3
$\begingroup$

I have data on ~100 subjects: blood values taken on different days (day 1, 4, 7, and 11). The subjects undergo different treatments (but only one treatment per subject) and may develop different complications and I'd like to see whether these are reflected in the evolution of their blood values.

As the values on different days are not independent, simple linear regression seems wrong. The time sequences are also too short and the time points are not equally spaced, so I cannot use time series analysis. I have no experience with mixed linear models, but I doubt they are applicable on data with 100 groups of 4 (3 if I use differences) observations in each.

So, what are my options?

P.S. The existing answers to analysis of measurements over time, How to compare short (9 points) time-series, and Predicting on data consisting of many independent short time series? don't seem relevant for my question.

$\endgroup$
4
  • 1
    $\begingroup$ Why do you doubt linear mixed models are applicable? IMO they're ideal for this kind of problem. $\endgroup$
    – Eoin
    Jul 6, 2023 at 8:56
  • $\begingroup$ @Eoin I know it's technically possible, but would the results be reliable? As I said, I've never done it before, I don't have a gut feeling. My doubts are due to the very small number of observations per group. I'd have one fixed effect (e.g. time difference) and ~ $100 \cdot K$ random effects, $K$ depending on the number of treatments I want to investigate. $\endgroup$
    – Igor F.
    Jul 6, 2023 at 9:10
  • 1
    $\begingroup$ To clarify, does each subject receive one treatment throughout the process, or does treatment vary over time within subjects? $\endgroup$
    – Eoin
    Jul 6, 2023 at 21:47
  • $\begingroup$ @Eoin one treatment per subject. I clarified the question. $\endgroup$
    – Igor F.
    Jul 7, 2023 at 4:59

1 Answer 1

3
$\begingroup$

Mixed effects models are well suited to analyze this kind of data. You will need to appropriately specify the fixed-effects structure according to your questions and the features of the data. The random effects structure will account for the correlations in the repeated measurements of each subject. You may also combine the use of random effects with a serial correlation structure for the error terms.

You may find more information on these models in my course notes.

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

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