What Information Can—and Cannot—Fix in a Labor Market

The Delhi field experiment gives a precise answer to the question that opened this series: labor-market information can change how people search without raising the overall probability that they are employed. This final synthesis connects the result to the starting beliefs and access constraints, the design of the intervention, the gendered search responses, and the later earnings and occupational patterns. It also turns the evidence into a conditional evaluation framework: before judging an information or digital-matching programme, ask which constraint it changes, for whom, through which search margin, and whether the change reaches a durable job outcome.

The answer is not that information is weak, nor that employment statistics are the only outcomes that matter. It is that information enters a labor market through people who already have expectations, networks, constraints, and options. It can recalibrate a forecast or make a route easier to use. Those changes can be consequential even when no new vacancy is created, but they can also stop at registration or application. The difference between those possibilities is the central analytical lesson of the Delhi study.

The answer is a two-part mechanism

The first part is belief correction. Participants entered the experiment with high aspirations and expected wages above the urban occupational benchmarks used by the researchers. Men were more upward-biased than women. For men, especially those whose expectations were furthest above the benchmark, personalized information reduced reported confidence that they were already in or moving toward an ideal job and lowered their perceived earnings gain from that job. When the expected return to a costly search falls, fewer applications or fewer search days can be an economically coherent response.

The second part is access relief. Women entered with lower employment exposure, were less likely to apply even when they had searched, and relied relatively less on the informal channels that men used for leads and referrals. The intervention gave them another route to vacancies. Women's overall search intensity did not fall, but their use of digital modes increased while offline search declined. This is not a universal gender rule; it is a response to the interaction between prior exposure, available channels, and the content of the information.

The same package could therefore move two different margins. For some men, it changed the perceived payoff of searching. For some women, it changed the cost and accessibility of finding relevant opportunities. The observed behavior—less search in one group, different search in another—makes sense once those mechanisms are separated.

flowchart LR A["Starting priors and access constraints"] --> B["Information design and take-up"] B --> C{"Primary constraint changed"} C --> D["Belief calibration"] C --> E["Lower cost of finding relevant work"] D --> F["Search intensity"] E --> G["Search channel and quality"] F --> H["Applications and employer response"] G --> H H --> I["Employment probability"] H --> J["Earnings and occupational category"] I --> K["No detected aggregate increase"] J --> L["Selective later gains for some women"] H -.-> M["Final offer, contract, acceptance, and duration not fully observed"] N["Vacancy supply and employer decisions"] --> I

The diagram has two important features. First, the branches are conditional. Information does not have one behavioral effect independent of the problem it addresses. Second, the final outcome depends on more than participant behavior. Vacancy supply, employer decisions, contract terms, acceptance, and persistence can all stand between an application and a good job. A programme can move the left side of the diagram without moving the employment box, or can change job quality for people who were already working.

Search quantity, search channel, and job quality are different achievements

The experiment's strongest behavioral result is not a simple increase in search. By the second endline, treatment reduced the full sample's likelihood of having searched by 3.5 percentage points, driven mainly by men. Men's any-mode search fell by 4.2 points overall and by about 5 points in the Personalized group; digital-search use also declined for men in the Personalized group. Among women, by contrast, there was no significant reduction in whether they searched at all. Their digital or online search increased by 4.3 points at six months and 6.0 points at twelve months, while offline search fell.

The difference between “more search” and “different search” is substantive.