This chart.
This chart has bothered me since I first saw it 16 years ago. It’s fascinating and the data is pretty accurate, but the chart title leads you to believe something that is only partly true: that the rate of adoption of new things gets faster and faster and so, implicitly, that this is an exogenous feature of technology.1
The Harvard Business Review used a version of the chart above in an article titled “The Pace of Technology Adoption is Speeding Up”. They made explicit the notion that things aren’t just faster, that they’re speeding up:
“Many people suggest that rates of new product introduction and adoption are speeding up, but is it really, across the board? The answer seems to be yes….It took decades for the telephone to reach 50% of households, beginning before 1900. It took five years or less for cellphones to accomplish the same penetration in 1990. As you can see from the chart, innovations introduced more recently are being adopted more quickly.”2
First, it’s never okay to say “look at this crazy chart: can’t you just see that it proves what I want it to prove?” Second, comparing the adoption of landlines and cellphones is not comparing apples and oranges, and comparing apples and oranges is what you would have to do to say that technology adoption is speeding up generally. (And third, this chart doesn’t show technology adoption, it shows product adoption. I know, this is nitpicking, but since my next post is about the pace of technology adoption, I need to point it out.)
But it feels right, doesn’t it?
Well, #?@* your feelings. Below is a chart of two technologies being adopted, almost eight decades apart.3 It doesn’t show any change at all in the rate of adoption of a fundamental technology.
What’s really going on here?
The rate of adoption of innovations is pretty well studied, and there are many factors that influence it. The book to read is Everett Rogers’ Diffusion of Innovations. Most of this post is derived from Rogers’ work.4 I’ll go through the main factors Rogers says influence the rate of adoption, illustrating each with an example. I tried to think of examples that had one factor as the dominating factor, but every factor comes into play to some degree in each example.
1. Relative Advantage
The chart below shows the percent of US households that owned a refrigerator and stand-alone freezer. The rates of adoption are the slopes of the lines. You can see that the refrigerator was adopted faster than the freezer. This was primarily because a refrigerator is more useful to most people than a freezer. It has a relative advantage.
Relative advantage is usually the biggest driver of the rate of adoption, as you might imagine: how much better is this product than the alternative? “Relative advantage is a ratio of the expected benefits and the costs of adoption of an innovation. Subdimensions…include economic profitability, low initial cost, a decrease in discomfort, social prestige, a savings of time and effort, and immediacy of reward.”5 Which of these is most important depends on the innovation and the adopter.
It’s obvious why relative advantage dictates who adopts the innovation, but it also influences the rate—how quickly they adopt it—in a couple of ways. First, people go through a process to decide if they will be better off if they adopt an innovation. This process involves factors like relative advantage, but each individual needs a certain amount of clarity about how strong the factor is before they can decide whether and when to adopt the product. I’m sure, in your own life, you’ve been presented with an innovation and asked yourself “is this really better, or not?” People make claims, people make counter-claims, it’s hard to find objective advice, and it’s especially hard to figure out if someone else’s advice applies to you. There’s just a lot of uncertainty, and this is at the crux of the adoption decision.
One way people deal with this is to look at the credible claims about how much relative advantage the product has. If the relative advantage is very high, people will adopt the innovation even without certainty because there is enough of a margin for error. If the new things looks like it is ten times better than the existing thing, people will become convinced to adopt it more quickly than if it seems to be 10% better.
As an aside, part of relative advantage has been found to be how soon you get the advantage after adoption. This is the “painkiller vs vitamin” distinction entrepreneurship pundits always talk about. The idea of relative advantage is a more thoughtful way to think about this.
Second, things change over time. Products get cheaper, they get better, and the advantage becomes more objective. In the 1970s, when cellphones cost $3000 and people said things like “Why would I want people to be able to call me when I’m out?”, the relative advantage was both less (because the cost was higher) and less certain (because the benefit was unclear.) But phones came down in price and people who used them could vouch to their friends for their usefulness. Both of these things sped up the rate of adoption.
2. Compatibility
That leads us to the next example, cellphones and landline telephones. Cellphones were more expensive than landline telephones and seemed less relatively advantageous in the context of their introduction: having a cellphone over having a landline is less of a benefit than having a telephone over having nothing. But cellphones were adopted much more rapidly than landline telephones. The reason, again, is pretty obvious: in the early days of the landline, when fewer people had them, there were fewer people to call. Telephones are only useful when you have someone to call. In addition, it wasn’t yet an accepted part of everyday life to call your mother every Sunday. Phones were less compatible with people’s lives. “An innovation can be compatible or incompatible with (1) sociocultural values and beliefs, (2) previously introduced ideas, and/or (3) client needs for the innovation.”6
Compare the compatibility of the cellphone versus the compatibility of the landline. Cellphones fit with many already-existing sociocultural values: keeping in frequent touch with friends and family, exchanging important information via voice, the interrupt-driven nature of calling someone, etc. Landlines did not. Cellphones were also compatible with the previously introduced telephone: when you got a cellphone, you could call anyone, not just other people with cellphones. Cellphones were much more compatible and so they were adopted more quickly.
Innovations must also be compatible with users’ needs, or at least what users perceive their needs to be. Cellphones, again, had the advantage here. People knew they needed telephones in the 1990s; that need was less clear in the early 1900s.
Generally, people judge innovations based on their experiences with previous technologies. Innovations that don’t fit their mental models are adopted more slowly. This again, is just an uncertainty thing: will this innovation improve my life? If it’s hard to imagine, people will hesitate. If it’s easy to see, people will be more likely to adopt. Of course, if compatibility is too high, it slows adoption because it no longer seems very innovative. The flip-side is that if one of the virtues of an innovation is novelty, that innovation may be adopted more rapidly because it is incompatible. Art, for instance.7 But generally, “The more radical and disruptive an innovation and the less its compatibility with existing practice, the slower its rate of adoption.”8
3, 4, 5. Complexity, trialability, observability
An especially pointed refutation of the idea that innovations spread more rapidly is the personal computer. Above, I compare its adoption to the other two important media technologies of the 20th century, the radio and television. You can see that the time for the PC to go from 10% adoption to 25% adoption, and time for it to go from 25% adoption to 75% adoption are both substantially longer than for radio or television.
| 10% to 25% | 25% to 75% | |
|---|---|---|
| Radio | 6 years | 6 years |
| TV | 4 years | 5 years |
| PC | 9 years | 15 years |
It was glamorous to buy a radio or TV when they were introduced, cart it home, deploy the antenna, and listen to/watch shows. It was intimidating to bring home a personal computer in 1981 and try to figure out what to do with it. PCs are just more complex. This, as well as trialability and observability, may be less important than the first two, but they still have an impact. Trialability is whether you can test out an innovation without spending a lot of time or money. Observability is the degree to which you can see other people benefiting from the innovation. If you can try something, you will be quicker to adopt it because the trial reduces your uncertainty about whether the innovation would benefit you. Likewise, if you can see others benefiting from the innovation, you will more easily be able to imagine how it might benefit you.
Think about an innovation that spread extremely rapidly and you can see these in action. Take ChatGPT. I read somewhere that LLMs spread more quickly than any other product in history. If by adoption we mean people trying the product, it was certainly an overnight sensation. Here’s how Rogers’ factors would explain it:
- Relative advantage: the product seemed better than other ways of generating meaningful text, and it was undoubtedly cheaper, being free to try.
- Compatibility: Typing text into a box and expecting something useful to be returned is something we have all become used to, so ChatGPT had high compatibility.
- Complexity: I assume one reason OpenAI released the LLM as a chatbot is because we easily understand how to use chat to communicate.
- Trialability: It was easy to try the product and there was no downside to doing so. There was no onerous signup, you didn’t have to give a credit card, etc.
- Observability: There was no media in which you could avoid reading about someone experimenting with LLMs. Unless you were in a cave, you observed others using Chat GPT.
ChatGPT scores high on each of these metrics, explaining why it was adopted so quickly.
These five factors—relative advantage, compatibility, complexity, trialability, and observability—are all attributes of the product, and Rogers thinks they explain 49-87% of the rate of adoption.9 There are non-product variables that explain the rest.
6. Who is the adoption decision maker?
Who is deciding to adopt the innovation affects adoption rate. The quickest adoption is when an authority mandates it. This may be a government requiring seat belts in cars or your corporate overlords demanding you use Slack. The least quick is when decisions must be made collectively. “The more persons involved in making an innovation decision, the slower the rate of adoption.”10 In between are optional decisions made by individuals. These tend to be faster than collective decisions and slower than authority decisions.
7. How does news of the innovation spread?
The faster news of the innovation is spread, the faster it can be adopted. The first factor here is the communications channels used. Showing a friend your new car is slower than posting on social media about the new Taylor Swift album. The second is the nature of the social system the adopters are in. If it’s highly connected, innovations spread faster. If it is open to new ideas, they are adopted more quickly, etc. The last variable is how much effort is being put into promoting the innovation. An innovation backed by a massive advertising campaign will spread more quickly than one that is not, all else being equal.
The general speed of adoption
Faster information flows argue that innovations should be adopted more quickly than in the past, supporting the thesis of The Chart, but let’s dig a little deeper.
First, yes: the general speed of communication in our society has increased almost exponentially in the last two hundred years. We are also a far more socially interconnected society than we have been in the past. But the speed of communication is plateauing. Communications channels are now effectively instantaneous. Similarly, the density of social interconnectedness seems to be at its human limits (though this might be my own shortsightedness showing). To the extent this has sped up adoption, it will not speed up adoption more. As usual, we noticed a trend at its tail end. Those who make a singularity argument about speed of adoption are wrong.
In any case, the slow increase in adoption speed over the last century or two is swamped in any given fifty year period by other factors. Innovations tend to come in waves, building off each other. The taming of electricity and understanding of electromagnetism led to electric lights and electric motors and electronics, among other things, and these led to a plethora of innovations in the 20th century: a cascading tree resulting in an exponentially increasing (for a time) number of innovations.
The first of these innovations were adopted slowly. Even when there was an obvious relative advantage (electric lighting, for instance), the technology was not compatible: there was not much electricity generation and distribution. When electricity generation and distribution started to be built, there was uncertainty about whether it would end up being high-voltage direct current or lower-voltage alternating current. Etc. Every step of the way, fundamental decisions about how to make the innovation compatible had to be made. But every decision that was made was one fewer decision that future innovations needed to make. As time went on and the underlying technologies were incorporated into the socio-economic paradigm, every innovation in that technology’s lineage became more compatible. If you squint at the graph up top you can see this: in general, the further from 1894 an innovation associated with electrification is introduced, the faster is it adopted. But once we get to innovations associated with IT, in 1971, adoption slows down again. From there, IT innovation adoption speeds up over time to today, when the good ones are adopted very quickly.
As the information technology revolution ages, innovative information technologies are not only more compatible, they seem less complex because they are familiar. They are also generally easier to trial and observe. All of these things speed up adoption. But this means the rate of adoption is cyclical: it is slow at the beginning of technology waves and fast at the end. We have limited memories, so we don’t remember the change from fast adoption at the end of the last wave to slow adoption at the beginning of the current wave. We just see the speeding up, not the standing start. This isn’t just a feature of now, it was a feature of the late stages of the last wave also, in the 1960s.11
There’s a danger to this type of thinking also, both from a consumer point of view and from a corporate strategy point of view. At the beginning of a technology wave, innovations will be adopted quickly only if they have really high relative advantage. Consumers start to associate quick adoption with useful products. But later in the technology wave, innovations that are part of that wave will be adopted quickly because they are less complex and more compatible, even if they have low relative advantage. But this is a trap: useless products will be adopted quickly and their quick adoption will cause others to think they are probably useful and then also adopt them. GE had about $1 billion in sales of electric knives in 1966 (invented in 1964) and in 1971 about one in three US households owned one.12 Even though electric knives are overall less useful than regular knives, people bought them because they were electric.
Innovations change society when they are adopted more broadly, not when they are adopted more quickly. Some products are adopted quickly by a few and then peter out. It took the electric knife about seven years to get to 33% adoption; it took the electric blender three times as long.13 But there are far more households with blenders today than electric knives. The electric knife was familiar and compatible, but had low relative advantage. People bought it, they tried it, they abandoned it. The blender, on the other hand, was less familiar when it launched but far more useful.
This mistake can lead to bad business decisions. The VCR was adopted quickly at the tail end of its technology revolution. At the same time, the PC was being adopted more slowly because it was at the beginning of its associated revolution.
This slow adoption caused the then established computer companies (IBM, DEC, etc.) to stay out of the PC market for years, so outsiders dominated the early years. I would argue, in fact, that almost all of the conditions, aside from relative advantage, that cause very fast adoption of a new product work against startups. Slower adoption is caused by uncertainty, after all, and uncertainty is the place where entrepreneurs thrive. If a new product is compatible, not complex, trialable, etc., as well as being a good business proposition, established companies will be quick to bring it to market or be fast followers. This, unfortunately, means that the best strategic position may requires startups to endure slower adoption rates.
If the speedup of adoption from faster and broader communication has reached its limits, then “consumption spreads faster today” is technically true, but not something to get excited about. And if some new information technology product is adopted overnight, that doesn’t mean it has staying power. Fast adoption merits, more than anything else, skepticism.
I’m not saying that slow adoption is good, just that it might be unavoidable if you have a truly revolutionary product. And I’m not saying that fast adoption is bad, just that there is a danger in simply believing “things are adopted faster”: these are probably not the things that are going to change the world.
The other thing that bothers me about this chart is that it is the worst cited thing I have ever seen in a piece written by academics (though par for the course for the New York Times). This bothers me at a visceral level. This instantiation was published on February 10, 2008 as a New York Times op ed, “You Are What You Spend”, written by W. Michael Cox and Richard Alm, who were, at the time, the senior vice president/chief economist and the senior economics writer at the Federal Reserve Bank of Dallas, respectively. They are now both at SMU. The chart in the NYT is credited to Nick Felton who, since he is one of the leading graphic designers in the world, I assume did the graphic design. The Harvard Business Review republished the chart sometime later, completely redesigned, and credited it to Felton. This just left me confused: did they think Felton was the person who sourced the data, or did Felton actually redesign the chart for the HBR? In any case, neither place told me where the data came from. This is both annoying and unprofessional. Why should I believe data that doesn’t cite its source? I dug deeper and found another paper written by Cox and Alm and published by the Dallas Federal Reserve in 1997: “Time Well Spent”. This is a more detailed version of the op-ed. It has an earlier version of the chart, but the data sources cited are vague and not easily replicated. In another piece by Cox and Alm where the chart is shown, the cite is simply “Authors’ calculations”. Sigh. In any case, in the below I use data from Our World in Data, “Technology Adoption by Households in the United States” who seem to have gotten most of it from the awesome Horace Dediu, who seems to have gotten most of it from the US Census, though not all. The data is pretty similar to the chart. ↩
McGrath, Rita, “The Pace of Technology Adoption is Speeding Up”, published November 25, 2013, updated September 25, 2019, The Harvard Business Review, https://hbr.org/2013/11/the-pace-of-technology-adoption-is-speeding-up, accessed June 9, 2024. ↩
Chart from Jovanovic, B., & Rousseau, P. (2005). General purpose technologies. Handbook of Economic Growth, 1(05). https://doi.org/10.1016/S1574-0684(05)01018-X, p. 1194. ↩
I’m using the fifth edition: Rogers, Everett M., Diffusion of Innovations: Fifth Edition, New York: Free Press, 2003. ↩
Rogers, p. 233. ↩
Rogers, p. 240. ↩
Lievrouw, Leah A., and Janice T. Pope. “Contemporary art as aesthetic innovation: Applying the diffusion model in the art world.” Knowledge 15, no. 4 (1994): 373-395. ↩
Rogers, p. 247. ↩
Rogers, p. 221. As a source he cites the previous edition of his book, which seems like cheating. I spent some time looking for independent confirmation of this number but found only other academics citing Rogers as authoritative, as I just have. I would love to know how Rogers got these numbers. ↩
Rogers, p. 221. ↩
See, for instance, the 1970 book Future Shock. ↩
https://www.foodandwine.com/lifestyle/kitchen/electric-carving-knife-history, https://www.chowhound.com/1638203/kitchen-tools-not-popular-anymore/ ↩
US Census Bureau, Major Household Appliances, various years. ↩