Analyses of animal management and facility data showed that the variability between pig production companies is as important as within them. High homogeneity among farms within in the same company is a common practice in the Spanish pig companies and may be the same in countries were production is highly integrated. Oliveira et al.  evaluated factors affecting both mortality rate and feed intake of GF pigs in an integrated Spanish company and observed a narrow spectrum of husbandry and management practices among the different farms. Studies investigating factors that affect profitability in GF pigs were performed using batches from farms belonging to a unique company [6, 11] however, other studies [5, 8] looked at factors that affect profitability across companies using larger databases maintained by US Department of Agriculture (USDA). Both approaches can be used to study the variability within and among companies separately and provide different but complementary information.
Trimester of placement was an important factor responsible for variation in both FCR and sqrtMORT in four and five companies out of seven, respectively, as well as with OD. In most companies pigs placed in warm seasons had better performance. Maes et al.  also found higher mortality in pigs placed during the cold season compared to the warm season, and suggested that it could be due to respiratory diseases due to poorly ventilated buildings when trying to maintain indoor temperature. Thereby, its importance can vary among companies depending on facilities, management, or geographic location. All farms studied were located in northern Spain, and had similar outside temperature ranges, however there may have been differences in the ventilation of these farms depending on the age of the pig facility. On the other hand, environmental conditions were not recorded and trimester of placement may be correlated to other factors like quality and vailability of ingredients or meat demand. As a limitation of the study, not all companies reported the same range of categorical responses regarding trimester of placement. Three companies provided information of batches placed in the warm and cold season whereas others provided information of batches placed in all seasons. Although this difference should not cause important bias a priori, it has to be considered for future studies.
Variables such as breed of the pig-sire, sex, and sex segregation in batches were closely correlated among farms and the type of pig produced and they were pooled in a unique variable, BREGENSEG. Split-sex pens, use of entire males, and Pietrain-sire were normally used to produce “light” or “industrial” pigs whereas mixed-sex pens, which had White (Landrace, Large White, or their commercial crossings) and/or Duroc-sire pigs were preferred combinations to produce “heavy” pigs, although these combinations were also used in industrial production. In the present study six companies produced industrial pigs and one company produced heavy pigs. Only the industrial pig companies were included in the OD because of the specific production conditions of the heavy pig companies. According to Corrêa et al.  and Gispert et al. , the Pietrain boar is the most used breed of sire due to its high genetic potential converting feed into muscle tissue instead of fat tissue, improving feed efficiency. Differences observed in the companies presenting variability on this factor and in OD confirm and quantify this effect in the conditions of the current study.
Only two companies performed circovirus vaccination (ComA and ComC) in the GF phase. Circovirus disease (PCV2) was first described in Spain in 1997 and the infection is present in almost 100 % of Spanish pig farms . However, a commercial vaccine against PCV2 was developed and introduced to the market in the last decade . Currently, several companies produce this vaccine and it is used in different phases of production (sows, nursery piglets, and GF pigs). In the present study, companies that did not use this vaccination in GF phase may have done it in sows or piglets. Studies conducted by Segales et al.  and Jacela et al.  obtained a reduction in mortality and an improvement in feed efficiency when circovirus vaccine was performed in GF pigs. Similar results were obtained for ComA but not for ComC and OD. Differences in severity of the circovirus disease or in quality of the vaccine used by different companies could account for these differences.
Mixing piglets from different origins has been shown to increase disease transmission and decrease performance . Maes et al. [6, 19] found an increase in mortality in batches that had pigs from multiple origins in a study using data from one Belgium pig company. In agreement, our data shows that batches using pigs from a single origin had better FCR and sqrtMORT both in the OD and in most of the companies presenting variability in this factor.
An analysis of the housing facilities showed that the percentage of slatted floor and type of ventilation system were significant factors only in ComB. Batches that had lower than 50 % of slatted floor or automatic ventilation control had lower sqrtMORT. In addition, automatic ventilation also led to a reduced sqrtMORT in OD. The optimal proportion of slated floor should be related to the density of pigs in the pen, since pigs use specific areas of the pen to feed, rest and defecate, where resting in a dry and solid floor is a priority . In adittion, barns with fully slatted floor may have higher emissions of ammonia and other noxious gasses compared to those with partially slatted floors resulting in more respiratory problems and/or pulmonary lesions [21, 22]. The efficient removal of gases and moisture may depend also on the type of ventilation control system used. Choi et al.  observed higher profitability in nursery pigs housed in barns with automatic ventilation compared with those housed in manual ventilation. In contrast to our results, studies carried out by Losinger , Maes et al. , and Oliveira et al.  did not observe any influence of floor type or ventilation system on performance.
The type of feeder showed no effects in the individual companies in the current study, but it was an important facility factor in OD, where herds equipped with multi-space feeders had poorer FCR. Gonyou and Lou  and Myers et al.  concluded that single-space feeders with drinker may improve performance in GF pigs. On the other hand, Maes et al.  did not find any benefit using feeders with incorporated drinker.
Age of the barns, routes used to supply medication, and water source were factors not included in the OD model because there were missing values for a high number of batches. Thus, these factors were only studied at a single company level in order to obtain information about its effects. In three out of four companies, herds managed in facilities less than 10 years old had better FCR or lower sqrtMORT than herds that were managed in facilties that were more than 10 years old. There is a paucity of information regarding age of facilities on GF performance, however one study showed that age of facilities did not affect GF mortality rate . Batches using only in-feed medication also had lower sqrtMORT. According Miller et al. , in-water and injection medication are more effective in sick pigs (for therapeutic purposes) whereas in-feed medication is associated with preventive medication. Thus, the use of medication only in feed may be more common in herds with excellent health. Results about water source were contradictory among companies and no clear conclusion can be drawn.
In agreement with our study Maes et al.  observed lower mortality in smaller batches. However Oliveira et al.  did not find an effect of batch size on mortality and feed intake. Our findings suggest that an all-in-all-out management system in small batches may improve health status.
Finally, IBW was included as a covariate in all the models to account for the large variability of IBW among companies unlike the FBW which was similar among companies. Forcing IBW was decided based on the experience of the authors with different data sets. Data from a single company may not have enough range of IBW to reach significance, however IBW was included to account for the corresponding variability. Lower IBW increased sqrtMORT and improved FCR in some companies, as also observed by Larriestra et al.  concerning mortality.