Rust in Numbers

I met Norm two decades ago. I recall him clothed in slacks and a button-down — I think it was the sort of shirt that featured a breast pocket expressly for carrying mechanical pencils. Norm wore glasses, didn’t have any hair, and had a big bass voice. I heard that voice a lot, because I stood next to him in a small choir run by the Presbyterian church just outside Princeton, N.J. I clung to Norm. I was not a strong musician. The melodies lured me to their line. Norm kept me locked in on the harmony.

Back in those days, we often chatted before practice began. That’s how Norm discovered that I was writing my dissertation about life insurers and the ways they valued people, invented new categories of difference, and forecast the spans of lives. 

As I was explaining my research to him, he recalled how back in his days as an engineer at the phone company, he used to forecast the spans of lives too. He was not prophesying human deaths. No. He forecast the deaths of machines.

At the next practice, Norm handed me a copy of an old book, with his name written in black pen capital letters along the top. It had a soft, mostly unadorned cover the color of a paper bag and stained by years of spilled coffee. It was over 200 pages long.¹

Just a few pages in, I found the book’s first image, a chart showing age in years along the horizontal axis and percentages climbing from 0 to 100 along the vertical axis. A line drawn on the graph showed a curve that looked like a backward S: it began flat then dropped steeply before leveling off.²

The book identified it as a “survivor curve” and it modeled an unidentified, hypothetical population. Whatever constituted this modeled group, they were not humans. At first, 100% of that population aged 0 years old survived. Nearly all still survived at 2 years, but by a little more than 7 years the population had been halved. The rate of decrease began to slow after that and the final individual died at the age of 20. 

A few pages later, the book presented a slew of similar curves. These, though, ran across an axis of years ranging from 0 to 100 and each curve bore the name of a nationality: Australia, Sweden, United States, Germany, and India.³ Even without the labels, I could have readily identified these as survivor curves for people. They followed a distinct course, one that began with a steep decline caused by the percentage of survivors plummeting in the first two years. This was a characteristic statistical stain, a mark of the tragedies endured by so many families who lost a child in those first, fragile years.

Life insurance companies had varied names for the columns of numbers they derived from these graphs. In an optimistic mood, one called them life tables. Otherwise, they were mortality tables. These were the fundamental building blocks of life insurance, an industry that entered its modern phase in the middle of the 18th century, when the first large British insurers began using such tables to model the likelihood that those they insured might die.⁴

The book’s graphs, these quantifications of human mortality, were meant to train engineers in how to read a survivor curve. More importantly, they grounded everything that was to come after in the legitimacy of an industry that by the 1930s was offering about $100 billion in protection to policyholders.⁵ Life insurance had already convinced the world of its scientific basis. The authors of that book borrowed that aura as they explored the life expectancy of things that never lived.

The next survivor curve, a few pages later, described the collective fate of “190 pile and frame trestles.”⁶ These were railroad supports, the sort of thing that held aloft the rails and prevented trains from plunging to the ground. Important stuff, and stuff that one should keep in good working order or replace when it could no longer be repaired. According to this curve, a pile and trestle might last as long as 22 years, but by its eighth birthday things were already starting to get pretty dodgy.

Flipping through the rest of the book, I learned that pumps in water works have an “average life” of 21.3 years compared to only 5.3 years for telephone switchboards.⁷ The life of an electric lamp is better expressed in hours (around a 1,000) than years.⁸ A railroad tie made from Douglas fir lives a couple more years, on average, than one made white oak and a manure spreader outlives an automobile by more than three years.⁹

Norm had spent a career with these sorts of facts. But I found them a bit odd. I mean, do telephone switchboards have life spans? They were never alive to begin with. So why had this metaphor taken root? What did it mean to an engineer like Norm and to those who came before him? 

This first installment in a two-part essay shows how the engineers used the life span as a tool for facing the inevitable problem of mechanical decay. The second part reveals how the metaphor boomeranged, how it transformed into a tool for managing the engineers themselves. Norm spent a career with these facts, while employers spent time and energy predicting the length of engineers’ careers and planning for the retirement of Norm and his generation of workers. They quantified a kind of late-century career (backed by a well-funded pension) that provokes nostalgic sighs today.   

Suppose you are an engineer participating in the age of industrial expansion in the United States. It is the late 19th century. You find yourself responsible for ensuring the quality, safety, and efficacy of machines that millions of people have come to rely upon every day. You keep the trains on time. You ensure telephone calls connect. You keep the lights on, the water running.

You face myriad challenges. There are, for one, risks of obsolescence. Inventors design new tools, features, or systems and the engineer may have to scrap perfectly good stuff to make room for something even better. The public too, the consumers of your goods and services, may cause headaches. Say you maintain a street car network where riders climb on at the rear, by the tracks, but passengers prefer entering from the sidewalk and demand street cars that make that possible. A railway engineer in 1913 reflected on this situation, complaining about “the too critical public, often voiced through demagogues” which could force him to scrap a fleet of otherwise functioning rear-entry cars.¹⁰  

Leaving aside the public, the owner of your capital goods presents a very great challenge. Andrew Carnegie famously razed a fully operational, brand new rolling mill, because it became possible to build a new one capable of generating steel even more efficiently.¹¹ The mill wasn’t inadequate, which is how engineers might talk about a machine that could no longer live up to the needs of the firm. Carnegie was simply single-minded in reducing his operating ratio—his costs divided by his revenue—and investing in a new mill would bring down expenses.  

Now, say a machine managed to stay relevant and could keep up with the workload that you required of it. Still, this machine would eventually need to be replaced. Operation wears away even the best-made capital good. You might, like an early-20th century civil engineer, talk about the “perishable parts of the plant” being “consumed, just as fuel is consumed though at a slower rate.”¹² A railway engineer warned his peers in 1913 that an apparatus “may seem to be lasting indefinitely, but it is burning up as surely as is the coal which is shoveled under boilers.”¹³

Your machines consume themselves, slowly and surely, or soaring demand makes them instantly inadequate, or a new fashion renders them suddenly obsolete. You can repair and mend wear and tear. Or you can purchase new machines and new parts using money you have either squirreled away in your allocated budgets or money you will need to ask the machine’s owner to borrow. The accountants taught you that this money is a depreciation expense.

Depreciation is about to become a battlefield, but, as an engineer, you don’t yet think of it as such. That’s not to say you don’t consider the cost of wear and tear. It’s a reality that you think about a lot. In fact, you try to keep a close eye on the state of repair of all the material and apparatus your company employs. 

You may even want to keep records of how long some of your machines or materials last, but it is costly and difficult to do so. In 1911, for instance, an industry committee circulates a questionnaire that asks railways for information about the ties that they used to support their rails — can they provide an accounting of when those ties were first laid and when they were removed? The answer is, mostly, no. One manager laments that “with the class of help we can get on this work it seems almost impossible to get an accurate record.” That manager’s railway instead started keeping records of smaller “experimental sections” of track.¹⁴ Even if you or your company desire to predict the length of usefulness of your property, employing trained people to keep track of so many miles of track or pipe or cable is expensive.

Something like a survivor curve for materials could be a stand-in for actual data. You could use it to leverage a small amount of information about your physical plant. It would allow you to predict depreciation and so make estimates for how much maintenance to plan for, all without having complete records. There have even been some initial forays into using these tools of actuaries to think about non-living things. In 1905, a German official was among the first researchers to attempt to create a survivor curve for industrial goods: for wooden telegraph poles, as part of a study of the efficacy of different treatments of the wood.¹⁵  

But few such studies exist. After all, to plot the survivor curve, one first needs a substantial store of accumulated data, and that is very hard to come by.

The thing is that you don’t usually need very precise estimates of depreciation anyway. You just need a general idea so that you can make a case for socking away some revenues to keep up with repairs. And in the daily work of deciding or predicting what needs to be replaced, you don’t think a mathematical formula could ever really tell you how an average machine will depreciate. You will instead rely on a careful examination of the things in front of you. You will rely on your judgment.¹⁶

Historically, this is what experts have done. The historian of science Theodore Porter explains in his book, Trust in Numbers, that even when engineers or actuaries in the nineteenth and early-twentieth centuries used calculations as a tool for guiding their decision making, they tended to be skeptical of relying too much on formulas and calculations alone. They believed that good decisions can be informed by numbers, but ultimately require judgment. It was a balance.

Soon, though, across the 1920s and 1920s, you will decide that a re-balancing is called for. You will decide that you want more numbers when it comes to valuing depreciation. It will be worth it to you to gather more data and to put more energy into calculating and forecasting the life spans of machinery. Why?

The reason is that you work for a monopoly that provides a key public service — and political pressures are about to rise. 

In the 19th century, industrialists had built railroads that spanned continents, trailed by telegraph and telephone wires. The resulting revolutions in transportation and communication came alongside a revolution in organization: the modern corporation took shape and increasingly took control. A bureaucratic arms race ensued, as states and companies faced off, each borrowing techniques and tools from the other, each hoping to exert control over the other.¹⁷

Governments at every level–from cities up to the feds–face pressure from mass movements of farmers and workers and officials want to ensure that the corporation you work for does not use its monopolistic power to fleece all those who rely upon it. In an attempt to rein it in, your company’s rates are regulated.

Depreciation becomes a contest between the regulator and the regulated. Those of you who work for the telephone company or the railroad or the waterworks will use your experience and your knowledge of your physical plant to provide your bosses with numbers indicating the amount of money needed to keep the enterprise functioning, to prevent it from being consumed by use. Your peers—fellow engineers employed by states, municipalities, and the federal government—will make their own estimates. The companies’ estimates will usually say the cost of depreciation is higher than the regulators’ estimates.

As the fight of expert against expert over depreciation valuations grows more fierce, you may decide to rely more heavily on numerical tools. And so too, you might prefer to have something objective, like an actuary’s survivor curve, to point to and say: “look, this curve provides a number for the years a water pump will survive. There’s no point arguing further.” 

Your motivation to generate precise numbers about depreciation using authoritative methods builds. A key factor here is what the US Supreme Court decided in 1909. In the case Knoxville v. Knoxville Water Company, the Court affirmed that every public service company should be guaranteed a profit—no rates (which must be approved by regulators) should be set so low that it couldn’t offer returns to its investors.¹⁸ The decision also added that rates needed to be high enough to cover the costs of depreciation. There were reasons to worry here: the case also made clear that Knoxville Water ran a deeply corrupt business, inflating its stock and engaging in rampant self-dealing. You might worry whether such a company can be trusted to actually use consumer money to fend off depreciation. 

And yet, there is something utopian in this arrangement, something the engineers can love. Many of you, after all, do wish you would have the freedom and resources to maintain your factories and fend off decay. It is powerful and important that governments affirm the necessity of reinvesting in productive capacity. It is an argument for a society’s collective responsibility to respond to decline and decay.

In practice, the decision puts a premium on collecting data about how long property lasts. This data will be used in court and in regulatory hearings when the government tries to lower your rates. The government will also try to gather data, presumably to make the opposite argument. After Knoxville, it starts to pay to invest in keeping careful records about industrial property.

Today’s apologists for the mass surveillance of individuals by technology firms sometimes speak of the “digital exhaust” left behind by users on the Internet. That metaphor rings false, since companies like Google actively engineer their systems such that users generate this so-called exhaust. It would be more true to say that many of the largest private enterprises of the nation—the railroads, the telegraph and telephone purveyors, utilities for cities big and small emitted vast plumes of data exhaust. An industry for generating numbers sprang up within larger industries in order to squeeze more money out of consumers. 

Companies after Knoxville devoted new energy to surveying all of the property they used in running their business.¹⁹ They began creating records of how long that property lasted, how long it remained useful. They did this to argue for higher rates from regulators. But, incidentally, they made it possible for risk calculations to embrace a new part of the human-built world.  

A midwestern researcher, Edwin B. Kurtz, was there to take advantage of the data these monopolies left behind.

Kurtz completed an undergraduate degree in engineering at the University of Wisconsin in 1917,²⁰ at a moment when the Midwest was a hotbed for all sorts of reform, from progressive calls for the direct election of senators to the installation of a national income tax. The second decade of the 20th century also saw the spread of a new kind of insurance scheme by which employers paid a premium against the risk that their workers would be injured. This was called workmen’s (now worker’s) compensation. Kurtz imagined a parallel program for the nation’s productive machinery.²¹  

Companies in Kurtz’s scheme would not pool their resources, though. Instead, each would be responsible for creating an insurance fund in-house. In essence, the company would take out a policy on each of its pieces of property. Each year it would pay premiums set in accordance with the predictions of Kurtz’s life (or morality) tables, as well as assumed rates of return on investments and costs of replacement. When a piece of property had to be retired, the company would redeem the policy, and get back money to replenish the switch, rail, pipe, fan, car, or cable, or whatever it had lost. Through such a scheme, “the property thus becomes perpetual,” wrote the idealist student.²²

It was an engineer’s ideal, exactly the sort of thing one might expect from an eager student. Kurtz wanted companies to keep precisely calculated reserve funds that would force them to save to pay off depreciation. 

For his thesis, Kurtz built a catalog of life tables projecting the fates of industrial goods. Consider the goods he began with: water pumps, wooden telegraph poles, railroad ties, electric street lamps, electric railways, and telephone cables. Why these? Because there was data for them, thanks to Knoxville. Kurtz did not generate the observations for his life tables. He found them among the records of regulatory hearings and engineering publications. 

After completing his thesis, Kurtz went on to do graduate work and eventually landed a job as a professor of electrical engineering at Iowa State College, which he would help turn into a powerhouse institution for calculating industrial risk. By 1930 he had designed a new way of thinking about machines and the way they deteriorated. The demands of accountants and lawyers involved in rate regulation after Knoxville had turned the spotlight on depreciation. Kurtz helped to usher the field toward an actuarial era. Actuaries were the original specialists in the scientific study of risk. In the early days of British life assurance, actuaries were mathematically competent leaders of firms. 

With the rise of major US firms in the 19th century, actuaries became professionals more focused on data and mathematics than on business more broadly. These actuaries began by taking observations made in the world and then fitting them to a curve.²³ For a very long time, actuaries drew curves by hand through their data, relying on their eyes and judgment to find the shape that best approximated the average of the data points and all their noisy imprecision. Astronomers did this too, to devise smooth lines from spotty data collected by looking at the skies. 

The actuaries eventually deployed mathematical models for determining a curve to fit their observations of populations as they lived and as they died. Edwin Kurtz marveled that the actuaries’ favorite model equations for smoothing their data into a curve seemed to work even better for machines than for people. He took it as proof that even the human-built world succumbed to natural laws and asserted grandly that “observations regarding the life of physical property can be classed as natural phenomena.”²⁴  

Robley Winfrey, a native Iowan, attended Iowa State in the early 1920s. He became Kurtz’s protégé and proselytizer.²⁵ He did not win adherents through pretty words or authoritative arguments. Instead, he offered a tool kit that looked like a book. This was Statistical Analyses of Industrial Property Retirements, the book Norm lent me in church choir when it was already 70 years old and in a revised edition.

Engineers like Norm picked up Winfrey’s book for its catalog of curves. Each survivor curve presented a certain type of machine or material. In his original studies, Kurtz had identified seven different types of physical property, classifying them  according to their distinctive survivor  curves and attributing to each a set of characteristics: from property relatively resistant to obsolescence like waterworks (a pump was a pump, a main was a main) to goods that evolved more quickly in industries sensitive to change, like railroad cars or ties.²⁶ Winfrey expanded the taxonomy to include 18 distinct type curves, each drawn by hand, which could be reproduced on “transparent graphs” to ease identification.

Say an engineer, like Norm, encountered some new data pertaining to a company’s drill presses. That engineer could plot out how many of the drill presses had been replaced or “retired” in the first few years: this constituted an initial or “stub” curve. Next the engineer could try lining up transparent graphs from Winfrey’s catalog over top of it. Lay one transparent sheet over the plotted stub. Did it fit? No? Move on. Then, bingo, once a stub matched with the beginning of one of Winfrey’s curves, the engineer could predict how long the rest of the company’s drill presses would survive. 

In that 1935 text, Winfrey departed from his mentor in one small, but significant way. He stopped writing about “mortality tables” or “mortality curves” and used only the terms “survivor tables” or “survivor curves.” “The term mortality suggests human beings and not inanimate objects,” he reasoned.²⁷ He wanted to make the distinction clear. This wasn’t about people.

He titled his book Statistical Analyses of Industrial Property Retirements though. And that muddied the water in unexpected ways. An engineer retired a piece of machinery that no longer functioned. Maybe technological changes had rendered it obsolete. Perhaps the machine had simply worn out from years and years of grinding, day in and day out.

But in the early 20th century, it became more common for corporations to talk about retiring people too. And those corporations would transform once again the tools of risk. A technique born in the oceans that came to land, a method that moved from people to machines, would be remade once more. It was coming for workers.