The difference with or without a correction for refraction is basically negligible. At most you are looking at 7-10% that can be attributed to refraction, which cannot fulfil the requirements for explaining the discrepancies in the results, because it is too large.

That's the last pillar of the defence for long distance observations, and refraction isn't enough. You can observe silhouettes of maintains over even longer distances which cannot be attributed to refraction at all.