Frequency distributions of sows with return to estrus
within each potential risk factor were obtained through
FREQ procedure (SAS, 2000). Before running the models of
logistic regression, the possible collinearity among variables
studied was examined by Chi-square test (PROC
FREQ) or by correlation analysis (PROC CORR; SAS, 2000)
although the results of these analyses are not presented.
Because LL and WEI were correlated, LL was chosen as a
factor to be included in the model instead of WEI. The
Hosmer and Lemeshow goodness-of-fit test statistic was
used to describe how well the respective models predicted
the outcome variable. Better-fitting models resulted in
higher (thus non-significant) Chi-square P-values (Hosmer
and Lemeshow, 2000).