RESEARCH STANDARD
Research Methodology
Before anything is published under the word research it has to disclose its dataset, observation period, geography, sample size, metric definitions, exclusions, outlier treatment, the analyst, the reviewer and its stated limits. Those disclosures are decided and recorded before the analysis runs, because a criterion chosen once the numbers are known is a justification rather than a criterion. Work that cannot meet the floor is published as commentary and labelled as commentary, or it is not published.
- Dataset name
- Observation period
- Geography
- Sample size
Try it, edit values and click buttons
API request
POST /v1/payments
{
"amount": 24900,
"currency": "CAD",
"capture": true,
"metadata": { "order_id": "4821" }
}Response
Click "Send request" to see the API response
Processing payment…
201 Created
{ "id": "pay_7xK2m", "status": "approved", "amount": 24900 }Demonstration data only
What every benchmark must disclose
Dataset name
A stable identifier for the extract the figures came from, so a number in a chart can be traced back to the run that produced it and a later recalculation can be compared against it rather than quietly replacing it.
Observation period
Exact start and end dates. Not recent, not the last year, not trailing twelve months without saying which twelve. A period that is not pinned cannot be reproduced and cannot be compared against the next release.
Geography
The market the data was drawn from, stated at the level it was actually drawn at. Canadian data is reported as Canadian data; it is not restated as North American because the larger word sounds more impressive.
Sample size
The number of merchants or accounts in the population after exclusions are applied, given as a count rather than as a word like many. Where the count is small the count is still published and the limitation is stated alongside it.
Transaction count
The number of transactions the metric is computed over, which is a different number from the sample size and answers a different question. A figure computed over many merchants and thin transaction volume is fragile, and a reader is entitled to see that before relying on it.
Metric definitions
Every metric defined in full, including its denominator. Effective rate, for instance, is total fees for the period divided by total processed volume for the same period. A metric named but undefined means whatever the reader assumes, which is how two honest parties end up disagreeing about a fact.
Inclusion and exclusion criteria
Which records entered the population and which were removed, recorded before the extract is pulled. Criteria written after the results are known are not criteria, and an exclusion that appears only once a result is unwelcome is a finding about the analyst rather than the data.
Outlier treatment
Whether outliers were trimmed, winsorized or kept, at what thresholds, and what difference the treatment made to the published figure. Silent outlier removal is the most effective way to produce a number that is technically computed and practically false.
Mean or median, and which
Which central measure is reported and why it was chosen. Where mean and median diverge materially, both are shown, because in payments they usually diverge for a reason the reader would want to know about.
Aggregation level
The unit a figure is computed at: per transaction, per merchant, or per merchant-month. A per-transaction average and a per-merchant average of the same data are different numbers, and only one of them answers a given question.
Analyst
The function that produced the analysis, recorded on the page. Internally, a named individual is accountable for it, on the same basis as every other published page.
Reviewer
The function that checked the method and the arithmetic, which is never the function that produced them. A benchmark reviewed only by its own analyst has not been reviewed.
Last recalculated
The date the figures were last regenerated from source data. It is deliberately separate from the date the prose was last edited, because a page whose words changed and whose numbers did not is not a fresh benchmark.
Stated limitations
What the dataset cannot support, written where the findings are and not at the foot of the page in smaller type. A limitation a reader has to infer has not been disclosed.
The gate a benchmark passes before publication
The question and the criteria are written down first: what is being measured, over what period, for which population, with which inclusions and exclusions, and which metric answers it.
The extract is pulled against those criteria and the dataset is named, so the run stays identifiable after the chart exists.
The analysis is computed with the outlier treatment and the aggregation level applied as specified rather than as convenient, and both mean and median are inspected.
A reviewing function that did not produce the analysis checks the method, the arithmetic and the limitations, and can send it back rather than annotate it.
The full disclosure block publishes with the findings, and the last-recalculated date is set to the date the numbers were generated, not the date the page went live.
Related policies
Research data ethics
De-identified aggregates only, small-segment suppression, governance approval.
ExploreFact-checking policy
Where a number is allowed to come from.
ExploreEditorial policy
Who writes, who reviews, and what moves a review date.
ExploreComparison methodology
The scoring model behind every processor comparison.
ExploreCorrections policy
What happens when a published figure turns out to be wrong.
ExploreCheck your processing costs
An effective rate computed from your own statement rather than an average.
Explore
Questions about this policy: Research Methodology
Can I get the underlying data?
No. Published research uses de-identified, aggregated data and the underlying merchant records are not released to anyone. What is published is the aggregate, the method and the limitations, which is enough to interrogate a conclusion. The research data ethics policy sets out why the raw data does not leave.
Why will RapidCents not publish an average saving?
Because the number would not mean anything. What a business pays depends on its card mix, its average ticket, its channel split and its monthly and equipment charges, and an average across merchants who differ on all four describes none of them. Fee Check computes an effective rate from your own statement instead, which is a figure about you.
What happens when a published figure changes?
It is recalculated and the last-recalculated date moves. Where the earlier figure was wrong rather than superseded, the corrections policy applies as well and a correction notice is published on the page saying what the figure was and what it now is.
Is a vendor survey research?
Only if it discloses the same set: who was asked, how many answered, over what period, in which market, with the question wording and the exclusions. A survey reporting a percentage without its denominator is a marketing asset, whoever published it and however the chart is drawn.
Does RapidCents publish research about itself?
Yes, under the same standard and with the commercial interest disclosed. Being the publisher is not a reason to relax the disclosure requirements; it is a reason for a reader to apply them harder.
Take the next step
Talk to a RapidCents specialist
RapidCents Fee Check reads a processing statement and shows interchange separately from the markup. Upload a statement for an instant breakdown, or open a merchant account and start accepting payments on one account.
- No obligation
- Canadian payment specialists
- Secure statement upload
