Bayesian analysis using stochastic programming to generate MCMC parameter estimates
Budget: $50 – $0 AUD
Hi. I have data from a community RCT
The design was randomisation to two arms (Control, Workshop), but by finalisation needed to have three (control, intervention, Intermediate), where intermediate is the people who were randomised to receive the Workshop intervention but never finished it.
There were 120 participants in 60 expectant couples with data collected at 5 time points - time -3, 3,12,24 months and then a late extension at 8-10 years = mean 105 months (time 0 was delivery of baby).
Data is gender, age at commencement, group (C,W,I) and returns on two self-report surveys. One survey has six items generating a single variable. The other has 89 items each a 0-4 likert, generating one main scale (the sum of participant responses) and three subscales.
There is substantial attrition with patchy return, such that while 35 returned the last one, only 18 of the 120 recruited returned surveys at every time-point.
Its an attempted partial replication.
When the original was published in 2005, rmANOVA was a publishable analysis for this sort of data. This is not the case now.
I have basic training in GLM-type stats up to basic ANOVA/rmANOVA models.
Some work has been done for me already with mixed effect linear regression modelling, but I have just been advised the unequal time-intervals and patchy attrition would be much better handled by a bayesian stochastic approach.
I need an analysis reported which considers any bias arising from failure of the initial randomisation, any bias or impact arising from the subsequent attrition, examines group trajectory on the self-report measures over time, and also addresses the issue of potential relatively low power with risk of inflated effect size or overestimated significance that arises with these lowish participant numbers when analysed with frequentist mixed model analysis techniques.
I would like to know time-zone when you might be available for zoom-type conversation. The ideal candidate will have done work with social science/psychology/health studies before, and have the capacity to complete within a month or so.
Looking forward to hearing.
Jamie
The design was randomisation to two arms (Control, Workshop), but by finalisation needed to have three (control, intervention, Intermediate), where intermediate is the people who were randomised to receive the Workshop intervention but never finished it.
There were 120 participants in 60 expectant couples with data collected at 5 time points - time -3, 3,12,24 months and then a late extension at 8-10 years = mean 105 months (time 0 was delivery of baby).
Data is gender, age at commencement, group (C,W,I) and returns on two self-report surveys. One survey has six items generating a single variable. The other has 89 items each a 0-4 likert, generating one main scale (the sum of participant responses) and three subscales.
There is substantial attrition with patchy return, such that while 35 returned the last one, only 18 of the 120 recruited returned surveys at every time-point.
Its an attempted partial replication.
When the original was published in 2005, rmANOVA was a publishable analysis for this sort of data. This is not the case now.
I have basic training in GLM-type stats up to basic ANOVA/rmANOVA models.
Some work has been done for me already with mixed effect linear regression modelling, but I have just been advised the unequal time-intervals and patchy attrition would be much better handled by a bayesian stochastic approach.
I need an analysis reported which considers any bias arising from failure of the initial randomisation, any bias or impact arising from the subsequent attrition, examines group trajectory on the self-report measures over time, and also addresses the issue of potential relatively low power with risk of inflated effect size or overestimated significance that arises with these lowish participant numbers when analysed with frequentist mixed model analysis techniques.
I would like to know time-zone when you might be available for zoom-type conversation. The ideal candidate will have done work with social science/psychology/health studies before, and have the capacity to complete within a month or so.
Looking forward to hearing.
Jamie
Related categories:
Matlab and Mathematica
Statistics
Health
R Programming Language
Statistical Analysis