On prediction, Nate Silver’s “The Signal and the Noise”

Title The Signal and the Noise: The Art and Science of Prediction
Author Nate Silver
Publisher Penguin UK, 2012
ISBN 1846147530, 9781846147531
Length 544 pages

Available on Amazon Use Google Books to read the first chapter.

RD Comment: Highly recommended reading. Reading this book reminded me of M&E data I had to examine on a large maternal and child health project in Indonesia. Rates on key indicators were presented for each of the focus districts for the year prior to the project started, then for each year during the four year project period. I remember thinking how variable these numbers were, there was nothing like a trend over time in any of the districts. Of course what I was looking at was probably largely noise, variations arising from changes in who and how the underlying data was collected and reported.This sort of situation is by no means uncommon. Most projects, if they have a base line at all, have baseline data from one year prior to when the project started. Subsequent measures of change are then, ideally, compared to that baseline. This arrangement assumes minimal noise, which is a tad optimistic. The alternative, which should not be so difficult in large bilateral projects dealing with health and education systems for example, would be to have a baseline data series covering the preceding x years, where x is at least as long as the expected duration of the proposed project.

See also Malkiel’s review in the Wall Street Journal (Telling Lies From Statistics). Malkiel is author of “A Random Walk Down Wall Street.” While a positive review overall, he charges Silver with ignoring false positives when claiming that some recent financial crises were predictable. Reviews also available in The Guardian. and LA Times. Nate Silver also writes a well known blog for the New York Times.

Predicting the achievements of the Katine project

September 2010: This post provides information on a revised proposal for a “Predictions Survey” on the achievements of the Katine Community Partnerships Project, a project managed by AMREF and funded by the Guardian and Barclays Bank, between 2007 and 2011.

Background Assumptions

The Guardian coverage of the Katine project has provided an unparalleled level of public transparency to the workings of an aid project. As of August 2010 there have been approximately 530 articles posted on the site, most of which have specifically about Katine. These posts have included copies of project documentation (plans, budgets, progress reports, review reports) that often don’t enter the public realm.

Ideally this level of transparency would have two benefits: (a) improving UK public knowledge about the challenges of providing effective aid, (b) imposing some constructive discipline on the work of the NGO concerned, because they know they are under continuing scrutiny not only locally, but internationally. Whether this has actually been the case is yet to be systematically assessed. However I understand the effects on the project and its local stakeholders  (i.e b above) will be subject to review by Ben Jones later this year, and then open to discussion in a one day event in November, to be organised by the Guardian.

So far there have been two kinds of opportunities for the British, and other publics, to be engaged with the public monitoring of the Katine project. One has been through posting comments on the articles on the Guardian website. About 30% of all articles have provided this opportunity, and these articles have attracted an average of 5 comments . The other option has been by invitation from the Guardian, to make a guest posting on the website. This invitation has been extended to specialists in the UK and elsewhere.  Multiple efforts have also been made to hear different voices from within the Katine community itself

The Predictions Survey would provide another kind of opportunity for participation. It would be an opportunity for a wide range of participants to:

  • to make some judgments about the overall achievements of the project, and
  • to explain those judgments, and
  • to see how those judgments compared to that of others, and
  • to see how those judgments compare to the facts, about what has actually been achieved at the end of the project

In addition a Predictions Survey would provide a means of testing expectations that greater transparency can improve public knowledge about the challenges of providing effective aid.

My proposal is that that the Prediction Survey would consist of five batches of questions, one for each project component, on a separate page. Each question would be a multiple choice question, but associated with an optional Comment field. People could respond on the basis of their existing knowledge of the project (which could vary widely) and/or extra information about the website obtained via component specific links embedded at the head of each page of the online survey e.g. on water and sanitation. Questions at the end of the survey would identify participants’ sources of knowledge about the project (e.g. obtained before and during the survey, from the website and elsewhere).

A 1st rough draft survey form is already available to view. Any responses entered at this stage may be noted, but they will then be deleted and not included in any final analysis.  The final design of the survey will require close consultation with AMREF and the Guardian.

Intended participants in the survey

  • UK public, reached via the Guardian
  • Uganda public, reached via Ugandan newspapers (likely to be more of a challenge)
  • AMREF staff, especially in Uganda, Kenya HQ and UK
  • The Guardian and Barclays, as donors
  • Monitoring and Evaluation specialists, reached via an international email list

Hypotheses (predictions about the predictions)

  1. We might expect that AMREF would be able to make the most accurate predictions, given its central role. But aid agencies are often tempted to put a gloss on their achievements, because of the gap that sometimes emerges between their ambitions and what can actually be done in practice.
  2. We might expect that participants who have been following the Guardian coverage closely since the beginning might be better informed and make better predictions than others who have become interested more recently. But perhaps those participants are still responding on the basis of their original beliefs (aka biases)?
  3. We might expect M&E specialists to make better than average predictions because of their experience in analysing project performance. But perhaps they have become too skeptical about everything they read
  4. We might expect the Guardian and Barclays staff to make better than average predictions because they have been following the project closely since inception and their organisation’s money is  invested in it. But perhaps they only want to see success.
  5. We might expect the highest frequency choices (across all groups) to be more accurate than the choices of any of the above groups, because of a ” wisdom of crowds” effect. The potential of crowdsourcing was of interest to the Guardian at the beginning of the project, and this survey could be seen as a form of crowdsourcing – of judgements.

This list is not final. Other hypotheses  could be identified in the process of consultation over the design of the survey

There may also be other less testable predictions worth identifying. For example, about the effects of this Prediction Survey on the work done by AMREF and its partners in the final year up to October 2011. Might it lead to a focus on what is being measured by the survey, to the detriment of other important aspects of their work?  If AMREF has a comprehensive monitoring framework and the prediction survey addresses the same breadth of performance (and not just one or two performance indicators) this should not be a problem.


The fourth and final year of the project starts in October 2010 and ends in October 2011.

The finalisation of the design of the Predictions Survey will require extensive consultation with AMREF and the Guardian, in order to ensure the fullest possible ownership of the process, and thus the results that are generated. Ideally this process might be completed by late-October 2010

The survey could be open from late October to the end of March 2011 (six months before the end of the project). All responses would be date stamped to take account of any advantages of being a later participant

A process will need to be agreed in 2010 on how objective information can be obtained on which of the multiple choice options have eventuated by October 2011.

A post 2011 follow up survey may be worth considering. This would focus on predictions of what will happen in the post-project period, up to 2014, the year of the vision statement produced by participants in the September 2009 stakeholders workshop in Katine.

“In 2014, Katine will be an active, empowered community taking responsibility for their development with decent health, education, food security and able to sustain it with the local government”


The participation of the Guardian and AMREF will be very important, although it is conceivable that the survey could be run independently of their cooperation

Assistance with publicity, to find participants, would be needed from the Guardian and Barclays

Advisory support is being sought from the One World Trust

Advisory support from other other organisations could also be useful

The online survey could be designed and managed by Rick Davies. However responsibility could be given to another party that was agreed to by AMREF, Guardian and Barclays.


  • The survey design needs to be short enough to encourage people to complete it, but not so short that important aspects of the project’s performance are left out
  • The description of the objectives used in the survey needs to be as clear and specific as possible, but also keep as close to AMREF’s original words as possible (i.e. as in the 4th year extension proposal, and using the M&E framework, now being updated)
  • Participants will be asked to make a single choice between multiple options, describing what might happen. These options will need to be carefully chosen, so there are no obvious “no brainers”, and to cover a range of plausible possibilities
  • It may be necessary in some cases (e.g. with some broadly defined objectives) to allow multiple choices from multiple options
  • I have heard that AMREF will be conducting a final evaluation in late 2011, using an external consultant. This evaluation could be the source of the final set of data on actual performance, against which participant’s predictions could be compared. But will it be seen as a sufficiently independent source of information?