Pens with pigs which died or had to be removed due to illness were excluded from the statistical analysis of production or pen hygiene, resulting in a sample size of 61 pairs of 0.67 m2/pig pens (1708 pigs), 65 pairs of 0.73 m2/pig (1820 pigs) and 67 pairs of 0.79 m2/pig pens (1876 pigs; 5404 pigs in total). This was to avoid the problem that removing pigs results in an increase in the space allowance per pig for the remainder. Power calculations before the experiment suggested that at least 56 repetitions would be needed to detect a difference of 10% per annum in the index of production with a power of 80. Thus even this reduced sample size met this requirement.
Production data were analysed at the level of pairs of adjacent pens (28 pigs) that share a feed trough. For analysis of start and end weight, daily gain, feed conversion, carcass lean meat and gross margin per pig, ANOVA models were fitted to determine treatment differences using the procedure MIXED in SAS (F ratios were generated as the test statistic). Pig weight at the start of the trial was fitted as a covariate and farm and room were included as random factors in the model. Interactions between treatment and farm were non-significant, so this interaction term was removed from final fitted models. Mortality was analysed in a logistic regression using the procedure GENMOD in SAS (chi-squared (χ2) values were generated as the test statistic).
Pen hygiene data were analysed at the pen level (14 pigs). Of the 1788 records collected, only 1200 could be used after pens with fewer than 14 pigs were excluded. The occurrence of faecal contamination in the experimental pens was analysed in an ANOVA with the procedure MIXED in SAS with days since the start of the trial as covariate (Fratios were generated as the test statistic). The number of pens which had contamination of the solid floor area, the number of pens requiring manual cleaning, the cleanliness of the slatted floor and pig cleanliness were analysed in a logistic regression analysis using the procedure GENMOD in SAS with age of the pigs as covariate (chi-squared (χ2) values were generated as the test statistic).