You could do this a few different ways. I would start with testing for linear change over time and for a group difference in linear change. I would include week number as a continuous variable, a dummy code for participant=1 vs. non- participant=0, and a week by participant interaction. If the interaction is statistically significant then there is a difference in growth between the groups.

You would need at least a random intercept, you could add a random slope for time as well but given that your sample is so small it might not work.

see http://en.wikipedia.org/wiki/Restricted_maximum_likelihood

and http://www.ltcc.ac.uk/courses/REML_and_LMM_Notes.php

I think the first important point to remember is that a GLMM does not analyse data, the person using the GLMM does (GLMM is the hammer and you are the carpenter). That is important because you can construct mixed effects models that are non-sensical but will give you results (you can hammer a nail anywhere in a house, it does not mean that it will stand and be safe afterwards). So before embarking on a MEM development, I would strongly suggest to read one of the good books on it (eg, start with this paper: http://www.sciencedirect.com/science/article/pii/S0169534709000196).

If you are in a hurry, you need to consult with a statistician. Once your model is developed given your study design, there are many algorithms available now to fit such model to data (these are two separate processes). So yes REML/ML, but now there are very versatile approaches that are more promising given the problems with MEM fitting processes such as MCMC approaches (mcmcglmm in R for example), or Lasso, or INLA.

REML is the default in SPSS, I wouldn't change the default for this example.

Do your fixed effect model selection with ML and your random effect model selection with REML. Finalise the estimation of your final model with REML.

REML might be better for such a small sample, also there might not be a need to compare several fixed effects models to test the stated hypothesis. If you did want to test other fixed effects, you could use p-values.

First, plot your data and check whether your data meet assumptions behind LME analyses.

Second, you got two treatment arms with a repeated-measure design. LME can be used to adjust the covariance structure of the data. Look for subject effect. #ICC

Health data normally include several confounders and interaction variables. So, you might want to include an interaction variable, at least. E.g., Use product of subject group (with or without technology) and time period when survey was conducted. I agree with Letoynia’s comments.

Remember to check your model; you can quickly do this by the coefficient of determination, R2.

‘What part of the output would you look at for your results?’ – Depends upon your research question and hypotheses. This analysis will give you regression coefficients for fixed and random effects with respective confidence interval, standard error and significance level. In addition, you can include another stat measure, effect size. Basic principles are quite similar to regression analyses. Always read more literature or consult a statistician when in doubt. Good luck!

Hope this helps!

Cheers,

Amit

IF you want some help with background material on mixed models and their interpretation ( and how to fit in R , Stata and Mlwin and soon Spss) then have a look at the Lemma site at the University of Bristol - there is a free online course

http://www.bristol.ac.uk/cmm/learning/online-course/index.html

For more help on mixed models for longitudinal data see

Repeated measures (panel models) which is at

http://www.bristol.ac.uk/cmm/software/mlwin/mlwin-resources.html#multipleresp

this is more MLwin focussed, but therr are some general resources too.

FInally, there is

Developing multilevel models for analysing contextuality, heterogeneity and change using MLwiN, Volume 2; Kelvyn Jones, VS Subramanian

Publisher: Unversity of Bristol

https://www.researchgate.net/publication/260772180_Developing_multilevel_models_for_analysing_contextuality_heterogeneity_and_change_using_MLwiN_Volume_2?ev=prf_pub

Book Developing multilevel models for analysing contextuality, he...

Chris Charlton has now completed the conversion of some of our online Lemma training materials into SPSS. Two Modules have been completed.

• Module 3 on using Multiple regression

• Module 5: Introduction to Multilevel Modelling.

Importantly the latter show you how to use SPSS syntax to calculate higher- level residuals.

You can freely use these materials to teach yourself, and/or use them in courses you are teaching.

The course already has material for MLwiN, Stata and R.

http://www.bristol.ac.uk/cmm/news/57.htmlhttp://www.bristol.ac.uk/cmm/news/57.html

http://www.bristol.ac.uk/cmm/learning/online-course/index.html

http://www.bristol.ac.uk/cmm/learning/online-course/index.html

Would like to know the link between GLMMs and multilevel modeling

@AA Sunethra

Multilevel modeling is an application of GLMMs. 

thank Letoynia

What is the different between GLMM and LME? Is it dependent on the type of dependent variable?

The study is "effect of exposure to mother's voice recording on high frequency component in premature infants".

independent variable " exposure to maternal voice"

dependent variable " HF component"

Random Intercept = Subject ID

Random Effect = Class (numerical) PMA ( post manstrual age), CAT (categorical) PMA

Fixed Effect = Gender, Group, Class (numerical) PMA

Covariate =Overall mean number of minutes of exposure to maternal speech and contact (e.g., kangaroo care, infant massage).

Covariance Structure =Unstructured

what is the best statistical analysis for this study? i initially decided to test GLMM.