Consider a study of a large population, say n people, over time and for 2 or at most 3 calls. The dimension of the collected data is (n, 3). The data set is small in width but large in length. Do you think this type of data could represent the within variability?
Do you think considering complex models for handling this variability worths it in terms of cost and time, if so are they capable to model that type of variability?
Hello,
You have data from a big group of people (*n* people), and you check on them a few times (say, 2 or 3 times). So, your data looks like:
Person 1: Measurement at Time 1, Time 2, (maybe Time 3)
Person 2: Measurement at Time 1, Time 2, (maybe Time 3) and so on for all *n* people.
Since you only have 2 or 3 measurements per person, the "width" (number of time points) is small, but the "length" (number of people) is big. With just 2 or 3 measurements, you can see some changes, but it’s limited.
-If someone’s values go up and down, you can track that. But with only 2 points, you only see a straight increase/decrease. With 3, you might see a small trend (like up-down or down-up).
The more measurements you have, the better you see patterns. So, you can see some within-person changes, but it’s not super detailed.
Hope that helps.
Thanks!!
Hello Nitish Jha
Thank you for your reply.
So, do you think, the within-person variability should be considered?
As you mentioned these data are not well capable to represent the within-person variability, so should we remove it?
I wonder if ML and AI could handle this challenge?
Bests
Hasan MISAII
importance is not how to measure, just to capture an experiment of interest. the length is not so important, the thing is how you have taken data, the sampling is more important because it explains if the length and repetitions are enough. the theory will always say if its neough, but need to read about the concept of sample: its the way you recopile and build the analysis based on how data was representing the variable of interest. good day
this data can stand for within-subject variability, since it indicates how people change over those few calls.
In my point of view, using complicated models to account for this variation is probably NOT worth the time and expense. With just 2 or 3 data points per individual, simpler models can actually pick up changes. Complicated models would be at a high risk of overfitting and would not provide much more insight based on the restricted number of data points per subject. Although complicated models theoretically could be used, their power to represent subtle variation is greatly impaired by the sparse repeated measures.
- Badges
- Science topic
- Similar topics
- Engineering
- Electrical Engineering