Eyes vs. Fingers: The Battle Royale of Data Entry Accuracy

Quick, imagine knowing nothing about corneal staining (okay, check 😊), and you've been tasked with rating eye disease severity.

Now look at the two forms below. Which one would you rather use to rate the images?

Form A probably jumped out. And you didn't need to know a thing about corneal staining to pick it.

Your eyes made the decision before your fingers ever touched the form.

Keep in mind that both forms ask the same questions and collect the same ratings. But getting those ratings into the form is a very different experience. One makes the job considerably easier.

So what's going on? Turns out there's a lot more here than meets the eye.

Let's start with probably the most intuitive part of what's happening: spatial relationships. The source already has a shape. The staining appears in a particular location on an eye. The question is whether the form preserves that relationship or throws it away.

When the Form Throws Away the Map

When we talk about accurate data entry, the discussion usually centers on what the fingers do: put values into fields.

The familiar techniques are worth knowing and using: skip and branching logic, showing or hiding fields, required fields, valid ranges, data type restrictions, input masks and formatting rules, cross-field consistency checks, and conditional validation rules.

All these matter. They help keep out-of-range values, missing responses, and contradictory answers from making their way into a dataset.

But notice where the attention is focused: on the data. Is the value present? Is it allowable? Is it formatted correctly? Does it agree with the values around it?

What's largely missing from that discussion is the work required to produce and enter it.

Form B is what happens when we primarily organize around the fields.

It's a logical layout. The locations run down the rows. Right and left eyes run across the columns. Ten fields. Compact. Organized. Easy to understand when you're thinking about the structure of the dataset.

But that organization comes at a cost for the person entering the data. Consider what happens when they see staining in the lower part of the left eye:

1. Translate what they're seeing into the word Bottom.
2. Find the Bottom row on the form.
3. Remember that the left eye is OS.
4. Move across to the OS column.
5. Find the dropdown.
6. Enter the rating.

Now do it again. And again. Ten times, for multiple participants and visits.

Some of that work is required by what we're trying to collect. The rater still must examine the eye and decide whether the staining is diffuse, localized, or not applicable. The form can't eliminate the judgment we're trying to collect.

But much of the remaining work is different. Translating where the staining appears into Bottom, finding the Bottom row, remembering that the left eye is OS, moving to the correct column, and locating the field don't help the rater judge the staining.

We created that work with the form.

Each step creates another opportunity for error before the value ever reaches the field. Get one of them wrong and the dataset says the disease was somewhere it wasn't.

The particularly troublesome part is that this type of error may not be caught by validation rules because the value itself is perfectly allowable. It may generate no warning, query, or even stand out during cleaning. Someone lost track during all that mental mapping, and the data simply record something other than what was observed.

This is a problem you can design out of the form.

Preserve What's Already There

Look at what happened to the source. The staining started as spatial information. Its location on the eye was part of the observation.

Form B received that structure for free and threw it away.

Then it asked the person to rebuild the structure using labels, abbreviations, rows, and columns. The observation started spatial. The form turned it into a lookup problem.

Form A does something much simpler. It preserves the structure that was already there.

The two circles correspond to the two eyes and appear in the same left-to-right order as the images above them. Each circle is divided into the same spatial regions the rater is examining.

Now the process collapses: see staining toward the bottom of the left eye, move to the bottom of the left diagram, enter the rating.

Less to translate. Less to remember. Less to search for.

Form B moves the work to the person. Form A moves the work into the design.

That's the value of spatial correspondence in this task. When location already carries meaning in the source, preserving that relationship in the form can eliminate mental mapping that has nothing to do with making the observation itself.

Anatomy is an obvious example. But the same question applies anywhere the source contains meaningful spatial information.

Now Let's Make It Harder

It would be easy to finish here with a simple rule: If the source is spatial, make the form spatial.

Except that rule is wrong.

We've concentrated on one part of our framework: the source. What happens when we bring in the person?

Next in Part 3, the example also involves anatomy and spatial location. Following everything we just learned, the answer should be: make the form look like the anatomy.

Right? Not quite. We'll see why the person doing the work can completely change the answer.