Scope a piece of analysis on production data

[emily-prudence-dfe]

We'll be importing old data from DQT and want to get a feel for what kinds of data problems we might encounter and how to deal with them.

For example, what do we do:

• when induction period start dates are before end dates?
• when induction periods for the same teacher overlap? (started looking at this in #575 but definitely unfinished)
• with really old records where the ECT has already completed induction - do we need these?

[joe-harrison-dfe]

Next Steps:

• Speak to Nathan to understand what bad shape data there is currently and what to look for
• Speak to Rob to understand potential size of the problem with bad shape data.
• Map out the potential problems and initial thinking of solutions.

[gracecbrennan]

Made a start on this but need to pick up next steps.

[gracecbrennan]

Wants to check in with Nathan and Rob on this

[joe-harrison-dfe]

Updates from discussion with Rob:

For the 'bad shape' data occurrences:

• start date later than end date - normally down to human error, some shortcomings in data return process but majority of the time comes from updating a record following a discussion with an AB and an error occurs.
• IP start date year in the future- Rob imagines the number of these records being quite low, although there is no block for ABs returning a start date in the future. One year in future is quite a long time and should be marginal.
• Duplicates - For any given TRN there can be duplicate on the AB and the start date but something else that differs, this is due to a shortcoming on DQT. Duplicates are quite prevalent but are normally from certain ABs. Numbers of this will probably be in the 1000s
• IPs where an LA is the AB still without an end date and associated outcome - This will probably just reduce over time as there are a number of records where the AB has just forgotten to provide an update.
• In a lot of these cases it won't be hard to fix moving forward at all. It is more jus the extent of problem to take across as DQT isn't great.

Next steps:

• speak to Nathan to understand bit more about the data cut he used to observe these results
• understand more about our scope for importing DQT records to better size the potential problem
• Once scoped can start to think about impact and if specific solutions are necessary.

[joe-harrison-dfe]

Chat with Nathan:

• Used all IPs of anyone who had served Induction since 3rd September 2021, since ECF reforms kicked in, People who had started induction a long time ago but hadn't completed it. They will likely be in a worse shape.
• DQT records fall into two types - overall induction stuff (complete/fail/ongoing) and then IPs. Top level migrated into the TRS. We will be migrating the IPs into the CPD service but only the periods that are relevant to teachers that haven't yet completed. For partially completed inductions, we will need to migrate over any terms they completed induction in.
• Currently working with the records that have LAs as ABs to record any outcomes so they don't have to step in and do it. Might consider just recording them as ended.
• No cut off for people who have done any induction, want all partially completed records.
• CPD only has concept of teaching school hubs as ABs. 150 ish LAs down as ABs so we will have to migrate them across to CPD and teach CPD what an LA is.
• 500+ teaching schools still need to be migrated. Not a current option but need to reflect back to a current AB.
• 1 National provider that might have some attributed to that.
• Start date > End Date - more often when an ECT does induction across multiple schools, ABs typically get start dates from the school. The school often answers when they started induction as when they started at their particular school.
• Start date year in the future, - very few but often keyboard error where they mistype a year. Might need some period of cut-off for registering start date in the future. Or a pop up asking if you are sure. Need validation around the entry there. Biggest gap for most records was around 6 months (a term)
• Duplicates - In terms of statutory guidance, can only be serving induction with one AB at one time. Start date should never overlap across two ABs.
• Not really any other common issues from the analysis. Opportunity to implement some simple validation for the migration and not do like for like replacement.

Next Steps:

• Talk to Rob about conducting analysis on DQT records to size numbers of 'bad shape' data
• Discuss validation rules to implement to resolve some issues (no start dates after end dates etc.)

[gracecbrennan]

Rob will be coming back in the next week or so with results on this.

[claire-hughez]

Joe has chased Rob, not this week. Got wednesday as a deadline.

[gracecbrennan]

Joe is going to write up findings here