Mass melding statictics. {You can have this}

[Kestiel]

Sorry if this has already been said, or to be a party-pooper but none of the OP's findings are even approaching statistical significance. If you're okay with a 95% confidence level, and your margin of error is tolerable at 5% (bare minimum really, usually <3% is used with 99% confidence level) you're looking at a minimum of 384 trials required.

However if we go with 99% CL and 3% Margin of Error we're looking at 1,843 trials to get a significant sample size. Fifty is a miniscule fraction of the latter, and roughly only 13% of the former required trials. So as it was said before by somebody else; definitely not statistically relevant and the values shown (27% reported by the game compared to his rate of 18%) are well within the tolerance levels for the small amount of trials done.

For anybody interested, for his sample size and desired outlook his margin of error for a 95% Confidence Level is 13.9%. If we were to adhere to a more stringent 99% it would be, 18.2%. Though I pity the person who tries for statistical relevant data by themselves on any worthwhile melds.

[Molly_Millions]

Sorry if this has already been said, or to be a party-pooper but none of the OP's findings are even approaching statistical significance.

This. 50 attempts means nothing in the grand scheme of things. You'll need 1000+ attempts to prove anything without a doubt.