Konsta Happonen
LightNews LightNews
banner
konsta.happonen.eu
Konsta Happonen
@konsta.happonen.eu
420 followers 1.4K following 680 posts
Youth researcher. Bayesian surveyor of inner worlds. Tired baritone.
Posts Replies Media Videos
konsta.happonen.eu
Went to see Alexander Ekman's COW at the National Ballet yesterday. Most fun I've had in a while! A delightful tongue-in-cheek performance that nonetheless left me thinking about some of the more serious themes presented.
A white sculpture of a cow hanging from the ceiling.
November 1, 2025 at 12:33 PM
konsta.happonen.eu
I recently smashed our last coffee server against the kitchen sink, so we were forced to buy a new one. This one's not as functional as the Hario server I destroyed, but just looking at it sparks joy every morning.
A Kruve coffee server under a Hario V60 dripper.
October 30, 2025 at 10:50 AM
konsta.happonen.eu
m2 on the other hand is a Bayesian model with weakly informative N(0;5) priors. The model is identified despite one group (group = A & y = TRUE) being completely empty, and shows an OK fit to the data.
Standard errors of m2 are 0.3–2.8. Posterior predictions show a pretty good fit to observed probabilities.
October 28, 2025 at 1:42 PM
konsta.happonen.eu
Take a look at these univariate logistic regression examples. The sample size is 81, but group A only has observations of y == FALSE. m1 is a frequentist maximum likelihod fit to the data. Look at the standard errors (1193 logit units :D)!
Counts of y (TRUE/FALSE) across groups (A,B,C). Group A only has observations of FALSE. The frequentist model has standard errors of 1192.83 on the log-odds scale.
October 28, 2025 at 1:42 PM
konsta.happonen.eu
I understand why the city does this. Maple leaves are tough and don't decompose especially quickly. Still, it seems criminal to deliberately turn a beautiful yellow blanket into brown goo.
A machine turning yellow maple leaves into brown smush. Brown smush.
October 27, 2025 at 12:05 PM
konsta.happonen.eu
Toisaalta
October 24, 2025 at 8:47 AM
konsta.happonen.eu
set.seed(42) has xkcd random number generator vibes.

xkcd.com/221/
int getRandomNumber() { return 4; // chosen by fair dice roll // guaranteed to be random
October 22, 2025 at 8:02 PM
konsta.happonen.eu
Look what arrived in the mail today! It's my copy of the pink book of marginaleffects by @vincentab.bsky.social

#rstats
The pink book of marginaleffects
October 21, 2025 at 12:41 PM
konsta.happonen.eu
Archie (1981) seems to have called it classical algebraic coupling of two variables, which is a mouthful, but could be maybe shortened to something useful.

doi.org/10.1097/0000...
Type 3 Mathematic Coupling Classical algebraic coupling of two variables. This type of mathematic coupling is frequently found in the literature. It occurs by addition, subtraction, multiplication, or division of one variable by another variable. Consider a measured variable, X to which a random number, a, is added or subtracted to form a second variable Y = X a. If one plots on a vertical axis (ordinate) the calculated variable Y = X + a and on the horizontal axis (abscissa) the value of X, then a strong relationship between the Y co-ordinate and X co-ordinate is guaranteed because of coupling. Classical examples of data produced by addition and subtraction coupling are Figure 1 of Archie.5 Figure 2 of Gale et al. 6 Figure 1 of Fuller et al.7 and Figure 5 of Scharf et al.8 In each example cited the Y variable equals the X variable plus or minus some other measured variable or number. Thus, when a regression analysis is applied to the data, it is expected to be highly significant because of coupling. While there may be some true functional relationship between the X and Y variables it is clearly obscured by the strong linear X and Y mathematical coupling.
October 14, 2025 at 8:21 PM
konsta.happonen.eu
More seriously, I try to manage my caffeine intake by sometimes swapping coffee for chinese tea at work. Added bonus: brewing the tea is fun!
A gong fu tea brewing set
October 7, 2025 at 2:35 PM
konsta.happonen.eu
Looks cool! I already almost finished this poster, but maybe next time. Strange coincidence, but getting the bibliography to format properly was one of the great struggles of today.
Known Issues The bibliography does not work properly and must be done manually
October 6, 2025 at 6:16 PM
konsta.happonen.eu
And to you. Best day in the calendar.
A cinnamon roll.
October 4, 2025 at 6:43 AM
konsta.happonen.eu
It's a case of Jensen's inequality: the average weighted variance is not equal to the variance function evaluated at the average sample size. Not sure how well this example scales to more complex cases, but this is how I've tried to build intuition.
Variances of the mean for three samples sizes(33, 50, and 67). The red dashed line shows the average variance for the sample sizes 33 and 50, which is larger than the variance for the group wiht 50 observations.
September 28, 2025 at 11:03 AM
konsta.happonen.eu
However, because the variance of the mean is not linearly related to sample size, the variance of a population estimate for the mean with n_A = n_B = 50 is not equal to an estimate using unequal group sizes.
## Assume groups have have the same variance var_naive <- (0.5^2 * (1 / 50) + 0.5^2 * (1 / 50)) #>0.01 var_true <- (0.5^2 * (1 / 33) + 0.5^2 * (1 / 67)) #>0.0113071 ## Inverse of the design effect: var_naive / var_true #> 0.8844
September 28, 2025 at 11:03 AM
konsta.happonen.eu
Erittäin totta. Heikon edustavuuden lisäksi avoimet kyselyt ovat herkkiä silkalle sabotaasille. Kyselytutkimuksen lippulaiva Pew teki tästä hiljattain kokeenkin.
Pewn kokeen tulokset, joiden mukaan 12 nuorista avoimen nettikyselyn vastaajista kertoi olevansa valtuutettu ydinsukellusveneen operoija.
September 23, 2025 at 7:53 AM
konsta.happonen.eu
Reading a history of the social sciences and it contains this not too veiled insult of social psychology.
In social psychology, the main trends in the early postwar years were two: continuation of the effort to subject social behavior to experiment, leading, as some have claimed, to the progressive removal of the "social" as such from the field; and the continuation of efforts begun during the 1930s to make social psychology more relevant to social and political problems. Both of these trends were complicated by the impact of émigrés from Nazi Germany.
September 23, 2025 at 6:43 AM
konsta.happonen.eu
I have just discovered discrete time markov chain models and am *enthralled* #rstats
The estimated posterior distribution of a state transition matrix. Data is the mvad dataset from the TraMineR package. Estimated stationary state distribution for the previous transition matrix
September 21, 2025 at 9:29 AM
konsta.happonen.eu
I totally missed that the #IgNobel prizes were awarded yesterday! My favorite one is probably the nutrition prize, for observing that rainbow lizards prefer quattro formaggi over other types of pizza.

improbable.com/ig/winners/?...
NUTRITION PRIZE [NIGERIA, TOGO, ITALY, FRANCE] Daniele Dendi, Gabriel H. Segniagbeto, Roger Meek, and Luca Luiselli, for studying the extent to which a certain kind of lizard chooses to eat certain kinds of pizza. REFERENCE: “Opportunistic Foraging Strategy of Rainbow Lizards at a Seaside Resort in Togo,” Daniele Dendi, Gabriel H. Segniagbeto, Roger Meek, and Luca Luiselli, African Journal of Ecology, vol. 61, no. 1, 2023, pp. 226-227. <doi.org/10.1111/aje.13100> Rainbow lizards (Agama agama) are common in suburban areas throughout Africa, and have an opportunistic foraging strategy, with arthropods being the main prey source. In a coastal resort in southern Togo, West Africa, several individuals in a population were observed while feeding regularly upon non-natural human-made food (pizza) and showing a clear preference for a given type of food versus others that were offered (‘four cheeses’ being the preferred one). The fact that all monitored individuals fed upon a same type of pizza suggests that they may have some chemical cues attracting them.
September 19, 2025 at 6:06 PM
konsta.happonen.eu
Mandatory absolute unit:
A tweet by the Museum of English Rural Life featuring an image of a massive sheep and the text "look at this absolute unit"
September 18, 2025 at 6:05 AM
konsta.happonen.eu
TIL that rstanarm has a secret* dev version that includes a stan_surv function for Bayesian survival modelling in #rstats

*not actually secret
Survival Analysis Version The feature/survival branch on GitHub contains a development version of rstanarm that includes survival analysis functionality (via the stan_surv modelling function). Until this functionality is available in the CRAN release of rstanarm, users who wish to use the survival analysis functionality can install a binary version of the survival branch of rstanarm from the Stan R packages repository with: install.packages("rstanarm", repos = c('https://stan-dev.r-universe.dev', getOption("repos"))) Note that this binary is static (i.e. it is not automatically updated) and is only hosted so that users can access the (experimental) survival analysis functionality without needing to go through the time consuming (and sometimes painful) task of installing the development version of rstanarm from source.
September 17, 2025 at 6:16 PM
konsta.happonen.eu
Phew!
Croissant, coffee and fruit
September 17, 2025 at 8:36 AM
konsta.happonen.eu
The "is this" meme where the butterfly is cheese and a silksong player asks "is this git gud?"
September 13, 2025 at 4:34 AM
konsta.happonen.eu
Hey #rstats and especially {targets} peeps! What are some good ways to modularise long _targets.R files? I'm generating all figures for a book in one script and it's getting a bit unwieldy. I'm thinking of breaking the script into shorter subscripts by e.g. chapter to make it more readable.
A targets script that is ~ 2000 lines long
September 12, 2025 at 1:14 PM
konsta.happonen.eu
For my workflows I expect the dictionary approach to be how this new feature finds its way to my plots. Satisfying.
It has also been entrenched in some workflows to use a ‘data dictionary’ or codebook. For labelling purposes these dictionaries often contain column metadata that include labels or descriptions for variables (columns) in the dataset. To make it easier to work with column labels, we added the labs(dictionary) argument. It takes a named vector of labels, that can easily be generated from a data dictionary by setNames() or dplyr::pull(). dict <- tibble::tribble( ~var, ~label, "species", "Penguin Species", "bill_dep", "Bill depth (mm)", "bill_len", "Bill length (mm)", "body_mass", "Body mass (g)" ) ggplot(penguins, aes(bill_dep, bill_len, colour = body_mass)) + geom_point(na.rm = TRUE) + # Or: # labs(dictionary = dplyr::pull(dict, label, name = var)) labs(dictionary = setNames(dict$label, dict$var))
September 12, 2025 at 10:43 AM
konsta.happonen.eu
My sick leave ends tomorrow so I'm slowly trying to ease into working on computers again. Two things I've remembered today while doing that: 1) base R graphics can be so crisp and beautiful (and {tinyplot} simplifies making them), 2) writing Stan code is FUN.

#rstats
A horizontal barplot with labels created using the package tinyplot. Stan code for a toy monotonic regression. A monotonic regression.
September 11, 2025 at 11:49 AM