What the Messages Changed First: Attention, Applications, and Search Quality

The first visible response to the Delhi intervention was not employment. It was activity on the partner platform. This chapter examines what happened in the weeks after each message, how to distinguish an immediate response from a cumulative one, and why the Personalized and Non-Personalized groups can produce different patterns of activity without one simply being “more motivated.” The practical judgment is modest but important: an early increase in applications shows that information reached behavior, not that it produced a good match.

The distinction matters because search is a process spread over time. A message may prompt a person to open an app today, browse vacancies tomorrow, apply next week, and then receive no response. Another message may cause the person to leave the platform for a government portal or a skills programme. Looking only at the final number of applications can hide both the timing of attention and the way the search route changes.

The first message changed next week's behavior

The researchers linked administrative records from the partner platform to the timing of the six messages. They used an event-study design with individual and week fixed effects to estimate how online applications changed after each message. In plain language, the analysis compares a participant's application activity around a message with their activity before that message while accounting for stable differences between people and common week-to-week changes.

Before the first message, the platform average was 1.552 online applications in the relevant period. In the week after Message 1, treated participants made 0.152 additional applications. The paper describes this as a 9.8% increase over the pre-message average. This is a meaningful first-stage result: the message did more than exist in a phone inbox. It was followed by an observable change in the platform behavior the intervention was designed to encourage.

The timing is also informative. A response in the week after contact is consistent with attention, a reminder, or newly usable knowledge about how to search. It is less likely to be explained by a twelve-month change in employment conditions, although the estimate still does not tell us whether the applications were relevant or successful. The result therefore belongs near the beginning of the causal chain: information was followed by search activity.

sequenceDiagram participant J as Job seeker participant M as Message stream participant P as Partner platform participant G as Other job portal Note over M: Messages arrive at roughly fortnightly intervals M->>J: Message 1 prompts search guidance J->>P: Apply during the following week alt Personalized content M->>J: Point toward relevant preferences and vacancies else General content M->>J: Provide broad market and search information end J->>P: Explore and submit applications M->>J: Later messages include skills and government resources alt Continue on partner platform J->>P: Continue searching else Follow another route J->>G: Search government opportunities end

The diagram shows why message timing cannot be separated from message content.