# Epistemic Spot Check: The Fate of Rome (Kyle Harper)

post by Elizabeth (pktechgirl) · 2019-08-24T21:40:01.164Z · score: 42 (19 votes) · LW · GW · 3 comments

## Contents

```  Introduction
Claims
1%-30% log distribution
60%-90% normal distribution
80-100%, c – log distribution
0-10% log distribution.
90-100% c – log distribution
80-100% c – log distribution
Conclusion:
None
```

# Introduction

Epistemic spot checks are a series in which I select claims from the first few chapters of a book and investigate them for accuracy, to determine if a book is worth my time. This month’s subject is The Fate of Rome, by Kyle Harper, which advocates for the view that Rome was done in by climate change and infectious diseases (which were exacerbated by climate change).

This check is a little different than the others, because it arose from a collaboration with some folks in the forecasting space. Instead of just reading and evaluating claims myself, I took claims from the book and made them into questions on a prediction market, for which several people made predictions of what my answer would be before I gave it. In some but not all cases I read their justifications (although not numeric estimates) before making my final judgement.

I expect we’ll publish a post-mortem on that entire process at some point, but for now I just want to publish the actual spot check. Because of the forecasting crossover, this spot check will differ from those that came before in the following ways:

1. Claims are formatted as questions answerable with a probability. If a claim lacks a question mark, the implicit question is “what is the probability this is true?”.
2. Questions have a range of specificity, to allow us to test what kind of ambiguities we can get away with (answer: less than I used).
3. Some of my answers include research from the forecasters, not just my own.
4. Due to timing issues, I finished the book and a second on the topic before I did the research for spot check.
5. Due to our procedure for choosing questions, I didn’t investigate all the claims I would have liked to.

# Claims

Original Claim: “Very little of Roman wealth was due to new technological discoveries, as opposed to diffusion of existing tech to new places, capital accumulation, and trade.”
Question: What percentage of Rome’s gains came from technological gains, as opposed to diffusion of technical advantages, capital accumulation, and trade?

1%-30% log distribution

Data:

• The Fall of Rome talks extensively about how trade degraded when the Romans left and how that lowered the standard of living.
• https://brilliantmaps.com/roman-empire-gdp/ shows huge differences in GDP by region, implying there was a big opportunity to grow GDP through trade and diffusion of existing tech. That means potential growth just from catch up growth was > 50%.
• Wikipedia doesn’t even show growth in GDP per capita (with extremely wide error bars) from 14AD to 150AD.
• Rome did have construction and military tech (https://en.wikipedia.org/wiki/Roman_technology)
• It also seems likely that expansion created a kind of Dutch disease, in which capable, ambitious people were drawn to fighting and/or politics, and not discovering new tech.
• One potential place where Roman technology could have contributed greatly to the economy was lowering disease via sanitation infrastructure. According to Fate of Rome and my own research, this didn’t happen; sanitation was not end to end and therefor you had all the problems inherent in city living.

Original Claim: “The blunt force of infectious disease was, by far, the overwhelming determinant of a mortality regime that weighed heavily on Roman demography”
Question: Even during the Republic and successful periods of the empire, disease burden was very high in cities.

60%-90% normal distribution

The wide spread and lack of inclusion of 100% in the confidence interval stem from the lack of precision in the question. What distinguishes “high” from “very high”, and are we counting diseases of malnutrition or just infectious ones? I expected to knock this one out in two minutes, but ended up feeling the current estimates of disease mortality lack the necessary precision to answer it.

Data:

Original Claim: “The main source of population growth in the Roman Empire was not a decline in mortality but, rather, elevated levels of fertility”
Question: When Imperial Rome’s population was growing, it was due to a decline in death rates, rather than elevated fertility.

80-100%, c – log distribution

“Elizabeth, that rephrase doesn’t look much like that original claim” you might be saying quietly to yourself. You are correct- I misread the claim in the book, at least twice, and didn’t catch it until this write-up. This isn’t as bad as it seems. The claims are not quite opposite, because my rephrase was trying to explain variation in growth within Rome, and the book was trying to explain absolute levels, or possibly the difference relative to today.

Back when he was doing biology, Richard Dawkins had a great answer to the common question “how much is X due to genetics, as opposed to environment?”. He said asking that is like asking how much of a rectangle’s area is due to its length, as opposed to its width. It’s a nonsensical question. But you assign proportionate responsibility for the change in area between two rectangles.

Fate‘s original claim was much like asking how much of a trait is due to genetics. This is bad and it should feel bad, but it’s a very common mistake, and I give Fate a lot of credit for providing the underlying facts such that I could translate it into the “what causes differences between things” question without even noticing.

Since weak framing wasn’t a systemic problem in the book and it presented the underlying facts well enough for me to form my own, correct, model, I’m not docking Fate very harshly on this one.

Original Claim: “The size of Roman merchant ships was not exceeded until the 15th century, and the grain ships were not surpassed until the 19th.”
Question: “The size of Roman merchant ships was not exceeded until the 15th century, and the grain ships were not surpassed until the 19th.”

0-10% log distribution.

This is true within the Mediterranean, but if  you check Chinese ships it’s obvious it’s off by at least 100 years, possibly more.

Original Claim: too diffuse to quote.
Question: The Roman Empire suffered greatly from intense epidemics, more so than did the Republic or 700-1000 AD Europe.

90-100% c – log distribution

https://en.wikipedia.org/wiki/List_of_epidemics shows a pretty clear presence of epidemics in the relevant period and absence in the others.

Original Claim: too diffuse to quote.
Question: Starvation was not a big concern in Imperial Rome’s prime.

80-100% c – log distribution

https://en.wikipedia.org/wiki/List_of_famines shows Roman famine in 441 BC (the Republic) and isolated famines from 370 on, but pretty much validates that during the prime empire, mass starvation was not a threat.

# Conclusion:

My fact checking found two flaws:

1. An inaccuracy in when ships that exceeded the size of Roman trade ships were built, and/or forgetting China was a thing. The inaccuracy does not invalidate the author’s point, which is that the Romans had better shipping technology than the cultures that followed them.
2. Bad but extremely common framing for the relative effects of disease mortality vs. birth rates.

These is well within tolerances for things a book might get wrong. I’m happy I read this book, and would read another by the same author (with perhaps more care when it refers to happenings outside of Europe), but they are not jumping to the of my list.

Is The Fate of Rome correct in its thesis that Rome was brought down by climate change and disease? I don’t know. It certainly seems plausible, but is clearly advocating for a position rather than trying to present all the relevant facts. There are obvious political implications to Fate even if it doesn’t spell them out, so I would want to read at least one of the 80 million other books on the Fall of Rome before I developed an opinion. I’m told some people think it had to do with something military, which Fate barely deigns to mention. In the future I hope to be a good enough prediction-maker to put a range on this anyways, however wide it must be, but for now I’m succumbing to the siren song of “but you could just get more data”.

[Many thanks to my Patreon patrons and Parallel Forecast for financial support for this post]

PS. This book is the first step of an ongoing experiment with epistemic spot checks and prediction markets. If you would like to participate in or support these experiments, please e-mail me at elizabeth-at-this-domain-name. The next round is planned to start Saturday August 24th.

comment by Tetraspace Grouping (tetraspace-grouping) · 2019-08-24T23:11:37.520Z · score: 4 (4 votes) · LW(p) · GW(p)

What do the probability distributions listed below the claims mean specifically?

comment by Elizabeth (pktechgirl) · 2019-08-25T17:52:41.327Z · score: 3 (2 votes) · LW(p) · GW(p)

it's a 95% confidence interval for where the actual probability lies.

comment by jacobjacob · 2019-08-25T18:01:26.324Z · score: 11 (10 votes) · LW(p) · GW(p)

I think of it like "95% confidence interval of where the mean of Elizabeth's estimate would land after 10 hours of further research".

I've found personally this format is often useful when giving quick probability estimates. Precision is costly.