Hi,

I am using Cronbach's alpha to test the reliability of an index I want to create. I am getting values of around 0.9 which I am very happy with, but sometimes SPSS says that alpha would be improved if I removed one of the variables. What sort of cut-off point should I use for this? Some of the improvements are very small (e.g. 0.002). Is it worth removing these questions?

I have several datasets each asking a range of questions which overlap quite a lot. In one case, even though an improvement is very small, it would make an improvement to remove it in every dataset which uses it. Would this mean I should remove it even if just looking at one dataset I wouldn't?