ROI data overview

The model needs two kinds of data:

How often an organization contributes to a project. This is used to infer how much value the organization perceives.
How much it invests, in units and dollars. This is used to value contributions and to compute ROI.

This page lists the sources and how each is used. Detail on the conversions is in Contribution labor and Financial expenditures.

Sources                            Processed datasets                Used for
────────────────────────────       ───────────────────────────       ─────────────────────────
Activity records, meetings,        Contribution frequency            Stage 1: markups
event registrations, sponsor-      (monthly yes/no per pair)
ships, memberships, courses        Contribution levels               Stage 2: allocation, and
                                   (yearly quantity and price        inference (value, ROI)
                                   per type)

Unit of analysis

Everything is indexed by:

Organization (written i in the technical pages)
Project (j)
Time (t): months for participation, calendar years for contribution levels
Contribution type (k)

The universe is member organizations that have an LF account mapped to them and are affiliated with a project. Expenditure data is only available for these organizations.

Contribution types and sources

Type	Source	Quantity	Price	Fixed unit?
Code	LFX Insights activities	FTE from activity counts and hours	Code salary	No
Community	LFX Insights activities	FTE from activity counts and hours	Community salary	No
Meetings	Meeting attendance	FTE from attended hours	Meeting salary	No
Event attendance	Event registrations	Number of paid registrations	Average net ticket price plus event overhead	No
Event sponsorship	Event sponsorships	1	Total sponsorship	Yes
Project membership	Memberships	1	Pro-rated annual fee	Yes
Linux Foundation membership	Memberships	1	Fee split across key projects	Yes
Education courses	Course purchases	Units sold	Median net revenue per unit	No

Expenditure for a type in a year is quantity × price. For fixed-unit types the quantity is always 1, so the price carries the dollar amount. The model treats these types differently in estimation. See Stage 2: CES allocation.

How the data is used

Dataset	Used for
Contribution frequency	Stage 1: a monthly yes/no participation record per organization-project pair, which drives the markups
Contribution levels	Stage 2 and inference: yearly quantity and price per contribution type
Markup distribution settings	Calibration of markups, anchored to the LF Research Open Source ROI Survey

Contribution frequency (stage 1 input)

A monthly record with an active flag for each organization-project pair:

A month is active if the organization had any code or community activity on the project in that month. Code and community activity are pooled. Meetings, events, memberships, and courses do not count toward the flag.
A pair is included if it has at least one active month on or after the project's start date. The record starts at the later of the model's start date and the project start date and runs to the end of the observation window. Projects with no start date are dropped.
Each month also carries the previous month's active flag.
The stage 1 participation model uses organization and project effects, the previous month's active flag (to control for past activity), and the pair's investment scale and observation window (so that team size and project age are not mistaken for perceived value). See Stage 1: markups.

Contribution levels (stage 2 and inference input)

One record per organization, project, year, and type, restricted to records where quantity and price are positive. Records are kept only between a pair's first and last year with any contribution of any type, and within the observation window. A year in which a type has no activity simply has no record for that type, which matters for how complete years are defined in estimation.

Counting each resource once

Summing any breakdown of an organization's investment must never count the same resource twice. Three rules guarantee this.

A unit of work is credited to one project. An activity is credited to the project it was recorded against and is not fanned out to parent projects. The same contribution can still be recorded once per project, repository, and integration that ingested it. For example, Linux kernel history imported through a fork would make the fork look larger than the kernel. The model keeps one copy per contribution, preferring the upstream repository over a known fork. See Contribution labor.
Money is credited to the project that sold it, over its real term. Memberships are spread across the years of their real term. Event registrations, sponsorships, and meetings are recorded once, against the project that owns them. See Financial expenditures.
Foundations keep their own spend and do not roll up children. A foundation pair holds the foundation's own events, membership, and foundation-level repositories. Its sub-projects are separate pairs, so summing an organization's projects gives a true total.

Placeholder organizations such as "Individual - No Account", "Self-Employed", and "Unknown" pool unrelated people and are not organizations. They are excluded from estimation and results.

Observation window

The window starts on 2010-01-01 and ends at the start of the next month, so it always reaches the current month. Due to data availability, meeting attendance is limited to the last 24 months.

Calibration data

The target distribution of markups (see Model settings and defaults for its mean and standard deviation) is set to be consistent with the LF Research Open Source ROI Survey. It is a configuration input of the model, not data observed for each organization.

Known limitations

Hours per activity are domain-knowledge estimates, not measurements (details).
Salaries and overhead multipliers are flat parameters, not organization-specific.
Event currency conversion uses a fixed exchange-rate table. Registrations in other currencies are dropped.
The cost of a Linux Foundation membership is distributed evenly across an organization's key projects.
Meeting costs are attributed to the project where the meeting occurred, with no fan-out to parent projects.
Only member organizations with a mapped LF account are covered for expenditure data.