Skip to main content

What do final results look like? (Employee polls)

Overview for when final results and their reports are made available

Once a Survey peer group reaches at least three verified responses, the Requester’s dashboard begins showing partial results. Results are always released in safe batches of three or more responses.

When a Survey closes (all slots filled), both the Requester and Respondents receive access to the full dataset and report.

You can read more about how we release partial results here

When a Poll has at least 2 survey peer groups completed, it starts to show its own results report (we don't technically show a poll report when it only has one survey peer group completed - because, well, there's nothing to roll up). If a poll has more than 2 surveys and at least 2 are completed but some survey peer groups are still collecting answers, we already start to show the Poll results view: We show it with those surveys that have closed, and later, when your outstanding survey peer groups also close, the roll-up poll view will automatically update with the additional results of the surveys that have eventually closed.

Report structure

Poll reports and survey peer group reports are similar in structure since one (the poll report) is a roll-up of many (survey peer group reports). Each report includes:

  • Individual anonymized response rows (one per respondent)
  • Median and mean values for each compensation field
  • Distributions and ranges (including low / high quartiles)
  • Aggregations by company size, with recalculated medians and means for each of our four company-size groups (when enough data exists)

Respondents see the same data as Requesters — there is no “lite” version.
The only difference is timing: Requesters see high-level results update as batches are released; the Requester and Respondents see the final report with all data once the Survey closes.

In poll report views, even if we are representing multiple surveys in one view, we do mark which survey a given result came from. This allows you to see, on a given chart, that a particular data point came from survey #1, while the dot near it came from survey #2. If you ever, of course, want to only see the report of Survey 1 without the visual noise of other surveys, you can view the Results report of that specific survey. That always remains available

How we choose subgroups for analysis

All analysis choices are, by definition, opinionated. There are reasonable ways to slice this data differently, and we expect these approaches to evolve over time.

Depending on the nature of the data point, we analyze results using one of two subgroup lenses:

  • organization size
  • total compensation quartile

Organization size is used when a benefit or compensation component tends to change meaningfully as organizations grow. This is why organization size is a mandatory field for all polls.

For example, certain benefits become more common or more generous at larger companies, making company size the more informative dimension. In other cases, organization size adds little explanatory power.

When organization size is less relevant, we instead analyze results based on total compensation quartiles, which often better reflect how compensation-rich an environment is overall.

That “given X” choice — organization size or compensation quartile — is how we decide how to analyze prevalence and quality.

How we analyze certain benefits

For some benefits, we apply a bit of internal “secret sauce” to allow meaningful computation.

Healthcare-related benefits, when provided, are evaluated based on both prevalence and relative quality. Respondents rate these benefits on a scale (e.g. none / standard / good / best) relative to their local market. Internally, we assign scores to those labels, which allows us to:

  • Compute summary statistics
  • Compare benefit quality across compensation quartiles
  • Represent benefit strength in a way that’s comparable across respondents

How numerical benefits are presented

Numerical benefits — such as time off, parental leave, or retirement matching — are typically analyzed by total compensation quartile, not company size.

In the backend, we calculate baseline (“base 100”) medians or averages, then compare subgroup medians or averages against that baseline. This gives you a directional signal showing whether, given overall compensation levels, benefits tend to be better than expected, average, or below expectations.

Why we used total comp quartiles for this analysis

Tightly matched peer groups are a core reason Salary Confidential surveys are meaningful in the first place. While some spread in total compensation is expected, we assume that respondents within a survey share broadly comparable responsibilities and scope.

Because of that, total compensation becomes a useful first dimension for understanding relative positioning: this peer is paid more than that peer, despite having similar scope.

If we relied on organization size instead, we’d often end up with flat or uninformative analysis. Many well-crafted Salary Confidential surveys deliberately target peers working at similarly sized organizations — in those cases, organiztion size simply doesn’t introduce meaningful variance.

By using total compensation as our “given X,” we preserve analytical signal even when respondents come from different organizations of similar scale. This allows you to meaningfully compare compensation practices across peers — even when they work at organizations of roughly comparable size — without losing resolution in the data.

A note on non-response

Missing or zero values are not used in calculations. However, we always show response counts.

For example, we’ll indicate if only one respondent provided data for a specific benefit in the 50–75% compensation quartile — and that four respondents fall into that quartile overall. This helps you assess how much weight to give a particular figure.

Data fields with more re-identification risks are presented separately

Certain freeform text fields (location, 'additional details') present higher risks of re-identification for the respondents. We represent their data not attached to the specific individual compensation record, but at the survey peer group level instead. We also randomize the order that the content of these fields appear relative to the order of the compensation records so they cannot be 'lined up.'

Certain metrics are blended indices; certain metrics can be suppressed for privacy

Certain contextual metrics are potentially de-anonymizing: Company stage, company age (which we collect for polls that are using our extended interview around equity compensation) are collected but not given back in reports. Instead, they get blended to create custom indices that tell a story relevant to understanding equity compensation.

Organization size is collected as a pure number from respondents, but given back as a band and the band is gently 'fuzzied' so its boundaries are probabilitistic (in other words, our "Small" band is really a "small-ish" band size because at the edges, a small-size company can sometimes be banded in Medium, or perhaps not). We do this because Company size has the potential to also de-anonymize respondents, so we present it in results report in blurry ways that represent 'order of magnitude', rather than perfectly known. Also, when we detect scenarios that are too likely to becoming de-anonynizing (a single company size record being very isolated from than other results), we suppress this information altogether from the report. The report will let you know if this happened, and you will see "Organization size withheld"

A little extra lift for your own numbers

Survey reports include an optional comparison tool.

If you’re evaluating an existing offer — or your current package — you can enter your own numbers directly in your browser (nothing is sent to us) and see how they compare to the dataset.

You could do this in Excel, of course. We just figured we’d save you a few clicks — and maybe give you a quiet confidence boost before a conversation with your manager or recruiter.

This comparison tool currently applies to compensation only. We don’t support benefit comparisons yet — that territory gets complicated quickly, and for now we’ve chosen not to go there.

Come chat with us on Reddit

More than many other parts of Salary Confidential, how we present reports benefits from user feedback.

We can’t anticipate every way you might want to analyze your data, and we know there are cultural and contextual nuances we’re still learning.

If you have thoughts, questions, or ideas, come join the discussion at
reddit.com/r/salaryconfidential

Disclaimer: Your report may not include every possible view or chart

The Requester decides which questions to include in a survey, beyond the base set of required compensation data (including so-called “fair value” numbers — read why these are mandatory). We have extended question sets for equity compensation, benefits information, performance pay, freelance pay -- all of these are optional extensions

As a result, while this article describes the full range of reporting features available on Salary Confidential, it’s normal for a given report to include only a subset of them. If a question wasn’t asked, there’s nothing to analyze — and the corresponding charts or views won’t appear.

Some data collection is also market-specific. For example, you won’t see “401k cap and match” if you’re using a European poll model. In some cases, there is no direct equivalent benefit in a given market; in others, we simply don’t yet have a question that maps cleanly to local practices.

So if a report feels “lighter” than the examples discussed here, it’s almost always because of choices made upstream when the poll was designed — not because anything is missing or hidden.

Finally, Salary Confidential ships new features regularly. This article reflects the current reporting capabilities, but you may not see every view if your poll was created before a given feature existed. Polls are designed by the Requester based on the platform’s capabilities at the time of creation, and when new question types or report views are introduced, we don’t retroactively modify the design of existing polls.

Updated May 2, 2026