I realise there are better forums to ask, but does anyone have experience of multilevel imputation using mice in R

[StealthPolarBear]

And be willing to answer some idiotic questions?

Yes all variables in one data frame.
I think its complaining because my binary variables have three levels and I'm trying to use logreg, but the 3 are 1,0 and NA!

Are they proper binary, or factors? Can you filter out the NA values?

Sorry, are you clustering or building a predictive model? It sounds like you want to build a multi-class logistic regression model where the target variable is either 1,0 or NA? Is that right? Or are you doing that to impute missing values before you cluster?

Mines I want mice to treat the nas as missing
I want to use mice to generate 100 or whatever imputations, and then I'll pool the results

But if you think that might be the problem, filter them out for now, and see if it works. When you use str(dataframe) what type is your input variable?

Maybe build it up, starting with one input and debug that?

Really, it is difficult without seeing script or specific errors

What about replacing the N/As with nulls first?

str(dataframe)
Thank you - I think that will be helpful. I need to know what types it thinks my variables are
I'm sorry, I know it's really frustrating, I don't really understand what I'm asking :( I do appreciate people's help

Yes it thinks all my binary are ints. Grr

Agh I bet binary need to be true/false

No experience of MICE but reading the documentation would polyreg work instead of logreg?

I'm using polyreg for unordered categorical data.
Which i suppose you could argue binary data is!
Why can't I find any examples I understand??

OK I've done something and it's run and brought me back a 'complete' dataset with no NAs.
I used pmm for binary but don't understand why I needed to.
Hmm

Does it look sensible if you plot it? If so, I'd be tempted to just accept it! I know what you mean about understanding the examples though - I once spent several hours trying to get a particularly complicated boxplot to use the right colours!

I read this as mice amputation Anyway, as you were......

I need to calculate proportions on the pooled dataset, and obviously I can also do it using the original dataset with the missings. And yes, if it looks OK, I'm going to accept it. I have someone helping me but she uses stata and so struggles with the syntax.

Is your original dataset big enough to be statistically robust if you ignore the missing values? If so, wouldn't it be better to go with that rather than model to impute the missing ones? Especially if you don't understand how it is doing it. Explainability is important too, I think.

I read it as mice in amputations and thought you were perhaps being a bit mean.

Too minty that would have been my preference ebut apparently it had to be this way

Thank you for giving me a happy reminder of what it's like sitting in the pub with my mates.

I usually treat it as a drinking game so, by the time they've spilt integers and recalibrated the coefficient of the flux capacitor, I'm ready to wobble off home.

Does R studio try and do everything for you? I've never used it, I'm quite old school... Just taught myself some Python though, that seems good for most things. But I like to write the code myself and decide what it does!

And that’s Numberwang.

Not that I've noticed, or if it does I stop it as I don't understand.
It completes variable names etc which I find helpful

Last night my imputation was running well. This morning it seems to be imputing some of the values I'd expect it to but not others. I bloody hate thos stuff

No idea what mice is but do know R

Have you done much R? You make it sound like you just haven't defined the variable

Also not being able to see the code is tricky to comment

And have you tried stackexchange ?

No, this is my first r project. Though I have coded before.
I am struggling with very basic stuff.
One of my variables is age. One of the NAs has been imputed, another hasn't. Its so odd