I'm would like to use pcaMethods to imput missing values. From the documentation each possible method is associated with a % of missing values they are robust to:

SVDimpute seems to be tolerant to relatively high amount of missing data (>10%)

How is this supposed to be interpreted:

• Max 15 % of the entire matrix?
• Max 15 % in each row?
• Max 15 % in each column?

Cheers

Kristoffer