesn’t sound great, but how could you know? Fortunately, that item was included in previous UXPA salary surveys, as shown in Figure 1, establishing context that the net 0% in 2024 was definitely not great.

Figure 1: Net UX jobs (% added minus % lost) by year.

Echoing our advice about relative percentages, beware of net percentages that are presented without documentation of the absolute percentages. In the article where this figure was originally published, we also provided a table of all the absolute percentages and sample sizes.

Table 1 compares the different types of percentages discussed in this article. All have their role in UX measurement, but all are a bit different, so be sure you know the type of percentage you’re trying to interpret. Keep in mind that all are difficult to interpret without some comparative context.

Table 1: Summary of the properties of three different types of percentages.