Handling unequal time spacing in EMA studies

[Lily Martin]

Hi there! 

In terms of handling unequal time spacing in EMA studies, it seems that apart from interpolation and/or ignoring the issue, the two main options are: 

1. Add NA rows for “overnight” (and any missing surveys), with just 1 row of NA for consecutive overnight/missing surveys versus many stacked rows of NA (to help with convergence issues). 
2. Treat time as an exogenous variable to “de-trend” the data. From Arizmendi et al. (2021): “Trend stationarity, or constant mean over time, is an assumption of time series analysis (Shumway, 2003). One can correct for violations of this assumption by including a “time” variable in u. This time variable can be represented as a linear or non-linear process (Molenaar, De Gooijer, & Schmitz, 1992). In its simplest form, time is represented as a vector numbered from 1 to T, where T is the number of time points.” 
   • I just wanted to confirm my understanding of this approach. Let’s say someone had 75 out of a possible 100 surveys. Would their time variable show 1-75 consecutively, or would it show 1-100, for example, but with some time values missing (e.g., if they missed their 6th survey, 6 would be missing from the time variable)? Would the approach retain an NA row for this person's 6th survey (like in the NA approach), or would it just be missing? That is, with the exogenous the variable approach, would everyone have 100 rows or no? 

Earlier in this thread you'd mentioned there's no definitive "right" way to accommodate unequal spacing (November 2024), and that ignoring unequal spacing may be okay in some circumstances (September 2025). I'm wondering if at this point there's any more guidance on when it might be best to use the various approaches? 

Thank you very much,

Lily 

[Katie Gates]

Hi Lily, 

Thanks for the thoughtful question. 

1. Yes, this seems to be what most people are doing. 

2. I'd like to clarify this point. Detrending the data doesn't really help with unequal spacing. It simply removes any linear or systematic changes in the mean. 

The status seems to still be the same- for these intensive longitudinal data, each researcher seems to choose whichever approach makes most sense to them based on assumptions and qualities of their data and design. I'd love to see some work on this provide a clear heuristic for people to follow! 

If anyone on this list knows of a good solution please respond all. 

Best,

Katie