First and third results are coherent. They both represent a probability of 0.9. Because 9 out of 10 is 9/10 = 0.9, and 135 out of 150 = 135 / 150 = 0.9. Nevertheless, second result is not coherent because 32 out of 50 is 32/50 = 0.64. I think this number must be discarded, and use 0.9 as the experimental probaility. That means that for 300 trials you should obtain x / 300 = 0.9 => x = 0.9 * 300 = 270. Then, the answer is that you can expect that a yellow ball will be selected 270 times out of 300.