# Theoretical Discussion of Applicability and a Practical Example of Using Statistical Second-Generation Techniques to Analyze Causal Relationships in Animal Experiments

## TL;DR

This paper discusses using advanced statistical methods to analyze causal relationships in animal experiments, aiming to reduce the number of animals used.

## Contribution

The paper introduces second-generation statistical methods to veterinary medicine for analyzing complex causal relationships.

## Key findings

- Second-generation methods like SEM and PLS allow flexible modeling of causal relationships in animal experiments.
- These methods can be applied in veterinary medicine for breeding, epidemiology, and ecotoxicity studies.
- Using these methods may help reduce the number of animals needed by extracting more data from each animal.

## Abstract

In animal experiments, causal relationships in life processes are, for example, investigated with pharmacological interventions, surgeries, or genetic manipulations. With the aim of obtaining more data from one animal and thereby potentially even reducing animals in line with the 3R principle, collecting multimodal data from single animals could be used to statistically identify important links in life processes. We therefore suggest and give examples for the use of second-generation statistical models such as structural equation modeling (SEM) or the partial least squares method (PLS). Second-generation statistical methods are often used in social sciences. Applications in veterinary medicine are found in breeding management of farm animals, epidemiology, veterinary care, and an ecotoxicity study in fish. Second-generation statistical methods allow flexible modeling to simultaneously calculate causal relationships between constructs in several layers. Observed variables serve the definition (formative model) or measurement (reflective model) of the latent variables, which present processes that cannot be directly observed. The factor-based common factor model investigates if empirical measurements of a concept agree with the suggested theoretical nature of a concept. Its alternative is composite-based and uses constructed artifacts. The theoretical question arising from this is whether factors or composites exist in medicine.

In animal experiments, causal relationships in physiological or disease processes are investigated by using interventions. Applying second-generation statistical methods could be used to identify important links in life processes. This article as a first step describes how second-generation statistical methods that are often used in social sciences are currently applied in veterinary medicine, including a single-animal experimental study, or an ecotoxicity study in fish. It explains how second-generation statistical methods allow flexible modeling to simultaneously calculate causal relationships between constructs in several layers. It continues with a discussion on how theoretical concepts from this statistical approach could be transferred to experimental or medical data. As an applied example, an investigation on a data set analyzed with a second-generation method is presented, showing how this allows us to calculate relationships between variables within a complex theoretical model. Limitations of the use of second-generation statistical methods as strict requirements on the data sets are overcome by technical developments; however, causality cannot be established by statistically testing hypothesized causal structures. Using second-generation statistical methods in the future might promote obtaining more data from one animal and thereby potentially even reducing animals in line with the 3R principle.