The current study aimed to exam the effects of sample size, response scale, and derived scores on estimating Almehrizi generalized alpha and generalized beta reliability coefficients. To achieve the aim of the study, a scale of students' attitudes towards mathematics has been used with a test TIMSS2019 and three copies have been made so that they represent the main samples, the first is for quadruple scaling, the second is for triangular scaling, and the third is for binary scaling. Also, five derived scores were extracted in all main samples and the sample of the study consisted of 5117 male and female students from the fourth-grade students in all schools of the Sultanate. The sub-samples were formed for each of the three scales with different sizes (30, 200, 1000). The reliability coefficients were calculated for the generalized alpha and the generalized beta in each of the main samples and random samples for all types of raw and derived scores, and this was done through the R program. The coefficients in all the raw scores and the derived scores in the main samples were calculated according to the type of score and grading, then the descriptive statistics were extracted for the generalized alpha and the generalized beta reliability coefficients in random samples through 1000 iterations and compared them with the reliability coefficients of the main sample, and finding the bias and the root mean squares error (RMSE).The study reached a high score in the level and accuracy of the estimates of Almehrizi generalized alpha and the generalized beta reliability coefficients in all types of raw and derived scores, through what was shown by the estimates of reliability coefficients in all types of scale scores in random samples and their closeness to the reliability of the main sample, as well as what was shown by the bias and the root mean squares error (RMSE) which are close to zero.