iiinova — where your digital self keeps evolving

Ruben C. Arslan @ruben.the100.ci · 14/09/2026

I had Claude update the chart. But it appears to be mainly home office and transit. I'm not sure being social took a big hit, since ATUS doesn't track being social via media, but there is change in the data (predating the pandemic, see last fig). IDK how this affected by ++nonresponse though.

Source: Own analysis of Bureau of Labor Statistics American Time Use Survey microdata, 2003-2025,
replicating the measure in Sharkey (2024). Weighted means; bars are 95% CIs from ATUS replicate weights.
2020 uses the BLS special-period weight; its replicate CI is not computable from the standard set.
ATUS collection was suspended 1 Oct - 12 Nov 2025 during the federal shutdown. Note truncated y axis.

Source: Own analysis of BLS American Time Use Survey microdata. Categories are ATUS tier-1 activity codes and account for 98 of the +102 minute total: the rest is an unspecified code. Location is not collected for sleep and grooming, which are assumed to occur at home, so personal care is at home by construction.

Source: Own analysis of BLS American Time Use Survey microdata. Places are the ATUS location codes, with all transport modes pooled and the codes added after 2003 (gym, bank, post office) folded into "other place" so the series are comparable. The eleven categories sum to -102 minutes, the exact mirror of the rise in time at home.

110

Ruben C. Arslan @ruben.the100.ci · 13/08/2026

maddening. researchers shouldn't abide such policies not to disclose massive errors. pubpeer first, then the journal gets a chance to do the right thing.

The original claim that science was becoming less disruptive has appeared in more than 300 news stories worldwide.

The Belgian researchers submitted their critique to Nature in October 2023. It took the journal 32 months to accept it. While the critique was under review, Nature still published an editorial and ran a News Feature drawing on the original conclusion. During those years, the original paper was cited more than 1,000 additional times and used in at least 16 policy documents.

"Whereas the software bug was fixed by the open-source community in months, it took years for the scientific correction to be published," says Ginis. "And all this time, the journal's policies prohibited us from speaking to the press while our critique was under submission."

Other researchers in the field have suggested restricting the analysis to papers with at least 10 references and citations. When they follow this guideline, the Belgian team finds a slight, statistically significant increase in disruption. Still, this is not their conclusion.

0287

Ruben C. Arslan @ruben.the100.ci · 31/07/2026

it's good, but it starts with a misleading statistic that its own source doubts (180,000,000 IQ points saved through iodine). Real effect is likely much smaller than 15 IQ points based on RCTs and the original study comes to this estimate very circuitously.

140

Ruben C. Arslan @ruben.the100.ci · 22/05/2026

interesting. with EA3 from 2018 instead of EA4 from 2022. I wonder why.

There are two main findings of our study. First, the reform led to a decrease in the association between genes and socioeconomic outcomes. This is evidenced by the negative interaction between the PGI for education and the reform across all three outcomes—education, income, and wealth—providing support for hypothesis 1b. Second, we find no significant interaction effects between parental SES and the reform, indicating that the relationship between parental SES and socioeconomic outcomes remained stable after the reform. Thus, we do not find evidence supporting hypothesis 2a or 2b. Our result of a decline in the role of the PGI for education after the reform is consistent with some previous studies (Conley and Domingue 2016; Okbay et al. 2016; Barcellos et al. 2018), while it contradicts others (Colodro-Conde et al. 2015; Lahtinen et al. 2023). Notably, our finding closely mirrors that of Barcellos et al. (2018), who found a decreasing association between the PGI for body mass index and health outcomes after the 1972 educational reform in England. This similarity is important because both studies focus on England and use a similar research design—leveraging a school leaving age reform, using PGIs, and implementing an RDD. In contrast, differences in the methodological approach and the context may explain the conflicting results with other studies. For instance, Colodro-Conde et al. (2015) used twin models instead of PGIs, while Lahtinen et al. (2024) examined Finland’s comprehensive school reform, which postponed tracking from age 11 to age 16. We know from previous literature that, in contrast to the reforms expanding free compulsory education, the reforms suppressing early tracking are especially beneficial for high-ability children from lowSES families (van de Werfhorst 2018).

130

Ruben C. Arslan @ruben.the100.ci · 08/05/2026

I've wanted an R fiddle for a while, so I had Claude make one. fiddle.rforms.org - completely client-side thanks to webr, no cookies, no external reqs without permission - link to scripts via url param (? or #) - rmd/qmd support

3242

Ruben C. Arslan @ruben.the100.ci · 21/04/2026

Can't resist reposting this.

Measuring self-citation using the patented 100% CI folding rule

280

Ruben C. Arslan @ruben.the100.ci · 14/04/2026

from the grant funder. who said anything about awards in court cases? why would people look for and report bugs that don't meet the threshold? it is not hourly pay. nothing wrong with that imo.

200

Ruben C. Arslan @ruben.the100.ci · 14/04/2026

If after your initial pour, the scale shows 25 grams, you'll add more until it shows 30 grams. But the scale still has error. What has changed is that now the true value is dependent on the measured value (X -> T) and the error is independent of X, not of T.

120

Ruben C. Arslan @ruben.the100.ci · 14/04/2026

Say, you want to measure how much müsli I eat for breakfast, I pour it into my bowl and you weigh my bowl before I eat it. The measured weight depends on the true weight (T -> X). The error of the scale means your measured value is slightly off the true value. Classical error: X=T+E

Illustration of classical measurement error

110

Ruben C. Arslan @ruben.the100.ci · 27/03/2026

no you were thinking of this one www.sciencedirect.com/science/arti... but it wasn't mayo!!! it was corn starch. gotta get the details right.

020

Ruben C. Arslan @ruben.the100.ci · 27/03/2026

journals.sagepub.com/doi/full/10....

130

Ruben C. Arslan @ruben.the100.ci · 11/03/2026

All I know is I've not seen people in the field who use RSA go beyond one trait per model, nor use Lasso. I once did LOOIC model comparison with four t2(trait,pref) splines but never published. I think the psych lit usually focuses on interpreting the derived RSA params. osf.io/download/v6m35

Significance Testing after Model Selection
Assume that the first step of investigating a (simple or complex) congruence hypothesis
revealed that the respective hypothesis-conforming model did not fit the data significantly
worse than the full third-order model. Then the coefficients of this model must be tested for
significance in a second step, to determine whether the suggested effects are significant and in
the expected direction. One open methodological issue refers to the question of how the model
selection uncertainty from the first step can be considered when testing the coefficients in the
second step, in order to avoid potential inflation of Type I error rates (Berk et al., 2010; Leeb
& Pötscher, 2006). Different strategies have been suggested to account for such post-selection
inference (e.g., splitting the data into a model selection sample and a coefficient testing
sample, Berk et al., 2010; computing confidence intervals that consider all other models that
might have been chosen by the model selection procedure, besides the one that was in fact
chosen, Berk et al., 2013; or computing a confidence interval whose coverage probability is
secured given the selected model, e.g., Lee et al., 2016; Rügamer & Greven, 2018). However,
there seems to be no consensus yet on which method is the "most valid" one. Thus, we cannot
yet make a well-informed recommendation on how to account for post-selection inference in
cubic RSA. In the illustrative R Code, we decided to reduce the alpha level to .01, but this
CUBIC RSA 52
approach should be seen as preliminary and should be adapted on the basis of future
methodological work on the question of post-selection inference.

130

Ruben C. Arslan @ruben.the100.ci · 09/03/2026

This is me living up to my new year's resolution that I'll keep getting mad about low ceilings in measurement and elsewhere.

me being mildly cross at best about a low ceiling

070

Ruben C. Arslan @ruben.the100.ci · 06/02/2026

Cystic fibrosis is highly heritable. Polydactyly which we just discussed is highly heritable. Lactase persistence is highly heritable (and not universal!). Here's a footnote from @kph3k.bsky.social's book that makes the point clearly IMO.

The evolutionary biologist Richard Dawkins made the point that ge ne tic
causes should be defined as difference makers even for relatively simple pheno-
types that are intuitively “ge ne tic,” such as eye color. He wrote, “The ‘effect’ of any
would-be cause can be given meaning only in terms of a comparison, even if only
an implied comparison, with at least one alternative cause. It is strictly incomplete
to speak of blue eyes as ‘the effect’ of a given gene G1. If we say such a thing, we
really imply the potential existence of at least one alternative allele, call it G2, and
at least one alternative phenotype, P2, in this case, say, brown eyes.”
He continues with an example of two genes both related to skin pigmentation:
“To be sure, A, the gene whose protein product is the black pigment, is necessary
in order for an individual to be black. . . . But I shall not call A a gene for blackness
unless some of the variation in the population is due to lack of A. . . . The point
that is relevant here is that both A and B are potentially entitled to be called genes
for blackness, depending on the alternatives that exist in the population (emphasis

added). The fact that the causal chain linking A to the production of the black
pigment molecule is short, while that for B is long and tortuous, is irrelevant.”
Fi nally, Dawkins pointed out that natu ral se lection is concerned with differ-
ences: some versions of genes become more common than others because those
versions cause differences in fitness. Evolution requires a comparison.
Failure to appreciate the fact that ge ne tic causes, like all other causes, are dif-
ference makers that imply a comparison to some alternative is one major flaw in
the reasoning of a still widely cited essay by the phi los o pher Ned Block. He wrote
(emphasis added), “Ge ne tic determination is a matter of what causes a charac-
teristic: number of toes is genet ically determined because our genes cause us to
have five toes. Heritability, by contrast, is a matter of what causes differences in
a characteristic: heritability of number of toes is a matter of the extent to which
ge ne tic differences cause variation in number of toes (that some cats have five
toes, and some have six).” Block’s error should be readily apparent. What causes
a characteristic is, by definition, what causes differences in a characteristic. To say
that a gene G1 causes us to have five toes is to imply the existence of an alternative
allele and an alternative phenotype— having a gene other than G1 would cause you
to have a diff er ent number of toes.
In fact, the fact that genes are difference makers can be empirically illustrated
using Block’s example of having five toes. Two of the genes that determine toe
number are EVC1 and EVC2. Rare mutations in these genes cause polydactyly
(extra fin gers and toes), as well as short ribs, teeth abnormalities, and cardiac
defects, a syndrome known as Ellis– van Creveld syndrome. The EVC genes code
for a protein that is found on the little hairlike projections that surround each
cell; the protein helps cells communicate wi…

220

Ruben C. Arslan @ruben.the100.ci · 03/02/2026

So unless I've missed something in his book, they seem to incorrectly cite Rutherford as saying it's not verifiable in principle. He says we should look to genomics, but EP as a field has barely engaged with genomics with few exceptions.

Since its inception, the eld
of evolutionary psychology has been accused of such lack of
rigor by critics who often assert that evolutionary psycho-
logical hypotheses cannot, in principle, be tested empirically
(e.g., Gould & Lewontin, 1979; Rutherford, 2020). In this

150

Ruben C. Arslan @ruben.the100.ci · 03/02/2026

So, let's see what a more modern critic has to say. AFAICT Rutherford doesn't mention EP, though he does criticise adaptationism. I don't see him saying it's unfalsifiable, he criticises assuming that any heritable variation is adaptive. That's fair! Evolution is also about mutation and drift.

$Arguments that the presence of brain disorders at high frequency in Ashkenazi Jews might explain the enrichment of genes that boost brains is woolly conjecture, and can be abandoned with current data to hand. “Winnowing through persecution” is also merely idle speculation, and has no place in a decent scientific paper. These are fractionally more sophisticated versions of the evolutionary crime we call adaptationism*—the assumption that natural selection is responsible for specific human behaviors, rather than happenstance or processes that are neither positive nor negative, but have simply drifted into existence. In the genomic age we are capable of actually seeing the parts of the genome where selection has taken place, and there are population-specific mutations that indicate positive selection of particular genes as adaptations to the local environment. Pigmentation, specific diets, resistance to diseases (such as malaria), and other traits are demonstrably local adaptations that are part of humankind’s success in colonizing the world. Adaptationism is an error because in many cases it results in untestable hypotheses, but ones that are appealing because they sound superficially convincing—Blacks are good sprinters because of selection during slavery; Jews are intellectually gifted because their history of persecution enriched$

170

Ruben C. Arslan @ruben.the100.ci · 03/02/2026

yeah, in women validity is as bad as estradiol, also unpublished. but pranj mehta has published critiques, so people should know. also widely understood in medicine.

020

Ruben C. Arslan @ruben.the100.ci · 03/02/2026

New meta finds no correlation between testosterone and risk preference, buuuuut testosterone was mainly measured using 2D:4D (nonsense) or saliva (still bad) and we know from Frey et al. 2017 that the risk measures don't tap into a coherent preference. www.sciencedirect.com/science/arti...

violin plot of results by testosterone measure

violin plot of result by risk pref measure

4274

Ruben C. Arslan @ruben.the100.ci · 02/02/2026

I also like this one from Emily Oster. What's the secret ingredient that books borrowed from the library or read to you by a paid caretaker just do not have? Surely, it must be LOVE

“First, parenting matters. Much more consistent than any of the associations in these studies is the association between parenting and child outcomes. Having books in your house and reading them to your kid is going to matter much more than what books they have at day care. This seems to be true even though your child probably spends as many waking hours with their care providers as with you. I don’t think we know precisely why this is the case, although it may be that you as the parent are the most consistent influence your child has.”

060

Ruben C. Arslan @ruben.the100.ci · 25/01/2026

To me, the funniest thing about the BART except individual differences in it not being related to important real life risk taking is that it isn't even representative of how real life balloons burst. psycnet.apa.org/record/2021-...

Markus Steiner exploding balloons for science.

191

Ruben C. Arslan @ruben.the100.ci · 18/12/2025

Maybe treat it like a small, potentially biased pilot sample — which you get for free and for more item/scale pairs than any pilot sample would have patience for.

140

Ruben C. Arslan @ruben.the100.ci · 18/12/2025

There's still some parts we struggle with. In the pre-trained model, the scatter plot looks like a 🥐. We would like our fine-tuned model's scatter plot to look like a 🥖. But currently the prediction interval (golden area) still looks like one of the misshapen loaves I tend to bake.

the golden area is the prediction interval.

130

Ruben C. Arslan @ruben.the100.ci · 18/12/2025

Finally, @bjoernhommel.bsky.social's and my paper introducing the SurveyBot3000 is officially out in AMPPS. It's a fine-tuned language model that guesstimates correlations between survey items from text alone. Not perfectly, but useful for search, for example. journals.sagepub.com/doi/10.1177/...

Fig. 3. Scatter plots of the synthetic and empirical estimates, validation study (Stage 2). Showing N = 30,135 item-pair correlations, N = 257 scale reliabilities, and N = 1,568 scale-pair correlations for (top) the pretrained SBERT model and (bottom) the fine-tuned SurveyBot3000 model. SBERT = all-mpnet-base-v2 model.

Fig. 4. Prediction error of the synthetic estimates, validation study (Stage 2). Our prediction model allowed the error term to vary freely according to the predictor, the synthetic estimate. The thin-plate splines show that some synthetic estimates were predictably more accurate.

Fig. 5. Accuracy by domain. Accuracy differed across domains. SurveyBot3000 accuracy (colored) was always higher than SBERT accuracy (gray). Results were largely consistent whether accuracy of items was tested (left, circle) within domains or (right, cross) across domains.

Fig. 1. Multistep training procedure for the SurveyBot3000, which produces synthetic estimates of interitem correlations. (a) Pretraining base model (SBERT). (b) Fine-tuning SurveyBot3000. (c) Validation. SBERT = all-mpnet-base-v2 model.

28134

Ruben C. Arslan @ruben.the100.ci · 06/12/2025

Lefthandedness also tends to yield low estimates of heritability. Interestingly, both handedness and sexual orientation are traits that most people (including on the political left) consider quite innate.

Plot from a master thesis I supervised, building on Willoughby et al. Lefthandedness, breast cancer, and sexual orientation stand out as traits where lay estimates for being mainly genetic are much larger than twin study estimates.

120

Ruben C. Arslan @ruben.the100.ci · 06/12/2025

The landscape is about "tiny initial differences [that] have large and irreversible final consequences". But most of what behaviour geneticists study is normally distributed, less Waddington landscape, more Galton board, where the ball "scampers deviously down", where differences often balance out.

The Galton board in which many small influences sum to a normal distribution.

The shot passes through the funnel and issuing from its narrow end, scampers deviously down through the pins in a curious and interesting way; each of them darting a step to the right or left, as the case may be, every time it strikes a pin. The pins are disposed in a quincunx fashion, so that every descending shot strikes against a pin in each successive row.
The principle on which the action of the apparatus depends is, that a number of small and independent accidents befall each shot in its career. In rare cases, a long run of luck continues to favour the course of a particular shot towards either outside place, hut in the large majority of instances the number of accidents that cause Deviation to the right, balance in a greater or less degree those that cause Deviation to the left.

111

Ruben C. Arslan @ruben.the100.ci · 04/12/2025

New paper by @ninanu.bsky.social: Estimating the Reliability and Stability of Emotional Variability Across Time online.ucpress.edu/collabra/art... We used Bayesian censored location scale models to estimate reliability and latent stability of positive and negative emotional variability.

Stabilities of intraindividual Emotional Variability
Note. Stabilities of emotional variability, estimated on latent and manifest levels, separately for negative (A) and positive emotion (B) and each data set (Uni2work and COVID data sets). The latent correlations shown are from the two-wave BCLSMs; the manifest correlations are from the separate, one-wave BCLSMs. Each estimation is accompanied by its corresponding 95% credible interval in brackets.

Stabilities of Mean Emotion
Note. Stabilities of mean emotion, estimated on latent and manifest levels, separately for negative (A) and positive emotion (B) and each data set (Uni2work and COVID data sets). The latent correlations shown are from the two-wave BCLSMs; the manifest correlations are from the separate, one-wave BCLSMs. Each estimation is accompanied by its corresponding 95% credible interval in brackets.

Interest in emotional variability as an interindividual difference is growing. Yet, basic features of the construct, such as whether it can be measured reliably and whether interindividual differences remain stable over time, have not yet been investigated extensively. To improve our understanding of the construct, we examined two longitudinal data sets, each comprising two waves of daily assessments, one with a 3-month and the other with a 16-month retest interval. Overcoming a key methodological limitation of past approaches, we used Bayesian censored location scale models as an alternative modeling approach that accounts for biases introduced by bounded rating scales. The results showed that the variability estimates from the models had reliabilities around rel = .64, which can be sufficient for group-level predictions. Additionally, the latent stability of r = .60 provides evidence for stable individual differences in emotional variability, suggesting that it is more than just a transient state. Our findings partially confirm previous studies, but also suggest that the reliability and stability of intraindividual emotional variability may have been underestimated because of biases due to censored distributions, something that has often been neglected in past work. We ran exploratory analyses to further examine the influence of external events and individual life transitions on emotional variability and found that emotional variability was potentially responsive to environmental changes due to the transition from studying to working life in one data set, and due to the COVID-19 pandemic in the second data set.

1316

Ruben C. Arslan @ruben.the100.ci · 03/12/2025

Want to make nice graphs with me, starting next year? I'm hiring for a position at the University of Witten/Herdecke. uni-wh.softgarden.io/job/61280592...

‘A fragmented field: Construct and measure proliferation in psychology.’ (2025)

From ‘Language models accurately infer correlations between psychological items and scales from text alone.’ (2025)

From ‘Not within spitting distance: salivary immunoassays of estradiol have subpar validity for predicting cycle phase.’ (2023)

Work in progress with cycle tracking data from the app Clue

26051

Ruben C. Arslan @ruben.the100.ci · 03/12/2025

wdym it's silent on it? they did the program and trained the teachers, and that was it. I think they would have mentioned otherwise. predictable for you and me maybe, but people keep trying this and simply don't follow up.

220

Ruben C. Arslan @ruben.the100.ci · 03/12/2025

Fadeout of cognitive training remains one of the more replicable findings in psychology in this preregistered study of 300 preschool children. Well-done study with a 4 year follow up. The language gains either faded or the control group caught up. journals.sagepub.com/doi/full/10....

Fig. 3. Effects of the intervention. In (a) are shown effects 7 months after the intervention; in (b) are shown interaction effects 7 months after the intervention. Models show the effects of the intervention on expressive-language skills in the grade 1 posttest, with an interaction between expressive-language skills in the pretest and after the intervention. Standardized coefficients (with 95% confidence intervals) are shown, except for the intervention dummy variable and the interaction where y-standardized values are indicated. Rectangles and circles contain observed variables and latent variables, respectively. Solid arrows point to significant regression or factor loadings (arrows from latent variables to their observed indicators); the arrow with dotted lines shows nonsignificant regression. The interaction (Fig. 3b) is illustrated by the arrow from the black circle to expressive language in the posttest. **p < .01.

Fig. 4. Long-term effects of the intervention on expressive-language skills. Model showing the effect of the intervention on expressive-language skills in the grade 4 posttest. Standardized coefficients (with 95% CIs) are shown, except for the intervention dummy variable where y-standardized values (equal to Cohen’s d) are indicated. Rectangles and circles contain observed variables and latent variables, respectively. Solid arrows point to significant regression or factor loadings (arrows from latent variables to their observed indicators); arrows with dotted lines show nonsignificant regression. **p < .01.

24619

Ruben C. Arslan @ruben.the100.ci · 02/12/2025

I'd argue the simplifying assumptions for the ratio aren't tenable and we can't generalise 2x to unobserved traits. Nevertheless, when I estimate the ratio as below, carrying the uncertainty, I get 2.13±.16. That's the same estimator, is it not (except I allowed a resid cor, but that reduces SE)?

bf_snp <- bf(WGS_h2 | se(WGS_h2_SE, sigma = TRUE) ~ 1)
bf_twin <- bf(Twin_Heritability ~ 1)

fit_ratio <- brm(
bf_snp + bf_twin,
data = dat %>% filter(Trait != "" & !is.na(Trait)),
chains = 4,
iter = 4000,
warmup = 1000
)

# Extract draws and compute ratio
draws <- as_draws_df(fit_ratio)

# Check variable names
# get_variables(fit_ratio)

ratio_draws <- draws$b_TwinHeritability_Intercept / draws$b_WGSh2_Intercept

# Summary
data.frame(
Estimate = mean(ratio_draws),
Est.Error = sd(ratio_draws),
Q2.5 = quantile(ratio_draws, 0.025),
Q97.5 = quantile(ratio_draws, 0.975)
)

100

Ruben C. Arslan @ruben.the100.ci · 02/12/2025

I realized I don't understand what "it" is in this sentence. A meta analysis of genomic h2? I ran a metaregression of twin h2 on genomic h2 for all traits, ignoring uncertainty in twin h2 (at my peril) for 41 traits and got a ratio of .44 ± .16.

> brms_eiv
Family: gaussian
Links: mu = identity
Formula: Twin_Heritability ~ me(WGS_h2, WGS_h2_SE)
Data: dat %>% filter(Trait != "" & !is.na(Trait)) (Number of observations: 41)
Draws: 4 chains, each with iter = 4000; warmup = 1000; thin = 1;
total post-warmup draws = 12000

Regression Coefficients:
Estimate Est.Error l-95% CI u-95% CI Rhat Bulk_ESS Tail_ESS
Intercept 0.44 0.05 0.35 0.53 1.00 4362 5107
meWGS_h2WGS_h2_SE 0.44 0.16 0.13 0.75 1.00 4307 5781

Further Distributional Parameters:
Estimate Est.Error l-95% CI u-95% CI Rhat Bulk_ESS Tail_ESS
sigma 0.12 0.01 0.10 0.15 1.00 5872 6274

110

Ruben C. Arslan @ruben.the100.ci · 01/12/2025

You'd look for a suspicious lack of effects around 0.55 for small studies in these funnel plots. Or evidence of pub bias for large rMZ and small rDZ. I wasn't around back then but an 'everyone knows' conspiracy is harder to make work than selection on significance for counter-intuitive/sexy effects.

100

Ruben C. Arslan @ruben.the100.ci · 23/11/2025

Rares: Contrast his summary with the abstract of Wainschtein et al. Focusing on ultra-rare to the exclusion of rare/de novo/structural variants seems like cherry picking a threshold in this context.

Amazingly, with three different method and datasets, the estimated “narrow-sense” heritability comes in at ~30% (shown above in blue). What about rare variants? In the two studies where common variant / GWAS heritability was estimated (Young et al. and Wainschtein et al.), it accounted for ~85% of the total narrow-sense heritability estimate, indicating that rare variants make a relatively small contribution on average and GWAS provides an only slightly lower bound on total narrow-sense heritability. As tentative support, Wainschtein et al. conducted an exploratory analysis of the heritability from 760 million “ultra-rare” variants and found that they added essentially nothing (~0.12%) to the total heritability estimate on average.

Low minor allele frequency variants in low linkage disequilibrium (LD) with neighboring variants were enriched for heritability, to a greater extent for protein-altering variants, consistent with negative selection. Our results imply that rare variants, in particular those in regions of low linkage disequilibrium, are a major source of the still missing heritability of complex traits and disease.

Graph from Wainschtein et al. showing rarer alleles have bigger individual effects.

162

Ruben C. Arslan @ruben.the100.ci · 23/11/2025

Geography: also misleading. Yes, we know of evidence where genes <-geography -> education, but we also have really obvious evidence for genes -> geography -> education (e.g., people move to university towns/big cities for their careers). So, yes, overcorrect. You can't just adjust for geography.

$After adjusting for geographic clustering the heritability estimates dropped further (38-40%). After adjusting for more geographic clustering, the heritability estimates dropped even further (32-34%). As a comparison, we can look at the same covariate adjustment for standing height and see that there is essentially no impact whatsoever. In fact, education and IQ exhibited by far the strongest evidence of stratification compared to the other traits analyzed (which were largely anthropometric or blood/lipid-oriented). GREML-WGS heritability estimates for three traits with a variety of covariates. Data from Wainschtein et al. (2025) The authors stopped at 100 geographic clusters, but does this adjustment fully correct for environmental biases (or maybe over-correct?) — no one really knows! We do know that a substantial fraction of the apparent educational attainment “heritability” is actually indirect associations from stratification or parental effects, and that disentangling these associations requires genotyped family data. So in a fundamental sense, GREML-WGS is the wrong tool for this question (direct narrow-sense heritability) for these traits (stratified behavior)4.$

260

Ruben C. Arslan @ruben.the100.ci · 18/11/2025

arxiv.org/pdf/2508.01390 have you seen this?

Box 1 — A Case Study in LLM Pollution Detection and Incidence In a recent study we
conducted with participants recruited via Prolific, we found that a lack of bot protection measures
led to an alarming rate of seemingly bot-generated answers. This example is used here illustratively
to demonstrate the types of challenges that can arise in practice rather than to report new empirical
findings. In our initial pilot study, we only tracked copying and pasting on a page containing an openended question. Notably, 45% of participants engaged in one or both of these actions, suggesting
potential cases of Partial LLM Mediation and Full LLM Delegation. Even to an untrained eye,
some of these responses were obviously generated by LLMs, representing, for instance, cases of
overly-abstract summarisations of all previous instructions. Given these results, we implemented the
following measures for the main study:
• Implemented two reCAPTCHA v2 checkpoints at the beginning and middle of our survey, asking participants to confirm that they are not bots by checking a box and potentially completing
a short validation test (0.2% failed to advance);
• Activated Qualtrics’ version of reCAPTCHA v3, which uses information about users’ behaviour
and past activity to assign each a score between 0 and 1, with lower scores indicating likely
bot activity (2.7% participants scored lower than 0.7, a threshold we determined on the basis
of our piloting, with the lowest score being 0.2);
• Added a honeypot question that included white text invisible to humans, but readable to bots
scraping all content (1.6% participants included our study-irrelevant keyword “hazelnut”; out
of these 16 participants, only 2 had low reCAPTCHA v3 scores);
• Prevented copy-pasting, while still tracking which participants attempted these actions (4.7%
participants attempted to copy and/or paste text).
While these incidence rates were less concerning than those observed during piloting, we still identified
identical nonse…

1120

Ruben C. Arslan @ruben.the100.ci · 18/11/2025

18s gemini 3 pro. Also, maybe you could do better when you were an eight year old child, not sure this is generally true ;-)

010

Ruben C. Arslan @ruben.the100.ci · 13/11/2025

maybe we're talking past each other. with this minimal sim, the pgi-top-ranked is worse than a random pick 30% of the time. is this what you mean? I don't disagree with that. But you have to choose somehow, and random picking clearly does worse more often.

100

Ruben C. Arslan @ruben.the100.ci · 13/11/2025

how is this not putting the decision problem on the family level?

100

Ruben C. Arslan @ruben.the100.ci · 06/11/2025

Great that you show the raw data, but it makes me wonder. Did you try modelling this as an ordinal outcome? Because the interaction effect looks like it could be fully explained by the item floor in ideation limiting variation at high levels of functioning. journals.sagepub.com/doi/10.1177/...

Simulated data showing a spurious interaction arise due to a ceiling effect.

120

Ruben C. Arslan @ruben.the100.ci · 28/10/2025

I built a DAG diagram with garden hoses for teaching. Pictured: a collider bias diagram, inspired by a blocked pipe situation I experienced (which I credit with giving me the intuition though it also ruined my belongings in the flooded cellar).

me with some garden hoses connected in a X -> Z <- Y fashion. If I shut the valve at Z, water from X spills out at Y

611122

Ruben C. Arslan @ruben.the100.ci · 13/10/2025

Ok, I gave it my best shot, though of course it depends on the pictures. Chocolates in a bowl wrapped or unwrapped etc.

$> library(tidyverse) > > squish <- function(x) { + x <- round(x) + x[x>7] <- 7 + x[x<1] <- 1 + x + } > nonserious <- 0.1 > N <- 500 > dt <- tibble( + serious = sample(c(F, T), N, prob = c(nonserious, 1-nonserious), replace = T), + poop = if_else(serious, 1, sample(1:7, N, T)), + chocolate = if_else(serious, squish(rnorm(N, 6.5, sd = 0.5)), sample(1:7, N, T)), + ) > > table(dt$serious) FALSE TRUE 44 456 > hist(dt$poop) > hist(dt$chocolate) > > effsize::cohen.d(dt$poop, dt$chocolate, paired = T) Cohen's d d estimate: -4.866682 (large) 95 percent confidence interval: lower upper -5.409463 -4.323901$

260

Ruben C. Arslan @ruben.the100.ci · 29/09/2025

Our fragmentation paper is now finally out! I put some of the dumb quips that didn't make the cut in the alt texts. journals.sagepub.com/doi/10.1177/...

A fragmented field. If we conceive of the constructs and concepts studied in behavioral science as a map, we would find that it is highly fragmented and directions are hard to come by. Scientists can hardly stand on each others' shoulders if they cannot manage to meet on common ground.

Fragmentation has worsened, not decreased, as the field has grown. Partly, this happens because we have too many reverse Columbuses, who, in search of prestige, set out to find a new continent, but just end up renaming India. But partly, we face a real, solvable search problem when trying to connect our fuzzy constructs and flexible measures.

Most measures are used only once. To be clear, we do not want to prevent or reduce refinements of existing constructs and measures. Revisions, translations, and other refinements can contribute to a more coherent, organized literature and improve measurement. We are most concerned with the measures conceived with limited planning and released into the literature without much commitment or much of a life expectancy. In ontologies, these are sometimes referred to as “orphan nodes.”

23412

Ruben C. Arslan @ruben.the100.ci · 11/07/2025

I made a "bubble" quiz for my class. They got the frequency of veganism and smoking quite wrong, but were fairly good on frequency of "migration background" and "never travelled abroad" even though they were selected on those traits too.

161