Abstract: The object of the reseach was to describe the effect of the negative questions in a scale on the reliability of the scale and the procedure for recoding the negative questions as well as the hypothesis testing method for the questions in a scale, to investigate and to compare the reliability of a scale obtained with different methods, to compute the parallel reliability between two groups of the positive questions in a scale, to analyze the validity of a scale. The method of the parallel reliability, the method of the parallel reliability between two groups of the positive questions, the indpendent t-test method, the canonical correlation, the factor analysis, the methods of computing the Cronbach's coefficient α, the Theta coefficient θ and the Omega coefficient Ω were used. ① Based on the surveillance data of "The Scale of Training the Rural Doctors in Beijing", the comparison of the Cronbach's coefficient α before and after recoding the negative questions was made. ② By using independent t-test to test the null hypothesis of the critical ratio of every question respectively, two questions were sieved and the Cronbach's coefficient rose from 0.892 9 to 0.899 3 after sieving. ③ The simple parallel reliability(r₁=0.602 7)was much less trustworthy than the parallel reliability between two groups of the positive questions(r_p=0.832). ④ The following indicates the whole scale, α =0.899 3, θ = 0.881 3, Ω =0.956 5 and r_p=0.823 0 which showed the scale had excellent reliability. ⑤ For the convenience of discussion, we used one model of the scale named "the reasons of the rural doctors to stay in village or to leave from village" to obtain a distinct statistic structure with the factor analysis method, which showed that the questions in the model could be divided into two groups: one for practical questions, the other rational questions. Endpoints: ① It is important to recode the negative questions in the scale, to compute the reliability and the validity of the scale after sieving the questions without significance. ② It is trustworthy to assess the reliability of the scale with α, θ, Ω and the parallel reliability between two groups of the positive questions except the simple parallel reliability. ③ The statistical structure of the scale obtained by the factor analysis has a very distinct structure, but not unique. The explanation of the statistical structure is based not only upon the statistics but also upon the professional knowledge. And the statistical structure can be a clue to modify the scale.

Key words: reliability, validity, scale