In the second part of the study, more difficult responses (those with low association value) were substituted for the original responses on the list (e.g., black-forgotten, cat-funeral, salt-membrane). In the habit family hierarchy, these new responses would initially have less habit strength than the original high-association responses. At the beginning of learning of the hard list LO-A > MED-A > HI-A, consistent with Spence-Spence theory. (WELL, NOT EXACTLY, LO > MED=HI). As learning of the difficult list progressed, when the new responses were correctly given reinforcement occurred and the habit strength of the new response increased. At the end of training of the difficult list, the new responses became dominant over the old ones, and this resulted in a cross-over where now HI-A > MED-A > LO-A (AGAIN, NOT EXACTLY, HI>MED=LO). If you only consider the HI and LO anxiety groups, the results confirm directly to theory.