Dear all, I am applying a multilevel modelling into my dataset. Level 1 is induvial patients data and level 2 is community. Outcome variables is "Health service utilisation (Yes/No)" and level 2 predictor variable is "community location (rural/urban)". I applied multilevel logistic regression and found that rural communities significantly less utilise health care compared to urban. However, when I checked into the number of clusters in urban and rural communities, out of total 30 clusters in the dataset, I found a single cluster in the rural and 29 in the urban. So, my concern; is it safe to apply multilevel model here even if the 2nd level predictor have a single cluster in the rural? could anyone of you please have answer to this question with references to cite.
Hi!
Reading this study (doi.org/10.1371/journal.pone.0225427) can offer valuable insights to help you.
Hi,
Your concern about having only one cluster in the rural community for the second-level predictor is important. In multilevel modelling, a shortage of clusters can lead to challenges such as unstable estimates, difficulties in generalization, and potential biases. Having just a single cluster for the rural group hinders accurate estimation of variability, which can result in uncertain estimates and errors.
Imagine you're observing rural communities like individual stars in a dark sky. With only one star shining, the statistical lens struggles to capture the full picture, leading to uncertain estimates. This can magnify predictor significance and introduce bias akin to illusion.
Analogous to a tapestry with a single thread, your ability to understand intricate variations within and between clusters falters when there's only one cluster. While there's no fixed rule for cluster minimums, aiming for 5-10 clusters per group is common. A solitary cluster raises caution about the credibility of classification algorithms.
You can navigate this challenge by re-grouping the communities and creating other meaningful clusters. Alternatively, you may want to explore other analyses that suit limited clusters or consider traditional regression without the multilevel approach.
Best regards.
Meera Tandan, Using Python and fabricated sample data, I generated results from my code. If you provide your dataset, I can assist you in creating similar outcomes. Let's connect on WA; https://wa.me/+923440907874.
- Badges
- Science topic
- Similar topics
- Mathematics