How is organization size handled while preserving anonymity?

Organization size is an important piece of context in compensation data, but it can also be identifying in small, targeted samples.

Salary Confidential is designed to preserve the usefulness of organization size without exposing it in a way that could be traced back to an individual. Because it is one of the most deterministic element of context we collect, we treat it with special care. In certain cases, we may end up never displaying it back in your survey results.

We do not expose exact organization size

We collect organization size as a number, but we never show it directly.

Instead, we represent it as an approximate category (“Small”, “Medium”, “Large”, “Enterprise”).

And we intentionally do not publish the exact boundaries of these categories. Because:

The boundaries are not strictly enforced
Results near the edge of a category may appear in an adjacent one
A organization shown as “Medium” may in reality be at the upper end of Small, or the lower end of Large

This reflects a simple truth:

If there was a boundary size at 20,000 between "XL" and "XXL" (we don't have such bands), the differences at the boundary (for example, 19,500 vs 21,000 employees) are not meaningfully categorical — even if strict banding would suggest they are.

Categories are “soft”, not exact

Organization size categories are designed to be directionally informative, not precise.

Near boundaries, results may shift between adjacent categories
This uncertainty is intentional and part of the privacy model

As a result:

You cannot assume that a result in a category definitively belongs there as a strict number. The math we do makes the drift between categories be proportional so 2-person organization cannot vault into anything we'd call "Medium", but we don't want you to get too attached to the details of this math, which we keep deliberately opaque.
Categories should be interpreted as approximate scale, not exact thresholds

Ordering within a category is not meaningful

Within each category, we do not preserve any meaningful internal ordering.

When results are grouped under the same category, they should not be interpreted as ordered by organization size
Two companies in the same category may differ in size, but that difference is intentionally not represented

In practice, organization size alone is a weak signal for comparison — many other factors matter (industry, growth, structure, geography).

Preserving precise ordering within a category would add limited analytical value, while increasing the risk of re-identification in small samples.

So we prioritize: order of magnitude over precise ranking

When we suppress organization size entirely

In some cases, even approximate categories are not safe to show.

If a result is too isolated from the rest of the group, we suppress organization size entirely.

Because category boundaries are intentionally soft, nearby categories can sometimes provide cover. A result near the edge of a category may plausibly appear in an adjacent one, allowing it to blend with neighboring responses.

However, this only holds when that overlap is realistic.

If a result remains clearly distinct from the rest of the group — even after accounting for boundary fuzziness — it is treated as unsafe.

For example:

One respondent works at a very small organization
All other respondents work at much larger organizations

Even if those organizations fall into an adjacent category, the gap may still be too large for meaningful overlap. In that case, the smaller organization remains effectively identifiable.

Requesters know who they invited, but Salary Confidential does not know whether:

multiple respondents from smaller organizations were invited but only one responded
or only one such respondent was ever part of the sample

Given that uncertainty, we default to the more sensitive case: when a result cannot plausibly blend with others, we suppress organization size for the entire peer group.

This is part of how we protect respondents in targeted surveys, where the audience is known and context can be used to make inferences.

But even when we suppress organization size from published results, we still use internally where it is safe

Although we do not expose organization size directly, we do use it internally where it adds value and does not create risk. In this case, we use precise organization size numbers (no need to blur them), and we use them regardless of whether the organization size band is shown on its own in final results

For example: in surveys that include extended equity questions, organization size can be used as one factor in blended metrics (such as liquidity-related indices)

Because organization size is combined with other variables, it cannot be isolated or traced back to a specific respondent, so we can always use it precisely and safely.

Why we don’t show organization size in partial results

We never show organization size in partial results: we cannot yet determine whether the final dataset will be safe as results begin to come in.

Organization size is only evaluated for display once the survey is complete.

The decision to hide organization size is evaluated at the peer group level

Safety is evaluated at the survey peer group level, not globally.

If organization size is unsafe in one peer group, it is suppressed for that peer group
This does not prevent other peer groups (within the same poll) from showing organization size if they are safe

However, we do not 'bring back' organization size for an unsafe peer group, even if, at the overall poll level, the combined data might appear large enough. Since survey peer groups are always available as a filtered result view, the global poll results view doesn't change the underlying risk for a peer group with an unsafe scenario

So a scenario is unsafe at the peer group level, it is treated as unsafe in all results views

A note about data visualization in results reports: our charts adapt when organization size is being withheld

In certain cases, we entirely remove a data visualization when organization size is a key organizing principle ('Pay relative to organization size') for a survey peer group results view. There's just nothing to show anymore.

In other results views, we present this survey's results in their own 'size withheld' group: this allows the pay data to exist on the same scatter plot as the other peer groups, but yes, this data can't purely roll up with the other peer group(s) in the poll.

In all cases, we make it clear if you are viewing a data visualization that expects organization size but isn't able to use for a subset of results.

Summary

Organization size is collected as a precise input, but never exposed directly
Results are shown as approximate, soft categories
Category boundaries are intentionally not strict, this means that as a organization size nears for example the upper-end of "small", it may end up being shown as a medium-banded organization. Or vice-versa.
There is no meaningful ordering within a band.
Organization size is used internally when safely combined with other factors
It is suppressed entirely when even approximate representation is not safe

This approach allows us to preserve useful context while protecting respondent privacy — even in small, targeted datasets.