Link by AgEcon
Random allocation for experimental studies.
Define your arms and how likely each one is — with weights, and with groups inside groups if you need them. You get one address. Everyone who opens it is sent to one questionnaire, chosen at random, and you get the record of who went where.
Free, takes a minute, no password. Don’t have an ORCID iD? Get one here. Participants never sign in — they just click.
How it works
Define the arms
One row per questionnaire, with a weight. 1 and 2/3 means three fifths
against two fifths — the weights are ratios, not percentages you have to balance. Need
“half to control, the other half split in two”? Put the two inside a group.
Share one link
Email it, put it in a panel, print it as a QR code. Participants need no account and see nothing: they land on the questionnaire. A code travels with them, so their answers can be matched back to the arm.
Report the allocation
The page shows how many went where, intended against actual. The record downloads as a CSV, which is what a referee will ask for when you write that participants were randomly assigned.
What it does differently
Balanced by construction
With a 1:1 design and sixty people, an independent coin flip lands 38–22 more often than you would like. Link reduces your ratio to whole people, shuffles that block and deals from it: every completed block puts the arms exactly on target, while the order inside the block stays unguessable.
Previews don’t break your design
Paste a link in a chat and something opens it to build the preview; some mail systems open every incoming link. Each of those visits would have eaten a slot and skewed your arms before the first real participant — and the totals would still have looked right. Crawlers and HEAD requests get a preview page and no assignment.
Nobody shops for an arm
Someone who reopens the link gets the same arm as before, and does not take a second slot. The return is counted, because how often people came back is a fact about your study — not an annoyance to hide.
No addresses kept
To tell a returning visitor from a new one you only need to know that something repeats, not who it belongs to. Link stores two one-way fingerprints — one for the device, one for the network — and neither can be turned back into an address. No email is ever collected.
Questions
How do I set unequal probabilities?
Weights are ratios between sibling arms. 1 and 2/3 gives 60% and 40%.
You can write whole numbers, fractions, decimals or percentages — 3,
7/4, 0.4, 40% — and mix them. Nothing has to add
up to 100.
Can I nest groups?
Yes, up to four levels. A group takes its own share first, then divides it among its arms. So “half to control, the other half split 1:2 between two treatments” is three arms at 50%, 16.7% and 33.3% — and you never have to work those numbers out: the builder shows them as you type.
How do I match answers to arms?
The redirect adds pid (a random participant code) and arm to the
questionnaire address. In Qualtrics these arrive as embedded data, in Google Forms as a prefilled
field, in LimeSurvey as URL parameters. Anything already in the incoming link — a panel’s
respondent id, a campaign tag — is passed through too, so you don’t lose it.
Can I change the arms after I have started?
No, and that is deliberate. People already assigned were assigned under the old rules; changing the design halfway gives you two studies that cannot be analysed as one. Close this study and create another.
Who can see the results?
Only you. Unlike a Slot poll, the study page is not visible to whoever holds the link: a participant who could see how the arms are filling might work out their own assignment. Details are in the privacy notice.