iiinova — where your digital self keeps evolving

Juha Karvanen @juhakarvanen.bsky.social · 09/09/2026

New paper in JMLR with Otto Tabell and Santtu Tikka: “Clustering and Pruning in Causal Data Fusion.” We derive conditions for reducing causal graphs while preserving conclusions about identifiability and non-identifiability. www.jmlr.org/papers/v27/2...

Three causal diagrams illustrating graph reduction. The first shows the original causal graph, the second shows a pruned graph after irrelevant variables are removed, and the third shows a clustered graph in which three variables are combined into one node labeled T.

0101

Juha Karvanen @juhakarvanen.bsky.social · 01/09/2026

When does“variance explained” actually mean causation? In our new preprint, Olli Saarela and I develop graph-based causal variance decompositions. The framework clarifies when components of an ordered variance decomposition have causal interpretations. arxiv.org/abs/2608.27140

arxiv.org

Graph-based causal variance decompositions: When "variance explained" means causation

Recursive application of the law of total variance decomposes the marginal variance of an outcome into components attributed to explanatory variables and a residual component. The resulting decomposit...

06617

Juha Karvanen @juhakarvanen.bsky.social · 27/08/2026

Potential outcomes or causal graphs? Rather than choosing sides, our new article shows how the Neyman-Rubin and graphical frameworks complement one another, with examples where each has particular strengths. doi.org/10.1080/0003...

Three causal diagrams illustrating M-bias, a trapdoor structure, and a complex front-door structure. The diagrams show observed and unobserved variables connected by directed and confounding relationships.

010

Juha Karvanen @juhakarvanen.bsky.social · 13/11/2025

In longitudinal studies, dropout leads to a monotone missing data pattern. We show in a new article that monotonicity sometimes enables and sometimes prevents the identification of the full law, i.e., the joint distribution of actual variables and response indicators. openreview.net/pdf?id=kVthd...

Two graphs where monotonicity prevents identification

063

Juha Karvanen @juhakarvanen.bsky.social · 24/06/2025

In a new preprint, we study causal effect identification with multiple data sources. We show that certain clustering and pruning operations of the causal graph are identification invariant. This means that we may use the smaller graph to make conclusions on the larger graph. arxiv.org/abs/2505.15215

arxiv.org

Clustering and Pruning in Causal Data Fusion

Data fusion, the process of combining observational and experimental data, can enable the identification of causal effects that would otherwise remain non-identifiable. Although identification algorit...

030

Juha Karvanen @juhakarvanen.bsky.social · 10/04/2025

Our paper on the value of information for a risk-averse decision maker was published. Koski, V., Karvanen, J. Risk aversion in the value of information analysis: application to lake management. Stochastic Environmental Research and Risk Assessment (2025) doi.org/10.1007/s004...

Figure on the value of information as a function of different parameters

010

Reposted by Juha Karvanen

ELLIS Institute Finland @ellisinstitute.fi · 04/02/2025

ELLIS Institute Finland is hiring Principal Investigators in AI + machine learning. World-class resources for research incl. LUMI supercomputer, generous starting package & professorship affiliation with a university in the world’s happiest country! Apply by March 9: ellisinstitute.fi/PI-recruit

34030

Reposted by Juha Karvanen

Matti Vihola @mattivihola.bsky.social · 07/01/2025

Reminder about these positions - DL tomorrow (8 Jan).

023

Reposted by Juha Karvanen

Risto Heikkinen @riskyristo.bsky.social · 18/12/2024

The SBEDE model is released! It is a statistical model for debiasing systematic biases in expert predictions and ignoring experts who have not proven their competence #BayesianStats The article includes a real data portfolio optimization application with stock analysts' target prices rdcu.be/d3n69

rdcu.be

A Bayesian model for portfolio decisions based on debiased and regularized expert predictions

001

Juha Karvanen @juhakarvanen.bsky.social · 02/12/2024

University of Jyväskylä has opened calls for a tenure-track professor and a senior lecturer in statistics: ats.talentadore.com/apply/assist... ats.talentadore.com/apply/senior... In addition, applications are invited for the JYU Visiting Fellow Programme: ats.talentadore.com/apply/jyu-vi...

ats.talentadore.com

Assistant or Associate Professor in Statistics

The department of [Mathematics and Statistics](https://www.jyu.fi/en/science/maths) is seeking to recruit an Assistant or Associate Professor (Tenure Track) in Statistics starting 1st August 2025, or ...

044

Juha Karvanen @juhakarvanen.bsky.social · 24/11/2024

Scholar Goggler summarized my research topics. scholargoggler.com

A word cloud. "Causal", "data" and "physical" have the largest font size.

030

Juha Karvanen @juhakarvanen.bsky.social · 11/11/2024

Santtu Tikka and I wrote two preprints on identification in missing data problems. The results have interesting implications for multiple imputation. Multiple imputation and full law identifiability, arxiv.org/abs/2410.18688 Monotone missing data: a blessing and a curse, arxiv.org/abs/2411.03848

arxiv.org

Multiple imputation and full law identifiability

The key problems in missing data models involve the identifiability of two distributions: the target law and the full law. The target law refers to the joint distribution of the data variables, while ...

093

Juha Karvanen @juhakarvanen.bsky.social · 11/11/2024

Forestry is important for the economies of Finland and Sweden and provides interesting problems also for statisticians. In a recent work done in collaboration with Skogforsk, we optimized the inventory decisions in operational forestry. doi.org/10.1093/biom...

The optimal mix of inventory methods as a function of the inventory budget.

020

Juha Karvanen @juhakarvanen.bsky.social · 08/11/2024

Simulating values from a known distribution is a basic task in statistics. But how to simulate from a counterfactual distribution? We consider this question in a recently published paper. jair.org/index.php/ja... The proposed algorithm is applied to fairness analysis in credit-scoring.

020