stupid question about ENT suite

At the moment I have very stupid soubt about the results of the ENT
suite. I can't explain myself in a good way this thing: supposed I
have a string with some bias, when I test the string considering bits,
the results are out of randomness limits, when I consider bytes, the
results are into the limits. Probabilistically speaking I can't
explain it. Has anybody a suggestion?

Thanks

D. Tischler