Omitting the hyphen can sometimes change the meaning of sentences or at least they can become ambiguous. This can occur especially in papers that describe statistical tests or introduce algorithms to evaluate p-values, but one may also describe methods that have nothing to do with statistics, and still calculate p values from t tests (but not the p-values using statistical t-tests). In this kind of context, the hyphens would really be necessary, even if writers usually try to avoid notations that could get easily confused.
Example (with a bad choice of notations): We would like to find a set of strong association patterns and evaluate the probability that the result would have occurred by chance. In the first phase, we search for the z best patterns with some goodness score. So, after the search phase, we will have z scores (but the z-scores). Then we evaluate the best patterns with a randomization test. We generate t random data sets and evaluate the score of the z:th best pattern in each data set. So, we perform t tests (but not the t-tests) and output the score of the z:th best pattern. We find out that p values (but not the p-values) of all t score values are better then the original z:th best pattern had. Therefore, we could estimate that the probability of getting z so good patterns by chance is p/t.