Estimated heritabilities for starting body weight were comparable to previous estimates in turkeys
[7, 11]. Similarly, estimated heritabilities for RFI, feed intake and body weight gain were also close to prior estimates in turkeys and comparable to results reported for broilers
[7, 12, 13]. However, the heritability of RFI was lower than estimated by Aggrey et al.
 and that of FCR (0.05) was also notably lower than previous estimates in turkeys
. These lower heritabilities may be due to the inclusion of all data points in this study, thereby increasing the phenotypic variance for each trait. As expected, heritabilities of all three RIG traits were similar to those estimated for the component traits (RFI and RG), since they are linear combinations of the latter. This similarity in heritabilities between RFI, RG and RIG was also previously reported in beef cattle
The negative genetic correlation between FCR and body weight gain (−0.64) has also been observed for other meat-type poultry
[12, 13, 15, 16] and is most likely due to the inverse relationship between FCR and its component trait, i.e. body weight gain. The phenotypic and genetic correlations between FCR and RFI were lower (0.15 and 0.36) than previous estimates in turkeys and broilers
[7, 13, 14], which is again likely due to the phenotypic variability found in FCR for the data (standard deviation of 0.96). The phenotypic correlations of zero between MMW and the residual feed efficiency traits (RFI, RG, RIG, RIG2, RIG3) were due to the inclusion of MMW in the regression model used to calculate each of these traits. For the same reason, RFI had a zero phenotypic correlation with feed intake and RG a zero correlation with body weight gain. Due to confounding factors of the RIG traits (RIG, RIG2, and RIG3) being linear combinations of RFI and RG, the phenotypic and genetic correlations of RIG were strong with both RFI and RG, and were not estimable for RIG2 and RIG3 with RFI and RG.
Potential reasons for choosing RG as a selection tool are the contrasted results observed in the phenotypic and genetic correlations between RFI and other traits of interest. If selection decisions were made purely on the basis of RFI, it is possible that slow growing birds with low feed intake rank high. The favorable genetic correlations of RG with both feed intake and body weight gain (−0.41 and 0.43), would lead selection on RG to produce faster growing birds with lower feed intake. The combination of RFI and RG into different traits (RIG, RIG2, RIG3), depending on the emphasis placed on RFI versus RG, captures the benefits of both composite traits. In contrast to RFI and RG, the RIG traits were phenotypically and genetically correlated with both feed intake and body weight gain (Table
3). The different weights on RFI and RG in the three RIG traits led to different phenotypic and genetic correlations with feed intake, with RIG2 having the weakest correlations and RIG3 the strongest correlations, with the opposite being true for correlations with body weight gain.
The use of RFI in animal breeding programs is becoming more and more prevalent across species. However, it may be more advantageous in some species than others. Comparing results from Table
4 with similarly grouped RFI animal studies in beef cattle shows a marked difference. In a study by Montanholi et al.
, the most efficient beef bulls ranked based on RFI (low) had significantly lower DFI than the medium and high groups. This was also the case for Irish performance tested beef bulls, for which again the most efficient bulls ranked based on RFI had significantly lower DFI than the medium and high groups
. However, in both studies, ADG of the low RFI bulls was higher or equal than the ADG of the medium and high RFI groups. In our study on turkey, the low RFI group had significantly lower feed intake than the medium and high groups, but also had significantly lower body weight gain. This could make the use of RFI less beneficial in the turkey compared to beef cattle.
When birds were divided into groups based on RG, with the high group being the most efficient, the high group had the greatest body weight gain over the test period and a slightly lower feed intake than the medium group (Table
5). The partitioning of birds into groups based on RIG led the most efficient group (high) to have both significantly lower feed intake and significantly higher body weight gain than the medium and low groups (Table
6). This demonstrates the benefit of combining RFI and RG into a single trait. The most efficient birds based on RFI ate less but had poor body weight gain, the superior birds based on RG had the best body weight gain but poorer feed intake, and the high birds based on RIG had both excellent body weight gain and feed intake. Interestingly, when looking at the most efficient groups based on each trait, average FCR was highest for the RFI group (2.78), followed by RIG (2.57), while RG and RIG2 had the same average FCR (2.54).
As shown in Table
7, RIG2 was the closest to the ideal combination of low DFI and high ADG. While RIG2 offered the best result in this study, the weights placed on RFI and RG were not optimized; optimizing the weight could reduce feed intake to reach a 5 kg body weight gain even more. In addition, our results were based on a four-week period in the life cycle of the turkey (16–20 weeks). Thus, if the effects obtained during this period apply to the entire lifespan of the birds, cost savings due to lower consumption may be greater. Unlike RFI, RIG traits have moderate correlations (phenotypic and genetic) with body weight gain, yielding greater ADG and faster growth rates. While this can lead to improvements in feed efficiency, over-emphasizing body weight gain in a selection index may have detrimental consequences because it can have a negative impact on conformation traits such as hip and leg structure, as well as on footpad and breast skin health. Poor conformation traits can also lead to poor walking ability and other undesirable consequences
Investigating the genetic and statistical parameters for traits such as RFI, RG, RIG, RIG2 and RIG3 contributes valuable knowledge to the poultry industry. However, we must also consider previous research on linearly regressed traits in animal breeding. Koch et al.
 preferred the use of RG over RFI, both Herd et al.
 and Arthur and Herd
 used RFI, and Berry et al.
 denoted the advantages of RIG. Each trait has advantages and disadvantages but all are composite traits calculated using linear regressions on the component production traits. As such, the genetic and phenotypic properties of these efficiency traits can be predicted from the genetic and phenotypic parameters of their component traits
[6, 20]. Adding any of the linearly regressed traits to a multiple trait selection index may prove useful but similar results may be achieved by applying the appropriate weights to their component traits (feed intake, body weight gain, metabolic mid-weight)