On the Decline in Research Productivity

Photo by Nicolas Thomas on Unsplash

We live in interesting times. Technology continues to advance at breakneck speed and the fourth industrial revolution is upon us. What were once concepts relegated to science fiction have become integral to our everyday lives. Our smart phones allow us to connect with virtually anybody in the World in real time, and I’d bet that most of us are completely unaware of the extent that artificial intelligence and machine learning pervades the interactions we have with our phones (you can read more here).

One might assume that this advancement may in part be attributed to the compounding effect that newer technologies have in raising our productivity levels. That would seem to make sense; with access to the World’s body of knowledge at our fingertips and virtual collaboration ever easier, surely the large strides that we are making are being facilitated by the improvements made in the past which compound our future output?

The data would seem to disagree with this hypothesis. In a fascinating paper released by Bloom, Jones, van Reenen and Web, an argument is put forward that while our economic growth rate and rate of technological advancement is being maintained, the effective number of researchers required to maintain that growth rate has increased enormously, giving rise to a dramatic reduction in the research productivity of each researcher.

Put another way, maintaining the rate of technological advancement that we have enjoyed since the second industrial revolution has come at a price; in this case the cost is in the number of research hours required to invent and industrialize a new technology.

In fact since the 1930’s, while the Total-factor Productivity growth-rate has hovered below 5 % per annum, the number of researchers required to fuel this growth has risen by a factor of almost 25 times!

TFP-num-researchers-graph
Growth in number of researchers with TFP Output (Bloom et al).

Bloom et al applied this insight to a number of industries including a study of the fabled Moore’s Law, pharmaceutical innovation and mortality, and crop yields in the US.

Let’s take a closer look at two: Moore’s Law and Agricultural Yield.

Moore’s Law

Moore’s Law has been hailed as a key driver of economic growth during the last fifty years. The observation is named after Gordon Moore, a co-founder of Fairchild Semiconductor and Intel, and who observed that the number of transistors per integrated circuit doubles approximately every two years. This observation first made in 1965, has held remarkably constant for over 50 years. This is an absolutely stunning growth rate which translates to an exponential growth rate of 35 % per year!

moore-law
Moore’s Law (Wikipedia)

While the output side of Moore’s Law shows constant exponential growth, it is only by looking at the corresponding input side of the equation that we see the full picture. The constant exponential growth has certainly come at a cost: today there are 18 times more researchers working on semiconductors than in 1971.

Put another way, as a result of the declining research productivity, it is around 18 times harder today to generate the exponential growth that underlies Moore’s Law than it was in 1971.

Effective number of researchers to maintain Moore’s Law (Bloom et al).

Crop Yields

A second very interesting example in the decline of research productivity is in the evolution of crop yield growth over time on US farms. With agricultural yield being of such importance to the well-being of an economy, there is a wealth of published data showing the growth in yield over time. Yields on four US crops were measured; Corn, Soybeans, Cotton, and Wheat, and all show that the yields across the four crops studied, roughly doubled over the course of 55 years (1965 to 2015). This equates to an annual growth rate of around 1.5 % per year.

Increase in US crop yields over time (Bloom et al).

In order to achieve these compounding growth rates, the number of active researchers working in the biological efficiency (cross-breeding and genetic modification) and crop protection and maintenance (including the development of herbicides and pesticides) fields has increased by a factor that ranges from 3 to 25 depending on the crop and research measure.

Yield growth and research effort by crop (Bloom et al).

As with Moore’s Law, the large increase in the number of researchers required to maintain historic yield growth rates equates to substantial reduction in research productivity with factors ranging from 6 for wheat to 23 for both Corn and Soybeans. On average, research productivity in the area of crop yield declines by between 4% to 6 % per year.

My Thoughts

My biggest takeaway from the research presented here is that incremental gains in the advancement of a technology are progressively move difficult to achieve as that technology matures. Advancement, whether in medicine, agriculture, medicine or technology, is hard fought and exhibits a clear law of diminishing return. However, this has not stopped companies from matching the increased demand for additional researchers with supply in order to continue to match the growth rates of previous years.

This demonstrates to me just how critical it is to the industries studied that historic growth rates are maintained. The marginal cost of increasing research input to maintain output must be surpassed by the resulting increase in demand for more advanced products. The increased demand from industrial sectors that are growing rapidly, such as the semi-conductor industry, more than compensates for the growing R & D burden placed on those companies to maintain their growth trajectory. Neglecting or reducing R & D in one of these industries would relegate that company to the graveyard, as the technological advances which result in today’s state-of-the-art are deemed obsolete tomorrow.

Simply put, with such an explosion in demand for ever faster CPU’s, the semiconductor industry with continue to suffer the decline in research productivity in order to achieve the gains associated with Moore’s Law. Similarly, with populations still expanding across the globe, crop yields must necessarily increase in order to feed the growing population. Here again, the increase in demand for crops must necessarily outstrip the increase in R & D costs associated with improving and growing crop yields.

I don’t think that one should look at the decline of research productivity in a negative light at all. Scientific research is a high paying and highly skilled profession and should certainly be encouraged as this is how we as humanity advance. This is especially true when one considers that the companies producing the goods and services that we all consume, shoulder a large portion of the increased research burden.

The paper does however cause me to question the sustainability of the current status quo. How much longer will the diminishing returns in new idea generation associated with these sectors continue to outstrip increased demand for the product/service?

Product life cycle graph
Product life cycle graph

Perhaps this is just another angle from which to approach the Technology Life Cycle; as products/technologies reach maturity, incremental improvements become harder and harder to effect. With increased competition, there is also an increased burden on the incumbents to remain a market leader, leading to an increased R & D spend to maintain their dominance. All of this however, is good for you and me as the consumer, as the increased competition plus increased research spend results in improved technology at reduced prices.

Perhaps also we are just nearing the peak of both the semiconductor and traditional agriculture industry before disruptive innovations such as quantum computing and brand new farming methods render the current technology obsolete?

aeroponics-system
New farming methods like Aeroponics could change the future face of farming (farmcrowdy).

Additional Reading

Leave a Reply

Your email address will not be published. Required fields are marked *

This site uses Akismet to reduce spam. Learn how your comment data is processed.