baseline Archives - GeoPoll https://www.geopoll.com/blog/tag/baseline/ High quality research from emerging markets Wed, 18 Nov 2026 10:24:00 +0000 en-US hourly 1 https://wordpress.org/?v=7.1.4 https://www.geopoll.com/wp-content/uploads/2017/12/favicon-2.png baseline Archives - GeoPoll https://www.geopoll.com/blog/tag/baseline/ 32 32 Baseline and Endline Surveys: How to Run Them Remotely Without Losing Comparability https://www.geopoll.com/blog/baseline-endline-surveys/ https://www.geopoll.com/blog/baseline-endline-surveys/#respond Sat, 14 Nov 2026 10:06:25 +0000 https://www.geopoll.com/?p=26120 A baseline survey is only as useful as the endline you compare it to. If the two waves measure different things, reach […]

The post Baseline and Endline Surveys: How to Run Them Remotely Without Losing Comparability appeared first on GeoPoll.

]]>
A baseline survey is only as useful as the endline you compare it to. If the two waves measure different things, reach different people, or use different modes without a plan, the change you report may be an artefact of the method rather than the program.

That risk has grown as more evaluations move to remote data collection. Phone, SMS, WhatsApp and web surveys make it possible to reach communities that field teams cannot easily visit, and to do it faster and at lower cost. They also add new ways for waves to drift apart.

This guide covers how to design baseline, midline and endline surveys that hold up when you compare them, with a focus on remote and mixed-mode data collection in emerging markets.

What baseline, midline and endline surveys are

The three waves measure the same indicators at different points in a program’s life.

Wave When it runs What it answers
Baseline Before the intervention starts, or as early as possible Where are participants starting from? What targets are realistic?
Midline Partway through implementation Is the program on track? What needs to change before it ends?
Endline At or after the program ends What changed, by how much, and for whom?

The baseline sets the reference point. Without it, an endline can describe conditions but cannot show change.

The midline is optional on short programs but valuable on long ones. It gives implementers a chance to adjust while there is still time and budget to act.

The endline closes the loop. On its own it shows a snapshot. Compared with the baseline, it shows movement. Compared with a comparison group, it helps show whether the program caused that movement.

Example: The Seed Marketing Report, produced by GeoPoll and Resourced under the Seed Marketing Insights & Adoption (SMIA) program, ran three waves of data collection with farmers in Kenya and Uganda, from a 2023 baseline to a 2025 endline. Because the same themes were measured at each wave, the study could track how seed sourcing, variety switching, purchasing barriers and marketing channels shifted over two years, rather than describing a single point in time.

1. Start with indicators you can measure the same way every time

Comparability starts at the questionnaire. Each indicator in your results framework needs a question, or a small set of questions, that will be asked identically at every wave.

  • Lock the wording. Changing a question between baseline and endline, even to improve it, breaks the comparison. If a baseline question was flawed, keep it and add the improved version alongside it.
  • Lock the answer options and scales. A 4-point scale at baseline and a 5-point scale at endline cannot be compared directly.
  • Lock the recall period. “In the last 7 days” and “in the last month” produce different numbers for the same behaviour.
  • Fix the translation. Use the same approved translation in every wave and every language. Back-translate once, then version-control it.
  • Keep question order stable. Earlier questions shape answers to later ones. Moving a block can shift results.
  • Use standard instruments where they exist. Validated tools such as the reduced Coping Strategies Index or the Food Consumption Score already have tested wording and make results comparable with other studies.

Write a short indicator reference sheet before baseline fieldwork. For each indicator, record the exact question text, the calculation, the population it applies to, and the target. The endline team, who may not be the baseline team, will rely on it.

2. Decide early: panel or repeated cross-section

The biggest sampling decision is whether you will interview the same people at endline or a fresh sample from the same population. Make it before baseline, because a panel design needs things a cross-section does not.

Factor Panel (same respondents) Repeated cross-section (new respondents)
Measures Change within individuals or households Change at population level
Statistical power Higher for the same sample size Lower; needs larger samples to detect the same change
Main risk Attrition between waves Sample composition shifts between waves
Needs at baseline Consent to recontact, multiple contact details, unique IDs A documented, repeatable sampling frame and method
Best for Beneficiary outcomes, livelihoods, behaviour change Population awareness, perceptions, coverage

A hybrid is common: recontact the baseline sample at endline, and top up with fresh respondents to replace those lost. Analyse the panel and the top-up separately before combining them.

Whichever you choose, size the sample for the endline, not the baseline. Work out the smallest change the evaluation must detect, then add a buffer for attrition or non-response. See our guide on how to determine sample size for the core calculation.

Document the sampling frame, stratification and selection method in enough detail that a different team could repeat it exactly. For a cross-section, the endline is only comparable if it draws from the same frame in the same way. Our explainer on sample frame and sample error covers what to record.

3. Choose modes for every wave, not just the first

The same question can produce different answers depending on how it is asked. Respondents tend to give more socially acceptable answers to a live interviewer than to a self-completed SMS or web survey. Long answer lists behave differently when read aloud than when shown on a screen. These are mode effects, and they can look like program impact.

The safest design uses the same mode, or the same mix of modes, at every wave. When that is not possible, plan for it.

  • Match the mode to the instrument. SMS suits short instruments with simple answer options. CATI handles longer questionnaires, complex skip logic and sensitive topics. Mobile web and WhatsApp suit image-based questions and smartphone users. In-person interviews (CAPI) reach offline populations.
  • Keep the mode mix stable. If baseline was 70% CATI and 30% in person, aim for a similar split at endline, and record the mode for every interview so you can control for it in analysis.
  • Run a bridge study when you must switch. If baseline was in person and endline must be remote, interview a subsample in both modes. The difference gives you an estimate of the mode effect to adjust for.
  • Design for the hardest mode. If any wave may run by phone, write the baseline questionnaire so it works when read aloud. Short stems and few answer options transfer best.

For a fuller comparison of modes, see CAPI, CATI and CAWI research methods. If you are considering AI-led phone interviews for later waves, our practitioner’s guide to AI enumeration explains why a bridge study matters.

4. Plan retention at baseline, not at endline

In a panel design, the people you lose between waves are rarely random. Those who move, change numbers or stop responding often differ from those who stay. High attrition can bias endline results even when the remaining sample looks large enough.

Most retention work happens at baseline:

  • Get consent to recontact. Explain that you will call again, roughly when, and why.
  • Collect more than one contact. A second phone number, a household member’s number, or a community contact gives you a fallback when the main number fails.
  • Assign a stable respondent ID. Link every wave to it, not to a phone number that may change.
  • Stay in touch between waves. A short check-in or midline keeps contact details current and respondents familiar with the study.
  • Use consistent, timely incentives. Small airtime or mobile money rewards paid on completion support participation across waves. Pay them the same way each time.
  • Plan the endline contact protocol. Set the number of call attempts, the times of day and the fallback contacts in advance, and apply them evenly.

At endline, compare the baseline profile of respondents you retained with those you lost. If they differ on key indicators, report it and consider weighting to correct for it. Our explainer on nonresponse bias covers the options.

Example: For UNICEF’s High-Frequency Survey in Thailand, GeoPoll completed 2,583 CATI interviews in the first round, reaching respondents through verified random digit dialling. Every respondent consented at round one to be recontacted for the second round, which gave the follow-up wave a ready, nationally representative pool to work from.

5. Add a comparison group if you need to show attribution

A baseline-to-endline change shows that something moved. It does not show that the program moved it. Prices, weather, elections and other programs all change outcomes too.

If the evaluation must show attribution, you need a comparison group surveyed at the same waves, with the same questionnaire and the same mode.

  • Randomised control groups give the strongest evidence but must be built into program design from the start.
  • Matched comparison areas or households are a practical alternative when randomisation is not possible. Match on baseline characteristics that predict the outcome.
  • Difference-in-differences compares the change in the program group with the change in the comparison group. It needs baseline data for both.

Remote data collection lowers the cost of surveying a comparison group, because you do not need field teams in additional locations. That makes attribution more affordable on programs that would otherwise skip it.

Example: When Farm Radio International commissioned GeoPoll to measure the impact of its radio programming on banana farmers in Uganda, the SMS survey covered both listeners and non-listeners in the target districts. Among listeners, 83% said they had adopted practices promoted on the program. But listeners and non-listeners reported adopting new practices at similar rates overall (85% and 88%). Without the non-listener group, the adoption figure alone would have overstated the program’s distinct effect.

If a comparison group is not feasible, say so in the evaluation design. Then strengthen the analysis with contribution evidence: midline data, qualitative interviews with participants, and context data on what else changed.

Baseline checklist: before you field

Use this list to confirm the baseline will support a credible endline.

  • Every indicator has fixed question wording, answer options and recall period
  • An indicator reference sheet records question text, calculation and target
  • Translations are approved, back-translated and version-controlled
  • Panel or cross-section design is decided and documented
  • Sample size is calculated for the change the endline must detect, plus an attrition buffer
  • Sampling frame, strata and selection method are written down in full
  • Mode or mode mix is chosen with later waves in mind, and recorded per interview
  • Consent to recontact, backup contacts and stable respondent IDs are collected (panel designs)
  • Incentive type, amount and payment method are set for all waves
  • A comparison group is included, or its absence is justified
  • The questionnaire has been piloted in every language and mode
  • Raw data, codebook and cleaning scripts will be stored where the endline team can find them

How GeoPoll supports baseline and endline surveys

GeoPoll has run multi-wave surveys for development and humanitarian partners for more than a decade. During the West Africa Ebola response, we ran thirteen rounds of nationally stratified surveys in Liberia and Sierra Leone for the USAID Bureau for Africa and FHI360, with 1,000 completes per country per round. More recently, the Seed Marketing Report tracked farmers in Kenya and Uganda from baseline to endline across three waves. Both depended on the principles in this guide: stable instruments, documented sampling and consistent modes across rounds.

For baseline, midline and endline work, we offer:

  • Multi-mode data collection: SMS, CATI, mobile web, WhatsApp, the GeoPoll app and in-person CAPI, combined by design and recorded per interview.
  • Our own respondent panels and call centres in Africa, Asia and Latin America, so samples can be recontacted across waves.
  • Survey design support, from indicator mapping and questionnaire review to sampling plans that hold up at endline.
  • Translation and localisation in the languages respondents actually use.
  • Mobile airtime and mobile money incentives paid on completion.
  • Comparison group fieldwork in the same waves and modes as the program group.

Planning an evaluation or responding to a baseline RFP? Contact the GeoPoll team to discuss design, feasibility and timelines.

Frequently asked questions

What is the difference between a baseline survey and an endline survey?

A baseline survey measures conditions before a program starts. An endline survey measures the same indicators when the program ends. Comparing the two shows how much changed over the life of the program.

Can a baseline survey be done remotely?

Yes, as long as the target population can be reached by phone or online and the questionnaire is designed for the mode. Remote baselines are common for short instruments and in areas that are hard for field teams to access. Where some groups are offline, mix remote and in-person modes and keep that mix stable across waves.

Should I interview the same people at baseline and endline?

Interview the same people (a panel) if you need to measure change in individuals or households. Use a new sample from the same population (a repeated cross-section) if you need population-level estimates. Panels give more statistical power but require a retention plan from baseline.

What if the baseline was done in person and the endline must be remote?

Run a bridge study. Interview a subsample both in person and remotely, estimate the mode effect, and adjust for it in analysis. Report the switch and the adjustment in the evaluation.

How large should a baseline sample be?

Size it for the endline. Start from the smallest change the evaluation must detect, calculate the sample needed to detect it, and add a buffer for attrition or non-response.

The post Baseline and Endline Surveys: How to Run Them Remotely Without Losing Comparability appeared first on GeoPoll.

]]>
https://www.geopoll.com/blog/baseline-endline-surveys/feed/ 0