iiinova — where your digital self keeps evolving

Jonas Schöley @jschoeley.com · 26/09/2026

Incredibly proud of Ricarda Duerst @mpidr.bsky.social who defended her thesis on "Learning from past mistakes". Ricarda started with the idea that a prediction is only as good as similar predictions in the past turned out to be and applied it to demographic forecasts. hdl.handle.net/10138/637205

051

Jonas Schöley @jschoeley.com · 11/09/2026

Der Podcast lohnt sich. Hier @maxikniffka.bsky.social wie sie ihr Thema souverän in Rostock vor versammelter Gesellschaft und Politik präsentiert.

020

Jonas Schöley @jschoeley.com · 02/09/2026

Future death rates, wars, epidemics, disasters are uncertain. Does this matter for our individual uncertainty about the age of death? Not much, finds Ricarda Duerst in "The contribution of forecast uncertainty to lifespan uncertainty". Lifespan is intrinsically uncertain. doi.org/10.4054/DemR...

011

Jonas Schöley @jschoeley.com · 18/06/2026

Some of the stuff the LLM community produces is just hilarious for an outsider to witness.

011

Jonas Schöley @jschoeley.com · 21/05/2026

Fertility stats are conventionally denominated in births per *woman* - as if men have no agency. @cdudel.bsky.social & @demomapper.bsky.social hold men responsible by calculating trends in male fertility by age. Check out their male fertility collection on demoscapes: demoscapes.org/collections/...

0102

Jonas Schöley @jschoeley.com · 05/05/2026

What is the main reason for premature male mortality in Mexico? @ikashnitsky.phd and Aburto answered this question in 2019 for every Mexican state, every year since 1990 and all ages from 15 to 50. Their rich results are available on demoscapes. demoscapes.org/collections/...

093

Jonas Schöley @jschoeley.com · 30/04/2026

I don't understand this comment... On an unrelated note, have you seen this cartogram below? Give it some days and you *really* start to see.

120

Jonas Schöley @jschoeley.com · 30/04/2026

@angelacar.bsky.social et al. show a nice age effect, where men start to marry in larger numbers right around hitting age 30. This age threshold is quite stable, no matter how old men were, when they moved in with their partner. Explore in greater detail at demoscapes.org/collections/...

120

Jonas Schöley @jschoeley.com · 30/04/2026

www.ratemygithub.com/u/jschoeley So LLMs are great at roasting. I would be somewhat upset if a human made these observations and told me - They are all true, of course - But I'm not upset at all being analyzed like that by a machine. This must be why we all got so comfy being intimately surveilled.

000

Jonas Schöley @jschoeley.com · 29/04/2026

How many siblings does a recently born child in China likely have? @demography.bsky.social etal have estimated global kinship by age since 1950 and can answer the question (and a million others). Check out their results as interactive heatmaps on demoscapes.org. demoscapes.org/collections/...

0104

Jonas Schöley @jschoeley.com · 29/04/2026

demoscapes.org is a visual atlas of demographic surfaces hosted at @mpidr.bsky.social. It is also my passion project growing out of my 2016 human mortality database explorer. Did you publish something containing Lexis surfaces? Consider reaching out and having your work featured on the site.

69330

Jonas Schöley @jschoeley.com · 20/04/2026

That must be a maximum entropy estimator. If one has no data or relevant prior information Laplaces principle of indifference leads to, "meh, prolly 50/50". Good that we're kept informed on the latest whale statistics.

020

Jonas Schöley @jschoeley.com · 20/04/2026

There's a whale stranded on the German Baltic coast and the intensity of news reporting matches the crises of past years. The live news ticker borders on sarcasm.

010

Jonas Schöley @jschoeley.com · 03/03/2026

You got it. The question on our mind is to distinguish between the 3 stylized scenarios below. We've seen a and b in the data, but have not found strong evidence for c, which would be the most consequential one.

010

Jonas Schöley @jschoeley.com · 03/03/2026

"Temporary shock or lasting scar?" By 2024 nearly all high-income countries remained below their pre-pandemic life expectancy trends. We identify four distinct mortality shock patterns since 2020. Full analysis in our preprint: www.medrxiv.org/content/10.6...

7168

Jonas Schöley @jschoeley.com · 28/01/2026

1992 two optimistic takes on the future were published: Fukuyama reflected on global convergence towards liberal democracy while Lee & Carter revolutionized mortality forecasts with a model of ever lasting log-linear progress in mortality improvements. From my recent talk on Lee-Carter with crises.

020

Jonas Schöley @jschoeley.com · 09/01/2026

Author of tricolore here, the library used for color coding this map. This page shows how to create maps like this yourself: github.com/jschoeley/tr.... Here's an example for the whole of Europe. Happy for a citation to www.demographic-research.org/volumes/vol4... if you use this coding.

151

Jonas Schöley @jschoeley.com · 08/01/2026

The Geography of the Labour Force Composition in the Netherlands by Ate Poorthuis (doi.org/10.1111/tesg...). One of the best uses of the centered ternary balance scheme (www.demographic-research.org/articles/vol...) I've seen so far. Build with tricolore github.com/jschoeley/tr... @ikashnitsky.phd

12210

Jonas Schöley @jschoeley.com · 15/11/2025

Hah. My phone's gps location switched from south of Gotland to Kaliningrad while on a flight from Helsinki to Berlin. It was a clear night, so I guess stellar navigation was an option.

110

Jonas Schöley @jschoeley.com · 10/10/2025

Jim Oeppen showing latitudinal gradient in life expectancy at @hmdatabase.bsky.social symposium in Paris.

081

Jonas Schöley @jschoeley.com · 11/09/2025

Sigh... another academic project website turned into an advert for online gambling. Give me 90s html not updated in decades over this anyday.

030

Jonas Schöley @jschoeley.com · 13/08/2025

Surprisingly, mortality selection only plays a minor role in explaining the rapid drop in the risk of death following birth. Even within relatively homogeneous subgroups, mortality changes rapidly with age, suggesting an intrinsically dangerous transition period following birth.

100

Jonas Schöley @jschoeley.com · 13/08/2025

I estimated the changing distribution of mortality risk over the first year of life which has a massive right tail, so the average risk of death is a really bad estimator for the typical risk of death at any point in time. Should be enough heterogeneity for mortality selection...

100

Jonas Schöley @jschoeley.com · 13/08/2025

To make this calculation I needed to have a reasonable estimate for the variability of mortality at various stages throughout infancy. So I estimated that using birthweight, APGAR score and gestation at birth as predictors, getting a discrete mixture distribution over >200 strata.

100

Jonas Schöley @jschoeley.com · 13/08/2025

In my neonatal mortality paper I applied the Vaupel-Zhang equality to quantify how much the apparent change in the risk of death over the first days of life is merely mortality selection, and how much is due to actual changing risks in population subgroups.

100

Jonas Schöley @jschoeley.com · 13/08/2025

The left side of the equation is how steeply the risk of death changes at age x for the whole population. It is equal to the average individual level change in the risk of death at that age minus the heterogeneity/variance of the risks of death across the population at that age.

100

Jonas Schöley @jschoeley.com · 13/08/2025

Why is that? Mortality selection. Those with a higher innate risk of death tend to die earlier, leaving those with a lower risk in the population. Over time, due to this selection, the average risk of death in a population declines, even without individual level decline.

110

Jonas Schöley @jschoeley.com · 13/08/2025

The Vaupel-Zhang equality proves that we will observe a declining risk of death over infancy, even if individual risks do not change by age, as long as individuals differ in their risk of death. Jim called this "Heterogeneities Ruses".

100

Jonas Schöley @jschoeley.com · 13/08/2025

Think about infants during the first year of life. Their risk of death is highest at the day of birth, but rapidly falling afterwards. Question is: Is that only true in a statistical sense for the whole population or is it, true for individual infants as well?

100

Jonas Schöley @jschoeley.com · 13/08/2025

In statistical parlance, it's a result about the relationship between the hazard function of a mixture distribution over the positive reals, and the hazards of the mixture components. I've re-derived it in that statistical sense in the paper: www.demographic-research.org/volumes/vol5...

100

Jonas Schöley @jschoeley.com · 13/08/2025

The Vaupel-Zhang equality features in my recent publication on mortality selection and convergence in neonatal mortality. JimV mentioned that he'd like this equation inscribed on his tombstone. It's a super general consequence of unobserved heterogeneous mortality, his speciality. Let me show you:

140

Jonas Schöley @jschoeley.com · 05/08/2025

You're right. On a gestational age scale we see a continuos exponential increase in the risk of death from post- to pre-natality. This does not extrapolate to conception though. @jnobles.bsky.social knows about the very early part. epc2024.eaps.nl/uploads/241135

010

Jonas Schöley @jschoeley.com · 05/08/2025

Taylor's law holds for heterogeneity in the risk of neonatal death. As the average death rate in a cohort of newborns declines over the first month of life, the variance of death rates declines as well in a power-law fashion. Read more in my piece on post-natal mortality selection and convergence.

020

Jonas Schöley @jschoeley.com · 05/08/2025

Kitagawa's decomposition uses the product rule for finite differences. Thus it applies to every demographic statistic expressible as a difference of weighted sums. Below, I use it to decompose changes in the variance of mortality rates into mortality selection and convergence components.

010

Jonas Schöley @jschoeley.com · 04/08/2025

Average mortality rates are a horrible measure of typical mortality during the first weeks of life. On the day of birth modal mortality is 189 times lower than average mortality. Read more in my recent piece on selection in neonatal mortality.

1113

Jonas Schöley @jschoeley.com · 01/08/2025

In this piece in honour of Jim Vaupel I pull the veil from hidden heterogeneity. I show how the observed distribution of frailties changes in a cohort of newborns over the first month of life and quantify mortality selection and its impact on the age pattern of neonate mortality. Fully reproducible.

151

Jonas Schöley @jschoeley.com · 03/07/2025

All of this discrete modeling on time series of counts can also be expressed in the continuous language of survival analysis. The distribution of the observed lifetimes is then a convolution of the distribution of expected lifetimes with a distribution of displacement times.

120

Jonas Schöley @jschoeley.com · 03/07/2025

A single slice of the time-varying distribution of displacement times allows for some relevant inference on the life-time lost of those who died at that time.

120

Jonas Schöley @jschoeley.com · 03/07/2025

The displacement matrix can be transformed into a time-varying distribution of displacement times.

110

Jonas Schöley @jschoeley.com · 03/07/2025

I use a simple heuristic to estimate the displacement matrix whereby I proportionally distribute all deficit to the excess time points. Other and better solutions are possible. The resulting model is one, where the observed deaths are expressed as expected deaths re-shuffled in time.

110

Jonas Schöley @jschoeley.com · 03/07/2025

Excess death modeling suggests a huge excess during the heatwave followed by a persistent moderate death deficit in the following fall and winter.

110

Jonas Schöley @jschoeley.com · 03/07/2025

I apply the method to time series of daily French deaths following the 2003 European heatwave (data by @rchung.bsky.social).

110

Jonas Schöley @jschoeley.com · 03/07/2025

The displacement matrix can then be transformed to yield the distribution of displacement times. Here, we shown by how much the observed deaths at t=2 have been displaced from their expectation.

110

Jonas Schöley @jschoeley.com · 03/07/2025

Given enough constraints, a unique solution for the displacement matrix can be found.

110

Jonas Schöley @jschoeley.com · 03/07/2025

The displacement of expected deaths across time to form observed deaths can be written down in matrix form as a transport model.

110

Jonas Schöley @jschoeley.com · 03/07/2025

The central idea is that, controlling for migration and population dynamics, observed deaths are expected deaths displaced in time. During a mortality crisis, excess deaths are future deaths advanced in time, leaving a deficit where they would have originally occurred.

110

Jonas Schöley @jschoeley.com · 03/07/2025

During C19 it was often argued that most excess deaths are not to people with a remaining lifespan of a few weeks, as that would imply a large deficit in deaths after the excess. I formalize this intuition proposing a displacement model, linking expected to observed deaths via a displacement matrix.

110

Jonas Schöley @jschoeley.com · 03/07/2025

Everything in mortality is a tempo effect. Let me walk you through my recent work on "A convolution formulation of mortality displacement" a.k.a. "Lifetime lost from time series of counts", last shown at Nordic Demographic Symposium. Happy to give invited talks. www.jschoeley.com/events/2025-...

1123