The Diagnosis: Single-column forms won. Across 702 desktop participants, the single-column layout was completed in 102.3 seconds against 117.8 seconds for the multi-column version.
That gap of 15.5 seconds is 13% of the multi-column completion time, at p = 0.00412. The two 95% confidence intervals do not overlap, though they come within half a second of touching.
Baymard still finds extensive multicolumn layouts in 16% of the checkouts in its e-commerce benchmark. Ten years after this study, the mistake is far from retired.
Every form design guide tells you to use one column.
Almost none of them measured it.
The authorities everyone cites, Baymard and Nielsen Norman Group, publish qualitative usability findings with no sample size attached. The most-linked quantitative alternative is an eye-tracking study run on 8 people years before responsive design was standard.
Speero ran the comparison on 702 people. So here are the numbers for single or multi-column forms, and here is what they can and cannot settle.
What the study measured
Speero ran a single-variable comparison inside a live survey in June 2016, through CXL Institute.
Half of participants saw a linear, single-column layout. Half saw the same questions in a multi-column layout. Same questions, same wording, same field types, one difference.
356 people saw the single-column version and 346 saw the multi-column one, 702 in total, all on desktop. Results compared with a two-sample t-test.
Here is that output as text, because for ten years these figures existed only inside an image file and no search engine or answer engine could read them:
Single-column form. Mean completion time 102.338 seconds. 95% confidence interval 95.009 to 109.667. n = 356.
Multi-column form. Mean completion time 117.791 seconds. 95% confidence interval 110.188 to 125.395. n = 346.
Difference of means. 15.453 seconds, standard error 5.369, p = 0.00412.
The original write-up reported the gap as 15.4 seconds. The test output gives 15.453, so 15.5 is the closer round number, and we use it below.
What the numbers actually say
Now that the baseline is on the page, the finding can be sized.
15.5 seconds against a 117.8 second form is a 13% reduction in completion time, for a change that costs you a CSS rule.
Two things sit on top of that headline and both matter more than it does.
The size of the effect is much less certain than its direction. With a standard error of 5.369, the 95% interval around the difference runs from roughly 5 seconds to 26 seconds. That range is our arithmetic on the published standard error rather than a figure the study reported. The single-column form was faster. Whether it was faster by a trivial amount or by a fifth of the total time is something 702 people cannot settle.
The study's form ran just under two minutes. Yours probably does not. The same 15.5 seconds is worth a very different amount depending on where you start.
Which gives you the first action. Instrument your own form and get its median completion time, so you are working from your denominator instead of borrowing ours.
Why single or multi-column forms differ in completion time
The mechanism is documented even where the measurement is not.
Baymard's account is the clearest: "the different columns of empty fields and selectors draw users' attention in multiple directions, making it more difficult for users to correctly interpret the form."
A single column has one possible reading order. A multi-column layout has several, and the reader has to pick one before answering anything. That decision costs time on every row, and it is a decision the interface should have made for them.
Baymard's usability rounds show two consequences. Fields get skipped because they sat in a column the user never scanned, and fields get completed that did not apply.
The 15.5 seconds is the price of resolving an ambiguity that a single column never creates.
Completion time is not conversion
This is where the study needs a caveat it never got.
A faster form is not automatically a better-converting form. Completion time is a proxy for friction, and friction correlates with abandonment, but this study measured neither abandonment nor conversion.
There is a sharper version of that limitation, and it is the most important sentence in this post.
All 702 participants completed the form. The study can only see people who finished. Anyone who opened a multi-column form, decided it looked like work, and left is invisible in this dataset. If multi-column layouts drive abandonment, and Baymard's field observations suggest they do, this study systematically understates the cost.
So read the 13% as a directional signal about friction rather than a forecast of lift. Preference and speed tell you which direction to build. Behavior on your own traffic tells you what it is worth.
Can you test single or multi-column forms on your own traffic
Probably not, and that is the honest reason a 2016 study still earns its place.
Form pages are usually low-traffic surfaces with low baseline conversion, which is the combination that makes a detectable effect expensive. Run the numbers with the A/B test calculator before you commit to a test, and use the test bandwidth blueprint to see how many experiments your traffic actually supports at each step.
Most teams find their form page cannot referee a small effect in a reasonable window. When that is true, the should I run an A/B test checklist is a better starting point than the test queue, because some decisions should be settled by evidence you already have.
The choice between single or multi-column forms is one of those decisions. The direction is established, the mechanism is understood, and the cost of complying is a CSS change.
Save the traffic for the questions nobody has answered yet.
What about mobile?
The study is desktop-only, and that limit is worth stating plainly rather than burying.
On a phone, a single column is mostly forced on you by the viewport. That makes this a desktop finding, and it applies most directly to two situations.
Desktop form design, where you have the horizontal room to make a bad choice.
And responsive layouts that collapse to one column on small screens and fan out to two or three above a breakpoint. That second case is where teams reintroduce the problem without noticing, because the mobile view looks fine in review and the desktop view is the one carrying the multi-column layout.
Check your form at your widest breakpoint. That is where the 15.5 seconds lives.
What this study still does not answer
The original write-up was honest about its limits and that honesty is worth keeping.
The form variations used a fixed set of questions, specific field types, and one label treatment. The result does not transfer automatically to every form.
Three questions this dataset cannot settle, in the order we would test them:
Abandonment. The one that matters most and the one the design excluded by construction. A study that measures starts as well as completions would tell you what layout is actually worth.
Label placement. Above the field, beside it, inside it as a floating label. Widely argued, thinly measured.
Field types. Multi-select against radio buttons, across devices, is the follow-up we flagged in 2016 and still have not seen measured properly by anyone.
If you want to see how we structure a question like that before running it, problem-statement focused hypothesis is the format we use.
The answer
Use a single column.
The measured evidence is 102.3 seconds against 117.8 seconds across 702 desktop participants, a 13% reduction at p = 0.00412. The mechanism behind it is well documented. The cost of complying is close to zero.
What the evidence does not give you is a lift forecast, and anyone quoting this study as a conversion number is quoting something it never measured.
Three things to do with it:
Instrument your form's completion time, so you know whether your form sits above or below the two-minute mark this study ran at.
Check your widest breakpoint for a layout that fans out above a certain width.
And if a stakeholder wants the multi-column version anyway, put this study in front of them. It is still the only one on the question with a sample size worth arguing about.
Our review stars study uses the same approach on a different question. If you want this kind of evidence built for your own funnel rather than borrowed from ours, that is what our conversion rate optimization work does.
Study details. Fielded June 2016 through CXL Institute, published October 2016, updated 2026. n = 702 desktop participants, 356 single-column and 346 multi-column. Single-column mean 102.338 seconds, multi-column mean 117.791 seconds, difference 15.453 seconds, standard error 5.369, p = 0.00412 on a two-sample t-test. Every participant completed the form, so the study carries no information about abandonment.
The original statistical output from the 2016 test, kept here as the source for every figure above.
.avif)
.svg)
