£42k for a top-class software engineer? It's no wonder uni research teams can't recruit
Lack of recognition and support for quality engineering part of science's reproducibility problem, MPs hear
Lack of support for software engineering is holding back efforts to improve reproducibility and openness in the UK's scientific research, a panel of MPs was reliably informed yesterday.
With data analytics at the heart of a great number of leading-edge scientific fields, the need for support from software engineers has never been greater, but it is sadly lacking, according to two researchers speaking to the Parliamentary Science and Technology Committee.
[Universities] think: 'Oh, you're an IT person. We've got a grade for that. How many people do you manage? None? In that case, it's £42k a year' ... and that just doesn't work
Dr Jessica Butler, analytical lead and research fellow at the University of Aberdeen, said universities were not set up to support people like software developers or statisticians, who have no ambition to be published in a "fancy" academic journal.
"They want accurate statistics. They want to write code that works. They could easily work for a company and get paid six times as much and get promoted," she said of the engineers.
Dr Ben Goldacre, director of the University of Oxford's Nuffield Department of Primary Care Health Sciences, told MPs that his team employed software engineers who could get "very good low six-figure salaries" outside academia.
"They're willing to halve their salary to work with me if it's an interesting project. They're not willing to decimate it down to £30k or £40k," he said. That's $40k or $53k.
The problem was university pay grades do not recognise the skills of software engineers, Goldacre added.
"A full-stack, commercial-grade software developer is a different set of skills to the person who fixes your printer when it's broken. [Universities] think: 'Oh, you're an IT person. We've got a grade for that. How many people do you manage? None? In that case, it's £42k a year' ... and that just doesn't work," he said.
Goldacre said his team had managed to "hack" the system by using money from research to employ its own developer and build software tools other academics can use, but the practice was not commonplace.
"There is almost no open competitive funding available anywhere for anybody who wants to build these kinds of intermediate [software] tools. They are intellectually and academically creative, and incredibly powerful and high impact in terms of the good that they do, but there's almost no open competitive funding to do that kind of work, so that kind of work is not done," he said.
Software engineering in these teams was usually done as a "hobbyist activity" by scientists whose main roles lay elsewhere. Although they were capable, they were not a substitute for qualified experienced software engineers, he said.
"It's research software engineers that are at the heart of this new way of working. You need people who are really good at writing software, not people who are researchers who dabbled a little bit here and there. It should be an independent, high status and legitimately funded activity."
The problem of recruiting and retaining software engineers links back to the so-called reproducibility crisis in science.
- Academics tell Brit MPs to check the software used when considering reproducibility in science and tech research
- Hubble Space Telescope restored to service: No repeat of those missing messages, but here's a software patch anyway
- What would machine learning look like if you mixed in DevOps? Wonder no more, we lift the lid on MLOps
- Genetically modified E coli bacteria produce ink for 3D printing programmable objects
In its written evidence to the committee, joint academics group the Software Sustainability Institute said about 69 per cent of research is produced with specialist software, which could be anything from short scripts solving a specific problem and complex spreadsheets analysing collected data, to the millions of lines of code behind the Large Hadron Collider and the Square Kilometre Array.
"With many studies, research published without the underlying software used to produce the results is unverifiable," the submission said.
Goldacre told the committee today: "The true and complete story of how you produce your results is the code that you wrote, attached to the data that you executed it across. But unfortunately, our model of how we publish scientific findings was built in the 19th century." ®