0
$\begingroup$

I am currently working on an ecological momentary assessment research project. I can't seem to find a clear answer in literature review to this question.

My problem is when observations reach to around 2,000 entries - we need to be clear which reading is from a specific participant perhaps belonging to a specific group as they make many entries at different time intervals each day.

What is the best way to fix the dependency issue of same participant data?

EDIT: I should mention that we are looking to make predictions of anxiety from various EMA entries multiple times a day for a few months now. This will utilize a machine learning approach. We really aren't sure if classification will work and perhaps will best be served with regression. But it is hard to tell since we aren't sure what the multiple readings from the same participants will do to the data. I am concerned with analyses specific to machine learning (how can we clean the data and do predictions with machine learning with repeated measures in a time series?)

$\endgroup$
5
  • $\begingroup$ Welcome to psych.SE. I'm not really clear on the question... Are you asking about how to code participant IDs? How to perform analysis on a mixed between/within-subject design? With ML-based regression or PCA, I think it's safe to add more variables rather than less, as the algorithm should automatically ignore those that don't affect results - is that what you are concerned about? You can also just generate sample training sets that contain some known patterns (with some noise) and test them to see how it goes? $\endgroup$
    – Arnon Weinberg
    Jun 13 at 18:56
  • $\begingroup$ @ArnonWeinberg So I guess I'm more concerned with performing analysis on repeated measures (that being data collected from the same participant many times). If we do a simple regression, that could skew the continuous relationship since at one hour the score can be lower than another. More variables is good, but I'm confused with how to group them so interpretation of analysis makes sense. Some methods I've come across is 1. a group-by-time interaction analysis and 2. using k-clustering downstreamed to supervised learning. I'm not sure if there is a standard though $\endgroup$ Jun 14 at 18:54
  • $\begingroup$ In that case, I recommend editing the question to make it clear that you are looking for help with analysis of a within-subjects (repeated measures) design. Standard approaches include ANOVA or MLM. $\endgroup$
    – Arnon Weinberg
    Jun 15 at 18:19
  • $\begingroup$ I'm concerned with machine learning though and how we could do predictive analyses on data using EMA though. Would ANOVA or MLM be used in something like this? $\endgroup$ Jun 17 at 20:03
  • 1
    $\begingroup$ I'd recommend you work with a statistician and offer them authorship. $\endgroup$
    – Bryan Krause
    Jun 17 at 21:33
0
$\begingroup$

LSTM or GRU neural networks can be used to model time-series data in machine learning. Likewise, moving average, autoregression, autoregressive-moving average, and the autoregressive integrated moving average models can be used to model time-series data.

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.