Insights

When a dataset comes back messy, the instinct is to blame enumerators or respondents. In our experience, the real culprit is usually the instrument itself. Here are the five mistakes we fix most often.

1. Questions without constraints

If a field accepts any number, someone will eventually type a mother's age as 3 or a household size of 200. Every numeric question should carry a sensible range constraint, with a clear message that tells the enumerator what went wrong.

2. Skip logic that was never tested end to end

Relevance conditions interact. A question that depends on two earlier answers can behave perfectly in nine scenarios and fail silently in the tenth. Before fieldwork, walk through every realistic path in the form, not just the common one.

3. Free text where categories belong

Open text fields feel flexible, but "Amhara", "amhara region", and "AMH" are three different values to your analysis software. Use select questions with a well-designed choice list, and keep "Other, specify" as the safety valve.

4. Labels written for the office, not the doorstep

A question that reads well in a proposal can be unaskable in a live interview. Write labels in the exact words an enumerator should say aloud, and put clarifications in hints rather than cramming them into the question.

5. No unique ID strategy

Duplicate and missing IDs are the most painful cleaning problem because they are discovered last. Build IDs from structure, for example a cascading location code plus an auto-calculated sequence, instead of asking someone to type them by hand.

None of these fixes is glamorous. Together they are the difference between a dataset you defend and a dataset you apologize for.

Skip logic in XLSForm lives in the relevant column: a question appears only when its expression is true. The syntax is simple; the discipline is what takes practice.

Start with the variable, not the question text

An expression like ${has_children} = 'yes' refers to the question's name, so name your variables to be readable: has_children beats q17 every time someone reviews your logic.

Combine conditions carefully

Use and and or deliberately, and group with parentheses: ${age} >= 15 and (${status} = 'married' or ${status} = 'union'). When an expression grows past two conditions, write out its truth table in plain language before trusting it.

Remember that hidden questions keep their answers

If a respondent answers a question and a later change makes it irrelevant, most platforms clear it, but behavior varies with edits and re-opens. Decide what should happen in those cases and test it on your actual platform.

Test with scripted scenarios

Write five short fictional respondents, for example "a widow with two children who declined the income module", and enter each one into a test deployment. Scenario testing catches interaction bugs that reading the spreadsheet never will.

Done well, skip logic makes interviews shorter, kinder, and cleaner. Done casually, it hides entire modules from the exact respondents you needed to hear from.

Because all three major platforms share the XLSForm standard, a well-built form can move between them with modest changes. That means your choice is really about hosting, features, and support.

KoboToolbox: the accessible default

KoboToolbox offers free hosting for humanitarian and nonprofit use, a clean web form builder, and simple exports. For most NGOs and students, it covers everything needed for a standard survey. Its limits show up in very large projects and advanced quality control workflows.

ODK: control and flexibility

ODK is open source. With ODK Central you can host your own server, keeping data fully under your control, which matters for some ethics boards and government agreements. The trade-off is that someone has to run that server, so budget for technical capacity.

SurveyCTO: enterprise assurance

SurveyCTO is a commercial platform with strong encryption, monitoring dashboards, and features like audio audits that let supervisors verify interview quality. Large research studies and organizations with strict compliance needs are usually the ones that justify its subscription cost.

A simple decision path

1. Tight budget, standard survey: start with KoboToolbox.
2. Data must stay on your own server: choose ODK with ODK Central.
3. High-stakes research with audit requirements: cost out SurveyCTO.

Whichever you choose, invest in the form itself. A rigorous instrument on a free platform beats a careless one on an expensive platform every time.

For years, monitoring and evaluation in many organizations meant one thing: filling templates for the next donor report. That is changing, and digital data collection is a big part of why.

Real-time review is replacing end-of-project cleaning

When submissions arrive on a server the same day they are collected, supervisors can query suspicious records while the team is still in the area. Errors get corrected at the source instead of being discovered a year later in an evaluation.

Standardization across sites is the quiet revolution

Programs that once received a different spreadsheet from every field office are moving to shared digital forms with indicator calculations built in. Consolidation stops being a quarterly crisis and becomes a single export.

Data responsibility is now part of the job

Funders and ethics boards increasingly ask hard questions about where data lives, who can see it, and how consent was recorded. Form design plays a role here too: collecting only what you need is the first data protection measure.

What this means for your next project

Budget for instrument design the way you budget for training or transport. A monitoring tool that produces analysis-ready data changes what an M&E team can do with the same staff and the same money.