Udder health, veterinary costs, and antibiotic usage in free stall compared with tie stall dairy housing systems: An optimized matching approach in Switzerland

Observational studies are important in livestock science. As treatment is not assigned randomly in such studies, selection bias can be a problem. This is often addressed by matching methods. However, if treatment and control groups differ considerably in their characteristics, it might be necessary to additionally prune observations that lack overlap in the opposite group. “Matching Frontier” method was developed because pruning observations manually often results in suboptimal solutions. The feasibility of the approach for animal health and welfare issues was tested in an observational study evaluating the effect of free stall housing and increased lying comfort on udder health, veterinary costs, and antibiotic usage in Swiss dairy farming. Data were collected in a survey with 1835 Swiss dairy farmers (response rate 28.3%). The treatment group (n = 179) comprised farmers participating in a voluntary animal welfare program that, in addition to free stall housing, required increased lying comfort. Farmers in the control group (n = 229) kept their cows in tie stalls. Using the Matching Frontier method, treated units were matched to control units based on five confounders. Subsequently, observations were pruned to achieve sufficient balance and overlap between the two groups. The effect of the program on the eight outcome variables was finally estimated using linear regression. Farmers in the treatment group had a lower incidence of clinical mastitis (−3.66 per 100 cow-years, −25%, p