Schools testing has been in the news again recently… are SATS etc useful objective measures of a school’s performance? Or do they add unnecessary stress and bureaucracy?

Well I think we can all agree more objectivity and less stress are good things, and most of us would probably go further and say that SATS aren’t doing either of those jobs. But kids are so unique! And testing is so essential! How on earth can we do both?!

Well, sorry. If there’s one field that is actually good at summarising hundreds of thousands of individuals in a heterogeneous population, it’s biology. So here’s A Biologist’s Alternative to SATS. Let’s call it… STATS:

• Pick 5-10 measures that are easy to test and cover a wide range of measurable markers of kids’ lives – say, a couple each of literacy and numeracy tests, some critical thinking, standard IQ and general knowledge. Plus, happiness / wellbeing and physical health.
• Assemble a mixed team of inspectors, governors, academics and teachers. Have them sample, say, 20 schools from a wide range of areas and rank them.
• Then test the kids in those schools using our metrics. Also collect information on their dates of birth, sociological factors (parents’ status, wealth, postcode, commuting distance, screen time – there’s loads of ways to do this), etc.
• Now we can construct a GLMM (a slightly-but-not-too complicated statistical model – or else use machine learning stuff like HMMs or neural networks, although I suspect getting enough data would be hard) to model each kid’s scores as a function of their school’s ranked quality given their sociological background.
• Here’s the important bit: we take the test scores of the 25th, 50th, and 75th percentiles and label them ‘below’, ‘on’ and ‘above’ average respectively. But we won’t translate these expected quartile scores directly into national targets because we know the makeup and weighting of school sizes and types across the country will vary greatly and nonlinearly.
• Instead the model itself provides a national benchmark, not a standard. This will be used to model the expected scores for a given school (and students) given the same sociological information, most of which can be imputed from child benefit statements, addresses and the like.

Why would this system – more complex to set up and quite data-intensive – be any better than the current one? Here’s a few reasons:

1. We know development is multifactorial. So is this model.
2. We know sociology greatly affects kids’ life chances, so let’s explicitly account for it. If the upshot on that is more effort alleviating poverty than endlessly tweaking the school system, great.
3. We can publish the tests’ relative weightings in the model so teachers/parents know which should be more emphasised.
4. Grade inflation would be easy to abolish, simply by updating the model every year or so.
5. The grading of schools would be simpler and integrated. Most schools will be ‘on-average’ – this is implicit – so the horrific postcode lottery will end and parents can agree to focus on improving their local school, which is better for their commute and their kids’ sanity.
6. Regional or municipal variations due to differences in sociology will also be apparent, and can be evidenced and tackled.