A physician survey result is a vote with turnout.

The report you buy prints the turnout on page two, in small type, and never mentions it again. This one puts it in the header of every page, publishes the scoring, shows you every question before you field, and hands back every raw row on the day it is collected.

Set up your hospital Open the console

Field monitor · a seven-site health system · demo data

Medical Staff SurveyIn field · closes 15 Sept

36%

332 of 923 respondedtarget 50%

The mark on the track is the target this hospital set for the result to count as a mandate. 130 more responses reach it. Margin of error ±4.3 points, finite population corrected.

Questions in the core, four to eight minutes, fixed so that every year and every hospital compares

36%

Live response on the demo campaign, 332 of 923, against the 50% target this hospital set

Reporting groups, of which 15 fall under the floor and are rolled up rather than blanked

Respondents below which no score renders, printed on every report so it cannot be quietly changed

A report that shows its working

Every figure carries its base. Every withheld cell carries the reason it was withheld. Every composite carries the internal consistency it achieved in your sample, not the one quoted in a manual.

Every figure on the right is the demonstration campaign as the console computes it, down to the withheld specialty and the level it was reported at instead. The same numbers export to CSV, and an exported figure that disagrees with the one on screen fails the build.

Medical Staff Survey 2026

a seven-site health system · all facilities · reporting floor 5

Turnout

Invited923

Responded332 (36%)

Margin of error±4.3 points, FPC applied

TargetNot met. 130 short of 50%

Engagement

Favorable55.1% of 1,627 answers

Mean3.60 of 5

Alpha0.83

Neurological Surgery

Respondents1

Shown atService line, Neurosciences

Shown at service line level (Neurosciences, 9 respondents) because Neurological Surgery had 1 respondent and the reporting floor is 5.

Independent physicians answered at 19.4% against 36% overall, the only group materially under-represented. Read every organization-wide figure with that in mind.

Six decisions somebody made on your behalf

None of these is an accident of technology. Each one is a choice, and each one is more comfortable for the vendor than for the hospital paying for it.

You sign before you read

The items arrive after the contract, and the ones that get retired take your trend line with them. Here you read every question in the console before a single one is fielded.

The scoring is not published

A percentile arrives with no method attached. The first independent psychometric study of the leading physician instrument appeared thirty-two years after it went into use.

Turnout is a footnote

The response rate sits on page two in small type, and every percentile after it is computed as though turnout were irrelevant.

Half the specialty rows are blank

Suppression is applied and then nothing is offered in its place, so the chief of staff who asked about their own service gets an empty cell.

Your own questions cost extra

Custom items are billed. So are extra cuts, and on one public price list, so is reading the free-text answers your own staff wrote.

Your history is not yours

Last year lives on their platform. Leaving means leaving the trend line behind, which is the whole reason nobody leaves.

Where this differs

The norm

Here

Item bank visible only after contracting

Every question shown before you field, with the reason it is asked

Scoring method proprietary

Published in full at /method, and the code is the documentation

Percentile rank as the headline

Percent favorable with its base, the mean beside it, and a Wilson interval

Drivers by undisclosed index

Johnson relative weights, with the model R-squared printed beside them

A withheld cell can be recovered by subtraction

Complementary suppression, so it cannot

What this does not have yet

A national physician benchmark

There is no norm set here, and there will not be one until five organizations have contributed. Until then your comparison is your own prior year, imported from whatever vendor you were with, and the console says which of the two it is showing you. A percentile built on two hospitals names the other hospital.

An outside audit

No third-party penetration test and no SOC 2 report. The confidentiality rules are enforced in code and each one is proved by a check that fails the build, which is a different kind of assurance and not a substitute for the first two. If your organization needs either, it gets scheduled rather than claimed.

Nobody else can price the answer

Some instruments ask whether pay feels fair. None of them can say what the answer is worth, because none of them sits next to compensation benchmark data. This one does, and the difference is a finding you can act on rather than a score to admire.

My compensation is fair for the work I actually do.
I believe my compensation is competitive with what I could earn elsewhere.
I understand how my compensation is determined.

The third scores worst in most medical staffs, and it is the one that can be fixed in a quarter without spending a dollar. The second raises a question that needs benchmark data to answer, and Fair Market Fast already holds it. That is the sentence no other survey vendor can finish.

Everything that produces a number on this platform

FavorableTop two on a five-point scale. The not-applicable count is kept in view rather than dropped from the base.
CompositesThe mean of each respondent's own item mean, not the mean of the item means. Those differ whenever people answer different subsets, which in a routed instrument is always.
ReliabilityCronbach alpha computed in your sample and printed. Below 0.70 the composite is flagged and you are told to read the items instead.
SuppressionA floor you set, printed on every report, plus complementary suppression so a withheld cell cannot be recovered by subtracting the visible ones from the total.
DriversJohnson relative weights. The model refuses below sixty complete responses and refuses outright when two items are effectively identical.
IntervalsWilson score, not Wald, because favorable percentages in small cuts sit near the ends of the range.
ComparisonWelch's t, corrected with Benjamini-Hochberg, and the correction is stated.
Margin of errorFinite population corrected. A medical staff is a small closed population, and the uncorrected formula makes a census look like a poll.
Quality flagsStraight-lining, low variance and speeding are computed and reported, never silently dropped.

Written out in full at /method, and proved by a runnable check rather than asserted. The arithmetic is tested against hand-derived and published values, not against its own previous output, because a snapshot test of a statistics library certifies only that it is consistently wrong.

Read the method before you decide anything

The scoring, the suppression rules and the demo console are open. The instrument itself is not published: every client reads all of it in the console before a single question is fielded, and nobody else reads any of it.

Set up your hospital Read the method