Comparison of four methods of variance components estimation for heritability of economic traits in Murrah buffalo

First lactation records on 2 107 Murrah buffaloes, maintained at 4 military daily farms (Ambala, Ferozpur, Lucknow and Meerut) and 3 UP State Governme nt Farms (Babugarh, Mathura and Madurikund) during 1961-88 were used to com pare 4 methods of varjance components estimation viz. Henderson method-3 (H M-3), maximum likelihood (ML), restricted maximum likelihood (REML) and min imum variance quadratic unbiased estimation (MIVQUE). Traits included In th e study were age ar first calving, first lactation milk yield, first lactat ion length and first carving intel val. Sire and residual variance componen ts were estimated from a model containing season, and year of carving and f arm as fixed effects and sires within farm as random effect. Sire variance components obtained from HM-3 and MIVQUE for ail traits except first lactat ion milk yield were significantly (P<.05) different from sire variance comp onents obtained from likelihood procedures. ML estimates of sire and error valiance components were smaller than obtained from other procedures. Estim ates of sire components of variance for first lactation length and first ca lving interval obtained fi om HM-3 and MIVQUE were negative. Likelihood est imation (REML and ML) procedures estimates confined to the parameter space. Heritability estimate for age at first calving obtained fi om HM-3 and MIV QUE procedures was moderate. The heritability estimate for first lactation milk yield obtained from HM-3 was highest (0.25<plus/minus>0.08). ML estima tes of heritability for first lactation milk yield was lowest (0.21 +/-0.07 ). The estimates of heritability for fil st lactation length obtained from ML and REML were almost of similar magnitude. Result revealed that likeliho od methods always yield positive estimates of heritability and therefore sh ould be the methods of choice over the fitting constants procedure (HM-3) f or unbalanced data in mixed models.