Problem Diagnosis and Intervention Logic · Lesson 2

Map causal pathways and assumptions

6 min read

Lesson objectives

  • Apply map causal pathways and assumptions to a development case, explaining inputs and decision limits.
Reading tools

A clear employment outcome does not make the pathway to it clear. For a service linking jobseekers with firms, drawing one arrow from ‘training’ to ‘jobs’ is insufficient. Separate inputs such as trainers, firms' time, and data; activities such as skill diagnosis, training, and interviews; immediate outputs such as completed learning plans and arranged interviews; intermediate outcomes such as stronger performance on a practical task; and final outcomes such as sustained paid work. Every arrow is a hypothesis that can fail. Attendance may be high while skills stagnate because the curriculum misses job requirements. Skills may improve while hiring does not if firms freeze recruitment. Naming the vulnerable arrow allows design changes before the whole budget is spent. Put an observable sign and a measurement date beside each arrow, selecting measures that distinguish plausible explanations of failure instead of filling a dashboard with every available count.

Distinguish operational assumptions from external ones. The initiative team can change session times, send reminders, and audit an employer list; it cannot ensure that the local economy creates vacancies. If employers keep selecting applicants with prior experience regardless of improved skills, adding training hours will not solve the fit problem. Maintain a short assumption register with the statement, current evidence, likelihood of failure, consequence for the decision, a low-cost test, and the person who will review it. Prioritize assumptions able to reverse the funding decision or alter the service design. A pathway diagram alone does not establish causality; it clarifies what to test. Nor does ‘we have an employer partner’ imply a hiring commitment. Specify a number of interviews or trial tasks, job types, a deadline, and a documentation method before treating a partnership as confirmed delivery capacity.

Inspect effects that spill beyond direct beneficiaries. A trainee might take a job that would otherwise have gone to a nonparticipant. That is displacement, not necessarily a net increase in area employment. An employer might gain productivity while the beneficiary incurs travel or childcare costs. Counting new jobs among participants is therefore insufficient for estimating net economic benefit or distributional fairness. Map who receives each benefit and who pays each cost, then separate evaluation questions: did trainees improve their position, did total work or production rise, and did the distribution of opportunity change? These questions are linked but require different comparisons and measures. If the initiative is small and market data are thin, mark displacement as unverified instead of assuming zero. Before monetizing employment gains, ask whether wages proxy additional production or transfer income from someone else; economic appraisal will return to this distinction.

A mechanism can be tested before final impact. Consider the ‘practical test to interview’ link: if passers cannot obtain interviews because partners have no vacancies, better training will not suffice. Ask each partner for actual trial tasks, job types, and interview decision dates, then link each candidate's path by identifier without publishing names in the committee report. The test output is verified transition capacity, not a promise of employment. If passers receive interviews but fail on a task absent from the curriculum, the remedy is different: change content and retest the skill measure. This design separates weak demand from poor skill fit. Set a stopping rule before the trial, such as failure to provide enough specified tasks by an agreed date. A numerical threshold chosen after seeing the results would turn a decision rule into a retrospective justification.

Different pathway links require different evidence. A reliable attendance log can measure session reach, but skill gain needs a task assessed with stable criteria before and after training or against a suitable comparison. Sustained employment needs timed follow-up and verification of the work definition. Area-wide impact requires an estimate of what would have happened without the service, and perhaps an assessment of displaced jobseekers. Mark each node with ‘known’ and ‘unknown.’ If records identify 150 passers but do not link them to the 30 trial tasks, do not call 30/150 a verified person-level transition; present it as a conditional capacity ratio. A small record-linkage error can become a large claim about program effectiveness. During design review, have a colleague trace five deidentified records from offer to outcome, checking that each transition preserves the person as the unit and records the event date.

Lesson summary

Map causal pathways and assumptions