Can I plug my own expected frequencies into a chi squared test or do I have to use the model described in text books (i.e. summing rows and columns)?

For example, if the expectation is that the ethnic composition of a workforce should mirror that of the local community, can I use the workforce numbers and census numbers in a chi squared?

If not, which test should I use to test congruence?

Many thanks,

Sarah

Sarah, not sure about this, but I think you could. I think a chi squared dist is only for used in distribution that exhibit bias, though (right or left biasness in their curves).