2.4. Statistical analysis
The ‘sampling of all occurrences’ was used (Lehner, 1992) to generate accurate frequency and duration of data. A block-data design was chosen to achieve data recording within a reasonable expenditure of time. Out of 438 litters, 20 sows without any crushed piglets (NC-sows) were randomly sampled by using the uniform distribution procedure (SAS, 2005). For each NC-sow, one sow with one or more crushed piglets within the first 48 h after parturition (C-sow) was selected, using the matching criteria of parity and farrowing date (season) (Table 3). Piglets’ deaths were assessed visually on the farm by the scientist conducting the study, and crushing was determined in combination with video analysis. Crushing was characterised as ‘no movement of trapped piglet after a change in the sow's posture’, as described by Vieuille et al. (2003). The selected sows were healthy, showing a body temperature of below 39.5 °C within 3 days after parturition. Exterior traits were determined by a linear scoring system before lactation (Hellbrügge, 2007). Only sows without extreme deviations were considered.