Well at some point this link may be helpful:

https://www.google.com/search?q=Missing+Data+repeated+meas.+dv+r&rlz=1C1CHBF_enUS847US847&oq=Missing+Data+repeated+meas.+dv+r&aqs=chrome..69i57j33.32596j0j8&sourceid=chrome&ie=UTF-8

I am not a psychologist but rather a pessimist. If your question translates to how do I find a pattern that is not showing in my data,. I guess the only way is to collect different data where the pattern might be showing. Sorry I can't be more helpful good luck I would choose R because it does more things than SPSS and you at this point have no idea what you might . MORAL: Proper Prior Planning Prevents Poor Performance

I think R

Before modeling effects in RCTs or similar designs with missing follow-up data (which, in my experience, is not an exception but the rule) researchers should always try to understand the mechanism of missingness and perform some sensitivity analyses.

Historically, Rubin and others have made the distinction between three mechanisms of missingness:

In sum, you should at least examine the frequency of missingness in the follow-up assessments and test whether outcomes, exposures, and confounders are significantly related to participation (MAR vs. MCAR, see point 2 above). If data are MAR and you apply a multiple imputation approach, you should be on the safe side. As far as I know, these steps can be done both in SPSS and R, though I would prefer R for its greater flexibility in modeling and handling of data.

I hope this helps you to go on.

Recommended literature:

Working with missing values

Acock (2005): http://www.jstor.org/stable/3600254

Missingness in RCTs

Faria et al. (2014): https://rdcu.be/bMpnJ

Multiple Imputation

White et al. (2010): https://onlinelibrary.wiley.com/doi/abs/10.1002/sim.4067

Inverse Probability Weighting

Moodie et al. (2008): https://doi.org/10.2202/1557-4679.1106

Handling of MNAR

Leurent et al. (2018): https://rdcu.be/bMpgu

Thanks very much Alexander Pabst

To make sure I understand correctly, is "indicator variable Z of participants (1) vs. non-participants (0)" meaning participants without missing data (1) and participants with missing data (0)?

Also, would you run this regression three times for each timepoint (1, 2, and 3 months), or would you do one regression predicting the variable Z across the 3 time points?

Olivia Cody, yes, I mean a binary variable that differentiates between participants and dropouts at follow-up. Be careful, however, to distinguish between item non-response (data only partly missing) and unit non-response (data completely missing, i.e. dropout).

Since it's more a sensitivity analysis than a fixed algorithm, I would do both: single regressions for dropout at each of the three time points and an additional regression for cumulative dropout. This should give you a comprehensible impression of data quality. As a basis for deciding between MCAR or MAR, I would primarily consider the dropout at follow-up 1, though.

Perfect, thanks.

One more question for you Alexander Pabst , is there a name for this technique? I see it described in Faria et al. (2014), but it appears to just be referred to as using logistic regression.

If there isn't a formal name, would it be appropriate in a protocol paper to say: "We will conducted a sensitivity analysis using logistic regression to look for any associations between the odds of missingness and observed/assigned factors, covariates, or outcomes."

Thank you for you help!

Olivia Cody, I don't believe there is a formal name for this technique; one could refer to it as missing data analysis. I suggest to write: "We conducted a sensitivity analysis on missing data, using logistic regression in order to examine whether drop-out at follow-up is associated with observed/ assigned factors, covariates, or outcomes at baseline."

Great thanks!