because they're better than nothing. also provided that they use the same sampling methodology, with the same biases, repeatedly, the polls can still be useful to show how a candidate's voter share is changing from one polling period to another, even though the actual predicted share itself might not be reliable