Earth’s biodiversity is facing an anthropogenic extinction crisis and yet conservation efforts are chronically underfunded. There is a tremendous need to act as efficiently as possible to conserve biodiversity – evidence-based approaches are essential to this mission. Biodiversity conservation is undergoing an evidence-based revolution, emulating techniques to synthesise evidence pioneered in medicine, such as systematic reviews, meta-analyses, and subject-wide evidence syntheses, which have had success in summarising the evidence on what works in conservation. To date, however, the biases and consistencies in this evidence base have neither been quantified nor explored in detail. This is important to facilitate further evidence-based decision-making in conservation and to improve the reliability and relevance of the evidence base. In line with its title, this thesis is structured into quantifying and addressing two types of biases in the evidence base for conservation: within-study and between-study biases. Both types of biases represent fundamental challenges that must be overcome to ensure evidence-based decision-making becomes more commonplace in biodiversity conservation practice and policy.

Within-study biases affect the reliability (internal validity) of research findings, which are known to hinge upon the choice of study design used to collect data. Many study designs are used to test the effectiveness of conservation interventions, including ‘gold standard’ randomised experiments (in the medical sciences) and various types of observational designs (often used when randomised experiments are too hard to implement cheaply, ethically, or practically). However, no large-scale, direct, and quantitative comparisons of the relative reliability of different study designs have been made and therefore little is known about how much more trust should be placed in results obtained using one design over another. I tackle this issue by quantitatively estimating the relative reliability of results obtained by commonly used study designs in ecology. In the first Chapter, I simulate the performance of different study designs using empirically derived estimates of the magnitude of study design bias from 51 ecological datasets obtained from a range of studies around the world. In the second Chapter, I build on these simulations by digging deeper into the raw datasets, conducting pairwise comparisons of the estimates given by different designs within each dataset. I also develop a hierarchical Bayesian model to quantify the relative reliability of study designs, enabling meta-analyses to account more effectively for the bias and variance introduced by studies. This approach attempts to tackle the challenging issue of combining study results obtained using different study designs, which has been a hotly debated issue in evidence synthesis.

Understanding between-study biases, or biases affecting the wider literature’s distribution and coverage, is crucial to prioritising future conservation research and action. In my third Chapter, I use the Conservation Evidence database (comprised of quantitative tests of conservation interventions) to quantify the spatial, taxonomic, bioclimatic, and design-related biases to show the severity of the knowledge gaps for amphibian and bird conservation. Entire orders of amphibians and birds were either poorly represented or absent in the evidence base, whilst more credibly designed studies were located, almost exclusively, in North America, Europe, and Australasia. In addition, fewer studies were conducted in locations with more threatened amphibian and bird species. These results run counter to the mission of conservation, suggesting that places with the greatest need for conservation often lack credible evidence. In my fourth Chapter, I investigate how much evidence exists for certain local questions. This is important because decision-makers typically prefer evidence that is locally valid and relevant to their specific setting. I quantify how much evidence exists within certain distances of a given decision-maker anywhere in the world, and then demonstrate, on average, how little relevant or credible evidence exists for most decision-makers. This work reinforces that there is a serious mismatch between where we test conservation interventions and where they are needed, and that providing decision-makers with locally relevant evidence is a major challenge.

This thesis demonstrates the fundamental importance of study design in determining the reliability of study findings, whilst also highlighting important knowledge gaps and biases in the literature that tests conservation interventions. Based on the findings of this thesis, I provide several recommendations and possible solutions to improve the evidence base for conservation, and to ensure that evidence-based decision-making and practice becomes more widespread and successful.