Johan Henrik Koskinen

This semester I am teaching Multivariate Methods and GLM. I have supervised a number of master’s and PhD students in various areas of applied statistics, including measurement error in labour force participation, survival analysis in Criminology, and social network analysis applied in a variety of areas. I am always looking for research students interested in advancing modelling and inference in statistical network analysis. See PhD projects

My main research interests centre on statistical modelling and Bayesian inference for networks. Networks present massively interdependent binary (mostly) data, either cross-sectionally or longitudinally, and methodology for network analysis has evolved to become a thriving field of statistics.

Representative Publications

Koskinen, J.H. & Snijders, T.A.B. (2023). Multilevel longitudinal analysis of social networks. Journal of the Royal Statistical Society Series A, Vol. 186 (3), 376–400 (arXiv preprint arXiv:2201.12713) https://doi.org/10.1093/jrsssa/qnac009

Koskinen, J., and Daraganova, G. (2022). Bayesian Analysis of Social Influence. Journal of the Royal Statistical Society: Series A (Statistics in Society), Vol. 185 (4) 1469--2325.  https://doi.org/10.1111/rssa.12844

Stys, P., Verweijen, J., Muzuri, P., Muhindo, S., Vogel,C., and Koskinen, J. (2020) Brokering Between (not so) Overt and (not so) Covert Networks in Conflict Zones, Global Crime, 21(1): 74-110. DOI: 10.1080/17440572.2019.1596806

Bright, D , Koskinen, J., Malm, A. (2019). Illicit network dynamics: The formation and evolution of a drug trafficking network, Journal of Quantitative Criminology, 35(2): 237-258. DOI: 10.1007/s10940-018-9379-8

Koskinen, J., Wang, P., Robins, G., Pattison, P. (2018). Outliers and Influential Observations in Exponential Random Graph Models. Psychometrika, 83(4), 809-830. DOI: 10.1007/s11336-018-9635-8

Koskinen J., Caimo, A., Lomi, A. (2015). Simultaneous modeling of initial conditions and time heterogeneity in dynamic networks: An application to Foreign Direct Investments. Network Science, 3(1): 58-77. DOI: 10.1017/nws.2015.3

Koskinen, J. H., Robins, G. L., Wang, P., Pattison, P. E. (2013). Bayesian analysis for partially observed network data, missing ties, attributes and actors. Social Networks, vol. 35(4), 514-527. DOI: 10.1016/j.socnet.2013.07.003

Lusher, D., Koskinen, J., Robins, G., (2013). Exponential Random Graph Models for Social Networks: Theory, Methods and Applications, Cambridge University Press, NY.

Snijders, T.A.B., Koskinen, J.H., & Schweinberger, M. (2010). Maximum likelihood estimation for social network dynamics. The Annals of Applied Statistics, Vol. 4(2), 567–588. DOI:10.1214/09-AOAS313

Koskinen, J.H. & Snijders, T.A.B. (2007). Bayesian Inference for Dynamic Social Network Data. Journal of Statistical Planning and Inference, Vol 137 (12) 3930-3938

Grants

Covert Networks: How to learn as much as possible about the structure of a network from sampled subnetworks (Grant W911NF-21-1-0335 for Proposal 79034-NS), Army Research Office [co-investigator; US lead M. Schweinberger]

Multidimensional networks of users and car purchases (Proposal No 2005661), National Science Foundation US [Named non-US partner]

Current Research Students

Jonathan Januar, Missing Data Mechanisms in Covert Networks [funded by W911NF-21-1-0335]

Potential PhD project areas

I welcome PhD candidates in the following broad areas.

Statistics for complex data. Network data describe the status of pairwise interactions between social or other units. Data may be the volume of trade between each pair of countries, for all countries; information for each pair of students in a school class whether they are best friends or not; what people committed a crime together or not, etc. Statistical methodology for networks is one of the most vibrant fields of statistical research. The interest generally lies in the many unexplored ways in which you may try to adapt current statistical modelling and inference techniques to the massively multivariate and highly interdependent network data.
 
Improved estimation methods for network models. A couple of statistical models for network data – exponential random graph models, for cross-sectional data, and stochastic actor-oriented models, for longitudinal data – have established themselves as the current best methods for analysing social networks. These models are successful in inferring such things as whom you are likely to know, and whether you only get to know people that think like you or whether you start thinking like the people you know. Statistical inference for these classes of models have to rely on computational methods, such as Markov chain Monte Carlo, by necessity, yet the combinatorial complexities or the data prevent these modelling frameworks from benefitting from the advances that have been made in standard statistics, such as variational Bayes, Machine-learning, etc.
 
Inference from “not so good” network data. Analysing network data, such as terrorist networks or criminal networks, already requires sophisticated statistical models and inference techniques. However, no matter how sophisticated these techniques are, data are always assumed to be perfectly observed and measured. In real life, we have no idea of how to define what it means to be “friends”, if the people I consider my friends even know me, or if you think that we are just friends but I think that we are more than that. To make matters worse, we can never do a census of who is connected to whom, and we invariable have rely on a mere sample on the network.