7PAVMALM – Multilevel and Longitudinal Modelling
Summative coursework assignment
2022 Resit assignment
Assignment summary
You have been provided with a labelled dataset containing a subset from an original study. Your task
is to undertake a theoretically and empirically meaningful analysis that requires the fitting of a series
of multilevel and longitudinal models.
The total amount of marks for this resit assignment is 60.
The final mark will contribute 70% towards the total mark of the module.
Background to the SOCRATES dataset:
The SOCrATES Trial – a Study of Cognitive Alignment Therapy in Early Schizophrenia.
This randomised controlled trial compared a 5-week CBT programme plus routine care to supportive
counselling plus routine care and routine care alone in a multi-centre trial randomising 315 people
with DSM–IV diagnosed schizophrenia and related disorders in their first or second acute admission
(Lewis et al, 2002; https://doi.org/10.1192/bjp.181.43.s91). There were 6 post-randomisation
exclusions, so the final dataset contains 309 people.
The primary outcome of the trial is the Positive and Negative Syndrome Scale (PANSS), a continous
measure of symptom severity ranging from 30-210, where a higher score indicates more severe
symptoms. A score of less than 70 indicates that a person is considered in remission, and a
secondary binary outcome records if a person’s score is less than 70 or 70 or more.
For the purposes of this assignment, the two intervention arms have been combined, and are to be
compared to the routine care alone arm.
The variables and their labels that are included in the dataset are:
idnumber Patient no#
therapy Therapy condition
interven Intervention or control
centre Centre
sex Sex
episode Admission episode
yearsofe Years of education
substmis Substance misuse
dup Duration untreated psychosis: weeks
logdup log10dup
ageentr Age at entry to study: years
panss0 Baseline panss total
panss1 Six week panns total
panss3 3 mnth panss total
panss9 9 mnth panss total
panss18 18 mnth panss total
therapis Therapist identifier
nosess Number of therapy sessions
panss0remis PANSS remission at baseline
panss1remis PANSS remission at 6 weeks
panss3remis PANSS remission at 3 months
panss9remis PANSS remission at 9 months
panss18remis PANSS remission at 18 months
Assignment details
For this assessment, consider the questions presented below and try and answer them. You should
discuss what you found and not simply reprint output from Stata. Remember to justify your choice of
statistical models and approach to the analysis.
1) Include a front sheet with your student number. Do not put your name anywhere on the
submission. Please include the word count per question, where this is specified. Tables and
figures do not count towards the word limit.
2) Present properly labelled tables, and make sure that the tables and figures are standalone
with captions describing their contents.
3) Give detail about the methods chosen for analyses and any alternative choices that could have
been considered, more so than in standard research reports.
4) At the end of the report, include your labelled do file (you can copy and paste it in word) with
the commands you have used to carry out the analyses.
5) Your answers may include a combination of text, tables and/or figures. Choose the most
appropriate way to present the findings.
6) Maximum 3000 words (excluding tables/ figures/do file) and upto 5 figures/tables.
Questions
1. Summarise the binary outcome variable (PANSS remission) between the combined
intervention arm and routine care alone arm over time.
Note: you could use graphs or tables to display these.
[5 marks]
2. Using an appropriate generalised linear mixed model, estimate the treatment effect
of the combined intervention arm to routine care alone on the PANSS remission
outcome at 9 months. You should check the robustness of your results by
performing suitable sensitivity analyses, and describe in statistical terms the methods
chosen for analyses and any alternative choices that could have been considered.
[15 marks]
You should consider the following:
An appropriate longitudinal model, based on scaling of the time variable
The appropriate metric to report the treatment effect
Any additional sources of clustering in the data
An appropriate random effect structure, based on model comparisons
Choice of any baseline variables to include in the model
Validity of underlying statistical assumptions
Graphical displays to summarise the findings from the modelling
3. Using a Generalised Estimating Equations approach, estimate the treatment effect of
the combined intervention arm to routine care alone on the PANSS remission
outcome. You should check the robustness of your results by performing suitable
sensitivity analyses, and describe in statistical terms the methods chosen for analyses and
any alternative choices that could have been considered.
[15 marks]
You should consider the following:
An appropriate correlation matrix, based on scaling of the time variable
The appropriate metric to report the treatment effect
Choice of any baseline variables to include in the model
Validity of underlying statistical assumptions
Graphical displays to summarise the findings from the modelling
4. Using a Generalised Estimating Equations approach, estimate the treatment effect of
the combined intervention arm to routine care alone on the continuous PANSS
outcome. You should check the robustness of your results by performing suitable
sensitivity analyses, and describe in statistical terms the methods chosen for analyses and
any alternative choices that could have been considered.
[15 marks]
You should consider the following:
An appropriate correlation matrix, based on scaling of the time variable
Choice of any baseline variables to include in the model
Validity of underlying statistical assumptions
Graphical displays to summarise the findings from the modelling
5. Summarise your findings from questions 2, 3, and 4 in the form of a research report
abstract results section. You may wish to highlight the difference between marginal
and conditional effects, and which of your analyses correspond to these.
[5 marks]
6. Include your STATA do or log file. This should follow good programming standards
(header/commented throughout). The log file should be error free and the do file
should be able to replicate your findings
[5 marks]