Hm. It's not that the test can't technically count because it was given out of level, it's just that you are limited in the conclusions you can draw. If a 4yo is tested and scored on 5yo norms, and scores above average, you can say that this is a lower bound on the score he would have gotten had there been norms for his age. You can't know how *much* higher the score would have been -- it might not have been higher at all. If a 4yo is tested and scores average or below on 5yo norms, you can't make any claims at all based on the data.
In terms of IQ tests, there are options for 4yo that do have continuous norms (most notably SB-5), but it's not shocking to consider using a 5yo test out of level. Some of the tests even offer "out-of-level" testing such that there *are* norms for younger kids on some subtests but it's assumed that most kids that age wouldn't take those subtests. K-ABC-II does that, although my recollection is that 4yo don't have huge options in that realm. DAS-II gives the tester considerable leeway. WISC does not (norms don't start until age 6, the one-year-overlap with the WPPSI is for age 6-0 to 6-11 and is really designed for out-of-level *low* testing for 6yo suspected of intellectual disability).
Personally, for the core IQ test, I would try to choose a test that has norms, even out-of-level ones, for the kid's age. I do a lot of interpretation, comparing performance on different subtests, and that really becomes much harder to do when the numbers are fuzzy. But I can see other professionals making other reasonable choices.
I'll typically only use a test with no norms for the kid's age when I'm trying to document a very specific strength or weakness and I'm pretty darned sure that the scores are all going to go in the interpretable direction (examples: 11yo who couldn't decode three-letter short-vowel words, documented this on diagnostic reading tests that don't have norms for kids that old; kid who ceilinged out the long-term memory subtests on the K-ABC but was too young for the delayed recall subtests, gave them anyway).
