there is no guarantee that one block after or before an empty block are lined by the same miner.
They don't have to be mined by the same miner, you would have to exclude the mining pool (miner) from the block n-1, you don't need it, I think if you fully understand the purpose of this analysis it will be very easy to understand what data you will need.
Phill is trying to investigate if Bitmain did not include transactions on purpose, first let's understand how do empty blocks come by:
When miner
A tells other miners that he found block number say 10, the other miners will start working on block 11 immediately, while doing so, they will download block number 10 and check it's transactions, then they will delete those transactions from their mempool because if they happen to include a transaction which was already included in block 10, their block will be invalid, so they need to delete the old transactions and only include those transactions that were not mined in block 10, now while doing that sometimes miner
B finds the solution to the block, but because of the fear of someone else finding the answer too, they will rush to send the block as fast as they can, and since they are not done with checking all transactions form the previous block they will not risk adding any transaction and they will send it empty with only 1 transaction which is the block rewards.
Now the above is pretty normal, anyone will do it, nobody will blame them for it, but the issue is, what if miner
B found the block AFTER checking the previous transactions, cleared his mempool, was able to add new transactions, then he finds the solution, refuses to add transactions and send an empty block? this is the core difference, miner
B is excused when the time between block 10 and 11 is too short, they will say okay 10 seconds are not enough to check the previous transactions, he rushed to mine an empty block all good, if the time is 50 seconds for instance, that will immediately a flag, people will question miner
B and say you had 50 bloody seconds!! which is more than enough to include new transactions, why did you NOT include them? are you trying to delay transactions to make bitcoin seem too slow?
That's what the analysis is all about, hope it's clear now.
I can do it for 1-year data of AntPool, from blockchair.com with above assumption (in seconds of timestamps). After having raw results with assumption, we might move further with correct timestamps in seconds.
It's highly unlikely that you will find variance in minutes, timestamps in seconds are crucial.
But for given data from blockchair.com, I have to assume figures for seconds are always 00, that somewhat biases results.
You can use
https://blockstream.info/