I am (when I have access to the Internet) somewhat glued to the FiveThirtyEight coverage of the US presidential election (their elections podcast is also awesome). At the heart of their coverage is “The Model”. They take data from published opinion polls, feed this into their model and then publish the likelihood of Trump or Clinton winning the race. The model allows them to understand and, to a certain extent smooth, the inherent biases and uncertainties in different polls with different techniques. They also factor in none polling factors (like the state of the economy).
So what they have is a simplified version of the United States. It’s based on some theories about how polling data relates to actual turnout validated against data from previous electoral races.
We are poised to find out how good a model it is.
Models are useful, possibly vital, in terms of understanding the world. We all have models in our head. In order to be able to predict with any likelihood the impact our actions will have on others, we need a model of how they are likely to behave. Luckily our brains handle all this for us in the background so we don’t have to be conscious about it.
If you ever imagine that an action you undertake will have a particular consequence you have developed a model.
I’m pretty confident that when I write a news release it will be picked up and used by a journalist. This is, of course, because of my great skill and experience. It’s also because I have a model of how newsrooms work, of the values of my target journalist, of the sort of writing style that will appeal to them.
This model was originally handed to me in my professional training and I have improved and refined it based on the experience of 20 years of communications roles. It’s a pretty good model. It’s not foolproof but most of the time its good enough.
We tend not to be very explicit about our models and that can be a problem. If we don’t tell others the basis of our models they find it hard to assess how confident they can be of our predictions. It’s also harder for others to understand how new data mesh with our existing model (and reinforce it or suggest it should be changed). And they can’t challenge the fundamental assumptions underpinning our model. Which is something we should welcome.
This may not matter that much in terms of getting news articles in local papers. But it really matters when we talk about the impact that changes to the funding of social care (for example), or housing allocation policy, or educational selection at age 11.
We have subconscious models for all these aspects of social policy. They are probably poor fits for reality. If we really want to improve services for the public we need to state our models, show how they fit the evidence and be ready to change them in the light of new data.