Methodology: how QA rates are sourced and graded
Formula version 1.0.0: 1928 accepted evidence entries, 79 rejected, 0 excluded as outliers; 25 published rate records; 456 withheld routes or figures.
From source to published figure
- Collection: one country per collector run; every candidate page is logged, accepted or not.
- Verification: a verifier opens each URL, confirms the quoted figure on the page, and sets the verification date or a rejection reason.
- Computation: weighted percentiles under the versioned formula.
- Thresholds: a rate, salary or overhead record needs at least 2 independent sources; engagement model pages also list figures with 1 source, marked single-source (not a confidence grade); pages need minimum counts per type. Withheld items are on the not-published page.
Median, p25 and p75
Median (the middle value of the weighted observations for a combination), p25 (the 25th percentile: a quarter of weighted observations sit below it) and p75 (the 75th percentile: three quarters of weighted observations sit below it) are weighted percentiles of the accepted observations for one combination.
Confidence grades
Every rate, salary and overhead record carries a confidence grade (A, B or C, set by observation count, source classes and spread under formula 1.0.0); single-source figures on engagement model pages carry none:
- A: high confidence: at least 3 observations from at least 2 source classes and a spread ((p75 minus p25) divided by the median) of 0.6 or less.
- B: medium confidence: at least 2 observations from at least 2 classes and a spread of 0.9 or less.
- C: indicative: at least 2 observations from 1 class, or a spread above 0.9 with at least 2 classes.
- Single source: not a confidence grade; used only on engagement model pages for figures with fewer than 2 independent sources, with the source count printed.
Refresh cadence
FX refreshes each quarter; rate and salary sources expire after 18 months, overhead sources after 24 months, provider facts after 12 months. Changes are dated on the changelog. Corrections: data@qaratescompare.com; Corrections are reviewed within 5 working days.
Detail pages: source classes and rejections.
Questions
Where does the QA rate data come from?
Each rate, salary and overhead record needs at least 2 independent sources with a verification date (engagement model pages also list figures with 1 source, marked single-source); 1928 evidence entries are accepted in data version 1.0.0.
How is the confidence grade for a QA rate assigned?
Grade A needs at least 3 observations from 2 source classes and a spread of 0.6 or less; grade B needs 2 observations from 2 classes and a spread of 0.9 or less; grade C covers 2 observations from 1 class or a wider spread.
What formula turns rate observations into a published median?
Formula 1.0.0 takes weighted percentiles (p25, median, p75) of accepted observations after the outlier rule, with monthly fees divided by 160 hours.
How are outlier rate observations handled?
An observation above 2.5 times or below 0.4 times the running median of its combination is kept in the evidence file and excluded; 0 observations are excluded as outliers in this version.
How old can a rate observation be before it is excluded?
Rates and salaries older than 18 months, overhead rates older than 24 months and FX rates outside the current quarter are rejected.
Why are some country and role combinations not published?
456 routes or figures are withheld in this version, each with its reason on the not-published page.
Figures are medians of published observations converted to USD at the quarterly ECB reference rate shown in the footer. They describe market ranges on the verification date, not quotes. Monthly and annual figures assume the hours stated on this page. Confidence grades are defined on the methodology page. This site does not recommend providers.