The experimental material generated from six diverse parents, comprised three crosses namely, Cocorit 71 x A-9- 30-1, HI 8062 x JNK-4W-128 and Raj 911 x DWL 5002. In each cross combination one of the parents (A-9-30-1, HI 8062 and Raj 911) had larger flag leaf area. Twelve basic generations, involved in these studies were two parents, F1 and F2, first backcross generations with both parents (BC1 and BC2), where BC1 was the cross between F1 x female parent and BC2 was F1 x male parent, their selfed progenies (BC1F2, BC2F2) and second backcross generations (BC11 , BC12, BC21, BC22) i.e. the BC1 and BC2 plants again crossed with both original parents (BC1 x female parent; BC1 x male parent and BC2 x female parent; BC2x male parent). These twelve populations of each of the three crosses were evaluated in randomized block design with three replications in two parallel experiments, one sown on 20th November (normal sown condition) and other sown on 20th December (late sown condition) in the same cropping season. Each replicate was divided into three compact blocks. The crosses, each consisting of twelve populations were randomly allotted to the blocks. All the twelve generations were then randomly allotted to twelve plots within a block. The plots of various generations contained different number of rows i.e. each parent and F1plots consisted of 2 rows, while each backcross generation in 4 rows and F2 and the second cycle of backcrosses in 6 rows. Each row was 5 m long accommodating 33 plants spaced 15 em apart, row to row distance being 30 cm. Border rows were provided at the beginning as well as at end of experimental rows in each block. The experiment was planted at Research Farm of Rajasthan Agricultural University, Agricultural Research Station, Durgapura, Jaipur, Rajasthan, India. The length and the maximum breadth of the flag leaf of the main spike of the each sampled plant was measured in centimeters and area was calculated following Simpson's (1968) formula as: Flag leaf area = (flag leaf length x flag leaf breadth) x 0.79. The data were recorded on 15 random plants in each parent and F1, 30 plants in each backcross generation and 60 plants in each F2 and second backcross generations in each replication under both the environments.
Standard statistical procedures were used to obtain means and variances for each generation and each environment separately, as suggested by Snedecor and Cochran (1968). While calculating variances, replicate effect was eliminated from total variances to obtain within replicate variance. These variances were used to compute standard error for each generation mean in each environment. Joint scaling test proposed by Cavalli (1952) were used to estimate genetic parameters by 3-parameter non-epistatic model [m, (d), (h)], 6-parameter model assuming digenic epistatic interaction [m, (d), (h), (i), (j), (l)], 10-parameter model, which allowed specification of digenic and trigenic non-allelic interactions [m, (d), (h), (i), (j), (1), (w), (x), (y), (z)].
The estimates of gene effects were obtained by weighted least square technique. Initially twelve equations were developed by equating observed generation means with their expectations in presence of digenic and trigenic interactions as proposed by Hill (1966). Generation means and their expectations were weighted, appropriate weights being the reciprocals of the square standard errors. Twelve simultaneous equations so obtained were solved by way of matrix inversion as follows:
Where, M= the column vector of the estimates of the parameters; S = the matrix of score (right hand side); J = the information matrix; J-1 = the inverse of information matrix J and is a variance-covariance matrix.
The adequacy of a model was tested by predicting twelve generation means from the estimates of each of the 3, 6 and 10-parameter model by the comparison of the weighted deviations of the observed and expected generation means in the form of chi-square test with 'n-p' d.f., which provides a test of the goodness of fit of a model. In this situation 'n' is the number of statistics or generations and 'p'. is the number of parameters. The estimates of chi2 (n-p) is obtained as:
Where, Oi = is the observed mean of ith generation; Ei = is the expected mean of ith generation; Wi = is the weight of ith generation, which is calculated as:
Relative magnitude of various gene effects in percent were also calculated by dividing the estimated value of each parameter by 'm' and multiplying it by hundred.
Components of heterosis were calculated as the difference between the mean value of F1 generation and that of its better parent was taken as a measure of heterosis. From the weighted least square estimates of components of generation mean, components of heterosis in the presence of digenic interaction were calculated as follows (Jinks and Jones 1958):
The formula given by Hill (1966) was used to calculate the component of heterosis when trigenic interactions were also present:
Percent heterosis over better parent and inbreeding depression were calculated as follows:
Parameters (h), (l) and (z) were not affected by the degree of association 'r' therefore interpretation of the different interactions in this study was based on the basis of magnitude and relative signs of these parameters (Hill 1966).