Hi all, my name is Bobby Berkowitz and I’m a researcher at a non-profit organisation in South Africa that focuses on financial inclusion. We work across many countries and historically have managed large-scale household surveys. These surveys typically use design weights for sampling and weighting data.
However, in pursuing sustainable data collection, we have an American service provider using stan to implement a structured MrP approach to modelling weights for household surveys across several countries.
A focus for my organisation is building capacity on the African continent. When it comes to structured MrP and stan, our targets would be academic institutions and national statistical bureaus.
From what I can tell, in this area (national statistical bodies implementing largescale household surveys, which traditionally covers social science and demography), there is little use and knowledge of Bayesian statistical methods and possibly zero stan use.
So inspired by recent post from @vianeylb and @martinmodrak, I’m looking for ideas, suggestions, connections and such for how to go about building practical capacity for implementing something as specific as structured MrP in the contexts we are focused on and for the applications we are using?
Hi,
first - helping to build capacities in Bayesian modelling on the African continent is definitely something I would personally like to see happen. I also think it aligns with broader goals of the Stan community, but I don’t speak for the whole community. I would definitely consider some volunteering on my side for such a project, but Mr. P is totally out of my area of expertise.
From my (limited) experience however it is important to remember that most infrstructure/community projects fail - building communities is just hard and requires a lot of investment. My advice would be to focus on building personal relationships with individuals first and then form a core of the community around those individuals - people you have good relationships with would at least not ignore your emails :-) Once you have the core, it gets easier to expand - people will notice someone is doing something (blogs, events, …) regularly and may start to lurk/hang around.
Unfortunately I don’t have any connections on the African continent. I also don’t think this forum has many users from Africa (which is a shame). I only know that @betanalpha and @seantalts ran a Bayesian workshop for KEMRI Wellcome Trust in Nairobi, so they might be able to share some insights/contacts.
@bobbyb I wasn’t quite sure I understood - are you looking to train folks in your org on MrP or to find an institution on the African continent that would collaborate with you?
KEMRI seemed pretty biology / medicine focused from our brief time there, but maybe @georgegithinji knows folks in Kenya? I also know that in Nairobi there is an office of Lendable that I believe uses Stan, but that’s a private company and not an institution and probably(?) not using MrP per se.
There is a huge gap in training and any efforts to bridge this gap would be very welcome. I cannot speak for everyone, but at KEMRI-Wellcome (https://kemri-wellcome.org/) we conduct research in several areas to improve human health. If you are thinking of running a course on bayesian statistics, you can get in touch with our training department (https://ideal.kemri-wellcome.org/about-us/) and we can take it up from there. They facilitated @seantalts and @betanalpha excellent course here in Kilifi. Let me know if you are keen.
You’re right to raise thinking about where to best position and fit such training.
I think it is helpful to focus in on the technical sampling side of household surveys. Because this is the closest fit with our work that an organisation like KEMRI is likely to implement frequently.
We haven’t worked with KEMRI before but I am currently conducting cognitive interviewing with KEMRI-associated researchers (Pamoja).
We are currently planning a workshop in Q1 2020 with FSDK targeted at KNBS and CBS to build capacity in some of the content focus areas we co-developed for their FinAccess survey.
So it looks like we have a good platform opportunity to work towards. My thinking is to incorporate a focused workshop session to introduce this specific application of Mr P to the relevant people working in household sampling and weighting, since KNBS are responsible for the official national databases to do such work But perhaps this is also opportune to coordinate something with KEMRI training on such applications at the same time.
@georgegithinji, thinking specifically of the household sampling survey side of things, do you know of people at KNBS or people in and outside KEMRI who have worked with KNBS and worth talking to around this kind of opportunity?