Monitoring change and evaluating mitigation

Monitoring tracks changes in conditions and implementation. Evaluation asks whether an intervention made a difference, for whom, and through what mechanism. Both are necessary for project-induced in-migration, but they require different evidence.

A workforce report can show how many people a contractor employs. A settlement survey can estimate changes in residence. A service record can show use or interruptions. None, alone, establishes whether migration caused an adverse effect or whether a management measure prevented it.

Start with a theory of change

For each major measure, explain how it is expected to change the impact pathway. If recruitment information is available away from the gate, fewer people may need to travel speculatively to obtain information. If off-site water delivery improves, households may spend less time collecting water and use fewer unsafe alternatives. If referral arrangements become accessible, people may obtain support earlier.

Identify the assumptions that connect the action to the outcome. Recruitment information needs to be trusted and reach job seekers. A water upgrade needs power, maintenance, distribution, and affordable access. A referral arrangement needs provider capacity and safe use.

This makes it possible to interpret failure. An action may be incomplete, an assumption may be wrong, or an external change may overwhelm an otherwise effective measure. Different explanations require different corrections.

Separate population, implementation, and outcome indicators

The following indicator set is proposed for adaptation to a project. Definitions and thresholds need agreement with responsible actors and affected people.

For every indicator, state the numerator, denominator, unit, population, spatial unit, observation period, data source, collection method, frequency, owner, privacy rule, and revision process. Define what a missing value means. An indicator without these elements may change meaning between reporting rounds.

Build a sampling and observation strategy

Choose frequencies that match the speed of change. Contractor workforce and accommodation records may need weekly review during mobilisation. Rental and service conditions may need monthly checks around peak demand. Household outcomes may justify less frequent surveys, with additional rounds when a major phase changes.

These are example frequencies, not prescribed standards. Consider seasonal cycles, staff capacity, cost, and the time needed to act on a finding. Collecting data more often than it can be interpreted and used does not necessarily improve oversight.

A household panel follows the same people and can reveal changes within households. Attrition becomes a material problem if those who leave are also those most affected. Record reasons for loss where safe and feasible, compare retained and lost respondents, and report the resulting limits.

A repeated cross-section describes the population at each round, including new arrivals, but changes may reflect changing composition. A useful design can combine a panel with refreshed settlement sampling. Weighting and uncertainty need to follow the actual sample design.

Use stable identifiers for settlements, facilities, indicators, and actions. When administrative boundaries or definitions change, preserve an explicit correspondence and show which comparisons remain valid. Do not silently replace an old series with a differently defined one.

Distinguish demographic change from project-induced movement

At its simplest, population accounting is:

B and D represent births and deaths; I and O represent in- and out-migration; and E represents measurement, coverage, or boundary-related discrepancies. The residual after natural increase is not automatically project-induced migration.

Arrival histories can identify reasons people report for moving. Project schedules, recruitment records, other investments, and regional shocks help test alternative explanations. Record multiple reasons where relevant instead of forcing one cause. Intentions at arrival may also differ from later experience.

For daily service demand, residential population may be the wrong denominator. Commuters, market users, and rotating workers can consume services without becoming usual residents. Use the appropriate demand population and avoid counting a person twice within the same estimate.

Evaluate mitigation with an appropriate comparison

Before-and-after changes can be informative, but they may reflect seasonality, inflation, another investment, or wider economic conditions. Where feasible, compare outcomes with places or groups that were similarly exposed to wider changes but differently exposed to the intervention.

Difference-in-differences subtracts change in a comparison group from change in the assessed group. Its causal interpretation depends on assumptions, including credible parallel outcome trends without the intervention (Gertler et al., 2016, Chapter 7).

For a fictional rent example, comparable rents rise from 200 to 260 units in the project area, while comparison-area rents rise from 180 to 207. The difference in changes is:

The calculation is descriptive until the design supports attribution. Several pre-intervention periods can help assess pre-trends, but do not prove the assumption. Check whether project workers also enter the comparison area, whether housing quality changes, whether rent records include the same market segment, and whether households relocate between groups.

Evaluating a recruitment reform is also different from evaluating the project's overall effects. Define the intervention, its start date, the affected group, and the outcome precisely. If reforms are introduced where conditions are already deteriorating fastest, simple comparisons may be biased by that selection.

Where a causal design is infeasible, use contribution analysis: specify the mechanism; verify implementation; examine the timing and distribution of change; seek independent evidence; and test rival explanations. Report a bounded conclusion, such as evidence consistent with reduced gate-based job seeking, rather than a numerical causal claim the data cannot support.

Set triggers that lead to action

Triggers need a response owner, time limit, and feasible action. Separate conditions requiring immediate protective action from trends requiring investigation. The examples below are planning proposals.

Numerical thresholds should reflect applicable technical requirements or locally justified management limits. Record the basis. Do not import a percentage or ratio solely because another project used it.

Interpret complaints and service records as situated evidence

Administrative records are generated through institutions and access conditions. Rising clinic attendance can indicate better access; falling attendance can reflect a closed road. More complaints can indicate improved confidence in reporting. Record system changes, outreach, and eligibility alongside the counts.

Protect sensitive information. Use restricted data for necessary analysis and carefully reviewed summaries for public reporting. In small populations, combinations of place, age, arrival period, or incident type can identify someone even without a name.

Make review a management process

A useful review combines indicator trends, critical actions, qualitative findings, uncertainty, and decisions. It asks which assumptions remain credible and which measures need adjustment. Invite affected groups to challenge the interpretation through safe and accessible channels.

Retain versioned data, definitions, analysis scripts, and decisions. Explain revisions to past estimates instead of concealing them. Plan who will maintain the records and services after construction, and how monitoring will address demobilisation, household departure, and remaining livelihood dependence.

The field tools provide an indicator specification and evidence ledger. The worked example illustrates how scenario assumptions can be changed and the resulting service priorities recalculated.