Delayed, but it’s done – this will be the final post of commentary on the research Mark Whalley and I conducted last year and had published in SSR In Depth this summer. It’s open-access so please do have a look and let me/us know what you think. We’re also presenting on the data in January at the ASE Annual Conference.
In the earlier posts of this sequence I discussed the context and the data we collected, as well as the recommendations I’ve built on what we found. Please note that Mark and I collaborated on the paper but these posts are my opinions, some of them going significantly further than the data supports in isolation.
- Why we thought it was worth asking the questions about job satisfaction and retention
- Context, questions asked, negative and attrition factors
- Reported retention factors and how they vary between low and high-risk groups.
- Recommendations for colleagues in different positions
This post will be have a slightly different focus; I’m going to describe the challenges I had with data-collection, in the hopes it will help others avoid the traps I fell into, and think out loud about further research that might be fruitful and/or interesting. I am open to collaboration either as part of my day job or in my own time, so please do get in touch if this is something you’d like to discuss.
Data collection
- Teachers are really bad at completing surveys.
- Longer surveys get fewer responses.
- The busier a teacher is, the less likely they are to complete a survey.
- There aren’t many physics teachers in England in the first place.
Taken together, these facts mean that it’s hard to get enough responses to treat the results with confidence. The solution we came up with was an incentive, and after some discussions I secured the budget to provide a voucher to every person completing the survey. This obviously cost more than a prize draw, but avoided the complications of ensuring randomisation etc. To provide a choice, we gave the option of donating the money instead (to Education Support). So we set up the questions on one anonymous survey, linked the completion page to a separate one to gather contact info, and shared the link on social media.
Roughly an hour later we had over 200 responses, the vast majority of which were filled with random garbage and a procedurally generated email address in a transparent attempt to get the voucher. Because of my inbuilt cynicism – I mean, I have three kids and spent ten years in a classroom with other people’s teenagers – the voucher process was not automated. We identified the fraudulent attempts, started a fresh survey and tried again.
This time it took a couple of hours to be discovered, but we got about a thousand submissions. There was Latin in the free text boxes and random numbers in the data fields. A handful looked genuine but most were clearly bot-generated. A later version which was only emailed still got corrupted, suggesting the link had been shared online somewhere the spammers had access to it.
We considered asking participants to use a school-based email address for the voucher, but decided this would reduce people’s confidence in the anonymity. (Although I should emphasize the two surveys were separated by design so it was impossible to link the answers to the participant, even for me.) The platform we used, strangely, didn’t have a built-in Captcha option as a question you could include. We did look at whether we could use an external service for that to enable the voucher request, or a trap question, but it was too complicated.
In the end, there was only one solution – putting my pattern recognition up against the scammers. I ended up figuring out boundaries for several of the questions and having Excel grade the answers as green (possible) amber (dubious) or red (clearly a scam). For example, a participant who claimed to be at a school with 64 students but alongside 20 science teachers was unlikely to be genuine. Although it was frustrating, it was also really interesting when I got into the data, especially as several columns showed a smooth bell curve rather than the clusters we’d expect to see in the real world.
- very short time taken for survey completion
- inconsistent participant qualifications (no GCSEs, no A-levels, but doctorates in physics and engineering? really?)
- unrealistic school numbers, student and staff
- incompatible reported FTE vs timetable load
- random or repeated phrases in the free text fields
In the end, I filtered around 5000 responses down to just under 100; those, with any possible identifying information such as time of completion removed, were shared with Mark for the data analysis.
My advice: when creating an online survey, set up at least some questions to allow impossible answers and plan on how to identify them quickly. Free text fields will be auto-filled by some bots and this can provide an obvious clue. Decide in advance what your yes/maybe/no ranges will be.
Next steps
I’m still looking at the dataset, for example to see whether there’s a correlation between responsibility roles and job satisfaction – in particular if colleagues are less grumpy about SLT if they’re already a middle leader. I think it would be really interesting to see whether any of the same issues show up for subject specialists who have to teach out of their exact specialism but within the same department. How to French teachers feel about a Spanish-heavy timetable, for example? I’d like to know how other factors, specifically age and gender, affect both job satisfaction and attrition factors. Recent work done looking at the exodus of women in their thirties from teaching suggests to me that alongside the recommendations in both the published paper and my commentary, you can make a big difference for those teachers by offering flexible working and matched timetables. We didn’t ask those questions because it counts as sensitive data and triggers a whole new world of GDPR and ethics trauma.
If I was doing this again, I’d simplify the questions and ask colleagues to prioritise the possible changes. If they could only have three or five changes from a shortlist, which would make the biggest difference? The benefit of this is that we could separate the possible changes based on the tiers I used in my last post, so HoDs find out which of the things they have control over are worth fighting for.
One of the things we looked at was deprivation score, and it didn’t make as much difference as we expected. (There were noticeably more physics specialists in less deprived areas though.) I’d really like to see the teaching unions investigate this angle to see if there’s a relationship between the deprivation index of a teacher’s home compared to their workplace. How does this vary by subject and seniority? For example, for some years I travelled from a relatively high deprivation area to teach physics in a leafy suburb. I’m prepared to bet that although it’s noisy, there’s a similar signal nationally.
I’d like to repeat my suggestion from the last post; we need a national programme of exit interviews for all departing physics teachers. Why do they go and where are they going? What might change their mind? Link it to an ongoing job satisfaction survey and we can see how much difference complaining about workload in years 2 and 3 makes to the chances of them leaving after year 5.
Give me some money and I’ll run a nationwide anonymous exit survey for every physics specialist leaving a state school. I’ll find out where they’re going and why. I’d hope schools are doing this now, but why on earth isn’t there a standardised set of questions for every departing teacher part of the Ofsted requirement for a school? Don’t add it to the league table, but anonymised to a regional level this would be a valuable tool. Add a threshold for X% of teachers ticking the same box for a particular school which should be a warning sign for SLT and Ofsted. (Heads Round Table, call me.)
Final thoughts
I really hope someone has been reading these posts – maybe there will be an influx as term restarts. Or maybe I should give up on blogging the old way and create a substack, but that just feels weird. If you’d like to discuss any of these ideas, please add a comment here, email me or find me on social media; I’m currently experimenting with BlueSky.




