On the structure of linear-time reducibility

In 1975, Ladner showed that under the hypothesis that P is not equal to NP, there exists a language which is neither in P, nor NP-complete. This result was latter generalized by Schoning and several authors to various polynomial-time complexity classes. We show here that such results also apply to linear-time reductions on RAMs (resp. Turing machines), and hence allow for separation results in linear-time classes similar to Ladner's ones for polynomial time.