Menu Search
Jump to the content X X
Smashing Conf Barcelona

You know, we use ad-blockers as well. We gotta keep those servers running though. Did you know that we publish useful books and run friendly conferences — crafted for pros like yourself? E.g. our upcoming SmashingConf Barcelona, dedicated to smart front-end techniques and design patterns.

Why Perceived Performance Matters, Part 1: The Perception Of Time

Those of us who consider ourselves developers, including me, are very task-oriented. We like to be guided towards optimal results, and we find ourselves uncomfortable when there is no clear path to follow. That is why we all want to know how to do things; we like step-by-step tutorials and how-tos. However, such guidelines are based on certain theories, deep knowledge and experience.

Further reading on Smashing: Link

For this reason, I will not provide you, the reader, with a structured answer to the question of how to make a website faster. Instead, I aim to provide you with the reasons and theories for why things function in certain way. I will use examples that are observable in the offline world and, using principles of psychology, research and analysis in psychophysics and neuroscience, I will try to answer some “Why?” questions, like:

  • Why do time and performance matter?
  • Why don’t we like to wait?
  • Why does faster not always mean better in the online world?
Deconstructing Performance6

“Alice: How long is forever? White Rabbit: Sometimes, just one second.” — Lewis Carroll. “Alice’s Adventures in Wonderland.”

In addition to these, we will cover psychological aspects of some practical cases, like performance optimization of an existing project, how to deal with the better performance of a competitor’s website and how to make users barely notice any waiting for your services. Hopefully, after understanding these “Why?” basics, you will be ready to look outside of some structured “box” the next time you are involved in an optimization process and will make your very own conscious path towards an optimal result.

Every how-to, tutorial, guideline has a simple question "Why?" at its roots7

Every how-to, tutorial and guideline has a simple “Why?” at its root. (View large version8)

Basic Concepts Link

“Well, I never heard it before, but it sounds uncommon nonsense,” says the Mock Turtle in Alice’s Adventures in Wonderland. So, first things first. Let us settle on basic terminology and principles. In the following paragraphs, I will define the main concepts that you will find throughout the article.

Time can be analyzed from two different points: objective and psychological. When we talk about time that can be measured with a stopwatch, we’re talking about objective time or clock time. Objective time, though, is usually different from how users perceive time while waiting for or interacting with a website, app, etc. When we talk about the user’s perception of time, we mean psychological time or brain time. This time is of an interest to psychologists, neuroscientists and odd individuals like me.

Psychological Time in our brain usually differs from Objective Time on the clock9

Psychological time in our brain usually differs from objective time on the clock. (View large version10)

Performance optimization is a process of improving the delivery speed of services, feedback or any other type of response action in order to meet a user’s expectation.

We will call the form of communication that occurs when users have to wait for a response from a system a human-to-computer style. On the other hand, when a system barely has any delay in response, we will call that a human-to-human style, similar to the regular conversations we have in real life.

A performance budget, as Tim Kadlec11 defines it12, “is just what it sounds like: you set a budget on your page and do not allow the page to exceed that.” For the purpose of this article, we will consider the performance budget to be directly connected to time — for example, loading a page, responding to a user’s input, etc.

Armed with these definitions, let’s go down the rabbit hole.

Part 1: Objective Time Link

Now, here, you see, it takes all the running you can do, to keep in the same place. If you want to get somewhere else, you must run at least twice as fast as that!

The Red Queen in Lewis Carroll’s Through the Looking-Glass

As we defined it earlier, objective time is time that can be measured with a clock. In this first article in our series, we will discuss how to use this time more efficiently for our projects.

“Remember That Time Is Money” Link

This phrase, credited to Benjamin Franklin in his Advice to a Young Tradesman, still proves true in our contemporary world. According to a recent study13, it takes only 3 seconds for a visitor to abandon a website. For every 1 second of improvement, Walmart experienced an increase in conversions of up to a 2% on its website. When auto-parts retailer AutoAnything cut loading times in half, it experienced a 13% increase in sales. Users value their time, and a human-to-human style of communication is increasingly expected.

A human-to-human style of communication is increasingly expected from online systems14

A human-to-human style of communication is increasingly expected from online systems. (View large version15)

New ways of communication require a new way of thinking in the design and development of online systems. In the web industry, we have been talking about speed and performance for quite a while. We’ve been emphasizing the importance of fast communication, and we’re always trying to reach certain goals in seconds, kilobytes, frames per second and so on. As a result, we know how to make our websites fast using technology. Furthermore, we keep getting advice from leading developers and companies about preferred times in user-to-website interactions. In November 2014, Dimitri Glazkov, Jochen Eisinger and Chris Harrelson, in their “State of Blink16” keynote about the Blink rendering engine, suggested a platform success model2117 based on interaction times.

Platform success model18

Platform success model from Google19 (View large version20)

After quite a few years of the “2-second loading time” recommendation, the platform success model2117 technically sets the limit on page-loading time to 1 second. Usually, the average developer or project manager is satisfied with an explanation like, “Longer times affect rankings in search results.” But what recommendations are these explanations based on? How much freedom do we have in setting certain time-based goals or a performance budget? And, what is even more interesting than simple numbers, can we be creative with these durations, playing with them to deliver better experiences to our users? These are the main questions we will cover in this article.

Choosing A Performance Budget Link

As defined earlier, a performance budget is a limit that we set on certain time-related operations of an online system and that we do not allow the system to exceed. Usually, human-to-human conversations involve quite short time intervals between a question being asked and a response being given by the interlocutor. In psychology, we should keep four important spans of time in mind when dealing with short time durations:

Psychological time durations22

Psychological time durations (View large version23)
  • 0.1 to 0.2 seconds
    Research24 points to this interval as the range of the maximum acceptable response time to simulate instantaneous behavior — an interval within which users barely, if at all, notice a delay.125
  • 0.5 to 1 seconds
    This is the maximum response time for immediate behavior. This interval is usually the response time from an interlocutor in a human-to-human conversation.226 Delays within this interval are noticeable but easily tolerated by most users. During this time, the user must receive an indication that their command (such as their click of a button or link) has been accepted, assuming that the signal was not already sent during the previous interval.
  • 2 to 5 seconds
    Psychologist Mihaly Csikszentmihalyi27 defines a flow or optimal experience as a state when people experience concentration, absolute absorption in an activity and deep enjoyment.328
  • 5 to 10 seconds
    According to the National Center for Biotechnology Information at the US National Library of Medicine29, the average attention span of a human being dropped from 12 seconds in 2000 to 8.25 seconds in 2015. Guess what? That is 1 second less than the attention span of a goldfish! As a simplification — and to emphasize our superiority to the goldfish — we consider 10 seconds as the absolute maximum time of a user’s attention span.430 The user would still be focused on their task but would become easily distracted. This is the time for the system to engage the user in the process; if this is not done, then the user will most probably be lost forever.

Now, if we apply these time definitions to the aforementioned platform success model31 suggested by the Blink team, we can translate its emphasis on a 1-second loading time and a 0.1-second response time into more human-to-human language:

  • The loading page should happen immediately.
  • Users should get feedback from a given action instantly.
Load pages immediately, and give feedback on any given action instantly.32

Load pages immediately, and give feedback on a given action instantly. (View large version33)

Humans have inner stopwatches that have been tuned well since our cave-dwelling ancestors. Hence, you can safely rely on the time spans above when choosing a performance budget for your next web project. But time on its own is useless unless put into some context. Consider an example. What do you think about in response to the statement, “A process will take 10 seconds to complete.” Take your time.

A process will take 10 seconds to complete34

“A process will take 10 seconds to complete.” (View large version35)

Done? I’ll bet the first thing you thought was, “What process?” Your brain iterated over possible options based on your life experience, current lines of thinking, mood and so on. Then, once your brain settled on a dominant guess, it evaluated your expectations about that process’ duration against the declared 10 seconds.

That being said, most often we don’t pick a random time duration on its own, but rather choose time in a certain context. First, the context could be set by our own constraints, such as a need for performance optimization in order to meet users’ expectation of a human-to-human style of communication. Secondly, the context might also be set by competitors; in this case, we’re dealing with a chase-the-leader scenario to reach the performance level set by others. We will look at both scenarios below.

The Need for Performance Optimization: The 20% Rule Link

Here is a situation that most of us can relate to to some extent. Your customer says that the search function of your website is slow. Your measurements show an average of 5 seconds to render search results. After employing some optimization techniques, you bring it down to 4.5 seconds. Happily (well, as happily as one could possibly be with a 4.5-second rendering time), you send an email to the customer with this information. All you get in return is, “Doesn’t look any faster to me.” That usually hurts. Let’s try to understand why the customer did not notice any difference.

Sometimes half of a second is just not enough36

Users will not necessarily notice your optimizations, so it is too early for a party. (View large version37)

In 1834, one of the founders of experimental psychology, Ernst Heinrich Weber38, postulated a law that defines difference threshold or a just noticeable difference (JND) as the minimum difference in stimulation (weight, light, etc.) that a person can detect most of the time. Later, Weber’s student Gustav Theodor Fechner39 applied the law to the measurement of sensations, setting the basis for the science of psychophysics40. This work by Weber and Fechner is known to us as the Weber-Fechner Law. The law is still regarded by many scientists as arguably the most important tool in understanding perception.

Time is not an exception to the Weber-Fechner Law. Practical experiments in psychophysics show that time intervals are prone to a JND of between 7% to 18% on average for shorter periods (with a duration of less than 30 seconds). Based on this, a good rule of thumb is to simplify the Weber-Fechner Law into a 20% rule. That is, in order for users to barely see a difference in time duration, it has to be changed by a minimum of 20%.

20% Rule

The Weber-Fechner Law can be simplified into a 20% rule for short time durations.

Let’s go back to our example of a client noticing no difference in the rendering of search results after our optimizations. Originally, the search results were returned in 5 seconds. Using the 20% rule, we can tell that, for the user to even notice the difference, the new search results should be at least 1 second faster:

5 seconds × 0.2 = 1 second

That is why the client simply did not notice the 0.5-second improvement.

We’re talking about performance optimization, but this technique works in the opposite direction. If, for example, you are developing a feature that slows down your web page, you could apply the 20% rule to determine whether the performance decrease will be noticed by users at all. Allowing our code to be a bit slower without harming the user experience is called regression allowance.

Note: When we talk about 20%, we are talking about a “just noticeable” difference. “Noticeable” does not mean “meaningful.” In order for users to appreciate your performance optimization, you should go well beyond this threshold.

But what do you do when a competitor comes to market with significantly better performance for a comparable feature? What if a 20% regression allowance relative to the competitor’s time is not even technically approachable for us? In this case, we should at least aim for what G. Moore, in his book Dealing with Darwin41, calls neutralization.

Neutralization, Or Chasing The Leader Link

Time neutralization occurs when the time difference between two services is noticeable but does not influence the user’s preference of one service over another. In this case, reliability, usability and other quality-related factors of the service play a much more significant role. In other words, time neutralization is a good position to be in when you have better services than competitors.

Chasing the leader might be tough42

What to do when your competitors deliver the same functionality but much more quickly? (View large version43)

Again, getting back to our example of search results taking 5 seconds to appear, assume that a competitor comes to the market with a very similar search service: the same functionality, the same results and the same feel. The problem is that their search results are revealed in 2 seconds, not 5. Applying the 20% rule to our 5 seconds would not make sense in this scenario. We’re not competing with our own time; rather, we have to match the competitor’s. We can make a couple of logical assumptions here:

  1. We cannot decrease the search response time to 2 seconds. (If it were that easy, we probably would have done it already.)
  2. If 2 seconds is out of reach, the next best solution would be to use the 20% rule of regression allowance relative to the competitor’s time: 2 seconds + 20% = 2.4 seconds.

At 2.4 seconds, users will not notice any difference between our search results and the competitor’s. But if we cannot achieve 2 seconds, then 2.4 seconds is probably also out of reach.

We can match neither 2 seconds directly, nor a 20% regression allowance of competitors’ time44

We can match neither the 2 seconds directly nor a 20% regression allowance of the competitor’s time. (View large version45)

When comparing two durations (2 and 5 seconds, in our case), we’ll find a “magical” psychological threshold between them. Time durations longer than this threshold will be perceived by the user as being closer to 5 seconds. Time durations shorter than that threshold will be perceived as being closer to 2 seconds. Surprisingly, studies in animal timing46, particularly those conducted by R. Church, M. MacInnis, P. Guilhardi and others at Brown University, show that this threshold is not simply the average of the two durations. This threshold proved to be predictable and is found at the geometric mean, instead of an arithmetical one. Research with human subjects confirm the same finding.

From mathematics, we might remember that the geometric bisection between two numbers can be found using the following formula:

√(A × B)

Applying this formula to our numbers, we get:

√(2 × 5) ≈ 3.2 seconds
The threshold of the difference between two times can be found through the geometric mean of the two47

Geometric bisection as an illustration of neutralization (View large version48)

For our 2- and 5-second results, 3.2 seconds is the threshold of neutralization. At this point, users will notice a difference, but the difference will be perceived as being unimportant to their choice of service. Factors such as the quality of the service and reliability will play a much more important role, but managing these is beyond the scope of this article.

With this, we finish our basic analysis of objective time. By now, we should have a good understanding of where some widely adopted industry standards49, such as page-loading time and the response time for a system, come from. We also have a guideline for setting a performance budget50, and we understand how to deal with time when we have to improve the performance of a website51 or match that of a competitor’s52.

But as we have mentioned earlier, another kind of time usually matters much more to users than the kind measured with a stopwatch. Ladies and gentlemen, let’s meet in part 2 and talk about the psychological time.

To be continued…53

Endnotes Link

1 This is the reaction time of the average person to a simple stimulus, such as catching a falling pen or jerking one’s hand away from a hot cup. This is probably the most important time we should keep in mind.

2 According to G.A. Miller, most adults can store five to nine simple items in their short-term memory54. Research55 — confirmed by a number of experiments, in particular the one by M. Greene56 and A. Oliva57 of MIT — shows that “the processing of complex natural images and visual object recognition” takes no more than a tenth of a second. This means that short-term memory processing of five to nine items such as this should take about 0.5 to 0.9 seconds. Hence, 1 second is considered to be the maximum time for the user’s flow of thought to stay uninterrupted.

3 While optimal experience time fluctuates according to subjective parameters, for most tasks that the average user faces on the web, the time of concentration falls between 2 to 5 seconds. This is why, for some years, we operated with 2 seconds as the optimal page-loading time.

4 Of course, the state of focused attention in general spans far beyond 10 seconds and can reach 20 to 30 minutes. (This long-term attention is called “selective sustained attention.”) However, first, this time span is much more exposed to fluctuation due to different uncontrolled circumstances; secondly, not many website owners out there can brag about their website being actively used by the average user for more than a couple of minutes.

(ah, ml, al)

Footnotes Link

  1. 1
  2. 2
  3. 3
  4. 4
  5. 5
  6. 6
  7. 7
  8. 8
  9. 9
  10. 10
  11. 11
  12. 12
  13. 13
  14. 14
  15. 15
  16. 16
  17. 17
  18. 18
  19. 19
  20. 20
  21. 21
  22. 22
  23. 23
  24. 24
  25. 25 #endnote1
  26. 26 #endnote2
  27. 27
  28. 28 #endnote3
  29. 29
  30. 30 #endnote4
  31. 31 #success-model
  32. 32
  33. 33
  34. 34
  35. 35
  36. 36
  37. 37
  38. 38
  39. 39
  40. 40
  41. 41
  42. 42
  43. 43
  44. 44
  45. 45
  46. 46
  47. 47
  48. 48
  49. 49 #success-model
  50. 50 #performance-budget
  51. 51 #20-percent
  52. 52 #neutralization
  53. 53
  54. 54
  55. 55
  56. 56
  57. 57

↑ Back to top Tweet itShare on Facebook

Denys is a fronted developer & public speaker living and working in Norway. Being 2-in-1: an art school graduate and an engineer, Denys is passionate about psychology, physics, history, drawing. In his day-to-day job he enjoys getting to the heart of the matter of things and processes. Originally on “CSS side”, for the last years Denys has been building javascript applications, still breaking CSS, abusing HTML and working with performance optimisations of pretty much all aspects of the fronted toolset.

  1. 1

    Wonderful article. Performance is too often overlooked and it shouldn’t be. The more designers and developers take into account the importance of performance, the greater the experience on the web that can be had for all. Not to mention the increase of comprehension of the content being delivered. Well done Denys, well done.

  2. 2

    Very nice article. Very interesting the 20% threshold rule.

  3. 3

    I think I have information that can be helpful here. Once psychologists wanted to study how long a “moment” lasts. They came up with an idea to check how long it takes to welcome a person. Turns out, shaking hand, greeting or hugging – last about 3 seconds (I experimented on myself by kissing – confirmed :). And so it can be concluded, that for our brain a moment lasts 3 seconds.

    Unfortunately I can’t give you source at the moment, maybe someone else can.

    It always helped me to understand the “magical” psychological threshold that you’re writing about and why 5s/3s difference will be perceived as being so important.

    Thanks for the great article!

  4. 5

    Great article. The referenced use cases from walmart and others on the impact of response times on revenue are frequently used. Maybe because of this people find them exagerated. I was fortunate enough to interview and host Gopal Brugalette from Nordstrom, a performance engineer recently for some Dynatrace meet ups in Australia. He found a 2.5 second response time was optimum. A 0.5 second increase in response time reduced conversions by 11%. He graphed and presented the data, and as I understand they make feature and function decisions based on end user performance impact on conversions. It was awesome to validate theory with real data. Thanks for the post, and look forward to part 2.

    • 6

      Thanks Dave! Would love to see the research by Nordstrom – seems like they might have some exciting analysis.

  5. 7

    great article to see it.

  6. 8

    Thank you all very much for the comments and valuable advises.

  7. 9

    Nice article – really enjoyed reading it!

    One thing that’s really important is understanding the psychological effect of time as well. Five minutes in pain is a lot different than five minutes of lying on a beach with a martini. Here’s a great article I’d like to share on the matter:

    Our success in life can also depend on our understanding of time: In business, with our clients and with our personal relationships.

    • 10

      Thanks for the insight. Though, stay tuned for the Part 2 of the article and I will elaborate much more on the psychological aspects and whether we should always aim for speed or not ;) Should be interesting.

  8. 11

    This is a great article and something that’s close to our hearts at We have lots of meaningful data around the improvement of page load times and what that means from a transactional perspective for eCommerce companies that are real….not just the big guys. The Walmart, Amazon and Google case studies have been kicking around for years, but if I get 20,000 visitors to my site a month, what does an increase in speed look like for me and how much do I need to spend to achieve 20%?

    What I really like about this article is the psychological aspects of time, speed and what’s perceived to be fast. All too often we hear about edge cases that are based on isolated incidents (“feels fast to me” or “CEO said the site was slow so we need to do something about it”). The best case to manage these expectations is using Real User Monitoring data and find out what’s going on in the browser for 95% (95th percentile) of your users or your “real browsing / buying traffic.” They are the folk that count.

    • 12

      Well, this is all correct of course. Though I am not sure why Wallmart, Google or Amazon do not count as “real browsing / buying traffic.” for you. Or do I miss the point here? Anyway, of course, to gain 20% performance improvement using technical means might be plain impossible. And this is what I am going to talk about in Part 2 of the article. Stay tuned!

  9. 13

    kalyankumar Bethi

    October 5, 2015 7:25 am

    Very nice and a wonderful article.

  10. 14

    George Gooding

    October 5, 2015 7:59 am

    Actually, the human vs. gold fish attention span factoid is just that, a factoid. It’s been debunked a bunch of times. The 8 second figure is how long humans spend figuring out whether a web page they’re looking at is worth their time or not, not a measure of their attention span. The gold fish figure seems to be an invention.

    You should cross this out in the article and for once explain to people that this claim is bogus.

    A full run down of why it’s a made up statistic can be found a bunch of places, I found a good one here:

    • 15

      Thanks for your comment, George! That’s a nice one and I don’t think that this particular information about the gold fish vs. human attention span is 100% correct and has been studied properly. Though I see widely-accepted information based on some research and surveys. We can argue forever proving that what the surveys actually show is right or wrong but nevertheless there is a bunch of articles on this topic.

      There is one problem with the article you mention: the only link proving that “It’s been debunked a bunch of times” is a link to an article in The Guardian, that in it’s turn provides a text that is hard to confirm: the only link it has is a link to a non-existing “research”. Nevertheless, I searched the publications by Proff. Philip Gee of Plymouth University, mentioned in the article and have found the only research he and the goldfish had been involved into together (and I guess, this is the research that The Guardian is writing about) — “Temporal discrimination learning of operant feeding in goldfish (Carassius auratus)”, J Exp Anal Behav. 1994 Jul; [].

      The problem is that “Temporal discrimination learning of operant feeding” and “attention span” are a bit different things. It’s training ability vs. attention (used in training in particular). So, the only real research your article indirectly refers to mentions different aspects of the goldfish’s brain capabilities.

      So I will leave the right to decide whether this information is right or wrong to the readers, considering these comments. But this piece of information is very far from being essential to the point of the article ;) Hope you find the rest of the article credible and worth reading.

    • 16

      … and concerning the human attention span it all depends on how you define “attention”. It important to distinguish transient attention and focused attention. In the article I am talking about the former and in a footnote #4 I mention that “focused attention in general spans far beyond 10 seconds and can reach 20 to 30 minutes”.

      So I think we agree that there is just a misunderstanding here about different types of attention.

  11. 17

    Marc Zovighian

    October 11, 2015 6:18 pm

    Great great article.

    When will part 2 be available?

    • 18

      Denys Mishunov

      October 12, 2015 9:20 am

      Glad you liked it, Marc. Part 2 is in the review process and, hopefully, you can expect it rather soon ;)

  12. 19

    Great write up on a topic that’s fascinated me for a long time!

    There is one small fault however. The 20% rule is only accurate for very short intervals. As you go beyond 1s, the JND is more like 1/8 (or 13%).
    This part of research has just not reached every applied scientist yet. You can find the original research cited in a paper on the experience of time we presented at a design conference:

    Lassi A. Liikkanen and Paula Gómez Gómez. 2013. Designing interactive systems for the experience of time. In Proceedings of the 6th International Conference on Designing Pleasurable Products and Interfaces (DPPI ’13). ACM, New York, NY, USA, 146-155. DOI=10.1145/2513506.2513522

    Funny how this evokes a similar comment as did the RAIL story on Smashing from the other week.

    • 20

      Denys Mishunov

      October 12, 2015 9:17 am

      Thank you for your comment, Al! The research is quite interesting and is even more valuable due to the References part. You are correct about 13%; though, even this number is not accurate enough. You might have overlooked the following in the article: “time intervals are prone to a JND of between 7% to 18% on average for shorter periods” but this supports the research you mention.

      The 20% rule is not a law or research-based value. This is exactly what is said in the article: “a good rule of thumb to simplify the Weber-Fechner Law” that levels fluctuations of practical results (those research-based 7% to 18%). This is simply a more convenient tool to use in our day-to-day jobs.

      Thank you very much for the valuable link.

  13. 21

    Very stimulating article – much more in-depth than most generic guest articles on performance. More compelling reason for me to come back to Smashing (rather than free icons ;-)) – to seek out meaningful in-depth articles.

    Time is tough to measure because as you hint, it is a concept made up by our brain to measure change. That’s why strategically placed animations can make an experience seem faster. So can interim feedback.

    There is also external factors in this perception like user motivation, brand goodwill etc. which I hope you will touch on later articles.

    • 22

      Denys Mishunov

      October 18, 2015 10:17 am

      Hi savant,

      Thank you very much for your comment. Glad you find it useful and take the article as an intrigue for the next chapters. Your comment about the animations is totally correct; being visually appealing (well, at least that is how they should be done) animations are also a valid tool for managing user’s perception of time. Great to see that you get the idea even after this first part. So I hope you will enjoy the next parts of the article where we dive more into psychological aspects of the performance. Stay tuned and feel free to comment on those.

  14. 23

    Tommy O'Keefe

    November 3, 2015 7:31 pm

    Any idea when we can expect part 2 of this series? I’ve been looking forward to it!

    • 24

      Denys Mishunov

      November 4, 2015 1:19 pm

      Hello Tommy,

      You can’t imagine how I am looking forward to it myself ;)

      As far as I am aware, Part 2 is at it’s last stage before being published. Part 3 (99.9% finished) will be sent for the review right after that. But there are a lot of other great articles waiting to be published, so the process takes some time. Thank you very much for the patience.


↑ Back to top