4 ms·
So much wrong. - Regarding missing values, if you have a special number in your survey to indicate it, then preprocess your data to replace it with an NA. - R
by murbard2 12y ago
So much wrong.
- Regarding missing values, if you have a special number in your survey to indicate it, then preprocess your data to replace it with an NA.
- Regarding 80 bit doubles. Adding a few bits of precision may be just all you need in some cases, but that's a crap shoot. If you have catastrophic cancellation, then rewrite your algorithm to avoid the numerical instability. Use compensated sums for instance, or work in log-space. There are many ways to solve this problems in a much more robust way than adding more precision.
- Plenty of algorithms can handle missingness. You can treat them as hidden variables and use something like the EM algorithm or MCMC sampling.
- wodenokoto 12y agoIn his example you have several missing value indicator that each contain some sort of reason as to why it's missing. Collapsing that to NA would lose data, which is why the post argue against it.
- murbard2 12y agoIf it's important and your algorithm can handle it, then it's data, not missing data, and there's no need to be concerned with the underlying representation. If it's not important or your algorithm can't handle it, then you don't lose anything by collapsing it to NA.