Dominic Cummings has got something right

His analysis of what is wrong in government is often right, but his solutions are too often magic bullets that won't help. Government as a whole and number 10 specifically do need better access to expert analysis to formulate better policy. But the problem isn't a lack of sophisticated modern decision making tools but a lack of capacity to do even the most basic analysis relevant to policy options.

Dominic Cummings' blog-as-recruitment-advert has become somewhat infamous. And not just because he wants to put "...data scientists, project managers, policy experts, assorted weirdos…" at the heart of government. Tom Chivers wrote a fair review. Jonathan Portes suggested the problem was the politicians not the advisors. Some others were more dismissive.

In this latest blog and many previous ones he has severely criticised the quality of government decision making. In this blog, for example, he argues, I think quite correctly, that :

Political ‘debate’ and the processes of government are largely what they have always been — largely conflict over stories and authorities where almost nobody even tries to keep track of the facts/arguments/models they’re supposedly arguing about, or tries to learn from evidence, or tries to infer useful principles from examples of extreme success/failure.

This sounds right. 

But the Cummings solution is build a team of experts in a whole bunch of advanced tech tools like artificial intelligence, advanced forecasting models and game theory. This is where I think his solutions are not going to solve the problem.

Yes, much government decision making is broken. Yes, there is a terrible lack of capacity to do good analysis of big problems. But the solution that will have a big impact is to improve the basics, not to bring in a plethora of advanced high-tech magic bullets. The assumption that we need radical new techniques to solve difficult problems is wrong. Sometimes simple things work.

I can explain why with some references to history and my own experience of working with government of the big problems of the NHS.

He is right that there is a deficit of capacity at the centre of government to do analysis and deliberation

First, an example of where he is right. In the book The Blunders Of our Governments, Anthony King and Ivor Crewe make these relevant points. 

The first point where Cummings would entirely agree is this:

In other words, what we observed – and still observe today – is not a sequence of unrelated episodes but a pattern. It would seem to follow that, if the incidence of blunders is to be reduced, it is the British governing system, and the ways in which people function within that system, that needs to change.

They also make another point that Cummings has not put much emphasis on: there is no point in having the right policy if you don't understand how to implement it.

No feature of the blunders we have studied stands out more prominently – or more frequently – than the divorce between policymaking and implementation and, in human terms, between those who made policies and those charged with implementing them.

Another point where, it seems, their analysis and Cummings analysis agrees is that Number 10 does not have much capacity to do good analysis (which is presumably why he is recruiting more people who can do it).

...Certainly the size of the British prime minister’s staff – those individuals with significant political clout who work directly to and with him or her – is minuscule by international standards….

...The heads of government of almost all European countries have more robust staff support than prime ministers in the UK.

To express this in different language, the UK PM finds it hard to scrutinise the policies proposed by their ministers either for realism when the ideas are proposed or for progress as they are being implemented because number 10 simply doesn't have access to the capacity or skills to do those jobs. 

This is a critical issue that Crewe and King argued in 2013 and Cummings argues now. Both good decisions about which policy to pursue and good monitoring of the implementation of the policy require a great deal of capacity for analysis. 

The consequences of not having this capacity at the centre of government is epitomized by the Lansley clusterfuck of an NHS reorganisation. Neither Cameron nor anyone else in government paid any attention to what Lansley was doing until very late in the day and, even when panic ensued when they found out, the changes did little to mitigate the consequences.

Better analytical capacity at the centre of government can yield big results

But things have not always been like that.

The Blair government made some major changes to the NHS too. But, unlike Lansley, some of those changes worked and led to very visible improvements. Waiting lists for elective treatments were reduced from years to weeks; Delays in A&E were reduced from catastrophically poor to amazingly good (for several years >98% of patients were admitted or discharged within 4 hours of arrival). 

I was involved with some of the efforts to deliver those changes and the details of what happened are why I support the Cummings plan to add more capacity at the centre of government but am skeptical about the details.

The unit that was instrumental in making the changes happen was the Prime Minister's Delivery Unit (known as the PMDU). This was staffed by experts, many externally recruited. And they were not afraid to bring in other outsiders for specific tasks. I was one of those advisors, originally brought in because I knew a decent amount about system and process modelling. My original task was to build models to test whether the 4hr target was feasible (the PMDU had already worked out that the target would have significant clinical benefits and would transform public perception if the NHS could meet it). This seems like the sort of thing Dominic Cummings is aiming to do: use advanced techniques to solve hard problems.

But it didn't take me or the PMDU long to work out that advanced techniques were not required. The problem wasn't an issue of system design in an unavoidably complicated system, it was the much simpler problem that almost nobody had ever bothered to collect even the most basic data about why long waits happened. So most hospitals had no idea which causes of delay were the most common. Was it waiting for staff availability? Was it finding a free bed? Was it delivering medication from the pharmacy? Was it a slow patient registration process? Was it a slow triage process? Nobody knew. And not knowing meant nobody knew where to start looking to see what could be improved.

So I developed a tool to help hospitals collect that data. A simple tool based in Excel supported a selection of hospitals to identify the major causes leading to long waits just by collecting a little bit of data from every patient for a whole week. The tool gave instant feedback (with simple but carefully crafted data visualisation) about where they should focus their attention if they wanted to reduce the waits (for a more sophisticated later example of the sort of insights available see my blog on analysing A&E data). Combined with other clinical and process expertise in the PMDU this became a big part of the plans individual hospitals developed to fix the problem. I worked with the PMDU, DH and many hospitals for about 3 years to identify their local problems and to ensure they implemented the fixes (For most of that time I produced the weekly national performance reports for the DH/PMDU/NHS team so they could monitor the success of the interventions and direct attention to laggards and backsliders). 

We learned many lessons about improvement. Some require a little insight into process management and even queueing theory but nothing a schoolchild couldn't learn. The biggest lesson requires no technical expertise at all: the most important delays in A&E are caused by things outside the A&E (no free beds is a downstream hospital problem that can't be fixed by anything the A&E department can do, a lesson that seems to have been forgotten by many as I have repeatedly pointed out). Streaming minors to a simpler, faster process than majors can radically improve speeds for everyone compared to having a single queue where the minors keep getting bumped in priority by the far less frequent majors. This doesn't need a complex model to explain when you see it working.

In 2002 more than 30% of patients waited more than 4hr to be discharged or admitted (as I write this the december 2019 performance is now worse than this for the first time since then) but by 2005 fewer than 2% of patients waited this long. Having the capacity at the centre of government to analyse the problem and implement effective solutions was critical to this improvement. 

The key lesson for me (and the one that applies to Dominic Cummings efforts) were that this huge improvement was a result of a handful of simple things done well not the result of shiny advanced tools. The keys were:

• Know the right change to aim for (PMDU know how concerned the public were about A&E performance and they had good clinical evidence to persuade staff than changes would be worth doing for clinical reasons not just for better public feedback)
• Collect the right data to identify what the problems are and where they occur and use simple analysis to focus on the key problems
• Understand the detail of the process you are trying to improve and the changes to it that are realistic
• Spread that expertise to the local teams who actually have to get things done
• Monitor the progress of the changes to check that they work and intervene with further support if local teams falter

In other words, a huge change was delivered with simple analysis and good project management peppered with a little understanding of how A&E departments worked.

In the last decade the performance in A&E has declined and is now as bad as it was before these changes were made. And we now, again, have a widely held belief that A&E performance is an intractable problem. It isn't. We once knew the key solutions and they were effective.

And one of the key reasons we are currently struggling is because what was learned in the 2000s has been largely forgotten by the leadership in the NHS and at the centre of government.

The decline was kicked off by Gordon Brown when he made little use of the units at the centre of government set up by Blair. This was compounded by Cameron who abolished them on the basis that they were just a bunch of central bureaucrats he could easily do without. A decade later, we can see that this was a very serious mistake.

Cummings is right in general but wrong in detail

My conclusion is that Cummings idea to put more capacity at the centre of government decision making is a good one. But his emphasis on clever analytical techniques and new thinking is wrong. We didn't need to introduce fancy new tools (Artificial Intelligence, Machine Learning, agent-based process simulation or advanced forecasting tools) to solve the A&E problem. We just needed to do very basic analysis properly.

The problem isn't a lack of clever, sophisticated modern tools: it is a lack of the most basic analysis. If Number 10 doesn't have a basic understanding of what is happening, government will make bad decisions. The key to achieving big improvements is not advanced techniques, but doing the basics right. 

What Cummings seems to be proposing–more analytical capacity at the centre of government–is good. But his focus on clever new techniques is misguided. The problem in the way British government works is not a lack of advanced analysis: it is a lack of the most basic analysis. Fix that rather than searching for magic bullets.