This is from the USMLE website:
"During the defined time to complete the test items in each block, you may answer the test items in any order, review your responses, and change answers."
As far as adaptive testing, I could not find anything on the USMLE website regarding it. According to the First Aid series, the algorithm adjusts difficulty according to blocks of questions. Do well on one block, next block is harder. Do poorly on a block, next block is easier. This would seem to imply that you would still be able to skip questions/change answers within each block, as you can now.
From the Middle States Association of Collegiate Registrars and Officers of Admission:
(
http://www.msacroa.org/proceed97/ses13.html)
"The USMLE steps may use a variant of CAT known as computer adaptive sequential testing (CAST). CAST works just like CAT, but adapts entire blocks of items to each examinee. By using pre-constructed blocks, CAST provides better control over the quality of test forms that every examinee sees.
"CAST uses blocks of items arranged in several stages of testing. Each item block is comparable to a test book in the current paper and pencil exam. These blocks will fixed in length and each block will be separately timed. For example, a block could have 60 items with a time limit of one hour. Scores are recomputed after each block; those scores are used to select the next item block to administer. This process of scoring and selecting blocks is repeated for a fixed number of stages. At the end of the block, the examinee can either take a programmed break or go on to the next block. The combination of item blocks adaptively administered to each examinee will be fully content balanced. Final scores will reflect the different difficulty levels of the various modules. However, scores will not be provided at the test site."
Again, this implies that decisions will be made on a block-by-block basis rather than question-by-question.
Here is what the University of Louisville med school thinks will happen:
(
http://www.medschool.louisville.edu/medstudents/Second/2_step1boards.html)
"Computer Adaptive Sequential Testing (CAST):
A system that uses the computer to dynamically select questions that provide the most precise information possible about each examinee. The test bank will consist of a large number of modules. Performance on one module determines selection of the next module- if you have a difficult module and do well on it, you will receive a second module of comparable difficulty. If you do poorly on it, you will receive an easier module. A concern was raised regarding how this will affect student anxiety during the test (if you get really easy questions, does that mean you are doing poorly on the exam?) The response was that the differences in the difficulty levels are subtle, and faculty members were not able to accurately assess what the computer called difficult or easy. So it will not be apparent to students, while taking the exam, whether they are in a hard or easy module. The difficulty of each module will influence exam scoring- an algorithm incorporating the difficulty level of each module and the number of questions answered correctly will be used."
Again, block-by-block, not question-by-question.
So, who knows what will actually happen, but evidence seems to suggest that the adaptive testing algorithm used by the USMLE will be block-by-block decisions allowing the test taker to skip questions/change answers within a block.
For now, it is certain that adaptive testing has not been implemented yet.