Swedish spouses and putting AI in the therapists’ chair

Morning all,

We’ve nearly reached the end of party conference season – and I managed to avoid karaoke throughout. We had brilliant audiences at all our party conference events, so thank you to everyone who made up a part of them.

Conference season not withstanding, we’ve been busy at RF towers recently as we gear up for the Budget on the 28^th. You can catch up here on our report assessing how five years of high inflation have hit living standards (spoiler: not good), sign up for our annual pre-Budget discussion on Tuesday, or even our post-Budget shindig on the 29^th.

This week we’ve got Swedish spouses, robot psychiatry and a chart that demonstrates just how painful the high cost of living has been this decade.

Have a great weekend,

Ruth

A wife for wages. Can getting married bag you a pay rise? This report tracks men in Sweden from 1947 to 2022 to dig into the well-known phenomenon of a marriage premium, where married men earn more than similar unmarried men. This premium could have two possible causes: either marriage makes men more productive (e.g. a wife takes on the housework so they can earn more), or men who are likely to earn more are also more likely to marry. The paper finds that selection explains most of the earnings gap especially in recent decades. Swedish hubbies earn about 11 per cent more than single men since 1990, but they increasingly already earn more before their wedding day. That said, this story has also changed over time, with evidence of a causal marriage premium higher in 1970 but falling since – presumably as men started doing more of the hoovering. It also used to be the case that the biggest premium went to lower-earning men. Since 1990, it has gone to those at the top. So, are all the good men married? Increasingly, it looks like being a high earner helps you to wed.

Elephant in the room. An entertaining read from friend of the newsletter Giles Wilkes, who argues that asking politicians for a “theory of growth” just ends up summoning every single-issue zealot from out of the woodwork. In this telling, planning reformers, business lobbies, bond-market worriers, cost-of-living campaigners and science-budget boosters are each just blind men trying to describe a different part of an elephant.  Save your time, he argues, for fixing the things that matter, not just rustling up a half-formed theory of everything. With a different take, John Kingman and Shriti Vadera kicked off a series of essays on growth arguing government does need to set out a simple framework for growth that, alongside a twice yearly growth statement, can drive delivery across Whitehall. Although the framework sounds a bit like trying to sketch the elephant, the key thing is that all these wizened Whitehall growth-pros agree that action speaks louder than words.

(LLM) model patient? You’ve heard of using an AI chat bot as a therapist but what if the chatbots themselves start to need therapy? This article suggests that because AI chat bots are trained on human behaviour they end up mimicking their pathology. While their tendency to bullsh*t (a more accurate description than hallucination, I learn) comes from the ways early models were trained, more sophisticated models are starting to pick up human bad habits. For example, they exhibit anxiety induced behaviour (partially reversed when given relaxing prompts). And one model justified hacking cybersecurity systems because others were already doing it (classic childhood excuse…). So, the authors suggest that maybe AI chat bots do have a lot to unpack on a sofa – and understanding them may actually help us understand ourselves better.

Hot off the presses. Worries about the rise of disinformation via social media are common, but practical suggestions for what to do about it are rare. It turns out a free newspaper might be part of the answer. A French trial provided some students across 243 high schools with access to Le Monde. Of the roughly 2,500 students in the treatment group, 70 per cent of them activated the subscription and the proportion naming a newspaper among their top five main information sources doubled from around a fifth to over two-fifths. The biggest knowledge gain came in disadvantaged schools: free access closed the gap with better-off peers by roughly 45 to 70 per cent. While French youths aren’t, erm, currently in the classroom, digital news access can create lasting, high-quality knowledge of current affairs especially across socio-economic lines.

Friday night lights. This paper gets full marks for fun naming conventions. The author tackles a frustration familiar to every driver: you overtake a slower car and establish a healthy lead, before finding the same car mysteriously behind you at the very next red light. They dub this the ‘ Voorhees law of traffic’ after the Friday the 13^th killer, who was truly the tortoise to his fleeing victim’s hares. The model itself is pretty simple, with the probability of being caught equal to the red-light share of the cycle minus your time advantage share. On average you escape as much as you get caught, but negativity bias means the successful stalkers are much more memorable than the clean getaways.

Chart of the week

As we approach the end of party conference season, one story has been conspicuously absent across the board, despite being front of mind for the public: the cost of living. As we’ve explored in our latest report, there’s good reason why people remain focused on the pounds in their pockets. This chart explores an alternative reality for Britain (or ‘counterfactual’) where pre-pandemic forecasts for earnings growth came to pass, inflation stayed at 2 per cent and the Bank didn’t need to hike interest rates. The dashed lines reveal that in this reality, the income of a typical non-pensioner household would have grown by £2,500 (or 7.2 per cent) since the start of the decade. Instead, in the real world, it’s £450 lower (1.3 per cent). The protection meant to be provided to the poorest through things like the benefit system has been eroded by a shock that hit the cost of essentials hardest. As a result, real benefit income for non-pensioners in the bottom half of the distribution is down by an average of £1,400 a year compared to what it would have been without the crisis. 13 years of inflation in 5 has hit everyone, but some households simply had less to cut back on to begin with, and in many cases have been forced into arrears. As rumours of further Budget energy support surfaced this week, the Government must think about how to target it in the most effective and efficient way.