This work was inspired by a blog post by Paul Allison which I highly encourage you all to read: statisticalhorizons.com/missing-at-r...

Disclaimer: the algorithm is conservative. It can tell you that MI will be valid when the m-backdoor criterion is met (assuming your DAG is correct). But where it tells you that the m-backdoor criterion is not met, this only means the MI may not be valid, not that it won’t be valid

Using a simple modification to the algorithm we show how to determine whether MI will be valid in subsamples restricted to participants with observed values for some of the incomplete variables

Additionally, we incorporate the work of Little and Zhang who showed that even when MI in the full dataset may not be valid, it may be valid in a subsample restricted to participants with observed values for a subset of analysis variables: doi.org/10.1111/j.14...

Our algorithm provides easy to use guidance on how to implement the m-backdoor criterion in your work

Building on the work of Mohan and Pearl, Maya Mathur and colleagues have developed the m-backdoor criterion which indicates using DAGs which paths need to be closed (and whether they can be closed) in order for valid imputation osf.io/preprints/os...

The definition of MAR (required for valid imputation) becomes tricky when there is more than one incomplete variable. This has led to suggestions to move away from the MCAR/MAR?MNAR classification system. doi.org/10.1093/ije/...

Not an ALSPAC participant but I strongly object to people born the same year as me being described as in "middle age"!

I've just begun my journey into grant writing/internal project applications and it is starting to feel more and more like an exercise in click bait production. Why on earth is it more desirable to be vague and punchy than actual explaining what you want to do using field specific standard language

@timpmorris.bsky.social @ahlqvistviktor.bsky.social @maartenvsmeden.bsky.social

...modelling effects from these simulated patients.

I can see a (perhaps naïve) similarity to g-computation (I recognise this is just a tool for estimating marginal effects) but even there the analyst is using predictions based on models that used patient data as opposed to using the models to simulate a control patient and then...

Just adding "The meta-analyst decides that the accumulated evidence is in fact a pileup." as an additional favourite

Congratulations Viktor!!

I'm even less convinced by these certificates now

bsky.app/profile/bkle...