The effect of stratified randomization on size and power of statistical tests in clinical trials