3 level data structure and sample size

Welcome to the forum for MLwiN users. Feel free to post your question about MLwiN software here. The Centre for Multilevel Modelling take no responsibility for the accuracy of these posts, we are unable to monitor them closely. Do go ahead and post your question and thank you in advance if you find the time to post any answers!

Remember to check out our extensive software FAQs which may answer your question: http://www.bristol.ac.uk/cmm/software/s ... port-faqs/
Post Reply
shanekav
Posts: 36
Joined: Wed Feb 20, 2013 12:55 am

3 level data structure and sample size

Post by shanekav »

Hi,

We would be grateful for some input if possible.

We are dealing with a three-level nested structure – L1 - 295 people / L2 - 84 houses / L3 - 14 organisations.

Questions:

1. At level 1 the households only have small numbers of people (between 1 - 5). Are there any issues we should be aware of? We are primarily interested in predictors at the household level. Will it be possible to look at slopes across households?

2. The inclusion of a second level in the data (i.e. households) improves the deviance statistic. But, the inclusion of level 3 (organisations) leads to little change (2484 to 2480), which appears to be borderline significant (allowing 1 DF between the models). However the VPC suggests that organisational level contributes roughly 12% of the variance (output below). Can/should we include this third level? If we include it, can we include predictors at this level?

Many thanks in advance,

Shane

Capture.JPG
Capture.JPG (26.75 KiB) Viewed 3251 times
GeorgeLeckie
Site Admin
Posts: 430
Joined: Fri Apr 01, 2011 2:14 pm

Re: 3 level data structure and sample size

Post by GeorgeLeckie »

Hi Shane,

Probably worth pointing out that these are general multilevel modelling queries that would apply equally in any multilevel software or commands rather than being any specific to MLwiN. So do look in multilevel modelling textbooks at the sections about sample sizes at different levels.

1. You should be able to allow the slopes of individual-level predictors to vary across households as you have a healthy enough sample of households (84).

2. Adding organisation random effects only leads to a modest improvement in the deviance because there are only 14 organisations and so the organisation variance component is estimated imprecisely. Think you are going to struggle to find significant regression coefficients on any organisation-level covariates as you have such a small sample size here.

Best wishes

George
shanekav
Posts: 36
Joined: Wed Feb 20, 2013 12:55 am

Re: 3 level data structure and sample size

Post by shanekav »

Hi George,

Many thanks for taking the time to respond.

Shane
BoyceRon
Posts: 1
Joined: Sat Mar 17, 2018 11:07 am

Re: 3 level data structure and sample size

Post by BoyceRon »

Hi George, what would be the number of organization after which you could say it's significant enough to make a noticable difference?
Post Reply