2018-08-26 Analytics Middle Tier Special Interest Group
Pre-Read Materials
Ed-Fi Analytics Middle Tier Design Proposal (PDF)
Agenda
Welcome & thanks
Brief overview of the proposal
Q&A - clarification of the proposal
Discussion topics:
Can you envision using these views in real-world projects?
Are there other use cases that would be immediately useful, beyond Early Warning System?
Thoughts on sharing your solutions with the community?
Are you concerned about data quality with real time analytics?
Attendees
Notes
Q&A
Are you considering temporal / multi-year in this model? → Not at this time, ODS as it is. Views could be modified to take any customizations into account.
EWS model is predictive. Would be nice to compare year-over-year → Yes this is a good point. Anyone could easily extract the EWS data into a warehouse to look at the impact of interventions and/or assess the efficacy of the predictive model.
Are there performance concerns with these views? → In a word, yes. Best practice will be to load the data elsewhere before letting the BI presentation layer access it - not supporting direct query mode from presentation to ODS. Two possible solutions:
Replicate / log ship / restore from backup to another copy of the ODS
Many BI tools have their own "data mart" under the hood that can pull data in from source systems, with storage in highly efficient structures (i.e. column store).
Is there a plan for an API on these views? → Not currently being considered.
Which version of the Ed-Fi Data Model will this be built on? → initially supporting the 2.x data model. If/when demand is proven out, will work on Data Model v3.x support.
Can you add last modified date to the dimensions, to support polling for updated records? → Yes
What do you think about fully flattening the fact and dimension data into a single object? → Some modern platforms are extremely efficient at handling data in this way, but we need to support a broad range of solutions. Keeping with the star/snowflake provides that broader support, while allowing the BI data modeler to either transform into a flattened table or use calculated columns to give the appearance of a flat table within the analytics engine.
What about multi-tenant support for educational service centers that have multiple districts? → We should be able to support this. Stephen will talk to Sherod Keen to talk about additions to the dimensional data model.
Does the security model support the concept of district staff (e.g. Assistant Superintendent) overseeing multiple schools? → Yes, this is feasible.
Assuming that you have an "Assistant Superintendent" classification descriptor, map that to the "School" scope instead of "District" scope in the
StaffClassificationAuthorizationScopetable.So long as your SIS loads the API's
staffSchoolAssociationproperly for these individuals, theUserAuthorizationview will have one row per school for each Assistant Superintendent, while continuing to have a single row withSchoolPermission = "All"for the Superintendent.
Is it possible to support charter networks, where there is oversight for schools in multiple districts? → We should be able to add this. Stephen to talk with Ed Comer.
Caveat: the Alliance is not sure if anyone is using the charter school capabilities in the ODS/API yet.
In a generalized solution it would help to have dimensions for classroom and subject, and to add state summative test results along with the classroom grade. → Good suggestions, will look into it.
The EWS model developed by the Alliance in 2017 relied only on SIS-provided data, but now that we've had a strong push for Assessment vendor support, we should be able to add the assessment results to the warning system.
Can the views support all of the data from the Ed-Fi Dashboards? → Not at this time, as we're releasing a minimum viable product aimed at self-serve analytics.
For more on Ed-Fi Dashboards, see Call for Participation: Join our Working Group on Analytics & Visualizations
Special Topics
Can you envision using these views in real-world projects?
Many "yes" responses
Post-meeting response: In addition to directly using them, they should provide a good hands-on education for how to query the ODS.
Are there other use cases that would be immediately useful, beyond Early Warning System?
None mentioned, group will think about it.
Thoughts on sharing your solutions with the community?
Experiment with ways to share.
Importance of keeping views generalizable / componenitized for improved re-usability.
Post-meeting response: suggestion to use GitHub for work in development and the Exchange for milestone releases.
Are you concerned about data quality with real time analytics?
Some expressions of concern. Melissa Straw offered to comment on this in the future.
In the class room, there is a feeling that near-realtime is more important than perfect data.
Post-meeting response: might not be a good idea to provide analytics to end-users "in the early days of ... an implementation". Wait till the installation is mature. In the views, limit the returned columns to those which are well-supported.
Next Steps
Alliance working on:
Incorporating feedback into the design
De-normalized views
Quicksight starter kit
v2 of PowerBI starter kit
Guidance on using the views
Special Interest Group to meet again on Sept 7 at 2 pm.