Skip to content

The UN Plot to Force Bayesianism on Unsuspecting Americans (penalized B-Spline edition)

Mike Spagat sent me an email with the above heading, referring to this paper by Leontine Alkema and Jin Rou New, which begins:

National estimates of the under-5 mortality rate (U5MR) are used to track progress in reducing child mortality and to evaluate countries’ performance related to United Nations Millennium Development Goal 4, which calls for a reduction in the U5MR by two-thirds between 1990 and 2015. However, for the great majority of developing countries without well-functioning vital registration systems, estimating levels and trends in child mortality is challenging, not only because of limited data availability but also because of issues with data quality. Global U5MR estimates are often constructed without accounting for potential biases in data series, which may lead to inaccurate point estimates and/or credible intervals.

We describe a Bayesian penalized B-spline regression model for assessing levels and trends in the U5MR for all countries in the world, whereby biases in data series are estimated through the inclusion of a multilevel model to improve upon the limitations of current methods. B- spline smoothing parameters are also estimated through a multilevel model. Improved spline extrapolations are obtained through logarithmic pooling of the posterior predictive distribution of country-specific changes in spline coefficients with observed changes on the global level.

The proposed model is able to flexibly capture changes in U5MR over time, gives point estimates and credible intervals that reflect potential biases in data series and performs reasonably well in out-of-sample validation exercises. It has been accepted by the United Nations Inter-agency Group for Child Mortality Estimation to measure countries’ progress in reducing U5MR, and to evaluate their performance with respect to Millennium Development Goal 4.

Excellent stuff (except for the y-axes that go below zero, the ridiculously big font of the headers in Figure 4, and Tables 2, 3, and 4 (yuk!)). I’m really glad to see this stuff becoming generally accepted. Spagat says it’s being used by “that bastion of Bayesian Analysis – UNICEF.”

My patriotic former colleagues at Berkeley would be spinning in their hypothesis tests were they to see this.


  1. I should note that your image is a Soviet black helicopter, which is even more sinister. Is Bayesianism a Communist plot?

  2. jrc says:

    I really wish I had more time to look at this paper today, but since I don’t I was hoping I could out-source a question. When using DHS data to calculate annual child mortality rates, there is a spike in number of births about 5 years before the survey (children under 5 are subject to a long extra questionnaire and measurement, so enumerators/households will often just say the child is actually 5). This messes with the projected fertility/mortality rate 4-5 years before the survey was taken (reference below*, and I’ve seen it in some of my work), because only living children have their ages shifted in this way. The spline could smooth this out if you have multiple rounds of DHS data (so that another survey’s time series also provides an estimate for those funny years) but I wonder if that was something the author’s were thinking about and/or specifically address – I didn’t see it in my first pass through. My gut feeling, and my experience looking at DHS U-1 mortality rate time-series, is that this is a problem when you want a really fine resolution, but less of a problem in the broader trends except for at the far-right end of the graph. But if you are trying to predict past the far right….

  3. John Mashey says:

    Well, I see that Ted Cruz thought that UN Agenda 21 was created to have the black helicopters come and take Texans’ golf coruses away (or something like that), but I had no idea the perfidy extended this far.

    But this is not the first black helicopter invasion of statistics.
    It reminds me of a well-known statistician, whose report to Congress referenced an article by Tom Valentine
    ‘”MAGNETS In Your Future” was an obscure fringe-science magazine, for which Valentine wrote articles and later served as Editor. He had a long history of writing on fuel-less engines, psychic surgery (books, see Amazon) and conspiracy theories, for a tabloid, The National Tattler. His Bio states of that work:
    “(Miracle editor—had to come up with a miracle a week!)” …
    His later talk show often promoted “black helicopters” conspiracies:

  4. K? O'Rourke says:

    > an informative prior distribution was used, based on an analysis of these biases in the previous 2012 round of UN IGME estimates

    I believe the explicit modelling of biases (enabled by Bayesian machinery) makes this more notable than just being “Bayesian”.

    It will also make it a tough read to discern if they have succeeded in doing more good than harm as that will depend mostly on the models and especially the priors for those biases.

Leave a Reply