Should i use multiple imputation for monotone and intermittent missing data?

[CharlesL]

I’m trying to decide on a robust method to handle missing values in a longitudinal survey dataset I’m analyzing, and I’m stuck on whether multiple imputation is the right approach for my case. My concern is that the missingness seems to be monotone for a key variable but intermittent for others, and I’m not fully confident my model assumptions are correct for the imputation process.

[Jack88]

I've seen that pattern too. The key variable would be missing for a bunch of time points (monotone), while other items pop in and out. I tried just carrying the last observation for the monotone part and it made the overall trend look smoother, but I worried it hid real changes.

[OliverT]

We actually tried multiple imputation once, and the results flipped a bit depending on whether we assumed MAR or used a different imputation model. The missingness mechanism felt shaky, so I stopped short of fully trusting the imputed numbers.

[JohnFS]

Another angle I considered was modeling the data with methods that tolerate missingness more natively, like likelihood-based approaches or mixed models, rather than imputing. It helped a bit but I still worried about the intermittent gaps skewing the estimates.

[Charles53]

Do you think the real issue is the missingness pattern or something about how we collect or define the key variable?