Collinearity and complete separation are two well-known conditions that render indeterminate classical model parameter estimates, causing serious problems, such as for example instable estimates, unusually large standard errors, and unreliable statistical inference. In contrast, priors assumed by Bayesian methods stabilize estimation, so that Bayesian methods are anticipated to efficiently handle the problem of indeterminate parameters. To assess and showcase the harm of indeterminacy caused by collinearity and complete separation in classical and Bayesian generalized linear regression models, this master thesis focuses on both simulated and real data. Based on existing methods to quantify collinearity, this master thesis develops a method to quantify the extent of complete separation and provides several examples of the harm caused by indeterminacy due to collinearity and complete separation. Furthermore, this master thesis provides a systematic account of the impact of standardization of covariates on the extent of harm caused by collinearity and complete separation in classical and Bayesian generalized linear regression models.

To test the superiority of a new dental material for preparation of crowns of implant restorations, multiple studies are conducted and the Voss fracture load of crown specimens is measured. However, preparation of crown specimens for the Voss test can be very time- and cost-intensive. Therefore, the number of crown specimens required for a dental material study must be reduced while preserving the validity of statistical findings. To reduce the number of crown specimens in a dental material study, this project develops an approach to elicit information about the fracture load from historical studies and to combine this information with current data. Moreover, this project clarifies whether aggregated or individual specimen data are better suited to accomplish this task.

Randomized controlled trials are the best practice approach to generating reliable scientific evidence. Although randomized controlled trials are very carefully designed, one in four initiated randomized controlled trials is prematurely discontinued, primarily because the insufficient recruitment rate of trial participants inhibits the collection of the envisaged target sample size. This master thesis uses simulations and real data to systematically assess the ability of open-source tools to detect insufficient recruitment rates. Moreover, it provides recommendations on how these tools could be further refined to adequately support the design of randomized controlled trials.

The handgrip strength is a non-invasive measure of muscle strength and an important indicator of the general fitness and health of children, adults, and elderly people. Swiss standards for males and females aged 18 - 96 are based on handgrip strength measurements obtained for less than 30 persons in each age category. So far, the credibility of quantiles provided by these standards has not been fully assessed. This master thesis uses simulations and real data to assess the credibility of quantiles and to provide recommendations for the design of further refined standards.