Synopsis

School districts operating under the COVID-19 pandemic have adapted quickly to collect information on how each student is participating in instruction for each school day. Is the student “remote”, “on campus” or in a “hybrid” modality, a mix of on campus and remote instruction? Even within these categories, there are sometimes variations in how the student is participating, such as “remote asynchronous” and “remote synchronous.”

In this document, we refer to the method of student participation in instruction as the student “learning modality.”

Learning modality data is obviously critical operational data today, but will remain important data as districts and state education agencies consider outcomes, efficacy, and equity of instruction in the current school year and beyond.

This Working Draft provides guidance on Ed-Fi aligned and integrated data models to capture learning modality. It describes a continuum of three methods designed to support the needs of both local and state education agencies, as well as to support a focus on both actuals and planned modality time.

This document represents a further iteration of the models presented in the whitepaper Marked Present: Recommendations for Building a Framework to Measure Attendance Data Across Learning Models published by the Ed-Fi Alliance.

General Discussion

There are three Ed-Fi-based data designs proposed by the Alliance to capture modality: a Program and Calendar approach, an Attendance approach, and a Program Aggregate approach.

• The Alliance believes that the Program and Calendar method should be the top option considered by agencies if possible. 
• The Attendance approach is also a strong and viable option. The main reason that it is not the primary recommendation is that our research suggests that this model will in some cases require more effort and business logic on behalf of SIS systems, as school districts have not generally captured modality natively in this design. However it is possible to map from the Program and Calendar model into this modal.
• The Program Aggregate approach should be avoided if possible. It is provided as a solution to cases where one part in the data exchange should not (for policy or other reasons) receive the more granular student data. This occurs in state reporting in some cases for example.

We base this recommendation on our research from the Marked Present whitepaper: that research suggests the Program and Calendar data model aligns best to the models in which modality data has been stored in actual student information systems (SIS) in school districts and to the models developed by SIS systems for learning modality data. It is therefore likely simpler for most student information systems to implement data exchange based on this model.

There are a few reasons why a continuum of options are covered rather than attempting to align on a single model.

1. We have found that organically all these models have emerged in field work. That is so because agency and ecosystem strategies have naturally taken different approaches on how to manage data and drive data quality during a very challenging time.
2. The models have different strengths and limitations, and therefore each may be more appropriate to an agency’s needs and goals based on where that agency is today and the choices it has made already in its data management.
3. Some agencies, such as state education agencies (SEA), will have policy reasons that prevent usage of the primary proposed model. For example, some SEAs will be required to gather aggregate data for policy or practical reasons, or to use attendance-based approaches to ensure maximum fidelity to actuals.

Resources

A set of Visio files of these models has been made available: Ed-Fi Working Draft 7.vsdx

Program and Calendar Method

In this approach, the Program data element is used to capture the basic “learning cohorts” into which students are divided – e.g., “remote”, “on campus”, etc.

Students are associated with these Programs via StudentProgamAssociation. If a student moves between programs (e.g., a “remote” student goes “on campus”) new records for StudentProgramAssociation are written and/or updated with each move, capturing the dates of these changes (via StudentProgramAssociation BeginDate and EndDate).

Figure 1: Elements of the proposed data model for Program and Calendar and Program Aggregate method. New elements shown in green. Click to expand.

It is further recommended that there be a Program for each cohort that is on a different calendar schedule.

For example, let us imagine a fictitious school district called Grand Bend that has 4 such buckets: remote students, on campus students, and 2 groups of hybrid students that rotate days on and off campus (“hybrid A” and “hybrid B”). The school might have programs (ProgramNames) that look like this:

• “on campus”
• “hybrid A”
• “hybrid B”
•  “remote”

In this manner, these cohorts may not map 1-to-1 with learning modalities: in this case, there are 2 “hybrid” cohorts, each modeled as a separate program.

That may not seem ideal, but there are two reasons for this approach. First, as we will see below, it allows one to connect the learning cohort to calendaring information specific to that cohort. Second, the experience of the “hybrid” cohorts may end up being qualitatively different, so it is potentially useful to capture this aspect of the student learning modality.

To capture where the cohort was on each calendar day, the CalendarDate element in Ed-Fi has a collection of ProgramLearningModalityType elements. This element consists of a reference to a Program and a collection of LearningModality elements. 

Figure 2: The connection of CalendarDates to Programs and Learning Modality is performed via the ProgramLearningModalityType collection. New elements shown in green. Click to expand.

 LearningModality is a component entity with 3 fields designed to capture the modality and the amount of time in the modality:

1. ModalityType – a descriptor, e.g., “on campus”, “remote”
2. ModalityTime – a decimal, e.g., 100, 25, 5.5, 65
3. ModalityTimeType – a descriptor desiring the units for the time: e.g., “hours, ”“minutes” or “percentage"

Using these entities, each date on the calendar can be annotated with an account of the modality for each learning cohort on each instructional day.

For example, a single calendar date (e.g., October 10, 2020) for the fictional Grand Bend school district might be described using these facts:

• On October 10, 2020, the “on campus” program was “on campus” for 440 minutes
• On October 10, 2020, the “remote” program was “remote” for 440 minutes
• On October 10, 2020, the “hybrid A” program was “remote” for 440 minutes
• On October 10, 2020, the "hybrid B" program was "on campus" for 200 minutes"and "remote" for 240"minutes

To show how these statements would map into the data model, here is a chart showing the mapping of elements of these facts:

Note that the last rows show why LearningModality is a collection on ProgramLearningModalityType: this is needed to capture cases in which a cohort has multiple modalities on a given day. Note also that this chart and its facts show modality measured in minutes, but this could also be performed in fractions of the instructional day - e.g., "the “on campus” program was “on campus” for 100% of the day.

Program Aggregate Method

This method is likely more appropriate for state education agencies seeking to gather aggregate totals. As mentioned above, in some cases agencies may be required by policy to gather aggregates only, and there are also important reasons, such as data privacy, to have options that limit the data on individuals that is collected and or exchanged.

For capturing the aggregate learning modality of a cohort, this method uses a LearningModality collection attached directly to the Program (LearningModality is the same element as presented in the "Program and Calendar" method above).

Figure 3: Elements of the proposed data model for Program Aggregate method. New elements shown in green. Click to expand.

Using this element, one can express facts such as:

• The program “on campus” was “on campus”  for 1400 minutes per week
• The program “remote” was "remote"  for 100% of the semester
• The program “hybrid A” was "on campus" for  40% per week
• The "hybrid B" program was "on campus" for 60% of the semester and "remote" for 40% of the semester

* Note that in actual data exchanges it is a best practice to use decimals to reflect percentages, i.e., 1.0 = 100%, 0.45 = 45%, etc.

Note that in this case the LearningModality.ModalityTimeType is combining units and durations, similar to a rate: “minutes per week” “percentage for the semester” etc.

Obviously, this aggregate capture, while likely good enough for some cases, will not provide a precise picture of student modality when modality of individual students changes. If students move between instructional cohorts, the aggregate totals at the program-level will likely not translate exactly to how the student participated, even when factoring in the dates on which students changed programs/cohorts.

Attendance Method

Attendance-based capture will likely be more rare than the program-based approach, as our research suggests that most agencies are not capturing data in this fashion in their source systems. It is nevertheless a strategy that has been adopted by some, and has some advantages: data is tracked at a very granular level, and it focuses strongly on capturing actuals, and so can be the most accurate method in a time where there are many shifts of individual student modality if well-operationalized.

This method requires that a campus use positive attendance tracking; that is, not only are absences marked (“negative attendance”), but that attendances are also marked.

In this model, the agency simply uses the same LearningModality collection on StudentSchoolAttendanceEvent, to express facts such as

• On October 10, 2020, student A was ‘on campus’ for 100% of the day”
• On October 10, 2020, student B was ‘on campus’ for 120 minutes and ‘remote’ for 360 minutes”

Figure 4: Elements of the proposed data model for Attendance method. New elements shown in green. Click to expand.

Sample records using the attendance approach might look as follows: