In the twenty-first century, research has become universally accessible. Anyone with a high school education and an internet connection can read original research.

This is a quiet modern miracle.

A critical reader is no longer forced to rely on conflicting authorities; where there’s controversy, one can turn to the actual evidence on each side. Everyone can, in principle, have an informed opinion.

In a companion article, I will walk you through how to find the most important research on a topic. Before that, though, I want to correct a common misunderstanding about how to read scientific evidence. Let’s suppose you’re trying to read the research on a question — say, ‘is sleep training harmful?’. A naïve way to read the research is:

• Find all the relevant studies.*
• Sort through them and throw out anything that looks dubious.
• Weigh up the ones that find ‘pro’ and ‘con’ evidence, and decide how strong the cases are.

I am delighted when people put in this effort; nothing annoys me more than appeals to authority. But, unfortunately, this process often gives misleading answers. The point is beautifully illustrated by a paper that concludes:

Parachute use did not reduce death or major traumatic injury when jumping from aircraft. (Yeh, 2018)

*I use ‘study’ and ‘experiment’ interchangeably below. Meanings differ subtly between fields; I’ll use both terms for ‘empirical evidence’.

That… unusual… paper is making a very serious point.

Science is not just about experimental evidence; it’s also about theory. Experiments strengthen our beliefs in theories, and theories tell us far more than experiments on their own could. If you only look at experimental evidence, you will get a misleading impression of the research. If you read to the end of this article, I’m confident I can convince you of that. You won’t need any advanced scientific or statistical knowledge; just the ability to think for yourself. I’m going to walk you through a case study, and I’ll explain all the material you need to understand it.

If you want to read more about the confusion that arises when people misunderstand how to read science, I’d recommend this excellent blog post:

For avoidance of doubt: I am taking a case study from child development, because that’s where I know the research. The case study is on circumcision, because both the theoretical background and experimental evidence are unusually compact. But these are sensitive areas…

• Child development research is not the same as popular ‘parenting advice’. In some cases, popular advice often tells people the opposite of what the research says.
• For non-US readers: I am not trying to make a point about any religion(s). A majority of male newborns are circumcised in the US.

Stop!

I need you to do something. It’ll take twenty seconds, and you need to do it now, or you won’t be convinced by this article.

The case study below looks at circumcision of newborns. Before you read on, please try to predict what long-term psychological effects circumcision might have, if any.

(This is a technique for countering the cognitive biases we all have. It’s only after you try predict something for yourself that you can get an objective sense of how impressive other people’s predictions are.)

.

.

.

.

Theory

The Force of Evolution

Can you remember your first crush? Can you remember how it colored everything? You felt that way because evolution exerts huge forces on our psychology. We have strong drives to behave in ways that make us more likely to survive and to have children.

In the environment in which we evolved, around a quarter of children died in their first year, and only half made it to puberty (Volk, 2013). This enormous evolutionary pressure means babies and children have strong drives that would have made them more likely to survive (and later procreate) in their original environment.

Of course, the world has changed. Infant mortality has dropped dramatically, and so many evolved drives aren’t needed for survival now. But our world has changed too fast; evolution will need many, many generations to ‘catch up’. So we, and our children, are stuck with the drives that fit our environment thousands of years ago.

These ideas let us make some remarkable predictions about child development. To understand them, we’ll need a few concepts from evolutionary biology…

Developmental Plasticity

An individual organism can ‘grow up’ to have different adult forms depending on what environment it grows up in. This is called developmental plasticity. For example:

• The fruit flies reflect a general biological rule that smaller forms are better suited to higher temperatures. (See Bergmann’s rule and its developmental-plasticity version, the Temperature Size Rule.)
• For the butterfly Bicyclus anynana, lower temperatures happen in the dry season, when eyespots would interfere with camouflage (Lyytinen, 2003).

A simplified but useful picture is that organisms have switches which are set in early life, and which affect what kind of adults they turn into. We’ll come back to these switches in a moment.

Life History Theory

Evolution allows more than one way to ‘do well’. Life history theory studies the evolutionary strategies that different species use. The most important kind of variation is between fast and slow strategies.

Organisms which follow fast life strategies reach sexual maturity quickly, have many offspring and invest little in each one. They also invest less in their own upkeep; they ‘live fast and die young’. Organisms which follow slow life strategies do the exact opposite. You can see the fast-slow continuum illustrated here:

These different strategies are a result of tradeoffs. Organisms have a limited supply of resources, so they can’t do everything well. If they invest more in their own upkeep, i.e. have longer lifespans, they have fewer resources for reproduction. An organism can only survive a long growth phase if there is sufficient parental investment to protect it while it is growing. And so on.

Fast and Slow Strategies in Individuals

Life history theory has also found fast-slow trade-offs for individuals in a species (Réale, 2010). So within one species:

• Some individuals follow faster strategies. They start having children earlier, have more children and invest less in them. They also live less long.
• Some individuals follow slower strategies. They start having children later, have fewer children and invest more in them. They also live longer.

Individuals tend to pick strategies that fit their environment in an evolutionary sense. Here’s an example: if you are in a dangerous environment, it doesn’t make sense to wait long to have offspring. There’s too much risk that you’ll die before having any at all. So the ‘fast’ strategy is better suited to dangerous environments.

Remember this? When individuals alter the way they develop to fit their early environment, that’s just developmental plasticity. Choosing between fast and slow life strategies is a kind of developmental plasticity; it’s just that the ‘switches’ being set affect not only physical form, but also behaviour.

For example, if an individual faces a dangerous early environment, that will set the biological ‘switches’ which make the danger-suited fast strategies more likely. These switches affect an organism’s risk tolerance, libido, and other psychological factors; those psychological effects push the organism to ‘live fast and die young’, to have children early, to have many children and to invest less in them.

Where are the Switches?

In vertebrates, the most important behaviour-affecting switches seem to be located in the hypothalamic-pituitary-adrenal axis, more often called the HPA axis. This is part of the stress response system; in humans it looks like this:

The stress response system is highly plastic: its lifelong behaviour changes in response to early experience, and particularly early stress (del Guidice, 2011; Gunnar, 2009; Levine, 2005).

Now, when I said individuals ‘choose’ strategies, I was being a little loose with language. Even in primates, it’s not a matter of conscious choice; the early environment induces biochemical changes in the stress response system and other parts of the brain. The long-term biochemistry of the brain changes in response to the early environment — the levels of different hormones change. Those hormonal changes affect risk tolerance, libido and other observable psychological factors.

Recap

What I’ve said above applies to all vertebrates, but we’re particularly interested in humans. To sum up:

• Danger in the environment causes stress, which is reflected in increased cortisol levels.
• In early life, large or sustained cortisol rises (& other similar changes) ‘flip switches’ in the stress response system.
• These flipped switches result in different lifelong levels of hormones.
• Those changes in hormone levels push an individual towards increased risk-taking, risky sexual behaviour and other parts of a faster evolutionary strategy.

This is a very simplified picture of life history theory; if you want a proper introduction, there’s an excellent one in (del Guidice, 2016).

The Nature of Danger

I’ve outlined how humans react to dangerous early experiences by switching to fast life strategies. But: what counts as dangerous?

Remember that the ‘choosing’ a life strategy isn’t conscious; it’s an automatic biological process. It’s something that we’ve evolved to do. So what matters is what was dangerous in the environment we evolved in. In many cases those things are no longer dangerous, but evolution hasn’t caught up.

If you stand just outside a cage containing a snarling lion, your adrenaline and cortisol levels will likely rise even though your conscious mind knows you are safe. What’s true for adults is doubly true for babies, whose perception of danger is almost entirely hard-wired. If something was dangerous in the evolutionary environment, babies will perceive it as dangerous, their cortisol levels will rise, and so on.

… and that’s it for the theory. Now to a case study where we can see how effective the theory is…

Neonatal Circumcision — the Experimental Evidence

Until recently, we had no experimental data on long-term psychological effects of circumcision. However: circumcision is known to cause enormous increases in the cortisol levels of newborns (Gunnar 1981).

As noted above, the stress hormone cortisol is produced by the HPA axis, where key developmental switches are located. So life history theory gave us a prediction: circumcision would push men towards ‘fast’ strategies.

Miani (2020)

The first substantial study to test these predictions was Miani (2020). I’ve bolded their key predictions about the long-term psychological effects of circumcision:

Consistent with the life-history theory, we hypothesize that adults who underwent neonatal circumcision, compared to those who did not, will display alterations in socio-affective processing conforming to the fast life-history strategies, namely, decreased reliance on social environment and increased perceived stress and sexual activity.

Specifically, here we test whether early-circumcised men, compared to men who did not undergo neonatal circumcision, display higher attachment insecurity and emotional instability (i.e., impaired functioning in emotional, motivational, and social domains), lower empathy and trust, higher sexual libido and unrestricted sociosexuality (i.e., preference of short- over long-term mating), as well as higher stress and risk taking attitudes.

We can sum up these predictions like this:

I do want to emphasise that prior to the Miani paper there was very little experimental evidence about the long-term psychological effects of early circumcision. Here’s what there was:

• Bollinger (2011). Found a link to alexithymia, or difficulty idenfitying emotions.
• Bauer (2013) and Frisch (2015). Found links to autism.
• Ullmann (2017). Looked at only 9 circumcised men (!) and found no significant results.
• Studies on sexual function, reviewed in (Morris, 2013) which concludes circumcision had no effect.

Alexithymia and autism can cause problems with empathy. But apart from that, Miani (and co-authors) had no experimental evidence supporting any of their predictions. They were made just on the basis of life history theory.

Findings

What did the paper find? This graphs sum things up:

So, for example, if you look at this part…

You can see that early-circumcised men have higher sexual drive and higher stress levels than uncircumcised men. Which is exactly what life history theory predicted. Here’s a ‘scorecard’ showing how the experimental evidence fit the other predictions:

This is a remarkable confirmation of life history theory.

Parachute Skepticism

There is no stronger proof of a theory than this: it makes new predictions, and later experiments prove those predictions correct. For example, when Einstein formulated general relativity, he used it to make two new predictions; it’s was after those predictions proved correct that people believed the theory to be correct.

If you tried to make predictions yourself, you’ll have a sense of why this is. There are so many different things that circumcision could have caused that it’s inconceivable that the predictions just happened to be right. The only plausible explanation is that life history theory encapsulates some essential truth about the world. The details may be modified and refined over time, but at heart, the theory must have captured something real.

When people miss the importance of theory, and try to look at experimental evidence in isolation… well, that’s parachute skepticism again.

Here is the crux of this article. If you look at the Miani paper in isolation, there are many, many holes you can poke in it; for example, it doesn’t correct for multiple hypothesis testing. It’s all too easy to dismiss it and conclude that we have no evidence that neonatal circumcision has long-term effects. But that conclusion utterly fails to explain how life history theory made such effective predictions. It’s like watching someone ace an exam and concluding that they guessed all the answers.

So here as in many other areas of science, the experiments on their own don’t tell you anything; it’s the combination of theory and experiment that is compelling.

In some cases, when people focus exclusively on experiments, they’re doing that for good reasons. For example, they might work on drug discovery — the human body is so complex and so messy that our theories of it are very poor, so we are forced to rely mostly on raw experimental evidence.

More often, there’s an insidious phenomenon at play. As Kraft (2015), a study of motivated reasoning and scientific beliefs, notes:

Partisans … systematically denigrate, depreciate and counterargue evidence that is contrary to their political views but accept uncritically the supportive evidence.

In the social sciences, you can reject any research in a seemingly scientific manner. Saying ‘causation is not correlation’, talking about the replication crisis, pointing out possible confounders and, of course, parachute skepticism will let you disregard any paper.

You can imagine a perfect philosopher-scientist who does this systematically and lives with a deep sense of doubt about the world. But that’s not what happens in practice. As Kraft (2015) and much other research have found, people systematically use those superficially scientific techniques against just the findings that make them uncomfortable.

I’ll have more to say about this in a sequel article. For now, I’ll just plead: when you see people use these skeptical techniques, ask yourself if they’re applying them in a balanced way. Are they being skeptical about findings they are threatened by and those they are comfortable with? For example, in child development, are they being equally skeptical about things found to be damaging to children and things found to be beneficial? Or do they just denigrate, depreciate and counterargue the findings they feel threatened by? If so, that’s classic science denial.

Which theories work?

This article has been about how you should not read science. I’d like to close by telling you more about how you should read it. Unfortunately, that’s just hard. I’ve written about the importance of theories, but not all theories in social science are backed by solid evidence. There’s a world of difference between

• theories that only ‘explain’ pre-existing data
• theories that have made new predictions, later proven correct

The difference between these is like the difference between taking an exam paper you’ve seen beforehand and one you haven’t! But if you’re coming to a field from scratch, it’s difficult to get a sense of which theories fall into which category. You have to work through the material yourself to be really convinced.

I do strongly feel that whenever an expert makes an assertion, they should sketch the evidence — both theoretical and experimental — that supports their position. A roadmap, if you like, so you know where to start reading. I’d encourage you to hold everyone to that standard; it helps keep people honest!

Afterword: Reading about child development…

I know some are readers are here because you’re interested in how children develop, so let me close by giving you some pointers to further reading.

I wrote about life history theory and circumcision because they can be discussed compactly. But the most powerful theory, the one that has proven to have the most explanatory power, is attachment theory (Bowlby, 1958; Ainsworth, 1978).

Before I go any further: attachment theory is not the same as attachment parenting! Attachment parenting was originally called immersion parenting; William and Martha Sears changed its name to give it a veneer of scientific credibility.

The most detailed experimental work confirming the predictions of attachment theory is covered in The Development of the Person (Sroufe, 2005). It’s a remarkable book, and if you want to understand more about how children develop, I’d recommend it in the strongest terms.

Beyond that, watch out for easy answers; anything that suggests ‘one simple trick’ will improve children’s wellbeing is likely preying on parents. Watch out for behavioural findings that don’t have an explanation; as in the example above, solid work is generally backed by evolutionary explanations and grounded in biological and neurological evidence. And finally, watch out for political bias. Child development, sadly, has to deal with more than its share of politics, and that can mean that headline findings are misleading.

…or not reading about child development!

In case you’re a parent, and that seems overwhelming, let me say this. You don’t need to read deeply into child development to be a good parent. Just as children have evolved to survive, you’ve evolved to help them survive. Babies will let you know when something seems threatening to them!* If you follow your instincts and soothe them, things are likely to turn out fine.

*NB. This is not true if you keep ignoring babies when they cry. Crying uses a lot of energy (Barr, 1998), so babies stop crying if parents don’t respond — but they remain just as stressed as before (Middlemiss, 2012).

So my advice to parents would generally be, relax, do what comes naturally, and don’t worry about keeping up with the latest research. Your time is much better spent having fun with your children!

Thank you for reading.

One headache is that every publisher uses their own naming convention for PDFs, meaning that organising them is a pain. This article describes how to automagically rename PDFs so that it’s easy to organise them. Here’s an example of the end product:

I name files with the first author and date, but this method works with other naming schemes too.

Tools

The tools I use to organise my papers are:

• Dropbox, so I can access the PDFs on all my devices
• Goodreader, a great PDF reading/annotation app on iOS
• Zotero, which organises PDFs

Zotero is meant to be a central store of PDFs itself, but I don’t use it for that; it doesn’t handle duplicates well and I spend more time un-confusing it than it’s worth. Also it doesn’t work well with PDF highlighting software. I only use it to rename PDFs sensibly, and I remove them from Zotero when I’m done.

You don’t need to use Dropbox and Goodreader to follow this guide, but if you do you’ll end up with a lovely interface like this:

One-time Setup

These are the steps you need to follow just once on a PC, to set it up for renaming files. I’ve only done this on Windows, though it should work on Mac/Linux.

1. Install Zotero

2. Download the Zotfile plugin from here; clicking ‘Download’ should get a file with a name like zotfile-5.0.16-fx.api.

3. Click on Tools/Addons in Zotero:

4. Click on the ‘settings’ wheel at the top right, select ‘Install Add-on From File’ and point it at zotfile-…-fx.api.

If it works, you should see this new entry in the Add-ons window:

5. Now you need to tell Zotero how you want to name files. Click on Tools/ZotFile Preferences. Open the ‘Renaming Rules’ tab, and set your rules like this:

The ‘{%a} {%y}. {%t}’ is the key part that produces this naming scheme:

You can tinker with the settings if you want something different. I do recommend leaving ‘Truncate title after . or : or ?’ switched on, as those characters cause problems for various tools.

6. Open up File/Preferences in Zotero and check you have ‘automatically retrieve metadata for PDFs’ and ‘automatically rename …’ switched on:

Renaming PDFs

1. Put the PDFs you want to rename into one folder on your PC. I have a Dropbox folder called ‘Transfer’ that I use for this; any papers I grab, on any device, go straight in there.
   The steps below won’t delete any files from the Transfer folder, so if anything goes wrong, you can just remove files from Zotero and return to this step.
   I normally rename around 100 papers at a time. The first time you try this, I’d limit it to 20 files or so, to get the hang of the process.
2. Start Zotero. This guide assumes you don’t use it for long-term storage, so you shouldn’t see any papers in it:

3. Drag the PDFs from the ‘Transfer’ folder into Zotero. After a little while, you’ll see something like this:

That progress bar takes forever. You may want to have a …

4. Close the dialog; you’ll see most of the papers given sensible names:

5. Typically 5–10% of papers aren’t detected properly. If you sort by ‘Item Type’, you’ll see them listed as ‘Attachment’. These need manual renaming. Select them in Zotero, and then drag them out of Zotero and into to a sensible folder on your PC (e.g. ‘Transfer/To Rename’).

Now go back to Zotero and delete those undetected papers!

5. Make an empty folder to store the automagically renamed files; e.g. ‘Transfer/Renamed’. Open Tools/ZotFile Preferences and set ‘Custom Location’ to point to the folder you just created:

6. Close the dialog and select all the articles (Ctrl + A). Right-click and select Manage Attachments/Rename Attachments.

8. Click ‘OK’; you’ll see a long progress display scrolling up on the right. When it’s done, go to ‘Transfer/Renamed’; you should see something like this:

No original files have been deleted/renamed, so if anything goes wrong you can delete the ‘Transfer/Renamed’ folder and retry.

9. Some of the papers may have a missing year or author; it’s easiest to fix that by hand at this stage. This is also a good time to manually rename the ‘To Rename’ files from step 4.

10. Once you’re happy with the results, delete the original files from the ‘Transfer’ folder. Also delete all the files from Zotero, so you don’t come back in a couple of months and wonder if they are safe to delete.

And, you’re done! Sort the renamed papers into whatever folders you want to by topic. Don’t miss all the time you used to spend in libraries.

Maybe miss it a little.

Great, so if you were to write a science based, research backed rebuttal to this chapter alone, it would recommend what exactly? No daycare? Daycare in particular family instances at particular ages?

I ask knowing full well I likely won’t be able to meet the ideal, but I personally would prefer to know what that ideal is to attempt to get closer to it.

I didn’t want to get embroiled in the ‘parenting wars’, so promised to DM the commenter. And then this happened…

I am writing this article to avoid fielding requests for DMs for weeks and weeks. It’s for people who have genuine, upcoming choices about what kind of childcare to use. It’s not a commentary on anyone’s parenting choices, and it’s not for people who have to use daycare.

The most common piece of feedback I get is ‘I wish I hadn’t read this’. If you don’t have realistic choices, please stop reading.

Science and Authority

Books on parenting usually do this…

That’s the exact opposite of science. The whole point of the Enlightenment, of the dawn of reason, was that we moved away from appeals to authority. The only reason to believe a scientific finding is that someone points you at the supporting research. After each section below, I’ve linked to the most important research on the topic.

For academics: I’m discussing topics where the evidence from psychology, biology and economics converge, so that we can be quite confident about findings. If you feel anything is not adequately justified, please let me know here — but please absorb the evidence from outside your own discipline before you do. (Also NB I am writing for a general audience! Look at the substance, not the language.)

Center Care

There’s a lot more research on center-based daycare than on other childcare options. I’ll go over that first and then talk more briefly about in-home care, etc.. Effects are always measured relative to children staying at home.

One important caveat: most research looks at the average effect on children. Many children are affected much more or less than average. I’m not going to write ‘on average’ before every sentence — but, please, take it as read.

Summary of effects

First, here are the effects of 15–30 hrs of daycare a week, broken down by age.

• For ages 3+, there are few downsides and substantial advantages. Daycare boosts both cognitive skills (literacy and mathematics) and social skills as measured in the first few years at school.
• For age 2, the findings are more mixed. This is the best age to start in terms of boosting later cognitive skills, but children are more likely to act out and be angry when they reach school.
• For age 1, childcare may improve cognitive skills a little, though certainly less than starting at age 2. But it also has even larger negative effects on later behavior in school. There is no boost to social skills.
• For children aged 0–12 months, daycare likely damages cognitive skills and children’s later behavior at school is even worse. There is no boost to social skills.

If there’s one thing you take away from this post, please let it be this: age makes a huge difference.

Compared to 15–30 hours, 30+ hours a week before 4 or so doesn’t give any cognitive benefits but makes children more likely to act out and be angry when they reach school. For children from “middle class and affluent families”, that much time in daycare has about two-thirds the negative effect on behavior of having “a moderately depressed mother”. Children spending long hours in any kind of out-of-home childcare have been found to be three times as likely to have “elevated levels of aggression”.

Note the age and long-hours effects are separate: putting 6-month-old children in daycare for long hours has a particularly large negative effect on later behavior.

All of that is for a family of average socioeconomic status (SES).* That term covers the spectrum from poverty/deprivation (lowest SES) up through working class families and then to affluent middle class families (high SES). Children from lower SES families benefit from starting earlier, and children from higher SES families from starting later.

*In earlier versions I wrote ‘income’ to avoid using a technical term, but that was misleading. Researchers often use family income as an easy-to-measure proxy for SES, but child outcomes are affected much more by how much people read to children, etc., than by money. The more enriched the home environment, the later the optimal time for starting daycare.

Age matters. Hours matter. SES matters.

Sources: The effects on cognition and the link between center care and later externalising behavior (“acting out”, etc.) at schoool have been found many times. See e.g. (NICHD, 2002), (NICHD, 2004),(Melhuish, 2004) and (Stein, 2013). (Loeb, 2007) is the best single source, though, as it splits results by age and SES, and considers ages 0–5. Its findings about young children are borne out by later studies such as (Fort, 2016), (Kottelenberg, 2014) and (Morrissey, 2010). Over a dozen papers have shown negative effects of non-parental care in the first 12 months; see (Im, 2018) for a review. For differences in outcome by SES, see (Melhuish, 2015) and also (Votruba-Drzal, 2004) for cognitive aspects.

(McCartney, 2010) and (Huston, 2015) investigate the causes of the behavioral effect. Watch out for studies based on maternal assessments; those are known to be biased by maternal mental health and stress levels (e.g. Chen, 2017), and were abandoned by the NICHD SECCYD research team in the mid 2000s (Solheim, 2013). Reviews tend to be contaminated by studies with maternal questionnaires, so you really need to look at the original papers here.

The modern classic on SES and parenting styles is (Lareau, 2011). See (Francesconi, 2016) [quoted below] on parenting styles having much more impact than money.

Cognition vs Behavior

Those results might suggest a trade-off between cognitive skills and behavior, but that’s a shaky interpretation for two reasons:

• Parents likely impact cognitive skills more than daycare, but daycare impacts behavior somewhat more than parents. So e.g. parents can improve children’s literacy by reading to them at home, but it’s much harder to fix behavioral problems caused by daycare.
• Cognitive boosts probably fade out, although it’s not completely settled; skills needed to regulate behavior persist through life.

Sources: See e.g. (NICHD, 2002) study on parenting, but watch out for the fact that their measures include the effect of shared genes. For cognitive skills, contrast e.g. (Baker, 2019) and (Magnusun, 2004) which find fade-out with (Vandell, 2016) which shows persistence. On behavior, (Vandell, 2020) traces more time in daycare centers through impulsivity, risk-taking, behavior problems and police contact up to age 26, and has the relevant references to the older work.

(Baker, 2019) is a particularly useful large real-world study of the universal childcare expansion in Quebec; it gives some of the strongest evidence we have that cognition fades out while behavioural effects persist, and indeed that all the phenomena discussed in this article happen ‘in the wild’. I’ll say more about it below.

Social Skills

The part that I most often see parents get wrong is this: they assume sending young children to daycare will improve their social skills. It doesn’t: the only large study of this finds more time in daycare is linked to poorer social skills in primary school.

A likely reason is that children typically start playing alongside each other around 2 or 2½; playing together starts ~6 months later.* For children under 3, their main social environment isn’t other children; it’s you, their adult carers, who teach them about turn-taking and so on. Daycare ratios mean less adult attention per child, so we’d expect social skills to develop more slowly.

*1-year-olds can play with each other in calm, familiar environments with one-to-one adult support. Very few daycares provide that, though.

With less adult attention in daycare, children learn to interact aggressively instead — a study finds around one incident of physical aggression per 9 mins per child, excluding play-fighting. That in turn likely raises children’s stress levels, which I’ll discuss next.

Sources: The studies mentioned are (NICHD, 2005) on social skills and (Goldstein, 2001) on aggression. (Rubenstein, 1976) found 3% of peer interactions in home settings were aggressive, vs 40% in group settings. The remaining material is very standard, so easier to digest from textbooks like this than papers.

Cortisol and Stress

Why does daycare usage affect some children’s behavior years down the line? There’s strong evidence that it relates to children’s stress levels in daycare.

Measuring stress observationally is known to be unreliable, so researchers normally look at levels of cortisol, a stress hormone. There are many studies showing that e.g. substance abuse during pregnancy or growing up in adversity permanently affect children’s cortisol response, and that in turn has long-term consequences for behavior, memory and likelihood of mental illness.

The normal pattern (blue line below) is for cortisol levels to start high in the morning and decline during the day.

In daycare (pink line above) cortisol levels go up and up through the day. That is, children become increasingly stressed throughout whole-day daycare. This finding is very consistent and turns up in studies in (at least) the US, France, Germany, the Netherlands and Norway.

The graph above is for children in daycare morning and afternoon: children who are in for half-days show a less elevated cortisol response or even a normal one. This fits with the finding that children who spend 30+ hours a week are more likely to show behavioral issues later on in school. Higher quality settings cause smaller cortisol rises, which fits with their having smaller effects on behavior (as discussed below). Elevated cortisol depresses the immune system, which is one reason children catch so many bugs in childcare.

Aside: the research focuses on the long-term effects of stress, but I find that too clinical. If we’re making toddlers so stressed that we can measure effects on their behavior five years later, that’s upsetting in and of itself. Plus long-term risk of mental illness worries me more than bad behavior.

A lack of responsive, individual care from adults has been proposed as a cause for the cortisol rise. (That’s not a criticism of daycare workers; it’s just really, really hard for one person to look after 3 or 4 babies at once.)

Sources: the pathway from early childhood stress to an altered cortisol response to later health issues and mental illness is well established; (Hunter, 2011) is a review and (Miller, 2016) presents relevant references compactly. For the childcare-cortisol link specifically, the classic review/meta-analyses are (Geoffroy, 2006) and (Vermeer, 2006). The review (Vermeer, 2017) updates the findings. Following references from those Vermeer articles reaches most other sources, so I won’t repeat them here. But note especially (Roisman, 2009), which finds time in daycare centers predicts altered cortisol levels at age 15. (Drugli, 2017) is too new to be mentioned in the reviews. See (Bowlby, 2007) on children’s need for a secure attachment to a responsive adult in daycare settings.

Quality

High quality daycare reduces all the negative effects I’ve discussed above, although it doesn’t eliminate them. It also has long-lasting positive effects on educational outcomes. That’s the good news. The bad news is that it’s hard to find; only 10–15% of daycare in the US is high quality, and many child care centers offer “disturbingly low quality programs”.

My friends always say that they’ve personally been lucky and found fantastic daycare settings. The sad thing is that researchers have found that nearly all parents rate their settings as great. Compared to trained observers, parents overrate infant-toddler rooms by about two standard deviations. That’s like saying parents overrate heights by six inches or IQs by 30 points!

The literature on child care quality is extensive. I’m not going to try to summarise it here. I know some of you will want to learn to evaluate quality yourself, but it’s not something you can pick up from an internet article — or at least, I worry that I’ll do more harm than good if I try. Except: look for settings with as few children per adult as possible. That’s been shown to make a difference.

Sources: Many sources in ‘Summary of effects’ above also discuss the effect of quality. (Cryer, 1997) is the classic on parents overrating quality. If you want to look into quality more, start with (Burchinal, 2017), which reviews the ‘modest’ links between care quality and outcomes; its bibliography points you at the key studies.

Effect Sizes

Researchers talk about small, medium and large effect sizes. Those are technical terms. Even Cohen, who invented them, warned against using them out of context. Unfortunately, they’re abused all the time to push various agendas to laypeople. I want to try to give you a sense of how large daycare effect sizes are in a way you can understand.

Suppose we were talking about an (imaginary) nutritional program. Rather than telling you it has a small-to-medium effect size, it’s more helpful to say that it will, on average, make your children an inch taller as adults. You can get a sense of what that means. It won’t make them into giants, but it’s noticeable. An extra inch would push them from the 50th percentile into the 63rd percentile. In a class of 30 lined up by height, they’d move about 4 places from the middle.

That’s the rough average effect size for all the things that affect children. Breastfeeding, lead exposure, maternal depression all seem to come in at around “one inch” of whatever each of them affects — IQ, behavior or whatever. Some are a little larger, some a little smaller, but they’re all around that amount. For middle-class children, the effect of long daycare hours on angry/disruptive behavior also falls into this ballpark.

There are a few important things to note:

• Even with the boost from that nutritional program, many children will grow up to be short. (Just, on average, an inch less short than they would have been!) Similarly, even with extensive exposure to daycare, many children will grow up without behavioral problems or later mental health issues. Daycare and the other factors shift the odds; they don’t guarantee bad outcomes.
• A small shift in the average makes a big difference to the tail. If everyone was an inch taller, there’d be three times as many American men over 6'4". The same applies here: the ‘inch’ of worse behavior caused by long hours in childcare results in triple the number of children with ‘elevated aggression’.
• Remember average effects are measured. Your child may be affected much less than average, or they may be affected much more. That applies to both the positive and negative effects from daycare — remember age and hours affect the mix of positive/negative outcomes.

Sources: this is mostly basic statistics, but also draws on numbers from the sources mentioned in ‘Summary of Effects’ above. There’s also a lot of research on what factors make children more or less affected by daycare; see ‘Further Reading’ below for more on that if interested.

Other forms of childcare

To run a good scientific study, you need hundreds of data points. It’s (reasonably) feasible to find and engage 200 children in daycare centers and 200 cared for by their mothers. It’s much harder to round up, say, 200 children cared for by nannies. So most research compares center care & maternal care; research on anything else is sparser, so this will be a short section. Here are the key points:

• All forms of relative care seem to be as good as each other. Mothers, fathers, grandmothers — doesn’t matter. Indeed, there are large studies comparing single fathers and single mothers and the differences are minimal.

Let me emphasise here: I have never seen any research supporting the idea that women are better than men at caring for children.

• Time with professional childminders (a.k.a. in-home daycare providers) can cause later behavioral problems, but much less so than daycare centers. Childminders do not however boost cognitive skills of older children as half-days in daycare do. (Usual caveat: such boosts probably fade out, whereas behavioral effects have long-term consequences.)
• How do nannies (who come into your home) compare to professional childminders (in their own homes)? Research is sparse, but it seems that nannies may cause fewer emotional problems than childminders, and (like childminders) do not give cognitive boosts.

Sources:

Many of the sources I discuss in the ‘Summary of Effects’ section also talk about relative care and non-center out-of-home care, especially (Morrissey, 2010), (NICHD, 2004) and (Stein, 2013).

Beyond those, (Downey, 1998) is a (large) comparison of single fathers and single mothers. For grandmother care, see (Fergusson, 2008). Note that both these last papers fail to correct for extensive multiple hypothesis testing, and so statistically speaking we would expect them to find one or two positive hypotheses among the mass they test even if all null hypotheses hold. Both do, and I have not read too much into it as AFAIK their findings have not been replicated. In both papers the results for externalising behavior in particular are negative.

On nannies vs childminders, (Stein, 2013) compares them directly and finds that time with childminders causes more emotional and behavioral problems than time with relatives or nannies. (It finds childminders are on par with daycare, but that’s not consistent with other research.) (Dettling, 2000) and (Groeneveld, 2010) found cortisol levels rose through the day with low-quality childminders; (Gunnar, 2010) finds process quality affects cortisol but that structural quality does not. (Structural quality is objective things like caregiver training and experience; process quality is fuzzier things like whether caregivers are warm and caring.)

Other considerations

I get a lot of feedback asking about tradeoffs inherent in childcare choices, especially on family income. It’s hard to say much here as some of you will live in places where childcare is free, while others pay an entire second income for it. This may be useful, though:

Carneiro and Heckman (2003) and Cunha et al. (2006) present evidence that child cognitive and non-cognitive skills diverge at early ages across families with different levels of permanent income during childhood.

Levels of permanent income are highly correlated with family background factors such as parental education and maternal ability, which, when statistically controlled for, largely eliminate the gaps across income classes.

(Francesconi and Heckman, 2016)

That is, most of the difference in child outcomes between the highest and the lowest income families is due to how people parent, not having more money. (That doesn’t mean money is unimportant for child outcomes in all families — near the poverty line it makes a big difference.)

[Heckman is not just a Nobel Laureate, but one of the most influential economists in the world. We’re extremely lucky that he has an interest in early childhood.]

Further Reading

(Huston, 2015) is an excellent general review. (Bradley, 2007) and (Philips, 2010) are a bit outdated but short and readable. (Burchinal, 2015) has the more recent work; one thing to watch out for is that although it highlights issues with maternal questionnaires, it doesn’t catch all the papers that rely on them. (Melhuish, 2015) is also good, though be sure you look at the ‘general population’ half rather than the ‘disadvantaged backgrounds’ half. (Jacob, 2009) is a widely cited review of the effects of childcare on behavior. (Sylva, 2010) is a whole book of findings from a large British study.

I’ve left out some topics to keep this article at a sane length. Quality, for reasons mentioned above. Attachment theory, which goes a long way to explaining why younger children find daycare so stressful. How child temperament interacts with everything else. And finally gender — many studies suggest that behavioural effects are worse for boys. The reviews above will point you at more on these topics.

Case studies

I really want to stress that it’s a mistake to say ‘center care is good’ or ‘center care is bad’. It all depends on age, hours and family SES. The following case studies should underline this.

Quebec

Quebec implemented universal childcare from birth in 1996. The US NBER summarised the key paper in an article called ‘Canada’s Universal Childcare Hurt Children and Families’. Some quotes (my emphases):

Along virtually every one of these dimensions [of child well-being], children in Quebec see their outcomes deteriorate relative to children in the rest of the nation over this time period.

[..] this policy resulted in a rise of anxiety of children exposed to this new program of between 60 percent and 150 percent, and a decline in motor/social skills of between 8 percent and 20 percent.

[..] families became more strained with the introduction of the program, as manifested in more hostile, less consistent parenting, worse adult mental health, and lower relationship satisfaction for mothers.

• The paper picks up something that is specifically happening to preschool children, ramping up from 1996 to 2001. Nothing happens to older children, which makes it unlikely that it’s e.g. the Quebec economy.
• Quality standards were increased in 1996 as part of the program roll-out. Overall quality was actually pretty bad, but still ‘comparable to the quality of care provided in many other countries’. 25% of the centers were good quality, which is about double the figure for the US.
• Economists Steven Lehrer and Mike Kottelenberg heard about the results and flat-out didn’t believe them. So they replicated them. Lehrer writes:

Baker, Gruber and Milligan’s work is 100 per cent correct. It’s robust. If anything, in our own work, including a paper that came out in Canadian Public Policy and won “best paper award” for that year, it actually finds the effects get larger over time, on average.

• Most strikingly, the crime rate goes up by about 20% when the children who were in daycare become teenagers. That’s consistent with (Vandell, 2020), which traces more time in daycare centers through impulsivity, risk-taking, behavior problems and police contact up to age 26.
• The program covered ages 0–4; the negative findings were disproportionately driven by the youngest children. Hours seem to be long. And the program mainly changed the behaviour of advantaged families. So the negative outcomes should not be surprising. (Age matters. Hours matter. SES matters.)

Boston

A recent study looked at a 1996 expansion of universal preschool in Boston.

• The study was only looking at 4-year-olds.
• Hours were relatively short; at most six hours a day (30 hours a week).
• The intake was ‘relatively disadvantaged with high shares of non-white and low-income students’.

All three of these are linked to good outcomes. (Age matters. Hours matter. SES matters.) And that’s exactly what the study found—in Boston, preschool improved high-school graduation and college attendance, and decreased juvenile incarceration.

—

Of course, not all studies fit the age/hours/SES pattern perfectly — for example, overly regimented, academic pre-K can be detrimental to four-year-olds (Lipsey, 2018). But age, hours & SES capture the big trends.

Policy

You didn’t ask about policies, but I’m going to tell you (briefly) anyway, as it’s the only way to make life better for all our children. Feel free to skip this section.

Disclaimer first. If I tell you guaranteed parental leave makes children better off, you’ll think I’m left-wing. But if I tell you early exposure to/long hours in daycare are bad for children, you’ll think I’m a social conservative. I hate how every issue has become polarised nowadays. I want to build better environments for all our children, and I’m open to any policies that get us there.

• You sometimes hear that investing in early childhood has huge returns. (The ‘Heckman Equation’.) That’s specifically true for disadvantaged children. It’s based on an analysis of two high-quality interventions (1962 Perry Preschool and 1972 Abecedarian). The 200 or so children in those projects were all poor, and nearly all African American.
• Hours matter. There’s a big difference between being in a center 8am–6pm (50 hours a week) and being in a center 9am-3pm (30 hours a week).
• Above all else, age matters. Journalists keep missing this. If you muddle together all the evidence from studies on 12-month-olds and studies on 4-year-olds, it looks like the evidence on childcare is mixed. But that’s a completely wrongheaded thing to do.

I understand why many people desperately want to be able to put babies into childcare without any consequences. But that’s like wanting geoengineering to fix climate change without any costs. We live in a world of trade-offs, and in this instance the costs are paid by our youngest children.

What policies leave children and parents better off? For a start, paid parental leave is fantastic. It’s a hard political fight as it costs businesses, and they’ll push back hard — but we should fight for it. Pre-K for 4-year olds is also great and may be an easier fight.

Massive expansion of high-quality care for younger children is possible but very, very expensive; probably the best childcare system in the world is in Sweden, which spends 2% of GDP on it. Even there, parental leave means few children enter the system before 18 months (OECD, 2006). High quality care for younger children takes so much manpower that nannies/parental leave will be cheaper than center care, due to rent, etc.. A sad political reality is that proposals to ‘expand high quality childcare’ tend to be massively underfunded.

There’s also one policy no-brainer which doesn’t cost anyone. That’s to replace government-run childcare programs for babies and toddlers with equivalent cash transfers to parents. (For this to work, the government has to regulate childcare quality because — as noted above — parents can’t measure it.) What would this policy mean for parents?

• Parents who need to can still spend the money on professional childcare. But because fewer people will use center care (‘reduced demand’), the cost of center care will go down, so parents using it will end up with cash in hand.
• For other parents, the money will mean that they, or relatives, are able to look after their babies themselves.

So everyone ends up better off. Why would we not do this?

Conclusion

Well, I guess I never answered the question.

Great, so if you were to write a science based, research backed rebuttal to this chapter alone, it would recommend what exactly? No daycare? Daycare in particular family instances at particular ages?

I ask knowing full well I likely won’t be able to meet the ideal, but I personally would prefer to know what that ideal is to attempt to get closer to it.

I hope you now have the information to answer that yourself, because there are a lot of nuances which I’m nervous glossing over. But if you want my own answer, for the average child:

• The best behavioral and cognitive outcomes come from starting half-days in daycare around 2½. Switching to full days provides no benefits and long days may worsen behavior until around 4.
• Before 2½, any relative as carer gives the best outcomes. Failing that, nannies are probably better than childminders (in-home daycare) and both are certainly better than daycare centers.
• All of the negative effects of non-relative childcare are more pronounced for younger children; childcare choices in the first 12 months make the most difference, as children are particularly dependent on their carers then.

Last major revision: Nov 2022

We’re often told that if we avoid waste and live in harmony with nature, we can stop climate change. But…

Plastic bags don’t cause global warming. Recycling often does. Cattle are responsible for about 10% of global emissions. And reusable nappies cause more warming than disposable ones.

In this article, I’ll go through the things someone in a developed country can do reduce climate change, and show you how effective each actually is, backed up with the science.

The Little Planet

If I tell you that going vegetarian will reduce your greenhouse gas emissions by 1.1 tons a year, is that a little or a lot? It’s hard to know. Tons of emissions are too abstract; what we care about is actual global warming. But telling you that going vegetarian would reduce the world’s temperature by 0.000000000019 °C in 2100 doesn’t help either; there’s no easy way to take in a number with that many zeroes.

The reason that number is hard to understand is that we have 7.8 billion people on a huge planet. Anything one of us does is going to have a miniscule effect on global warming — but between us, we have a pretty large effect.

If the size of the planet and the number of humans make things hard to understand, there’s an easy solution. How about we imagine dividing the world into 7.8 billion tiny planetoids, each with just one person on? If we imagine that happens today, then I can show you the impact of your future actions on your little planetoid and the numbers will all be sensibly sized.

It’s time to meet a friend.

The Little Prince lives on a Little Planet. Coincidentally, it’s exactly 7.8 billion times smaller than Earth, and so 7.8 billion times more sensitive to greenhouse gases. So when the Little Prince flies a mile, the Little Planet warms up by exactly the same amount the Earth would if all 7.8 billion Earthlings flew a mile each. It’s remarkably easy for the Little Prince to see how he causes warming on the Little Planet because his actions aren’t mixed in with 7,799,999,999 other people’s actions.

Throughout the article I’ll tell you how much your actions would affect your very own Little Planet’s temperature in 2100. For example, for an average American, going vegetarian reduces their Little Planet’s 2100 temperature by 0.12 °C. That’s a sensibly sized number! We can compare it to a goal like ‘2°C of warming by 2100’, and see how much it helps.

In case you’re short on time, there’s a table at the very end that shows all the actions and their effect on the Little Planet.

Science and Sin

Like a Mills and Boone novel, an article on climate change has a well-established structure…

Memorable statistic, sad anecdote. Depressing statistic, terrifying statistic. Finger-pointing. Guilt, guilt. You are a terrible person for not doing more about climate change.

Can we agree not to do that here? I’m not going to subtly finger-point and you’re not going to imagine I am. Journalists love talking about blame, and how to dole it out, but that’s a) not at all backed up by the science and b) pretty stupid, because many people are ‘to blame’ for each unit of emissions.

The point about blame is a bit abstract, so here’s an example. If I buy an iPhone, who’s ‘to blame’ for the emissions — me, Steve Jobs or Apple shareholders? If I fly for work, is that on me or my employers? These aren’t questions science can answer.

Science can tell us how much I could reduce emissions by keeping my old phone for a year, or how much Apple could by using greener energy. There’s overlap there: Apple and I are going after the same emissions. But that’s okay; the real world is messy like that.

The best climate researchers talk about the impact of choices and actions rather than pointing fingers, and that’s all we’ll do here. In this article I’ll always show the impact of your choices and actions, assuming other people do their usual thing.

Method

Full technical details and sources for my method are in a long, dry post, but there are a couple of important things you should know up-front:

• I’ll always be giving numbers for average people in the US and UK. If you eat a cow a day, you’ll have a lot more impact on emissions when you go vegetarian. I’ll give you enough details to adjust everything for your own circumstances.
• The longer you have left on Earth, the more you’ll affect 2100 temperatures. It’s 2020 now; I’m going to assume you’ll be around until 2070. I had to pick something, and 50 years is a round number. You can approximately adjust for your own life expectancy if you like; halve the numbers for 25 years, and so on.

Note that I’m ignoring your past emissions; we’re not pointing fingers here.

Actions

Because our Little Planets are made by magically cutting the Earth into little Earth-like pieces, 2 °C, 2.5 °C, 3 °C and other warming levels cause the same amount of flooding, heatwaves and other events that they would on the Earth. All the 2100 cooling effects we find should be thought of compared to those targets. So for example:

• Actions that cool the Little Planet by 1.0 °C are a huge deal; that gets us from 3 °C in 2100 to 2 °C in 2100, which makes a huge difference in practice.
• Actions that cool the Little Planet by 0.1 °C are well worth doing; they reduce flooding, etc., by a noticeable amount.
• Actions that cool the Little Planet by 0.01 °C or less have a tiny effect on climate change. Focusing on them may make things worse by diverting effort from substantive actions.

For each action, I’ve linked to a long, dry post with full sources. All the nuances and caveats and scientific arguments and many, many numbers are packed into those posts, to keep this one readable.

Right — enough burbling. Let’s go!

Recycling

In theory, if you recycled everything you could possibly recycle and you were separating paper, metal, plastics, glass, etc., into different bins and the recycling process was the best possible, you’d cool the Little Planet by about 0.01 °C in 2100. (0.012 °C for the US, 0.007 °C for the UK.)

That’s one-hundredth of a degree, not one-tenth. It’s about the warming produced by travelling from New York to Detroit and back once a year on a train. A major reason why recycling has such a tiny impact on emissions is that processing recycling is itself an energy-intensive process.

In practice recycling doesn’t even save that 0.01 °C:

• A lot of recycling is mixed, and separating paper, plastics, etc., takes even more energy.
• Even after separation, not everything is recycled; due to contamination and other factors, a lot goes to landfill or incineration.

As a result, the effect of recycling on climate change is miniscule. Trying to slow climate change by recycling is like trying to lose weight by cutting your hair. Depressingly, recycling is the most common action people cite when surveyed about climate change.

Plastics recycling needs a special shout-out. As the charts below shows, most plastic recycling is exported to countries with bad waste management practices, where it will end up incinerated on open rubbish dumps, releasing toxic fumes, or dumped into the ocean.

tl;dr: Recycling doesn’t make a dent in climate change. Recycled plastics are often actively harmful, poisoning people or ending up in the sea.

Sources, and a shorter discussion of plastic bags here.

Travel

On average, each person in the US or UK flies about 5,500km each year, which gets you from New York to London. Cutting out those flights (in economy class) would cool the Little Planet by 0.11 °C in 2100.

That ‘on average’ is pretty misleading, though. About half of those people don’t fly at all. So the ones that do fly must fly twice the distance — 11,000 km, or London to New York and back again each year. Cutting that trip out would cool their Little Planets by 0.22 °C in 2100. That’s for economy, by the way; the figure for business class is 0.60 °C.

If you fly more, the rule is that flying 5,000 km/3,000 miles each year warms the Little Planet by 0.1 °C in 2100. So 10,000 km/6,000 miles each year is 0.2 °C, and so on.

Incidentally, the carbon offsetting schemes sold by major airlines don’t do what they claim to and won’t make much difference to emissions.

What about driving? The average person in the US drives about 20 miles a day in a car that does 25 MPG. (That’s as a driver, not a passenger.) That ends up warming the Little Planet by about 0.50 °C in 2100. That’s very dependent on car size and fuel efficiency, though; driving the same distance in a SUV causes about 1.00 °C of warming.

You emissions will also be proportional to the distance you drive. If you drive 40 miles a day in an average US car, that’ll cause about 1.00 °C of warming; if you drive 10 miles a day in an average US car, that’ll cause about 0.25 °C of warming. Cutting out 4 miles a day, or 20% of your distance driven, reduces warming by 0.10 °C.

If you switch to an electric car, you can roughly halve those figures; for an average person that reduces warming by 0.25 °C. You have to be a little careful, though, because a surprising amount of the warming figures I’ve given come not from burning petrol but from making cars; manufacturing just one typical US car might cause 0.05 °C of warming, or slightly more for an electric car. So if you want to minimise emissions, you should wait until your previous car is unusable before buying an electric car.

All the warming numbers for the UK are about half those for the US because cars are smaller and people drive less.

Sources for flying here and driving here.

Food

There’s not really a polite way to say this...

Cows and sheep burp.

They burp methane, which causes far more warming than carbon dioxide. Methane breaks down into carbon dioxide over time, but even by 2100 a ton of methane will have caused at least 10x the warming of a ton of carbon dioxide.

This means that the beef, lamb and dairy products that come from cows and sheep account for a large chunk of the embedded emissions in western diets. Other meat contributes a little as well, as you can see here:

All of this means that someone in the US or UK can reduce their emissions by going vegetarian or going vegan. The figures for the two countries aren’t that different:

• Going vegetarian cools the Little Planet by about 0.12 °C in 2100.
• Going vegan instead cools it by about 0.15 °C in 2100.

Most of that 0.12 °C from going vegetarian comes from giving up beef and lamb; other meats have a much smaller effect. You can also reduce emissions a noticeable amount by avoiding air-freighted food.

Sources.

I’m nervous writing about the next topic. Research suggests that when people feel climate change science is critical of their choices, they reach by attacking the science or attacking scientists’ own lifestyles.

Please don’t.

38% of young Americans think that ‘climate change should be a factor in a couple’s decision about whether to have children’. If you share that belief, this next section has numbers you might find useful. If not, I’d skip ahead. If you already have children, I’d skip ahead; this article is about choices going forwards, not blame or guilt.

Children

For an average person in the US, having one child increases the Little Planet’s 2100 temperature by 2.2 °C. The figure for the UK is 1.4°C. Figures for all developed countries are in the same range; those for developing countries are much lower.

(Those figures assume your child is splitting their time between your Little Planet and another parent’s Little Planet. If you’re choosing to have a child on your own, please double them.)

If you’ve been following the numbers so far, you’ll have noticed that these figures are much larger than any others we’ve seen: little people have a startling effect on a Little Planet. For an average person in the US, having one less child has about five times as much effect as giving up driving completely. One professor of climate science puts it starkly:

Unless you will ever contemplate lighting a forest fire, the decision to reproduce is probably the biggest carbon choice you will ever make.

It is deeply sad that something so natural and essentially human has so much impact on climate change. If you take away one idea from this article, it should be that climate science is indifferent to our sense of what is good and bad. Recycling feels so rewarding and green, but has no impact on warming. Having children…

Well, I think I’ve said enough.

You might hope that if you raise your children to be environmentally conscious, they’ll have a much lower impact. The research stabs us again here:

• ‘Tolerable’ lifestyle changes can only give a 30% reduction in energy use and emissions.
• People can be pretty effective at teaching children to recycle, which has about as much effect on warming as singing Kumbaya. But there’s almost no correlation between parents’ energy use and their children’s, even before they leave home. So you’re unlikely to get near to that 30% reduction.

None of that means that how you bring up your children doesn’t affect warming. If you managed to reduce the emissions from having one child by 10%, that’s 0.22 °C for the US, or about as much as you’d save by switching to an electric car. Having a child causes so much warming that even saving 10% is a big effect.

Writing this section has made me melancholy. If reading it has had the same effect on you, I am really sorry. There is some upbeat news, though: those numbers mean that supporting family planning and women’s education can have a huge impact on climate change, far more than e.g. planting trees.

Sources and more. This covers the research and the many misunderstandings of it in the media. It may be less dry than the other sources links, but be warned, it’s long!

Home Energy Use

We’ve now talked about nearly all the actions that people list when asked about slowing climate change. There’s one more that’s generally missed, which is to reduce home gas and electricity use. There are a couple of easy things that can be done here, such as hanging up clothes to dry rather than using dryers, but to really make a dent you need to install heat pumps and improve insulation.

I’m not providing numbers here because they vary a lot depending on where exactly you live. But very, very roughly speaking, concerted action here is probably comparable to switching to an electric car.

Summary

Here’s a summary of all the numbers we’ve found so far.

All of these are for the average person; the sections above tell you how to adapt them to your own situation. Also, please don’t take these numbers to be too exact; climate science involves a lot of estimates. What you should take away from this is:

Recycling has a negligible effect on the Little Planet. Driving, flying and diet have significant effects. How many children you choose to have has a much larger effect than anything else.

Please don’t fixate on the part about children if it raises strong feelings. For the average person in either country, changes to driving, flying and diet can reduce warming by about 0.5 °C.

Half a degree less warming is a big deal. Some people claim that individual actions can’t impact warming that much. There’s no research behind that claim, and our numbers show it’s just plain wrong.

I’ve written a post showing you how individual choices affect global warming. You really should read that first. This is a long supporting post telling you about the research, the (frequent) misunderstandings of it in the media and then the adaptations needed to get meaningful warming numbers.

Children

I’m nervous writing about this topic. Research suggests that when people feel climate change science is critical of their choices, they reach by attacking the science or attacking scientists’ own lifestyles.

Please don’t.

I don’t feel I’m in a position to criticise anyone’s choices or lifestyle. For many people, how many children to have is too personal a choice for any external factors to matter. I’m absolutely fine with that. I really don’t like telling people how to live their lives. If you aren’t playing a piano next door, and you aren’t telling me how to live my life, I’m likely fine with whatever you’re doing.

What I’m not fine with is people being shown distorted science. I get mad that criminal incompetence with statistics has made people think Toxoplasma gondii infections change people’s behaviour. I get mad when a popular author misleads people into thinking childcare choices make little difference to their children’s wellbeing. And I get mad when the same journalists who rail against climate change denial reject the scientific consensus on the links between population and climate change.

So, please read this if you want to know what the research says, and don’t read in any subtext. If you share the view of the 38% of Americans between 18 and 29 who ‘believe a couple should consider the negative effects of climate change when deciding whether or not to have children,’ reading this may be informative. If not, please don’t take research findings as personal attacks on your worldview.

As for what the research actually says…

That graph shows you different actions someone in a developed country could take, and the area of a circle is proportional to the emissions reduction from an action. The estimate for having a child applies in each year they are alive, not just the year you have them.

To be clear, there’s a range of plausible values for the actual emissions/year from having a child, just as there’s a range of plausible values for the effect of emissions on temperature, and I happen to think the most widely cited estimate of 58.6 tons/yr is much too large. But just as there’s a consensus that man-made emissions affect temperature, there’s a consensus that having a child has a larger impact on emissions than any other action. A professor of climate science puts it starkly:

Unless you will ever contemplate lighting a forest fire, the decision to reproduce is probably the biggest carbon choice you will ever make.

How Bad are Bananas, Mike Berners-Lee (2013)

To emphasise: I know that when people tell you that driving/flying/eating meat/having a child causes a lot of emissions, they’re almost always trying to get you to stop. I’m not. I think the job of scientists is to give people their findings as clearly and accurately as possible, without exaggerations and without omissions, and stop there. Science isn’t a weapon for manipulating people.

Enough. I’ll return to frequent misunderstandings of the research in a minute, but first we need to do see what that research actually says.

Research

58.6 tons per child per year

The figure 58.6 tons of emissions per child per year is widely quoted in mainstream media articles. It and the other numbers in the Guardian infographic above were taken from a paper called (The) Climate Mitigation Gap. That paper draws on an earlier paper, Reproduction and the Carbon Legacies of Individuals, which I’ll call Carbon Legacies. The authors of Climate Mitigation Gap also produced a follow-up FAQ.

There’s a chain of the Telephone game going on here, which takes a bit of work to unravel. First, Carbon Legacies says that

our goal is to quantify the consequences of the childbearing decisions of an individual

To that end, they adding up half the child’s emissions, a quarter of any grandchildren’s emissions, and so on, up until an (unspecified) end date. They model different future emissions trajectories. Under the constant emissions trajectory, they find emissions of 9441 tons for the US, 2498 tons for Russia and 2026 tons for Japan.

Next, Climate Mitigation Gap takes these figures and produces this infographic:

The leftmost column is the key one — just focus on that. To obtain the per-country figures marked on that left-most column, they divided the figures from Carbon Legacies by the life expectancy of a woman in the US/Russia/Japan. Then (I think) they average the figures for the USA (~120), Russia (~40) and Japan (~20) to get the final figure of 58.6 tons/year, which is the height of the green bar in the graph.

It should be clear at this point that 58.6 tons per child per year isn’t one we can use verbatim. There are several issues:

• Emissions are added up until an end date, which isn’t specified in Carbon Legacies. The choice of date makes a big difference to the final figure.
• Carbon Legacies considers several emissions reduction trajectories, but they are pretty arbitrary, not grounded in the data.
• Dividing the Carbon Legacies figures by life expectancy is not really meaningful…
• and nor is averaging over three randomly chosen developed countries.

I don’t mean to be too damning of the papers, by the way. Climate Mitigation Gap was primarily trying to make the point that in a developed country, having a child has a far greater impact on warming than any other action, and it succeeds in that.

How Bad are Bananas

Berners-Lee provides his own calculation for the emissions from having a child; in the 2011 edition of How Bad are Bananas, he cites 688 tons for an average US child and 373 tons for the UK, based on the assumption that emissions will drop by 3.9% a year.

3.9% a year is the reduction rate needed to hit the UK’s old Paris goals for 2050. As before, I find that emissions reduction trajectory somewhat arbitrary; we should be working forwards from the data to find likely trajectories, not backwards from targets set by politicians. 3.9% is well outside the range of projections.

The other issue with Berners-Lee’s analysis is that (for ‘simplicity’) it doesn’t consider grandchildren, etc.. I think that as with Climate Mitigation Gap, he is mainly trying to point out how large the numbers are here, but leaving out grandchildren makes a big difference — if you have a child in 2021, they will have grandchildren around 2050, and those grandchildren will produce 50 years of emissions before 2100.

Other Research

There haven’t been other quantitative estimates of the warming resulting from a child, at least in the 350 or so papers that cite Climate Mitigation Gap. There is plenty of related work, though. Everything I have found concurs that having that having children has a very large impact on warming.

This webpage has an alternative calculation of the emissions from having a child, finding that it increases an individual’s annual emissions by about 40%. (It’s not peer-reviewed, but that’s no reason not to take it seriously; we care about the argument, not the arguer.) Unfortunately it assumes the US will fully decarbonise in 2080, which is incredibly optimistic: the US Annual Energy Outlook for 2020 projects total US emissions in 2050 to be about equal to those in 2020. Greener policies may well shift that down a little, but carbon neutrality in 2080 is not remotely likely.

Science versus Journalism

If you read Climate Mitigation Gap and see what they’re actually claiming to do, all they say is that they consider individual choices and

calculate their potential to reduce greenhouse gas emissions in developed countries.

As part of this, they find that choosing to have fewer children has far more effect on emissions than any other action. As we saw above, that hasn’t been contested by any subsequent research.

That ought to be the end of the story. Unfortunately, what happened was that ‘the recommendation to have ‘one fewer child’ was picked up in the media and subject to considerable controversy’. Publications from National Review and Breitbart on the right to the Guardian and Vox on the left attacked politicised versions of the research. I want to dispense with their assorted straw-men here, so you don’t inherit their misunderstandings. (You can skip this section if you just want to see the calculation.)

Having fewer children will wipe out the human race

If you literally assume that everyone does the exact same as you, any action you take will have devastating consequences. If you don’t have children, the human race is wiped out. If you don’t drive, the automotive industry collapses and there’s a massive recession. And so on. The sane way to assess individual actions is at the margin; your individual choice not to have a child reduces the fertility rate by a tiny amount, rather than wiping out humanity.

We need more workers to support the elderly in 50 years

That assertion doesn’t actually contradict the research findings? But for what it’s worth, fertility has less impact on GDP than people assume because the working-age population have to support both the elderly and children: taxes go to pensions but also to schools. A higher fertility rate means fewer pensions per worker but more schools.

To be more quantitative: this paper weighs the factors against each other and finds that the socially optimal policy involves taxing people for each child.* NB. no-one’s seriously suggesting doing that; it’s just an economic analysis that shows that the ‘more workers’ argument is not too convincing.

*Socially optimal policy is an economics term I’d need way too much space to unpack.

People have a right to have children

Yes. I know. The authors of Climate Mitigation Gap weren’t questioning that, any more than they were advocating sending the secret police into homes to force people to recycle. One of the authors said:

‘I knew this would be a sensitive topic to bring up. Certainly it’s not my place as a scientist to dictate choices for other people. But it is my place to do the analysis and report it fairly.’

And in case that sounds too clinical, she continues:

“Having a child […] is a vote of hope,” she says. “It’s a vote that the world is going to be a better place and we can actually tackle this challenge.”

[…] in her own life, she and her fiancé are deciding whether they want to cast that vote of hope for themselves.

“Because we care so much about climate change, it is a factor we’re considering. But it’s not the only one.”

This is blaming the developing world for problems caused by the developed world

Climate Mitigation Gap is only about the impact of individual choices in developed countries. It says so in the second sentence of the abstract.

To be absolutely clear, the research is unequivocal in showing that the impact of having a child in a developing country is far smaller than in a developed one.

Don’t blame parents for their children’s emissions; that’s double-counting

Words like ‘blame’ and ‘responsibility’ don’t turn up even once in Climate Mitigation Gap! It only talks about the effect of individual choices on emissions. This straw man particularly annoys me because the Vox piece that popularised it is from an author who rails against science denial. (He also completely misses that the study is only about developed countries.)

Journalists love blame; it sells copy. But it’s totally unscientific. Why? Because for every unit of emissions, there are many, many people who caused that unit of emissions to be released.

If you have to fly for work, that creates emissions. Are you to blame, for choosing that job, or is it your boss’s fault for sending you abroad? Or maybe we should blame the owners of your company? Actually, if you have a pension, you own a chunk of all the big companies out there, so we need to blame you again.

All of those people can act to reduce those same emissions. That’s why Climate Mitigation Gap and other good research talk about actions, rather than blame.

Global warming is the fault of neoliberalism/capitalism/big business/fossil fuel companies and individual action can’t impact it

I have looked really hard and I cannot find any research to support this common sentiment, interpreted widely. Best I can determine, government policies and individual actions can have about the same effect on the warming rate. (Details for another article.) For now, I hope the ‘Little Planet’ formulation in the main article makes it clear that individual actions in the developed world can have a large impact on warming.

Emissions and Sins

Hidden in many of these straw men is the idea that emissions are sins. If there’s global warming happening, someone, somewhere has done something bad and we need to point a finger at them and jump up and down. This idea leads naturally to the viewpoints discussed above: if your children’s emissions are their sins, they can’t be your sins.

If there’s one thing I want you to take away from this article, it’s that thinking of emissions as stemming from sins is just factually incorrect. Setting aside indirect emissions via children, the research says that for the US ‘tolerable’ lifestyle changes can only give a 30% reduction in energy use and emissions; a 50% reduction ‘would require dramatic changes which we believe would be unacceptable to most people’. (Other research suggests that individual actions can only reduce emissions by 6–16%, but 30% is consistent with the numbers in my main article.)

One reason that larger reductions are hard is that you pay tax, and the government spends that in ways that produce emissions — not because government is evil but because it’s giving you healthcare, education and so on. Another is that emissions are entrenched in everything you consume and do, often in ways that are hard to see. For example, typical use of reusable nappies will result in higher emissions than disposable ones, because washing them uses emissions-rich electricity.

What this means is that however green your life is, you will have a large impact on global warming. If you have children, then however green their lives are, they will have large impacts on global warming.

Of course that doesn’t mean you shouldn’t have children. For many people, having children is more important than anything else, and that’s natural. But for 38% of young Americans climate change is a factor in the decision. If you don’t share their view, please let them read about the research without jumping to science denial.

Calculating the Impact

I’m not comfortable using any of the numbers above, because they all depend on very arbitrary assumptions about how fast emissions will decrease and about how many generations to consider. So we’ll redo the Climate Mitigation Gap calculation with those issues fixed.

Emissions Decrease

The US, the UK, and other developed countries are reducing emissions per person, if more slowly than most people think. I ended up writing a whole other (long) article on this, which has all the sources. Here are some highlights for per-capita emissions:

• UK consumption emissions dropped 1.2% a year.
• UK emissions are projected to drop 1.7% a year to 2035.
• US consumption emissions dropped 0.54% a year.
• US emissions are projected to drop 0.71% a year to 2050.

For reasons I explain in the article, I’m projecting an average annual drop of 1% to 2% in per-capita emissions. Here’s what that means for total emissions:

For the main article, I’ll assume 1.5% a year. But I’ll also show you below how changing this number affects the carbon impact of each child.

Grandchildren and great-grandchildren

Because we’re looking at the temperature at a particular date (2100), there’s an objectively correct way to decide how many generations to track. We want to consider all the emissions before 2100, and only those. In practice, if you decide to have a child now,* you’ll likely have grandchildren around 2051 and great-grandchildren around 2081.

*Or, well, 2021, unless you’re some kind of supermutant.

So we need to sum up emissions from:

• your child from 2021 to 2099
• your grandchildren from 2051 to 2099
• your great-grandchildren from 2080 to 2099.

Let’s firm that up. In an ideal world I’d find the probability distribution for parental ages at the times children are born, and use it to find the probability distributions for the years in which children, grandchildren and great-grandchildren were born. Unfortunately I can’t find that data. We’ll have to approximate things using the average ages of mothers and fathers when their children are born.

Those graphs tells us that the average age of US mothers and fathers when they have children are around 28 and 31. (NB. we’re averaging over all children, not just firstborns.) I could average those to give 30.5, but it won’t make a difference in practice if we go with 30. So, as I suggested at the start of this section, we’ll assume all grandchildren are born in 2051 and all great-grandchildren in 2081.

How many children does each US adult have? On average each woman will have 1.728 children; that’s called the total fertility rate (TFR). The UN projects that it will stay very close to 1.8 through the next century, so I’ll use that figure. Remember that’s children per woman; since the gender ratio is very close to 50–50, we get 0.9 children per adult.

That means on average, having one child means you will end up with 0.9 grandchildren and 0.81 great-grandchildren. This is a shorthand for saying that there’s some probability you’ll have no grandchildren, some probability you’ll have one grandchild, etc, and that if you average over all the possible ways things could go, the number of grandchildren will be 0.9. This matters because it means that if you have one child, the average (or expected) emissions from your grandchildren in a year will be 0.9 times the emissions one person produces in that year. As everywhere else in emissions calculations, that expected figure is the one to use.

That was all for the US. For the UK, the average ages are 30.4 for mothers and 33.4 for fathers, averaging to 31.9. TFR is projected to be between 1.77 and 1.78 between 2050 and 2100; that’s about 0.89 children per adult. All of these figures are close enough to the US ones that I’ll stick with those — given the wide range of estimates for emissions, trying to get things to be overly precise is meaningless.

The Final Calculation

We’re now ready to do the final calculation. As a reminder, we’re computing the impact of one child; we assume 0.9 grandchildren in 2051 and 0.81 great-grandchildren in 2081. Rather than using a fixed annual emissions reduction, I’ll show you the effect of a range of parameters. For comparison, I’ll also add your ‘direct’ emissions, assuming you’ll live another 50 years.

I could use current estimates for UK/US emissions to give the final carbon impact of each child in (mega)tons, but there’s a more convenient approach which lets us handle both countries at once. I’m going to measure the impact of each child as a multiple of current (2020) annual emissions. So e.g. for the US, where 2020 emissions are 21 tons a year, 100 years would mean 2100 tons, i.e. 2.1 kilotons, of emissions.

Here we go…

The emissions reduction rate is on the x-axis, and the emissions from your children on the y-axis. The green lines indicate the historical/projected emissions decrease rates from the last section.

With a 1.5% rate (‘UK Historical’), your descendants will emit almost exactly twice the emissions that you do — i.e. the red line is twice the height of the blue line. (The exact number is 2.02x, but we’re not working to that level of accuracy.) And changing the rate doesn’t change that by very much; at 1.0% they emit 2.2x your emissions, and at 2.0% they emit 1.8x your emissions.

There’s one last thing we need to correct for. Suppose I decide to have a child, and my wife decides to have a child. 1 + 1 = 2, so that’s two extra children, right? Well, no. We’ve double counted — statistically speaking, a decision to make a child is most often jointly made by two adults. If you like, you can think of your descendants as splitting their time between your little planet and your partner’s little planet. To compensate for that, we should halve the numbers above. That gives us this rule of thumb:

Half the carbon impact of a child and their descendants (to 2100) is close to your carbon impact, assuming you’ll be around until 2070.

More concretely, if we assume that emissions reduce by 1.5% a year, as in the methods post, we find that half the carbon impact of having a child is equal to 35.7x your 2020 emissions. That’s assuming your child has the same lifestyle that you do; let’s look at that assumption next…

Lifestyle

I’ve been told that a large family with a ‘green’ lifestyle might result in the same emissions as an small family with an average lifestyle. That’s prima facie plausible, but I just can’t square it with the research I’ve read. There are two problems.

First, as noted above, even the greenest lifestyle can’t reduce individual emissions by nearly as much as most people think. The food calculations gave us some inkling of this; you can halve your food emissions by going vegan, but that’s pretty much the limit. Another reason is that more people means more schools, more hospitals, more roads, and generally more government, all with associated emissions. Unless you teach your children not to pay taxes,* they have no control over those emissions. Per this paper, ‘tolerable’ lifestyle changes can only give a 30% reduction in energy use and emissions; a 50% reduction ‘would require dramatic changes which we believe would be unacceptable to most people’.

*Not recommended.

Second, I can’t find any evidence that you can have much long-term impact on your children’s emissions, let alone your grandchildren and great-grandchildren. The research tends to looks at older teenagers, and so doesn’t shed much light on long-term behaviour. That said, this finds a correlation of only 0.13 for saving electricity, which is a good proxy for reducing emissions; it’s likely to go down after they leave home.

It’s also worth noting that if you have a large family then your children are also slightly more likely to have large families (correlation of 0.12–0.15), increasing your descendants’ emissions.

Putting these two effects together, I can’t see that how you bring up your children could shift your descendants’ expected emissions by more than 10%. (Using the 0.13 and 30% figures above would only suggest 4%, but if you assume parents have total control over young children’s emissions, you get a somewhat higher figure. I’ll go with 10% in the main article, but the real figure may be a little lower.)

On the Little Planet

If you’ve read the main article, you’ll know we’re turning emissions figures into something more comprehensible by illustrating their effect on a Little Planet which is 7.8 billion times smaller than Earth.

We noted above that half the carbon impact of having a child is equal to 35.7x emissions in 2020. (Remember, the child splits their time between two Little Planets, so we only want half their emissions.) In 2020, the average person in the US produced 21 tons of emissions, and the average person in the UK produced 13 tons of emissions. That means having a child in the US generates 750 tons of emissions, and a child in the UK generates 464 tons.

Now, a Little Planet warms by 2.96 °C for every 1000 tons of emissions. That tells us:

• For an average US person, each child increases their Little Planet’s 2100 temperature by 2.2 °C.
• For an average UK person, each child increases their Little Planet’s 2100 temperature by 1.4°C.

If you’ve seen other sections, you’ll note these are much larger numbers than anything else we consider.

Sorry.

It would be lovely if the numbers weren’t so harsh, but science isn’t often that kind. It’s just a depressing fact that for those of us in the developed world, the number of children we choose to have is the main way we’ll affect warming in 2100. After all our calculations and cross-checks, this picture turned out to convey the right idea, even if our numbers aren’t the same:

One final point I should be clear on. I noted in the last section that the research suggests that encouraging children to live a ‘green’ lifestyle can’t reduce their climate impact by a large percentage — maybe 10% or so. But 10% of 2.2 °C is 0.22 °C, which is quite a lot compared to other choices, like going vegetarian yourself.

So how you bring up your children will make a difference to the temperature in 2100; it just won’t counteract as much of the warming from having children as you might have hoped.

I’ve written a post showing you how individual choices affect global warming. You should read that instead. This is a dry, technical supporting post showing you where the numbers come from. If you read it, please read it after my post on methods. For reasons given there, I’ll be relying a lot on a book by Mike Berners-Lee for emission intensities of actions, although every figure will be cross-checked against other sources.

Driving

Cars, Cars, Cars

Thinking about emissions from driving is conceptually harder than any of the other actions we look at. The reason is that people buy cars, and making a car emits a truly prodigious amount of CO₂. Berners-Lee notes that people are particularly prone to underestimate this because they use lifecycle analysis rather than the (harder but better) input-output method. Here are some of the figures he gives for the UK:

For comparison, the average person in the UK emits 13 tons a year from everything they do. What this means is that if in a spirit of eco-goodwill you were to go out tomorrow and replace your perfectly functional car with an electric vehicle, you would substantially increase your lifetime emissions.

Most of those cars are around the 10 ton mark, so we can use our rules to figure out that each car made increases the Little Planet’s 2100 temperature by about 0.03 °C. (The Range Rover and other large cars cause twice that warming or more.)

Figures for the US are somewhat higher as cars are larger; Berners-Lee’s old (2011) edition cites 17 tons for a Ford Taurus. Manufacturing one of those would warm the Little Planet by 0.05 °C in 2100. A Land Rover Discovery SUV has 35 tons of embedded carbon, which is twice as much.

Things are complicated by the fact that more than one person typically owns a car, so it’s not obvious who is ‘consuming’ the car and its associated emissions. You could account all the emissions of a car to the first owner, or you could spread it out over all the owners according to how much they drive. As we’re interested in the actual emissions reduction from your choices, not in pointing fingers, there is one correct economic way of doing the accounting, but unfortunately I don’t have the data to apply it. (And it’d probably take another whole article to explain.) Berners-Lee uses the simple approach of spreading out the emissions over all the miles driven, and we’ll stick with that.

Spending the Same Money Twice

There’s one other technical point I’ll note and then ignore for simplicity. When computing Little Planet warming, I’m assuming emissions reductions of 1.5% each year due to technological improvements. But one of those technological improvements is electric cars! If I tell you that

• you can reduce emissions by switching to electric cars and
• emissions from driving will go down by 1.5% a year because of better technologies like electric cars,

I’m trying to count the same emissions reduction twice. If I want to split out the gain from electric cars as a separate item, I should use a slower reduction rate; looking at fuel efficiency for new cars suggests 1% a year is a more appropriate reduction rate for non-electric cars. As noted in the methods article, switching from 1.5% to 1% only changes the final figure by 10%, so I’m not going to try to correct here.

Manufacturing and Driving

There’s a useful fact from Berners-Lee that I’ll use several times below:

As a rule of thumb, about half the carbon impact of car travel comes out of the exhaust pipe. A few percent come from the fuel before it is burned. The rest, typically 40% of the footprint, is associated with the manufacture and maintenance of the car.

I’m going to cross my fingers and guess that maintenance is a small proportion, so the emissions split 40:60 ratio for manufacturing and for driving. I’ll call this the 40:60 rule.

I should say I am a bit puzzled because it’s hard to square those numbers with this graph, also from Berners-Lee:

According to that graph we should expect manufacture and transport of petrol to emit about half what burning it does. So I’m not sure how one gets ‘about half’ of emissions from the exhaust pipe and only ‘a few percent’ from ‘the fuel before it is burned’. But I’m not going to rely on that split for fuel, so I’ll relegate this to ‘vaguely troubling’ and move on.

The UK

Miles per Gallon

Berners-Lee notes that the average UK car burns 36 miles per gallon (MPG) and emits 530g per mile or 0.33kg per km driven. This includes all the emissions needed to get the petrol into the cars, and the manufacture and maintenance of the cars.

Let’s check these figures. Since certain car manufacturers are notorious for rigging emissions tests, we’ll look for independent sources. This corroborrates that 36 MPG for petrol cars; the average including diesel cars is 38.8 MPG, but I’ll stick with 36 MPG.

This webpage uses data gathered by Which to estimate 14.3 kg of emissions per gallon burnt. Combining that with 36 MPG gives us 0.25kg per km driven. The same page estimates 0.099 kg per km of emissions from car manufacture and maintenance. Adding those gives a total of 0.34 kg per km, which is remarkably close to Berners-Lee’s estimate.

I can’t square that driving-manufacturing breakdown with Berners-Lee’s comment that ‘about half of the impact of car travel comes out of the exhaust pipe itself’. I’m also a little surprised because the webpage seems to uses a fairly crude version of life-cycle analysis and I’d expect it to give lower numbers than an input-output model. But the numbers above are the best I’ve got, so I’ll stick with them.

Distance Driven

What about distance driven? Berners-Lee says that the ‘UK average annual distance’ is 7600 miles (12,200 km), but I’m not sure whether that’s per car, per driver or per household. The Department of Transport estimates car driving accounted for a total of 278.2 billion vehicle-miles in 2019. Dividing by the UK population of 66.8 million, we find 6,700 vehicle-km per person per year. To avoid double-counting, we will attribute that to the driver. (It’s arguably better to split it equally between all passengers, but I don’t want to complicate the main article by making people do arithmetic.) It works out at about 11 miles driven a day.

Emissions

6,700 km times 0.33kg per km gives average annual emissions of 2.2 tons per person per year, which warms the Little Planet by 0.23 °C in 2100.

Using the 40:60 rule, we get

• 0.9 tons per person per year embedded in the new cars people buy, warming the Little Planet by 0.09 °C in 2100.
• 1.3 tons per person per year from driving those cars, warming the Little Planet by 0.14 °C in 2100.

The US

Miles per Gallon

The latest edition of Berners-Lee only covers the UK, but the 2011 edition has old data for the US. It notes that the average US car drove 22.4 miles per gallon and emitted 850 grams per mile. As with the UK figure, that includes manufacture and maintenance and so is higher than figures you’ll see elsewhere.

We have to be careful about using that directly, though. The 2011 edition of Berners-Lee estimated US emissions at 28 tons per person; the 2020 edition at 21 tons per person. That’s a 25% reduction, which we can’t gloss over. (In the same timespan, UK emissions per person drop from 15 tons to 13 tons, a 13% reduction.) Some of that reduction will be due to driving more fuel-efficient cars and greener manufacturing processes.

That graph is from the EPA, and shows fuel efficiency for new cars over time. You can see that it looked to be around 22.5 MPG around 2010; my guess is that Berners-Lee used that instead of an average over all cars on the road due to a lack of data about car ages. We’ll do the same and use 25.1 MPG.

Berners-Lee links 22.4 miles per gallon to emissions of 850 grams per mile. As noted above, he says that about 40% of that is from manufacture and maintenance of the car, and the remainder from the fuel. Assuming that ratio stays the same, we can scale down the emissions figures to adjust for the MPG; at 25.1 MPG we get emissions of 760 grams per mile, of which about 300 grams per mile from manufacture and maintenance and the remaining 460 grams per mile from the fuel. Note the same EPA page suggests only 353 grams per mile for fuel, probably because they aren’t accounting for the emissions from processing and transporting fuel.

We’ll stick with 760 grams per mile, which is 0.42 kg per km.

Distance Driven

The Bureau of Transportation Statistics estimates that in 2018 2.23 trillion miles were driven in ‘light duty vehicles with short wheel base’, which includes ‘passenger cars, light trucks, vans and SUCs’. (A small proportion of that may be driving for business, which should be counted via embedded emissions in goods and services, but I can’t find the data to make that correction.)

The US Census gives a population of 328 million in mid-2019. That’s works out at 6,800 vehicle-miles per person per year, or 10,900 vehicle-km per person per year. Unsurprisingly this is much larger than the UK figure of 6,700 vehicle-km per person per year as the US is huge and things are spread out. As before we’ll attribute that to the driver; it works out at about 19 miles driven per day.

Emissions

10,900 km times 0.42 kg per km gives average annual emissions of 4.6 tons per person per year, which warms the Little Planet by 0.48 °C in 2100.

In the main article, I’m going to round the figures to 20 miles per driver per day and 0.5 °C of warming, to make it easier for people to follow the arithmetic.

Using the 40:60 rule, we get:

• 1.8 tons per person per year embedded in the new cars people buy, warming the Little Planet by 0.19 °C in 2100.
• 2.7 tons per person per year from driving those cars, warming the Little Planet by 0.29 °C in 2100.

Reducing Emissions

Driving Less

One way to emit less is simply to drive less; if you drive 20% less, you’ll reduce your emissions by 20% or so. (It’s not quite exact, because you might eventually have to replace your car because it’s old, but I think that’s a minor effect.) So driving 20% less would cool the Little Planet in 2100 by 0.05°C (average UK driver) or 0.10°C (average US driver).

Switching Car

Another way is to switch to a more efficient car. Smaller cars can be made with fewer emissions and emit less while driving. The most efficient car is, of course, an electric car. For the UK, Berners-Lee gives the following emissions figures for driving a mile, inclusive of vehicle manufacturing emissions:

• 180g in a mid-sized five-door electric car.
• 530g in an average UK car at 36 miles per gallon.

I find it hard to square these numbers with the 40:60 rule, which would tell us that 40% of that 530g, or 212g, is from the manufacture & maintenance of that average car. Berners-Lee notes that making an electric car emits more than an equally sized petrol car because the battery is carbon-intensive, so I can’t see why the overall electric car figure is lower than 212g.

Let’s run a sanity check by looking at the emissions from the electricity used. This suggests an electric car will travel around 3.5 miles on one kWh. Berners-Lee notes that that kWh involves 340g from the UK grid. So we’re talking roughly 100g/mile (UK) just from the electricity. Berners-Lee also notes that manufacture of a Renault Zoe electric car works out at 105–110g/mile if it’s driven 100,000 miles. So a total of 180g/mile does make sense if you assume the car is driven a fair bit more than 100,000 miles or so in its lifetime.

This is confusing. Let’s check car sizes. A Renault Zoe weighs about 1.5 tons; this (not UK specific) site notes that an average car weighs 1.4 tons. So that’s not it.

It might be this (from the 2011 edition):

Big, expensive new cars have more of their embodied emissions attributable to each mile of driving. An older car that is still fairly efficient could beat a new Fiat 500 by virtue of having had its embodied footprint written off.

I don’t understand what’s going on here. And unfortunately, I can’t cross-check this stuff because most of the research out there uses lifecycle analysis rather than the input-output method. Lifecycle analysis sucks. It often misses half of emissions or more.

It is clear that switching to an electric car reduces emissions to around a third to two-thirds of a non-electric car, including the embedded emissions. This suggests a figure of a half, and even if it’s based on lifecycle analysis I’m going to go with that.

Driving Differently

Driving more slowly and avoiding stops and starts can also reduce emissions. Berners-Lee notes that driving on a congested road can triple your emissions, and that driving at 60mph saves 15% compared to driving at 70mph. And finally, emissions per mile driven are barely affected by the number of people in the vehicle, so you can save a lot by driving with more passengers. For four people, Berners-Lee notes that taking a train emits very slightly more than taking a small efficient petrol car.

There are many, many articles telling you how fast emissions need to decrease to limit warming to 2° C in 2100. This one’s different. I’ll show you how emissions are likely to decrease in developed countries, using both expert projections and historical trends.

I’ll be looking at two countries:

• The UK, because it has reduced emissions more than any other developed country. Looking at the UK tells us a lot about how emissions are reduced, and how other countries are likely to proceed in future.
• The US, because it accounts for such a large proportion of world emissions.

Throughout this post, I’ll use ‘ton of emissions’ to mean 1000kg of CO₂-equivalent greenhouse gases. See this post for more details.

The UK

Projections

Let’s start with some government projections…

Note the y-axis doesn’t include zero. Also that shows territorial emissions; that means emissions for any items produced in the UK, and excludes a) all imports and b) air travel.

Emissions were around 650 megatons in 2008 and 470 megatons in 2016. That’s a 28% drop in 8 years, or 4% a year, albeit in a period including the financial crisis. Going forwards, reductions slow substantially, going from 470 megatons in 2016 to about 360 megatons in 2035. That’s 23% in 19 years, or 1.4% a year. (The 95% confidence interval is 1.1%-1.8% a year.)

So that data says we’ve managed 4% a year, but will only manage 1%–2% a year going forwards. Why the sudden slowdown? Well… have a look at the black lines in each of these two graphs…

The left graph shows you how electricity has been generated over time; the black line is for coal. Now look at the right graph; the black line shows the total emissions produced in the UK. Both black lines drop sharply around 2010–2015: as renewables let us retire coal plants, emissions dropped very rapidly. What about the future? You can see from the left hand graph there’s very little coal left, so there’s no easy source of emissions reductions left either.

You can see the overall effect of the transition clearly in this graph showing emissions from different sectors, with electricity in red:

Technology adoption

We’d expect similar effects to happen in the future. At some point electric vehicles, storage technology, heat pumps, and other technologies will go mainsteam, following adoption curves like those in this image:

That’s a century worth of technology; each curve shows you how one particular technology spreads. Typically adoption is slow until about 10% of a population is using a technology, then very fast until 80% or so, then slow again. The next graph shows how different countries are progressing along the electric-car adoption curve:

The UK isn’t marked on that curve, but is near the bottom: in 2019, 3.3% of new registrations were battery electric vehicles. In the future we can expect the UK to

1. Move slowly up to the point G; in this phase emissions reduction will be small as few petrol cars are being replaced.
2. Move rapidly up the curve until electric cars account for about 80% of cars on the road. During this rapid adoption phase, emissions will drop rapidly as petrol cars go off the road.
3. Once about 80% of new registrations are electric, emissions will go back to declining slowly, as number of petrol cars isn’t dropping that fast.

All in all, this gives us a rapid burst of emissions reductions. We saw the same kind of burst when renewables replaced coal, and we’d expect similar bursts when technologies like large-scale electricity storage and heat pumps go mainstream.

So: emissions won’t actually drop at a steady rate. They will lurch downwards, dropping quickly when a new technology is mid-adoption and slowly the rest of the time. Each new technology will take a quick ‘bite’ out of emissions in some sector, just as renewables did in the electricity sector (red line below):

As emissions are reduced, it will become harder and harder to reduce them; emissions from air travel, for example, are very hard to reduce. So the absolute pace of emissions reductions, in megatons per year,* will slow down. We’ll come back to this later.

*I lied a little here. Emissions are measured in megatons per year, so the rate of change of emissions is measured in (megatons per year) per year. Don’t worry about it if it’s confusing.

Modelling future reductions

How should we model these fast-slow-fast-slow emissions reductions? Looking at each of electric cars, heat pumps, etc., in turn and figuring out when they’ll become mainstream is an enormous task, and hard to do with any accuracy.

Thankfully, economics gives us a better method. Per-capita economic growth happens as a result of many technological improvements.* As we saw in the 20th-century graph above, each technology is adopted in an S curve, so each technology gives a burst of economic growth. But in the long run, all of these average out to give fairly steady trend growth.

*If you’re an economist: yes, I should be talking about total factor productivity here if it weren’t too technical. Well done. Have a lolly.

We’ll use the same idea here, and use the average annual emissions drop over a long period to estimate the future trend. 1990 is the usual reference point for emissions reductions, so we’ll start there. Using data from here and here, we get an average drop of 1.76% between 1990 and 2035. What would that look like if we extrapolated it out to 2100?

I noted above that the more we reduce emissions, the harder it is to keep reducing them. If you like, we grab the low-hanging fruit first, so progress slows down as there are fewer, higher fruit left. One of the advantages of modelling reduction as a fixed percentage per year is that it captures this slowdown in reduction: you can see that the red line above becomes flatter and flatter over time.

Politics vs the Facts

The UK recently announced a new target of net zero emissions by 2050. You can see that the red line in my projection graph above doesn’t come anywhere close to that. In fact, the old UK target for 2050 was 120 megatons of emissions, and the UK is not remotely on track for that. The UK is even due to miss its goals for 2030. Even those 2030 goals are misleading because they exclude aviation emissions for UK flights, which have doubled since 1990 and are still rising.

What this tell us is that when governments set aspirational multi-decade goals, even ‘legally binding’ ones, those are political theatre. Another example is China — the central government has committed to net zero emissions by 2060 but its regional development plans include substantial construction of coal plants over the next decade. If we want to know how fast emissions are likely to decrease, we need to look at policies rather than goals.

Negative Emissions

Aspirational goals generally rely on negative emissions technologies, technologies that remove greenhouse gases from the atmosphere. I wrote a detailed review of the research on direct air capture, which tells us that it’s not realistic to expect it to reduce emissions by a significant amount. The same unfortunately applies to other negative emissions technologies. The research shows that they all have capacity constraints that mean their contribution will be quite limited. For example, to make a substantial dent in emissions by growing plants, we’d need to plant over agricultural land the size of India, causing massive food shortages.

I have yet to see any realistic negative emissions plans — typically proponents show that something could work on a small scale, and then assume everything be scaled up arbitrarily. (In economics lingo, they assume that supplies of their inputs are ‘perfectly elastic’.) As thing stand, reaching net zero carbon before 2100 is an unrealistic, techno-utopian goal; we might be lucky with some as yet undreamt of scheme, but it’s unlikely.

Correcting for Population

There is one reason to think emissions might drop faster than the red line in my graph above. At the moment, the UK population is rising relatively fast; by 2100, it will be rising much more slowly (if at all), as the next graph shows:

More people means more travel, more buildings, more food, and generally more emissions. And emissions-reduction technologies generally act by reducing emissions per person. For example, switching to electric cars reduces emissions per driver; if the number of drivers were to go up fast enough, overall vehicle emissions could still go up.

So we should get a better picture of the effect of technology by thinking of emissions per person dropping by a fixed amount per year. Before we do that, though, there’s one other improvement we can make…

Correcting for Consumption

Here’s another graph from the UK Dept. for Business and Other Gubbins:

Like all the graphs we looked at earlier, this shows a pretty substantial decline in emissions. That picture isn’t wrong, exactly, but it’s very misleading. Over time, more and more emissions-intensive manufacturing has shifted overseas — but people in the UK still buy things from abroad, which emitted greenhouse gases when they were made.

If you look at the UK’s consumption emissions, that is the emissions in all the things bought by people in the UK, you get a much less optimistic picture:

Between 1990 and 2017, there’s been an increasing gap between CO₂ emissions produced in the UK and CO₂ emissions consumed in the UK, showing that people in the UK have been importing more and more emissions-heavy goods. (I wish I could find consumption statistics for all greenhouse gases, but I can’t, so I’m using the CO₂ figures. For long-terms trends that should be fine as 75% of warming is from CO₂.)

That gap has made the production-based statistics the government likes to quote look very good, but it’s a little misleading; ‘exporting’ emissions to other countries is zero-sum and doesn’t actually global warming. It’s also not possible to keep exporting more and more emissions. In that respect consumption statistics will give us a better measure of long-run trends.

Putting it all together

Using data from here, we find consumption emissions per person dropped by 18% from 1990 to 2017, which is 1.2% a year.

I can’t find projections of consumption emissions, but as I just noted they can’t keep diverging from production emissions, so we can use the production projections from the start of this article. Correcting those for population, we get a projected drop of 17% from 2017 to 2035, which is 1.7% a year.

The US

We’ve covered all the key ideas while discussing the UK, so this discussion of the US will just focus on numbers. The ‘bible of energy projections’, which everyone cites, is from the independent Energy Information Administration’s Annual Energy Outlook 2020 (AEO2020):

This doesn’t cover all emissions (and I’ll confess I don’t understand what’s left out), but it covers most of the ~6.5 billion tons of current emissions listed by the EPA. So it’s a decent source for trends.

The central black Reference line in the graph is the case the AEO2020 considers most likely to happen, given current policies and expected technological advances. The Low Renewables Cost line is for the case where the prices of solar panels and wind turbines drop faster than expected.

If you’re surprised by how flat those emissions lines are, there are quite a few reasons:

• “Costs for renewables such as wind and solar have continued to decline as experience is gained with more builds. How long these high cost reduction rates can be sustained is highly uncertain.”
• As renewables technology improves, the cost of electricity will drop; as it drops and also incomes go up, people will use more electricity.
• As electricity becomes cheaper, some coal plants will become uneconomic and close. But there’s a core of coal plants that are inexpensive to run, which are projected to stay in use until 2050.
• Wind and solar aren’t always available. Although the AEO2020 models the growth of storage (batteries), a lot of gas will still be needed.
• The graph is for the US as a whole; the population will grow a lot by 2050. Emissions per person are projected to drop from 15.0 tons per person in 2020 to 12.7 tons per person in 2050 (Reference) or 12.1 tons per person in 2050 (Low Renewables Cost).

The AEO may be the bible of energy projections, but it has been criticised for consistently overestimating wind and solar costs. On the other hand, this will mostly affect emissions from electricity generation, which only account for about 20% of US emissions:

In case you’re wondering, the transportation projection does take into account growth in electric cars and increasing fuel efficiency; they’re just balanced out by economic growth and increasing air travel. The projection doesn’t take into account the fact that high-altitude emissions cause about 1.9x as much global warming — so the actual effect of transportation will be even higher.

Also, as the green bars in the next chart show, the scope for use of renewable power in industry is relatively small.

So while emissions may drop faster than the AEO projects, the difference is unlikely to be dramatic; even if emissions from electricity generation halve, that should only affect the final figure by 10%. Still, it would be good to check against another objective source. The AEO2020 so ubiquitous that I can’t find any other projections, so let’s try looking at historical data instead. Here are US emissions:

The sharp drop around 2009 is due to the financial crisis, so we don’t want to read too much into it; we either want to exclude it or take a wide window around it. I’ll do the latter. Using data from here, I find per-capita US consumption emissions decreased 14% from 1990 to 2017, or 0.54% a year.

How does that compare to the AEO2020? That projects an annual drop of 0.55% (Reference case) or 0.71% (Low Renewables Cost). The historical data is at least in the same ballpark, which is good.

Incidentally, if the US had stayed in the Paris Agreement, it would have committed to reducing its (production) emissions from by 80% from 2005 to 2050. That’s a 3.5% reduction each year; you can see that both projections and historical data fall very far from that lofty goal.* One major reason for that is that increases in efficiency are counterbalanced by GDP growth, as the graph below shows. To put that more simply, as time passes, each thing we consume involves less emissions, but we consume more things.

*Per-capita US consumption emissions decreased 21% from 2005 to 2017, or 1.97% a year. But much of that drop was due to the financial crisis, so it’s not a great basis for extrapolation.

Putting it all together

Here’s a quick recap of all the per-capita emissions figures we’ve found…

• UK consumption emissions dropped 1.2% a year from 1990 to 2017.
• UK emissions are projected to drop 1.7% a year from 2017 to 2035.
• US consumption emissions dropped 0.54% a year from 1990 to 2017.
• US emissions are projected to drop 0.71% a year from 2019 to 2050 under the AEO2020’s Low Renewables Cost scenario.

Any attempts to project emissions to 2100 are inherently uncertain, but the best estimates we can make using history and science will rely on those numbers.

Divergence

The most obvious fact that jumps out from those numbers is that emissions have decreased much faster in the UK than in the US; UK and US emissions have diverged. You can see that divergence between the USA and UK lines in this next graph:

In the graph, the divergence is represented by the UK and the US moving further apart over time. We need to be careful when extrapolating because this divergence is very unlikely to keep happening all the way to 2100. One reason for that is that because the UK has made more progress to date, it’s now harder for it to reduce emissions; the low-hanging fruit are gone.

Another way of looking at things is that in the long run, emissions decreases are driven by technologies like electric vehicles; adoption in the US may be slow right now, but it will get happen eventually. You see something similar with GDP per capita; countries that are behind tend to grow faster, in a ‘catch-up effect’, because it’s easier to copy technologies than invent them from scratch.

So in the long run, the USA and UK emissions lines are likely to drift down at roughly the same rate; perhaps USA emissions will always be above UK ones, but they won’t get further and further apart.

Projections

This means it would be a mistake to use our UK and US numbers to project emissions for each country separately; i.e. we shouldn’t estimate that per-capita emissions to 2100 will decrease by 1.7% a year in the UK and 0.7% a year in the US.

A better model is to estimate that in the long run, emissions will decrease at roughly the same rate in all developed countries. The reason for this is that the long-run rate of decrease will mostly depend on green technologies, just as long-run economic growth depends mostly on technological progress, and green technologies should eventually spread to all countries.

You can see from the last graph that the UK has reduced emissions faster than other developed countries, while the US has been slower than any country except Canada. That tells us that the US and UK are good choices to bracket the projected rate of emissions decrease in developed countries; it’s unlikely to be faster than the UK or slower than the US.

So —deep breath — our best estimate is that emissions per person will decrease between 0.5% a year and 2% a year in developed countries. It’s possible that there will be some miraculous technological breakthrough that results in faster reductions, and it’s possible that major wars or other geopolitical crises will displace climate cooperation and result in slower reductions or increases. But the range that’s supported by past data and expert projections is 0.5% to 2% a year.

There are a couple of reasons to think 0.5% is pessimistic. First, the AEO has been consistently bearish on renewables and it’s possible that even their Low Renewables Cost scenario is too pessimistic. Second, all the technologies that have been used in the UK will be adopted in the US eventually, so we’d expect the US to follow the faster UK trajectory sooner or later. Given those points, I think I’ll go out on a limb and make a…

…final prediction!

In developed countries, the average annual per-capita emissions reductions to 2100 will be between 1% and 2%.

Let’s see what that actually looks like. Using population projections from here, I put together these graphs showing total emissions for each country.

You can see that there’s a big difference between annual reductions of 1% and 2%. But even a reduction of 2% a year doesn’t bring us anywhere near ‘net zero carbon’ by 2100.

Ponies

Suppose I’m a politician, and I have to pacify lots of people who are upset with me over climate change. I have two options. I could implement policies that substantially reduce emissions now. That will upset lots of other people (maybe even some of the same people) by driving up prices, etc., costing me a lot of political support. Alternatively…

I could promise everyone a pony in 2050.

Of course, no one is actually going to get a pony in 2050. But that’s okay; I’m not sure I’ll be around, and even if I am I can always blame the next government for not being sufficiently committed to ponies. And I don’t have to implement any costly policies now, so I really don’t have anything to lose. Free ponies all around! And in 2100 you can have a unicorn!

If there’s one thing you should take away from this article, it’s that net zero carbon is a convenient political slogan, rather than a likely outcome. The UK’s old target, of a 80% reduction from 1990 to 2050, or 2.6% a year, was ambitious but just about plausible. When the UK was on track to miss that goal and promised zero carbon by 2050, that was just political theatre. In the same vein, governments overstate progress by using production emissions, excluding aviation and doing other not-quite dishonest things with the statistics, because there’s little cost to doing so.

The Inside and Outside Perspectives

I should say that I hope nothing I’ve said causes a sense of fatalism. I started this article I’ve taken an ‘outside’ perspective in this article, treating governments as objects of study, because I needed an objective estimate of likely emissions reductions for an upcoming project. But many readers will be taking an ‘inside’ perspective, lobbying to shift government policies. I certainly don’t mean to imply that that can’t have an impact; if the US had followed the UK’s trajectory, we would be on track for a substantially lower temperature in 2100. So please don’t stop on my account. Just… please, please look at what they are doing rather than what they are promising.

You can save more lives without any economic hit, just by changing the timing of restrictions and lockdowns.

THERE is a surprising connection between COVID and the economy — and it’s not what you think…

In the 1960s, economists thought they had found a trade-off between jobs and inflation. By letting inflation run higher, you could create more jobs; if you pushed inflation down, there would be fewer jobs.

‘Aha!’ they said, ‘inflation is bad, but unemployment is much worse. Let’s turn up inflation and create more jobs.’

What happened? Well…

The 1970s was a pretty dismal period, and not just because of polyester and disco.

It turned out the economists were quite wrong. Keeping more people in jobs didn’t just push inflation up; it made inflation go up faster and faster and faster.

It’s as if you pressed down a little on the gas pedal and, rather than just speeding up to a comfortable 50 miles an hour, your car sped up to 70 and then 90 and then 110 and showed absolutely no sign of stopping. When you panicked and eased back up, it would stay at 110.

It turns out your gas pedal doesn’t control your speed; it controls whether you’re speeding up or slowing down. Driving in this imaginary world would be very different; there would be one magic pedal position which would keep you at a level speed, and you’d be trying to keep the pedal at that position whether you wanted to go at a steady 20 or a steady 70.

That’s actually how the economy is handled nowadays. There’s one magic unemployment rate (gas pedal position) that keeps inflation (car speed) steady, and central banks (drivers) try to keep the unemployment rate at that position all the time. If they tried to keep unemployment low all the time, they’d be back to runaway inflation (your car going faster and faster):

Now, what does any of this have to do with COVID-19? Well, you often hear that lockdowns trade lives against livelihoods, health against the economy. How does that work? Well, the story goes, a lockdown clearly means an economic hit, but it also lowers the reproduction number R, helping us control this %#&@* virus. Or, if the economic hit is too painful, we keep R a little high, saving the economy.

Here’s the problem with that. R doesn’t tell us how bad the virus is. It tells us whether it’s getting better or worse. An R of 1.2 isn’t like a car that’s speeding; it’s like a car that’s getting faster and faster and faster. We can’t keep the economy open all the time in return for a higher death rate; we can only do it in return for a higher and higher and higher death rate. Like the 1970s economists, we’re chasing a trade-off that’s not actually there.

If you’ve read this far, you might think I’m arguing that we need to damage the economy to save lives. I’m not. I’m really not! I’m about to show you that the maths shows we can get something for nothing here.

I’ll need a toy mathematical model to show you how this works. To keep things easy to understand, let’s assume that we have two kind of weeks:

• Lockdown weeks, where the economy suffers but the death rate halves.
• Open weeks, where the economy does great but the death rate doubles.

The UK policy has been to keep a region open until COVID is widespread, and then lock down. Let’s picture this as 7 weeks of open economy followed by 2 weeks of lockdown, that is, 7 weeks of doubling followed by 2 weeks of halving. The number of deaths in each week will look like this:

Now comes the key point. What happens if we put in the same two weeks of lockdown — the same hit to the economy — but we put it in earlier? Let’s say we put the two-week lockdown in after four weeks of doubling. Here’s what we see:

You can see that we have far fewer deaths. At the same time, we haven’t made the economy suffer more; we’ve had the same amount of lockdown, just earlier. In fact, the economy is probably doing better, because people will be less nervous about leaving their homes.

In case you’re worrying about what happens afterwards, here are similar graphs extended out to 20 weeks. They represent:

• Left Graph. The government starts a two-week lockdown when there are more than 100 deaths in a week.
• Right Graph. The government starts a two-week lockdown when more than 10 deaths in a week, except right at the end when a vaccine is near; then they let the number of deaths go up to 100.

In both cases we have six weeks of lockdown (six blue columns), so the same economic hit. But in the case shown on the right we have far fewer deaths.

By the way, don’t read too much into the term ‘lockdown’ — I just mean ‘some policies that reduce R’. You can draw little red-and-blue graphs for any policies you like, and you’ll always find that having restrictions and lockdowns earlier saves more lives.

tl;dr

You can save more lives without any economic hit, just by changing the timing of restrictions and lockdowns.

I’ve written a post showing you how individual choices affect global warming. You should read that instead. This is a dry, technical supporting post showing you where the numbers come from. If you read it, please read it after my post on methods. For reasons given there, I’ll be relying a lot on a book by Mike Berners-Lee for emission intensities of actions, although every figure will be cross-checked against other sources.

Recycling

Transporting, sorting and processing recycled things all emit greenhouse gases; those have to be balanced against the emissions from creating new things. So it takes careful analysis to tell whether recycling actually reduces emissions.

Unfortunately, the research on net emissions from recycling is so sparse as to be frustrating. I’ve hunted hard and only found one paper using input-output analysis. It has the following wonderful quote:

It is important to remember that the recycling of waste materials is not an aim in itself, but a means of reducing the overall environmental load resulting from waste emission, and that any production and consumption activities, including recycling and waste treatment, also emit waste.

This is a simple point, but it’s hard to convey to people, because recycling feels so ‘green’. Before diving into the paper, I want to do some back-of-the-envelope calculations for the UK, to give us a sense of scale.

An upper bound: what can be recycled?

The graph below is from Berners-Lee, and explains how emissions per person break down in the UK. Have a read of the labels. As far as I can see, the only things that can be recycled by individuals are ‘other non-food shopping’ (7% of emissions) and possibly the packaging from ‘food’ (25%).

Let’s deal with the food packaging first. As a very, very rough rule of thumb, most things have a few kg of emissions per kg of material. (You can check that for both food and plastics by looking at Berners-Lee.) This means, very roughly, that the emissions from producing things are in proportion to their weights. Food packaging usually weights a lot less than food — think about the weight of your food shopping if you took out all the food — so we’d expect packaging emissions to be much less than food emissions.

And indeed, Berners-Lee’s emissions figures for takeaways say 2.5% come from clear plastic trays. Combining that with the 25% of emissions from food, I’d expect food packaging to be less than 1% of our emissions.

That means food packaging and non-food shopping should come in at under 8% of our emissions. That’s the most we could hope to save by recycling. That’s to say, if:

• We could recycle all non-food shopping
• The recycling process had no emissions
• Each 1 kg of recycled stuff displaced 1 kg of new stuff we would have made

then we could reduce our emissions by 8%. We could say 8% is an upper bound on possible emission reductions. In reality the figure will be significantly lower, as none of those assumptions will be true.

Packaging

Let’s try to get at this another way. I’d say that a lot of what we recycle is packaging. This estimates 2.4 million tons of plastic packaging was placed on the UK market in 2017. That’s about 36kg per person. Here’s Berners-Lee on the emission intensities of different plastics:

At about 4kg of emissions per 1kg of packaging, that’s around 140 kg of emissions from plastic packaging. Compared to the annual UK emissions of 13 tons per person, we see that about 0.1% of our emissions come from plastic packaging. So there is no way we can save more than that by recycling plastic packaging; indeed, emissions during the recycling process, etc., mean the actual figure will be lower.

What about other kinds of packaging? This table gives a breakdown:

Berners-Lee says printed virgin paper works out at 2.5–3kg of emissions per 1kg of paper. Working through, we find that’s about 200kg of emissions per person, or about 0.15% of our emissions for paper packaging. For steel he estimates 1.8 kg of emissions per 1kg of steel, which is negligible at only 0.01%. (Remember that’s for steel packaging. Steel in other things is a lot more.)

For wood, aluminium and glass Berners-Lee doesn’t give figures. This is just a rough calculation, so I’ll use Wikipedia to get emissions per 1kg material of 0.46 for wood, 8.24 for aluminium and 0.85 for glass. Plugging those in I find wood packaging is negligible and aluminium and glass packaging are each around 0.02% of emissions.

I was surprised that the aluminium figure was so low, so I obtained independent estimates of 207,000 tons of packaging times 13.5 tons of emissions per ton, giving 0.03% of emissions. Truncation error during lifecycle analysis might mean that figure is out by a factor of 2, but it’ll still be a tiny percentage.

So all in all, packaging looks like it only accounts for roughly 0.3% of our emissions. Again, that’s an upper bound on what we could save by recycling packaging.

The Research

As I noted, there’s really not much research here. This paper uses lifecycle analysis to determine the net emissions from recycling of source-segregated materials, that is, recycling that has been presorted into plastics, paper, etc.. Here’s the key table:

The dark grey bars on the left are emissions saved by not manufacturing new materials; the light grey bars on the right are emissions from the recycling process. It looks like on average recycling saves 1–2 tons of emissions/ton of recycled material. (Metals are much higher but are a small proportion, so it averages out.) We’ll use that fact later. Here’s a close-up of an interesting part:

Look at paper on the fifth line. You can see that the dark grey bar (savings) is a little longer than the light grey (emissions). So it looks like in total, recycling paper is a good way to reduce emissions. Looks is the key word here. There are two problems:

• Lifecycle analysis is notorious for missing as much as half of emissions. Both the dark grey and light grey bars could have quite different lengths. For many of those items (e.g. paper), it wouldn’t take much mismeasurement for the light grey bar to be the longer one; in that case, recycling would increase emissions.
• Single-stream recycling is the norm in the US and the UK. That is, paper, plastics, etc., are all mixed together. Separating these takes work! And that means more emissions, which are not accounted for in that table. Again, that means that recycling might increase emissions in many cases.

As an aside, that graph also shows that emissions from transporting recycling are negligible; we don’t need to consider them when deciding whether to recycle. There’s a similar graph in Berners-Lee that shows that landfill emissions aren’t a significant factor.

It’s really just the emissions from avoided production that have a significant effect.

I said at the start of this article that there was just one paper that analysed recycling using the robust Leontief input-output method. It looks at Japan, where:

The Japanese government has recently enacted a series of new laws to promote the recycling of used containers and packaging materials, discarded electrical home appliances and construction waste, among others.

It finds that a ‘proper combination of waste disposal and recycling processes’ can reduce CO₂ emissions by up to 2.1%. Given our discussions earlier in this article, it shouldn’t be surprising to see a figure this low.

As I said, there are no other input-output analyses out there, but we should still sanity check 2.1% against something. This analysis says that recycling waste can save 0.9 gigatons of global emissions in 2030, out of 70 gigatons total. That’s 1.3%. Significantly smaller, but same order of magnitude.

So, should we go with 2%? Not quite; there’s one more issue. All these numbers gauging potential emission reductions from recycling have been considering all the waste a society generates. That’s waste from households and from industry. But the individual actions we’re interested in only reduce household waste.

How much of a difference does this make?

Estimates vary a bit, but that source estimates municipal waste is only 14% of overall waste. (Also household waste is only a part of municipal waste, but we’ll assume you can recycle at work if you want to.)

I must admit to being somewhat frustrated at this point. The lack of reliable research makes this whole exercise like playing pin-the-tail-on-the-donkey. I’m going to give you a couple of datapoints and then a guess.

• The Japanese paper includes an analysis case excluding construction waste, and only finds emissions reductions of 0.7%. The details of that case are quite different to ours, though.
• This estimates 2 kilograms (4.51 pounds) of municipal waste per person in the US. That’s about 0.7 tons a year. We estimated earlier that recycling saves around 1–2 tons of emissions per ton of recycled material, so 0.7–1.4 tons a year. That’s 0.3%–0.7% of emissions for an average person in the US.
• We saw above that municipal waste is around 14% of total waste. Earlier we estimated all recycling, including industrial, could reduce emissions by 2%. Combining these suggests recycling municipal waste could reduce emissions by about 0.3%.

I’m going to grit my teeth and go with 0.5%. The real figure could be a lot lower or a little higher; it seems very unlikely that it would be more than 1%, though.

Combining these with Berners-Lee’s figures, we find these figures for 100% recycling:

• An average person in the US could reduce emissions by about 0.11 tons a year by recycling everything possible; that reduces 2100 temperatures on the Little Planet by about 0.012 °C.
• An average person in the UK could reduce emissions by about 0.065 tons a year by recycling everything possible; reduces 2100 temperatures on the Little Planet by about 0.007 °C.

To put that in context, the UK figure is very close to the emissions from an annual return trip from London to Glasgow. By train.

Beyond Arithmetic

All the research I can find considers recycling being done locally, with proper procedures. That’s not what happens in the real world. A lot of recycling is exported to developing countries. Until recently, China was the main importer, but it banned waste imports in 2018. Here are the main destinations for US plastic recycling:

The article that’s from says 20% to 70% of plastic entering recycling facilities around the globe is discarded because it is unusable. But discarded waste doesn’t just disappear; it has to go somewhere. The countries that import recycling generally have much worse waste management practices than in the US or EU. I used this data to find the proportion of waste that’s inadequately managed in each country in the image above:

What might it mean if your plastic recycling is inadequately managed? Well, it might be incinerated on open rubbish dumps, releasing toxic fumes. Or it might end up in the ocean; you can see Bangladesh, Thailand, Vietnam, the Philippines, Malaysia and Indonesia in the image below.

You can of course look into your own local recycling operation; if you’re lucky and everything is done locally, you don’t need to worry about this. If not, you’re likely to do more harm than good by recycling plastics. (I wrote more about the impact of plastic bags here if you’re interested.)

I can’t find good sources on other kinds of recycling. It’s probably safe to recycle things with high metal content, as those are valuable and likely to be recovered wherever they end up, but I’d check your local operation before recycling anything else.

I’ve written a post showing you how individual choices affect global warming. You should read that instead. This is a dry, technical supporting post showing you where the numbers come from. If you read it, please read it after my post on methods. For reasons given there, I’ll be relying a lot on a book by Mike Berners-Lee for emission intensities of actions, although every figure will be cross-checked against other sources.

Air travel

Emissions per passenger-km

Berners-Lee estimates 3.5 tons of emissions for a return flight in economy class from London to Hong Kong (19,300km). That works out at 181 grams of emissions per passenger-kilometre.

This estimate falls inside the range of 80–218 grams given by the IPCC for medium to long haul flights. The reason it’s towards the high side of the range is likely that Berners-Lee applies a factor of 1.9x to adjust for a well known effect whereby high-altitude emissions have a larger effect on global warming.

The UK’s official GHG conversion factors give another useful data point. They estimate 103 grams per passenger-km for long-haul flights; once the 1.9x factor is applied this again seems very consistent. Short-haul flights are 79 grams and domestic flights 145 grams. The figures for different types of flights are close enough that I don’t feel it’s useful to split out categories.

Having cross-checked it, I’m happy with 181 grams of emissions per passenger-kilometre in economy.

The UK

Berners-Lee estimates that in the UK personal flights account for about 8% of 13 tons, or 1.04 tons of emissions a year. Using the conversion factor from the methods article, cutting out such a flight each year would reduce the 2100 temperature on the Little Planet by 0.11 °C.

Now to cross-check that. Estimating how far the average person in the UK/US flies is hard, because international flights are not included in any country’s emission statistics. This 2013 study estimates that the average UK household emits 1.13 tons of CO₂ from flights a year. (That’s the mean; the median is 0! Only 41% of households fly.) ONS gives the average household size as 2.4, so that’s about 0.47 tons per person per year. Applying Berners-Lee’s 1.9x correction factor, we get 0.89 tons per person per year. Flights have been increasing rapidly since 2013, so that seems quite consistent with Berners-Lee’s estimate.

Combining our figures of 1.04 tons of emissions and 181 grams of emissions per passenger-kilometre, we find that the average person in the UK flies 5,750km a year.

That’s a pretty misleading statistic, though; as mentioned above, only 41% of UK households fly at all in any given year. Assuming that about 41% of people fly, we can split that 5,750km and say that in any given year:

• About 59% of people in the UK don’t fly at all.
• About 41% of people in the UK fly on average 14,000km a year.

The flight distance from London to Miami is about 7,100km, or 14,200km for a return flight. Close enough to 14,000km that I’ll use it in the main article. Combining 14,200km with Berners-Lee’s 181g/km estimate, that return flight (in economy class) emits 2.6 tons.* Applying the conversion factor from the methods article, we find that cutting out such a flight each year reduces the 2100 temperature on the Little Planet by 0.27 °C.

As you can see in the table above, Business class generates about 222.1/76.6 = 2.9x the emissions of Economy. So a business class return trip from London to Miami emits 7.5 tons; cutting it out reduces 2100 warming on the Little Planet by 0.79°C.

The US

The Energy Information Administration’s Annual Energy Outlook 2020 (AEO2020) says that air travel accounts for 1,020 billion passenger miles in 2020.

The AEO2020 isn’t clear on whether or not it includes international flights, but the St Louis Fed’s economic data (above) shows seasonally adjusted ‘U.S. Air Carrier Domestic and International, Scheduled Passenger Flights’ were heading towards 90 billion passenger miles a month before COVID-19. That’s 1,080 billion passenger miles, so I’m confident that the AEO2020 figure a) includes international flights and b) isn’t skewed by COVID-19.

Converting to SI units, the AEO2020 figure becomes 1,640 billion passenger km. A US Census webpage gives 2020 population at 330 million. Dividing, we find the average person in the US flies about 5,000 km a year, which is surprisingly close to the UK figure of 5,750 km a year.

As with the UK, that figure is misleading because many people don’t fly. This 2018 survey estimates that 48% of the US adult population flew in 2018. The 2016 version of the survey notes that nearly every adult who flew made some personal flights. Given that, I’m just about comfortable extrapolating the 48% figure to the whole US population. Using that, we find:

• About 52% of people in the US don’t fly at all.
• About 48% of people in the US fly on average 10,400km a year.

Conveniently, the flight distance from London to New York is 5,500km, or 11,000km for a return flight, which is close enough to 10,400km to use in the main article. Using the same methods as before, that return flight (in economy class) emits 2.0 tons; cutting it out each year would reduce the Little Planet’s 2100 temperature by 0.21°C. In business class, a flight would emit 5.7 tons; cutting it out each year reduces the Little Planet’s 2100 temperature by 0.60°C.

Offsetting

Standard economic theory provides a good case for letting people carbon offset their flights. The problem is that in practice it doesn’t work. I looked through the four largest US Airlines; two of them, Delta and United, let you offset your flights via their partners Conservation Coast and Conservation International. If you click on those links, you’ll see they’re running projects verified by Verra’s VCS program, which is the largest carbon offsetting program in the world.

… That’s great?

No.

Verra’s VCS program is an empty facade. ProPublica published an excellent expose about it, which I wrote about here. (I was pretty annoyed when I wrote that, so it’s a little intemperate.) To my mind, the most damning part is this:

[ProPublica’s Lisa Song] asked the CEO how the company could guarantee that the credits it continues to sell are actually helping to preserve trees if it might go as long as two decades without checking.

[Verra CEO] Antonioli said, “Anybody who is interested in buying such credits would want to do their due diligence and check to see what’s happening on the ground.”

Which is to say, Verra’s actual certification means nothing. United and Delta are ‘greenwashing’ their emissions. By misleading environmentally conscious fliers into thinking their flights have no effect, it’s likely that their offsetting schemes are increasing the number of flights and so total emissions.