any statistician around?

[bartholomew]

The problem:

In the patient population there is only 45% of the data on ethnicity of the patient-55% is missing.

Do you think that by analysing trends in the 45% of the observable data spectrum the conclusions reached can be valid for the whole population?

I think the answer is no. What do u think?

[loudickson71]

The answer is 'unanswerbable.'

You need to know the numbers in your sample, the standard deviation in the sample, and how salient ethnicity is to what you are measuring.

[CT]

If it was 45% of 100 then no. If it 45% of several thousands then maybe.

[Calvesy]

You could extrapolate by assuming these 45% are representative of the whole. But ethnicity is not mutually exclusive to health as there are, by definition, genetic factors, also literary, poverty, religious, dietary and cultural issues linked to ethnicity that may have affected the collection of data in the first place. For example, if black males are more likely to be too stubborn to see the doctor when they are ill, or if Muslim women are too embarassed to be examined for, say cervical and breast cancer, this will sway your estimate.

[Peter Pedant]

yes I think you should go ahead and analyse the 45% and then discuss whether it is exptendable to the whole population.

Then no-one is being misled and the reader can make his own decision.

if it is randomly missing and a large sample the answer is maybe

if it has been weeded according to some rule - take out all the afrocaribeans then the answer is clearly no