Adding to Zass comment, the best fit to α is being studied by IF, so far what we have is the upper bound of α, that means if we want a good confirmation rate in the network the α value can't be higher than that. Right now they are searching for the lower bound, which is more complicated because a low α permits malicious behaviours. You can read more about that here and here.

The probability of a transaction getting left behind, for various values of α and λ. λ is the rate of transactions, measured in units of transactions per h. Essentially h is the average time it takes to perform proof of work. Assuming it takes one minute to perform PoW (which is more or less the case), then λ is in transactions per minute.

Adding to Zass' answer, the best fit to α is being studied by IF, so far what we have is the upper bound of α, that means if we want a good confirmation rate in the network the α value can't be higher than that. Right now they are searching for the lower bound, which is more complicated because a low α permits malicious behaviours. You can read more about that here on the IOTA blog and in this IOTA foundation whitepaper, "Probability of being left behind and probability of becoming permanent tip in the Tangle".

[Source: The whitepaper mentioned above.]

The probability of a transaction getting left behind, for various values of α and λ. λ is the rate of transactions, measured in units of transactions per h. Essentially h is the average time it takes to perform proof of work. Assuming it takes one minute to perform PoW (which is more or less the case), then λ is in transactions per minute.