Latenttimuuttujamalleja kompleksisille datarakenteille
Päärahoittaja
Rahoittajan antama koodi/diaarinumero: 356484
Päärahoittajan myöntämä tuki (€)
- 453 691,00
Rahoitusohjelma
Hankkeen aikataulu
Hankkeen aloituspäivämäärä: 01.09.2023
Hankkeen päättymispäivämäärä: 31.08.2027
Tiivistelmä
In many fields of applied science, data on multiple, correlated responses are collected together with some additional covariates. A classical approach for modelling multiple, correlated data has been algorithmic' multivariate analysis, where the focus has mainly been in so-called ordination methods. Although researchers have recognized the limitations of such methods, their popularity mainly arises from their computational simplicity and availability in statistical software. Instead of ad-hoc algorithmic methods, a better approach for modelling multiple, correlated responses is via fully specifying joint statistical model for responses. Generalized linear latent variable models, offer a general framework for this. When using model-based approaches we explicitly account for key statistical properties of the data. We also have diagnostic, model selection and statistical inference tools readily available.
The development of joint modelling approaches started already in late 90's in social sciences. However, the computational complexities related to model fitting has limited their everyday usage. Now recent advances in computational methods have started a new era in research concerning joint models, and we have new tools available for fast and efficient model fitting. The research in the field of joint modeling has been very active during the past decade. However, all methods that are currently available in statistical software have their limitations when multivariate abundance data are high-dimensional, sparse, compositional and have complex correlation structures. Such data may for example be obtained using modern lab-based sampling and classification techniques
In this project we derive joint modelling methods for modern, complex ecological data and scalable computational solutions for model fitting. The methods will be made available for practitioners via non-proprietary R software allowing them to become the mainstream methods to be used instead of the out-of-date algorithmic approaches. We use ecological datasets to illustrate how the developed methods can be used for example for ecological monitoring of Finnish peatlands and forests. Such monitoring is crucial for evaluating the success of restoration programs Finland.
The development of joint modelling approaches started already in late 90's in social sciences. However, the computational complexities related to model fitting has limited their everyday usage. Now recent advances in computational methods have started a new era in research concerning joint models, and we have new tools available for fast and efficient model fitting. The research in the field of joint modeling has been very active during the past decade. However, all methods that are currently available in statistical software have their limitations when multivariate abundance data are high-dimensional, sparse, compositional and have complex correlation structures. Such data may for example be obtained using modern lab-based sampling and classification techniques
In this project we derive joint modelling methods for modern, complex ecological data and scalable computational solutions for model fitting. The methods will be made available for practitioners via non-proprietary R software allowing them to become the mainstream methods to be used instead of the out-of-date algorithmic approaches. We use ecological datasets to illustrate how the developed methods can be used for example for ecological monitoring of Finnish peatlands and forests. Such monitoring is crucial for evaluating the success of restoration programs Finland.
Vastuullinen johtaja
Päävastuullinen yksikkö
Seurantakohteet
Profiloitumisalue: Resurssiviisausyhteisö (Jyväskylän yliopisto JYU) JYU.Wisdom
